Output Measurement in the Service Sectors 9780226308890

Citation preview

Output Measurement in the Service Sectors

Studies in Income and Wealth Volume 56

National Bureau of Economic Research Conference on Research in Income and Wealth

Output Measurement in the Service Sectors

Edited by

Zvi Griliches with the assistance of

Ernst R. Berndt, Timothy F. Bresnahan, and Marilyn E. Manser

8$

The University of Chicago Press Chicago and London

ZVIGRILICHES is the Paul M. warburg Professor of Economics at H m a r d University and director of the Productivity Program at the National Bureau of Economic Research.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London 01992 by the National Bureau of Economic Research All rights reserved. Published 1992 Printed in the United States of America 01 00 99 98 97 96 95 94 93 92 ISBN (cloth): 0-226-30885-5

1 2 3 4

5 6

Copyright is not claimed for chap. 1 by Michael F. Mohr, chap. 6 by Dennis J. Fixler and Kimberly D. Zieschang, and “Comment” on chaps. 6 and 7 by Jack E. Triplett. Library of Congress Cataloging-in-Publication Data Output measurement in the service sectors / Zvi Griliches with the assistance of Ernst R. Berndt, Timothy F. Bresnahan, and Marilyn E. Manser. p. cm.-(Studies in income and wealth ; v. 56) Includes bibliographical references and index. 1. Service industries-United States-Labor productivity. I. Griliches, Zvi. 11. Series. HC106.3.C714 vol. 56 [HD9981.5] 330 s-dc20 [658.3’14] 92-17561 CIP

@The paper used in this publication meets the minimum requirements of the American National Standard for Information Sciences-Permanence of Paper for Printed Library Materials, ANSI 239.48-1984.

National Bureau of Economic Research Officers George T. Conklin, Jr., chairman Paul W. McCracken, vice chairman Martin Feldstein, president and chief executive ojicer

Geoffrey Carliner, executive director Charles A. Walworth, measurer Sam Parker, director ofjnance and administration

Directors at Large John H. Biggs Andrew Brimmer Carl F. Christ George T. Conklin, Jr. Don R. Conlan Kathleen B. Cooper Jean A. Crockett George C. Eads

Martin Feldstein George Hatsopoulos Lawrence R. Klein Franklin A. Lindsay Paul W. McCracken Leo Melamed Robert T. Parry

Peter G. Peterson Douglas D. Purvis Robert V. Roosa Richard N. Rosett Bert Seidman Eli Shapiro Donald S. Wasserman

Directors by University Appointment Jagdish Bhagwati, Columbia William C. Brainard, Yale Glen G . Cain, Wisconsin Franklin Fisher, Massachusetts Institute of Technology Jonathan Hughes, Northwestern Saul H. Hymans, Michigan Marjorie B. McElroy, Duke

James L. Pierce, California, Berkeley Andrew Postlewaite, Pennsylvania Nathan Rosenberg, Stanford Harold T. Shapiro, Princeton Craig Swan, Minnesota Michael Yoshino, Harvard Arnold Zellner, Chicago

Directors by Appointment of Other Organizations Marcel Boyer, Canadian Economics Association Rueben C. Buse, American Agricultural Economics Association Richard A. Easterlin, Economic History Association Gail Fosler, The Conference Board A. Ronald Gallant, American Statistical Association Robert S . Hamada, American Finance Association

Charles Lave, American Economic Association Rudolph A. Oswald, American Federation of Labor and Congress of Industrial Organizations Dean P. Phypers, Committee for Economic Development Charles A. Walworth, American Institute of Certijed Public Accountants

Directors Emeriti George B . Roberts Moses Abrarnovitz Gottfried Haberler Williard L. Thorp Geoffrey H. Moore Emilio G . Collado James J. O’Leary William S. Vickrey Thomas D. Flynn Since this volume is a record of conference proceedings, it has been exempted from the rules governing critical review of manuscripts by the Board of Directors of the National Bureau (resolution adopted 8 June 1948, as revised 21 November 1949 and 20 April 1968).

This Page Intentionally Left Blank

Contents

Prefatory Note

xi

Introduction Zvi Griliches

1

I. WHATIs BEINGDONE 1. Recent and Planned Improvements in the Measurement and Deflation of Services Outputs and Inputs in BEA's Gross Product Originating Estimates Michael F. Mohr Comment: Martin Neil Baily

25

2. Productivity Measurement in Service Industries 73 Edwin R. Dean and Kent Kunze Comment: W. Erwin Diewert Reply: Edwin R. Dean and Kent Kunze Comment: Robert E. Lipsey

3. Improvements in Measuring Price Changes in Consumer Services: Past, Present, and Future Paul A. Armknecht and Daniel H. Ginsburg

109

Comment: Robert E. Lipsey

11. ALTERNATIVE APPROACHES

4. Productivity in the Distributive "kades: The Shopper and the Economies of Massed Reserves Walter Y. Oi Comment: Sherwin Rosen vii

161

viii

Contents ~

5. The Real Output of the Stock Exchange Timothy F. Bresnahan, Paul Milgrom, and Jonathan Paul

195

IIIA. MEASURING THE OUTPUTOF BANKING AND OTHERSERVICES

6. User Costs, Shadow Prices, and the Real Output of Banks Dennis J. Fixler and Kimberly D. Zieschang Comments (follow chap. 7 ) 7. Measurement and Efficiency Issues in Commercial Banking Allen N. Berger and David B. Humphrey Comment: Frank C. Wykoff Comment (chaps. 6 and 7): Jack E . Triplett Comment (chaps. 6 and 7): Diana Hancock

219

245

IIIB. SELECTED INDUSTRIES

8. The Output of the Education Sector Dale W. Jorgenson and Barbara M. Fraumeni Comment: Michael Rothschild 9. Measurement of Output and Quality Adjustment in the Day-care Industry Swati Mukerjee and Ann Dryden Witte 10. Productivity in the Transportation Sector Robert J. Gordon Comment: Robin C. Sickles

303

343 37 1

11. Purchased Services, Outsourcing, Computers, 429 and Productivity in Manufacturing Donald Siege1 and Zvi Griliches Comment: M. Ishaq Nadiri 12. Dispersion and Heterogeneity of Firm Performances in Nine French Service Industries, 1984-1987 Elizabeth Kremp and Jacques Mairesse

46 1

IV. INTERNATIONAL ASPECTS

13. Measuring Final Product Services for International Comparisons Alan Heston and Robert Summers

493

ix

Contents

14. Measuring Public-Sector Output: The Swedish Report Richard Murray

517

Contributors

543

Author Index

547

Subject Index

553

This Page Intentionally Left Blank

Prefatory Note

Preliminary versions of the papers and discussions contained in this volume were presented at a conference held May 4-5, 1990 in Charleston, South Carolina. Funds for the Conference on Research in Income and Wealth are provided by the Bureau of the Census, the Bureau of Economic Analysis, the Bureau of Labor Statistics, the U.S. Department of Energy, the Statistics Division of the Internal Revenue Service, and Statistics Canada. We are indebted to all of them for their support. Executive Committee, May 1990 Charles Hulten, chair Ernst Berndt Geoffrey Carliner, NBER representative Carol Carson Rosanne Cole Frank de Leeuw Stanley Engerman

Zvi Griliches, NBER representative Stanley Lebergott Robert Lipsey Marilyn Manser Robert Parker Sherwin Rosen Fritz Scheuren Charles Waite

Volume Editors’ Acknowledgments The conference from which the volume was developed was organized under the auspices of the Conference on Research in Income and Wealth by a committee consisting of Ernst R. Berndt, Timothy F. Bresnahan, Marilyn E. Manser, and Zvi Griliches, chairman. They also served as the editorial committee

xi

xii

Prefatory Note

for this volume. We are grateful to the National Science Foundation and the Sloan Foundation for financial aid in this endeavor. We are indebted to many people who helped make this conference a success and helped in the production of this volume, but especially to Geoffrey Carliner, Jeanette DeHaan, Kirsten Foss, Ilana Hardesty, Charles Hulten, Jane Konkel, and Annie Spillane.

Introduction Zvi Griliches

The main point is that ingenuity cannot fully or effectively compensate for lack of basic information. Kuznets (1941, 111)

The continued growth of service sectors in almost all the developing economies has fascinated and occasionally alarmed economists and other observers. The recent record in the United States is displayed in figure 1 and table 1 . Overall, the relative growth of employment in services (excluding government) was rather slow until the early 1960s, the upward trend that had started in the 1920s having been interrupted by the Great Depression and the war years. The trend accelerated, however, beginning in the mid- 1960s. Why should this be viewed with alarm? Table 1 provides a partial answer: productivity as measured in the national accounts has grown significantly slower in services, especially in the early postwar period, 1948-60, and in the most recent decade, 1979-89. That slowness of growth, together with the rising share of services in nominal GNP and in employment, has been viewed as a major drag on the productivity growth of the overall economy and its competitive performance.’ There are at least two, possibly complementary, explanations of these phenomena. The first is slower technical change in services, resulting from their Zvi Griliches is the Paul M. Warburg Professor of Economics, Harvard University, and director of the Productivity Program at the National Bureau of Economic Research. The author is indebted to the Bradley, Guggenheim, and Sloan foundations for financial support. He has benefited from the comments of many of the participants in this conference on an earlier draft of the introduction. The author would like to thank especially Ernst Berndt, Robert Fogel, Robert Gordon, Robert Lipsey, Marilyn Manser, and Jack Triplett for their comments on the draft as a whole. 1, Note that “commodities” do include agriculture in table 1 but not in figure I .

1

2

Zvi Griliches

2.5

2 .o 1.5 1.o

0.5

I

0.0

D

Fig. 1 Relative share of nonagricultural employment: “Services” versus “commodities”

Source: Historical Statistics of the U.S., pt. 1, D127-41, and the Economic Report of the President, 1991, table B-43. Note: “Services” exclude government; “commodities” exclude agriculture.

Table 1

Services versus Commodities, 1947-1989 1947 or 1948

1960

1969

1979

1989

Share of GNP (%):* Current prices: Commodities Services Constant prices: Commodities Services

43.8 40.1

39.4 49.9

36.5 50.7

33.8 53.3

27. I 60.7

39.9 45.5

39.7 46.2

37.7 48.0

34.4 52.5

32.4 57.7

Implicit relative price:* Services/Commodities

80

109

110

104

126

108

99

I02

100

88

Relative productivity (GNP/hour):*

Sources: *Mohr (chap. 1, in this vol.), Survey of Current Business, A ril 1991 27 (for 1989), i and NIPA ofthe U S . , 1928-82: Statistical Tables (for 1947 and 1948). Ratio of shares in current and constant dollars. *Hours worked by industry from Survey ofCurrent Business, July 1990 and the NIPA of the U.S. Note: Hours and hence GNPihour series start in 1948. The numbers for 1947-69 are not fully comparable to the 1979-89 estimates. The latter are based on the newly revised methodology described in Mohr.

P .

3

Introduction

intrinsically more labor intensive nature, and a potentially higher income elasticity of the demand for them (see, e.g., Baumol 1967; and Baumol, Blackman, and Wolf 1985). Only the first part of this explanation is actually worrisome. But before one accepts it as a fact, one needs to consider the second explanation, the possibility that difficulties in measuring output and prices in services may have resulted in a mismeasurement of productivity growth in these sectors, a mismeasurement that accounts for some or even much of the observed contrast with the productivity experience of commodities. It was this latter possibility that motivated the Conference on Research in Income and Wealth to organize a conference on this range of topics, the edited proceedings of which make up this volume. An earlier conference, organized by Victor Fuchs (Fuchs 1969b), was held in Ottawa in 1967, over 20 years ago. The organizing committee of the current conference (Emst Bemdt, Timothy F. Bresnahan, Marilyn Manser, and Zvi Griliches, chair) felt that, because of the importance of the topic, newly available data, and further methodological developments, such a conference could contribute to a better understanding of the issues at hand. After a planning period of about a year, the conference was held in Charleston, South Carolina, on May 4-5, 1990. In organizing this conference we faced the problem that services are actually a rather amorphous concept, covering a heterogeneous set of industries. Figure 2 illustrates the different historical trends of its various components: a decline in the share of employment in the transportation, communication, and public utilities industries; a continued, relatively constant rate of growth in the share of trade, finance, insurance, and real estate (FIRE) industries; and a relatively sharp growth in the share of the not-elsewhere-classified service industries, especially in business, health, and personal services.* Similar disparities can be seen in table 2, which records the productivity experience of the different subsectors. The average levels of labor productivity were not significantly lower in the service sectors in 1948 than in commodity production, with the possible exception of retail trade. Over time, only retail trade and other services fell significantly behind the productivity developments in commodity production. In fact, it was only relatively recently that the average of GNP per hour in all the service sectors fell below that of commodity production. But productivity growth was indeed slower in the fastest-growing subsectors: retail trade, FIRE, and services, a fact that contributed to the overall decline in the relative performance of services as a whole. Because these are also the industries where output measurement may be most difficult, the suspicion is raised that some of the observed declines could be spurious-the 2. Government is excluded from these tables and figures because our focus is on productivity measurement and only a small fraction of this sector has reasonable productivity measures. Also, the sources and determinants of growth of the government sector are somewhat different and adequate coverage would take us much beyond what could be done in this volume. Two of the chapters in this volume, Jorgenson and Fraumeni and Murray, do discuss some of the measurement issues that arise in this sector.

4

0’35

Zvi Griliches

1

0.30-

0.25 -

,-0

-.-.-.::I-

0.20 -

.

,,-,----.I’

8

I

.--. ;-I

_---Finance an4_TrrdE_,.--‘

\

I I I

I

t\ Jl

1 1 1 I

*/-*

0----8---

/

/

I‘

0’

/

l

“Services”

, a

3 Fig. 2 Shares of total nonagricultural employment: Selected Sectors

result of our inability to observe and interpret the historical developments correctly. To explore these issues, the committee decided to divide the conference into two parts. First, it would include a description of how services output and prices are currently measured by the major official U.S. data-collection and data-construction units: The national income accounts (GNP “originating” by industry) produced by the Bureau of Economic Analysis (BEA) in the Department of Commerce; and the consumer price index (CPI) and the productivityby-industry series, both constructed and reported by different sections of the Bureau of Labor Statistics (BLS) in the Department of Labor.3 And, second, it would include a discussion of alternative approaches to the measurement of output in some of these industries and studies of specific subindustries and other related topics. Before reviewing the resulting studies in some detail, it may be worthwhile to say a few more words about the slippery concept of services and why it is so problematic. The difficulties of discussing services arise the moment one tries to define them. A standard dictionary, Webster’s Collegiate (1946), lists as its 15th def3. The PPI (producer price index) is not considered explicitly in this section since its coverage of services is limited to transportation and telephone services.

5

Introduction

Table 2

Constant-Dollar GNP per Hour in Selected Sectors

A. Levels (in dollars)

Sector

1948

I960

1969

1979

1989

Commodities, total Service sectors: Transportation, commercial & public utility Wholesale trade Retail trade FIRE Services

10.7

15.1

17.9

20.2

25.3

11.6 10.0 8.0 32.8 10.3

16.4 14.5 9.7 42.8 11.5

23.0 19.1 11 .o 48.7 13.2

30.2 19.7 12.1 51.1 14.5

38.2 25.6 14.4 51.4 15.2

1947-60

196M9

1969-79

1979-89

1948-89 (average)

2.9 2.1 -0.8

1.9 2.2 0.3

2.3 1.o -1.3

2.1 1.6 --0.5

2.9 3.0 1.6 2.2 0.9

3.8 3.1 1.4 1.4 1.5

2.4 2.6 1.8 0.1 0.5

3.0 2.3 1.4 1.1

B. Rates of Growth (per annum, %) Sector

~

Commodities, total Services, total Service - commodities Service sectors: Transportation, commercial & public utility Wholesale trade Retail trade FIRE Services

1.2

1.o

-.0.2 2.8 0.3

1.o

0.5 0.9

1.o

inition (among 20) of the word “service” “Any result of useful labor which does not produce a tangible commodity; as, railroads, telephone companies, laundries, and physicians perform services.” Note the negative definition: “not a tangible commodity.” A useful definition is provided by Hill (1977) in which the user (consumer) or the user’s goods are changed by the provider of the service. This definition captures the aspect of being worked on or moved either for persons, as in haircuts, physician visits (though not every psychoanalytical session results in a transformation), or airplane trips; or for goods, as in car repair, tailoring, or warehousing. One has to emphasize, however, the distinction between the production process of such activities, the question of legal ownership of the items being worked on, and the payment format-is the price paid for services rendered, a recompense for agency efforts on behalf of the principal, or a direct purchase of a commodity from a middleman? What is important in this definition is the recognition of the role of the purchaser in such a transaction, either because of her direct involvement in the activity and consequent contribution to its ultimate output or as a supplier of one of the major inputs to it. Although the quality of a commodity does not usually depend on the “quality” of its consumers (though the demand for it

6

Zvi Griliches

may), the output of a service activity may depend on the quality and/or effort of its consumers, as in teaching and related advisory services. Rather than discussing definitions, it may be more useful to take an operational approach and to examine what are actually called services in the national accounts and related statistical sources. The broadest definition of services corresponds to the nontangible, noncommodity notion: everything except agriculture, mining, construction, and manufacturing. This notion defines the scope of this volume but also leaves one wanting to quarrel with it from time to time. It includes transportation, communication, public utilities, wholesale and retail trade, FIRE, repair, personal, business, health, legal, and other services; and the activities of federal and local governments. It is troubled by the fact that electricity is tangible, that, although gas in bottles is a commodity, gas from a pipe is a service, and that in many cases the definition and measurement of an activity depend on rather arbitrary boundaries. If manufacturers conduct more of their trade activity out of their plant offices much of “trade” shifts back to manufacturing. Alternatively, buying prepared meals shifts output from the household sector to retail trade. Additional complications arise from the fact that governments and many nonprofit institutions do not sell their output directly, provide no relevant transaction data, and hence require a variety of more or less unsatisfactory imputation techniques. Although services are different, they are really not so different from goods as far as the problem of measuring output is concerned. Most of the problems afflicting the measurement of commodity output affect also the measurement of services, only more so. To measure the output of any activity we need to know its total receipts and have adequate information to construct an appropriate price index for it. To measure productivity, we need in addition parallel information on the inputs used in production (total costs and prices or units used). In either case, we need to know the relevant transaction unit and deal with the problem of quality change, which arises form the underlying heterogeneity of outputs and inputs and the continuing appearance of new products, varieties, and services, and the disappearance of old ones. Why is the problem more serious in some of the service sectors? Partly it is a data problem, but also, importantly, it is a conceptual one. Historically, much more data were collected on agricultural and manufacturing commodities and their prices than were collected on services. Censuses and annual surveys of service industries are a relatively recent development and are much less detailed in their coverage of inputs used. Many of the service industries produce intermediate products in areas with very little direct price coverage, such as computer programming, advertising, and information. The producer price index (PPI), formerly the wholesale price index, the major source of deflators for the GNP by industry series, does not collect service prices (except of air- and water-transport and telephone services). Because of this lack of data, a number of service industries series are deflated by makeshift deflators, and real output is assumed to grow proportionally to some measure of

7

Introduction

input and to lead to no observed productivity growth by assumption. The latter is true for the whole government sector, the contribution of various nonprofit organizations, such as universities, and such difficult-to-measure sectors as banking and business services. The conceptual problem arises because in many service sectors it is not exactly clear what is being transacted, what is the output, and what services correspond to the payments made to their providers. A whole section of this book is devoted to the discussion of how one should view and measure the output of the banking sector. Similarly, when an industrial firm keeps a legal firm on a retainer, what is the corresponding quantity of services? In several service sectors, such as business, health, and legal services, what is transacted is a delivery and exchange of information. Because of its extreme heterogeneity, it is rather difficult to price it efficiently, per bit transmitted, and therefore the resulting pricing structures are often nonlinear and not directly related to what was actually received by the consumer. This difficulty is reflected, for example, in the different pricing structures of various data services, such as Compuserve, Dialog, and Prodigy, and the associate complexity of evaluating their output. Over all this hangs the ubiquitous issue of quality change. The problem is general and pervasive. It affects the measurement of consumer durable purchases and the measurement of output in construction. In some service industries, with good data bases and relatively homogeneous outputs, such as communication or public utilities, the measurement problems are less severe than in some of the commodity sectors. But, in general, because of the underlying heterogeneity of transactions, the difficulty of making comparisons across time and space is even greater. In many service sectors output depends on the interaction with the user and thus is more difficult to standardize. Moreover, for many commodities, even for such rapidly changing ones as personal computers and stereo equipment, one has publicly available “specification” data, which report some of the characteristics for the individual items relevant to the measurement of output and performance. The same detail is not available on the performance characteristics of doctors, lawyers, and stockbrokers. Moreover, the necessary economic-engineering research that would tell us which of the characteristics and training levels are important for their successful performance has not been done. We are thus lacking the scientific base for the desired measurement procedures. It is best to discuss some of these issues in the more concrete context of specific industries and their special measurement problems. I shall turn, therefore, to a brief overview of the papers contained in this volume. The volume starts with three papers on the major sources of official U.S. data in these sectors. In the first paper, Michael F. Mohr describes the recent revision in the GNP-by-industry series and its effect on the measurement of output in the service sectors. The current revision represents a significant ad-

8

Zvi Griliches

vance on past practice, especially in its improved measurement of the intermediate inputs used in these sectors. For example, it reduces the growth in the output of health services during 1979-87 from the previously estimated 4.6 percent per year to 2.8 and reassigns the difference to the medical instruments, pharmaceuticals, and other supplier industries. But the revision is incomplete. Some major and growing subsectors, such as banking and business sectors, are still being extrapolated by input measures, eliminating productivity growth by definition. Some of the improvements come with their own problems: “revenue miles” are a reasonable measure of transportation services, but they leave open the question of quality change in these miles, a question discussed later on by Robert J. Gordon (chap. lo), and the question of the effect of various travel restrictions associated with special fares, a question considered by Paul A. Armknecht and Daniel H. Ginsberg (chap. 3). Similarly, measuring the output of brokerage services by the number of trades, treating $1,000 and $1,000,000 trades as equivalent, leaves something to be desired. Here, one could have probably constructed a reasonable index of commission rates from the Securities and Exchange Commission and other sources. One could also raise questions about the treatment of radio and television broadcasting, which under the current conventions is entirely an intermediate input, contributing only to the output of cereals and razor blades (except for public television, which is included in consumption and treated differentially). Implicitly, output in this industry is measured by the size of the audience. Thus, quality changes that expand the industry’s size are reflected in output, but quality improvements that occur in the competition for audience may show up in higher rates and be “deflated” away. Because advertisers are not interested just in minimizing the cost per person reached but also in maximizing the total size of the audience reached and the effectiveness of the message, the current procedure may be problematic without even raising the issue of what consumers get out of it and how it is related to advertiser costs, if at all. The double-deflation procedure (the subtraction of deflated intermediate purchases form deflated gross output to arrive at a real value-added concept) is itself troublesome, as is also the GNP by industry construction, which is based on a value-added measure of an industry’s output and is motivated by a desire for an unduplicated measure of national output. If one is interested in productivity measurement at the industry level and has some notion of a production function as a framework for it, the subtraction of intermediate inputs from gross output is appropriate only when these inputs are used in fixed proportion to output, when the ratio of their prices to final product prices remains constant, or when changes in their prices have no effect on the relative amounts of capital and labor used in production. Neither is a very likely occurrence. If one looks at the restaurant sector, one sees that the relative price of food purchased for away-from-home consumption to food purchased for home consumption (a proxy here for the cost of intermediate input) has risen by 1.4 percent per year between 1979 and 1986 (Survey of Current Business,

9

Introduction

July 1987). Similarly, in the retail food sector as a whole, the consumer price index (CPI) for food rose by 1.8 percent more per year, between 1979 and 1990, than the comparable PPI for consumer food. For productivity measurement purposes we would be much better off with explicit and separate series on gross output and intermediate inputs in constant prices. The duplication problem can be solved by using appropriate value-added weights, as was pointed out a long time ago by Domar (1961). To implement either this program or the current GNP-by-industry effort correctly requires a detailed set of productivity accounts covering the whole economy. At the minimum, as Mohr notes, we need a consistent and current set of input-output table^.^ But the 1982 input-output table was published only very recently (Survey of Current Business, July 1991), with a lag of almost a decade, and is based on the old standard industrial classification. Thus, it is already obsolete. The problem is actually deeper: it is not just the delay; it is the lack of the right underlying ingredients. To construct GNP by industry by current methods requires information either on profits and depreciation or on intermediate input purchases by industry. But, although most of the output, employment, and wage bill data are collected at the establishment level, profits and depreciation come from company-based IRS records and have to be allocated across industries based on scraps of obsolete information. At the same time, neither the census nor the annual survey of manufacturers collects a complete account of expenditures on all intermediate purchases, especially services. Moreover, the censuses of service industries often do not ask about purchases of intermediate inputs at all. Nor can the BEA in our system tell the other statistical agencies what and how to collect, to impose a consistently designed datacollection framework centered on the need for a coherent and high quality set of national income accounts. In the meantime one muddles through as best one can and the Mohr paper shows us both how much progress can be made even within the current constraints and also how far we have still to go. The BLS program on “Productivity Measures for Selected Industries,” described in the paper by Edwin R. Dean and Kent Kunze, benefits from not having to cover the whole waterfront. It concentrates on measuring productivity growth in those detailed industries where physical measures of output are available or where there is a reasonable price index for the deflation of the gross output data in current prices. But here concentration on physical measures of gross output and the lack of any information on intermediate (and capital) inputs creates problems of its own. Physical output units may also often vary widely in quality. Only electricity and gas provide us with reasonable measures of output. As noted earlier, passenger miles are a dubious unit of measure because they ignore timing and convenience considerations. Similarly problematic is the number-of-transactions approach to the measurement of banking output, as is indicated in the subsequent papers that discuss this 4. See Jorgenson (1990) for an attempt in this direction

10

Zvi Griliches

topic in greater detail. But measuring the output of service establishments by their deflated sales, by their “throughput,” it also questionable, as can be seen from Walter Y. Oi’s discussion of this topic. Looking at a fruit store and measuring its output by the number of oranges sold ignores the effort that may go into their arranging and culling. If the store stays open longer and makes itself more convenient, its measured productivity declines. If demand or supply shifts from radishes and onions to kiwis and strawberries productivity increases. The latter effect could be counteracted by deflation and weighting at the individual product level or by a CPI based on rapidly shifting weights. Neither is, unfortunately, the case in practice. Moreover, many of the CPIbased deflators, especially for electric appliances and electronic equipment but also for hotel services, may be missing quite a bit of the quality change (upgrading) occurring in all of these industries. Thus, although the spirit is willing, the execution is not always as strong as we or the BLS would desire. In his comment on Dean and Kunze’s paper, W. Erwin Diewert calls attention also to the rather strange mixture of weighting schemes used to construct these measures (see, e.g., Dean and Kunze’s fig. 2.4). Nevertheless, these measures do provide us with another very useful window on reality, on what is going on in our economy. Because the BEA figures are mostly published at a much higher aggregation level, and because they are value-added, not output, measures, it is hard to make a direct comparison to the parallel BLS measures. Gordon does this comparison for the productivity measures in transportation. In table 3, I present a few of the possible comparisons for some of the service industries (see also table 2.7 in Dean and Kunze). Given the different conceptual bases (gross output vs. value added and hours vs. persons), they are often rather close. Two noteworthy differences occur in banking and automotive repairs. In both cases they arise from differences in the measurement of output. For banking, the BLS approach, although imperfect, is still superior. In automotive repairs the difference arises, presumably, from the differential treatment of intermediate inputs. AlTable 3

Alternative Estimates of Productivity Growth for Selected Industries, 1979-1989

Sector Air transportation Petroleum pipelines Telephone Banking Hotels Automotive repair

BEA

BLS

0.5

1.9

1.1

0.4

5.4 0.0

5.3 2.3* -1.3 0.2

-1.1 - 2.6

Note: BEA = GNP (value added in 1982 prices) per person engaged in production; BLS Output per hour (From USDL 19-41, table 2). * 1979-88

=

11

Introduction

though conceptually accounting for them is an improvement, the resulting implication of a ten-year decline in labor productivity at over 2.5 percent per year is hard to believe and is worth additional investigation. Similarly, the conclusion of both the BEA and the BLS measures, that the productivity of hotels and motels has been declining, is dubious, though it could be explained by the downward trend in occupancy rates and the increase in various ancillary services, such as concierges. Taken over all, it is good that we have two different glimpses of the same phenomena, though I keep thinking that more could be done in explaining the differences between them. Providing an explicit reconciliation could be very informative. The paper by Paul A. Armknecht and Daniel H. Ginsburg describes the procedures currently used in constructing the CPI and some of its major services prices components. Because the CPI is the major source (together with the PPI) of the deflators used in the construction of “real” GNP, it is very important to the whole productivity-measurement enterprise. Its problems translate directly into output-measurement problems because, by and large, output is measured as deflated sales or as value added. Armknecht and Ginsburg lay out very clearly the major issues facing the CPI: weighting, new goods and services, and quality change. Solutions to these problems depend on the availability of resources and on an agreement on what is to be measured. Both of these have been in short supply historically. The weighting problem is perhaps the most obvious one: 1960-61 base weights were used for 14 years, from 1964 until 1978; 1972-73 weights were used for the next 9 years, until 1987, by which time they were 15 years outof-date. Current data are based on 1982-84 weights. Actually the problem is not as bad as it sounds because as of 1978 the CPI started adjusting some of the internal weights and shifting the product mix to be priced.s Probability sampling procedures introduced in 1978 in principle bring new items into the index within a five-year cycle. But such a two-to-three-year average lag is still too long in a world of rapidly changing products with most of the price declines occurring in the first few years after their introduction. Thus, for example, personal computers did not enter the index until 1987, and many of the new models do not live long enough to be caught in such a sampling cycle. A more general problem arises from the standard “linking” procedure for new goods in both the CPI and PPI: goods are defined too finely, and hence the gains from their appearance are “linked out” from the index. For example, video rentals are a lower-cost alternative to movies but the transition to them does not lower the entertainment price index as a whole. Nor does the appearance of generic drugs show up as a decline in pharmaceutical prices, because the generic version of a branded item is treated as a separate commodity. 5 . Overall weighting is probably less a problem (see Manser and McDonald 1988) for the CPI than for the GNP implicit deflator. The lag in the introduction of new products and the treatment of new outlets (Reinsdorf 1990) may be empirically more important.

12

Zvi Griliches

Armknecht and Ginsburg describe a very interesting attempt to use hedonic regression adjustments in computing the price of air travel, which could also be used to improve the parallel BEA and BLS estimates. They also discuss the very difficult issue of measuring the price of health services and health insurance. Here we come up squarely against the question, What is to be priced? What is the service of a physician-a consultation, a procedure, or a cure? How do we adjust for quality change if what the physician does or suggests is more effective than it used to be? Are insurance rate increases, caused by a rise in morbidity or by the increased use of new procedures, a rise in price or in quantity? Armknecht and Ginsburg discuss the dilemmas in this area and conclude that an increase in the utilization of the health-care system because of the appearance of a new disease should not affect the relevant price indexes even though health-insurance rates may rise. This is right, but it leads to the paradox of an increase in “real” GNP that is likely to be misinterpreted as a rise in the standard of living. Ideally, one would like to distinguish between the price index of living and the cost index of living, of keeping the consumer at some fixed utility level. The cost of living may change because prices have changed, or it may change because the physical and social environment has changed. Thus, one may wish to exclude taxes from the definition of the price index and above normal expenditures on heating or health from the dual definition of the level of living. The use of the concept of a household production function (Becker 1965), extended to include a separate disturbance to the technology of consumption, would break the identity between expenditures in constant prices and the associated indirect utility index. This breakage would lead us toward a redefinition of the GNP concept to allow for capital gains and losses resulting from natural disasters, epidemics, and the depletion or restoration of various natural resources. We are still very far from having the data bases necessary for doing this right, but it is worthwhile to try to work out the conceptual problems of extending the national accounts in these directions.6 In the meantime, however, “simpler” quality-change problems also require attention. Statistical agencies have been quite reluctant to move in the direction of pricing a “cure” or a “disease episode,” partly for conceptual reasons (see, e.g., the old exchange between Gilbert 1962 and Griliches 1962), but mostly because of the difficulty in collecting the relevant data. Here, the increased movement toward payments by diagnostically related grouping (DRG) may help. But ignoring the problem may also result in serious biases. For example, the new laser-based gall-bladder procedure has reduced significantly the total cost of treating such episodes. Because it is a different procedure, it does not show up as a decline in the “real price” of health services. But, assuming that it does not lead to an increase in the total number of oper6. These are not new ideas. See Kuznets (1941) and Nordhaus and Tobin (1972).

13

Introduction

ations, the effect of such a substitution is a decline in the output of the healthservices sector and the resulting reduction in resources used does not show up as a productivity increase. I believe that we can and should do a better job of tracking such changes and incorporating them into our measurement procedures. It should be noted, however, that, despite the implicit complaint above, the CPI is probably the best of all the statistical series produced by the U.S. government, in the extent of its attention to this range of problems and the effort it puts into guarding and improving the quality of the primary data that it collects. The next set of papers discusses the productivity experience of specific industries while at the same time straining against the conventional boundaries of national income accounting and trying to go beyond them, often quite far, in the quest for alternative measurement procedures. Walter Y. Oi reviews the measurement problems in the retail trade sector and the available empirical evidence on them. He emphasizes the importance of inventories and inventory services provided by this sector to consumers with the resultant implication of significant economies of scale (what he calls the economies of massed reserves). He also notes that the shifting boundaries of activity between manufacturers, wholesalers, retailers, and consumers make conventional productivity measurement both difficult and often misleading. Thus, for example, current measures of double-deflated output in the retail food industry underestimate its growth and the associated productivity increases because the CPI links out the price decline that occurred with the introduction and spread of chain stores by treating them as a separate commodity. At the same time, the BLS measure of the productivity of gasoline stations overestimates their productivity growth by excluding the growing use of consumer input in selfservice stations. A related measurement problem affects also some of Oi’s own cross-sectional comparisons: looking at sales per employee or similar measures across different-size stores underestimates the true extent of economies of scale because it does not take into account the lower price levels in the larger stores. It would be useful to have price measurement not only across time, as in the CPI or PPI, but also across space and type of outlet, holding “true” service levels constant. This task is very difficult, similar in magnitude to that undertaken by the United Nations International Comparison Project reported on by Alan Heston and Robert Summers in their paper in this volume.’ But without something along such lines we are unlikely to make real progress on productivity measurement in this area. To escape some of these difficulties Oi suggests that we may have to give up the quest for measuring productivity separately at each transaction level and concentrate instead on looking at the 7. Experimental work along these lines is currently being pursued by the BLS in its Interarea Price Program.

14

Zvi Griliches

“final” productivity of the product delivery process as a whole: cake on a plate or gasoline in the tank. Gordon pushes this idea even further below and suggests miles driven rather than gallons of gasoline in the tank as the final measure. Timothy F. Bresnahan, Paul Milgrom, and Jonathan Paul take up the very difficult question of the output of the stock market and the associated brokerage and finance and information services. Following the earlier lead of Samuelson (1957), they emphasize that much of the activity in the stock market is devoted to anticipating next week’s, next hour’s, and even next minute’s price, resulting in significant wealth transfers and rent dissipation. They show that the contribution of this activity to “real” productivity, either via its use in managerial compensation schemes or via its influence on investment decisions, is likely to be very small because the signal provided by the market is about the average value of the firm and that is only very loosely related to the marginal contribution of a particular managerial action or new investment project. On the other hand, because acquiring new information may result in a transfer of rents, this may induce excessive investments in such information gathering activities (see Hirshleifer 197 1, for a similar argument in a different context). Given the great decline in the cost of communication and computing, it is not surprising that both trading on and employment in the stock markets has expanded greatly. The authors develop a model of such activity that indicates that its growth may be a mixed blessing and that the number of transactions on the market (the measure used by the BEA to move its real output series) may not be a good indicator of its social product. Their point could be expanded to attack also the puzzle of why the contribution of computers to productivity growth is not visible in the usual measures (see Baily and Gordon 1988; and Donald Siege1 and Zvi Griliches, chap. 11, in this volume) and to inquire into the productivity consequences of the recent growth in telemarketing. Related issues of what is the output in the insurance sector and how it is to be measured were also discussed in a paper by A. Hornstein and E. C. Prescott (1991), not included in this volume. Conceptually simpler but empirically still very intractable is output measurement in the banking industry. A separate session with two papers and three discussants was devoted to this topic. Currently, even the nominal value of banking output is in dispute. For obscure reasons, discussed in the comment by Jack E. Triplett, interest received for loans made is not counted and instead an imputation is made that is substantially greater than the interest “foregone” on demand deposits, to reflect the flow of consumer services provided by banks. Over time, output is extrapolated by the BEA by employment growth and by the BLS Industry Productivity Program as a weighted average of checks cleared and loans made, neither of which seems fully to capture what banks are all about. To estimate the output of banks, Dennis J. Fixler and Kimberly D. Zieschang use a translog distance function (essentially a joint production func-

15

Introduction

tion) that relates eight different financial assets, such as different types of loans and deposits, to three “conventional” inputs: labor, physical capital, and materials. The resulting estimates produce an opportunity cost of funds series and are used to construct flows of financial services associated with each of the specific asset types. They imply an output index that rises by 8.8 percent per year during 1984-88, much faster than the parallel estimate by BLS of 3.6 percent per year and the miserly 0.7 percent per year estimated by the BEA. This index may still be an underestimate because it does not take into account important quality changes that have occurred in this industry from the point of view of its consumers, especially the spread of and the improved convenience in the use of automated teller machines (ATMs) and other electronic funds transactions. Another view of the productivity experience of banks is taken by Allen N. Berger and David B. Humphrey who use a cost function with a thick-frontier approach to measure their performance. They report an actual decline in total factor productivity in banking in recent years and attribute it to deregulation and the subsequent dissipation of rents. Because output is measured from the cost side of banks rather than from the utility of the services rendered by consumers, their output measure (or anybody else’s) does not capture the increased supply of services arising from the resulting competition for depositors. Berger and Humphrey are aware of this distinction. They show that some of their decline in output was passed on to consumers in the form of an increase in interests payments on deposits and also note that costs associated with ATMs resulted in increases in services that were enjoyed by depositors but could not be incorporated in their output measure. The Berger and Humphrey finding of widespread inefficiencies is challenged by Frank C. Wykoff in his comment. He interprets these inefficiencies as possibly arising from differences in location and from product differentiation, aspects of which are not fully taken into account in the estimated cost function. On the other hand, the recent experience of the banking sector makes their findings more credible. Jack E. Ti-iplett, in his comment, suggests a hedonic approach to output measurement in this industry, with data on both loan charges and service charges and the “free” services associated with different types of deposit accounts to be used in estimating price indexes for bank services. Such price indexes could then be used to deflate the nominal receipts of banks and produce a more appropriate quantity index of bank services. Education and health services are probably the most difficult sectors for output measurement. Even though we tried, we did not succeed in including a paper on the measurement of health-services output in this conference. But on education we do have the pioneering work of Dale W. Jorgenson and Barbara M. Fraumeni. In a series of papers (see also Jorgenson and Fraumeni 1989 and 1992) they have suggested a new measurement procedure and have implemented it on U.S. data. In essence, their approach defines the output of the educational system in a particular year as the net addition to human capital

16

Zvi Griliches

that occurs as the result of the various student bodies completing an additional school year. The value of this addition is derived from current wage-ageschooling relationships that are projected into the future and then discounted back into the present. The procedure adopted makes a number of controversial assumptions: it accepts differences in the existing wage structure as reflecting primarily differences human capital produced by the educational system and steps over issues of discrimination and selectivity by ability and by socioeconomic status (on the latter, see the earlier discussions in Denison 1964; Griliches 1970; and Willis 1986). It also assumes that leisure time is to be valued at the same wage rate as working time, an assumption that is questioned in Michael Rothschild’s comment on this paper. The use of current wages as fully reflecting the correct expectations about the future could also be questioned. It implies a set of capital gains-and-losses terms in the associated wealth accounts (see Jorgenson and Fraumeni 1989). The resulting estimates are ‘‘gross’’ in the sense that they do not allow yet for the input of student and teachers time, capital, and possibly most importantly, family time (including child rearing) used in “producing” some of this output. The former, excluding the child-rearing component is included in their subsequent paper for the Uppsala Conference (Jorgenson and Fraumeni 1992). In spite of these reservations, their numbers do draw attention to the fact that investment in human capital represents the major investment activity of this and other economiessomething that is often overlooked in the various policy debates. The paper by Swati Mukerjee and Anne Dryden Witte on the day-care industry could have, in principle, benefited from following the Jorgenson and Fraumeni lead. The effective output of that industry is some combination of parental hours “relieved” and the present value of the increase in the human capital of the children as the result of the various training and educational activities pursued there. The data are not available, however, to pursue such a path in this industry. Instead, Mukerjee and Witte use a cost-function framework and child hours (adjusted for age differences) as their primary measure of output and concentrate on developing a “quality of child care” measure based on staff hours per child. They show that this measure of quality affects costs significantly and also that it appears to have declined nationally as childcare institutions were expanding at a fast rate. Although questions can be raised about the exogeneity of staff-pupil ratios in such a cost-function context, their results do suggest that the growth in the output of this industry may be overestimated when the concomitant decline in its quality is ignored. Robert J. Gordon’s paper brings us back to more traditional ground. Reviewing the construction of output data in the transportation sector in some detail, it focuses especially on the railroad and air-transport industries and on the trucking industry. The paper is organized around the development of multifactor productivity indexes for these industries, incorporating his newly developed capital estimates (Gordon 1990). Especially noteworthy are his findings that deregulation did not reduce the quality of air transport significantly

17

Introduction

and that incorporating public infrastructure capital into the productivity accounts does not change the productivity growth story by much. Going beyond the current measurement conventions in including fuel efficiency gains in his measurement of the changing quality of capital equipment, he argues that the ability of an airplane to generate net revenue is the appropriate starting point for measuring the relevant aircraft price indexes. He also computes an estimate of consumer time saved as a result of the improvement in flight speeds. The latter calculation yields a very large number, but one that originates largely in the airplane manufacturing industry rather than in the air-transport industry and was realized primarily in the previous several decades. More recently ( 1978-87), multifactor productivity in the air transportation industry is estimated by Gordon to have grown at only 1.3 percent per year as compared to 5.5 percent in the first postwar decade. Compared to other service industries, however, the productivity performance of the transportation sector has held up reasonably well. The remaining four papers are more heterogeneous. The paper by Donald Siege1 and Zvi Griliches started out from the premise that, if the output of services is not measured correctly, its contribution should show up in the measured productivity of those industries that use these services. Because the census of manufactures reports on the purchases of some services, primarily communication and repair services, the idea was to correlate this information with multifactor productivity growth for the same four-digit SIC-level industries in manufacturing. In trying to implement it they ran into serious difficulties as a result of sampling problems and sample change problems in the underlying annual data. Their paper digresses, therefore, to consider other aspects of these data, including the growing use of foreign inputs and the associated mismeasurement of their prices, and the effect of the increased use of computers in manufacturing. Their main finding is a negative one: the recent recovery of productivity growth in manufacturing cannot be attributed to increases in purchased services, foreign outsourcing, or a decline in the quality of the data. They do find, however, a positive correlation between productivity growth in different four-digit-level industries and the intensity of their investment in computers. Elizabeth Kremp and Jacques Mairesse use the French survey of service industries to examine cross-sectional differences in productivity at the firm level for selected service industries in France. They construct a large and detailed panel-data set for their study and analyze the experience of over 2,300 French service firms. What strikes one there is the extreme heterogeneity in the experiences of these firms. For example, even though legal services firms have both a higher average labor productivity level and a much higher productivity growth rate (1984-87) than personnel supply firms (temporaryemployment agencies), the firm distributions of value added per worker overlap greatly in these two industries in levels and almost completely in rates of

18

Zvi Griliches

growth. They find significant industry as well as strong locational differentials (price-level differences?) in productivity levels but only very little in productivity growth rates. Nor is productivity related to firm size. Not having more detail on the output characteristics of these firms, it is hard to make much progress on the output-measurement issue (in the absence of decent deflators at the detailed industry and firm level). The availability of such survey data does, however, open up the possibility of studying other interesting questions (such as exit and entry behavior) about the functioning of service firms in a modem economy (see Kremp and Mairesse 1992, for a further analysis of these data). The paper by Alan Heston and Robert Summers reports on only one aspect of a much larger enterprise: the United Nations International Comparison Project. It describes the problems that arise in making international comparisons of service prices and the various solutions adopted in this work. It then presents estimates of nominal and “real” shares of services in consumption and GDP in 1980 for 60 countries, where “real” means that the various service flows are valued at a common average set of international prices rather than in varying domestic prices. Their main finding is that the real share of services rises very little with income but that nominal service shares rise primarily because of higher relative service prices in higher-income countries. This finding is consistent with Baumol’s hypothesis that the relative labor intensiveness of services raises their price as income and real wages go up. Combined with the relatively low-price elasticities estimated by Heston and Summers, it will result in an ever-growing nominal share of these industries in consumption and GDP. But the higher nominal shares do not imply a higher real consumption of such services, only a higher expenditure on them. The last paper in this volume touches on a very important topic: the productivity of the public sector. In it Richard Murray summarizes the results of a large-scale attempt in Sweden to measure the output and productivity of various public bodies and enterprises, such as police, weather forecasting, education, and hospitals. Although the measurement problems are horrendous, the Swedish study tried to develop outcome, rather than input or throughput measures. It compared the resulting output measures to total inputs used, including capital, and produced, in effect, a multifactor productivity index for the whole public sector. The findings of the Swedish study, which may surprise some, indicate a continued and pervasive productivity decline in most public-sector activities during the periods surveyed (primarily 1960-80). Murray discusses extensively possible biases that could arise from a neglect of positive qualitative developments but concludes, after examining a variety of indicators, that they are unlikely to have done so enough to overturn the original findings. For example, neither life-expectancy nor morbidity statistics showed much improvement in the period examined to negate the finding of productivity decreases in health services. Similarly, the observed crime and recidivism rates do not conflict with the finding of a productivity decline in

19

Introduction

the justice and police system. A possible interpretation of these findings is that various demand shifts have drawn in more resources into areas that have encountered sharply diminishing returns. In health services, new technological possibilities have expanded the scope of possible medical interventions with only marginal improvements in the overall health status of the population. At the same time, societal-environmental changes have increased criminal activities and reduced the effectiveness of the existing legal and police systems. What is surprising is that the same trends do not show up in the U.S. statistics. The BLS productivity measurement program for selected government services (reported in BLS Bulletin no. 2349 and reviewed in Kendrick 1991) yields an overall 1.4 percent per year improvement in labor productivity in the covered portion of federal government activities for the 1967-88 period. In part the difference arises from the fact that the Swedish estimates are for total factor productivity rather than just labor productivity, but mostly, I believe, because the U.S. numbers are based more on “activity” rather than on “outcome” measures. For example, in the case of the legal system the distinction is clear: the U.S. measures are based primarily on cases handled; the Swedish study measures cases solved. One may also question the veracity of the estimated 1.7 percent per-year improvement in the productivity of federal education and training activities or the 1.2 percent per year improvement (from 1967 through 1988) in the productivity of the U.S. postal service. Such measures need to be improved to take consumer and producer satisfaction more into account. The papers collected in this volume illustrate the great heterogeneity that hides behind the general label “services” and also the difficulty of getting a good handle on what is actually happening there. Contrasting them to the papers included in the Fuchs (1969) volume, one does find significant progress in the official data series and a wider understanding of the problem and difficulties involved in the measurement of any economic activity. Progress is reflected in the work reported here on transportation, education, and other service sectors. But several areas remain as difficult today as they were then. The problem of measurement in the health-care industry (not covered in this volume) are not much closer to solution now. Nor is the debate about the measurement of the output of banks, outlined by Gorman in the Fuchs volume settled yet. Neither d o we have the tools, currently, to resolve the measurement of output issues in retail trade and related service sectors, issues that were already discussed then by Barzel, Schwartzman, and others, and now restated and expanded by Oi in this volume. New issues are raised in this volume by Bresnahan, Milgrom, and Paul, by Oi, by Murray, and by others, but they all founder on the lack of relevant data about the uses of consumer time and household and firm activity outside the conventional market sphere. The exclusion of the household sector from the national income accounts

20

Zvi Griliches

was originally an expedient compromise. The time may have come, however, to move toward its inclusion in the next revision of the accounts. This inclusion will require an extension of the current population survey and/or the consumer expenditure survey toward the collection of much more data on time use and household activity. But without some new data of this sort we will not be able to evaluate productivity trends in many important service industries where the primary effect of technical change has been not in terms of the items themselves, but in what they accomplish when used in the household sector and how they substitute for consumer time and other purchased inputs. The growth of the entertainment industry, previously alluded to, the substitutions arising from movie video rentals, the rise of the fast food and take-out industries, and the introduction of microwave ovens have all reduced significantly the time used in household production and have actually raised the average quality of the final product. Such contributions of services to household productivity, to improvements in health, and to increases in human capital cannot really be measured without new and more extensive data. Although new datacollection efforts inevitably turn out to be incomplete and imperfect, they are still worth pursuing because the alternative of not knowing, of giving up, is a defeat-a defeat not only in terms of not having measured something that we would like to know but also because our understanding of what we do measure is heavily affected by where we draw such boundaries and how firm they stay put over time.8 In the meantime, the actual productivity situation may not be as bad as some of the crude numbers indicate. In some sectors, such as communication, where we have. good data, productivity is growing at a satisfactory rate. In others where our measurement efforts are still in their infancy, we should not overinterpret the numbers. Overall, I am more sanguine about the underlying productivity growth possibilities than some of the other commentators. It is true that baby-sitting may not be a beneficiary of productivity-improving developments, but the development of intercoms and the day-care industry have provided alternative and often more “productive” ways of satisfying some of the same wants. Similarly, taking Baumol’s favorite example, it still takes the same four people the same time to play one of Beethoven’s quartets, but their productivity, in terms of audiences reached has improved greatly, especially when recordings, radio, and television are taken into account. This fact is even true of live performances as the result of larger halls, alternative venues such as stadia, and lower real transport costs. The economies of scale inherent in modern communication media have created the phenomenon of superstars (Rosen 1981). In a real sense, Pavarotti is much more productive today than Caruso was in his own time (and also better paid), This trend may not be true, of course, of all services. In some areas, such as health and the criminal justice system, we may be facing sharply diminishing returns in spite of the many 8. I am indebted to Robert Fogel for some of the ideas in this paragraph

21

Introduction

technological improvements that may have affected them. But unless we improve our measurements in this area, both in terms of the availability of basic statistics and improvements in the conceptual frameworks for their interpretation, we will never know. It is the hope of this volume to have taken a small step in this direction.

References Note: This list contains also a number of relevant references not quoted explicitly in the introduction.

Baily, M. N., and R. J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 2:347-43 1 . Baumol, W. J. 1967. Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis. American Economic Review 57(3): 415-26. Baumol, W. J., S. A. B. Blackman, and E. N. Wolf. 1985. Unbalanced Growth Revisited: Asymptotic Stagnancy and New Evidence. American Economic Review 75(4): 806-17. Becker, G . 1965. A Theory of Allocation of Time. Economic Journal 75(294): 493517. Bureau of Labor Statistics. U.S. Department of Commerce. 1983. Measuring Productivity in State and Local Government. BLS Bulletin no. 2166. Washington, D.C.: Government Printing Office. . 1990. Productiviv Measures for Selected Industries and Government Services. BLS Bulletin no. 2349. Washington, D.C.: Government Printing Office, February. Denison, E. F. 1964. Measuring the Contribution of Education. In The Residual Factor and Economic Growth, 13-55,77402. Paris: OECD. Domar, E. 1961. On the Measurement of Technological Change. Economic Journal 71 :709-29. Fuchs, V. R. 1969a. The Service Economy. New York: Columbia Univ. Press. , ed. 1969b. Production and Productivity in the Service Industries. NBER Studies in Income and Wealth, vol. 34. New York: Columbia Univ. Press. Gilbert, M. 1962. Quality Change and Index Numbers: A Reply. Monthly Labor Review 85, no. 5 (May): 544-45. Gordon, R. J. 1990. The Measurement of Durable Goods Prices. Chicago: Univ. of Chicago Press. Griliches, Z. 1962. Quality Change and Index Numbers: A Critique. Monthly Labor Review 85, no. 5 (May): 532-44. . 1970. Notes on the Roles of Education in Production Functions and Growth Accounting. In Education, Income, and Human Capital, ed. W. L . Hansen, 71114. NBER Studies in Income and Wealth, vol. 35. New York: Columbia Univ. Press. Grubel, H. G., and M. A. Walker. 1989. Service Industry Growth. Vancouver, B.C.: Fraser Institute. Hill, T. P. 1977. On Goods and Services. Review of Income and Wealth 123(4): 31538.

22

Zvi Griliches

Hirshleifer, J. 1971.The Private and Social Value of Information and the Reward to Inventive Activity. American Economic Review 5 l(3): 561-74. Hornstein, A., and E. C. Prescott. 1991. Insurance Contracts as Commodities: A Note. Review of Economic Studies 58:917-28. Inman, R., ed. 1985.Managing the Service Economy. New York: Cambridge Univ. Press. Jorgenson, D. W. 1990. Productivity and Economic Growth. In Fifty Years of Economic Measurement, ed. E. R. Berndt and J. E. Triplett, 19-1 18. NBER Studies in Income and Wealth, vol. 54.Chicago: Univ. of Chicago Press. Jorgenson, D. W., and B. Fraumeni. 1989.The Accumulation of Human and Nonhuman Capital, 1948-84. In The Measurement of Saving, Investment, and Wealth, ed. R. E. Lipsey and H. S. Tice, 227-82. NBER Studies in Income and Wealth, vol. 52. Chicago: Univ. of Chicago Press. . 1992. Investment in Education and U.S. Economic Growth. Scandinavian Journal of Economics. Forthcoming. Kendrick, J. W. 1991. Appraising the U.S. Output and Productivity Estimates for Government: Where Do We Go from Here? Review of Income and Wealth 37(2):

149-58.

Kremp, E., and J. Mairesse. 1992. A Look at Productivity Level in French Service Industries, 1984-1988.Journal of Productivity Analysis. Forthcoming. Kuznets, S. 1941.National Income with Composition, 1919-38. New York: NBER. Manser, M., and R. McDonald. 1988.An Analysis of Substitution Bias in Measuring Inflation, 1959-85.Econometrica 56(4):909-30. Nordhaus, W. D., and J. Tobin. 1972.Is Growth Obsolete? Economic Research: Retrospect and Prospect. Vol. 5 , Economic Growth. 50th Anniversary Colloquium. New York: NBER. Ofer, G. 1973. The Service Sector in Soviet Economic Growth. Cambridge, Mass.: Harvard Univ. Press. Reinsdorf, M. 1990.The Effect of Outlet Price Differentials on the Consumer Price Index. Paper presented at the NBER Conference on Price Measurements and Their Uses. Washington, D.C., March 22-23. Rosen, S. 1981.The Economics of Superstars. American Economic Review 71(5):

848-58.

Samuelson, P. A. 1957. Intertemporal Price Equilibrium: A Prologue to the Theory of Speculation. Weltwirtshaftliches Archiv 181-2 19. Searle, A. D., and C. A. Waite. 1980.Current Efforts to Measure Productivity in the Public Sector: How Adequate for the National Accounts? In New Developments in Productivio Measurement and Analysis, ed. J. W. Kendrick and B. N. Vaccara, 333-56. NBER Studies in Income and Wealth, vol. 44.Chicago: Univ. of Chicago Press. Willis, R. J. 1986. Wage determinants: A Survey and Reinterpretation of Human Capital Earnings Functions. In Handbook of Labor Economics, vol. 1, ed. 0. Ashenfelter and R. Layard, 525-602. Amsterdam: Elsevier-North Holland. Wolfson, M. C. 1991. A System of Health Statistics: Toward a New Conceptual Framework for Integrating Health Data. The Review of Income and Wealth 37(1):

63-80.

1

Recent and Planned Improvements in the Measurement and Deflation of Services Outputs and Inputs in BEA’s Gross Product Originating Estimates Michael F. Mohr

The GNP-by-industry estimates, alternatively known as the gross product originating (GPO) estimates, are a widely used and closely monitored series prepared by the Bureau of Economic Analysis (BEA) as an integral part of the national income and product accounts (NIPAs). Compared to output measures (such as sales, value of shipments, or gross output), GPO measures have two main attributes: (1) they measure the GNP originating from the component industries of the U.S. economy and (2) the sum of industry GPO provides an unduplicated measure of the total output produced by the economy.I During the 1970s and especially during the 1980s, the GPO estimates have become the object of regular and intense interest by policymakers and economists investigating hotly debated, high profile, and closely associated economic issues of national importance. These issues include Productivity growth. Why has the rate of productivity growth of the aggregate U.S. private business and nonfarm economies declined since the mid1960s and especially since 1973? Why has the productivity growth of the services sector of the economy not rebounded since 1979 as robustly as has manufacturing? And is the post- 1979 productivity improvement in manufacturing real or just an artifact of the GPO measures? Michael F. Mohr is Chief of the GNP by Industry Branch of the National Income and Wealth Division at the Bureau of Economic Analysis. Views expressed in this paper are solely those of the author and do not represent official positions of the Bureau of Economic Analysis or the U.S. Department of Commerce. Comments from Robert P. Parker and Zvi Griliches on earlier versions of this paper are gratefully acknowledged. I , Although GPO is a value-added measure. that term is not used here because of possible confusion with census value added. GPO differs from census value added largely because GPO excludes, but census value added includes, services inputs. Students of productivity growth should also note that GPO and gross output are distinctly different measures. Even over expansion intervals, the mean growth rates of industry GPO and gross output are quite different for nearly all industries shown in table I . I of the text.

25

26

Michael F. Mohr

Structural change. Is the manufacturing share of the U.S. private domestic product declining? Is the United States becoming a service economy? And are the effects of structural change a net good or bad for the U.S. economy? Competitiveness. Are U.S. industries strong and healthy enough to compete in a world economy? Is foreign competition destroying the industrial and technical base of the United States?

As a consequence of their critical importance in the foregoing areas, the industry GNP estimates in recent years have been the subject of several studies investigating the possibility that flaws in the source data and estimating methods underlying the industry estimates are producing profoundly incorrect answers to the questions raised in the above issue areas. BEA acknowledged the potential significance of several of these criticisms in the July 1988 Survey of Current Business (SCB), when it announced its intention to reexamine the methodology underlying the industry GNP estimates and, where existing source data permitted, to undertake improvements to the estimates (Bureau of Economic Analysis 1988). The fruits of phase 1 of the GPO improvement program effort are now emerging; improved estimates for 1977-88, published in the January 1991 SCB, mark the first publication of GPO estimates since July 1988 (Bureau of Economic Analysis 1991). This paper focuses on the improved measures of services outputs, inputs, and GPO generated by phase 1 of the GPO improvement program; it also discusses future improvements planned for phases 2 and 3. Section 1.1 defines the services-producing industries included in the GPO estimates, and it demonstrates the growing role of services in the GPO estimates and in the U.S. economy from both output and input perspectives. Section 1.1 summarizes the recent literature criticizing the services measures underlying or produced from the GPO estimates. Section 1.3 outlines not only the methodology BEA uses to generate the current estimates of GNP for services-producing industries but also the measurement problems attendant to those procedures. Section 1.4 develops the major improvements in the constant-dollar measures of services outputs, inputs, and GPO that have been embodied in the recently published phase 1 estimates for 1977-88. Section 1.5 summarizes the overall GPO improvement program, and it outlines important planned future changes in methodology that will be instituted during phases 2 and 3 to further improve the services measures in the GPO accounts. Section 1.6 closes the paper with a discussion of critical source data deficiencies that can be addressed only through expanded data collection by other agencies. 1.1 Services in the GPO Estimates and the Economy 1.1.1

Services-Producing Industries

The industry detail currently contained in the GPO estimates is shown in table 1. I , which also gives the 1972 standard industrial classification (SIC)

27

BEA Measurement of Services Outputs and Inputs

definition of each GPO industry. Annual current- and constant-dollar GPO estimates, at the level of detail shown in table 1.1, have traditionally been published in tables 6.1 and 6.2, respectively, of the July SCB. Included in the existing GPO estimates are 60 private industries, which provide approximately two-digit SIC private-sector detail. Of these 60 private industries, 28 are commodity producing and 32 are services producing. Following convention, the services-producing or service sector of the private economy is defined here to included the detailed industries classified by SIC under transportation; communications; electric, gas, and sanitary services; trade; finance, insurance, and real estate (FIRE); and services in table 1.1. And, the commodity-producing sector is defined to include all the component industries classified by SIC under agriculture, forestry, and fisheries; mining, construction, and manufacturing in table 1.1.z 1.1.2 The Share of Gross National Product in Services-Producing Industries Table 1.1 also demonstrates the dramatic growth that has taken place in the share of total real GNP accounted for by services-producing industries. Between 1960 and 1988, the share of total real GNP originating in servicesproducing industries increased 10.5 percentage points-from 46.2 percent to 56.7 percent. Led by the rapid growth in the telephone and telegraph, wholesale trade, real estate, business services, and health services, most of the relative growth of services-producing industries occurred between 1969 and 1979. By 1988 these five industries accounted for more than 27.6 of the 56.7 percentage point share of GNP traceable to services-producing industries, compared with 19.2 of 46.2 percentage points in 1960. 1.1.3 The Services Share of Intermediate Inputs Consumed Another measure of the importance of services in the economy and in the GPO estimates is the value of services inputs consumed relative to the value of all intermediate inputs consumed by U.S. industries. Based on estimates for 1977 and 1985 derived from the recent methodology improvements, this input perspective on the importance of services is demonstrated in tables 1.2 and 1.3. For example, table 1.3 demonstrates a fact that may surprise many: real services inputs are not only a rapidly growing but also a very large share of the real inputs consumed by every major industrial sector of the private nonfarm economy. Indeed, services constitute well over half of the real cost of intermediates in seven out of ten nonfarm industry divisions in both 1977 and 1985. In addition, services relative share of such costs has also grown rapidly in seven of these industries between 1977 and 1985. Table 1.2 demonstrates that this relative growth has been particularly pronounced in the con2. This commonly used definition of commodity-producing industries incorporates industries such as agricultural services, mining services, and maintenance and repair construction, which might be more appropriately defined as services producing.

28

Michael F. Mohr

Table 1.1

GNP by Industry as a Percentage of Constant-DollarGNP for Select Years

Industry or Sector GNP Domestic industries (GDP) Private industries Commodity-producing industries Agriculture, forestry, & fisheries Farms Agricultural services, forestry, & fisheries Mining Metal Coal Oil & gas extraction Nonmetallic minerals, except fuels Construction Manufacturing Durable goods Lumber & wood products Furniture & fixtures Stone, clay & glass products Primary metal industries Fabricated metal products Machinery, except electrical Electric & electronic equipment Motor vehicles & equipment Other transportation equipment Instruments & related products Miscellaneous manufacturing industries Nondurable goods Food & kindred products Tobacco manufactures Textile mill products Apparel & other textile product Paper & allied products Printing & publishing Chemicals & allied products Petroleum & coal products Rubber & miscellaneous plastic products Leather & leather products Transportation & public utilities: Transportation

1972 SIC

Difference, 1988-1990

1960

1969

1979

1988

100.0 99.3 86.1 39.9

100.0 99.3 85.8 37.7

100.0 98.3 87.0 34.4

100.0 99.3 89.7 33.3

4.1

2.7

2.4

2.3

-1.8

3.7 0.4

2.4 0.3

2.0 0.4

1.8 0.5

-1.9 0.1

5.7 0.2 0.5 4.8 0.2

5.3 0.1 0.4 4.6 0.2

4.5 0.1 0.4 3.8 0.2

3.2 0.1 0.5 2.5 0.2

2.5 -0.1

9.8 20.3 12.2 0.5 0.3 0.8 2.3 1.7 2.0

7.6 22.1 13.8 0.6 0.3 0.8 2.1 2.0 2.4 I .5

5.2 22.3 13.6 0.7 0.3 0.7 1.7 1.8 2.7 1.9

4.4 23.0 14.5 0.6 0.3 0.6 0.9 1.6 4.2 2.2

-5.4 2.7 2.3 0.1 -0.2

37 1 372-79

1.1

1.5

1.5 1.6

1.5 1.1

1.3 1.5

-0.2 -

38

0.4

0.6

0.7

0.8

0.4

39

0.4

0.4

0.4

0.4

-

20 21 22 23

8.2 2.0 0.4 0.4 0.7

8.4 I .8 0.3 0.5 0.6

8.7 I .9 0.3 0.5 0.4

8.6 1.7

0.4 -0.3 -0.3 -0.1

26 27 28 29 30

0.8 1.3 1.1

0.9

0.9

1 .o

0.4

1.4 0.9 0.6

0.9 1.2 1.7 0.8 0.6

1.8 1.1 0.7

0.1 -0.2 0.7 0.1 0.3

31

0.2

0.2

0.1

0.1

0.1

4.4

4.3

3.9

3.7

01-02 07-09 10

11-12 13 14 15-17 24 25 32 33 34 35 36

1.o

1.3

0.1

0.4 0.6 1.1

-

3.6 -6.9

-2.3 -

-1.4

-0.1 2.2 1.2

-0.7

29

BEA Measurement of Services Outputs and Inputs

Table 1.1

(continued)

Industry or Sector

1972SIC

Railroad transportation Local & interurban passenger transit Trucking & warehousing Water transportation Transportation by air Pipelines, except natural gas Transportation services Communication Telephone & telegraph Radio & television broadcasting Electric, gas, & sanitary services Wholesale trade Retail trade Finance, insurance, & real estate Banking Credit agencies other than banks Security & commodity brokers, & services Insurance carriers Insurance agents & brokers, & services Real estate Holding & other investment companies Services Hotels & other lodging places Personal services Business services Auto repair, services & garages Miscellaneous repair services Motion pictures Amusement & recreation services Health services Legal services Educational services Social services & membership organizations Miscellaneous professional services Private households

Difference, 1988 1988-1990

1960

1969

1979

40 41

1.4 0.7

1.2 0.3

0.7 0.2

0.7 0.2

-0.7 -0.5

42 44 45 46

1.3 0.3 0.3 0.1

1.5 0.2 0.6 0.1

1.7 0.3 0.6 0.2

1.6 0.1 0.8 0.1

0.3 -0.2 0.5 0.0

47 481,482,489 483

0.2 1.2 1.0 0.2

0.2 1.5 1.4 0.2

0.2 2.3 2.0 0.3

0.3 2.7 2.4 0.2

0.1 1.5 1.4 0.1

49

2.1

2.4

3.0

3.3

1.2

5.5 9.2 12.4

6.1 8.8 13.0

6.2 9.1 14.4

7.4 9.9 14.5

1.9 0.7 2.1

61

1.5 0.1

1.5 0.1

1.7 0.2

1.6 0.2

0.2 0.1

62

0.3

0.4

0.3

0.8

0.5

63

64

0.9 0.6

0.8 0.6

1.1 0.5

1.0 0.5

0.1 -0.1

65-66 67

8.8 0.1

9.4 0.1

10.5 0.2

10.1 0.3

1.3 0.2

11.4 0.7

11.9 0.7

13.7 0.8

15.3 0.7

3.9 0.0

1.1

0.7 2.5 0.8

0.7 3.7 0.7

-0.4 2.3 0.2

50-5 1 52-59

60

70 72 73 75

0.5

1.0 1.9 0.6

76 78 79

0.3 0.2 0.5

0.2 0.2 0.4

0.3 0.2 0.4

0.3 0.2 0.5

0.0 0.0 0.1

80 81 82 83, 86

2.5 0.9 0.5 0.9

3.0 0.9 0.6 0.9

4.0 0.9 0.6 0.9

4.0 1.0 0.6 0.9

1.5 0.1 0.0 0.0

84, 89

0.9

1.0

1.4

1.7

0.8

88

0.9

0.5

0.2

0.2 -0.7 (continued)

1.4

30

Michael F. Mohr

Table 1.1

(continued) 1972SIC

Industry or Sector Government & government enterprises Federal Government Government enterprises State and local Government Government enterprises Statistical discrepancy* Residualt Rest of the world*

91-97 01-89 9 1-96 01-89

1960

1969

1979

1988

14.4

14.0

11.8

10.5

7.4 6.5 0.8 7.1 6.5 0.6 -0.5 -0.7 0.7

6.6 5.8 0.8 7.4 6.9 0.5 -0.4 -0.1 0.7

4.3 3.5 0.7 7.5 7.0 0.5 0.0 -0.5 1.7

3.7 3.1 0.6 6.7 6.2 0.5 -0.2 -0.7 0.7

Difference, 1988-1990

- 3.9 -3.7 - 3.4

-0.2 -0.4 -0.3 -0.1 0.3 0.0 0.0

Note; Percentages for 1960 and 1969 are based on data published in Nataional Income and Producr Accounts of the UnitedStates, 1929-82: Statistical Tables and are not fully consistent with the 1979 and 1988 percentages, which are based on the revised estimates published in the January 1991 Survey of Current Business. *Current-dollar statistical discrepancy equals GNP measured as the sum of expenditures less charges against GNP-i.e., GNP measured as the sum of costs incurred and profits earned in production. Constant-dollar statistical discrepancy is equal to current-dollar statistical discrepancy divided by the implicit deflator for gross domestic business product. 'Equals GNP in constant dollars measured as the sum of expenditures less the statistical discrepancy in constant dollars and GNP in constant dollars measured as the sum of gross product by industry. *Production abroad that is attributable to factors of production supplied by U.S. residents less the production in the United States attributable to factors of production supplied by foreign residents. Production is measured by the net inflow of labor and property incomes.

Table 1.2

Services Share of Constant-DollarIntermediate Inputs of Nonfarm Industries, 1977 and 1985 (%) Industry*

Difference, 1985 - 1977

1977

1985

53.1 33.5 21.4 53.5 58.3 41.5

56.1 38.7 24.3 58.4 57.8 41.8

3.0 5.3 2.9 4.9 -0.5 0.4

76.1 60.8 94.1

78. I 66.5 91.9

2.0 5.7 - 2.2

61.0

64.8

3.8

~~~

Mining Construction Manufacturing Transportation Communications Electric, gas, & sanitary services Wholesale trade' Retail trade? Finance, insurance, & real estate Services

*Column includes only those industries shown in table 1.5 that are double deflated under phase I of GPO improvement program. +Intermediate input excludes cost of purchases for resale.

31

BEA Measurement of Services Outputs and Inputs Industrial Composition of Constant-DollarIntermediate Inputs Consumed by Manufacturing Industries, 1977 and 1985 (%)

Table 1.3 ~~

Difference, Input Type

1977

1985

All inputs Commodity inputs Agriculture, forestry, & fisheries Mining Construction Durables manufacturing Nondurables manufacturing Services inputs Transportation Railroad Local & interurban passenger Trucking & warehousing Water Air Pipelines, except natural gas Services Communication Telephone & telegraph Radio & television broadcasting Electric, gas, & sanitary services Wholesale trade Retail trade Finance, insurance, & real estate Banking Credit agencies other than banks Security & commodity brokers, services Insurance carriers Insurance agents & brokers, services Real estate Services Hotels & other lodging places Personal Business Auto repair, services, & garages Miscellaneous repair Motion pictures Amusement & recreation Health Legal Educational Social services, membership organizations Miscellaneous professional Other*

100.0 77.2 5.9 14.2 1.1 30.7 25.3 21.4 3.7

100.0

-

74.0 6.5

-3.2 0.6 -3.7 -0.6 -0.7 1.2 2.9 0.0 -0.2

1 .o 0.2 1.6 0.3 0.4 0.3 0.0 0.4 0.4 0.0 3.6 5.5 0.9 1.8 0.6 0.0

10.5 0.5 30.0 26.5 24.3 3.7 0.8

0.1 1.6 0.2 0.6 0.4 0.0

3.2 0.4 0.4 0.0 0.0 0.0 0.2 0.1 0.1

0.5 0.5 0.0 3.8 6.2 0.9 2.3 0.8 0.1 0.1 0.1 0.0 1.3 6.9 0.3 0. I 4.2 0.7 0.5 0.0 0.0 0.0 0.3 0.1 0.1

0.6 1.4

0.7 1.7

0.1

0.3 0.0 0.9 5.3 0.3

0.1

*Scrap and imports of commodities not produced in the United States.

1985- 1977

-0.1 0.0 -0.1 0.2 0.1 0.0 0.1

0.1 0.0 0.2 0.7 0.0 0.5 0.2 0.1 0.0 -0.2 0.0 0.4 1.6 0.0 0.0

1.o

0.3 0.1 0.0 0.0 0.0 0.1 0.0 0.0 0.1 0.3

32

Michael F. Mohr

struction, transportation, and retail trade industries; in each of these, services share of intermediate input cost grew by 5.0 percentage points or more. Although manufacturing shows the lowest relative share-2 1.4 percent in 1977 and 24.3 percent in 1985-this industry nevertheless accounts for the largest value of real expenditures on service inputs. Table 1.3 decomposes the real cost of intermediate inputs consumed by manufacturers; it shows that the share of intermediate input cost accounted for by commodity-producing industries declined by 3.2 percentage points between 1977 and 1985 and that the share from services-producing industries rose by 2.9 percentage points. It also shows that more than half (1.7 out of 2.9 percentage points) of the relative growth in manufacturing's consumption of service inputs occurred in services purchased from wholesale trade and from the business services group.3 By 1985, these two groups of services constituted 10.4 percent of the total real cost of intermediate inputs consumed by manufacturers, compared to 8.7 percent in 1977. Taken together, the data in tables 1.2 and 1.3 are suggestive of the critical contribution of services to important changes taking place in industry production processes and in interindustry relationships. Industries are lowering their cost of production and increasing their international competitiveness by procuring more of the activities-accounting, advertising, legal help, computer software, and temporary help, and so on-that they used to perform in-house from service firms that specialize in such activities.

1.2 Recent Criticisms of BEA Services Measures The recent literature on productivity, structural change, and competitiveness contains several studies that suggest deficiencies in the measures of services outputs and inputs produced from the GPO estimates through July 1988. These studies include contributions by Mohr and Christy (1986), Mishel (1988, 1989), and Baily and Gordon (1988), and Kelly and Wyckoff (1989). Mohr and Christy (M-C) observed that, although detailed industry GNP measures are the ideal output series for analyzing structural change, the GNP estimates for several service-sector industries are likely to have significant errors. The M-C study attributes these potential errors to deficiencies in the underlying methodology-particularly, the use of employment and earnings data to extrapolate GPO benchmarks in several service-sector industries does not allow for productivity change in those industries. Several of the industries singled out by M-C were scrutinized in the Baily 3 . The business services group i5 consistent with the definition in BEA's input-output (1.0) tables. It consists of services included in SIC 73, business services. as well as services defined in SICS 76, 8 I, and 89. The SIC composition of 1-0 industries is shown in Bureau of Economic Analysis (1984).

33

BEA Measurement of Services Outputs and Inputs

and Gordon (B-G) study for evidence of large measurement errors that may help to explain the yet-unexplained slowdown in nonmanufacturing U.S. productivity growth since 1973. The B-G study examines both the current-dollar data and the price deflators used to construct the GPO estimates in FIRE, retail trade, and transportation industries. In each case, B-G finds evidence of “large potential errors” in the measurement of real GPO, most of which result from the failure to measure properly real output growth and to account for quality change. Their analysis suggests that improved GPO estimates could be achieved by one or more of the following: (1) use of either new or better physical output quantity indexes to extrapolate base-year output estimates in transportation, finance, and insurance; and ( 2 ) the development of hedonic price indexes to deflate the output of the insurance, real estate, and airtransportation industries .4 Mishel examined in detail the methodology underlying the entire spectrum of prerevision GPO estimates and highlighted several major measurement problems. These problems included the omission of import prices from the input price deflators, the use of outdated and, in some cases, inappropriate relationships to distribute company-based profits and depreciation allowances to establishment-based industries in measuring current-dollar industry GNP, and errors in the measurement of the prices and value of service inputs. He conjectured that the combination of these problems in the old GPO series helped to mask a significant erosion in manufacturing’s share of GNP and in its productivity growth since 1979 and thereby caused a complementary understatement of output and productivity growth in the service sector. Finally, Kelly and Wyckoff (K-W) noted that reliable estimates of GNP by industry are important for assessing interindustry rates of productivity growth and innovation; for understanding complex interindustry relationships, and for monitoring important changes in these relationship^.^ They observed, however, that, although the input-output (1-0) tables provide the basic tool for achieving such estimates, the lack of up-to-date information on services used as inputs is a major impediment to improving the quality of the industry GNP estimates. Specifically, K-W examined the methodology underlying BEA’s annual 1-0 tables, which are used in the revised GPO estimates to compute the distribution of intermediate inputs consumed by U.S. industries. The KW study concluded that, because they are too sparse in services detail and are based on out-of-date input distributions from the 1977 benchmark 1-0 table, the annual 1-0 tables (and GPO estimates) do not adequately capture the rapidly growing importance of services inputs relative to goods inputs. 4. Nevertheless, B-G conclude that the net result of all their recommended measurement improvements would be but a small improvement in aggregate nonfarm productivity growth. 5. Although industry GNP is still widely used in labor productivity studies, the clear preference in the contemporary productivity literature is for total factor productivity studies employing industry gross output measures.

34

Michael F. Mohr

1.3 Estimation of Current-Dollar GPO: Past and Present Practice In principle, equivalent measures of current-dollar (CU$) industry GNP can be calculated from either of two methods: Method (1)

CU$ GNP

=

CU$ gross output

- CU$ intermediate inputs,

or Method ( 2 )

CU$ GNP = Sum of CU$ payments to labor and capital CU$ nonfactor charges.

+

The value of intermediate inputs shown in method (1) includes the cost of materials and services either purchased from other domestic industries or imported. The payments to labor in method (2) include not only wages and salaries but also supplements; payments to capital include profits, rent, and net interest; and nonfactor charges include depreciation, business transfer payments, indirect business taxes, and subsidies. Method 1 corresponds to the procedure that would be used to generate industry value added from a consistent set of production accounts or 1-0 tables; this is the procedure prescribed in the United Nations system of national accounts (SNA) literature.6 Presently, this method is not used by BEA because sufficiently detailed and SIC consistent annual production accounts are not yet available, as a result of source data limitations. As noted below, however, BEA intends to develop such accounts during phases 2 and 3 of the GPO improvement program. Consequently, method 2 is the procedure used both historically and presently by BEA to generate current-dollar GNP for all industries except farms and nonfarm housing services. Under this method, the components of industry GNP correspond exactly to the components of the income side of the NIPAs or charges against GNP (CAGNP).’ As such, the sum of industry GNP is identical to CAGNP, and, like CAGNP, total industry GNP plus statistical discrepancy is equal to GNP, which is measured from the expenditures side of the accounts. However, the source data actually used to allocate the components of CAGNP are in some cases poorly suited to obtaining consistent and precise SIC establishment-industry GNP estimates. Table 1.4 summarizes both the components and the major source data that are presently used to construct the current-dollar GPO estimates. In deriving industry GNP, BEA presently seeks to distribute each component of aggregate CAGNP on an establishment basis, using industries defined according to the 1972 SIC. As previously noted, the methodology used to effect these distributions has several problems. Included among them are the following. 6. See United Nations (1968). 7. See, e.g., Survey ofcurrent Business. July 1988. 36.

35

BEA Measurement of Services Outputs and Inputs

( 1 ) No one establishment-based data source covers either all the private industries included in the GPO or all the 14 factor and nonfactor components of the current-dollar GNP for a single industry; (2) The before-tax corporate profits, capital consumption allowance, nonfarm proprietor’s income, net interest, and pensions (in other labor income) components of CAGNP are derived from company-based rather than establishment-based industry data; and (3) establishment-industry distributions of the components of industry current-dollar GNP can be inconsistent (e.g., wages and salaries are based primarily on classifications assigned by The Bureau of Labor Statistics [BLS]; profits are based on classifications assigned by the Census Bureau; and nonfarm proprietor’s income are based on classifications assigned by IRS) . The accuracy of BEA’s estimated establishment-industry distributions for corporate profits before tax and for capital consumption allowance (depreciation) is the most questionable of all the CAGNP component^.^ Here, the primary source data are tabulations of corporate tax-return information. The IRS classifies a corporation into an SIC industry according to the industry that accounts for the largest percentage of its sales. Many companies, however, either are highly diversified or draw a high percentage of their profits from an industry that lies outside their principal industry. Depreciation and profits for IRS industries covered in the economic censuses are reallocated by use of a 1972 employment matrix that provides the Census Bureau’s establishmentindustry distribution for employment for corporations classified by IRS into specific SIC industries.1° Use of the employment matrix for this purpose has 8. Evidence of the difference between BLS and census establishment classifications is found in Office of Management and Budget (1990). 9. For the corporate net interest component of CAGNP, BEA makes no attempt to redistribute the IRS value, because no adequate basis exists for converting from a company-industry to an establishment-industry distribution. For the noncorporate CAGNP components, BEA assumes that the IRS data are already distributed on an establishment-industry basis, because noncorporate firms generally operate in only one business. 10. Algebraically, the employment-matrix model can be described as follows: (1)

C, = A , ? X ,

where C, = (n x 1) vector whose elements c,,represent company-based industry i profits or depreciation from tax-return data for year r; A,, = (n X n) matrix whose elements a,,represent the number of employees of company-based industry i who worked in establishment-based industry j during 1972; X, = (n x 1) unknown vector whose elements x,, represent either the profits or depreciation per employee in establishment-industryj. Model (1) is solved simultaneously by matrix inversion to yield: (2) X, = A,’ C, The solution vector X, thus represents the profiUdepreciation rates per employee that must exist in census establishment-based industries in order to redistribute current-year company-based industry profitsldepreciation in a manner that is consistent with the corresponding 1972 distribution of

Table 1.4

Major Sources for Current-Dollar GPO by Industry Industrial Distribution

Component of Charges against GNP Compensation of Employees: Wages and salaries

Employer contributions for social insurance Other labor income

Proprietors’ income with IVA: Farm Nonfarm: Proprietors’ income IVA Rental income of persons

Major Source Data

Distribution Available in Source Data

BLS tabulations of wages and salaries of employees covered by state unemployment insurance and Office of Personnel Management data on wages of federal government employees Federal budget data

Establishment

Trade association data and IRS tabulations of business tax returns

None

Department of Agriculture net income

Establishment

IRS tabulations of business tax returns BLS prices and Census Bureau inventory data Census Bureau American housing survey, BLS consumer expenditures survey, & IRS tabulations of business & individual tax returns

Company

None

Establishment Establishment

Data or Assumption Used if Establishment-Industry Distribution Is Not Available in Source Data

Social Security Administration and BLS tabulations Census Bureau and IRS tabulations

Assumed to be equivalent to an establishment-industry distribution

Corporate profits with IVA: Corporate profits before taxes

IVA Net interest: Corporate Noncorporate Business transfer payments Indirect business tax & nontax liability

Subsidies Current surplus of government enterprises Capital consumption allowances: Corporate Noncorporate

.

Census Bureau & Department of Energy data, relating establishment-industry & company-industry data

IRS tabulations of business tax returns

Company

BLS prices & Census Bureau inventory data

Establishment

IRS tabulations for business tax returns

Company

None

*

Company

*

None

Assumed to be equivalent to an establishment-industry distribution Industry-specific payments are estimated using IRS, FBI, ABA, & BAA data Industry-specific payments of nonproperty taxes are estimated using Treasury, Census Bureau, IRS, & state data; property taxes are based on BEA capital stock distributions

Federal budget data & Census Bureau data on state & local governments

* * IRS tabulations of business tax returns

*

None

Establishment

Company Company

Same as corporate profits before tax Assumed to be equivalent to an establishment-industry basis

Note: BLS = Bureau of Labor Statistics; IRS = Internal Revenue Service; FBI = Federal Bureau of Investigation; ABA = American Bankers Association; BAA = Best’s Aggregates and Averages; IVA = Inventory valuation adjustment.

*Same source as preceding line.

38

Michael F. Mohr

several major weaknesses; these affect the accuracy of the current-dollar GNP estimates for both services- and commodity-producing industries. The first weakness reflects the fact that the use of the employment matrix forces all the reallocations to take place between census-covered companies and establishments. As such, there is no employment matrix reallocation of profits or depreciation for agriculture, forestry, and fisheries; transportation; communication; electric, gas, and sanitary services; FIRE; and service industries numbered in the SIC ~ O S ,except for legal services. Thus, the only services-producing industries covered by the employment matrix are those numbered in the SIC 70s and SIC 8 1 . The second weakness is that, at best, an updated employment matrix is only available every fifth year prior to 1972, and it has not been updated since 1972." Application of the matrix for years since 1972 assumes no change in company-industry structure as a result of mergers, divestitures, or acquisitions that cross industry lines. There has been a considerable number of such transactions since 1972, and they often have the effect of changing the company classification as well as the underlying establishment distribution. Use of a prior year's employment matrix under such conditions can result in incorrect reallocations. The third weakness of the employment matrix is that it often misallocates profits and depreciation associated with assets leased through subsidiaries whose establishment-industry classification is different from that for the parent company. For example, many large manufacturers run leasing operations through their finance subsidiaries. These subsidiaries are frequently consolidated with the parent's tax return, which is classified in manufacturing. Because the employment matrix excludes financial activities, the profits and depreciation on these leased assets are allocated to manufacturing rather than to credit agencies. Despite these weaknesses, BEA has historically used the employment matrix for several reasons. First, in most industries, diversification is not a problem. Second, profits are typically a small part of industry GNP; therefore even large errors in profit distribution cause relatively small errors in industry GNP estimates. Third, there are offsetting errors in the allocations of profits and depreciation derived from the employment matrix. Fourth, BEA had hoped to improve its estimates of industry profits and depreciation by using Federal Trade Commission line-of-business data or Securities and Exchange Commission business-segment data, but neither alternative materialized as a viable substitute. Finally, BEA has planned to improve the employment

employment. Finally, given the solution vector X,, estimates of establishment-industry profits/ depreciation in the current year obtained as (3) C,, = 2, U ~ , ~ X (j , , = 1 , . . . , 17). 1 1 . Beginning with 1958 and ending with 1972, new employment matrices were developed for every quinquennial census year-1958, 1963, 1967, and 1972.

39

BEA Measurement of Services Outputs and Inputs

matrix by updating it annually and expanding it to cover all industries, but the Census Bureau has so far been unable to fund this program. Thus, for the lack of a better alternative at this juncture, the employment matrix method 2 continues to underlie the current-dollar industry GNP estimates derived from phase 1 of the GPO improvement program. However, it is anticipated that its use will terminate with the development during phase 3 of direct estimates of industry intermediate input consumption of sufficient quality to permit current-collar industry GNP to be estimated by method 1.

1.4 Estimation of Constant-Dollar GNP for Services Producing Industries: Past Practices and Recent Improvements 1.4.1 Methods of Estimation

Historically, constant-dollar industry GNP estimates were estimated using three different variants of the double-deflation procedure and two non-doubledeflation techniques: extrapolation and direct deflation. A description of these five estimating techniques follows. Double Deflation Method I. The constant-dollar (CO$) analogue to method 1 for computing CU$ GNP; CO$ GNP is computed as

CO$ GNP

=

CO$ gross output - CO$ intermediate input.

This variant is used only for industries where there exist direct and consistent gross output and intermediate input data that provide complete coverage of the industry. Method 2. The constant-dollar analogue to method 2 for computing currentdollar GNP; constant-collar GNP is computed in two steps as (1)

(2)

CO$ intermediate input = CO$ GNP

=

(CU$ gross output - CU$ GNP) intermediate input deflator ’

CO$ gross output - CO$ intermediate input.

This variant is used only when there exists complete and consistent industry gross output and GNP data. Method 3 . Indirect double deflation procedure; an industry GNP deflator is derived by using method 1 double deflation on industry gross output and intermediate input data that are consistent, but not compatible, with method 2 current-dollar GNP because they cover only part of the industry. This derived deflator is then used to deflate the current-dollar industry GNP derived from method 2. IZ 12. For example, method 3 was formerly used to obtain the real GNP of the electric, gas, and sanitary services industry because the previous measure of gross output and purchased input covered only the electric and gas components of the industry. Using method 1 current- and constant-

40

Michael F. Mohr

Nondouhle Deflation

Extrapolation. Constant-collar industry GNP is derived by extrapolating the base-year value of industry GNP by an indicator series, either CO$ gross output, the number of employees, or hours worked. Direct deflation. Constant-dollar industry GNP is derived by directly deflating current-dollar GNP; the index used for deflation is either gross output prices or earnings. 1.4.2 Past Practice The preferred procedure for obtaining industry constant-dollar GNP is the double-deflation procedure because it measures GNP in the same way it is defined and because, given the appropriate data, double deflation allows for changes over time in the relationship between gross output and intermediate inputs.') In the international and the United Nations SNA literature, double deflation is defined as the method 1 variant presented above.I4 Although many users of the GNP estimates assumed that method 1 double deflation has been historically employed for all private industries, table 1.5 shows that only two industries were and continue to be so estimated-farms and the nonfarm housing services component of real estate. Beyond these two industries, double deflation in one of the two other variants above had been used to obtain real GNP only for construction, manufacturing industries (expect petroleum and coal products), electric, gas, and sanitary services, and railroad transportation. Real GNP for manufacturing industries (except petroleum) was derived by method 2; the real GNP estimates for the three remaining industries were derived by using method 3. In brief, table 1.5 shows that, under past practices, real GNP for only three of 33 services-producing industries (including the housing services component of real estate) was derived by some form of double deflation. For the 30 remaining services-producing industries, real GNP for 18 industries was estimated by the extrapolation method; the direct-deflation method was used for 12 industries. Table 1.5 also indicates that the extrapolator used in 9 of the 18 extrapolated industries was based either on employment or on hours, and that the deflator used in 5 of the 12 directly deflated industries was based in whole or in part on earnings data. Put differently, before the phase 1 revision, the real GNP of servicesproducing industries representing 15.4 percent of private GNP and 23.5 percent of service-sector GNP in 1987 was based on a methodology that assumed away all or part of productivity change; the real GNP for services-producing dollar GNP estimates, these data were used to derive an implicit GNP deflator for electric and gas utilities, which was then used to deflate BEA's CAGNP-based estimate of current-dollar GNP for the entire electric,gas, and sanitary services industry. 13. See, however, n. 26. 14. See, e.g., United Nations (1979) and La1 (1990).

41

BEA Measurement of Services Outputs and Inputs

Table 1.5

Previously Published and Phase 1 Methods for Estimating ConstantDollar G P O for Services-Producing Industries Method

Industry Transportation: Railroad Local & interurban passenger transit Trucking & warehousing Water Air Pipelines, except natural gas Services Communication: Telephone & telegraph Radio & television broadcasting Electric, gas, & sanitary services Wholesale trade Retail trade Finance, insurance, & real estate: Banking Credit agencies other than banks Security & commodity brokers, & services Insurance caniers Insurance agents & brokers, & services Real estateX Nonfarm housing services Other real estate Holding & other investment companies Services: Hotels & other lodging places Personal Business Auto repair, services, & garages Miscellaneous repair Motion pictures Amusement & recreation Health Legal Educational Social services & membership organizations Miscellaneous professional Private households

Previously Published

Phase I

Double deflation (M3)' Extrapolation (0)s Direct deflation (P)' Extrapolation (0) Extrapolation (0) Extrapolation (0) Extrapolation (L)

Double deflation* Double deflation Double deflation Double deflation Double deflation Double deflation

Direct deflation (P,W) Direct deflation (W) Double deflation (M3) Direct deflation (P) Extrapolation (0)

Double deflation Double deflation Double deflation Double deflation Double deflation

Extrapolation (L) Extrapolation (L) Extrapolation (L) Extrapolation (0) Extrapolation (0)

*

*

*

Double deflation Double deflation Double deflation

Double deflation (Ml) Direct deflation (P) Extrapolation (L)

* * *

Extrapolation (0) Direct deflation (P) Extrapolation (L) Direct deflation (P) Extrapolation (L) Direct deflation (P,W) Direct deflation (P) Extrapolation (0) Extrapolation (0) Extrapolation (L) Direct deflation (W)

Double deflation Double deflation

Direct deflation (W) Direct deflation (W)

*

Double deflation Double deflation

*

Double deflation Double deflation Double deflation Double deflation

* * *

*Same method as used for previously published estimates. tIn the previously published estimates, two variants-M1 or M3-of double deflation were used to estimate GNP for services-producing industries; the variant used for a given industry is indicated by showing MI or M3 in parentheses. For a description of the double-deflation variants see the text, section 1.4. ?In the phase 1 estimates, (M2) double deflation is used for all industries except the nonfarm housing services component of real estate, which continues to be derived by (Ml) double deflation. (continued)

42

Michael F. Mohr

Table 1.5

(continued)

$Industries using labor input extrapolation are indicated by (L); industries using gross output extrapolation are indicated by (0). "Industriesusing direct deflation by earnings are indicated by (W); industries using direct deflation by gross output prices are indicated by (P);industries using direct deflation by both earnings and gross output prices are indicated by (P,W). #The real estate industry is listed in two parts because the estimates for the two parts are prepared using different methods.

industries representing 35.3 percent of private GNP and 54.1 percent of service sector GNP was based on a methodology that assumed fixed proportions either between real values of GNP, gross output, and intermediate input or between the prices of these respective measures; and the real GNP for services-producing industries representing 12.7 percent of private GNP and 18.5 percent of service sector GNP was based on indirect, or method 3, double deflation. 1.4.3 Recent Improvements The source of the limitations in the previous real GNP series can be traced largely to the lack of available source data, although for some industries newly available sources were not introduced into the estimating method.I5 The real GNP estimates derived from phase 1 of the GPO improvement program incorporate comprehensive improvements in methodology and source data. The phase 1 improvements can be subsumed into the three following categories: (1) double deflation; (2) gross output; and (3) intermediate input prices. Double Deflation

Double deflation is the core of the improvements incorporated into the GPO estimates during phase 1 of the improvement program. The extension of this procedure resulted in significantly improved measures not only of real gross output and GNP of services-producing industries but also of the services inputs consumed. Under past practice, the real GNP estimates for only two services-producing industries-railroads and electric, gas, and sanitary services-were obtained using some variant of the double-deflation procedure. Moreover, as noted above, nearly 80 percent of the real GNP from servicesproducing industries was based on a methodology that either assumed no change in labor productivity or assumed that no substitution occurred between value-added inputs (labor and capital) and intermediate inputs. And, the remaining 20 percent was derived by an indirect double-deflation procedure based on incomplete data. In contrast, under phase 1 the real GNP estimates for most services-producing industries-representing about 80 percent of the GNP produced by the service sector-are now derived by double deflation. 15. Source data problems in the early constant-dollar GPO estimates are discussed in Marimont (1969).

43

BEA Measurement of Services Outputs and Inputs

Out of the 33 services-producing industries shown in table 1.5, the GNP estimates for all but 10 industries and the nonfarm housing services component of the real estate industry are now obtained by uniformly employing method 2 double deflation-the same procedure is presently used for all commodity-producing industries, except farms. The real GNP for nonfarm housing services continues to be derived by method 1 double deflation. Four of the ten non-double-deflated industries are in FIRE: these four include banking, credit agencies other than banks, holding and other investment companies, and real estate (except nonfarm housing services). The remaining six omitted industries are transportation services, business services, motion pictures, social services and membership organizations, miscellaneous professional organizations, and private households. l 6 Gross Output This expansion in the double deflation of service-producing industries was made possible by the development of gross output estimates for each doubledeflated services-producing industry. These estimates were constructed by developing underlying 1977 and 1982 benchmark and annual extrapolator real gross output estimates at various levels of 1-0 industry and product detail. Table 1.6 describes this detail as well as the methods and source data used to construct the current- and constant-dollar values. Three items of particular interest are highlighted in table 1.6: (1) the high degree of product detail used to obtain the current and constant-dollar gross output for most industries; (2) the extensive use of output quantity indexes to derived the real gross output estimates for all transportation industries, gas and electric utilities, and security brokers and services; and (3) the use of a quality-adjusted cost index to obtained the real gross output of radio and television broadcasting. To deflate service outputs, approximately 100 true output deflators were either constructed from quantity extrapolation or selected from components of the consumer price index (CPI), the producer price index (PPI), and implicit price deflators prepared as part of the expenditure estimates of GNP. In all, approximately 120 current- and constant-dollar component series were developed and used to construct the gross output estimates for the 23 services16. Double deflation is not appropriate for private households because GNP for this industry is defined as employee compensation. 17. The cost index used to deflate the current-dollar gross output of radio and television broadcasting is an estimate of the cost to advertisers to reach 1,000 of viewing or listening audience. as opposed to the cost per unit of air time. The former better represents the extent that two media are providing more effective access by advertisers to targeted markets. It should be noted. however, that advertising revenues are not a direct measure of the gross output of programs produced by the radio and television industries, nor is the deflator discussed above a direct measure of the quality of programming from a consumer’s point of view. It should also be noted that the advertisingrevenues approach has the effect of making the entire output of the industry an intermediate, as compared to a consumer, good.

Table 1.6

Principal Source Data and Estimating Methods Used in Preparing Phase I Estimates of Gross Output for Double-Deflated Industries

Industry

Transportation: Railroad

Local & interurban passenger transit: Taxicabs Intercity buses School buses Local transit Trucking & warehousing

Water

Air: Domestic & international passenger

Domestic & international mail, freight & express

Current Dollars

Constant Dollars

Extrapolator or Interpolator of Benchmark Values?

Price Index for Deflation or Quantity Extrapolator of Base-Year Value

Total operating revenues for class I freight & AMTRAK passenger revenues (1977, 1982)

Composite index of IPD for class I freight, from revenue ton miles from AAR, and of IPD for AMTRAK, from passenger miles from NRPC

PCE (1977) Operating revenues from ABA (1977, 1982) Wages & salaries from BLS (1977, 1982) Operating revenues of private local transit systems from APTA (1977) For 1977-83, operating revenues for class I motor carriers of property from ICC; for 1984-88, Census Bureau annual survey (1977, 1982) Receipts from IRS tabulation of corporate tax returns (1982)

CPI for taxi fares Passenger miles from ABA Employment from BLS Passenger trips from APTA

Operating revenues of scheduled air carriers from DOT and the Federal Express from DOT and public sources (1977, 1982)

Separate revenue passenger miles for domestic and for international travel from DOT

*

Ton miles from DOT

Composite index of ton miles for deep-sea foreign transportation from BEA, ton miles for other water transportation from DOD, & tons for marine cargo handling from DOD

Separate ton miles for domestic and for international mail; separate ton miles for domestic and international freight and express.

Other Pipelines, except natural gas Communications: Radio & television broadcasting Telephone & telegraph

Electric, gas, & sanitary services: Electric utilities

Gas utilities

Sanitary services Wholesale trade Merchant wholesalers gross margins$

Manufacturers’ sales branches & sales offices (MSB&O) gross margins$

* Oil pipeline operating revenues from OGJ (1977, 1982) Advertising expenditures from M-E (1977, 1982) Revenues from FCC (1977, 1982)

Private class A and B revenues (adjusted for imports and cost of resales), from DOE and BEA; REA cooperatives revenues (adjusted for cost of resales), from USDA (1977, 1982) Revenues of gas pipelines (adjusted for imports) & of gas utilities (adjusted for state & local government utilities) from BEA & AGA (1977, 1982) Receipts from IRS tabulations of business tax returns (1977, 1982)

Gross margin rate times sales: For 1977-82, margin rate from Census Bureau quinquennial censuses & sales from Census Bureau annual survey; for 1983-88, both from annual survey (1977, 1982) Estimated operating expense rate times estimated MSB&O sales: estimated operating expense (excluding expense for equipment rental by wholesalers of commercial machines & equipment) derived by extropolating MSB&O operating expenses from Census Bureau quinquennial cen-

Composite index of IPDs for passenger, freight, and mail. Ton miles from AOP

Composite deflator based on cost per 1,000 of audience from M-E Composite deflator based on PPIs for local telephone service, toll telephone services, and directory advertising Kilowatt hours for investor owned and cooperatives from EEI

For gas pipelines, BTUs of gas for resale from AGA; for gas utilities, BTUs of gas utility sales to final customers from AGA CPI for water & sewage maintenance

1982 gross margin rate-weighted sales deflated by kind-of-business deflators derived from PPls

1982 operating expense rate-weighted sales deflated by manufacturing shipments deflators (at 3-digit trade level) derived from PPIs

(continued)

Table 1.6

(continued)

Industry

Agents &brokers (A&B) gross margin$

Computer & office equipment rentals

Excise taxes

Current Dollars

Constant Dollars

Extrapolator or Interpolator of Benchmark Valuest

Price Index for Deflation or Quantity Extrapolator of Base-Year Value

suses with estimated sales. Estimated sales derived by extrapolating MSB&O sales from Census Bureau quinquennial censuses with 4-digit manufacturing shipments from Census Bureau annual surveys, allocated to 3-digit MSB&Os using distribution by class of customer data from 1977 Census of Manufacturers (1977, 1982) Estimated earnings rate times estimated A&B sales: estimated earnings derived by extrapolating A&B earnings (commissions plus margins) from Census Bureau 1982 quinquennial census with estimated gross margin for merchant wholesalers. Estimated sales derived by extrapolating 3digit A&B sales from 1982 quinquennial census by corresponding 3-digit sales of merchant wholesalers from Census Bureau surveys (1977, 1982) Log linear interpolation between 1977, 1982, & 1987 computer & office equipment rentals earned by wholesalers of commercial machines and equipment from Census Bureau quinquennial censuses, & forward extrapolation at 1982-87 rate8 Excise taxes paid by wholesalers of petroleum, of alcoholic beverages, & of tobacco & tobacco products (SICS 517, 518, & 519) from BEA (1977, 1982).

1982 gross earning rate-weighted sales defalted by kind-of-business deflators derived from PPI

IPD based on ratio of historical to constant-dollar gross stock of office computer & accounting equipment

1982 excise tax rate times deflated sales for SICS 517, 518, and 519; sales deflated by kind-ofbusiness deflators derived from PPI

Sales taxes

Import duties Retail trade Eating and drinking places

Other retail: Gross margin$

Sales tax

Finance, insurance, & real estate: Security & commodity brokers & services: Security brokers & services: Commissions Mutual funds Undenvriting/selling new securities Trading & investment gains, excluding interest, & other revenues, excluding repro interest

Aggregate sales tax rate times aggregate sales (including excise taxes) of merchant wholesalers. For 1977-84, tax rates & sales from Census Bureau surveys; for 1985-88 sales from Census Bureau survey & tax rate from Census Bureau 1987 quinquennial census (1977, 1982) Import duties from BEA

1982 aggregate sales tax rate times sum of deflated sales and excise taxes paid by merchant wholesalers; sales deflated by kind-of-business deflators derived from PPIs

IPD for all merchandise imports from BEA

Sum of sales of eating & drinking places from Census Bureau annual survey & of sales taxes on food & on drink from BEA (1977, 1982)

For sales of eating and drinking places, IPD com-

Gross margin rate times sales at %-digit kind-ofbusiness detail, both from Census Bureau survey (1977, 1982) Sum of %-digit kind-of-business sales taxes from BEA (1977, 1982)

1982 gross margin rate (on sales with sales taxes) weighted sales deflated by kind-of-business deflators derived from CPIs 1982 sales tax rate (on sales with sales taxes) times sales deflated by kind-of-business deflators derived from CPIs

Securities commissions from SEC (1977, 1982)

Numbers of public securities orders from SEC & BEA IPD for securities commissions

Revenue from sale of investment company securities from SEC (1977, 1982) Profits (loss) from undenvriting/selling from SEC (1977, 1982) Gain (loss) on trading & investment accounts from SEC less BEA estimate of interest earnings on trading accounts plus other revenues less interest

posed of CPIs for meals and drinks away from home. For sales tax on meals, 1982 sales tax rate (on sales with sales taxes) for eating places times deflated sales of meals. For sales tax on drinks, 1982 sales tax rate (on sales with sales tax) for drinking places & deflated sales of drink.

New securities registrations for each sale from SEC IPD for GNP (continued)

Table 1.6

(continued)

Industry

Revenue of self-regulatory organization Commodity brokers Insurance carriers

Insurance brokers & agents

Real estate-nonfarm housing services Services: Hotels, rooming houses, camps, & others Personal: Laundry, cleaning & garment services Shoe repair shops, shoe-shine parlors

Current Dollars

Constant Dollars

Extrapolator or Interpolator of Benchmark Valuest

Price Index for Deflation or Quantity Extrapolator of Base-Year Value

earned on margin account from SEC & less BEA estimate of interest from repos Revenues earned by exchanges from SEC (1977, 1982) Residual estimate (1977, 1982)

*

Sum of life insurance company net premiums for health insurance from ACLI, PCE expense of handling life insurance, & nonlife insurance company net premiums (adjusted for losses) for auto, accident, and health, property, & workers' compensation from A. M. Best Company (1977, 1982) Receipts from IRS tabulations of business tax returns (1977, 1982) PCE for owner & tenant-occupied nonfarm dwellings (1977, 1982) Receipts from Census Bureau quinquennial census & annual survey (1977, 1982)

*

IPD composed of lPDs for commissions, underwritingiselling & GNP IPD composed of BEA implicit deflators for health, life, & workers' compensation, & CPIs for auto & property insurance

Composite deflator computed as sum of insurance carrier deflators weighted by commissions from A. M . Best Company. IPD for PCE

Laventhol & Honvath room-rate index

CPI for laundry & dry cleaning CPI for other apparel services

Photographic studies (portrait) & miscellaneous personal services

Beauty shops Barber shops Funeral service & crematories Automotive repair, services, & garages: Automotive rental & leasing without drivers Automobile parking, repair services, & other auto services Miscellaneous repair: Electrical repair shops Watch, clock, &jewelry repair Reupholstery & furniture repair Miscellaneous repair shops Amusement & recreation, except motion pictures: Dance halls, studios, & schools and amusements & recreation services, n.e.c. Theatrical producers, bands, orchestras, & entertainers Bowling alleys & billiard & pool establishments Commercial sports

Health services: Offices of physicians, osteopathic physicians, dentists, & other health practitioners

* * *

*

IPD composed of CPIs for other entertainment services, personal financial & legal services, CPI for beauty & barber shops, & BEA earnings & expense index for clubs & fraternal organizations CPI for beauty parlor services for females CPI for haircuts & other barbershop services CPI for funeral expenses CPI for other auto-related fees

*

CPI for auto maintenance & repair

CPI for appliance & furniture repair CPI for other apparel services CPI for furniture repair Average annual earnings from BLS

*

CPI for other entertainment services

*

CPI for admissions

*

CPI for participant sports

*

IPD composed of CPI for other entertainment services & BEA implicit deflator for pari-mutuel net receipts

*

IPD composed of CPIs for physicians, dentists, & other professional medical services (continued)

Table 1.6

(continued)

Industry

Nursing & personal care facilities Hospitals Medical & dental laboratories Outpatient care facilities Health & allied services, n.e.c. Legal Education: Private education & libraries

Private education housing & meals

Current Dollars

Constant Dollars

Extrapolator or Interpolator of Benchmark Valuest

Price Index for Deflation or Quantity Extrapolator of Base-Year Value

*

HCFA index of input prices BEA composite deflator composed of HCFA index of input prices & CPI for hospital room CPI for other professional services, medical services CPI for professional medical services CPI for other professional medical services CPI for legal service fees

Sum of nonprofit hospital expenses & profit hospital revenues, both from AHA (1977, 1982) Receipts from Census Bureau quinquennial census & annual survey (1977, 1982)

* * *

Sum of PCE for private lower & higher education, private commercial & vocational schools, & private libraries (1977, 1982) Sum of FCE for elementary, secondary & higher education housing & meals (1977, 1982)

IPD composed of BEA deflators for private lower education, private higher education, private commercial & vocational schools, & private libraries IPD composed of PCE deflators for elementary & secondary education housing & for higher education housing

Notes: A&B = agents and brokers: AAR = Association of American Railroads: ABA = American Bus Association; ACLI = American Council of Life Insurance; AGA = American Gas Association; AOP = Association of Oil Pipelines; APTA = American Public Transit Association; BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; CPI = Consumer Price Index (BLS); DOD = Department of Defense; DOE = Department of Energy; DOT = U.S. Department of Transportation; EEI = Edison Electric Institute; FCC = Federal Communications Commission; HFCA = Health Care Finance Administration; ICC = Interstate Commerce Commission; IPD = implicit price deflator; IRS = Internal Revenue Service; M-E = McCann-Erickson; NRPC = National Railroad Passenger Corporation; OGJ = Oil and Gas Journal; PCE = personal consumption expenditure; PPI = Producer Price Index; REA = Rural Electrification Administration: SEC = Securities and Exchange Commission; and USDA = U.S. Department of Agriculture. *Same source as preceding line. t The year(s) in parentheses represents the benchmark input-output (1-0) table to which gross output is directly benchmarked +Gross margin, which is used to measure the gross output of most of the wholesale and retail trade industry, equals sales minus cost of goads sold. §The 1987 quinquennial census, in addition to the 1977 and 1982 quinquennial censuses, was used to benchmark the operating expense, equipment and rental revenues, and sales of MSB&O.

51

BEA Measurement of Services Outputs and Inputs

producing industries (including nonfarm housing services) that are double deflated under phase 1. Intermediate Input Prices

The estimation of composite deflators for intermediate inputs consumed by industries was significantly improved over past practice in three areas: (1) intermediate input weights; (2) imported inputs; and (3) services inputs. Intermediate Input Weights. The weighting scheme used to construct the new composite deflators differs from that used previously in three important dimensions: First, it incorporates a 1972 SIC-based version of the benchmark I0 table for 1977; this eliminates an inconsistency in past practice that arises because 1-0 industries are defined differently than SIC industries.’*Second, it employs unique annual SIC-based weights for every year between 1977 and 1985 (1978-81 weights are derived by interpolation) rather than 1977 1-0 weights reflated to 1982 prices. Third, it includes separate weights for domestic and imported inputs rather than assume that all inputs are domestically produced. These new annual weights were constructed by intensive use of both 1977 benchmark 1-0 work files and unpublished annual 1-0 work files for 1981-85, with the 1985 weights also used for 1986-89.19 Imported inputs. Improved measures of industry constant-dollar intermediate inputs were developed by decomposing the value of each 1-0 defined product consumed by an industry into imported and domestically produced components, by deflating each component with corresponding import and domestic prices, and by computing composite Paache input price indexes for each industry. In general terms, the procedure involves using the 1977 benchmark I0 work files to allocate each imported product class between final demand and intermediate consumption and to allocate the latter across consuming industries in proportion to product-class inputs purchased by each industry. These 1977 proportions are then used to allocate annual estimates of imports by product class, taken from the annual 1-0 tables; in all, more than 1,400 distinct imports are identified and priced. The phase 1 input pricing model incorporates approximately 645 distinct import prices taken from the BLS international price program, from the Bureau of the Mines for mineral products, and from the Commerce Department’s 18. The SIC-based output and input industry definitions used in the GPO differ from 1-0 based definitions for several reasons including: ( I ) In some cases, 1-0 splits an industry out of a larger GPO industry, but in other cases, 1-0combines industries across detailed GPO industries; (2) 1-0 redefines out the secondary products produced by an industry to the industry where it is primary; and (3) 1-0 treats new and maintenance and repair force-account construction performed by nonconstruction industries as part of the construction industry. A discussion of 1-0 conventions is found in Bureau of Economic Analysis (1984). 19. See Bureau of Economic Analysis (1984, 1990) for information on the benchmark and annual 1-0 tables.

52

Michael F. Mohr

national energy accounts (updated by BEA) for energy products.20BLS compiles three different classifications of price indexes for imports-standard international trade classification (SITC), SIC, and end-use category. The import price chosen for each 1-0 product class is determined as follows; If the product class matches, an SITC price is used; if no match is obtained on the SITC, then an appropriate SIC-based price index is used; and, as a last resort, enduse price indexes are used. Because BLS price indexes do not exist for many product classes prior to 1982, either (1) higher-level import prices or (2) matching domestic prices were used to extrapolate the available BLS prices back to 1977. However, less than 26 percent of the 1977 value of imported inputs was so deflated by these two methods in 1977; by 1982 less than 1 percent of the 1977 value of imports was so deflated. Services inputs. An important by-product of the phase 1 effort to develop de-

tailed gross output deflators for services-producing industries is a significant improvement in the deflation of purchased services of all industries. Under past practice, only 30 broad categories of service inputs were identified, and the real input estimates for all of these categories were obtained by deflating with implicit two-digit GPO or earnings deflators. In contrast, the phase 1 estimates of services inputs consumed by 50 double-deflated industries are obtained by using unique annual purchases weights for every year between 1977 and 1985 (with the 1985 weights also used for 1986-89). Moreover, these improved input weights provide detail for more than 300 types of services deflated using more than 100 distinct output-based services deflators; only nine GPO or earnings deflators are used in the new methodology. Table 1.7 shows the services input detail and the prices used to deflate each services input under phase 1. 1.4.4 Effect of the Phase 1 Improvements Table 1.8 illustrates the effect of the recent improvements in terms of their effect on the percentages of GNP accounted for and on the rates of growth experienced by service-sector industries. Specifically, the table compares the unrevised industry GNP shares for 1979 and 1987 with those generated under phase 1 of the GPO improvement program; it also compares previous and revised industry growth rates for the 1979-87 period. Shown also are 1988 industry shares and 1979-88 industry growth rates; these measures are available only in terms of the new methodology of phase 1. It is apparent from table 1.8 that the new methodology does not fundamentally rewrite economic history. Industries with the largest share of constantdollar GNP and the fastest growth in real GNP before revision are also the largest and fastest growing after revision. Nevertheless, several notable effects 20. The national energy accounts are maintained by the Commerce Department’s Office of Business Analysis.

53

BEA Measurement of Services Outputs and Inputs

Table 1.7

Principal Sources of Phase 1 Service Input Prices for Double-Deflated Industries

Service Input Agricultural services Railroad transportation: Dining car receipts, business travel Other passenger train services Rail freight Other railroad services Local and interurban passenger transit: Services from local private transit systems Taxicabs Other Trucking and warehousing Water transportation Transportation by air: Domestic passenger International passenger Mail Freight and express Other air services Pipelines, except natrual gas Transportation services: Private car line services Other Communication services: Telephone services Telegraph services Television services Other communication services Electric, gas, & sanitary services: Electric utilities Gas pipeline Gas utilities Water, sanitation, & other Wholesale trade: Merchant wholesaler & agents & brokers Manufacturers’ sales offices & branches Rental of gasoline tanks & pumps Retail trade: Eating & drinking Other

Source of Price Index IPD for agricultural service gross output CPI for food away from home CPI for intercity train fares IPD for freight gross output IPD for freight gross output IPD for local transit system gross output CPI for taxi fares IPD for intercity bus gross output IPD for trucking and warehousing gross output IPD for water transportation gross output IPD for domestic passenger gross output IPD for international passenger gross output IPD for domestic and international mail gross output IPD for overseas freight and express gross output IPD for transportation by air IPD for pipelines, except natural gas gross output IPD for boxcar rental IPD for transportation services GPO IPD for telephone gross output PPI for telephone toll service IPD for radio & television gross output IPD for telephone & telegraph gross output PPI for electric power IPD for gas pipeline gross output IPD for gas utility gross output CPI for water & sewage maintenance IPD for merchant wholesalers & agents and brokers’ gross output IPD for manufacturers’ sales offices & branches gross output IPD for machinery, equipment, & supplies wholesale trade gross output IPD for eating and drinking gross output IPD for other retail trade gross output (continued)

54

Michael F. Mohr

Table 1.7

(continued)

Service Input Banking: Imputed service charges Other Credit agencies: Savings & loan imputed service charges Other Security and commodity brokers & services: Securities underwriting Securities trading Services allied with exchange of securities Other services Insurance carriers: Automobile insurance Nonlife insurance services, except automobile Other insurance services Mortgage & loan insurance Insurance agents & brokers, & services Real estate services: Nonfarm business rental & property management Farm rental Rent paid by nonprofits Royalties for oil & gas mining Royalties, except oil & gas mining Commissions paid to real estate dealers Condominium association fees & assessments by cooperatives Other Hotel & lodging places Personal services: Funeral & burial expenses Other Business services: Local, national network, & spot TV advertising Radio advertising Magazine & supplements advertising Newspaper advertising, national, classified & local Direct mail advertising Other advertising

Source of Price Index IPD for financial services furnished without payment by commercial banks CPI for personal financial services IPD for financial services furnished without payment by savings & loan associations CPI for personal financial services IPD for underwriting gross output IPD for securities commissions gross output IPD for security & commodity brokers, & services gross output IPD for GNP CPI for automobile insurance IPD for insurance carrier gross output CPI for property and household insurance

IPD for new nonfarm residential buildings and IPD for GNP IPD for insurance agents & brokers, & services Rental rate per square foot from BOMA IPD for rental value of farm housing PCE IPD for capital consumption allowance of nonprofit organizations IPD for oil & gas extraction gross output IPD for PCE IPD for new nonresidential building construction CPI for home maintenance & repair services IPD for real estate GPO Laventhol and Horwath room-rate index CPI for funeral expenses CPI for laundry & dry cleaning McCann-Enckson cost index for network & spot TV advertisements McCann-Erickson cost index for radio advertisements McCann-Enckson cost index for magazine advertisements McCann-Erickson cost index for newspaper advertisements McCann-Erickson cost index for direct mail McCann-Erickson composite cost index

55

BEA Measurement of Services Outputs and Inputs

Table 1.7

(continued)

Service Input Maintenance, cleaning, disinfecting, & exterminating Photofinishing Other business services Automotive repair, services, & garages: Repairs, tire retreading, parking, & washing Other Miscellaneous repair services: Radio, TV, refrigeration, & air conditioning, & electrical & electronic repairs Other Motion picture services: Production & allied services Distribution & allied services Amusement & recreation services: Sports, recreation, & amusements Other commercial recreation & amusements

Theatrical, dance, symphony, & spectator sports productions Health services: Physicians services Other Legal services Education services: Vocational schools, except high schools Higher education & related services Social services Membership organizations: Membership organization expenses Business associations Professional organizations Miscellaneous professional services: Noncommercial museums & art galleries Accounting, auditing, & bookkeeping services Other

Source of Price Index CPI for home maintenance & repair services IPD for film development PCE IPD for business services gross output CPI for automobile maintenance & repair CPI for other auto-related fees CPI for appliance & furniture repairs

Average annual earnings for miscellaneous repair shops & related services from BLS Average annual earnings for motion picture production & allied services from BLS Average annual earnings for motion picture distribution & allied services from BLS IPD for sports & recreation camps IPD based on PCEs for sightseeing, commercial participant amusement n.e.c., sports & recreation camps, & commercial amusement (n.e.c.) CPI for admissions

CPI for physicians CPI for other medical professionals CPI for legal services IPD for commercial & vocational schools PCE IPD for private higher education PCE Average annual earnings for job training & vocational rehabilitation services from BLS

BEA earnings & expense index Average annual earnings for business associations from BLS Average annual earnings for professional membership organizations from BLS IPD for miscellaneous professional services GPO CPI for personal financial & legal services fees IPD for miscellaneous professional services gross output

(continued)

56

Michael F. Mohr

Table 1.7

(continued)

Service Input Government enterprises: Postal services: Unallocated services 1st-class mail 2d-class mail 3d-class mail, bulk rate 3d-class mail, nonprofit bulk rate 4th-class mail Penalty fees Money orders Pension benefit guaranty insurance Share insurance for member credit unions Insurance protection for commercial bank deposits Share & deposit insurance Services to members of Federal Home Loan Banks Imported services: Rail freight transportation Water transportation (n.e.c.) Gas utilities Tire retreading

Source of Price Index

PPI for U.S. postal service, all types PPI for 1st-class mail PPI for 2d-class mail PPI for 3d-class bulk mail PPI for 3d-class nonprofit bulk mail PPI for 4th-class mail PPI for special services and fees PPI for special services BEA earnings & expense index for life insurance No price change assumed Product of index of FDIC ratio of deposit insurance fund to insured deposits & fixedweighted GNP deflator IPD for GNP, fixed weighted IPD for financial services furnished without payment by savings & loan associations PPI for railroad freight IPD for imports of passenger water transportation services Unit prices for imported natural gas from DOE PPI for tires & inner tubes

Notes: For this table, services consist of the primary outputs of (1) private businesses in the agricultural services, transportation and public utilities, trade, finance, insurance, and real estate, and services industries as defined by the 1972 standard industrial classification, and (2) similar services provided by government enterprises. Prices for imported services are shown separately at the end of the table if they differ from prices used for corresponding domestic services. Sources of price indexes for gross output IPDs, except for business services and for miscellaneous professional services, are shown in table 1.6. The IPDs for the gross output for these two industries were estimated from the IPDs for GPO for these industries and from information on inputs from the 1-0 tables. Abbreviations: BEA = Bureau of Economic Analysis; BLS = Bureau of Labor Statistics; BOMA = Building Owners and Managers Association; CPI = consumer price index; DOE = Department of Energy; GPO = gross product originating; IPD = implicit price deflator; PCE = personal consumption expenditures; and PPI = producer price index.

of the new methodology can be found in table 1.8. Examples are highlighted below. Industry Shares of Real GNP

The service sector’s 1979 share of constant-dollar GNP has been revised down by 0.6 percentage points-from 53.1 to 52.5 percent; the commodity sector’s share has been revised upward by an offsetting amount-from 33.7 to 34.4 percent. The upward revision in the share from commodity-producing industries is traceable to mining, whose share increased from 4.1 to 4.5 per-

57

BEA Measurement of Services Outputs and Inputs

cent of real GNP and to durables manufacturing, whose share increased from 13.3 to 13.6 percent. The downward revision in the share from servicesproducing industries is traceable to the transportation sector, where the shares from railroads and transportation by air were both revised down by 0.2 percentage points, and to wholesale trade, whose 1979 share was revised down by 0.6 percentage points. Partially offsetting the combined -1.O percentage point revision from these three industries is an 0.4 percentage point upward revision in the 1979 share of real GNP attributable to electric, gas, and sanitary services. Turning to 1987, we find that the share of real GNP from servicesproducing industries is virtually unchanged from the previous estimated56.8 versus 56.7 percent, but the share from commodity-producing industries is revised up by 0.7 percentage points-from 31.9 to 32.6 percent. Both of these upward adjustments are balanced by a 0.8 percentage point downward revision in the residual-from 0.1 to -0.7 percent.2’ The upward revision in the commodity-producing share in 1987 is traceable to the agriculture, forestry, and fisheries and mining industries, each of whose 1987 shares were increased by 0.2 percentage points, and to nondurables manufacturers, whose share was increased by 0.4 percentage points. Although there is little difference between the previous and phase 1 estimates of the service sector’s share of real 1987 GNP, significant revisions did occur in two of the sector’s detailed industries: the share attributable to electric, gas, and sanitary services was revised up by 0.4 percentage points; the share attributable to health services was revised down by 0.5 percentage points. Finally, comparing the newly estimated 1988 shares of real GNP with the revised share estimates for the 1979 business cycle peak, we find that commodity-producing industries lost 1.4 percentage points between 1979 and 1988; services-producing industries gained 4.2 percentage points. These compare with minus 1.8 and plus 4.3 percentage points, respectively, between 1979 and 1987. Industry GNP Annual Average Rates of Change

Turning to average rates of change, table 1.8 shows many large differences between the previous and phase 1 industry GNP estimates for the 1979-87 period. All the particularly large revisions occurred in services-producing industries. For example, the changes recorded by railroad, local and interurban passenger, trucking and warehousing, and air transportation; wholesale trade, security and commodity brokers; and hotel and other lodging phases were all revised up by between 1.2 and 7.7 percentage points. Meanwhile, very large downward revisions were made to the changes recorded by water and pipeline transportation; radio and TV broadcasting; insurance carriers and insurance agents and brokers; auto repair services and garages; and health and legal 21. The residual component of the GPO estimates is the difference between aggregate GNP in constant dollars, measured as the sum of expenditures less the statistical discrepancy in constant dollars, and aggregate GNP in constant dollars measured as the sum of GPO by industry.

Table 1.8

Previous and Phase 1 Average Annual Rates of Change and Shares of Constant-DollarGNP for Selected Years (70) Share of GNP

1979

GNP Commodity-producing industries Agriculture, forestry, & fisheries Mining Construction Durables manufacturing Nondurables manufacturing Services-producing industries Transportation Railroad Local & interurban passenger transit Trucking & warehousing Water Air Pipelines, except natural gas Services Communication Telephone & telegraph Radio & television broadcasting Electric, gas, & sanitary services Wholesale trade Retail trade Finance, insurance, & real estate Banking

1987

Previous

Phase 1

Revision

Previous

Phase 1

100.0 33.7 2.4 4. I 5.4 13.3 8.6 53. I 4.3 0.9 0.2

100.0 34.4 2.4 4.5 5.2 13.6 8.7 52.5 3.9 0.7 0.2

0.0 0.7 0.0 0.4 -0.2 0.3 0.1 -0.6 -0.4 -0.2 0.0

100.0 31.9 2.5 3.1 4.6 13.7 8.2 56.7 3.5 0.4 0.2

100.0 32.6 2.7 3.3 4.6 13.4 8.6 56.8 3.8 0.7 0.2

1.8 0.3 0.8 0.2 0.2 2.3 2.0 0.2 2.6 6.8 9.2 14.4 1.7

1.7 0.3 0.6 0.2 0.2 2.3 2.0 0.3 3.0 6.2 9.1 14.4 1.7

-0.1 0.0 -0.2 0.0 0.0 0.0 0.0 0.1 0.4 -0.6 -0.1 0.0 0.0

1.6 0.2 0.7 0.1 0.3 2.8 2.5 0.3 2.8 7.6 9.6 14.5 1.6

1.7 0.1 0.8 0. I 0.3 2.7 2.5 0.2 3.2 7.5 9.6 14.7 1.6

Average Annual Rates of Change of GNP

1979-87 1988* Revision Phase 1 Previous Phase 1 Revision

197988,* Phase 1

0.0 0.7 0.2 0.2 0.0 -0.3 0.4 0. I 0.3 0.3 0.0

100.0 33.0 2.3 3.2 4.4 14.5 8.6 56.7 3.7 0.7 0.2

2.4 1.7 3.0 -1.3 0.2 2.72 1.8 3.2 -0.2 -6.0 -1.9

2.4 1.7 3.9 -1.5 0.6 2.2 2.2 3.4 2.1 1.7 -0.6

0.0 0.9 -0.2 0.4 -0.5 0.4 0.2 2.3 7.7 1.3

2.6 2.1 2.3 -1.2 0.6 3.4 2.4 3.5 2.1 2.0 -0.7

0.1

1.6 0.1 0.8 0.1 0.3 2.7 2.4 0.2 3.3 7.4 9.9 14.5 1.6

0.8 -0.6 1.7 -0.5 5.7 5.1 5.2 4.1 3.0 3.7 2.8 2.5 1.9

2.1 -9.9 6.2 -2.2 5.7 4.8 5.1 1.7 3.1 4.9 3.0 2.6 1.9

1.3 -9.3 4.5 -1.7 0.0 -0.3 -0.1 -2.4 0.1 1.2 0.2 0.1 0.0

I .9 -8.8 5.4 -0.2 5.8 4.6 4.9 2.0 3.9 4.7 3.5 2.7 1.6

-0.1 0.1 0.0 0.0

-0.1 0.0 -0.1 0.4 -0.1 0.0 0.2 0.0

0.0

Credit agencies other than banks Security & commodity brokers, & services Insurance carriers Insurance agents & brokers, & services Real estate Holding & other investment companies Services Hotels & other lodging places Personal Business Auto repair, services, & garages Miscellaneous repair Motion pictures Amusement & recreation Health Legal Educational Social services & membership organizations Miscellaneous professional Private households Government & government enterprises Statistical discrepancy Residual Rest of the world

0.2

0.2

0.0

0.2

0.2

0.0

0.2

6.3

6.3

0.0

5.7

0.3

0.3

0.0

0.6

0.8

0.2

0.8

10.8

14.5

3.7

14.3

0.9 0.6

1.1 0.5

0.2 -0.1

1.o 0.6

1.o 0.5

0.5

1 .o

3.3 3.7

1.2 2.5

-2.1 - 1.2

1.7 2.3

10.6 0.2

10.5 0.2

-0.1 0.0

10.3 0.3

10.3 0.3

0.0 0.0

10.1 0.3

2.0 7.3

2.1 7.3

0.1 0.0

2.2 7.1

13.5 0.7 0.7 2.5 0.8

13.7 0.8 0.7 2.5 0.8

0.2 0.1 0.0 0.0 0.0

15.9 0.6 0.7 3.6 0.9

15.3 0.7 0.7 3.6 0.7

-0.6 0.1 0.0 0.0 -0.2

15.3 0.7 0.7 3.7 0.7

4.5 0.6 2.5 7.4 3.5

3.9 1.9 2.2 7.4 1.7

-0.6 1.3 -0.3 0.0 - 1.8

3.9 2.4 2.3 7.2 I .7

0.3 0.2 0.4 3.9 0.9 0.6 0.9

0.3 0.2 0.4 4.0 0.9 0.6 0.9

0.0 0.0 0.0 0.1 0.0 0.0 0.0

0.3 0.2 0.5 4.6 1.o 0.6 0.9

0.3 0.2 0.5 4.1 1.o 0.6 0.9

0.0 0.0 0.0 -0.5 0.0 0.0 0.0

0.3 0.2 0.5 4.0 1 .o 0.6 0.9

1.7 2.8 5.1 4.6 3.6 3.1 2.5

2.2 3.6 5.7 2.8 2.6 2.6 2.5

0.5 0.8 0.6 - 1.8 - 1.0

-0.5 0.0

2.9 3.6 5.9 2.6 3.1 2.7 2.9

1.4 0.2 11.8

1.4 0.2 11.8

0.0

1.7 0.2 10.8

1.7 0.2 10.8

0.0 0.0

1.7 0.2 10.5

5.0

0.0 0.0

1.3

4.9 1.5 1.2

-0.1 0.0 -0.1

4.7 1.3 1.3

0.0 -0.3 I .7

0.0 -0.5 1.7

0.0 -0.2 0.0

-0.2 0. I 0.7

-0.1 -0.7 0.7

-0.2 -0.7 0.7

-

-

-

-

* 1988 industry GNP was not estimated until phase

1 of the GPO improvement program.

0.0

-0.1

0.0 0.1 - 0.8

0.0

I .5

-9.3

-8.8

0.5

- 7.3

60

Michael F. Mohr

services. The 1979-87 average annual rate of change in each of these industries was reduced by between 1 .O and 9.3 percentage points. As shown in table 1.8, the upward revisions made to the 1979-87 real growth rates of railroads, air transportation, and security and commodity brokers are particularly large. Below, we outline the sources of the revision in each industry in order to further illustrate the influence of the improved estimating procedures. In the case of railroads, the previous estimates of constant-dollar GNP were obtained by indirect, or method 3, double deflation. As such, a significant part of the revision in growth rates is due to the switch from method 3 to method 2 double deflation, but most of the revision is due to improvements made in the constant-dollar gross output estimates. The previous gross output estimates were obtained by using a composite implicit deflator based on the PPI for railroad freight and the CPI for intercity train fares. By contrast, the revised real gross output estimates are obtained by using a composite implicit deflator based solely on the physical gross output of freight and passengers transported, as described in table 1.6. In the case of air transportation, the previous estimates of constant-dollar GNP were obtained by extrapolation with constant-dollar gross output estimates for the industry. Thus, the revision in the real GNP growth rate reflects both the switch from output extrapolation to method 2 double deflation and revision of the constant-dollar gross output estimates. The latter revision stems largely from the incorporation of benchmarks from the 1977 and 1982 benchmark 1-0tables and of a quantity extrapolator for the output of domestic passengers transported (see table 1.6) in place of the CPI and personal consumption expenditure (PCE) deflators for airline passenger fares used in the previous estimates. Finally, in the case of securities and commodity brokers, the previous estimates of constant-dollar GNP were obtained by extrapolation with labor input and thereby assumed no productivity growth. The large upward revision in this industry’s real GNP growth reflects the switch from labor input extrapolation to method 2 double deflation. To implement double deflation in this industry, original estimates of the components of its real gross output were constructed. Of particular significance, the real output of the security brokerage activity is now estimated by extrapolating 1982 securities commissions with a quantity index representing the number of public orders received by registered exchanges and over-the-counter markets. And, the real output of securities underwriting/investment banking activities is now estimated by extrapolating 1982 fees for such activities with an index of the quantity of new issues brought to market by underwriters (see table 1.6). In summary, the new GPO estimates, derived from the aforementioned methodology improvements, indicate not only that the share of the U.S. economy accounted for by services-producing industries during 1979-87 was more than a half a percentage point smaller than previously estimated but also

61

BEA Measurement of Services Outputs and Inputs

that the increase in that share between 1979 and 1987 was a half point larger than previously estimated. However, in contrast to much publicized speculation, the relative growth in the service sector has not come at the expense of manufacturing.22On the contrary, reflecting the large increases in productivity that occurred after 1982 and the rapid growth in manufactured exports that occurred after 1986, the new GPO estimates indicate that manufacturing, especially durables, increased its relative share by 1.2 percentage points between 1979 and 1988. Finally, the revised GPO estimates for transportation industries (except water), wholesale trade, security and commodity brokers, and hotels and lodging places suggest, ceteris paribus, that these servicesproducing industries experienced substantially more productivity growth during 1979-87 than previously estimated.

1.5 GPO Improvement Program The GPO improvement program is a long-run and ongoing effort to improve comprehensively and systematically both the industry current- and constant-dollar GNP estimates by preparing consistent time series of production accounts, which will provide detailed and complete coverage of the outputs produced and the inputs consumed by each industry. The result of achieving this core objective will be a substantial reduction in most of the remaining methodology limitations and will provide GNP estimates that are better suited for measuring industry growth and productivity. The program is anticipated to be completed in three phases. Phase 1 has been completed and has produced extensive, but incomplete, improvements covering the period from 1977 to 1989.23Phase 2, scheduled for completion during September 1992 as part of the forthcoming comprehensive NIPA revision, will provide estimates that reflect most of the planned methodology and date improvements for the 1977-forward period. Finally, phase 3 will extend the improvement program back to 1958 and is expected to be completed during 1993, depending on available resources. In what follows, major improvements planned for phases 2 and 3 of the improvement program are discussed in detail. 1.5.1 Selected Improvements Six specific categories of improvements are to be implemented during phases 2 and 3. Included are the following: (1) improved current-dollar GNP

22. See, e.g., Mishel(l988, 1989) and Kelly and Wyckoff (1989). Each has been prominent in speculating not only that there has been a secular decline in manufacturing’s share of GNP since 1979 but also that this decline would be manifested in the new GPO estimates. 23. Phase 1 of the GPO improvement program was completed when revised 1987 and 1988 estimates and initial 1989 estimates were published in the Survey of Current Business, April 1991.

62

Michael F. Mohr

estimation; ( 2 ) expanded use of double deflation; ( 3 ) improved measurement of gross output; (4) improved measurement of intermediate inputs; (5) expanded industry detail; and (6) use of superlative indexes. Current-Dollar GNP Estimation

Under phase 1, current-dollar industry GNP continued to be developed by the method 2 technique discussed in section 1.3; during phase 3, currentdollar GNP estimates for most, if not all private industries, will be increasingly derived by the method 1 technique; that is, as the difference between current-dollar gross output and current-dollar intermediate input. The latter method will ensure that industry gross output, GPO, and intermediate input measures are internally consistent and it will make BEA’s estimating methodology for the GPO consistent with United Nations SNA accounting procedures. Double Dejlution

With completion of phase 1, double deflation is now used to derive the real GNP for 50 industries and the nonfarm household services component of real estate; these industries represent 87 percent of the 1987 real GNP originating in the private sector. In conjunction with the adoption of method 1 for currentdollar GNP estimation, the objective during phases 2 and 3 of the improvement program is to obtain the real GNP estimates for the component industries of the entire service sector (except private households) through method 1 double deflation, including the nine industries above private households in table 1.6.24Of these nine, particular effort will be directed toward business services, banks and other credit agencies, and real estate, except nonfarm business services. Gross Output

Preliminary current- and constant-dollar gross output estimates for 197789 were developed during phase 1 for the component SIC industries listed in table 1.7. In order to achieve the expanded double deflation objective, current- and constant-dollar gross output and intermediate inputs measures will be developed for the remaining non-double-deflated services-producing industries. The final estimates for 1977-forward will be constructed during phase 2; those for 1958-76 will be constructed during phase 3 . The methodology for producing these estimates is designed to generate nominal and real gross output measures that (1) are defined on a consistent SIC industry definition; (2) provide comprehensive coverage of every industry’s output; and ( 3 ) identify the major product composition of each industry’s gross output. This last change permits more accurate deflation than is possible in existing real gross output measures. 24. Seen. 16.

BEA Measurement of Services Outputs and Inputs

63

With respect to the current-dollar estimates, the new methodology involves the use of a consistent definition of gross output across industries, the development of five-year benchmarks consistent with benchmark 1-0tables, and the development of annual extrapolator series that are both consistent over time and consistent with the benchmark data sources. Construction of industry and product benchmarks and extrapolators have been and will continue to be based not only on intensive use of benchmark and annual 1-0account work files but also on considerable research to choose the “best” annual extrapolator for each industry and product. With the possible exception of some finance industries, the phases 2 and 3 final benchmark estimates for industry nominal gross output (GO) will be based on the following formula:

GO

=

receipts (including BEA coverage adjustment) - cost of resales

+ inventory change + commodity taxes + new force-account construction.

With respect to the constant-dollar gross output estimates, the improvements to be implemented during phases 2 and 3 involve both improving the manufacturing methodology and introducing a similar one for services-producing industries. Under the phase 1 methodology, the product composition and deflation of total shipments from each four-digit manufacturing industry is determined at the five-digit product-class level. Inventory change is currently estimated and deflated at the two-digit industrial level so that real gross output can be determined only at the two-digit industry level. For phases 2 and 3, however, current- and constant-dollar inventory change will be available at the four-digit manufacturing level, and benchmarked current-dollar gross output estimates will be generated and deflated at four-digit level. The effort to replicate the manufacturing procedure in services-producing industries, which began during phase 1, will be extended during phases 2 and 3 in several steps. First, benchmark 1-0work files will be used to identify and measure more of the products produced by each industry. Second, more and better source data will be incorporated in order to develop improved currentdollar extrapolators at this more detailed product level. Third, more detailed industry product deflators or quantity index extrapolators will be incorporated. Together these improvements are expected to produce better real gross output and implicit price deflator estimates for services-producing industries.

Intermediate Inputs Critical to achieving the improvements outlined above and below is the implementation and completion of a comprehensive project designed to produce improved estimates of the current- and constant-dollar services and other intermediate inputs consumed. Under phase 1, estimates of the current-dollar intermediate input consumed by each industry continued to be derived by the residual method-current-dollar gross output less current-dollar value added-and the constant-dollar estimates continued to be derived by employ-

64

Michael F. Mohr

ing industry-specific composite intermediate input deflators. During phase 3 , current-collar and constant-collar estimates of the detailed services and other intermediate inputs consumed by each industry will be constructed from 1-0 tables. In turn, these improved input estimates will permit method 1 estimation of both current- and constant-dollar industry GNP. The methodology used to construct the improved input estimates will go beyond that employed in phase 1 in several ways: (1) it will incorporate SICbased benchmarks developed from the 1958, 1963, 1967, 1972, 1977, and 1982 benchmark 1-0 tables, and from the latest annual 1-0 table; ( 2 ) it will partition the intermediate transactions matrix of each benchmark into three submatrices-energy, other materials, and services; ( 3 ) intrabenchmark interpolation will be conducted either on the basis of input cost share coefficients within each of the three submatrices or by incorporating previously unused industry data to move the cell values within a submatrix between benchmarks; (4) the final current-dollar cell estimates in each submatrix will be obtained by using a biproportional balancing algorithm and a comprehensive collection of row and column controls; and (5) the constant-dollar input estimates will incorporate both improved estimates of the imported and domestic composition of inputs and more detailed deflators for services inputs. Industry Detail The private-sector industry detail in the phase 1 GNP estimates is confined to 60, essentially two-digit industries. It is anticipated that the improvements in methodology discussed above will make it possible in phase 3 to significantly increases the number of industries in the GPO estimates. Although the industry count that ultimately will appear is uncertain at this time, an expansion to three-digit detail from the present two-digit detail appears possible for the mining; manufacturing; electric, gas, and sanitary services; and services industries. This expansion in industry detail will permit a much more refined study of productivity, structural change, and competitiveness issues than that possible from the presently published GPO series. Superlative Indexes

A major criticism of BEA’s existing aggregate real GNP estimates is that they are calculated by using a fixed base-year weighing formula. As a result, the aggregate real GNP estimates may not properly reflect price-induced substitution along given utility and production functions and thereby tend to overstate aggregate prices increases and understate aggregate real output increases. In addition, periodic shifting of the base year tends to reduce growth rates because the new index often overweighs goods whose prices have risen most rapidly between base years and whose real sales have, therefore, grown least rapidly. For these reasons, several observers have suggested computing aggregate annual real GNP and its growth by the use of chain superlative

65

BEA Measurement of Services Outputs and Inputs

index number formulas.25The BEA is planning to publish estimates of the growth in aggregate real GNP obtained by alternative index number formulas as part of the forthcoming comprehensive GNP revision (Young 1989). The above noted criticisms also apply to industry level real GNP estimates. The foregoing improvements in the measurement of gross output and intermediate input will result in an increase in the quality and quantity of the data necessary to develop estimates of the change in industry-level measures of aggregate gross output and intermediate inputs based on superlative index numbers. In turn, these estimates can be used to prepare implicit superlative index number estimates of the change in industry-level GNP.26 1.6 Concluding Remarks The BEA anticipates that the fully implemented GPO improvement program will significantly improve the services measures in the GPO estimates, eliminate most of the criticisms of the previous industry GNP estimates, and improve the credibility of productivity, structural change, and competitiveness analyses based on the industry GNP estimates. There are, however, limits to the degree to which either the historical or future estimates can be improved. In the first instance, going back in time runs directly into the source data constraints that in large part shaped the previous methodology with all its apparent potential for measurement error. The introduction of a new methodology and more intense mining and exploitation of existing data can produce significant improvements but they cannot completely mitigate measurement error traceable to limitations in the available source data. Since the early 1980s, advances in the Census Bureau’s annual coverage of service industries have made possible significant improvement in the estima25. Superlative index numbers are traced to Diewert (1976). A summary of the contemporary literature on aggregation theory and the production theory foundations of alternative superlative index formulas is found in Mohr (1988). chap. 2 and appendix. Triplett (1989) provides a comparison of the growth in producers durable equipment calculated from the conventional base-yearprice weighted quantity indexes and from alternative superlative index number specifications. 26. Two possible formulas for calculating the growth of industry real GNP from the growth in its gross output and purchased inputs come to mind: (1)

or (2)

CO$ GNP

=

CO$ GO - CO$ purchased input

log CO$ GNP = log CO$ GO

-

log CO$ purchased input.

The first formula, which is the standard double-deflation calculation of methods 1 and 2 in the text, is justified only if an industry’s production technology is additively separable between its value-added inputs and its intermediate inputs; i.e., intermediate and value-added inputs of all forms are either perfect substitutes or complements-partial elasticities of substitution are either infinite or zero. The second formula, however, is justified under the somewhat less restrictive condition of log linear (multiplicative) separability; is., intermediate inputs and value added inputs of all forms are exact substitutes-partial elasticities of substitution are finite and equal. See, e.g., Denny and May (1977).

66

Michael F. Mohr

tion of current-dollar gross output for many service industries. Nevertheless, there remains a substantial agenda of long-standing limitations that can only be overcome through expanded source data collection by BLS and the Census Bureau. Included in this agenda are the following: 1 . Expand economic censuses to all services-producing industries, particularly those in FIRE; 2 . Provide at least census-year coverage of not only the detailed types of materials but also of the detailed types of services inputs consumed by U.S. industries. 3. Expand the coverage of Census annual surveys to all industries and collect data on both materials and services inputs. 4. Collect annual quantity data by type of product or service provided by service sector industries. These data will provide improved estimates of real gross output and will provide weights for the development of qualityadjusted prices. 5. Expand the business services portion of the BLS PPI program. 6. Develop (by BLS) output and input price deflators that reflect both changes in the character and improvements in the quality of services produced. 7. Collect (by the Census Bureau) annual data on imported goods sold and purchased by establishments in wholesale and retail trade. This agenda has few new items. Over the years, BEA has supported Census Bureau and BLS data-collection initiatives in the aforementioned areas. In addition, several independent committees have prepared reports that recommended granting BEA, BLS, and Census Bureau increased budgetary authority to address these pressing problems. The earlier reports included the 1977 report of the advisory committee on gross national product data improvements and the 1981 report of the National Science Foundation panel to review productivity statistics. Unfortunately, the necessary resources have just begun to materialize and the problems still remain many years after several calls to action. In recent years, however, criticism of the industry GNP data has significantly raised not only the level of visibility of these problems but also the consequences of failing to adequately address them. For example, the April 1987 report of the working group on the quality of economic statistics to the Economic Policy Council noted: Because of difficulties of measuring quality in services, construction, and various high-technology products, current-dollar output in these industries may have been “over-deflated‘’ and real growth underestimated. . . . The solution is not entirely in BEA’s hands-BEA depends upon data produced by other government agencies and private organizations and cannot always readily bring about improvements in the quality of these data.

67

BEA Measurement of Services Outputs and Inputs

More recently, on January 25, 1990, Michael Boskin, chairman of the Council of Economic Advisers, issued a coordinated call to action on these problems when he released the recommendations of President Bush’s working group on improving economic statistics-“Improving the Quality of Economic Statistics.” 27 Most recently, these recommendations formed the basis for a comprehensive initiative for improving economic statistics in the president’s fiscal year 1992 budget. This initiative, the fiscal year 1992 Economic Statistics Initiative, includes programs that address the agenda of needs in services measures outlined earlier in this section.28As a result of these developments, prospects for effective action to deal with important source data deficiencies in services and other areas of the GPO estimates appear much brighter.

References Baily, Martin N., and Robert J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 2:347-43 1. Bureau of Economic Analysis. 1984. The Detailed Input-Output Accounts of the US. Economy, 1977, vols. 2 and 3. Washington, D.C.: Government Printing Office. . 1988. Gross Product by Industry: Comments on Recent Criticisms. Survey of Current Business 68 (July): 132-33. . 1990. Annual Input-Output Accounts of the U.S. Economy, 1985. Survey of Current Business 70 (January): 41-56. . 1991. Gross Product by Industry, 1977-88: A Progress Report on Improving the Estimates. Survey of Current Business 71 (January): 23-37. Council of Economic Advisers. 1991. FY 1992 Economics Statistics Initiative Improving the Quality of Economic Statistics. Press release, February 14. Denny, Michael, and Doug May. 1977. The Existence of a Real Value-Added Function in the Canadian Manufacturing Sector. Journal of Econometrics 555-69. Diewert, W. Erwin. 1976. Exact and Superlative Index Numbers. Journal of Econometrics 41106-71. Kelly, Henry, and Andrew Wyckoff. 1989. Missing Links: The Need for Better Data on Purchased Services. The Service Economy 3 (October): 1-6. (This study is part of a larger study undertaken by the authors at the Congressional Ofice of Technology Assessment. “Statistical Needs for a Changing U.S. Economy.” Background paper no. OTA-BP-E58. September 1989.) Lal, Kishori. 1990. Service Industries in the Business Sector of the Canadian Economy. Review of Income and Wealth 1 (March): 83-94. Marimont, Martin L. 1969. Measuring Real Output for Industries Providing Services: 27. A summary of the working group’s report appears in the Survey ofcurrent Business. February 1990.2. 28. See Office of Management and Budget (1991); for further information see Council of Economic Advisers (1991).

68

Michael F. Mohr

OBE Concepts and Methods. In Production and Productivity in the Services Industries, ed. V. R. Fuchs, 15-52. New York: NBER. Mishel, Lawrence R . 1988. Manufacturing Numbers: How Inaccurate Statistics Conceal U.S. Industrial Decline. Washington, D.C.: Economic Policy Institute. . 1989. The Late Great Debate on Deindustrialization. Challenge 32 (JanuaryFebruary): 35-43. Mohr, Michael F. 1988. Capital Inputs and Capital Aggregation in Production. BEA discussion paper, no. 31 (August). , and Paul T. Christy. 1986. Changes in the Structure of the U.S. Economy since 1960: A Primer. In Implications of Internationalization of the U S . Economy. U.S. Department of Commerce, Office of Economic Affairs Workshop on Structural Change, January. Office of Management and Budget. 1990. Comparative Study of Reporting Units in Selected Employee Data Systems. Statistical policy working paper no. 16. Statistical Policy Office, Office of Information and Regulatory Affairs, May. . 1991. Budgedof the United States Government: Fiscal Year 1992, pt. 2 , 320I . Washington, D.C.: Government Printing Office. Triplett, Jack E. 1989. New Measures of Producers’ Durable Equipment. Paper presented at Western Economic Association, Lake Tahoe, Nev. June. United Nations. 1968. A System of National Accounts, series F., no. 2, rev. 3. New York. . 1979. United Nations, Manual on National Accounts at Constant Prices, series M., no. 64. New York. Young, Allan H. 1989. Alternative Measures of Real GNP. Survey of Current Business April, 27-36.

Comment

Martin Neil Baily

There is a common perception that the slow growth of productivity in the U.S. economy since 1973 is attributable in some substantial degree to the mismeasurement of real output, particularly service-sector output. This proposition was fairly easy to refute with respect to the 1973-79 period. The growth slowdown was pretty much across the board, so that almost all the major sectors of the economy had experienced slower growth, and in fact the most serious declines in productivity occurred in the goods producing industries of construction and mining. The situation changed in the 1980s, however, a change that seems to be continuing into the 1990s. Fueled by huge increases in the quality of computers, productivity growth has recovered dramatically in the manufacturing sector. And the collapse of productivity in construction and mining has ameliorated. Meanwhile, the growth slowdown in service industries has intensified. The growth rate of labor productivity in services in the 1950s and 1960s was quite good, but it has become steadily weaker since then. The extent to which Martin Neil Baily is professor of economics at the University of Maryland and a research associate of the National Bureau of Economic Research.

69

BEA Measurement of Services Outputs and Inputs

the U.S. economy resumes more normal rates of productivity growth in the remainder of this century largely depends on the extent to which the (broadly defined) service sector is able to achieve improvements in productivity. Of course one possibility is that improvements in service-sector productivity are really taking place but we are not seeing them because of errors in the data. This was a question that Robert J. Gordon and I addressed in our 1988 paper (see full citation in Michael F. Mohr’s paper). And the story we came up with is a paradoxical one. There is a lot of evidence of egregious errors in the data, and many of these errors suggest that productivity growth in the service sector is being understated. On the other hand, it is very difficult to make the case that measurement errors account for a major part of the slowdown, either in service-sector productivity growth or in aggregate growth. The explanation for this paradox is that first, in many cases the measurement errors predated the slowdown in growth. And second, many of the errors are in industries that are partly or wholly intermediate goods suppliers, which means that improvements in the measurement of output in these industries does not change the estimates of aggregate real output. In a way, however, it is a relief to get the slowdown issue out of the way, because that takes the pressure off the statistical agencies. We can now get down to the serious business of tackling the many measurement errors that do exist in the data for the service sector. Probably the slowdown cannot be explained as a measurement problem, but the interest in this issue has prompted a major effort at data improvement. And irrespective of explaining the slowdown, it is very important to know how prices, real output, and productivity are doing in the service industries, the part of the economy that accounts for over half of GNP. Michael F. Mohr is the head of the branch at the Bureau of Economic Analysis (BEA) that prepares the data on gross product originating by industry and in his paper he describes the major effort that is underway to improve the quality of those data. It is an impressive effort, particularly so because the budget crisis keeps all of the statistical agencies squeezed for funds. In the past, value added in many parts of the service sector has not been computed using data for outputs and inputs and their appropriate deflators. In some cases the deflator for labor compensation has been used to deflate current dollar GPO, which has the effect of making real GPO growth in the affected industry depend largely on the growth in employment. The improvement program that Mohr describes will develop “industry current- and constant-dollar GNP estimates by preparing consistent time series of production accounts, which will provide detailed and complete coverage of the outputs produced and the inputs consumed by each industry” (see sec. 1.5). This program of improvement has been partially completed already and the remainder will be completed by 1993. In the next rounds of improvement, the BEA will also be exploring the use of superlative index numbers, a change that could make a big difference,

70

Michael F. Mohr

given the large relative price changes that take place in the economy. We can look forward to the results of this effort, including the BEA’s proposal to compute some divisia value-added numbers to compare with the standard figures for value added. The concept of value added as the arithmetic difference between gross output and purchased inputs is not one with much validity in production theory. In table 1.8 some of the fruits of the first phase of the program are shown, and one important issue brought out in the table is the share of output produced in services. Some people have argued that data errors have lead to an exaggeration of the share of GNP produced in the manufacturing sector with a corresponding understatement of the service-sector share. This idea is not supported by the revised data presented in the table. There was a rearrangement of manufacturing, giving a bigger share to nondurables and a smaller share to durables, but the total manufacturing share has not changed much at all. The data do show a rise in the service share, but this has come at the expense of the government sector. The revisions have resulted in a shift away from the share of services that are publicly provided. In looking at the growth rates of real output in the main service sector categories, Gordon and I can perhaps be forgiven a bit of “I told you so.” As we predicted, the improvements in the data have not lead to a big change in the estimate of overall growth-the slowdown has not been explained away. And the biggest changes in estimated growth have occurred in the transportation sector, where we suggested that growth was being understated. Despite the fact that the program of improvement that Mohr has described is an impressive one, there remain some serious problems to be tackled, most of which are out of Mohr’s hands and will surely need new funding to solve. In particular, the price indexes for banking and financial services, for medical services, for insurance and for the rental component of the real estate sector are very weak indeed, The increase in the quality of health services is not being captured by the current deflators and this problem also gets carried over to the insurance industry, to the extent that this industry is providing health insurance. Improvements in the quality of houses and office buildings are not well captured, despite the use of hedonics for the construction industry, so that real rental costs are probably being exaggerated and the real output of the real estate sector is then understated. The deflators for banking and financial services are also still weak. And while it is hard right now to argue that bank productivity has really been great over the last ten years, one can still be concerned that the contributions that innovations in this sector will be making to future productivity will be missed. Another problem is more directly in Mohr’s province. His paper shows that the improvements that are being made in the industry data are making heavy use of input-output tables. But the reference table that is being used is the 1977 table. This same table is even being used to carve up imports. BEA is making annual adjustments to the coefficients in this table, but it is still a

71

BEA Measurement of Services Outputs and Inputs

pretty old reference table to be using, given all the structural changes that have been taking place in the economy. We will have much more confidence in the revised industry numbers when more complete data finally emerge. A similar problem arises with the employment matrix-actually the problem is even worse. Adequate data are not currently available by which to allocate capital income by industry for firms that span several industries. In practice, Mohr has to use an industry employment matrix and even this is somewhat out-of-date. As he notes, capital income is not such a large fraction that this is going to throw off real output estimates by much. But there are situations where it is important to know the profit rate by industry. For example, Charles Schultze and I found that there was an apparent inconsistency between the manufacturing profit rate and the predictions of the neoclassical growth model. This may simply reflect allocational errors in capital income. Phase 3 of the improvement program will allow BEA to replace the figures that are generated by the employment matrix and hopefully this will improve the estimates of profit rates by industry. This is an enormously helpful paper that will hearten those of us who consume the data that Mohr’s office puts out. There has been an erosion of the statistical base in some areas, so it is good to see a place where things are improving. There remains much to be done, but we are grateful that so much is being done.

This Page Intentionally Left Blank

2

Productivity Measurement in Service Industries Edwin R. Dean and Kent Kunze

The U.S. Bureau of Labor Statistics (BLS) presently publishes productivity measures for 173 industry titles, of which 39 are for the broad service, or nongoods, sector. In 1975 the bureau published a total of 53 industry titles, of which 10 were in the service sector. At present, the published industries cover 42 percent of all workers in service industries in the private business sector. After placing the service industry productivity measures in the context of the larger BLS program of productivity measurement, we will provide in sections 2.1-2.3 basic information about the industry productivity measures, including the measurement model used by the bureau. Section 2.4 describes the methods used for specific industries; the discussion groups the industries into broad sectors, such as transportation, trade, and communications. Figures 2.1-2.7, should be examined in conjunction with section 2.4 because they illustrate the construction of several of the measures. Section 2.5 examines trends in the measures themselves, and section 2.6 discusses some comparisons of the BLS measures and the industry output measures developed by the U. S , Bureau of Economic Analysis (BEA). These industry measures form one part of a broad BLS program of productivity measurement. The methods used for these measures differ from the methods used for the other components of the program. Industry productivity measures are available for mining and manufacturing industries as well as for service industries. They are annual measures, employing gross output and hours of labor input, mostly at the three- and four-digit (but occasionally at Edwin R . Dean is associate commissioner for productivity and technology, Bureau of Labor Statistics, U.S. Department of Labor. Kent Kunze is a senior economist and project leader in the Office of Productivity and Technology, Bureau of Labor Statistics, U.S. Department of Labor. The authors wish to thank the staff of the Division of Industry Productivity and Technology Studies of the Bureau of Labor Statistics for their assistance with this paper. Special thanks go to Brian Friedman and Virginia Klarquist for their work on the figures and tables.

73

74

Edwin R. Dean and Kent Kunze

the two-digit) levels. These measures are prepared and published in an autumn press release and then in a bulletin several months later. Other major components of the bureau’s productivity measurement program are (1) quarterly measures of output per hour for major sectors of the economy, which are prepared using gross product originating (value added) and are published in press releases eight times a year; (2) annual measures of multifactor productivity for major sectors, which use value-added output and are published annually; (3) annual measures of multifactor productivity for two-digit manufacturing industries, which use gross output and are updated and published when resources permit; (4)annual measures of multifactor productivity for selected three-digit industries, which are published annually; (5) international comparisons of labor productivity in manufacturing, which are published annually for 12 countries; (6) measures of labor productivity for selected federal government functions, which are published annually; and (7) labor productivity measures for selected state and local government activities, which are published annually. The methods used for the various measures differ. For example, the industry measures of labor productivity use gross output, base-year weighted; the major sector multifactor productivity measures use value-added output, baseyear weighted, and the two- and three-digit industry multifactor measures use Tornqvist indexes of gross output.

2.1 The Model The industry labor productivity measures are computed as indexes of output per hour by dividing an index of output by an index of aggregate employee hours. For industries in trade and services (the service industry within the broad nongoods sector), measures are prepared relating output to the hours of all persons involved in producing that output, including self-employed and unpaid family workers. The index of output per hour is expressed as the quotient of an index of weighted output and an index of employee hours, as follows: Output index + Employee hours index = Output per hour

-

- A p - -

C~t,~qt,~ C1t.oqr.o ’

C1r.rqz.t’

where l,,r is unit labor requirements of product i in year r, where unit labor requirements are aggregate hours spent in producing product i divided by gross output of product i; qt,tis gross output of product i in year t , and “product” is understood to be either a good or a service. The output index compares the quantities of the various products in the current year with the quantities in the base period, each weighted by the employee hours expended per unit produced in the base period. The employee hours index compares the aggregate

75

Productivity Measurement in Service Industries

employee hours in the base and current periods. The employee hours data are the total hours expended by employees in establishments classified in the industry to produce the base-period and current-period composites. The productivity index, as described above, eliminates the effects of shifts in product mix on productivity. That is, because of the fixed, current-year output weights, labor productivity changes resulting from changes in the relative quantities of the various outputs will not affect the productivity indexes. These productivity indexes are affected only by changes in unit labor requirements of the individual products. The measurement procedures described above were developed a number of years ago and are intended to show the changes in total labor requirements of the industry that result from changing production processes for the various industry products (Seigel 1961).

2.2 General Description of Measures 2.2.1 output The industry output indexes are based on measured quantities of products or services provided by the industry. The unit of measurement of the quantity can be either a physical quantity such as passenger-miles, ton-miles, or kilowatt hours or a constant-dollar value of production. One of the primary objectives for output measures is to start with as much detail or disaggregation of the measured outputs as possible. For example, the output index of the electrical utility industry (SIC 491) is not simply the number of kilowatt hours produced. Instead, it is derived from indexes of the number of kilowatt hours sold to each of seven types of customers. The amounts sold to each type of customer are aggregated with specific weights for each type of service. Similarly, the output index of hardware stores (SIC 5251) is obtained by aggregating the deflated revenues of 23 different merchandise lines from all stores. The intent is to develop output indexes that correctly reflect the differing trends in the output of various products produced within the industry. As a general rule, weights are derived every five years from the economic censuses because only the economic censuses provide the detailed data needed for the disaggregation of the output and the development of the weights. For most industries that do not rely on census data, weights are still changed every five years for consistency within the measurement program. Although the above model states that the desired weights are unit labor requirements, this is not the case for many industries and, as a matter of fact, for most of the service industries. Industry output information is obtained from a wide variety of sources, both public and as private. Output indexes for trade, services, and manufacturing make extensive use of Census Bureau data. Other important federal government sources include the Department of Transportation, the Internal

76

Edwin R. Dean and Kent Kunze

Revenue Service (IRS), the Department of Energy, and the Department of the Interior. For deflated value series, industry price indexes are derived from the BLS producer price indexes and consumer price indexes (CPIs).

2.2.2 Employee Hours Indexes of employee hours are computed by dividing aggregate employee hours for each year by the base-period aggregate. Employee hours are treated as homogeneous and additive with no distinction made between hours of different groups of employees. For industries in which the self-employed are important, indexes are constructed for the hours of all persons, including paid employees, partners, proprietors, and unpaid family workers. Industry employment and employee hours indexes are developed from basic data compiled by BLS, the Census Bureau, and other sources. For most private nonagricultural industries, BLS publishes employment and average weekly hours data for production or nonsupervisory workers and employment data for all employees. The Bureau of the Census publishes employment and aggregate hours data for production workers and employment data for all employees. Average annual hours of nonproduction and supervisory workers are estimated from all available data. For trade and service industries, all-persons hours estimates are derived by summing the aggregate hours for paid employees and the estimated aggregate hours for the self-employed and unpaid family workers. In a few industries, labor input measures are simply total employee counts.

2.3 Characteristics of the Measures The above model of productivity measurement is very straightforward. However, when put to use, it can become complicated. When the bureau begins to study the possibility of developing a new industry productivity measure, the first task is to examine the available data. It is common to find that the data suffer from various deficiences. If the deficiencies are important and not correctable, the study is ended and no measure is developed. In other instances, special efforts are made to correct the data problems. There are a number of conditions considered during the examination of the data. It is important to find out whether, for the industry in question, there have been significant changes in the standard industrial classification (SIC) code over the time period considered, and, if so, whether adjustments can be made for the changes. Changes in the SIC codes indicate major changes in the type of products or services being produced or changes in product mix. If these new products or the changing mix of products cannot be introduced with acceptable weights, then the output and productivity indexes will not capture the correct output changes over time. A second important condition that is examined is whether the industry has become more or less vertically integrated over time. This condition is a partic-

77

Productivity Measurement in Service Industries

ularly important consideration when output indexes, based on a gross output concept, are constructed from deflated value data. Gross output (the output concept generally employed by the bureau) may not change as an industry becomes more or less integrated. However, labor hours could change with a change in vertical integration. Therefore measured productivity can change even though there are no changes in the production process. Changes in vertical integration are generally examined by studying changes in the ratio of value-added output to gross output. Two other important considerations are (1) what percentages of primary products are made within the industry, and what changes in the percentages have occurred over time; and (2) what percentage of output for the industry is composed of primary products. The first of these conditions is referred to as product coverage and the second as specialization. The reasons these percentages are important have to do with how the data are reported-a problem mostly in noncensus years. The amounts (values) of primary products produced for the year are reported on a “wherever made” basis; the amount of industry output is reported for primary and secondary products combined. The amount of industry primary products output is not known. In order to develop weights and match prices for each product group, it is necessary that both the coverage and specialization ratios be high. Historically, an industry measure has not been developed unless both percentages have been over 90 percent. Some of the industries for which the bureau publishes measures are regulated-most notably in the transportation, communications, and electric, gas, and sanitary service industries. Regulated prices of outputs may not reflect competitive market conditions, a fact that can have adverse effects on the output measures. When value weights are being used, the regulated prices are part of the weights used to compute the output indexes. In this case the weights may reflect neither unit labor requirements nor marginal costs of production, and the output indexes are not weighted correctly. It is difficult to determine how much effect the use of revenue weights, which contain regulated prices, has had on the output trends. The trend in the output index for the railroad industry was revised slightly downward when a change from revenue weights to labor cost weights was introduced in 1974. Certainly there has been some distortion in the changes of the output index for telephone communications during the regulated years because long distance rates were set artificially high to offset low local rates. For labor input there are also a number of potential data problems. Presently, establishment surveys do not collect hours data on supervisory and nonproduction workers. The hours for supervisory and nonproduction workers are estimated for each industry. In addition, household data on hours of work of self-employed and unpaid family workers are generally very thin at the industry level-a particularly acute problem for measuring labor input in the service sector, where most nonfarm self-employed and unpaid family workers

78

Edwin R. Dean and Kent Kunze

are working. As a result changes in annual hours of these workers are often erratic. The hours collected by the bureau’s establishment survey are based on hours paid and not hours worked, which is the more appropriate measure for labor input. However, results from the bureau’s hours at work survey show that for the private business sector the ratio of hours at work to hours paid has been reasonably stable since 198 1 (Jablonski, Kunze, and Otto 1990). Indeed, even for many of the nongoods industries this seems to be the case. For the transportation sector, however, the ratio increased nearly 5 percentage points from 1981 to 1989.’ Hence, actual hours of work have increased faster than hours paid. This would suggest that labor input increased 5 percentage points more than reported for this period and that labor productivity growth for industries in this sector was about 5 percentage points less, on average, than reported for this time period.

2.4 Description of Service Industry Measures As noted earlier, the bureau presently publishes productivity measures for 173 industry titles, of which 39 are for the broad service or nongoods sector. In 1975 the bureau published a total of 53 industry titles, of which 10 were in the service sector. Of the present 39 industry titles in the service sector, a total of 32 are for mutually exclusive service industries. The difference between the number of titles and industries arises from the overlap of measures for both a two-digit SIC code and a three- or four-digit SIC code within the two-digit group. There is at least one published measure for every major industry division in the service sector.2 As of 1987, the published industries covered 42 percent of all workers in service industries in the private business sector (table 2.1). (The private business sector excludes government and nonprofit institutions.) The reader should note that figures 2.1-2.7 provide diagrams intended to clarify how the various industry output measures are developed and to supplement the following verbal descriptions of the construction of the productivity measures for each sector. Transportation. The bureau publishes productivity measures for five industries in the transportation ~ e c t o r The . ~ employment coverage of these meal . Unpublished data provided by the Office of Productivity and Technology, Bureau of Labor Statistics. 2. The bureau does produce productivity measures for the Federal Government including the Postal Service. However, these measures are not included in this study. Technical notes describing detailed characteristics of all industry measures are available on request. 3. For the intercity trucking industry, SIC 4213 (part), two measures are produced; one is for freight trucking alone and one is for all intercity trucking. For railroad transport (SIC 401),two measures are produced using different output concepts, car miles and revenue ton miles. Thus, seven measures are produced for the five industries covered.

79

Productivity Measurement in Service Industries

Table 2.1

Service Industries Covered by BLS Productivity Measures and Employment Coverage for Major Service Sectors 1987 Employment (thousands)

Services producing sector Transportation Railroad transport, revenue traffic (SIC 401) Railroad transport, car miles (SIC 401) Buscarriers,classI(SIC4111,413,414)(part) Intercity trucking (SIC 4213) (part) Intercity trucking, freight (SIC 4213) (part) Air transportation (SIC 451 1,4521) (part) Petroleum pipelines (SIC 4612, 13) Transportation employment covered Communications Telephone communications (SIC 481 1) Communications employment covered Electric, gas, & sanitary services Gas & electric utilities (SIC 491, 492, 493) Electric utilities (SIC 491, 493) (part) Gas utilities (SIC 492, 493) (part) Utilities employment covered Trade Scrap & waste materials (SIC 5093) Hardware stores (SIC 5251) Department stores (SIC 531 1) Variety stores (SIC 5331) Retail food stores (SIC 54) Grocery stores (SIC 541 1) Retail bakeries (SIC 546) Franchised new car dealers (SIC 55 11) Auto & home supply stores (SIC 5531) Gasoline service stations (SIC 5541) Apparel & accessory stores (SIC 56) Men’s & boy’s clothing stores (SIC 561 1) Women’s ready-to-wear stores (SIC 5621) Family clothing stores (SIC 5651) Shoe stores (SIC 5661) Furniture, home furnishings, & equipment stores (SIC 57) Furniture & home furnishings stores (SIC 571) Appliance, radio, TV, & music stores (SIC 572, 73) Household appliance stores (SIC 5722) Radio, TV, & music stores (SIC 573) Eating & drinking places (SIC 58) Drug- & proprietary stores (SIC 5912) Liquor stores Trade employment covered Finance, insurance, & real estate

59,860 3,478 271 271* 20 448 293* 457 18 1,214 1,300 905 905 930 81 7 605* 212* 81 7 26,287 117 I76 2,033 248 3,191 2,749* 190* 922 346 681 1,242 116* 419* 281* 233*

Employment Coverage (%)

34.9 69.6

87.8

938 558* 379* 105* 274*

6,460 595 I 78 17,127 7,131

65.2 (continued)

80

Edwin R. Dean and Kent Kunze

Table 2.1

(continued) 1987 Employment (thousands) ~~

~

Emp1oyment Coverage (%)

~

Commercial banking (SIC 602) FIRE employment covered Services Hotels, motels, & tourist courts (SIC 701 1) Laundry & cleaning services (SIC 721) Beauty &barber shops (SIC 7231, 41) Beauty shops (SIC 7231) Automotive repair shops (SIC 753) Services employment covered Services-producing sector employment covered

1,562 1,562 20,734 1,558 468 770 687* 803 3.599 25.224

21.9

17.4 42.1

Source: Bureau of Labor Statistics. Note: Employment coverage of BLS measures is in italics. *Designates employment is aggregated at a higher level.

sured industries is 35 percent of the transportation sector, based on 1987 employment. Interestingly, the same industries represented over 50 percent of transportation employment in 1980. The decrease in coverage is attributable to declining employment in some measured industries and increasing employment in the yet-unmeasured industries within the sector. For example, employment in railroads (a measured industry) has dropped from 67 1,000 in 1967 to 271,000 in 1987; employment in transportation services (an unmeasured industry) has risen from 100,000 in 1967 to 296,000 in 1987. Conceptually, the output measures for transportation industries are relatively easy to define: output is the movement of goods or passengers over distance. This is a quantifiable definition. Industry output for this sector is based on physical quantities of ton-miles, passenger-miles, or barrel-miles. In both trucking and railroads, the index of ton-miles is adjusted for changes in commodity mix being transported. The adjustment factor is the difference between the price weighted growth rate of tons of commodities and the unweighted aggregate growth rate of tons of commodities. There are over 170 commodity lines for railroads and trucking. For the air transportation industry four separate measures of output are aggregated using revenue weights (fig. 2.1). Input measures are indexes of total hours for the railroad and petroleum pipeline industries. For trucking, air transportation, and bus carriers, labor input measures are indexes of annual employment only. A potential problem specific to trucking and railroads is the lack of data on average length of haul. To the extent that unit requirements are different for long-distance hauls versus short hauls, a bias occurs if average length of haul changes during the period studied. The adjustment for commodity mix changes, referred to above, may partially correct for this bias. Another pos-

81

Productivity Measurement in Service Industries

Total Domestic: Revenue Passenger-Miles Revenue Ton-Miles

Total International:

Revenue Passenger-Miles Revenue Ton-Miles

OUtDUt

/ / Weights

/

/

Index

Fig. 2.1 lkansportation: Air transportation (SIC 4511, 4521 part) Note: Physical quantities are combined using unit revenue weights to create an industry output index. Source; Bureau of Labor Statistics

sible problem, mentioned above, is the use of regulated output prices for deflating revenues or developing weights. If output prices, because of regulation, do not accurately reflect competitive market equilibrium conditions, then output measures can be biased. This problem is greater for the historical data than for recent data. Communications. There is only one industry measure within the communications sector-telephone communications (SIC 48 13), which covered 70 percent of total employment in the sector in 1987. Again, coverage has actually decreased over the past ten years as a result of rapid growth in the unmeasured industries (television, radio, and cable television broadcasting). Output indexes are generated as weighted aggregates of deflated revenues collected by four different categories of telephone services: local calls, measured toll service (MTS), wide-area toll service (WATS), and all other (this includes private line service). The revenue data are collected and published by the Federal Communications Commission. Deflators are derived from price indexes compiled and published by the BLS under its producer price index program. Revenues by type of service are used as weights. Labor input is an index of total hours derived from the bureau’s establishment survey data. Besides the possible problem of regulated prices, a measurement problem may exist because of flat-rate charges for WATS or local service. The price indexes for flat-rate services reflect changes in the service rate only and do not reflect changes in the volume of traffic or other additional services being provided with the local or WATS service. Hence changes in deflated revenues may not reflect total changes in outputs. Electric, gas, and sanitary services. The bureau publishes three productivity measures for industries in this group. The industries are electric utilities (SIC 491 plus part of SIC 493), gas utilities (SIC 492 plus part of SIC 493), and

82

Edwin R. Dean and Kent Kunze

the combination of the two. These measures do not include any governmentowned establishments. In 1987 the measures covered approximately 88 percent of total employment in this sector. Output indexes in the electric utilities industry are weighted aggregates of seven types of electric services measured in kilowatt hours (fig. 2.2). Services are differentiated by type of customer: residential, commercial, industrial, and so on. Weights are unit revenues for each service. For the gas utilities industry the output index is a weighted aggregate of four types of services; again the weights are unit revenues by type of service. The two industries’ outputs are aggregated using employee weights. Output data are collected by the Department of Energy, the Rural Electrification Administration, and the American Gas Association. Input indexes are derived from employment and average hours data collected by the bureau’s establishment survey. Trade. The trade sector is the largest of the service producing sectors. In 1987 over 26 million people worked in this area. The bureau publishes measures for 23 different industries of which 19 are mutually exclusive. These published industries cover 17 million workers or 65 percent of the trade sector. Only one of these measures-scrap and waste materials (SIC 5093)-is in wholesale trade. The remainder of the measures are for retail trade industries. With the exception of scrap and waster materials, output indexes for the trade industries are weighted aggregates of deflated sales of merchandise lines. Sales data, available from the census of retail trade and annually from Current Business Reports, are deflated by price indexes derived from CPIs. For census years the sales data are reported by merchandise lines. For noncensus years more aggregated sales data are reported. For the later, deflators are calculated by combining the prices with the base year weights. The number of merchandise lines varies by industry (table 2.2). Several types of weights are used in the BLS measures (figs. 2.3 and 2.4). The most commonly used weights are product group gross margins derived from the input-output tables produced by the BEA. Unfortunately gross margin data are not available for all merchandise lines reported. Labor cost weights are used for some years in the retail food store measure. Employment weights are used in the franchised new car dealer measure. Detailed all employee hours weights are used for department stores. Industry all person hour weights are sometimes used to weight four-digit measures into two-digit measures, as in retail food stores and total apparel stores. In some industry measures, gasoline service stations, for example, gross margin weights were not available for most of the services and products sold, other than gasoline. In these cases sales data are used for weighting products sold. The labor input indexes for most of the measures in retail trade are allpersons hours. The bureau includes measures of self-employed and unpaid family workers derived from either IRS data or current population survey data. Some measures, for example, department stores and franchised new car

83

Productivity Measurement in Service Industries Electricity - Kilowatt Hours

Residential Commercial Industrial Public streets and highways Other public authorities Railroads and railways Interdepartmental

\ -

Electricity Product Output Index

-Revenue

\ Employee Weights

Gas - Therrns Unit

Residential Commercial Industrial Other

__

Revenue Weights

Gas Product Output Index

/

Industry output Index

Fig. 2.2 Utilities: Gas and electric utilities (SIC 491, 492, 493) Nore: Physical quantities are combined to form output indexes using unit revenue weights. Indexes are aggregated to an industry output index using employee weights. Source: Bureau of Labor Statistics

Table 2.2

Number of Merchandise Lines for Retail Industries

Covered Retail Trade Industries Hardware stores (SIC 5251) Department stores (SIC 531 1) Variety stores (SIC 5331) Grocery stores (SIC 541 1) Retail bakeries (SIC 546) Franchised new car dealers (SIC 551 I ) Auto & home supply stores (SIC 553) Gasoline service stations (SIC 5541) Men's & boy's clothing stores (SIC 561 I ) Women's ready-to-wear stores (SIC 5621) Family clothing stores (SIC 5651) Shoe stores (SIC 5661) Furniture & home furnishings stores (SIC 571) Appliance, radio, TV, & music stores (SIC 572, 573) Household appliance stores (SIC 572) Radio, TV, & music stores (SIC 573) Eating & drinking places (SIC 58) Drug- & proprietary stores (SIC 5912) Liquor stores (SIC 5921)

Merchandise Lines Used 23 41 33 26 15 7 17 11 20 19 22 7 56 36 17 19 5 30

6

Source: Bureau of Labor Statistics.

dealers, are all employee hours measures because of the lack of suitable selfemployed and unpaid-family-worker information from these sources. Hardware stores (SIC 525) provide a good example of productivity measurement methods in retail sales. For a benchmark year (economic census year) annual sales are reported by merchandise line for the industry in the

84

Edwin R. Dean and Kent Kunze Consumer Price Indexes

’

Maintenance work # 2 ~~~loil Tobacco products Alcoholic beverages

Gasoline Service Station Industry Sales

Line Sales Weights

/ /

Weighted CPI Industry Deflator

-

-

Deflated Value output Index

//

Food at away home from home

Fig. 2.3 Retail trade: Gasoline service station (SIC 5541) Source: Bureau of Labor Statistics

CPI Deflated Merchandise Line Sales Eating Places SIC 5812 Groceries MealdSnacks Alcoholic drinks ETC.

Groceries Meals/Snacks Alcoholic drinks ETC. Groceries Meals/Snacks Alcoholic drinks ETC.

Gross Margin Weighted Sales

> WTS

-G--

Cafeterias

3 GM WTS

/ Drinking Places SIC 5813 Groceries Meals/Snacks Alcoholic drinks ETC.

Restaurants and Lunchrooms

__z

8!3

/

-

Refreshment ‘laces

Drinking ‘laces

All Person Hour Weights

58

/

Drinking Places output

Output

/

Fig. 2.4 Retail trade: Eating and drinking places (SIC 58) Nore: Merchandise line sales are deflated by CPIs or combined CPI deflators and then aggregated to industry segments using gross margin weights. Segments are further aggregated with employment and hours weights. Source: Bureau of Labor Statistics

census of retail trade. To provide output indexes between pairs of benchmark years, these detailed sales data are separately deflated by the appropriate CPIs and aggregated using base-year gross margin weights. The gross margin weights are introduced as a proxy for labor weights in the aggregation of quantities. The gross margin data are developed from BEAs input-output tables. Annual measures of output for hardware stores are developed from total industry sales as reported in current business reports; CPIs for all items sold in hardware stores; and from merchandise line sales reported in the most recent economic census. Annual industry sales, reported for the total industry,

85

Productivity Measurement in Service Industries

are deflated using an aggregation of CPIs. The price indexes are weighted according to the reported merchandise line sales from the most recent census. Table 2.3 lists the merchandise lines and CPIs used for hardware stores. The annual output indexes are bench marked to the benchmark year indexes derived, as described above, from the more detailed sales data published in each census of retail trade. Labor input for the hardware stores measure is an index of hours of all persons working in the industry. The number of employees and the average weekly hours of nonsupervisory workers are derived directly from the bureau’s establishment survey. The number of self-employed and unpaid family workers and their respective average weekly hours are derived from current population survey data. The average hours of supervisory workers are constructed from the census of population. These average hours are held constant between decennial census years. Average hours and employment by class of worker are simply multiplied and summed over all classes of workers in the industry. Finance, insurance, and real estate. Commercial banking (SIC 602) is the

only industry in this sector for which the bureau publishes a productivity measure. In 1987 commercial banks employed l .56 million people, or 22 percent of all workers in the finance, insurance, and real estate sector. The output measure for this industry is based on the number of transactions for three major banking activities: time and demand deposits, loans, and trusts (fiduciaries). (See fig. 2.5.) Each major activity is an aggregate of more refined measures. The indexes for these three activities are aggregated with fixed-year employment weights to obtain the output index for commercial banking. The employment weights were derived for 1967, 1972, 1977, and 1982 from data published in the Federal Reserve’s functional cost analysis (FCA). The components of time deposits consist of (1) demand deposits, and (2) time and savings deposits. Output indexes for both of these categories are constructed and aggregated on the basis of employment weights derived from the FCA. Time and savings deposits at commercial banks include all regular savings accounts, club accounts, certificates of deposit, and other time deposits. The output measure for demand deposits consists of two components-the number of checks written by the public and transacted through the banks, plus the number of electronic funds transfers (Ems) to the banks’ customer accounts. The two sets of numbers are added for each year, yielding the number of demand deposit transactions from 1967 forward. The output series for the number of checks is based on three surveys conducted in 1970, 1974, and 1979 in addition to annual data published by the Federal Reserve System. The three surveys are used as benchmarks to which the Federal Reserve’s annual data are adjusted by linear interpolation.

86

Edwin R. Dean and Kent Kunze

Table 2.3

Merchandise Lines, Sales, and CPIs Used for Hardware Stores, 1982 1982 Sales ($1,000)

Percent

Groceries & other foods Cigars, cigarettes, & tobacco Health & beauty aids

13,266 10,025 11,478

0.0016 0.0012 0.0014

Men’s & boys’ clothing, except footwear Women’s & girls’ wear, except footwear Footwear, except infants’ Curtains, draperies, & dry goods Major household appliances Small electric appliances

11,827

0.0015

Food at home Tobacco products Toilet goods & personal care appliances Men’s & boys’ apparel

6,454

0.0008

Women’s & girls’ apparel

8,256 6,977

0.0010 0.0009

Footwear Textile house furnishings

203,986 184,128

0.0252 0.0228

44,624 32,843

0.0055 0.0041

Household appliances Office machines, small electric appliances, etc Television Sound equipment

42,112 23,680

0.0052 0.0029

434,429

0.0537

6,183 271,409

0.0008 0.0335

3,047,705 1,640,569

0.3766 0.2027

784,706

0.0970

1,106,017

0.1367

Cars, trucks, & power vehicles Automotive fuels & lubricants

7,401 84,774

0.0009 0.0105

Auto tires, batteries & accessories Household fuels

94,733

0.01 17

14,740

0.0018

8,092,322 8,335,088 0.9709

I .oooO

Merchandise Lines

TVs Audio equipment, musical instruments, supplies Furniture & sleeping equipment Floor coverings Kitchenware & home furnishings Jewelry Sporting goods Hardware & tools Plumbing & electrical supplies Lawn & garden equipment & supplies Lumber & building materials

Total MLS used Total MLS reported in census MLS used as a percentage of MLS reported in census

CPIs

Furniture & bedding Floor & window coverings, infants, laundry, etc. Tableware, serving pieces, etc . Jewelry & luggage Sporting goods & equipment Weighted CPIs* Plumbing, heating, elec. & cool. supplies Weighted CPIst Maintenance & repair commodities New vehicles Motor fuel, motor oil, coolant Automobile parts & equipment Fuel oil, coal & bottled gas

*Weighted CPIs: lawn equipment, power tools, & other hardware; miscellaneous supplies & equipment (maintenance & repair commodities). ‘Weighted CPIs: lawn equipment, power tools & other hardware; lawn & garden supplies.

87

Productivity Measurement in Service Industries Number of Accounts: Employee Benefit Trust Personal Trust Estate Employee Benefit Agency Other Trust Accounts

Revenue - Trusts

Weights /

Number of Transactions: Time Deposit Checking Account and Electronic Fund Transfer

7

Deposits

\ -

\

Employee Weights

Commercial Banking output

Real Estate Loans Consumer Loans Credit Card Loans

Weights

-

Commercial & Other Loans

Fig. 2.5 Finance: Commercial banking (SIC 602) Source: Bureau of Labor Statistics

Loans are measured in terms of the number of new loans extended. The loan output measure is an aggregate of four types of loans: real estate, consumer, credit card, and, commercial and other loans. These loan outputs are aggregated by employment weights, derived from the FCA, for 1967, 1972, 1977, and 1982. The output measure for real estate loans represents the number of residential mortgage loans, the number of construction loans, and the number of commercial mortgage loans. Data used to derive real estate loans are obtained from the Federal Housing Association (FHA) and the U.S. Department of Housing and Urban Development (HUD). The index for consumer loans is a composite of the number of automobile loans, home improvement loans, personal loans, mobile home loans, and other installment loans. The weights used to aggregate the consumer loans output components are derived from American Bankers Association (ABA) data on the expense per average loan. The number of credit card loans is represented by the physical volume of bank credit card transactions. The output measure is based on the number of bank credit card transactions occurring within the United States as reported by the VISA card network and the Mastercard Association. The measure of trust department output is the number of accounts. The total number of accounts, by type, is combined on the basis of net income, as reported by the FCA. The output measure for commercial and other loans, for 1977 forward, is based on the number of loans as reported in the Federal Reserve’s survey of

88

Edwin R. Dean and Kent Kunze

terms of bank lending. Prior to 1977 no information on commercial loans is included in the banking output index. Labor input in commercial banking is measured by an index of allemployees hours from 1967 forward. The number of employees and hours are derived from BLS establishment data. Average weekly hours are available only for nonsupervisory workers. They are inputed to all employees. This procedure assumes that supervisory workers work the same number of average weekly hours as nonsupervisory workers. Services. The bureau publishes only five measures in the services division of the SIC system, which includes, for example, hotels (SIC 701 1) and automotive repair shops (SIC 753). The employment coverage for these five industries is the lowest of all the sectors in the overall service sector at just 17 percent. Furthermore, this number is somewhat inflated because this sector has the largest percentage of nonprofit establishments, and these are not included in the denominator of the coverage ratio. Outputs are aggregated indexes of deflated revenues. In general, deflators are constructed from appropriate CPIs and revenue weights. Labor input indexes are derived from the bureau’s establishment survey of employment and hours, the current population survey, and IRS data. The CPS and IRS data are used to estimate the number of self-employed and unpaid family workers. An example of productivity measurement in the service division is the automotive repair shop industry (SIC 753). Annual measures of output are constructed by deflating total industry receipts, as reported in current business reports, service annual survey, by the CPI expenditure category “automative maintenance and repair.” The annual output indexes are bench marked every five years to the receipts data published in the census of service industries. Figure 2.6 diagrams the construction of the output index for benchmark years. For a benchmark year, annual receipts are reported for 12 kind-of-operation groupings. The annual receipts are deflated by appropriate CPIs to the previous benchmark year, indexed, and combined to the three-digit level using base-year paid-employee data as weights. The receipts data available by kind of operation are for those establishments with payroll. To arrive at an all-establishment index, a coverage adjustment ratio is derived by dividing industry receipts of all establishments by receipts of establishments with payroll. The index of weighted receipts is multiplied by this coverage adjustment ratio to arrive at the final benchmark output index for each census period. Labor input for automotive repair shops is an index of hours of all persons working in the industry. The number of employees and the average weekly hours of nonsupervisory workers are derived directly from the bureau’s establishment survey. The number of self-employed workers is derived from IRS data. The number of unpaid family workers and the average weekly hours for

Productivity Measurement in Service Industries

89

Industry Receipts

Consumer Price Indexes

.

1 Deflated

Receipts

Type of Shop 7532pt 7532pt 7533 7534 7536 7537 7538pt 7538pt 7539pt 7539pt 7539pt 7539pt

Top and body repair shops

Paint shops Exhaust system repair shops l i r e re air sho s Auto g&s repkcement shops Auto transmission repair shops Auto repair shops, exc. diesel Diesel engine repair, auto Electric and fuel systems repair Radiator repair Brakes and wheel alignment Other auto repair shops, n.e.c.

Deflated

Matching '

CPl's

-

Receipts

by Typeof

Employee Weights

/

output SIC 753

Shop

Fig. 2.6 Services: Automotive repair shops (SIC 753) Nore: Individual shop receipts are only from establishments with payrolls. Industry output i s further adjusted at the total industry level to reflect the output of all establishments. Source: Bureau of Labor Statistics

Class of Worker

Employment Source

Average Weekly Hours Source

Nonsupervisory Employees

BLS EstablishmentSurvey

BLS Establishment Survey

Supervisory Employees

BLS EstablishmentSurvey

Census of Population

Partners

IRS Statistics of Income

Current Population Survey

Proprietors

IRS Statistics of Income

Current Population Survey

Unpaid Family Workers

Current Population Survey

Current Population Survey

Fig. 2.7 Employment and hours sources for service industries Source: Bureau of Labor Statistics.

self-employed and unpaid family workers are derived from current population survey data. The average hours of supervisory workers are constructed from the census of population. These hours are held constant between decennial census years. Average hours and employment by class of worker are multiplied and summed over all classes of workers in the industry. Figure 2.7 shows the construction of labor input for all the service industries described above.

2.5

Results

Tables 2.4, 2.5, and 2.6 show the average growth rates of labor productivity, output, and labor input, respectively, for the published industries in the service sector. Figures 2.8-2.14 show the measures of all the industries. Time periods have been selected according to business cycle peaks except for the

90

Edwin R. Dean and Kent Kunze

Table 2.4

Output per Hour for Service Industries, Average Annual Rates of Change (%) Annual Growth Rate

Industry (SIC) Transportation: Railroad transport, revenue traffic (SIC 401) Bus carriers, class I (SIC 41 11, 413, 414)(part) Intercity trucking (SIC 4213)(part)* Intercity trucking, freight (SIC 4213)(part)* Air transportation (SIC 451 1,452l)(part)* Petroleum pipelines (SIC 4612, 4613) Communications: Telephone communications (SIC 481 1) Electric, gas, & sanitary services: Gas & electric utilities (SIC 491, 492, 493) Electric utlities (SIC 491, 493)(part) Gas utilities (SIC 492, 493)(part) Trade: Scrap & waste materials (SIC 5093) Hardware stores (SIC 5251) Department stores (SIC 53 1 1) Variety stores (SIC 5331) Retail food stores (SIC 54) Grocery stores (SIC 541 1) Retail bakeries (SIC 546) Franchised new car dealers (SIC 551 1) Auto & home supply stores (SIC 5531) Gasoline service stations (SIC 5541) Apparel & accessory stores (SIC 56) Men’s & boys’ clothing stores (SIC 561 1) Women’s ready-to-wear stores (SIC 5621) Family clothing stores (SIC 5651) Shoe stores (SIC 5661) Furniture, home furnishing, & equipment stores (SIC 57) Furniture &home furnishings stores (SIC 571) Appliance, radio, TV, & music stores (SIC 572, 573) Household appliance stores (SIC 5722) Radio, TV, & music stores (SIC 573) Eating & drinking places (SIC 58) Drug & proprietary stores (SIC 5912) Liquor stores (SIC 592) Finance, insurance, & real estate: Commercial banking (SIC 602)

Change in Annual Growth Rate

1967-73

1973-79

1979-87

1967-73 to 1973-79

4.94

1.4%

8.9%

-3.5%

1973-79 to 1979-87 7.5%

1.3

- 1.3

- 1.0

0.0

3.6 3.4 4.6 7.1

3.2 4.1 4.8 0.7

2.2 2.8 3.3 0.4

-0.4 0.7 0.1 - 6.4

4.6

6.8

5.2

2.2

- 1.6

4.5 5.2 2.8

0.9 1.3 - 0.2

-0.5 0.7 - 4.5

- 3.6 - 4.0 - 3.0

- 0.6

5.3 3.4 3.6 4.8 6.4 0.9 5.2

2.6 3.2 - 2.7 -0.7 -0.3 - 1.9 0.2 2.3 3.7 2.1 0.8 3.5 -0.1 1.9 2.0

2.9 2.5 3.6 - 0.6 - 0.7 -0.7 - 2.7 1.2 3.3 3.4 2.9 2.7 4.6 2.2 0.8 4.0

5.1

1.4

5.5

-

I .9 1.2 1.2

2.9

1.2 - 3.9 - 1.8 - 2.7 - 1.6 - 1.3 - 2.8 - 1.2 -6.6

0.3 -

1.0

- 1.3

-1.5 -0.3

- 1.5 -4.3 0.0 0.4 2.0 -0.1 -0.4 -0.9 0.9 1.1 -0.4 0.8 1.8 1.1

-3.2

2.3 - 1.2 1.9

0.9

-3.7

-0.5

3.0

8.2

-2.5

I .o 6.4

3.4 2.4 -0.6 1.1 -0.7

5.6 9.1 - 1.2 0.0 -0.7

2.3

0.6

2.0

1.o

-1.6 -5.3 - 1.7

5.2 2.2 6.8 -0.5 -1.1 0.1 1.4

91

Productivity Measurement in Service Industries

Table 2.4

(continued) Annual Growth Rate

Change in Annual Growth Rate 1967-73 to

Industry (SIC) Services: Hotels, motels, & tourist courts (SIC 701 1 ) Laundry & cleaning services (SIC 721) Beauty & barber shops (SIC 7231.41) Beauty shops (SIC 723) Automotive repair shops (SIC 753)

1967-73

1973-79

1979-87

1973-79

1.8 0.5

1.4 -0.2

-1.6 -1.9 -1.0 -0.9 -0.5

-0.4 -0.6

1.1

0.3 -0.7

1973-79 to 1979-87 -3.0 -1.7 -2.1 - 1.2 0.1

*Labor input used is number of employees.

beginning and end points of the complete series. The end point, 1987, is the last year for which measures have been completed for all industries. The beginning point, 1967, is the first year for which measures are complete for most of the industries. During the first time period, 1967-73, of the 24 mutually exclusive industries, only one (bus carriers) experienced negative productivity growth ( - 1.3 percent). The industry with the highest rate of productivity growth was petroleum pipelines (7. l ) percent). Two industries, bus carriers and laundry and cleaning services, had negative output growth during this period. Telephone communications and electric utilities had the fastest output growth rates, 8.3 percent annual rates of increase in both cases. During the second time period, 1973-79, 10 of the 31 mutually exclusive industries experienced negative productivity growth. It is important to note that 19 of the original 24 industries had slower productivity growth during the slowdown period than in the earlier period. Six industries actually had negative output growth in this period, and 19 had slower output growth rates than during the first period. The industry with the highest productivity growth rate for this period was the telephone communications industry (6.8 percent annually), which also had the best increase in productivity growth from the first period to the next. Radio, television, and music stores had the fastest rate of output growth at 9.2 percent annually. Another strong performer in terms of output growth (6.8 percent) and productivity growth (4.8 percent) for this period was the airlines industry. For the same time period, 1973-79, variety stores had the worst productivity growth performance ( - 2.7 percent); family clothing stores experienced the greatest decline in the annual productivity growth rate from the first to the second period (6.6 percentage points). The petroleum pipelines industry also had a large turnaround in its productivity growth rate, dropping 6.4 percentage points from the previous time period. Two other industries showing poor peiformance during this period were gas utilities and electric utilities. The gas

92

Edwin R. Dean and Kent Kunze

Table 2.5

Output for Service Industries, Average Annual Rate of Change (%) Annual Growth Rate

Change in Annual Growth Rate

1967-73 1973-79

1973-79 to 1979-87

-7.2

-2.1% 2.8

0.5% -6.3

2.0 1.3 6.8 4.0

- 1.7

-5.1

- 2.3

7.0 -0.7

-4.2 -0.6 - 1.1

-3.6 -3.7 0.2 -4.7

8.2

5.4

-0.1

-2.8

2.5 3.4 -0.1

0.9 3.1 -5.9

-4.3 -4.9 -3.3

-1.6 -0.3 -5.7

to

Industry (SIC)

1967-73

Transportation: Railroad transport, revenue traffic (SIC 401) 2.2% Bus carriers, class I (SIC 41 11, 413, - 3.7 41 4)(Part) Intercity trucking (SIC 4213)(part) 7.0 Intercity trucking, freight (SIC 4213)(part) 5.5 Air transportation (SIC 451 1,4521)(part) 7.4 Petroleum pipelines (SIC 46 12, 46 13) 5.1 Communications: Telephone communications (SIC 481 I ) 8.3 Electric, gas, & sanitary services: Gas & electric utilities (SIC 491, 492, 493) 6.7 Electric utilities (SIC 491, 493)(part) 8.3 Gas utilities (SIC 492, 493)(part) 3.2 Trade: Scrap & waste materials (SIC 5093) Hardware stores (SIC 5251) Department stores (SIC 531 1) 5.8 Variety stores (SIC 5331) 2.3 Retail food stores (SIC 54) 2.2 Grocery stores (SIC 541 I ) Retail bakeries (SIC 546) Franchised new car dealers (SIC 551 1) 4.7 Auto & home supply stores (SIC 5531) Gasoline service stations (SIC 5541) 5.1 Apparel & accessory stores (SIC 56) 4.7 Men’s & boys’ clothing stores (SIC 561 1) 5.4 Women’s ready-to-wear stores (SIC 5621) 5.9 Family clothing stores (SIC 5651) 6.5 Shoe stores (SIC 5661) 2.9 Furniture, home furnishings, & equipment 7.3 stores (SIC 57) Furniture & home furnishings stores 7.8 (SIC 571) Appliance, radio, TV, & music stores 6.5 (SIC 572, 573) Household appliance stores (SIC 5722) Radio, TV, & music stores (SIC 573) Eating & drinking places (SIC 58) 3.8 Drug & proprietary stores (SIC 5912) 5.5 Liquor stores (SIC 592) Finance, insurance, & real estate: Commercial banking (SIC 602) 8.1

1973-79

1979-87

0.1% -0.9

0.64

1.9 3.8

4.3 2.9 -5.2 1.1 1.4 0.3 0.3 5.8 - 1.2 4.3 0.5 5.3 4.6 4.1 4.4

-3.7 1.8 2.0 -0.6 2.7 6.3 2.0 6.0 -1.3 7.3 8.8 3.0 7.9

3.0

3.7

6.6

13.9

2. I 9.2 3.2 1.5 0.8

6.7 16.9 2.6 I .5 - 1.7

4.6

4.9

5.1

-4.4

- 0.5 2.2 1.5 0.7 0.6 -0.9 2.4

- 6.3 -0.4 -5.0 -0.6 - 1.9 1.3 -2.9

0.5 3.2 1.7 -1.7 2.0 4.1 -1.1 3.5

-2.9 -7.5 -1.0

-4.8 0.1

-0.5 -4.0

-

3.5

0.7 7.3 4.6 7.7 -0.6 -0.1 - 2.5 0.3

93

Productivity Measurement in Service Industries

Table 2.5

(continued) Annual Growth Rate

Change in Annual Growth Rate 1967-73

Industry (SIC)

1967-73

Services: 3.8 Hotels, motels, & tourist courts (SIC 701 1) Laundry & cleaning services (SIC 721) -4.6 Beauty & barber shops (SIC 7231, 7241) Beauty shops (SIC 723) Automotive repair shops (SIC 753)

1973-79

1979-87

4.3

3.4

- 3.4

- I .0

-0.2 1 .0 4.0

1.8 3.1 4.4

1973-79

to

to

1973-79

1979-87

0.5 1.3

-0.9 2.4 2.0 2.1 0.5

utilities productivity rate dropped 3.0 percentage points and electric utilities dropped 4.0 percentage points from the first to the second period. During the latest time period, 1979-87, 11 of 32 industries experienced average annual rates of decline in productivity. Also, 17 industries had slower rates of growth for this period, compared to the previous time period. Of these 17 industries, 10 had slower productivity growth in the second period than in the first. The service industry with the highest productivity growth rate for this period was again radio, television, and music stores at an impressive 9.1 percent rate. It also had the highest rate of growth in output. The industry with the slowest rate of productivity growth was gas utilities (-4.5 percent annually). This industry also had the largest decline in productivity growth from the second to the third time periods.

2.6 Comparisons As stated earlier, the BLS develops measures of output and productivity for industries classified mainly at the three- and four-digit SIC levels. The BEA develops measures of output only for industries classified at the one- and twodigit SIC level of classification. In general, it is not possible to compare output measurement results because of the different levels of coverage. In addition, the output measures developed by BEA are based on valueadded or net output; BLS prepares gross output measures. The BEA measures are developed to show the contribution by each industry to GNP (see Mohr, chap. 1, this vol.). The BLS measures are developed for the purpose of measuring productivity change^.^ However, because of data limitations, the BEA cannot always measure value-added output using a double-deflation procedure 4. There are a number of studies that suggest that gross output measures should be used for productivity measurement at the industry level; at the aggregate level a value-added measure is appropriate. In order for value-added output to be appropriate at the industry level, strong separability must hold for capital and labor inputs with all other inputs.

94

Edwin R. Dean and Kent Kunze

Table 2.6

Hours for Service Industries, Average Annual Rate of Change (%) Annual Growth Rate

Industry (SIC)

1967-73

Transportation: Railroad transport, revenue traffic (SIC 401) -2.6% Buscarriers,classI(SIC4111,413, - 2.4 4Wpart) Intercity trucking (SIC 4213)(part)* 3.3 Intercity trucking, freight (SIC 4213)(part)* 2.0 Air transportation (SIC 451 I , 4521)(part)* 2.6 Petroleum pipelines (SIC 4612,4613) - 1.9 Communications: Telephone communications (SIC 481 I ) 3.5 Electric, gas, & sanitary services: Gas & electric utilities (SIC 491, 492, 493) 2.1 Electric utlities (SIC 491,493)(part) 2.9 Gas utilities (SIC 492, 493)(part) 0.4 Trade: Scrap & waste materials (SIC 5093) Hardware stores (SIC 5251) Department stores (SIC 531 1) 3.8 Variety stores (SIC 5331) 1.1 Retail food stores (SIC 54) 1 .o Grocery stores (SIC 541 1) Retail bakeries (SIC 546) Franchised new car dealers (SIC 55 11) 1.7 Auto & home supply stores (SIC 5531) Gasoline service stations (SIC 5541) -0.2 Apparel & accessory stores (SIC 56) 1.2 Men’s & boys’ clothing stores (SIC 561 1) 1.7 Women’s ready-to-wear stores (SIC 5621) 1.1 Family clothing stores (SIC 5651) 0.1 Shoe stores (SIC 5661) 1.9 Furniture, home furnishings, & equipment stores (SIC 57) 1.9 Furniture & home furnishings stores (SIC 571) 2.6 Appliance, radio, TV, & music stores (SIC 572, 73) 0.9 Household appliance stores (SIC 5722) Radio, TV, & music stores (SIC 573) Eating & drinking places (SIC 58) 2.8 Drug & proprietary stores (SIC 5912) -0.8 Liquor stores (SIC 592) Finance, insurance, & real estate: Commercial banking (SIC 602) 5.7

1973-79

-

1979-87

Change in Annual Growth Rate 1967-73 to 1973-79

1973-79 to 1979-87

1.3%

- 7.5%

0.4

-4.4 -3.4 - 4.4 1.9 -0.9

2.8 -4.5 -4.7 -0.7 5.2

-4.8 -2.2 -1.8 0.0 -4.2

-1.1

-2.2

-2.3

-0.6 -0.9 - 0.4

-0.3 -0.5 0.0

- 1.2

-2.6 1.9 3.3

I .3

1.3%

- 6.3%

1.5 2.1 0.0

1.2 1.6 0.1

1.7 -0.2 -2.6 1.8 1.7 2.2 0.1 3.4 -4.7 2.1 -0.3 I .7 4.8 2.2

-2.0 0.3 0.2 - 2.2 2.1 2.2 2.4 0.8 I .3 - 1.8 I .5 -3.5 0.8 4.2 1.5

-4.5 0.9 -2.1 0.6 4.7 0.2

0.4 0.4 0.3 0.5 0.2 0.8 -2.1 3.O -0.6 -3.2 -0.9 -0.6 -0.7

2.3

1.8

0.3

- 0.5

1.5

1.8

- 1.0

0.3

3.4 -1.2 6.6 3.9 0.4 1.6

1.8 -0.6 3.0 3.1 1.1 -0.6

2.5

- 1.6 0.6 - 3.7 -0.7

4.0

1.6

- 1.7

- 1.5

- 4.0 -3.7 0.8 - 1.7

1.1

I .2

0.7

- 2.2 - 2.4

Productivity Measurement in Service Industries

95 Table 2.6

(continued)

Annual Growth Rate

Change in Annual Growth Rate 1967-73

Industry (SIC)

1967-73

Services: Hotels, motels, & tourist courts (SIC 701 1) 2.0 Laundry & cleaning services (SIC 721) -5.1 Beauty & barber shops (SIC 723 I ,41) Beauty shops (SIC 723) Automotive repair shops (SIC 753)

1973-79 2.8 -3.2 - 1.2 0.7 4.7

1979-87 4.2 1.1 2.4 3.3 3.9

1973-79

to

to

1973-79

1979-87

0.8 1.9

1.3 4.3 3.7 2.5 -0.8

*Labor input used is number of employees.

500

400

300

200

100

1958

1970

1980

Fig. 2.8 Productivity in transportation industries

and must rely on procedures and data that may approximate gross output changes for certain industries (see Mohr, chap. 1, this volume). For this reason and because many two-digit SIC industries contain only one four-digit industry, some of the BEA and BLS industry output measures may be roughly comparable. Of the 39 service industries published by the BLS, there are 10 that may be roughly comparable. The industries are (1) railroad transport (SIC 40);(2) bus carriers (SIC 41); (3) intercity trucking (SIC 42); (4) air transportation (SIC

96

Edwin R. Dean and Kent Kunze

Telephone communications

500

Eleclrlc utilities

400

Gas utilities 11111,11111111

300

Hotels,motels,and tourlst courts 1-11-11-11-

Laundry and cleaning services

200 I..,

100 t l 1958

I

I

I

I

I

I

1970

I

I

I

I

I

I

I

I

1980

Fig. 2.9 Productivity in telephone communications, utilities, and selected services 1I958= 100 Retail food stores

300

.......,

Franchised new car dealers

250

200

150

100

1958

1970

1980

Fig. 2.10 Productivity in retail industries with measures beginning in 1958

45); (5) petroleum pipelines (SIC 46); (6) telephone communications (SIC 48); (7) electric, gas, and sanitary services (SIC 49); (8) commercial banking (SIC 60); (9) hotels, motels, and tourist courts (SIC 70); and (10) automotive repair shops (SIC 75). Table 2.7 shows the average annual growth rates of output for the three periods and the employment coverage of the BLS measure as a percentage of BEA coverage. As is evident from the table, employment

97

Productivity Measurement in Service Industries 1967~ 100

Women'sready-to-wearstores I.......

Family clothing StOrS Shoe stores

150

-

I

I

I

1967

I

I

I

I

1

1970

1

I

I

I

I

1975

I

I

I

I

I

1980

I

I

I

1985

Fig. 2.11 Productivity in four types of apparel and accessory stores 1967~ 100 350

300 250

-

Department stores

1I

1

I

varlety stores ,.I.... I

Furnlture & i

w

o

_..-

y

11111111111111

150

-

ll 1967

*. I

I

I

1970

I

I

I

I

1

1975

I

I

I

I

I

1980

I

I

I

I

I

**.

-1

1985

'

0.

Fig. 2.12 Productivity in retail industries with measures beginning in 1967

coverage is 70 percent or less for the following four industries: bus carriers (6 percent), intercity trucking (27 percent), telephone communications (70 percent), and automotive repair shops (68 percent). For these industries comparison seems tenuous, though they often have similar output growth rates for the time periods shown. The petroleum pipeline industry is the only industry with

Edwin R. Dean and Kent Kunze

98

1972=100 225

Hardware stores

200

Grocery stores

1 75

111111111,1,,1

150 125

...'

Retail bakeries

t1

Auto & home supply stores 1-11-11-11.

---

I

Liquor stores I - - -

**-

I.........-

*......--

......I.

-.......I

100

75

t l

I

I

7972

I

I

I

I

I

1975

I

I

I

I

I

I

1980

I

I

1985

Fig. 2.13 Productivity in retail industries with measures beginning in 1972 7 9 7 2 100 ~ 140

-

-

Commercial banking Beauty and barber shops

120

-

80

-

......I I

Automotive repair shops

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

I

Fig. 2.14 Productivity in commercial banking, beauty, and barber shops, and automotive repair shops

identical BLS and BEA coverage. The output growth rates differ little for the periods shown. The differences in growth rates in the railroad industry can be attributed to different methodologies of measurement. The BEA measure of output for the railroad industry is a value-added measure calculated using the doubledeflation method. The BLS measure for this industry is a gross output mea-

99

Productivity Measurement in Service Industries

Table 2.7

Average Annual Rates of Change in Output and Employment Ratios for Selected Industries, BEA versus BLS, 1967-1987

Railroad transport 1967-73 1973-79 1979-87 Bus carriers 1967-73 1973-79 1979-87 Intercity trucking 1967-73 1973-79 1979-87 Air transportation 1967-73 1973-79 1979-87 Petroleum pipelines 1967-73 1973-79 1979-87 Telephone communications 1967-73 1973-79 1979-87 Electric, gas, & sanitary services 1967-73 1973-79 3979-87 Commercial banking 1967-73 1973-79 1979-87 Hotels, motels, &tourist courts 1967-73 1973-79 1979-87 Automotive repair shops 1967-73 1973-79 1979-87

BEA (%)

BLS (%)

SIC 40 - 1.3 0.3 -6.0 SIC 41 - 2.5 0.2 - 1.9 SIC 42 6.8 2.5 0.8 SIC 45 6.0 6.0 1.6 SIC 46 5.8 3.1 -0.5 SIC 48 8.6 6.1 5.2 SIC 49 6.9 1.8 3.0 SIC 602 5.1 3.9 1.9 SIC 70 3.2 3.2 0.6 SIC 75 6.9 4.3 3.5

SIC 401 2.2 0.1 0.6 SIC 41 1,31,41 -3.7 -0.9 -5.4 SIC 4213 pt. 7.0 2.0 - 1.3 SIC 451 1,21 7.4 6.8 5.2 SIC 46 5.1 4.0

Employment Ratio (BLSIBEA) 90.0

6.0

27.0

76.0

100.0

-0.5

SIC 481 1 8.3 8.2 4.1 SIC 491,2,3 6.7 2.5 0.7 SIC 60 8.1 4.6 3.6 SIC 701 1 3.8 4.3 2.5 SIC 753 n.a. 4.0 3.3

70.0

89.0

90.0

94.0

68.0

sure based on revenue weighted ton-miles of freight and passenger-miles.5 The same difference in methods exists for the output measures of the electric, gas, and sanitary services industry, though there is little difference in the growth rates except for the latest period. The BLS measure is gross output 5. See Robert J. Gordon (chap. 10, this vol.) for a detailed comparison of these measures

100

Edwin R. Dean and Kent Kunze

based on revenue-weighted physical quantities of kilowatt hours of electricity and total therms of gas produced. The BEA measure is a value-added measure based on deflated receipts and expenses. For the air transportation industry, much of the difference in the output growth rates can be attributable to differences in the methodology. Up until 1983 BEA measured output based on deflated sales, the BLS on the other hand, has always measured the output as a revenue weighted physical quantity measure of ton-miles and passenger-miles (see Gordon, chap. 10, this vol.). In another productivity study of the airline industry, Caves, Christensen, and Tretheway (1983), arrived at an output growth rate of 5.5 percent from 1970 to 1980. The BLS output index shows the same growth rate for this period. The differences in the banking industry output measures are again attributable to differences in method, though both the BEA and the BLS methods differ from any discussed above. The BEA measure is an extrapolation of a base-year measure. Employment of persons engaged in production is used as the extrapolator (see Mohr, Chap. 1, this vol.). The BLS measure is based on weighted volume of different type of transactions completed. For the hotel, motel, and tourist court industry the major difference is in the deflator used to deflate revenue. The BLS uses an aggregate of CPIs; the BEA uses a price index derived by a trade association. Of the ten industries where it seemed probable that comparisons could be made, BLS employment coverage is 70 percent or less of BEA coverage for four industries, bus carriers, intercity trucking, telephone communications, and automotive repair shops. For four other industries (railroad transport; electric, gas, and sanitary services; air transportation; and banking) the two agencies used different methods of output calculation. Of the remaining two industries, the two agencies’ published output series differ considerably for one, hotel, motel, and tourist courts, especially in recent periods; they are similar for the other, petroleum pipelines.

References Baily, Martin Neil, and Robert J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 2~347-31. Caves, Douglas W., Laurits R. Christensen, and Michael W. Tretheway. 1983. Productivity Performance of U.S. Trunk and Local Service Airlines in the Era of Deregulation. Economic Inquiry. 21 (July): 312-24. Gollcp, Frank M . , and Mark J. Roberts. 1981. The Sources of Economic Growth in the U.S. Electric Power Industry. In Productivity Measurement in Regulated Industries, ed. Thomas G . Cowing and Rodney E. Stevenson, 107-43. New York: Academic Press. Jablonski, Mary, Kent Kunze, and Phyllis F. Otto. 1990. Hours at Work: A New Base for BLS Productivity Statistics. Monthly Labor Review 113 (February): 17-24.

101

Productivity Measurement in Service Industries

MXK, Jerome A. 1988. Measuring Productivity in Services Industries. Technology in Services: Policies for Growth, Trade, and Employment ed. Bruce R. Gulie and James Brian Quinn, 139-59. Washington, D.C.: National Academy Press. Nelson, Randy A., and Mark E. Wohar. 1983. Regulation, Scale Economics, and Productivity in Steam Electric Generation. International Economic Review 24 (February): 57-79. Siege], Irving H. 1961. On the Design of Consistent Output and Input Indexes for Productivity Measurement. In Output and Input Productivity Measurement, 23-46. NBER Studies in Income and Wealth, vol. 25. Princeton, N.J.: Princeton Univ. Press. U S . Department of Labor. Bureau of Labor Statistics. 1988. BLS Handbook of Methods. BLS Bulletin no. 2285, pp. 78-87. Washington, D.C.: Government Printing Office. . 1990. Productivity Measures for Selected Industries and Government Services. BLS Bulletin no. 2349. Washington, D.C.: Government Printing Office.

Comment

W Erwin Diewert

Edwin R. Dean and Kent Kunze are to be congratulated for presenting a clear exposition and discussion of the BLS’s program of productivity measurement. The focus of my comments will be on the bureau’s annual industry productivity measures, which utilize gross output and hours of labor input information for 173 industries. I have four major criticisms of these productivity measures: My first criticism is that these productivity measures are labor productivity measures and hence that they may be very imperfect indicators of changes in the industry’s total factor productivity. Total factor productivity measures are much more useful than labor productivity measures, and I will now attempt to explain why this is so. Each firm in an industry produces outputs and utilizes many inputs. A rough classification of a firm’s outputs and inputs into different broad categories could be made as follows: (1) sales or gross outputs; (2) purchases of materials and goods; (3) purchases of business services; (4) leasing of capital services; (5) labor inputs; (6) capital input: machinery and equipment; (7) capital input: computers; (8) capital input: structures; (9) capital input: inventories; (10) capital input: land and natural resources; (1 1) capital input: R&D stock and patents; (12) capital input: marketing, trademarks, and advertising; and (13) capital input: human capital and the skills of the firm’s workers. Categories 2-4 are intermediate input categories; 5-1 3 are primary input categories. Categories 6-10 are the traditional physical capital input categories (although the current system of national accounts ignores the contribution of land); 11-13 are the intangible capital input categories. The total factor productivity of a firm (or industry) going from period t - 1 to period t can be W. Erwin Diewert is professor of economics at the University of British Columbia and a research associate of the National Bureau of Economic Research.

102

Edwin R. Dean and Kent Kunze

defined as (Y,/Y, l)/(XJXr.l)where Y, is the firm’s real value added in period t (an aggregate of the quantities in categories 1-4, above, where intermediate inputs are indexed with a negative sign in the index number formula) and X, is the firm’s quantity of primary input utilized in period t (an aggregate of the inputs in categories 5-13, above). Measuring the total factor productivity of a firm, industry, or economy is a useful exercise because it gives us some indication of how much “free” output per unit of primary input was generated by the firm, industry, or economy going from period t 1 to period t . The gross output divided by labor productivity measures generated by the BLS can be defined (approximately) as (Q/Q,.,)/(L,/L,.l), where Q, is the gross output produced by the industry under consideration in period t and L, is the corresponding period t labor input (measured in unweighted man-hours). Thus the BLS labor productivity measures utilize information on only 2 of the 13 listed above: categories 1-5. Thus the BLS labor productivity measures generally do not closely approximate the total factor productivity measures (which should utilize information on all 13 categories of inputs and outputs). My second criticism of the BLS labor productivity measures is that they are biased (compared to the labor productivity measures used by other countries) and that they do not aggregate properly over firms and industries. The problem is that the BLS measures use gross output instead of real value added. In the last two decades, business services (and imports) have been growing faster than gross output. Hence, for the economy as a whole, real value added has grown more slowly than gross output. Thus the BLS labor productivity measures tend to be biased upward compared to labor productivity measures calculated in most other industrial countries that use real value added in place of gross output. Because the labor productivity measures calculated in most other industrial countries use real value added in place of gross output, the U.S. labor productivity measures are not comparable with the labor productivity measures calculated by other countries. The use of gross output instead of value added also leads to difficulties in aggregating the BLS labor productivity measures over firms or industries in a consistent manner. These aggregation difficulties are not discussed here: the reader is referred to the papers by Domar, Hulten, and Diewert.’ My third criticism of the BLS labor productivity measures is more technical and has to do with the BLS choice of index number formula for measuring gross output. Following Dean and Kunze, define qir as the gross output of product i in year t , define L,, as the aggregate number of labor hours spent in producing product i in year t and define e, = L J q , as the unit labor require1. W. E. Diewert, Aggregation Problems in the Measurement of Capital, The Measurement of Capital, ed. D. Usher, vol. 4 5 , NBER Studies in Income and Wealth (Chicago: Univ. of Chicago Press, 1980); E. D. Domar, On the Measurement of Technological Change, EconomicJournal71 (1961): 709-29; C. R. Hulten, Growth Accounting with Intermediate Inputs, Review ofEconomic Studies45 (1978): 51 1-18.

103

Productivity Measurement in Service Industries

ments of product i in year t . Dean and Kunze define their labor productivity measures as the left-hand side of the following equation:

Cl~lOqlt~Cl~l,qI, = C,(L,dq,Jq,/Xz(4,~qzJq,, = C1(qtr/q1o) Lc2iL!t = [C,(s,,/e,)(L,dCS.p)l / [CC,,/C,L@I.

In the numerator of the right-hand side of the above equation, the ith output growth rate qJq,,, is weighted by the ith labor share for period 0, L,dC,L,. In the denominator, we have the growth rate in unweighted labor hours going from period 0 to period t, C&,,&+fl. I am unaware of any discussion in the theoretical index number literature that justifies the use of such a strange output index. Even Irving Fisher did not consider such an odd index number formula.2 Thus my third major criticism of the BLS labor productivity measure is that the choice of index number formula used to calculate the gross output growth rate is totally unconventional and leads to the U.S. gross output, labor productivity measures being noncomparable with the labor productivity measures computed by other countries. A related criticism of the choice of index number formula by the BLS is that it is usually very difficult to compute exactly how many hours of labor L,( were required in period to produce the corresponding amount of the tth output in period 1. My final criticism of the BLS labor productivity measures concerns the way the data are collected from different sources. Often, the data or sales are collected from one source of survey, the data for price deflators from another source, and the data or man-hours from a third source. The end result is that the output data do not actually correspond to the labor input data. Thus the resulting labor productivities could be seriously biased, depending on sample sizes and the intersection of the survey frames. The cure for this problem is easy to state (but probably difficult to implement): instead of having 13 separate business surveys to collect data for each of the 13 major categories listed above, there should be a single business survey that collects price and quantity information on all 13 categories of outputs and inputs. The statistical unit to be samples should be the firm (or establishment), and comprehensive economic data should be collected for all of the inputs and outputs that the firm produces and utilizes. My conclusion is that the various U.S. statistical agencies (BLS, BEA, and the Census Bureau) should cooperate in the construction of comprehensive total factor productivity measures. It is simply too wasteful to have independent and unrelated measures of productivity. In fact, I think that the time is ripe for the creation of a comprehensive U.S. statistical agency. “Statistics USA” has a nice ring to it. 2. I . Fisher, The Making of Index Numbers (Boston: Houghton Mifflin, 1922).

104

Edwin R. Dean and Kent Kunze

Reply

Edwin R. Dean and Kent Kunze

W. Erwin Diewert’s comments on our paper and the BLS’s productivitymeasurement program suggest that the bureau’s labor productivity measures suffer from important shortcomings. As we had expected, Diewert’s comments proved stimulating and provocative. We have responses to each of his four major criticisms: First, he states that the BLS labor productivity measures are very imperfect indicators of changes in an industry’s multifactor productivity because they use only 2 of 13 possible inputs and outputs. Although he rightly points out the advantages of multifactor productivity measures relative to labor productivity measures, he underestimates the value and usefulness of these measures. As stated in our paper, the industry labor productivity measures form only one part of the bureau’s productivity measurement program. Although the bureau continues to increase the number of industry labor productivity measures, since the late 1970s the main focus has in fact been on developing multifactor measures both for industries and for the economy as a whole. The bureau has developed annual measures of multifactor productivity for all the two-digit SIC industries within manufacturing, five three-digit industries in manufacturing, and one four-digit industry in services (railway transportation). The first two of these measures were introduced in 1984, and all provide annual data from 1958 (or earlier) forward. The measures include all but the last three inputs on Diewert’s list of input and output data, though some of the measures also include energy, a component of production he did not mention. In addition, the bureau has developed measures of the effects of R&D expenditures and changes in labor skills on productivity growth for the economy. And it continues to publish major-sector multifactor productivity measures, first published in 1983, on a value-added basis. Between 1983 and the very recent past, the bureau was the only government statistical agency in the world with an ongoing program of multifactor productivity measurement. (Statistics Canada has recently joined us with some preliminary results.) The bureau’s emphasis on multifactor measures does not mean that labor productivity measures have no merit. In fact, as the bureau increases its multifactor coverage, we find that the trends and changes in the industry labor productivity measures often reflect the trends and changes in the industry multifactor productivity measures. Further, the labor productivity measures provide useful information in their own right and can be prepared with fewer resources and less developmental time. Second, Diewert states that the BLS labor productivity measures are biased and do not aggregate properly over firms and industries. It appears that the bias Diewert has in mind relates mainly to a comparison that might be made between the bureau’s gross output measures and value-added measures produced by other countries. Certainly gross output measures should not be com-

105

Productivity Measurement in Service Industries

pared to value-added measures. But, just as certainly, the bureau should not be choosing an output measure based mainly on the criterion of comparability with foreign countries. We should be concerned mainly with the appropriateness of gross output for our purposes. Finally, the issue of comparability with other countries is not a weighty one. We know of only two other countries that regularly produce industry productivity measures at a three-or four-digit level and in both cases the output measure is not a pure value-added measure. The bureau never intended for the detailed industry productivity measures to be aggregated. Although a common methodology is applied, each industry measure is tailor-made: each is produced using the best data available for that particular industry and the specific measurement techniques appropriate for that data. It would, therefore, be inappropriate to aggregate these measures. Suppose, on the other hand, that someone were to wish to aggregate these measures. As Diewert correctly points out, there are difficulties in aggregating industry productivity measures, but his statement that the use of gross output rather than value added is the source of the difficulties is questionable. Ifone were to aggregate industry multifactor productivity measures, there are reasons to prefer gross-output-based measures. Domar showed how grossoutput-based multifactor productivity measures can be aggregated. I Domar also stated that industry productivity measures should be developed using gross as opposed to net output. Domar’s conclusions have been strengthened by Hulten, who showed how major-sector multifactor productivity measures can be aggregated from gross-output-based industry productivity measures in the context of a flexible production formula, and by Gollop, who set forth the case that industry multifactor productivity measures should be based on gross output.* We would be the first to agree that the conclusions regarding the preferability of gross output industry multifactor measures may not carry over to labor productivity measurement, which is one of the reasons why BLS has generally published measures only for those industries that have not experienced strong changes in the ratio of value added to gross output. The bureau does produce value-added-based measures of productivity at the more aggregate levels. The results do not support Diewert’s statement that gross output labor productivity measures, compared to value-added measures, tend to be biased upward: from 1979 to 1988, the value-added measure of output increased faster than the gross output measure for the total U.S. manufacturing sector (a 3.1 percent annual rate for value added compared to a 2.1 percent annual rate for gross output). Third, Diewert states that the index number formula used by BLS for out1. E. D. Domar, On the Measurement of Technical Change, Economic Journal 71 (1961): 710-29. 2. Charles R. Hulten, Growth Accounting with Intermediate Inputs, Review ofEconomic Studies 45 (1978): 511-18; Frank M. Gollop, Growth Accounting in an Open Economy, Working Papers in Economics (Boston: Boston College, 1981).

106

Edwin R. Dean and Kent Kunze

put measurement has not been discussed in the theoretical index number literature and is unconventional. However, Irving Siege1 discussed this formula in several papers, including one paper published in an Income and Wealth Conference volume, cited in our paper. Finally, Diewert states that the data for the bureau’s industry labor productivity measures are collected from different sources and that therefore the output and input data do not correspond to one another. However, the fact that the bureau usually uses data collected from different surveys is not the hindrance to accuracy that Diewert suggests it might be. First, it is not the case that there are 13 separate business surveys to collect data for each of the 13 major categories listed by Diewert. Second, as long as the surveys use the same universe for constructing the sample frames and appropriate weights are used for each sampled unit, consistent estimates can be made. The sample unit for the bureau’s surveys of employment and producer prices, as well as the Census Bureau’s survey of output and employment, is the establishment. The great majority of the BLS industry productivity measures are constructed mainly with these data sources. There is no evidence that the labor productivity numbers are seriously biased because of sample sizes and intersection of sample frames as Diewert’s statements suggest.

COmment

Robert E. Lipsey

Several measurement issues that are passed over very lightly in Edwin R. Dean and Kent Kunze’s paper, by their simply describing the methods the BLS prefers, deserve more extensive consideration. The first of these is the preference for the use of physical quantities in preference to deflated values in the construction of quantity indexes. Because neither prices nor physical quantities are obtainable for all the products of most industries, any choice between them involves some assumption that omitted products behave in the same way as do covered products. If the sample of quantity changes is used, the assumption is that quantity changes for uncovered items move identically, on average, with those of covered items. If a sample of price changes is used, the same assumption is made for uncovered prices. If the degree of coverage were identical and the precision of product specifications were identical for price and quantity information, the choice would rest on the variance of price changes as compared with that of quantity changes. I would guess that the variance of price changes is smaller in most cases, and the preference of the BLS for quantity data needs some justificaRobert E. Lipsey is professor of economics at Queens College and the Graduate Center of the City University of New York and a research associate of the National Bureau of Economic Research.

107

Productivity Measurement in Service Industries

tion. Users of the data would probably like to see a comparison of the results of the two methods. Another decision that is not discussed is that no attempt is made to adjust for changes in the quality of labor inputs. Changes in the composition of the labor force must be important, for example, in retailing, where inexperienced and part-time workers have come to represent a much larger part of employment. The composition of the labor force in banking also may have changed with the increased use of computers. The use of employment weights may be appropriate when the emphasis is on measuring labor productivity, particularly if the main purpose of the measurement is to analyze changes in the demand for labor. Once the focus is on efficiency, as with multifactor productivity measures, that weighting no longer appears to be an obvious choice. For example, it must give very little weight to highly computer-intensive operations that may be the major growth areas in some service industries, such as banks. At least some discussion of the justification for this particular weighting scheme is needed.

This Page Intentionally Left Blank

3

Improvements in Measuring Price Changes in Consumer Services: Past, Present, and Future Paul A. Armknecht and Daniel H. Ginsburg

A major shift in the spending patterns of American consumers has occurred in the past 25 years. At the end of 1963, consumer services represented 34 percent of the market basket in the consumer price index (CPI) for urban wage earners and clerical workers (CPI-W). As of December 1989, consumer services represented 5 1 percent of the CPI-W. When the CPI for all urban families (CPI-U) was introduced in 1978, consumer services had a relative importance of 37.8 percent. At the end of 1989, they represented 55 percent of the CPI-U. There has been a shift from a predominantly commodity-oriented market basket to one that is services oriented. Over the past 25 years substantial progress has also been made in the measurement of price change in the services sector of the CPI. These improvements have included the timeliness and availability of data that measure expenditures on consumer services as well as the sources and methods used in measuring price movements. This paper presents a summary of the major changes that occurred during this period that affect the measurement of consumer services, the improvements currently underway, and some future improvements that are being considered for implementation with the next CPI revision. In part 3.1 we discuss improvements made when new surveys, new methodologies, new procedures, and new coverage were introduced into the CPI. These improvements include the availability of continuous data on consumer expenditures, continuous updating of outlets where consumers shop and the products and services they purchase, probability sampling within the retail Paul A. Armknecht is assistant commissioner for consumer prices and price indexes with the Bureau of Labor Statistics and member of the Conference on Research in Income and Wealth, National Bureau of Economic Research. Daniel H. Ginsburg is chief of the services section, Division of Consumer Prices and Price Indexes, with the Bureau of Labor Statistics.

109

110

Paul A. Armknecht and Daniel H. Ginsburg

outlet, the rental equivalence approach to home ownership, and the expansion of coverage to include more service items. In part 3.2 we summarize studies that are currently under way for measuring price change among service items. These include the use of hedonic regression models to improve price measurement and quality adjustment, new pricing procedures for tenants’ and auto insurance, and methods for handling quality changes in medical care services. In part 3.3 we present two areas for future improvements in measuring consumer services: The first area is an alternative approach to measuring price change for medical care by pricing treatments rather than specific services. The second improvement would be shifting to a flow-of-services approach for private transportation services.

3.1 CPI Survey Methods and Improvements during the Past Tiventy-Five Years 3.1.1

Conceptual Framework for the CPI

The CPI and the COL Index

The underlying theory of the CPI can resolve the conceptual and operational issues that arise in the course of developing and producing a price index. Pollak (1989), Gillingham (1974, 1983) and Gillingham and Lane (1982) have argued that the theory of a cost-of-living (COL) index should guide the CPI. The CPI is a modified Laspeyres index using a market basket of goods and services based on consumer spending patterns at some point in the past (called the base period). It measures the cost of purchasing the baseperiod market basket today. A true COL index measures the minimum cost of purchasing a market basket today that yields the same level of satisfaction as the market basket purchased in the base period. The major difference between the CPI and a COL index is that the COL index allows for the fact that consumers substitute products or services within the market basket in response to price change. For example, if the price of beef products doubles relative to poultry products, a COL index allows for the consumer to substitute chicken for beef so long as they maintain the same level of overall satisfaction. When this substitution occurs, consumers’ costs of purchasing the new market basket are lower than if they had purchased the original amounts of beef and chicken. The COL index holds the standard of living constant and allows both prices and quantities to vary. The CPI holds the standard of living constant by keeping quantities fixed and allowing only prices to vary. Because of this fact, the CPI provides an upper bound to the COL index. We can use the COL conceptual framework as a guide for the CPI. In this framework for the CPI, a consumer’s welfare is determined by the flow of consumption services received, where such services can be (1) directly pro-

111

Measuring Price Changes in Consumer Services

vided; (2) obtained coincidentally with the consumption of a nondurable good (in which case, the distinction between a consumption good and a consumption service is unnecessary); or (3) obtained from the use of a durable good owned by the consumer. In each case, satisfaction is derived from the act of consumption. Items that provide consumption services over several time periods are durable goods or assets. Purchasing such goods is not consumption; rather, it is investment, that is, the purchase of assets that provide consumption services over many time periods. Within this framework, the CPI should measure the change over time in the cost of the market basket of consumption goods and services consumed in the base period. For the consumption services provided both from directly purchased services and from nondurable goods, this implies observing market prices and transaction levels in the base period, as well as the subsequent time path of market prices. However, for the services provided by durable goods owned by consumers, the implicit price of such services must be estimated because market transactions do not take place each time the services are consumed. For many goods whose life expectancy exceeds one month, the cost of investment in the good during this time period is a reasonable proxy for the consumption of the goods’ services during the same period. If the life of a durable good is not too long, if the secondary (used) market and the rental market for the good are not well organized, if the good depreciates fairly rapidly, or if the good is not a very big part of consumption spending, then it may be acceptable to treat the good itself, rather than the service it provides, as the item in the consumer’s utility function and, therefore, as an appropriate item in the CPI. Because for many goods there is no real market for the services they provide (you cannot rent shoes), as a practical matter, there really is not any choice. However, this is not the case for all durable goods, including housing units and automobiles. You can rent them and resell them; they do not depreciate rapidly, and, however they are measured, they are an important element of consumer spending. Maintaining Constant Quality

The CPI measures the average change in prices paid by urban consumers for a fixed market basket of goods and services of constant quality. The fixed market-basket approach would call for measuring price changes for the same services originally chosen for the sample. Doing this through time, however, is frequently impossible. Some services, like passenger train service between two cities, disappear altogether from the market; others, such as intracity mass transit, are improved when a subway system replaces bus service. New services, such as video rental, emerge. If all improved and new services were exact replacements for the original and disappearing services in terms of providing consumers with the same level of satisfaction, there would be no problem with comparing the price of the new service with the old when measuring

112

Paul A. Armknecht and Daniel H. Ginsburg

price change. However, most new and replacement services provide consumers with levels of satisfaction different from the original. Some offer improved quality; others offer reduced quality. Thus, when a service disappears, it presents a problem for the construction of price indexes because a new or improved service may offer a differing level of quality and compromise the basic underlying assumption of measuring price change for services of constant quality. A variety of methods are used to handle the quality issue when substitute services occur. These are discussed in detail in Armknecht and Weyback (1989). When the quality of a substitute service is different from that of the previous item, an estimate of the quality difference must be made. In the CPI, the quality component in each substitution is handled in accordance with the information available. There are three principal methods for quality adjustments. The first is to estimate the quality change by observing the difference in market prices between the old and new services at the same point in time. For example, a transportation company offers a new service, y, to replace eventually a current service, x. After y is introduced, we can observe the market prices of x and y The current period is our overlapping month in which we collect prices for both services. The price change in the current period is computed as the price change in service x from its previous price. In the next period, we will collect only the price for service y, and the price change will be measured by the change in the price of y. The current period, where we have overlapping prices for the two services, provides the estimate of quality change. The quality change is the difference in the market price betweeny and x. This method of quality adjustment is referred to as overlap pricing. The second method for handling quality change is direct quality adjustment. Ideally, we would like to obtain the market value of the quality difference between an item and its substitute. Then we can adjust the price of the old item directly for the quality change and compare the current price with the quality-adjusted price to measure price change. The market value of the quality difference is not frequently available. If suppliers of the service are cooperative, they may be willing to provide information on the value of the quality change analogous to information provided annually to the Bureau of Labor Statistics (BLS) by the domestic automobile producers at the time of the model changeover. It may also be possible to estimate the quality change from other sources, such as component services for which prices were previously available, or by using statistical techniques such as hedonic regression models. In most cases of quality adjustments, the method used is an imputation procedure called linking. In this method no comparison is made between the price of the new and the previous service, but rather an imputed price change is used. The imputation for the price change is made using the average change of other similar services available during the comparison period. This method

113

Measuring Price Changes in Consumer Services

assumes that the underlying price change for the service not available would have been the same as those that are available. The quality difference between the two services is the observed price change in the two services less the amount of the imputed price change. 3.1.2 Weighting and Item Structure The market basket used for the CPI is based on the spending patterns of the urban population. Historically, revisions of the CPI market basket were based on one-time consumer expenditure surveys. For the 1964 CPI revision a consumer expenditure survey relating to spending patterns for 1960-61 was used. For the 1978 revision 1972-73 expenditure patterns represented the market basket. Beginning in 1979 a new program for conducting ongoing consumer expenditure surveys was instituted. Since then, it has been possible to monitor shifts in spending patterns on a regular basis to determine whether significant changes are occurring. The market basket for the 1987 revision of the CPI was the average of expenditure data from the 1982-84 surveys. We are currently in the process of developing new expenditure weights for the CPI from the 1986-88 surveys. These will be used to evaluate shifts in the CPI market basket. We will also test a reweighting of the CPI to measure the effect that the use of the most current market basket would have on price changes. If warranted, we could introduce an updated market basket for the CPI between revisions in 1992 or develop an experimental index using the more current weights. The item structure for the CPI consists of 184 item strata in which prices are collected for commodity groups such as white bread, carbonated drinks, and boy’s apparel. Within each item stratum, one or more substrata, called entry-level items (ELIs), are defined. There are 364 ELIs that are the ultimate sampling units for items selected in CPI areas. These represent the level of item definition at which BLS field staff begin item sampling within each outlet. For a detailed discussion of the sample design the reader is referred to chapter 19 of the BLS Handbook ofMethods (1988). The consumer expenditure survey (CE) provides the sample weights to implement item sampling. The CE consists of a quarterly interview survey and a diary survey. The interview survey collects inventories of items held by the respondent and expenditures for a full year on major consumer items (e.g., vehicles, durable goods, insurance policies); the diary survey records every purchase made during a two-week period by any member of the family. Each expenditure recorded in the CE is coded to one of the 364 ELIs used in the CPI classification structure. Estimates of annual expenditures for each item stratum by area are then produced. The average of these estimated annual expenditures for the 1982-84 period constitutes the expenditure weights currently used in the U. s. CPI. The CE also provides information for selection of items in the CPI sample rotation process whereby the entire sample is replenished on a five-year cycle.

114

Paul A. Armknecht and Daniel H. Ginsburg

The CPI consists of samples for 85 urban areas. Because of their size the 29 largest areas are self-representing; the remaining 56 are probability selected to represent medium- and smaller-size metropolitan areas and nonmetropolitan urban areas. To enable the CPI to reflect changes in the market place, item and outlet samples are reselected each year for 20 percent of the areas (about 17 cities) on a rotating basis. We will discuss briefly the item selection, and in the next section present the outlet selection process. The number of quotes assigned to each item stratum was established at the time the sample was designed by an optimum allocation model that used nonlinear programming techniques. (See Leaver et al. 1987.) The factors used in this allocation of quotes were the relative importance of the item-stratum expenditures to total consumer expenditures, the contribution of the item stratum to the variance of price change and the cost of data collection, constrained by the amount of budget. Each year, four regional universes of consumer expenditures (Northeast, North Central, South, and West) are tabulated for the 364 ELIs from the two most recent years of CE data. The regional universe is used to select an independent sample of ELIs within each item stratum in the region. The number of quotes assigned to each ELI is accomplished by using probabilityproportionate-to-size (PPS) sampling technique, where the probability of selection is the relative importance of the ELI expenditures within the item stratum. Figure 3.1 provides an example of ELI sample selection for the information processing item stratum at two different times. For the cities that had their samples rotated in 1984, the 1982-83 CE data were used. The typewriters and calculators ELI had the highest relative importance within the item stratum, followed by home computers. Therefore, typewriters and calculators ELI

Probability of Selection 1984 1987 (1982-83CE Data)

Home Computers Home Computer Software Telephones Typewriters, Calculators Other Processing Equipment

(1984-85 CE Data)

.25055 .13491 .13326 .27819 .20308

.34792 .18734 .19619 .21018 .05835

Probability of Selection is calculated as: ELI Total Expenditure divided by Item Stratum Total Expenditure Fig. 3.1 Sample Selection Example: Information-processingequipment for region-South

115

Measuring Price Changes in Consumer Services

had the greatest chance of selection. For the 1987 sample-rotation cities, these ELIs reversed positions, when more current expenditure data were used, with home computers having the highest probability of selection. Thus, the ELIs selected for area samples may change on the basis of their importance in the ongoing CE. 3.1.3 Selection of Outlets The second survey used for the CPI is the point-of-purchase survey (POPS), which determines the retail outlets from which consumers purchased goods and services. Since 1977 the POPS has been conducted annually in the 17 or so areas that are scheduled for sample rotation. The POPS is a household survey conducted over a four- to six-week period, usually beginning in April. Respondents are asked whether certain categories of items were purchased within a specified recall period. The recall period varies depending on the type of items purchased. Commodities and services are grouped into sampling categories (called POPS categories) based on ELIs. Some POPS categories consist of only one ELI; others are combinations. ELIs are combined into a single POPS category when certain types of commodities or services are generally sold in the same retail outlets. For example, POPS category 106, meat and poultry, consists of eight beef ELIs, six pork ELIs, four ELIs for other meats, and three poultry ELIs. These are combined because an outlet that sells beef also tends to sell other meats. For each category the respondent in the household is asked about purchases made within the stated recall period, the names and locations of places of purchase, and the expenditure amounts. The expenditure information by outlet within POPS category is tabulated for the city and used to draw a sample of outlets. Each POPS category within a geographic area is designated a number of outlets based on an optimum allocation design for minimizing variance and cost. Each outlet’s expenditures within a POPS category are used to compute the probability of selection. The sample of outlets is drawn using probability proportionate to size (PPS) procedures to select the designated number of outlets for each POPS category. Because the item sample based on ELIs from the CE and the outlet sample based on POPS categories are selected in separate processes for each geographic area, they must be merged before the sample can be finalized. A concordance exists to map each ELI to a POPS category. Sample ELIs are assigned to the outlets selected for the corresponding POPS categories. The number of price quotes assigned for an ELI in each outlet is equivalent to the number of times the ELI was selected for the region in the item sampling process. For example, video games, home computers, and home computer software are the three ELIs in POPS category 31 1 (electronic equipment for nonbusiness use). If the home computers ELI is designated to receive two quotes per outlet in item sample selection and home computer software and video games are each designated one quote, then each outlet for POPS category 3 11 has two quotes for home computers, one quote for software and one

116

Paul A. Armknecht and Daniel H. Ginsburg

quote of video games. On the basis of the optimum allocation model for the CPI sample design, POPS category 3 11 is only designated one outlet per area. Therefore, the sample sent to the field for collection in this geographic area will only have one outlet with four quotes. If the outlet selected does not sell products that fall within the definition of one of the ELIs, then that price quote is dropped from the sample. Thus, the number of price quotes assigned for collection in a sample outlet is determined through an itedoutlet sample merge. In the outlet sample process, an outlet may be selected more than once for a given POPS category, provided the expenditures reported for the outlet are large. An example of this is a major grocery store in a small nonmetropolitan urban area with a very large number of consumers. The outlet may also be selected for more than one POPS category, for example, a major department store. If an outlet is selected multiple times for a given POPS category, the same multiple of price quotes is assigned for collection for each sample ELI matching the category. If an outlet is selected for more than one POPS category, price quotes are assigned for collection for all sample ELIs matching within each of the POPS categories. Through this process new outlets are identified and incorporated into the CPI sample. For more details on this process the reader is again referred to the BLS Handbook of Methods (1988). To the extent that the ELI definitions are fairly broad, new products and services that emerge and fall within the current ELI definitions are eligible for inclusion in the sample. Products and services that do not fit within the existing ELI scheme cannot be included until new categories are defined at the time of the next CPI revision (Marcoot 1985). This was the case with home computers and software that were introduced in the early 1980s. They did not fit within the existing classification of CPI categories. During the 1987 revision of the CPI attempts were made to keep the ELI definitions broader than in the past so that as new products and services emerge they could be incorporated into the CPI sample. 3.1.4

Selection of Service Items for Pricing in the CPI, 1964-1987 Revisions

In the 1964 revision of the CPI all items purchased for consumption by urban wage and clerical households should have been eligible for selection in the market basket of goods and services priced. However, this was not possible. Only those items that could “be described reasonably definitively, and were bought and sold with some degree of regularity and in sufficient quantities to have a measurable price,” were given an opportunity to be in the sample of priced items. The determination of the specific items that were selected for measuring price change started with an analysis of information that was available from the 1960-61 CE. The expenditure data reported in the CE were categorized into 52 expenditure classes (ECs) that were set up, in a general way, to serve similar human

117

Measuring Price Changes in Consumer Services

needs, for example, furniture, fuels and utilities, apparel services, public transportation, and professional medical services. Within this EC framework, 8 12 items were identified as making up the universe from which the final item selection would be made using PPS methodology. The final sample contained 309 different items. In many instances these sampling frame items consisted of a fairly welldefined commodity or service, for example, train fares, and thus no additional selection was needed before moving on to the next stage, developing a detailed specification for use in pricing. In other cases, where the selected item contained a number of related, yet distinct subitems, further item selection, again using probability, took place, e.g., psychiatrist selected from the item category other medical specialist. The technique used in collecting price data for the CPI since 1934 has been to describe the commodity or service in such physical detail that a BLS field representative can identify it in subsequent months from among other similar items. In this way, BLS attempts to maintain constant quality of goods priced. A specification, therefore, contains a description of the physical characteristics of an item that are judged to determine its quality and influence its price. Specifications also may include features that aid in identifying the item, such as model numbers. In practice, the process of setting up the specifications for the 1964 revision involved identifying the volume selling unique item to represent the price movement for the selected item category, for example, pricing the “usual fee for an office visit to a general practitioner or internist for a regular patient” and “usual fee for a house visit, during the day, within the corporate limits of the city, to a regular patient,” to represent family doctor, office and home visits. These specifications frequently were developed after consultation with manufacturers, retailers, associations, and trade journals, and also drew from staff economists’ knowledge gained through extensive market studies of the various ECs that made up the CPI. Although an attempt was made to maximize the use of national specifications, this was not always possible. Climatic conditions or regional preferences necessitated deviations. For example, in Honolulu furnace repair, the charge for repair of automatic gas safety pilot in a single-family-occupied home or for repair of a pressure atomizing-type oil burner unit in a single family home could not be priced, and thus a waterheater replacement was priced instead to represent the heating-equipmentrepairs item. The forced selection of volume selling unique items for pricing in the CPI enabled BLS to publish detailed service indexes, representing very specific items. For example, under physician fees BLS published seven item indexes: (1) family doctor, office visit; (2) family doctor, home visit; (3) psychiatrist office visit; (4) herniorrhaphy, adult; and ( 5 ) tonsillectomy and adenoidectomy; (6) obstetrical care; and (7) pediatric care; office visit. Users of CPI data could, therefore, specifically track the price movement, over time, of these

118

Paul A. Armknecht and Daniel H. Ginsburg

seven specific physician services. However, since there are hundreds of services performed by physicians, the overall measurement of the price movement of physician services left much room for improvement. This was especially true because the CPI prices health insurance policies indirectly, and the priced physician services are used to move not only out-of-pocket expenses not covered by insurance but also the physician services part of health insurance. (See section below covering health insurance in the CPI). Using the volume selling item within an item category to represent the price movement for that item, meant that the CPI essentially was measuring price change for only a narrow, albeit frequently sold, group of unique items. Within the marketplace a vast array of items are sold, with different degrees of frequency. It, thus, seemed clear that broadening the selection criteria from just volume selling items would greatly improve the representativeness of the index, especially if the selection could be done at the sample outlet using probability sampling procedures, based on actual sales data. In this way, both frequently and infrequently purchased items would have a proper proportional chance for selection. In summary, the preselection of volume selling unique items for use in measuring price change in the CPI enabled BLS to publish very specific item indexes, but at the cost of not being able to include in the index measures of a very broad range of items to more closely represent the array of unique items actually purchased by the index population. If the price change for volume selling items was different from other items, a potential bias could result. The major emphasis for the 1978 revision was to introduce probability sampling methods for the selection of unique items for pricing within the outlet. For the 1978 revision seven major product groups-food and beverages, housing, apparel and upkeep, transportation, medical care, entertainment, and other goods and services-were subdivided into 68 ECs, which in turn were split into 265 item strata. Within each stratum one or more substrata, called ELIs, were defined, yielding 382 ELIs. As mentioned before, these ELIs became the ultimate sampling units for items and were used by BLS field staff as their initial level of item definition within an outlet. A multistage probability sampling process that BLS calls disaggregation was used to make the final selection of unique items for pricing in the CPI. Using this methodology all goods and services within an ELI were, ideally, given a chance for selection in proportion to their dollar sales in each outlet. The disaggregation process was designed to make the probability of selection of a unique item proportional to the dollar sales of the item in the outlet. As the number of items eligible for selection was progressively narrowed in each stage of this process, measures of sales were assigned to each of the groups identified. There were four methods available in measuring the sales of a group: percent of dollar volume sales, ranking, shelf space, and equal probability. The preferred method for measuring the sales of a group was the percent of

119

Measuring Price Changes in Consumer Services

dollar volume sales, that is, the percent that a specific group represents of the total dollar sales of all the groups listed in a specific stage of disaggregation. This method relies on the knowledge of the respondent of the dollar sales of the groups formed in the first stage of the disaggregation process. The respondent often refers to sales records or, when that was not feasible, provides estimates for the percent of sales for each member of the group. The time period on which these sales data were to be based was the 1Zmonth period prior to the time of initiation. (Certain exceptions existed for the 40 ELIs, such as women’s suits, that were identified by BLS as being potentially seasonal in at least one of the four primary census geographic regions. In such situations, the reference period related to each of two identified seasons that, when combined, covered the entire year.) If the total of the percentages provided by the respondent fell outside the range of between 95 percent and 105 percent, the respondent was asked to reexamine and adjust the values. The ranking procedure was used when the respondent could not provide specific percents of dollar volume sales for listed groups. Ranking required an ordering by the respondent of the identified groups from largest to smallest in terms of dollar volume sales. Specific percents were then assigned to each of the groups using a ranking table. (See appendix A, Disaggregation Sheet, Form 3400A.) The shelf-space procedure was used to estimate relative dollar volume sales when the respondent could not provide either percent of dollar volume sales or ranks. If the amount of shelf space for display of the units for selection was roughly comparable to the dollar volume sales for those units, and if those units on display accounted for at least 80 percent of the total ELI sales in the outlet, then the shelf-space method was eligible to be used. The shelf-space method rook into account both display space and unit price. The display volume times the unit price was used as the approximate proportional equivalent of dollar volume sales. The equal probability procedure had to be used if the respondent could not provide the percent of dollar volume sales, could not rank the relative importance of the groups listed for selection, and the shelf-space method could not be used. This method was to be avoided and used only if not to do so would result in a refusal. The equal probability method assigned percents to each of the groups listed solely on the basis of the number of units listed. The percents assigned were equivalent percentages (differing by no more than one percent). In general, disaggregation was limited to five stages in order not to jeopardize cooperation and accuracy. Thus, if after five steps of disaggregation a unique item was not identified, the respondent was asked to identify the volume selling item that fell in the group selected during the fifth stage of selection. The item identified by the respondent became the unique item to be priced over time. In order to facilitate the disaggregation process, BLS provided the field staff with numerous copies of disaggregation sheets that contain both work space

120

Paul A. Armknecht and Daniel H. Ginsburg

and three tables: random number table, ranking table, and equal probability table. Note that a series of unique random number tables were produced for use on disaggregation sheets. These disaggregation sheets were used to select each of the unique items that were priced for the index. The following example of determining the characteristics of a quote for beauty parlor services demonstrates this disaggregation process. (See appendix A for the referenced forms.) The BLS field representative visited the outlet and contacted the owner who agreed to participate. The owner indicated that eight services as found on the CPI checklist for beauty parlor services for females are provided, and the BLS field representative listed them in the first box of the disaggregation sheet. The respondent was able to provide an estimate of total revenue derived from each of these eight services. The percent of total revenue for each service was calculated. (Please see disaggregation sheets in appendix A.) Next, the field representative calculated a running (or cumulative) total on the basis of the percent of sales values. Then using the random number table on the disaggregation sheet, the field representative identified the selection number 73, because this was the first stage of disaggregation and there was only one item to be selected. Thus, haircut, coloring, and shampoo were selected during the first stage of disaggregation. In the second stage, the respondent provided a ranking of the four types of haircuts performed and the percentage of sales was calculated from the ranking table for four items, yielding trim. Following the ranking procedure three additional disaggregation steps were completed yielding: type of coloring-rinse; shampoo and set; and length of hair-short. Because five disaggregation steps had been completed, the respondent was asked to identify, for the final price factor in this outlet, the volume age of client for the above selected item, which was adult. The final step in unique item selection was to circle the selected item characteristics on the ELI checklist (shown in appendix A). Throughout the life of the 1978 revision, checklists were modified to ensure that the characteristics of disaggregated items could be unambiguously identified through the elements listed on each checklist. Through product research, current price-determining variables that make up the universe of the commodities and services that were eligible for selection were identified and incorporated into the checklist. The same disaggregation process was continued for the 1987 revision. This disaggregation process allows a much broader range of services to be covered or represented in the index, and, although fewer detailed indexes are thus available, the broader range of covered items yields much more accurate, albeit more general, published indexes. For example, rather than seven specific physician indexes being published between 1964-77, only one overall physician services index has been available since 1978. Where before the 1978 revision BLS priced only two types of visits for general practitioners, with the advent of disaggregation procedures, sixteen types of services are now priced-eight visits and eight procedures. (See general medical practice

121

Measuring Price Changes in Consumer Services

checklists, September 1977, in appendix A.) Note that ideally BLS would like all services performed by general practitioners to be included. However, test pricing indicated that the effort required was too time-consuming for physicians to participate in the disaggregation process. Thus, BLS had to limit the universe on which disaggregation was based to a subset of eligible services. Similar limitations were required for automobile and home maintenance and repair services to reduce the extent of the eligible item universe. 3.1.5

Measurement of Home Ownership in the CPI

Prior to the 1953 revision only the outright expenditures of home ownersthat is, mortgage interest, property taxes, property insurance, home maintenance and repairs, and ground rent and financing charges-connected with the ownership of a dwelling were included in the index. However, only rent was priced and the above home-owner expenses were moved by the change in rent. Expenditures to purchase a home-the down payment and payments toward mortgage principal-were considered to be investments (expenditures to purchase an asset) rather than consumer expenditures and, thus, excluded from the index. Up until World War 11, the imputation of home-owner costs to rents raised little question. Most wage and clerical workers, the only population covered by the CPI until the 1978 revision, obtained their shelter through renting dwellings. In addition, studies indicated that rental and owner housing were in competition and prices of each type of occupancy moved similarly. The widespread introduction of rent controls during World War 11, and their continuation after the war, combined with the increased importance of home ownership among wage and clerical workers focused attention on this concept of defining and measuring home ownership. After much study, a new durablegood theory of home ownership evolved during 1952. Behind the adoption of this theory was the general acceptance of the view that the outright purchase or mortgage payments on principal by home owners are merely substitutes for rent payments by renters and are made chiefly to provide shelter for the family and, only secondarily, for purposes of investment. Proponents of this durablegood theory felt that houses, to a degree, are similar in characteristics to other durable goods such as automobiles and appliances. Purchase of durable goods requires a sizable outlay; they possess substantial resale value; their life expectancy is very long and none are consumed immediately. So they all have an element of investment. Also, even though houses produce a marketable service-shelter-it was felt that, as long as owners choose to occupy their houses, no marketable service exists. The treatment of home ownership in the CPI as a durable good was defined as measuring the effect of price change on the cost of acquiring and maintaining homes (Lane 1979). The expenditure value for home purchase was derived as the product of the prevailing 1952 house prices and the annual rate of purchase. Mortgage interest was derived as the product of the rate of mortgage contracting among owners, the percent of consumer units who are home own-

122

Paul A. Armknecht and Daniel H. Ginsburg

ers, and the average per period interest contracted per mortgagor, which was calculated by using 1951 average market value of homes in sample areas, a derived ratio of mortgage to purchase price and appropriate terms, and interest rates. Mortgage interest is, therefore, considered the price of borrowing money for the purchase of a home. The index for mortgage interest, thus, was designed to measure the change in the amount of interest required in current markets at current rates to buy houses of the same quality and at the same ratio of loan to purchase price as in the reference year. Therefore, the change in interest was the product of the change in rates for new loans on new and existing homes and the change in market value of homes as measured by home purchase. To reiterate, home purchase and mortgage interest reflected expenditures only for those owners who acquired housing in the survey reference period. The expenditures for the remaining home ownership items-property taxes, property insurance, and home maintenance and repair commodities and services-reflected outlays for all home owners, regardless of when they purchased their homes. Under this durable-good approach all the above expenditure elements were moved by price changes obtained through the direct collection of prices for a sample of transactions associated with each item. Essentially this same definition of home ownership was followed from 1953-82. 3.1.6 Owners’ Equivalent Rent in the CPI With the release of the January 1983 CPI-U, BLS changed the homeownership component of the CPI to a flow-of-services approach. The change, first announced by Commissioner of Labor Statistics Janet L. Nonvood (1981), was implemented by means of the rental-equivalence technique for measuring the shelter cost of consumers who own their own homes. BLS delayed converting the companion index, the CPI-w, until the index for January 1985 because the CPI-W is used to escalate payments in many contracts and federal entitlement programs. It was deemed necessary to give notice to affected parties well in advance of introducing the change. In essence, the change converted the home-ownership component from a method that mixed investment and consumption elements to a flow-ofservices approach that measures only shelter consumption-the cost of shelter services consumed by home owners. The expenditures to purchase new housing units or the expenditures to replace that part of housing stock that has worn out are investments to purchase assets, things that provide a flow of services. It is the monthly cost of these services that the CPI should track. BLS implemented the flow-of-services approach using the rental equivalence technique, which estimates the change in cost of housing services for home owners from the cost of renting housing services. This conversion to rental equivalence followed many years of recommendations and research by BLS staff and by other government, academic, busi-

123

Measuring Price Changes in Consumer Services

ness, and labor economists and statisticians-see Gillingham (1980, 1983) and Gillingham and Lane (1982). In 1980, the bureau introduced five experimental measures (CPI-U-X 1-CPI-U-XS) to demonstrate the effect different home-ownership concepts and techniques could have on the all-items CPI. The measure known as CPI-U-X1 , which used a rent substitution technique, is the direct (although approximate) antecedent of the method the bureau now has adopted. To implement the measurement of price change for any CPI item, one needs a base-period expenditure weight and a monthly estimator of price change. The base-period expenditure measure for rental equivalence, or owners’ equivalent rent, is the rent the stock of owner-occupied units would bring if they had been rented. The last several CEs have asked home owners to estimate what their homes would bring if they were rented. A study by Francois (1989) of the responses to this question showed that the average estimates are reasonable. For purposes of calculating CPI base-period weights, BLS treated their responses to this question as actual consumption expenditure. The question defines implicit rent as net of utilities and household furnishings. The weights for other CPI items, notably those for home maintenance and repairs, household insurance, and major home appliances clearly included some expenditures that were also covered by the weight for owners’ equivalent rent. To avoid double counting these items, BLS adjusted the CE expenditures home owners reported for them before computing the final weights for each of the above items. The experimental measure, CPI-U-X1, simply used the U.S. rent index as a measure of the change in the implicit rents on owner-occupied housing units (henceforth, owner units). However, the sample of rental units was not representative of owner units so that when the CPI converted to rental equivalence in 1983, the rent sample was reweighted and augmented to measure price change for owner units. From 1983-86, when they were used to measure owners’ equivalent rent, the renter units were weighted according to the number of owner units in the universe they represented. For the 1987 revision of the CPI, BLS developed an independent measure of owners’ implicit rents. A new sample of renters was drawn to represent renters in owner neighborhoods. A sample of owner units was also drawn in the same area. On the basis of proximity and similarity of structural characteristics, each owner unit was assigned a small set of similar, nearby renter units. No renter unit could be assigned to more than three owner units. The change in implicit rent for each owner unit is the average change in the pure rent of its set of renter units. Pure rent is the actual rent paid, less the estimated value of utilities and furniture included in the rent. Once an implicit rent is calculated for each owner unit, the average change is obtained in the same manner as for rent. For more detail, see Kosary, Sommers, and Branscome (1984) and Lane and Sommers ( 1984).

124

Paul A. Armknecht and Daniel H. Ginsburg

This method ensures complete coverage of the owner universe. It provides a measure of change that is capable of distinguishing differences in rent movements in owner and renter areas. The new housing sample over represents renters in owner areas to facilitate matching owner units with nearby rental units. Using a multistage stratified design, the sample areas were selected in strata distinguished by tenure (percent owner occupied) and rent level (a surrogate for income). For a more detailed discussion of this the reader is referred to the BLS Handbook of Methods ( 1988). 3.1.7

Improvements in Other Service Items during Recent Revisions, 1964-87

In table 3.1, improvements to selected CPI service items and their basic pricing methodologies are listed for the last three CPI revisions. During the 1978 revision, BLS made several improvements in item strata representation in the CPI. For example, the following item categories were added to the list of items priced: housing at school, cable television, refuse collection, gardening services, tenants’ insurance, auto financing interest rates, vehicle rental, other medical professionals (e.g., chiropractors), hospital emergency room care, nursing and convalescent homes, admission to sporting events, club membership dues, and elementary and high school tuition. The following item categories were added for the 1987 revision: care of invalids while convalescing at home, ship fares, photographers’ fees, pet services, veterinarian fees, technical and business school tuition, and day care and nursery school tuition. In addition to expanding the list of eligible items during each revision, BLS also improved the pricing methodology for many items. This was particularly true in BLS’s methodology for handling health insurance in the CPI. Health Insurance in the CPI The expenditure weights in the CPI reflect direct household purchases of health insurance, including Medicare part B, plus employee contributions to insurance purchased through work. The share of premiums paid by employers is not included in the weights as these are considered to be business costs and not direct consumer expenditure. Between 1950 and 1963, BLS priced the most widely held Blue Cross/Blue Shield family group plan in each priced area. During this time period, premium changes were reflected as follows: Premium differences resulting from changed medical care prices associated with covered services were reflected. Premium differences caused by changed utilization (change in the frequency that the policy is employed to cover expenses incurred by those insured) were reflected because such changes were viewed as not directly affecting the amount of protection offered by the policy. Premium differences resulting from benefit changes were not reflected, as they were considered to be quality changes, affecting the level of coverage under the policy. As BLS approached

Table 3.1

CPI Service Improvementssince 1964

Service Hotels & motels

1964 Revision Sample from directories Regional relatives

Housing at school Electricity

Natural gas Telephone, long-distance

Water & sewer Maintenance & repair service

Not priced Priced through Federal Power Commission Small no. of preselected consumptions Small no. of preselected consumptions Measured change in residential revenue, when rate change occurred Priced same Washington-selected consumption in all areas Outlets selected from yellow pages Washington-selected services

Cable TV Refuse collection Gardening service

Not priced Not priced Not priced

Baby-sitting

Priced from state employment office who hardly ever placed babysitters

1978 Revision Sample from unemployment insurance file; then CPOPS Market basket relatives

1987 Revision Sample from CPOPS

Linked in May 1981 Priced through DOE

Market basket relatives to December 1989; from January 1990 on, moved by regional relatives Priced Priced by BLS

Consumption disaggregated at outlet

CE selected consumption

Consumption disaggregated at outlet

CE selected consumption

Priced actual sample of calls

Priced actual sample of calls Added non-AT&T carriers from CPOPS Same as 1978 revision

Determined consumption through disaggregation at each outlet Outlets from CPOPS Disaggregated from among Washington-selected services Gradually introduced at rotation Gradually introduced at rotation Minimum wage, then CPOPS Certainty item Priced from minimum wage

Same as 1978 revision Same as 1978 revision Priced Priced CPOPS Probability item Priced from minimum wage to December 1989; January 1990 on, moved by day care. (continued)

Table 3.1

(continued)

Service Domestic services Care of invalids convalescing at home Tenants’ insurance Auto insurance

1964 Revision Priced from state employment office who hardly ever placed domestics Not priced

Priced from minimum wage

Not priced Priced in Washington

Priced Priced part in Washington; rest in field Selected companies from A. M. Best tape Priced policy characteristics identified in policies disaggregated at each company Priced interest rates associated with disaggregated loan characteristics Auto & truck rental priced

Bureau rates & two deviating &/or independent companies Priced one set of liability limits & one car, full-size Chevrolet Auto financing

Weight moved by new car index

Vehicle rental

Not priced

Airfares

Priced a small no. of chiefly coach fares, rest first class; priced one carrier per city; trip based on average length of air travel from city-used to identify destination. Intercity bus fares-selected trips in Washington-CPI city to nearest terminal city; train fares-selected trips in Washington, based on the average length of trip from city

Other intercity transportation

1978 Revision

Not priced

Disaggregated to trip characteristics from CAB tape of actual trips taken from originating airport; type of fare also identified Intercity bus and train fares-disaggregated to trip based on scheduled trips originating in each sample city

1987 Revision Gradually changing from minimum wage to CPOPS Priced; sample from CPOPS Same as 1978 revision Priced mostly in field; rest in Washington Selected companies from A. M. Best tape Same as 1978 revision

Same as 1978 revision Same as 1978 revision, plus added other vehicle rental (e.g., trailers) Same as 1978 revision & provided additional broader instructions for dealing with discount fare characteristic changes; shifted to SABRE pricing Same as 1978 revision, plus added ship fares

Professional medical services: Physician services

Dental fees

Other professional Hospital services

Priced 7 preselected services

Priced 3 preselected services

Not priced Priced semiprivate & private room and 3 ancillary services until 1972; in 1972 dropped private room & added 6 more preselected ancillary services.

Priced a wide range of preselected services, with specific service priced disaggregated to each physician Disaggregated to a specific service from among those provided by the dentist Priced Disaggregated rooms and inpatient services from among the individual services available at the hospital

Same selection process, however, now disaggregate type of fee, if more than one exists for the selected service Same as 1978 revision

Same as 1978 revision, plus nurses Same as 1978 revision, plus added outpatient stratum that includes ER and all other outpatient services

Added pricing emergency room & nursing homes Entertainment services: Admissions

Fees for participant sports

Priced adult & child movie admissions & drive-in movie admissions

Priced only bowling & golf-green fees

Disaggregated from among the full range of admissions-movies, plays, operas, rock concerts, circuses, etc. Added admission to the full range of sporting events-baseball, tennis, horse racing, auto racing, etc. Disaggregated from among the full range of participant sportsbowling, tennis, golf, swimming, etc.

Same as 1978 revision

Same as 1978 revision

(continued)

Table 3.1

(continued)

Service

1964 Revision

Membership fees

Not priced

Fees for lessons

Priced only beginning piano

Other entertainment services: Photographers Pet & veterinary services Equipment rental Tuition & fees

Personal expenses: Legal fees Cemetery lots Accounting fees

Not priced

1978 Revision Priced a full range of membership organizations-golf clubs, fitness centers, auto clubs, credit cards, etc. Priced a full range of lessons-music, sport, dance, language, etc. Not priced

College tuition priced

College tuition, elementary & high school tuition priced

Day care priced from state employment agencies

Not priced

Priced short-form will

Disaggregated from a group of preselected services Not priced Not priced

Not priced Not priced

Nofe: CPOPS = current point of purchase survey.

1987 Revision Same as 1978 revision

Same as 1978 revision Added

Same as 1978 revision, plus added technical and business school tuition and day care and nursery school tuition Priced; sample from CPOPS

Same as 1978 revision Added Added

129

Measuring Price Changes in Consumer Services

the 1964 revision, problems with this direct method of pricing health insurance became apparent. Defining when a benefit change occurred was difficult at times. For example, was a new surgical schedule introduced to cover the increased cost of covered operations, or in part to provide an increase or decrease in the level of protection? Also, BLS research indicated that the proper methodology for handling changes in utilization should be to exclude their effect from the measurement of premium changes. BLS’s new view was that utilization changes, reflecting the intensity with which the policy is used, do affect the level of benefits, by affecting the risk level associated with those insured. Therefore, premium changes associated with utilization changes (like those associated with benefit changes) should not be reflected in the index. Health insurers indicated that it was becoming more difficult to provide the premium effect of the various identified benefit changes and that no premium data were available related to the effect of utilization changes. Thus, for the 1964 revision, BLS shifted to an indirect approach to pricing health insurance. Appropriate health insurance premium changes for the 1964 revision were defined to include the effect of changed medical care prices, administrative cost and surplus requirements, and the profit needs of commercial carriers. The index weight used for health insurance was broken down into two subweights, one reflecting the benefits companies pay out and the other representing the premium income retained to cover overhead and profits for commercial carriers. Although consumers do not directly purchase the services represented by these subweights, they pay for them indirectly through their premium payments. Also, as the prices of these elements change so do consumer payments. Under the indirect method of pricing health insurance the expenditure weights were first subdivided between claims and retained earnings on the basis of national Blue Cross/Blue Shield and commercial data. Within the claims portion, expenditure weights were further allocated to in-hospital and out-of-hospital costs and within these to ten priced medical care items: semiprivate room, private room, operating room, diagnostic X ray, physician’s office visit, tonsillectomy, herniorrhaphy, obstetrical case, and prescription drugs. The retained earnings portion was adjusted annually and also escalated from month to month on the basis of the average change in the prices representing claims. In this way, the relation between claims and retained earnings remains fixed between adjustments. The annual adjustment for changes in the ratio of retained earnings to premium income was based on national financial data for the Blue Cross/Blue Shield and commercial carriers as reported to the Health Care Financing Administration (HCFA). For convenience, these retention factors were calculated by expressing them as a proportion of benefits (claims) rather than of total income, because the weights for the priced medical care items correspond to benefits. This ratio adjustment reflects how retained income has either increased or decreased as a proportion of benefit payments over the previous year.

130

Paul A. Armknecht and Daniel H. Ginsburg

Effective in February 1972, the list of unique items priced to represent hospital charges was expanded in order to provide an improved estimate of price change. This expansion also affected health insurance as the claims portion was reallocated by dropping private rooms and adding laboratory fees, inhospital prescriptions (an anti-infective and a tranquilizer), electrocardiogram, physical therapy, intravenous solution, and oxygen. For the 1978 revision to the CPI, BLS continued to indirectly price health insurance premiums. However, the use of disaggregation to select unique items for pricing led to improved representation of the medical care items used to measure price changes for estimating the claims portion of health insurance. In addition, BLS shifted from reflecting the retained earnings adjustment all in one month to spreading the change over the entire year and thus minimized the effect in any one month. In the 1987 revision the indirect method for pricing health insurance was again employed. However, instead of using the medical care sample to move two sets of expenditures (out-of-pocket expenses and insurance benefits), the expenditures of the two types of payments were combined into one index. Thus, the expenditures for each medical care item are now the combination of the direct out-of-pocket expense for the item by the consumer and the indirect expense for the item paid from consumer purchased health insurance. See Ford and Sturm (1988). The current item labeled health insurance is the portion of premium payments retained by the insurer in the form of operating expenses and profit. During the 1987 revision BLS also shifted to new retained earnings’ sources. In recent years, the data reported to HCFA covering commercial carriers have shown inconsistencies, reflecting many revisions and changes in methodology. Also, both the commercial and Blue Cross/Blue Shield data obtained from HCFA suffered from lengthy lags between the reference period and the release of the data. These limitations of the data led BLS to investigate alternative data sources. After careful study improved data sources were located. The Argus health chart, which contains sample company data that must be reported to state insurance regulatory agencies, was selected for company commercial carriers. The Argus data are released with a shorter lag than HCFA data. In addition, BLS improved the timeliness of Blue Cross/Blue Shield reporting by obtaining data directly from the provider on a quarterly basis, rather than annually from HCFA. Over the past 15 years, BLS has twice tested the direct pricing of health insurance policies. The latest effort was during 1984 and 1985. For this test, a data collection document (checklist) was created to describe accurately the numerous variable qualities and characteristics of the many health insurance policies. The checklist was used to rate, classify, and differentiate between the various health insurance policies that were included. Health insurance schedules for insurance company policies available from a 1976-77 test pricing program were examined, and a sample of potential respondents was picked to be priced. The sample was chosen to represent

131

Measuring Price Changes in Consumer Services

commercial carriers and Blue Cross/Blue Shield, individual and group plans, and single and family plans and to cover different coverages, qualities, price ranges, geographical areas, and so on. Next, sample respondents were contacted to collect 1984 and earlier pricing and benefit package information on these (or other similar) policies that were initiated in the mid-1970s. From this back-pricing information, an experimental health insurance index was constructed for the period 1977-84. About 70 health insurance companies were contacted in the survey. Of the 22 companies that provided data, 16 responses representing 95 price quotes were deemed complete. The responses consisted of a fairly representative geographic distribution of policies. However, the response rate, which was poor, principally reflected the low number of the 1976-77 initiated policies that were still in force when the carriers were contacted again in 1984. Completed responses were described on the checklist along with the annual price trend of the rates they charged to subscribers. The reported health insurance premiums used in this survey were for policies purchased directly by consumers and policies purchased at work that reflect employer and/or employee contributions. Employees make considerable contributions to their own health insurance premiums. Data from the employee benefit survey conducted by BLS indicates that the percent of employees whose health insurance premiums are completely paid by their employers has been declining in recent years-from 54 percent in 1986 to 48 percent in 1989. Utilization data, one aspect of quality change, were also requested from insurance company respondents, but only five respondents were willing and able to provide limited utilization data. At that time, appropriate data on utilization and methodology to account for utilization (quality) changes in the direct pricing of health insurance were not available, and thus are not accounted for in the experimental index. This lack of data had serious effects on the acceptability of the direct pricing approach. Quality changes stemming from changes in benefit packages were handled in two ways: In some cases, insurance companies indicated the effect on premiums of a given benefit change, allowing the bureau to adjust these policies for quality changes. In those cases where the effect on premiums of benefit changes was unavailable and adjustments for quality could not be made, the policies were excluded from the index. The experimental index is based strictly on the premium changes of policies that did not change coverage plus those for which premium adjustments could be made for changes in coverage. Comparison of the results of the experimental direct price index for health insurance with the indirect method for the same period are shown in table 3.2. As is evident in the comparison, the direct pricing method indicates a faster rate of increase than the indirect method currently used in the CPI. The differences between the indexes resulting from the two pricing methods may be due to a number of possible causes. Each method measures somewhat different things. Some of the large price jumps in the direct pricing index in 1979-80 and in

Paul A. Armknecht and Daniel H. Ginsburg

132

Comparison of Alternative Health Insurance Price Indexes

Table 3.2

December 1977 1978 1979 1980 1981 1982 1983

Experimental Direct Pricing

Current Indirect Method

100.0 116.8 128.7 164.0 174.4 211.6 246.6

100.0 100.5 123.6 133.7 152.9 117.2 183.9

1981-82 were attributed by some insurers to shifts by many of the healthiest subscribers out of their current plans into health maintenance organizations or lower-cost, higher-deductible plans. This left the affected plans with an older, less healthy pool of subscribers who made a higher level of claims. The resulting increased costs for the health insurers may have caused some of the sharp increases in the direct pricing index. This type of utilization change is exactly the type of quality change that should not be included in the index, and the inability to obtain data on the premium effect of these utilization changes is the chief roadblock to developing a directly priced health insurance index.

3.2 3.2.1

CPI Improvements Currently Underway for Consumer Services Using Hedonic Regression Analysis to Measure Quality Changes

As previously mentioned, most cases of quality adjustments in the CPI use an imputation procedure (linking) that excludes a price comparison of the new and the previous service and estimates the price change as the average change of other similar services available during the comparison period. This method assumes that the underlying price change for the service not available would have been the same as those that are available. If, however, price change is different for the new service, we could be missing some price change. It is often the case that providers of services use the introduction period for new or replacement services as an opportunity to change prices. As an example of this situation consider airfares. The airlines at one point in time introduced a new set of discount airfares to replace supersaver fares. Originally, supersaver fares required a 30-day prepayment to obtain the reduced fare. The new discount fares introduced a lower price structure than the supersavers and required the 30-day prepayment. However, along with the lower fares came a 50-percent cancellation penalty, a quality difference between the fares that mayishould make them noncomparable. If no other airline price changes occurred during this month, the exclusion of the discount

133

Measuring Price Changes in Consumer Services

quotes w c d d result in an imputed price change of zero because all other airline fares (coach and first-class) remained the same. The index, therefore, misses any possible price change occurring with the introduction of the lower discount fares under current procedures. (This problem would exist even if other fares changed but at a different rate.) One method discussed earlier for remedying this situation would be to make a direct quality adjustment. We would like to obtain the market value of the quality difference between the original service and its substitute. Then we could adjust the price of the old item directly for the quality change. For example, when an auto insurer dropped the 50-dollar deductible on collision coverage and replaced it with a 100-dollar deductible, we were able to quality adjust for the premium difference because the new minimum service already existed in the market place. This situation occurs infrequently, however. When market information is not available, we could ask the provider of the service to estimate the value of the change from cost data. Changes in provider’s costs could be adjusted for normal profit margins and marked up to the retail level. The resulting price change serves as a proxy for consumers’ valuation of the quality change, which cannot be observed directly in the market place. The success of such an approach is contingent on the willingness and ability of providers to supply the cost data. In the case of the airline example just cited, the required information could not be obtained. Yet another approach would be to estimate the value of the quality change through statistical means because the quality change is usually manifest in some difference in characteristics. It might be feasible to measure the value of the quality change using hedonic regression techniques. The price for a service, P, is a function of the price, b,, for each of its characteristics, X , . In a linear form this would be: P

=

b,

+

K

b,X, 1=

I

+ e,.

The estimated parameters, b,, from the linear regression provide implicit prices for each of the K characteristics. Whenever a quality characteristic changes, an estimated value for the amount of the quality change can be calculated from the regression parameters. Griliches (1971), Triplett (1969, 197 1a, 197 1 b), Triplett and MacDonald ( 1 977), Early and Sinclair (1983), and Liegey (1990) have demonstrated the use of hedonic techniques to measure and adjust for quality change in price indexes. The general form of the regression model used in these studies is log linear where the natural logarithm of price is regressed on the observed values of the characteristics. Each parameter estimate, b,, in this form is interpreted as the approximate percentage change in price associated with a unit change in a quality characteristic, X,. In addition to providing a direct estimate of the quality difference resulting from a change in a characteristic, hedonic models enhance the analyst’s

134

Paul A. Armknecht and Daniel H. Ginsburg

knowledge of the overall quality composition of the services offered. The models can be used to identify those major price factors that contribute the bulk of the quality makeup of the service. These major price factors are, in turn, used as aids by the analyst on two important fronts. First, they help analysts more clearly distinguish between significant and insignificant price determining factors and provides statistical measures of the strength of the relationship. As a result, analysts can refine their criteria for determining whether a change in a service characteristic makes the previous and new services comparable or not. Second, the major price factors can be used to redesign the collection documents (checklists) used by CPI field staff. In ordering the price factors on the checklists according to their importance in determining quality, the field staff can select the best substitute service by matching the most important quality characteristics in the designated order. Preliminary studies in developing hedonic models for consumer services have been conducted for hospital-room stays, airline fares, intercity bus fares, and ship fares. We have not been successful so far with hospital rooms because a critical variable, level of nursing care (acuity), is not measured consistently among hospitals. We will have to improve our measurement of this variable by attempting to standardize the measure on our collection documents rather than by accepting each hospital’s current measurement system for billing purposes. The model for airline fares in table 3 . 3 indicates that there is a significant relationship between the natural logarithm of price and the various explanatory variables. The variables that contribute significantly to variations in price are distance (miles and miles squared), type of fare (first-class, coach), connecting flights, minimum-stay requirements, penalty for cancellation and two area dummy variables. The connecting flight variable is of interest because it may be a proxy for hub cities. There has been recent concern that fares from hub cities are higher because of lack of competition. Of primary interest to us is the coefficient for penalties. In our earlier example of the introduction of discount fares there was no information available on the amount of adjustment that should be made for the cancellation penalty. On the basis of our hedonic model we could quality adjust the previous supersaver fares, P,, for the 50percent penalty and then use the quality-adjusted price for comparison with the new discount fares in index calculations. The parameter estimate for penalty is -0.1409, which is then multiplied by 0.5, the amount of the penalty. The amount of the adjustment corresponding to these results in A In P = - .07045, which implies A P = - .06802 P , . If a supersaver fare of $200 was previously used, we would reduce it by $13.60 and use the resulting quality adjusted price of $186.40 as the previous price to be compared with the new discount-fare price in the current month’s index computation. Our hope is that, as these models are refined, we will be able to use the results to improve our ability to make quality adjustments for consumer services in the CPI. Similar studies are being conducted for selected commodities.

135

Measuring Price Changes in Consumer Services Hedonic Regression Results for the CPI Airfares Sample

Table 3.3 Variable

Parameter Estimate

INTERCEPT

4.9481*

0.1093

MILES

O.o006*

0.00004

-4.96E-08* 1.1703* 0.6892* 0.2737* 0.0985 - 0.0037 -0.1409* 0.2040* -0.0174 -0.0177 -0.1336 - 0.3038* - 0.0771 0.5623*

7.47E-09 0.1385 0.1191 0.0376 0.1236 0.0044 0.0590 0.1037 0.1689 0.1351 0.1127 0.1505 0.1222 0.2452

MILESSQ FIRST-CL FF-COACH CONNECT ONE-WAY ADV-PUR PENALTY MIN-STAY ALASKA HAWAII MEX-CRBN CANADA EUROPE FAR-EAST

Standard Error

t-value 45.270 14.683 -6.639 8.451 5.789 7.280 0.797 -0.840 -2.388 1.968 -0.103 -0.131 -1.185 -2.019 -0.631 2.293

Dependent variable: LNPRICE R2 = 0.8009 AdjustedR2 = 0.7909 No. of observations = 314 Source: Dale A. Smith, Hedonic Regression Analysis of Air Fares in the CPI. Unpublished research paper. Division of Consumer Prices and Price Indexes, Bureau of Labor Statistics. Notes; LNPRICE = Log of round-trip airfare. (One-way fares were doubled.) MILES = Number of miles (one way) from origin to destination. MILESSQ = Miles squared. FIRST-CL = Dummy variable-coded 1 for first class and 0 otherwise. FF-COACH = Dummy variable-coded 1 for full fare coach and 0 otherwise. CONNECT = Dummy variable-coded 1 for connecting flights between two separate carriers and 0 otherwise. ONE-WAY = Dummy variable-coded 1 for oneway trips and 0 for round-trips. ADV-PUR = Number of days of advanced purchase required for the fare. PENALTY = Penalty as a percent of the fare for a change in itinerary. MIN-STAY = Dummy variable-coded 1 if a minimum stay is required and 0 otherwise. ALASKA = Dummy 1 for trips with origin or destination in Alaska and 0 otherwise. HAWAII = variable-coded Dummy variable-coded 1 for trips with origin or destination in Hawaii and 0 otherwise. MEX-CRBN = Dummy variable-coded 1 for trips with destination in Mexico or the Caribbean and 0 otherwise. CANADA = Dummy variable-coded 1 for trips with destination in Canada and 0 otherwise. EUROPE = Dummy variable-coded 1 for trips with destination in Europe and 0 otherwise. FAR-EAST = Dummy variable-coded 1 for trips with destination in East Asia and 0 otherwise. *Denotes that the coefficient is significant at the 5% level.

3.2.2 Indirect Pricing of Automobile and Tenants’ Insurance in the CPI Currently the CPI expenditure weights for automobile and tenants’ insurance reflect 100 percent of premiums paid by policyholders as reported on the CE. These lines of insurance are priced directly, that is, the index is moved by changes in premiums of policies whose benefit levels are held constant. This treatment of insurance is conceptually unsound because most of premiums paid represent payments to a pool of funds that are redistributed as claims

136

Paul A. Armknecht and Daniel H. Ginsburp;

payments and as such should be included only in the insurance component of the index or only in the expenditure weights of the components for which the benefits are paid. At present there is some double counting, as noted below. Under the latter approach, which we prefer for the CPI, the insuranceexpenditure weight should then be limited to that portion of premiums paid that are not returned to policyholders as benefit payments, that is, as retained earnings. Automobile and tenants’ insurance in the CPI should then be moved by changes in these retained earnings similar to the treatment of health insurance. Assuming the portion of premium payments that contribute to retained earnings is 10 percent, the remaining 90 percent is returned to policyholders in the form of benefit payments, which are then presumably spent on the specific goods and services covered by the insurance. For automobile insurance, this would include such items as automotive body work, medical expenses, lost wages and legal fees. Currently with direct pricing these expenditures are included in automobile insurance. With the retained-earnings approach, that is, indirect pricing, the purchases made with benefit payments would be shifted from automobile insurance to the items purchased. In addition to reapportioning expenditure weights, a switch to indirect pricing would alter the index’s treatment of changes in utilization of insurance. Under the current system of direct pricing, price changes resulting from utilization changes are reflected in the index, as BLS can neither identify nor adjust for their effect on the specific policies for which prices are collected. To illustrate with automobile insurance, assume the Northeast has an unusually bad winter with record snowfall. This causes a 50-percent increase in collision claims over what is typical for the winter season. Utilization has increased as evidenced by an increase in the expected value to a typical insured’s claims payment. As a result, premiums increase, and this increase is reflected in the index. Given the same situation with indirect pricing, this increase in utilization would have a zero net effect on the index. In period N , benefit payments would increase causing retained earnings to fall. In period N 1, premiums would be increased to restore retained earnings and thus bring the index back to its original level. Proponents of indirect pricing suggest that this is appropriate because increased utilization suggests a higher quality policy. That is, as utilization increases, the expected value of the insured’s benefit payments increase and thus improve the quality of the policy. Therefore, a price change resulting solely from a change in utilization should not affect the index. Utilization should be held constant at the level measured during the base period. Another variable on which to compare direct versus indirect pricing is investment income. Some portion of retained earnings is invested and generates income. With direct pricing, the effect of a change in investment income is reflected in the index at least to the extent that this income is returned to policyholders in the form of higher or lower premiums. Given the highly compet-

+

137

Measuring Price Changes in Consumer Services

itive insurance market, it is reasonable to assume that, although part of an increase in investment income may be held as profit, insurance companies would be forced to return some portion of this income to policyholders in the form of lower premiums. Therefore, with direct pricing, an increase in investment income may be expected to cause premiums and hence the index to decline. Investment income would affect the index similarly with indirect pricing. Retained earnings are defined as premiums less benefits. Given the above assumption, that there is an inverse relationship between changes in investment income and premiums, there is also an inverse relationship between investment income and retained earnings. That is, as investment income increases, premiums and therefore retained earnings decrease. Using data from the A.M. Best Company, Inc., it appears that automobile and tenants insurance indexes based on changes in retained earnings would have moved more erratically than the current measures. This is shown by table 3.4, which indicates annual percent changes in retained earnings for private passenger automobile insurance and home owners’ multiperil insurance. These retained earnings are based on aggregates and reflect the financial reports of virtually all insurance organizations doing business in the United States. Retained earnings are defined by Best as administrative costs plus profits or minus losses, that is, economic profit. This is arrived at by subtracting claims losses and policyholder dividends from net premiums earned. This can be expressed as P - C - D = A G, where P = premiums; C = claims; D = dividends; A = administrative or underwriting costs; and G = economic profit (can be a gain or loss). Before the percent changes in table 3.4 were calculated, the retained earn-

+

Table 3.4

Insurance Industry Retained Earnings and CPI Insurance Indexes (% change)

Year

A. M. Best, Private Passenger Auto Retained Earnings

1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

+38.1 - 4.9 -11.0 + 3.4 - 16.3 - 7.2 + 6.0 - 14.7 - 13.9 +11.5

CPI-u, Auto Insurance

+ + +

7.6 3.5 5.8 7.3 5.4 + 8.6 + 9.1 7.8 + 12.1 +11.8

+ +

+

A. M. Best, Homeowners’ Multiple Peril Retained Earnings

+ 21.4 + 1.2

- 23.5

-25.0

+ 16.0

7.7 + 5.9 - 12.2 - 15.9 +47.9 -

CPI-u, Tenants’ Insurance

+ 4.5 + 10.2 + 10.2

+ 10.9

+

+ + + +

6.8 6.7 2.3 5.8 5.4

138

Paul A. Armknecht and Daniel H. Ginsburg

ings figure for each year was divided by claims for that year to correct for the effect on retained earnings of changes over time in volume of premiums written. Therefore, by looking at changes in these ratios rather than only changes in retained earnings, no change is reflected to the extent that premiums less dividends increase or decrease concurrently with claims. In summary, a switch from direct to indirect pricing of automobile and tenants’ insurance would result in the following changes: 1 . The CPI weights for the insurances would be greatly reduced reflecting only that portion of premium payments that accrues to insurance companies’ retained earnings. That portion of premium payments that accrues to benefit payments would be reapportioned among the CPI goods and services purchased by those benefit payments. 2. Utilization would be considered a quality variable. Therefore, all else being equal, the index level would be unaffected by a price change resulting from a change in utilization. Currently with direct pricing, a utilization change causes the index to move in the direction of the change. 3. The insurance indexes would be expected to move erratically and with no apparent pattern to the extent that insurance companies’ retained earnings continue to fluctuate as they have over the last five to ten years. The desirability of pricing insurance, indirectly by tracking retained earnings versus direct pricing, is then a function of how one defines the cost of insurance to the consumer. The feasibility of indirect pricing is a function of the availability of appropriate ongoing data for tracking retained earnings over time plus data that allows us to redistribute the insurance weights among the consumer expenditures made with benefit payments. For the first of these data needs, retained earnings data, the A. M. Best Company mentioned above has a wealth of annually updated data by company and line of insurance. In fact, as is evident above, they practically calculate our indexes for us. As to data for reweighting from insurance to expenditures made with benefit payments, the data appears to be less than forthcoming. Insurance industry statistical gathering organizations such as A. M. Best do not have data that meet our needs for reweighting. Studies and reports of individual insurance companies may provide some useful breakdowns. Also, the CE may be helpful. There are questions on the CE that ask what portion of expenditures for automobile repair were reimbursed by insurance. However, currently there is no such breakdown for other expenditures reimbursed by benefit payments such as medical expenses (other than those reimbursed by health insurance), legal fees, and, in the case of tenants’ insurance, dwelling contents. This means there is currently the potential for double counting these items. If indirect pricing is to be pursued, more research is required into alternative data sources and modifications to the CE might be needed to support this effort. A final question is: What, if any, effect would a switch to pricing private

139

Measuring Price Changes in Consumer Services

transportation as a flow of services have on the pricing of automobile insurance? This is unrelated to the question of whether to price automobile insurance directly or indirectly, but switching to a flow-of-services approach may change the type of insurance contract we wish to price. This was the case when rental equivalency, a flow-of-services approach, was adopted for the pricing of home ownership. The shift to rental equivalency entailed switching from the pricing of home owners’ insurance to tenants’ insurance. Similarly, the insurance component of private transportation priced as a flow of services would suggest the need for a shift from the pricing of automobile insurance on a consumer-owned car to the pricing of automobile insurance on a consumer-leased car, assuming insurance is not included as a part of the lease agreement. In fact, the norm for auto leasing contracts is for the lessee to purchase his or her own insurance. Furthermore, insurance companies treat policyholders with leased cars identically to policyholders with owned cars, that is, the insurance contract is the same whether the car is owned or leased. Therefore, the type of auto insurance contract priced would not change if we switched to a flow-of-services approach for pricing private transportation.

3.2.3 Develop Improved Methods for Handling Changes in Quality of Medical Care Treatment The discussion of quality adjustment for medical care centers around what information is available to the BLS field representatives at the time of collection, relating to change in treatment associated with the visit or procedure being priced. The information on quality of technological change is only as complete as what the respondent provides. At a hospital, for example, the respondents are varied, with accountants and comptrollers providing price codes and rate information. Nurses, technicians, and pharmacists usually provide treatment and supply details. Many times the BLS field representatives do not have the opportunity to discuss the details of specific changes in services or supplies with the most qualified person. However, overall changes in service that reflect quality or technological improvements are identified by our respondents in most cases, and one of the appropriate quality-adjustment methods previously discussed is used. In many instances BLS field representatives have uncovered these changes by questioning their respondents when a large increase or decrease in price occurs. We are increasing the intensity of training BLS field staff in interviewing techniques and expect this training to lead to more frequent identification of quality changes. Currently some hedonic regression analysis is being conducted for hospital services and physician services to determine whether any adjustments for quality can be made using regression coefficients. The problem with acuity in developing a hedonic model for hospital room charges was discussed earlier. Another problem with developing hedonic models is with the many varied procedures and supplies that are priced for the index. The mixture of services

140

Paul A. Armknecht and Daniel H. Ginsburg

may not allow a large enough sample at the detail needed for statistical estimation. The situation is being studied and further analysis is needed before it can be considered for us in making direct quality adjustments.

3.3 Future Improvements in Measuring Consumer Services 3.3.1 Examining Alternatives to Pricing Individual Medical Services Most of the services priced within medical care services are individual services, for example, a brief office visit for an upper respiratory infection or an emergency room visit for a laceration. Some researchers (e.g., Feldstein 1988) have taken exception to this methodology and have encouraged the bureau to pursue pricing by treatment. Pricing by treatment would define a condition and reflect price changes for all services and procedures that apply to treatment of that condition, that is, physicians’ fees, other medical professionals’ fees, pharmacy costs, and hospital services. Currently we are reflecting price changes based on diagnosis related groupings (DRGs) at hospitals in New York and New Jersey and had priced DRGs in Connecticut hospitals for approximately two years, from March 1987 to October 1989. Connecticut has recently returned to a non-DRG-based system. Approximately 365-400 DRG classifications were developed for specific illnesses and injuries. The DRGs followed somewhat those developed for Medicare patients. Each DRG includes the prescribed treatment and average length of stay. The fee includes all hospital-billed services necessary for treating the illness or injury, that is, pharmacy, laboratory tests, operating room, and hospital room charges. Introduction of required DRG-based payment systems in these states caused us to adopt their methodology because the hospitals within these states were unable to provide us with valid charges for the individual services that we had been pricing at the hospitals. Pricing by DRG has caused a few problems, the first being the all-inclusive fee structure. In the areas where DRG pricing is required, BLS decided that, for every current hospital room or other inpatient service, a DRG would be selected for pricing. When a DRG is priced, other services are included along with the room charge, and price changes may reflect changes in the other services. The DRG priced for a hospital room could include the use of operating room and pharmacy as well as a semiprivate room and, vice versa, the DRG priced for other inpatient services could include the cost of the semiprivate room. However, because the weight for hospital rooms reflects expenditures for room services as calculated from the CE, under the DRG system more than just room rate changes affect this index. The specific treatments and length of hospital stay associated with each DRG may change. Traditionally such changes have been considered quality factors for which quality adjustments would be required. An example of this

141

Measuring Price Changes in Consumer Services

is change to the average length of stay. Each DRG has a designated range for an acceptable length of stay. One patient may stay slightly longer than the previous patient for the same DRG, but as long as it is within the identified range the patients would be charged the same rate. The range may be adjusted annually by the state hospital commission, and this could result in a rate change. Such rate changes are reflected entirely as price changes in the CPI because the charge to the patient for the DRG treatment has changed even though a quality change may have occurred. Another problem with pricing by DRG is the detail of what services and supplies are provided. For example, a patient with DRG 123 was charged $3,500 for this all-inclusive treatment with a length of stay between three and six days. Now this same DRG has an expanded length of stay between two and seven days, and the charge has increased to $3,700. The increase in price could be the result of change in length of stay, additional X rays, higher staff salaries, or a combination of such factors and could correspond to the same or different quality of service. Currently the bureau is not readily able to obtain such exact detail when pricing DRGs. Thus, most likely the change in price in the above example would be reflected as a price increase of 5.7 percent. BLS will be researching what can be done to clarify the amount of detail for these all inclusive fees. Although standard CPI procedures currently price individual services at selected outlets, that should not imply that the complete cost of treating an illness is not being reflected in the index. An example is a patient giving birth by vaginal delivery with the professional medical services index reflecting changes in obstetrician, anesthesiologist and/or midwife and various hospital services, which reflect the changes in the cost of operating room, birthing room, pharmacy, laboratory tests, semiprivate room, and routine nursery. These changes would be reflected over multiple outlets and cities; thus, a price change for the complete service would be reflected in the CPI through measuring its individual parts. The method of pricing by individual service does provide more detail, and potential changes in quality can more easily be identified than when pricing a more inclusive fee such as with the DRG. There are areas priced within professional medical services that are somewhat analogous to pricing the total treatment, as provided by the practitioner. Some of these are represented within physicians’ services by allergists, obstetricians, surgeons, and orthopedists for treatment of an asthma attack, hysterectomy, hernia repair, and total hip replacement. However, due to specialization by physicians and the need for non-physician-based professionals such as physical therapists, the treatment of many illnesses or injuries cannot be accomplished at one office. When pricing by treatment, the use of multiple resources would compound the difficulties of obtaining the total price of the treatment. The DRG fee also would not be a good means of reflecting the cost of the total treatment because it does not include separately billed physicians’ services that the patient utilizes for the designated DRG. Although it will be

142

Paul A. Armknecht and Daniel H. Ginsburg

difficult to develop the methods for pricing total treatment, BLS is planning further research in this area.

3.3.2 Development of a Flow-of-Services Approach to Transportation Services The purchase of a durable good such as an automobile has traditionally been treated as a consumed item in the CPI, that is, all consumer satisfaction is attained at the time of purchase. The fact remains that a durable good continues to provide the consumer with satisfaction over an extended period of time as in the case of home purchases. In the case of automobiles the purchase provides the consumer with a flow of transportation services over a period of years. In addition, the purchase represents the acquisition of an asset that continues to have a market value for some period of time while providing the consumer with transportation services. For purposes of the CPI, we are interested in the value of the consumed service as opposed to the purchase of the asset. Over time we want to measure the change in price of the transportation service provided by the automobile if it can be separated from the purchase of the asset. This is analogous to the situation with home ownership. The major difference is that the value of the asset generally depreciates over time in the case of automobiles but appreciates in the case of housing. As with home ownership, there are two methodological approaches that can be explored for separating the service from the asset. The first approach is to measure the user’s cost of the service. The second is to measure the rental value of the service. In a user cost framework the measure of the cost of consuming automobile services has three components: The first component is the opportunity cost that can be measured by the forgone return on investment. If the car were sold and its value invested, there would be earned interest that is forgone from holding the asset. The second component is the appreciation that might occur in the value of an automobile of constant quality. This would be the capital gain on the investment. Finally, there is the depreciation that would occur because of the aging and deterioration of the car. A study measuring the transportation services of an automobile using the user cost methodology by Blanciforti and Galvin (1984) found that the user cost index fluctuated erratically due to the capital gains component. The best available measure for the appreciation component was the change in price of constant quality used cars after adjustment for depreciation. This measure of capital gains demonstrated extreme instability and affected the user cost index similarly. The indexes were also sensitive to the choice of interest rates on equity and varied by over 300 percent during the test period of 1978-83. The Blanciforti-Galvin results are not uncommon when asset rental prices are estimated through a user cost methodology. Their results are similar to others who have empirically tested user cost formulations for measuring the

143

Measuring Price Changes in Consumer Services

rental prices of owner-occupied housing and capital goods. Gillingham (1980) found that there was extreme volatility in the capital gains term of the user cost function for housing prices. Even with the use of moving averages to smooth the rate of return and the appreciation variables, there were still large variations in capital gains. Harper, Berndt, and Wood (1987) found similar effects in the volatility of the capital gain term when measuring the rental price of capital goods. Whether using an internal or external nominal rate of return user cost model, they found the rate of return and capital gains to be quite volatile and the primary sources of variation in the estimated rental price. The other flow-of-services measurement alternative is a rental-equivalence measure. In a steady-state world of uncertainty with perfectly competitive markets and no tax distortions, it can be shown that the user cost of a durable asset is equal to its rental value (Gillingham and Lane 1982). Therefore, at equilibrium a rental-equivalence approach should yield the same price index as a properly measured user cost approach. With the extreme volatility that the user cost measure produces, it would seem appropriate to determine whether a rental-equivalence measure could be developed for automobiles. There are potentially two distinct rental markets for automobiles: The first, currently measured in the CPI, is for short-term rentals of cars for private use. The second market, which has begun to emerge for private users in the last few years, is that for automobile leasing. The short-term market represents the use of cars by consumers for occasional use such as vacations or emergencies (when their own car is disabled) and is almost entirely composed of rentals of new cars that are at most a year old. Depreciation on these automobiles generally would exceed those of consumer-owned cars because of the intensive use by a succession of renters. This market would not be representative of the general population of automobile consumers because the depreciation and capital gains components differ. The more appropriate market for a rental-equivalence measure is the longterm rental market for automobile leases. The purchase of the automobile by the consumer provides a flow of transportation services over an extended period of time. The leasing of an automobile by an individual is generally for a four- or five-year period. This extended rental may, therefore, be an appropriate measure of the rental equivalence for automobiles. An empirical concern is whether there is sufficient activity in the market to get continuous consumer prices over time. Examination of the leasing market by Forbes in 1984 indicated that somewhere between 40 percent and 50 percent of all cars produced are leased. Most of these are for business or government use. Estimates of retail leases (Automotive News, September 26, 1988) indicate that about 9 percent of retail deliveries of new automobiles are leased annually. (Retail leases combine personal-use leases to individuals along with leases of nine or fewer vehicles to businesses.) Information from the 1988 CE indicates that 3 percent of consumer units have consumer leases for automobiles. With the

144

Paul A. Armknecht and Daniel H. Ginsburg

phase out of tax deductions for personal interest spending in the U.S. tax code, consumer leases may increase in prevalence because the interest on auto loans would no longer be deductible. Automobile Leases

The lease transaction consists of three parts-origination, servicing, and financing. Any or all of these services may be provided by the leasing source: an automobile dealership, an independent leasing company, or a bank with direct leasing operations. Lease financing may also be provided by the automotive credit companies, a third-party program, a bank’s direct leasing office, or a bank on a wholesale basis. Wholesale bank financing is the simple funding of a lease without ownership or servicing by the bank. Financing by a bank or credit company involves the lending of money at fixed rates for the term of the lease with the lender holding the title and lease payments as security. There are two basic types of leases-the open-end finance lease and the closed-end or walkaway lease. The two differ with respect to who is responsible for the residual value of the vehicle at lease termination. At the start of the lease the lessor estimates the value of the automobile at the end of the lease term and takes into account the cost of the car and the anticipated depreciation during the term. This is often done using a published estimate of the residual value from an industry guide such as the Leasing Black Book. The residual value is needed in order to calculate the monthly lease payment. When the lease terminates, the actual market value of the automobile is determined from an independent industry source and compared to the estimated value. Liability for the difference can rest with either of the two parties. With an open-end lease the lessee assumes responsibility for the vehicle’s residual value. If the automobile is worth more at lease termination than was estimated, the lessee essentially has paid too much and is reimbursed for the difference. If the car is worth less, the lessee must pay the difference up to three times the monthly payment (the legal limit). Under a closed-end lease the lessee has no liability for the residual value. Whatever the case, the lessee is responsible for any damage beyond normal wear and tear or mileage in excess of a prespecified limit. The closed-end lease is the more prevalent. Originally, dealers were hesitant to offer closed-end leases because they may have to absorb the residual value as a loss. In 1982 the automotive credit companies initiated a nonrecourse lease program that guaranteed the residual values in the lease. The credit corporation retained the title to the automobile, and the dealer had no obligation to buy back the car at lease termination. Dealers became much more interested in leasing arrangements as a means to boost sales and the leasing market has grown ever since. The monthly payment in an automotive lease is based on the cost of the vehicle (which is negotiable as with a purchase), the expected depreciation during the term, interest and service charges, and a maintenance charge (if

145

Measuring Price Changes in Consumer Services

vehicle maintenance is to be included in the contract.) Registration, licensing, state taxes, and usage and sales taxes may also be paid outright or on a monthly basis. Insurance against loss or theft of the vehicle is the responsibility of the lessee. Thus, the consumer is purchasing the use of the automobile for a fixed period and paying the costs (including interest) of the automobile’s services. Test Pricing of Automotive Leases

BLS attempted to test the pricing of consumer leases at their point of origination in new vehicle dealers and leasing agencies during 1987 and 1988. We found that the volume of leases at these sources was insufficient to enable price collection on a monthly basis. Between the new car dealers and independent leasing companies, there may be as many as 15,000 sources for lease origination. Given the 1987 unit volume of 900,000 retail leases (which include some business leases), this implies an average annual unit volume of only 60 leases per outlet. When this volume is further divided among the three to ten nameplates typically sold in dealerships as well as the varieties of lease terms (three to five years in length), it was almost impossible to obtain a price for a new lease on a similar vehicle in consecutive periods. Because the pricing of leases at the point of origination is not feasible, we plan to investigate whether we can build a sample frame for lease finance sources and price leases at these outlets. Lease originators typically just arrange leases for large financing companies that purchase the car and lease it to the consumer. Lease financing is much more concentrated than lease origination. Five companies (GMAC, FMC, CC, GE Capital, and Marine Midland Bank) finance an estimated 35 percent of all leases originated each year. If we can successfully develop a sample frame, our existing pricing methodology, which was proven to be effective in the earlier test, can be employed. Our chances for success should improve because the financing outlets should have the necessary volume to quote an actual price for the selected sample of leases in most periods. We would like to field such a test once the sample frame is developed. If pricing at lease finance outlets proves successful, we will expand our sample and develop an experimental automotive leasing index. This could lead to the possible development of a leasing-equivalence measure of transportation services in the next CPI revision. Development of an automotive leasing index is just one step in the development of a leasing-equivalence measure. Under a flow-of-services approach expenditure weights would also have to be developed for the total value of automobile services consumed by families owning cars in the base period. Such an expenditure weight is derived by estimating the rental value of all automobiles owned by families in the base period. The rental value of each owned car could be obtained by asking owners of automobiles in the CE to estimate the rental value. It is not clear that all owners have sufficient knowl-

146

Paul A. Armknecht and Daniel H. Ginsburg

edge of the rental market to make such an estimate. Clearly, those who are about to or have recently terminated leases should be able to make such an estimate. Alternatively, the rental value can be estimated as a function of the depreciation in the asset value of the car over time. To estimate the rental value in this manner we need information on the asset value and stock of consumerowned vehicles in the base period from the CE. BLS has already conducted some research into estimating rental values for automobiles owned by CE respondents by using a model similar to those used by the leasing finance companies. Research in this area will also continue.

3.4 Summary BLS continuously strives to improve the measurement of price changes for consumer services in the U.S. marketplace. In the past this has occurred primarily at the time of a major CPI revision by the inclusion of more service items that are eligible for price collection. Beginning with the 1978 revision of the CPI, two new surveys were introduced-a continuing CE and the current point of purchase survey (CPOPS). These surveys have markedly changed the ability to introduce new items into the CPI, to monitor shifts in expenditure patterns of U.S. consumers, and eventually, to update the market basket more frequently. New methodologies for sampling within the retail outlet have also aided in the inclusion of new services. Major conceptual changes have occurred in determining the price of the consumer services that should be measured in the CPI. The most notable change was for the price of shelter services that led to the use of a rentalequivalence measure of home ownership. In addition the price measurement of health, tenants’ and automobile insurance have changed as a result of research into better methods for measurement of the appropriate service that is being provided to the consumer. Considerable research is underway to improve BLS’s ability to measure only pure price change in the CPI and exclude price shifts that are the result of changes in quality. The use of hedonic regression models that isolate factors that contribute to quality changes appear promising. There will also be a continuing effort to explore alternate measures for medical care services as new technologies are introduced at a rapid pace. Finally, BLS continues to evaluate new measures that will resolve the problems of including the price change of consumer assets as measures of price change for consumed services. To this end, we are testing the feasibility of a leasingequivalence measure for automobile transportation services. If the leasing market for individual consumers grows to become a viable rental market, we may be able to improve our measurement for this consumer service.

147

Measuring Price Changes in Consumer Services

Appendix A m u

...

IIJB

ELI CHECKLIST

June 1975

avrrer code

ELI Code

b I k t Codc

oc-.

n-

Keq*. I,,,,

IPR-

“Tl’ic

.~. .

--._.

I _

01

lim PAFLJ jiRvICES KIR R34LCS --

iY-5 OF SERVICES

rnNEV/Wrn E99 Permanerx , brand, D1

Razor

COW,II.X; H1 H2

@ J1 52

K1 il

Dye. all o v e r Dye, t o u r h up Rinse F r o s t i n g , w i t h cap Frosting, with aluminum f o i l Streaking Tipping

I

F 1 Shampoo @Shampoo and Get

WIG SERVICE

AB99

AC1

V 1 Facial

w1 Manicure X1 P e d i c u r e

R 1 Cleaning S1 S e t t i n g T1 Shampoo

U1 For f u l l wig U2 For h a i r p i e c e

Y 1 Electrolysis AA99 Other,

COWLETE THE FOLLOWING ITEMS ONLY I F THEY ARE PRICE FACTOKS

AGE

Weekdays Weekends Other,

OF CLIENT

MPERIEKE ff BWllICIKi

Long Short

,2299 Data Collector comments pertinent to item priced:

1

Adult AD2 Child AD3 Student AD4 S e n i o r c i t i z e n

L€N3H OF MIR @

M 1 Bleaching, a l l over M 2 Bleaching, touch up N1 Conditioning P1 S t r a i g h t e n i n g

OTHER PERSOIJ9L CARE SERVICES

Q1 S t y l i n g

T I E CF MY OR KEK AB1 AB2

OlHER tWIR SERVICES

A E 1 Most experienced o p e r a t o r AE2 Regular s t a f f AE3 S t u d e n t

-_

148

Paul A. Armknecht and Daniel H. Ginsburg

65011-01 Beauty p a r l o r s e r v i c e s f o r females ELI DEFINITION - Includes a l l types of s e r v i c e s provided by beauty p a r l o r s f o r wmncn and g i r l s , such a s h a i r c u t s , shampoo, s e t , permanent, f a c i a l , manicure, pedicure, h a i r c o l o r i n g , bleaching, t i n t i n g and c a r e of wigs and h a i r p i e c e s . Hair c a r e o r o t h e r personal c a r e products a s well a s wigs o r h a i r p i e c e s purchased in beauty p a r l o r s a r e not included here. POPS

DEFINITION - POPS 3 - Personal c a r e % - v i c e in beauty s a l o n s and barber shops f o r females - H a i r c u t s , c o l o r i n g , s t y l i n g , manicures, f a c i a l s , and treatments in beauty salons o r barber shops.

This POPS category c o t i s i s t s of only one ELI: Fema 1es . F r o s t i nq

65011, Beauty Parlor Services f o r

Bleaching and/or dyeing smal 1 groups o f h a i r s t r a n d s a l l over t h e head.

S t r e a k i n q - Bleaching and/or dyeing groups of h a i r s t r a n d s a t i n t e r v a l s over the head, o f t e n around t h e f a c e o r down the p a r t . Tipping - Bleaching and/or dyeing t h e ends of t h e h a i r i n random f a s h i o n . Electrolysis

-

T h e d e s t r u c t i o n of h a i r r o o t s by e l e c t r i c needles.

DISAGGREGATION SUGGESTIONS - Type of s e r v i c e , any o t h e r p r i c e f a c t o r

149

Measuring Price Changes in Consumer Services

+

Bureau uf Labor Slat6licr Oisaggrsgation Sheet Form -A

O.M.B. NO. 122ow39

46

28.78

5.39.72

116.41.68.91

181 42

31.81

19.53.86

Zd.49.74.99

(91 67

49. W

18.52. 85

15.10.65.90

78

22. 72

12. 46. 79

5.30.55.80

171

iioi

7

14

114. 29. 43, 57. 71. 86.100

150

Paul A. Armknecht and Daniel H. Ginsburg

iai 142

131.81

119.53.86

fat 167

149.99

l18.52.85 /15.40.65.90

122.72

112.46.79 15.30.55.80

IIIOI 178 E p u l -Illy

12A.49.74.99

hl

wumbr 01 ltnm

Pan ~n Col"rn"C

Pest tn Column D

2

50

50.100

3

23

33. 67. 100

4

25

25. 50. 75. 100

5

20

20, 40. 60. 80.100

6

16

17. 33. 50. 67. 83. 100

7

14

14.

a.43, 57, 71. 86. loo

151

Measuring Price Changes in Consumer Services

Appendix B US. Department of Labor

Bureau of Labor Statistics

ELI Checklist Collectlo" P.r,od

CP.

QR

Varrla" code

ouote coda

Outlet Code

oc=

vc=

oc=

-

Dspt /Arrangement

Revondent

AR-

PUC*

Quantlt"

PR= Field Agent Mossage

QT=

sm

sz-

Footnoter

OrOg,"

FN-

OG-

FM=

ELI Number and Title

SPECIALTY A1 F a m i l y A2 A3 A4

Cluster Code

pmrws

5601 1

o r a c t i c e (FP) General p r a c t i c e ( G P ) General p r e v e n t i v e m e d i c i n e (GPM) I n t e r n a l m e d i c i n e (111)

VE-

SPECIALTY.SEt€CTP IS ME FliYSICIW'S: 61 82

B99

Primary s p e c i a l t y Secondary s p e c i a l t y Other,

TYPE OF SERVICE - VISITS

c1 c2 c3 c4

C1 OFFICE VISIT, a d u l t , m i n i m a l s e r v i c e C2 OFFICE VISIT, c h i l d , b r i e f s e r v i c e C3 HOSPITAL VISIT, a d u l t , i n i t i a l c a r e C4 HOSPITAL VISIT, a d u l t , subsequent c a r e C5 HOME VISIT, a d u l t , l i m i t e d s e r v i c e C6 EMERGENCY DEPARTMENT V I S I T , adolescent, intermediate service C7 CONSULTATION, comprehensive C8 HOSPITAL VISIT, r o u t i n e newborn c a r e

c5

C6

c7 C8

I ESCRIEE SPECIFIC VISIT S€I.E"l

299

01

ON WERE

I

Pall 1" Column D

of Items

Number

Post ,n Column C

2

50

50,100

4

25

25,50,75,100

5

20

20,40,60,80, 100

6

16

17.33,50.67,83.100

8

12

12,25,37,50,62,75,87,100

Data COllector commantr pertinent 10 ilem priced:

BLS 3 u M B (Rev. Octobsr 19751

504.3

eWi& September 1977

Paul A. Armknecht and Daniel H. Ginsburg

152

M,W,IEFJ

a d u l t , minimal s e r v i c e

SPECIFIC SERVICE

Established Datient Routine immunization f o r tetanus G1 Administered by nurse El F1

0 f f F I E VISITJ CPT 90000

HW

VISIT,

CPT 90150

?%E'&hl-fALW r E -W&!6T?%I4 % E5 F5

Established p a t i e n t Review o f r e c e n t h i s t o r y

H5 Ausculation o f h e a r t I 5 Adjustment o f medication

C6 EEKENCY

CPT 90200

F6 G6

a d u l t , i n i t i a l care

H6 I6

SPECIFIC SERVICE

E3 Naw p a t i e n t F3 B r i e f h i s t o r y and p h y s i c a l examination 63 I n i t i a t i o n o f d i a g n o s t i c and treatment programs H3 P r e p a r a t i o n of h o s p i t a l records

Review o f r e c e n t i l l n e s s Examination o f pharynx, neck, a x i l l a , g r o i n , and abdomen Interpretation o f lab tests P r e s c r i p t i o n o f treatment

C7 VNSIJLTATIW, CPT 90620

comprehensive,

SPECIFIC SEF/IE

E7 D e t a i l e d h i s t o r v F7 Thorough p h y s i c a l examination G7 Diagnosis o f i l l n e s s H7 Recommendation o f treatment I 7 In-office

-. -.__

C4 HOSPITAL VISIT,

a d u l t , subsequent care, extended s e r v i c e , CPT 90270

SPECIFIC SEWICE

E4 E s t a b l i s h e d p a t i e n t F4 D e t a i l e d review o f r e s u l t s of d i a g n o s t i c e v a l u a t i o n (Zncludes d i s c u s s i o n o f : physical findings, l a b o r a t o r y studies, x-ray examinations, d i a g n o s t i c conclusions, recommendation f o r treatment)

nEPA m .NT (W1)Y S I T

adolescent, i n t e r m e d i a t e s e r v i c e CPT 90521

€2 New p a t i e n t F2 Throat examination f o r a c t i v e t o n s i 11it i s

C3 HOSPITAL VISIT,

@ HGPITAL \/ISIT, r o u t i n e newborn care CPT 90285

SPECIFIC XWIE

€ 8 Physical examination o f baby F8 Conference w i t h parents

-

ADDITIOVL SERVICES INCLUW) IN FEE

h M

65 Determination o f b l o o d pressure

child, b r i e f service

SPECIFIC SENICE

adult, l i m i t e d service

OTHER CLAM FYING DATA

T99

w99

u99

x99

v99

Y99

153

Measuring Price Changes in Consumer Services

U.S. Department of Labor

3ureau of Labor Statistics i L I Checklist hllection Period

:P=

oc=

____

-

auote code

Outlet Code

Version Code

ac=

vc=

__

Dept.lArrangement

1espondcnl

1R=

AR=

'rxe

auantity

Size

Footnotar

OUp

'R=

OT=

sz=

FN=

aG=

'ield Awn1 Meuaga :M=

ILI Number and Title

CWSTER 01 - ENERN '"lICAL SPECIALTY A1 A2 A3 A4

Cluster Code

PHySICIppss ' SF WI tFS

5601 1

01

VE=

PRMTICE

F a m i l y p r a c t i c e (FP) General p r a c t i c e (GP) General p r e v e n t i v e rnedijcine (GPM) I n t e r n a l medicine (IM)

SPECIALTY SELECTm IS ME PHYSICIPN'S B1 B2 B99

Primary s p e c i a l t y Secondary s p e c i a l t y Other,

TYPE CF SERVE I - PROCWJRES 01 ELECTROCARDIOGRAM D2 BLOOD COUNT D3 URINALYSIS, r o u t i n e D4 CYTOPATHOLOGY (pap smear) D5 RADIOLOGIC EXAM o f forearm D6 RADIOLOGIC EXAM o f upper gastrointestinal tract 07 B A S I C AUDIOMETRY SCREENING TEST D8 B A S I C METABOLIC RATE MEASUREMENT

01

02

03 D4 D5 D6 07 08

j 1 :! I

I

E Q U A L P H O B A B I L I ~ TABLE Y Pull 8" Column 1)

2

-

25.50.75. 100

20,40,60,80, 100 17,33,50.67,83, 100 .~ 14,29.43,57,71.X6,lW

~

I2

1

12.25.37.50.62.75.87.1M)

~.

3LS MOOE (Rw. 0ctab.r 1975)

504 :

Revised September 197'

154

1

Paul A. Armknecht and Daniel H. Ginsburg

D1 ELECTROCARDIOrlPPM CPT 93000 SPECIFIC SERVICE 51 KI L1

Monitorins electrocardioqram Interpretation Report

I

D2 BLOOD CWM, CPT 85010

complete

SPECIFIC SERVICE

55 A n t e r o p o s t e r i o r view K5 L a t e r a l v i e w

D6 PflDIOLUiIC Wl o f

upper g a s t r o i n t e s t i n a l t r a c t , CPT 74240

SPECIFIC SER/IlT

SPECIFIC SERVICE

56 W i t h o u t K.U.B. K6 W i t h d e l a y e d f i l m s

52 RBC K2 WBC L 2 HGB M2 D i f f e r e n t i a l

D3 URINALYSIS, r o u t i n e , CPT 81000

complete

SPECIFIC SERVICE 53 K3 L3 M3

N3

PH Specific gravity Protein Tests. f o r r e d u c i n g substances such as glucose U s i n g microscopy

D7 BASIC AUDImTRY SCKENING TEST CPT 92551

SPECIFIC SERVICE 5 7 Pure t o n e K7 A i r o n l y

D8 BASAL m A B O L I C RATE ,F@UEEW CPT 89000

SPEC1FIC SEN1CE J8

P e r f o r m i n g BMR t e s t

SPECIFIC SERVICE 54 K4

Smears from g e n i t a l source Screening 2 s l i d e s

ADDITIONAL SERVICES INCLUW) I N FEE

OTHER CLARIFYING DATA

T9 9

w99

u99

x99

w99

Y 99

I

References Armknecht, P. A., and D. Weyback. 1989. Adjustments for quality change in the U.S. consumer price index. Journal of Oflcial Statistics 5: 107-23. Blanciforti, Laura A., and John M. Galvin. 1984. New approaches for automobiles in the CPI. In Proceedings of the Business and Economic Statistics Section, 64-73. Washington, D.C.: American Statistical Association.

155

Measuring Price Changes in Consumer Services

Early, J. F., and J. H. Sinclair. 1983. Quality adjustment in the producer price indexes. In The U S . national income and product accounts: Selected topics, ed. Murray F. Foss, 107-42. Chicago: Univ. of Chicago Press. Feldstein, Paul J . 1988. Health care economics. 3d ed. New York: Wiley. Ford, I. K., and P. Sturm. 1988. CPI revision provides more accuracy in the medical care services component. Monthly Labor Review 1 11: 17-26. Francois, Joseph F. 1989. Estimating homeownership costs: owners’ estimates of implicit rents and the relative importance of rental equivalence in the consumer price index. American Real Estate Urban Economics Association Journal 17237-99. Gillingham, Robert F. 1974. A conceptual framework for the revised consumer price index. In Proceedings of the Business and Economics Statistics Section, 246-52. Washington, D.C.: American Statistical Association. . 1980. Estimating the user cost of owner-occupied housing. Monthly Labor Review 103:31-35. . 1983. Measuring the cost of shelter for homeowners: Theoretical and empirical considerations. Review of Economics and Statistics 45:254-65. Gillingham, Robert F., and Walter Lane. 1982. Changing the treatment of shelter costs for homeowners in the CPI. Monthly Labor Review 105:9-14. Griliches, Z. 1971. Introduction: Hedonic price indexes revisited. In Price indexes and quality change, ed. Z . Griliches, 3-15. Cambridge, Mass.: Harvard Univ. Press. Harper, Michael J., Ernst R. Berndt, and David 0. Wood. 1987. Rates of return and capital aggregation using alternative rental prices. BLS working paper no. 170. Washington, D.C.: Bureau of Labor Statistics. Kosary, Carol L., John P. Sommers, and James M. Branscome. 1984. Evaluating alternatives to the rent estimator. In Proceedings of the Business and Economic Statistics Section, 410-12. Washington, D.C.: American Statistical Association. Lane, Walter F. 1979. The costs of home ownership. SellerlServices September/October. (Reprinted by Federal National Mortgage Association, Washington, D.C.) Lane, Walter F., and John P. Sommers. 1984. Improved measure of shelter costs. In Proceedings of the Business and Economic Statistics Section, 49-56. Washington, D.C.: American Statistical Association. Leaver, Sylvia G., William L. Weber, Michael P. Cohen, and Kenneth P. Archer. 1987. Determining an optimal item-outlet sample design for the 1987 U.S. consumer price index revision. In Proceedings of the 46th Session of the International Statistical Institute, 173-86. Geneva, Switzerland: International Statistical Institute. Liegey, Paul R. 1990. Quality adjustment of apparel commodities in the CPI. Paper presented at NBER Conference on Research in Income and Wealth, Workshop on Price Measurements and Their Uses, Washington, D.C., March. Marcoot, John L. 1985. Revision of the consumer price index now underway. Monthly Labor Review 108:27-39. Norwood, Janet L. 1981. Statement regarding changes in the consumer price index. BLS News, release no. 81-506. Washington, D.C.: Bureau of Labor Statistics. Pollak, Robert A. 1989. The theory of the cost-of-living index. New York: Oxford Univ. Press. Triplett, J. E. 1969. Automobiles and hedonic quality measurement. Journal of Political Economy 77:408-17. . 1971a. Quality bias in price indexes and new methods of quality measurement. In Price indexes and quality change, ed. Z. Griliches. Cambridge, Mass.: Harvard Univ. Press. . 1971b. The theory of hedonic quality measurement and its use in price indexes. BLS staff paper no. 6. Washington, D.C.: Bureau of Labor Statistics. Triplett, J. E., and R. J. McDonald. 1977. Assessing the quality error in output measures: The case of refrigerators. Review of Income and Wealth 23: 136-57.

156

Paul A. Armknecht and Daniel H. Ginsburg

U.S. Department of Labor, Bureau of Labor Statistics. 1988. BLS Handbook ofMethods. BLS bulletin no. 2285. Washington, D.C.: Bureau of Labor Statistics.

Comment

Robert E. Lipsey

This is a very informative paper about the many improvements that have been made in measures of service prices in the CPI over the last 25 years. My comments are not so much on the paper itself as reflections on issues raised by the paper with respect to a few of the many topics covered. The price of health insurance has clearly been troublesome for the BLS, and the BLS has changed the method of calculating it a number of times without settling on an ideal one. The present method is to separate health insurance expenditures into a part that is a payment for medical care and a part that is a payment for insurance, and pricing the medical care part by costs of medical services. That seems appropriate to me on the ground that health insurance is mainly one possible way of paying for medical care, and the price of the policy depends largely on medical care costs. If that is the case, the expenditure should be placed under medical care rather than insurance, except for the amount that covers the profits and expenses of the insurance companies. That would confine expenditure on insurance to the costs of spreading risk over time and among consumers, the appropriate coverage, I believe. Of course, the price of medical services is itself a difficult item to measure. Aside from the usual quality issues, one could ask how a rise in fees attributable to a rise in malpractice insurance costs should be treated. If there were no transaction costs, all the additional payments would be returned to the consumers. In that case, should there be any rise in the price measure for consumers as a group? On the other hand, what if half the additional malpractice insurance costs went into legal fees? Would the properly measured price rise be different? Some part of the rise in the price of medical services seems to be paying not for these services but for the operation of a lottery or wealth redistribution program. Because those insurance costs are built into medical and hospital fees and are not paid directly as insurance by consumers, they appear as costs of health care. This issue is not discussed at all here but may deserve more attention from the BLS. In the case of automobile insurance, the CPI prices the policy rather than, as with health insurance, the price of the service bought with policy proceeds. As with the health insurance, the indirect method seems more appropriate, for reasons given in the paper itself. It is true that auto insurance prices become much more volatile by this method. But that may be because auto insurance Robert E. Lipsey is professor of economics at Queens College and the Graduate Center of the City University of New York and a research associate of the National Bureau of Economic Research.

157

Measuring Price Changes in Consumer Services

prices are in fact quite volatile. A rise in investment earnings by casualty insurance companies does make it cheaper to insure your car. In both health and auto insurance, the BLS faced the problem of dealing with changes in utilization, the average frequency of claims under insurance policies. Where the change is the result of decisions by consumers that affect the composition of the insured population, as in the example cited in the paper, it seems appropriate that it be excluded from the measured price change. On the other hand, where the change is the effect of an alteration in the environment, such as a rise in the level of air pollution or an increase in the frequency of car theft, it should be recognized as a price change. If the price of health could somehow be calculated, rather than the price of medical services, the environmental effects would be captured there. For the present, however, we probably must be content with observing the effects of these environmental changes in the price of insurance. The most troublesome problem with the pricing of medical services is the fundamental one of defining what is being purchased, and that issue is not gone into in the discussion of quality. Presumably, a patient is interested in purchasing a cure or at least an improvement in his condition, not X rays, not magnetic resonance imaging, and not even a certain amount of physician time, although the last may represent a utility to some patients. Perhaps the cost of an operation should be multiplied by changes in the probability of failure in each period. That may not be available for all procedures, but it would not seem impossible to develop such success data for some of them. One could imagine such information for cataract operations, for obstetrical procedures, and for other medical procedures. At least, this should be the goal that BLS sets for itself in this field rather than the price of a procedure itself. Measuring the price of services raises another problem that is not discussed here. That is the fact that the purchase of a service often involves a substantial input of the consumer’s time. That aspect of price is often considered in analyzing the demand for services, such as different types of transportation service or different types of retailing, but it does not enter price indexes, as far as I know. Yet, a shift to self-service can involve a rise in the price of retail services, if consumers must spend more time in searching for products and selecting them. A shift to weekend store hours or to automatic banking, on the other hand, involves a decline in price because the consumer can use less valuable time for his or her input into the transaction. There is no discussion of banking services in the paper, but an attempt to measure their price should surely take account of the shift to automatic teller machines. I think this is a fairly general issue for service prices that deserves some consideration. Of course, the consumption of goods also involves a time input by consumers. However, we evade the problem because the time expenditure comes at a later date than the purchase, and we define consumption as purchase, except in the case of housing. It is harder to evade the issue in the case of services because the consumer’s expenditure of time occurs simultaneously with the purchase of the service.

This Page Intentionally Left Blank

4

Productivity in the Distributive Trades: The Shopper and the Economies of Massed Reserves Walter Y. Oi

Goods are of little use unless they can be put in the hands of the ultimate consumers, Some direct sales are still observed, but the vast majority of consumable goods are channeled through middlemen specialists who facilitate the movement of goods in time and space, consummate mutually advantageous exchanges by matching buyers and sellers, and supply product information and ancillary services that reduce transaction costs. Retailers rarely charge explicit prices for their services that are demanded by both producers and consumers. They earn their remuneration by introducing a spread between retail prices and wholesale costs. Wholesale and retail trade, which accounted for only 6.1 percent of total employment in 1880, provided fully 20.6 percent of all jobs in 1980. The size of the distribution sector is considerably smaller when size is measured not by employment but by man-hours, labor costs, or value added. Whatever metric we use, the share of the economy’s resources allocated to distribution is steadily growing. Wallis and North (1986) defined a broader concept of a transaction sector that dealt with both the exchange of goods and the protection and enforcement of property rights. They estimated that in 1970 over half of the nation’s resources were allocated to the transaction sector. Economic theory has largely neglected the distributive trades. Henry Smith (1948), W. Arthur Lewis (1970), and Bob R. Holdren (1960) represent some notable exceptions who mainly examined the place of the retail establishment. Stores in these models are differentiated by location and behave like firms in a monopolistically competitive industry. Bucklin (1972) and Ingene and Lusch (1981) directed attention to the consumer who has to devote time and resources to shopping. As individuals acquired more cars and bigger houses, they changed their shopping behavior, which, in turn, affected the derived Walter Y. Oi is the Elmer B. Milliman Professor of Economics at the University of Rochester.

161

162

Walter Y. Oi

demand for retail services. Average transaction sizes got larger, and so, in combination with the economies of massed reserves, provided an environment that encouraged the development of large supermarkets. National advertising of standardized brands lowered search costs for customers who now had less incentive to inspect goods. They had less to gain by relying on the reputation of the Broadway to stock good underwear. They could simply buy Jockey. Some retail functions that were previously performed by retailers were shifted forward to consumers and others backward to producers. Stores are open longer, and retail clerks have less store-specific human capital. The product lines of supermarkets have been greatly expanded; those of gasoline service stations have been narrowed. The changing structure of the service bundle supplied by retail firms and the economies of massed reserves complicate the problem of measuring productivity for the distribution sector. The received theory of production has to be augmented by including the inputs of consumers and producers in measuring the rate of technical progress.

4.1 Growth of the Distributive 'kades When each household was largely self-sufficient, there was little need for the services of middlemen. Technological advances in transportation, agriculture, and manufacturing were responsible for the division of labor and the growth of large-scale enterprises. The propensities to truck, barter, and exchange expanded as resources were specialized. More transactions were required to move goods to their highest valued uses, and these could be more efficiently performed by specialists and institutions that make up the sector called the distributive trades. The growth of this sector was documented by Harold Barger (1955) in Distribution's Place in the American Economy. In 1900, 62.0 percent of the 29.1 million persons in the labor force were in the goods-producing industries (agriculture, forestry, and fisheries, construction, mining, and manufacturing); only 8.2 percent were engaged in distributing goods. By 1980, only 26.5 percent were producing goods, and 25.5 percent were employed in wholesale and retail trade; confer table 4.1 . I For every 100 persons who were producing goods in 1880, 8.6 workers were employed in distributing those goods. This figure climbed to 13.2 in 1900, 46.4 in 1950, and 77.1 in 1980. This dramatic shift in the industrial affiliation of the labor force overstates the sectoral reallocation of labor. Head counts ignore differences in the length of the work week and the skill mix of the work force. Over the period, 1900-1980, average weekly hours in the goods-producing industries fell from 51 hours to 40.9 hours, compared to a drop from 65 hours to 32.3 hours in trade. The ratio of 1. The first two columns, which were taken from Barger, refer to the labor force; the last two columns are census data on employed persons. The two series are not strictly comparable. To the extent that unemployment rates by industry vary, they could result in slightly different percentage distributions.

163

Productivity in Distributive Trades

Table 4.1

Employment, Hours, and Earnings by Major Industry, 1900-1980

Employees (thousands): All industries Goods producing Manufacturing Transportation & utilities Trade Other* Ratio, 100 (tradeigoods)' Average weekly hours: Goods producing Manufacturing Trade Wholesale Retail Ratio, 100 (tradeigoods) Work hours per week (millions): Goods producing Trade Ratio, 100 (tradeigoods) Average hourly earnings ($): Goods producing Manufacturing Agriculture Trade Wholesale Retail Ratio, 100 (tradeigoods) Ratio, 100 (tradeimanufacturing)

1900

1940

1950

1980

29,070 18,020 6,340 2,100 2,391 6,559 13.27

53,300 22,190 11,940 4,150 7,180 19,780 32.26

55,813 22,368 14,469 4,346 10,385 18,714 46.43

97,639 25,857 21,915 7,087 19,934 44,761 77.09

51 52 65

N.A. N.A. 127

43 38 48 41.3 43.2 112

43.8 40.5 40.5 40.7 40.4 92

40.9 39.7 32.3 38.5 30.2 79

919 155 17

954 345 36

980 42 1 43

1,058

0.128 0.175 0.087 0.175 N.A. N.A. 136.7 100.0

0.454 0.633 0.253 0.536 N.A. N.A. 118.1 84.7

1.39 1.44 0.75 1.26 1.48 0.98 90.6 87.5

7.16 7.27 3.90 5.34 6.96 4.88 74.6 73.5

644 61

Sources: Employees, hours, and earnings 1900-1940 are taken from Harold Barger (1955); tables 1 and 5, and A l . Data for 1950 and 1980 obtained from Historical Statistics and Statistical Abstract ofthe U.S. (selected issues). Note: n.a. = not available. *Other industries include services, construction, government, and so forth. 'Ratio is defined as employment in trade divided by employment in goods producing industries.

the labor inputs in trade and the goods-producing industries measured by employment counts, (EJE,) rose from .136 in 1900 to .771 in 1980, but when this ratio is measured in man-hours, (MT/MG), it climbed from . l l to .61.* The quality of labor has improved over time, but these improvements have 2. The data shown in table 4.1 pertain to hours paid rather than hours actually worked. Employees in retail trade receive fewer paid holidays and vacations. K. Kunze reported that in 1985 the ratio of hours actually worked to hours paid (H,/H,) was 0.914 in manufacturing and 0.964 in retail trade. The secular decline in this ratio was surely steeper in the goods-producing industries. If we could have measured man-hours actually worked, the labor input ratio would have been less than 0.61 in 1980.

164

Walter Y. Oi

been uneven across sectors. Three bits of evidence suggest that there has been a decline in the relative quality of the retail work force. First, the percentage of women in the work force increased for the economy as a whole, but the increase was even greater in retail trade. Many of the retail jobs that were created in the postwar period were filled by women and teenagers who were new entrants with little work experience. Second, part-time employees accounted for 13.0 percent of all employed persons in 1950 and 20.9 percent in 1980. In manufacturing, the reliance on part-timers changed little, from 8.5 to 9.6 percent. However, it nearly doubled in retail trade, climbing from 12.5 to 24.5 p e r ~ e n tThe . ~ rapid growth in the use of part-timers can, I believe, be explained by the changing nature of retail transactions. Third, the greater use of less skilled employees is corroborated by the data on relative wages. In 1900, the wage index shown at “ratio, 100 (tradeigoods)” under “average hourly earnings” in table 4.1 was 136.7, meaning that retail employees earned hourly wages that were 36.7 percent above the wages of employees in the goods-producing industries. This wage index fell to 90.6 in 1950, and by 1980 wages in retail trade were 25.4 percent below the wages in the goodsproducing industries. This decline is even sharper when manufacturing is the comparison group, as shown in the next row of table 4.1. At least three factors contributed to this drop in relative wages: (a) employment within retail trade shifted toward the low-wage three-digit industries; (b) the relative demands for part-timers and females who are paid lower wages climbed faster in retail trade; and (c) retail trade unions lost much of their market power. Adjusting the labor input for these quality changes would bring down the estimate of the growth of distribution. Another measure of the resource reallocation can be gleaned from the value-added data. From 1950 to 1980, the share of total GNP generated by the goods-producing industries fell from 41.1 percent to 3 1.9 percent. The ratio of value added in trade to that in manufacturing, (VA,~~)VA,,~) climbed from .725 in 1970 to .786 in 1987.4These data still indicate a relative growth of the distribution sector, but the size of the growth is far smaller than that indicated by employment counts. Greater specialization was accompanied by a rise in the fraction of goods handled by middlemen. The proportion of consumable goods passing through retail outlets increased from 72 percent in 1869 to 87 percent in 1929. A faster rate of technical progress in producing goods also contributed to a movement of resources away from the goods-producing sector. The Bureau of Labor Statistics (BLS) has assembled data on output measured by value added X , em3. The three-digit industries within retail trade exhibited considerable dispersion in the percentage of employees on part-time work schedules. Over the 1950-80 period, it climbed from 11.4 percent to 39.1 percent in general merchandise stores, 26.6 percent to 45.3 percent in food and dairy stores, and 8.9 percent to 26.9 percent in gasoline service stations. 4. The value-added data were taken from table 9.12 in the Economic Report ofthe President, 1988.

165

Productivity in Distributive Trades

ployment E , and total hours H . These data shown in panel A of table 4.2 allow us to measure the relative importance of four sectors: goods G, manufacturing M , wholesale trade WT, and retail trade RT. The patterns are slightly different from those indicated by table 4.1, but they confirm the relative growth of the distributive trades. Labor productivity measured by output per hour in panel B, increased at an annual rate of 2.60 percent in goods, 2.69 for manufacturing, 2.45 for wholesale trade, and only 1.76 percent for retail trade. The time path of labor productivity for the entire period, 1947-87, is presented in figure 4.1. At the three-digit level, the BLS embraces a sales measure of output. These data are shown in table 4.3 for food stores E department stores D,and gasoline service stations G.5Between 1967 and 1987, total output in constant dollar sales increased by 35.0 percent for food stores, 124.5 percent for department stores, and 50.3 percent for gas stations. Labor productivity (output per hour) grew at annual rates of 0.72 percent for food stores, 3.09 for department stores, and 4.00 for gas stations.6 Using a sales measure of output, Ratchford and Brown (1985) estimated that for the period 1959-79 total factor productivity (TFP) increased at a rate of 0.47 percent a year for food stores, and at 2.07 percent for manufacturing.’ The slower growth rate of TFP in food stores in relation to manufacturing led these authors to conclude that in the years ahead an increasing share of the economy’s resources will have to be devoted to distribution. I shall argue that not all the productivity changes are the results of exogenous technical progress, but they can, in part, be traced to changes in the organization of production.

4.2 The Output of a Retail Firm Firms engaged in producing goods-extruding aluminum, fermenting grapes, or growing catfish-must establish channels through which their goods can reach the ultimate consumer. Direct sales by farmers and manufacturers were not uncommon at the turn of the century. In such an economy of vertically integrated firms, there is little need for a distribution sector. How5 . Hall, Knapp, and Winsten (1961) used a sales measure of output that assumes a fixed proportions technology. Changes in the gross margin are attributed to changes in input prices. A margins or value-added measure of output assumes that changes in the gross margin are due to changes in the services supplied by the store. Input prices are presumed to move in proportion to changes in retail prices. 6. These growth rates are the regression coefficients of log-linear trend equations using all of the available data. The r-statistics are reported in parentheses in table 4.3. 7. TFP based on a margins measure of output led to a rate of technical progress of 0.34 percent a year. The TFP based on a sales measure of output fluctuated over the two decades. It grew at a rate of 1.20 percent a year for the first seven years 1959-66, slowed to 0.91 percent for the next six years 1966-72, and actually declined at a rate of -0.62 percent for the 1972-79 period. The BLS estimates of output per hour in food stores also fell during the decade of the 1970s. Ratchford and Brown controlled for changes in the sales mix by constructing weighted averages of the annual rates of change in sales for three departments-groceries, meats, and produce.

166

Walter Y. Oi

Table 4.2

BLS Measures of Productivity by Major Industry Goods

A. Percentage distribution by industry: output: 1950 50.1 1970 46.0 1987 39.0 Employment: 1950 52.5 1970 43.3 1987 31.9 Hours: 1950 53.1 1970 44.7 1987 35.3 B. Index of output per hour (1977 = 100): 1950 46.8 1970 88.7 1987 120.8

C. Annual growth rate, 1950-87: output 2.55 Employment 0.17 Hours -0.05 Output per hour 2.60

Table 4.3

Manufacturing

Wholesale

Retail

27.1 27.2 27.1

6.5 8.2 9.4

12.6 11.6 11.9

30.1 30.5 21.4

5.6 6.6 6.8

16.6 19.8 22.2

28.2 30.6 23.4

5.5 6.8 7.2

17.4 18.9 19.0

49.8 80.3 132.9

48.1 84.3 117.6

59.3 87.0 113.0

3.24 0.59 0.54 2.69

4.28 2.05 1.79 2.45

3.07 2.32 1.30 1.76

Index of Output per Hour (selected retail industries) Food Stores

Index of output: 1958 1967 1979 1987

66.8 85.2 103.5 115.0

Index of output per hour: 1958 72.0 1967 95.5 1979 98.3 1987 92.8 Regression of log output per hour: Slope ,0072 r-value 4.10 Period 1958-1987 Nore; n.a. = not available.

Department Stores

Gas Stations

N.A. 64.6 107.7 145.0

N.A. 75.4 94.5 111.3

77.2 104.4 137.2

63.2 107.4 145.7

,0309 27.60 1967-1987

,0400 49.63 1963-1987

167

Productivity in Distributive Trades

113.8671 113.867-

- ------- --

A. Nonfarm Business

---.

HOURS/EMPLOYEE

.* .*

x

Q)

.** OUTPUT/HOUR

67.94

-

*.. 1

1

1

120.207-

- \

I

30

'-HOURS/EMPLOYEE '-* \

-

71.9828

I

Year

-

v 2

1

- *** I

..

...=OUTPUT/HOUR I

I

I

1

168

Walter Y. Oi

ever, as specialization developed, the number of transactions increased at an exponential rate. The emergence of a distribution sector was a logical step toward the goal articulated by 0. E. Williamson (1979), namely, to minimize the sum of production and transaction costs. Retail firms today supply composite bundles of services that may include some or all of the following: ( 1) exchange-they consummate transactions that transfer property rights to the goods that they handle; (2) a product line-they assemble and display an array of goods that are made available to customers, and they jointly supply product information; (3) convenience-they offer this product line at a location and time (store hours) that have the effect of reducing transaction costs; (4) ancillary services-they sometimes provide delivery, credit, and implicit warranties; and (5) production-they may engage in packaging and processing goods to put them in a more suitable form for the customer.* The derived demand for retail services depends not only on the consumer demands for final goods but also on the mix of services that are jointly supplied by the retailer. If a store buys local advertising or sets aside shelf space for displays, it may obtain price concessions from the manufacturer. Self-service establishments, especially cafeterias, realize lower gross margins because they are supplying fewer point-of-sale services. Variations in the quantity and quality of services can result in a dispersion of retail prices across stores and over time.

4.3 Technology and the Economies of Massed Reserves

A production function usually refers to a technical relation describing how inputs of labor and capital can be transformed into an output, f ( L , K ) = X . However, the production of retail services shares many of the properties that characterize the production of education and transport services. It differs from manufacturing in at least two important respects: First, the consumercustomer supplies an essential input that has to appear as an argument alongside labor and capital, yielding a function,f(l, K , N ) = X . Second, demands are random, and delays are costly and result in a stochastic output. Producing a person trip from Oblong to Normal calls for the inputs of both the trip taker as well as the transport mode. The duke of Buccleuch and one moral philosopher constituted the inputs that produced one qualified student who could matriculate at Oxford. Without a customer, a retail firm could not produce a transaction which is the raison d’Ctre for its existence. Time and resources have to be allocated to shopping, to search for the right product or the proper price, and to arrange to get the goods home. We usually carry our fresh fish home but ask to have our firewood delivered. Local advertising can obviously reduce search costs. The particular functions that are performed by the retailer 8. A retailer who jointly supplies delivery and credit is engaging in downstream vertical integration into transportation and finance. Making baked goods, prepared salads, and canned goods under private labels exemplifies upstream vertical integration. Lunch counters at variety stores and baby nurseries at department stores are other examples mentioned by Barger.

169

Productivity in Distributive Trades

and those that are left to the customer or manufacturer jointly determine the output of the retail trade sector. A retail firm ordinarily acquires the property rights to the goods it sells. (Exceptions are goods sold on consignment and sometimes catalog sales.) Stores maintain inventories, hire clerks, and stay open even when they have no customers. On the other side of the exchange, customers may have to wait to be served. Someone or something is almost always waiting. Idle resources are, however, productive when they are in a state of what W. H. Hutt (1939) called “pseudo-idleness.’’ All idleness could, in principle, be eliminated, but, to accomplish this, the synchronization of the arrival rates of customers, clerks, and just-in-time inventories would be prohibitively expensive. We can find idle resources in the goods-producing industries, but neglecting this idleness appears to pose no serious analytic or empirical difficulties. This is not so for the distributive trades. The cost functions of retail firms exhibit increasing returns that can, in part, be traced to the economies of massed reserves. E. A. G. Robinson (1958) pointed out that these economies are a consequence of the coordination and synchronization of activities that can be achieved only by large firms. They do not result from the law of large numbers. A clearer exposition of this distinction was provided by J. G. Mulligan (1983) and A. S. DeVaney (1976). These economies of massed reserves characterize the production function applicable to a retail firm. Retailing is, in some respects, similar to the repairman’s p r ~ b l e m In . ~ this problem, a firm has M machines. The probability of a breakdown follows a Poisson distribution with a mean arrival rate of A. The time required to repair a machine is exponentially distributed with a mean service time p.. If the firm employs only one repairman, there is some probability po that none of the machines will need servicing. In this event, the repairman is idle. If two or more machines breakdown, a queue develops, and unproductive, idle machines have to wait to be repaired. The addition of a second repairman raises labor costs but reduces the opportunity costs of idle machines. If A and CI. are technically fixed, one can solve for the optimum ratio of machines to repairmen ( M / R ) , which minimizes the sum of idle times of machines waiting to be repaired and of repairmen waiting for the arrival of a broken machine.

M R

=

1

+

();-I

This problem is isomorphic to one where a hospital serves a population of M potential patients who arrive to be treated. A Poisson distribution describes the probability of the number of patient arrivals. If the mean duration of a hospital stay is k, the sum of waiting times (empty beds waiting for patients 9. This is one of the queuing models analyzed by Gross and Harris (1974). Economic applications can be found in DeVaney (1976), Arrow, Levhari, and Sheshinski (1972), Syrquin (1972), and Levhari and Sheshinski (1970).

Walter Y. Oi

170

and sick patients lining up for a vacancy) is a minimum when the population to beds ratio satisfies equation ( 1 ) . If M and R are both doubled, the mean length of the patient queue falls, and the occupancy rate of hospital beds rises.'O These economies of massed reserves generate a cost function that exhibits increasing returns; unit costs are inversely related to firm size. The cost advantage enjoyed by the largest store has to be set against any cost disadvantages incurred by customers who have to travel longer distances or who have to wait in longer customer lines.

4.4 A Full-Price Model of Consumer Demand Goods are not acquired in continuous flows. They are purchased in discrete lots or batches. Time and resources are allocated to the complementary activities of buying, carting, and storing consumable goods. The costs of these activities are incurred by both the consumer and the retailer. In this section, attention is first directed to an inventory model that determines the size of each transaction and hence trip frequency. Next, this model is extended to analyze how a customer chooses a store. Finally, I examine the way in which the full price is affected by selling efforts and search costs. 4.4.1 An Inventory Model and the Optimum Basket Size

Following T. M. Whitin (1952), the total cost of consuming Q units of a single good is the sum of three components: (1) expenditures for the good, PQ; (2) shopping or setup costs, C , = ST, where S is the implicit cost of a trip, and T is the trip frequency per month; and (3) home-inventory costs, C,

13

= h - , where h is the unit cost of holding a home inventory whose average

size is

13 -

. When (is Q , S, h ) are all exogenous, total cost is a function of only

one decision variable, the basket or transaction size q. C = C(q) = PQ

(3 (3

+ C , + C,, = PQ + S - + h -

A larger basket reduces total cost if it reduces C , by more than it raises C,. The optimum basket size balances these two opposing effects and is attained when -C,' = CHI. (3)

q* =

10. Consider a community with several hospitals. Patient amvals are proportional to bed capacities. If the largest hospital has 60 percent of all beds in this community, it gets 60 percent of all patient arrivals. If all of the beds in a particular hospital are filled, the patient has to wait for a vacancy or balk and move to another hospital. The likelihood of this situation is inversely related to hospital size. The queuing model implies that the largest hospital realizes the highest bed occupancy rate. This principle is equally applicable to airlines. George Douglas and James Miller (1974) found that on a given route, the carrier supplying the largest number of available seats enjoyed the highest seat occupancy rate or load factor.

171

Productivity in Distributive Trades

Substitute for q* in c, and divide by Q to obtain the full price P*: (4)

P*

=

P

+y

=

P

+

g.

The full price is the sum of the retail price P set by the store plus an implicit buyer cost y, which is incurred by the shopper. Everyone who patronizes a given store pays the same price R but y can vary across customers resulting in a distribution of full prices. The implicit buyer cost is lower for those who confront lower cost parameters {S, h} and who demand larger total volumes per month Q. The square-root formula assumes that {S, h} are constants. Spoilage and limited storage capacity K‘ ought to produce a rising marginal holding cost function (MHC) like the curve depicted in figure 4.2. The optimum for this case results in a smaller basket q**, which is to the left of q*. A bigger refrigerator and more cupboard space shifts the capacity constraint K’ to the right, meaning a larger basket q** and fewer shopping trips per month. 4.4.2 Location and Store Choice The implicit trip cost S depends on, among other things, the distance D between home and store. Assume that this relation is linear, S = so s,D. ‘I Each customer is presumed to choose that store that provides him or her with the lowest full price. Consider a set of consumers who reside along a line of length L connecting stores A and B . If both stores charge the same price t store choice is determined by proximity. Store A captures everyone who resides to the left of the midpoint D,. If store A cuts its price, some customers located to the right of D , find that they can obtain a lower full price by going to the more distant store A, even though this means a higher implicit buyer cost. Because the implicit buyer cost is an increasing function of the distance to store A , (dy,/dD,) > 0, there is a critical distance D* at which the full price at store A is the same as that at store B . This critical watershed distance is located further from store A for those customers who demand more per month Q, face a lower unit inventory cost h, or incur a lower incremental trip cost s,. If Q, > Q2,individual 1 may patronize the more distant store A; person 2 finds that for him or her, the full price is lower by shopping at the closer store B . We could thus observe two individuals who live across the street from one another but who choose to shop at different stores. The number No who choose to shop at store A will depend on the size of the retail price saving (Pb - P,) and the distribution of customers, g(Z), which is the frequency of customers who face a cost penalty of traveling to store A of Z = (y, - yb). It is obvious that Z is negative for everyone who lives to the left of the midpoint D,. The customer traffic attracted by store A is thus

+

11. The value of the shopper’s time, the basket size q, the number of checkout lanes, and the travel mode could all affect the shopping cost parameters. The fixed component, so, will be larger for those who demand larger baskets; the incremental cost per mile, s,, is lower for those who drive rather than walk to the store. Although S is specific to the customer-store match, I shall assume that the trip cost parameters are the same for a given customer.

172

Walter Y. Oi

** *

q

q

Fig. 4.2 Optimum basket size

The incremental traffic due to a price change on the part of store A is simply the negative of the height of the frequency density evaluated at the size of the price saving, A = (Pb - P J ; that is, dN,/dP, = - g(A).’*The profitability of price competition will be greater, the larger is the elasticity of N , with respect to P a .

4.4.3 Retail Services and the Full Price Isaac Ehrlich and Lawrence Fisher (1982) developed a similar model in which the full price, P* = P + Vt, is the sum of the retail price P plus an implicit shopper cost that depends on the value of the shopper’s time, and the time required to purchase one unit of the good, t . This unit time requirement is inversely related to advertising A , in-store selling effort E , and the customer’s total purchase volume Q. If A and E are aggregated into a composite retail service input R , and if Q is held constant, the time requirement function simplifies to t = t(R) with t’(R) < 0. Ehrlich and Fisher argue that, if retailing is competitive, customers must confront the same full price, P*, at all stores.I3 Stores can still compete by cutting prices or supplying more retail services but are always subject to the constraint that P* is a constant. If 12. Total sales at store A depends on the conditional distribution of purchases Q given the size of the cost penalty 2. 13. This differs from my model of spatially differentiated stores where it is only the marginal customer located at D* in figure 4.3 who faces the same full prices at two competing stores. Those customers located nearer to a store enjoy an inframarginal rent.

173

Productivity in Distributive Trades

- Vt‘(R) > P,, (where P, is the price of the retail service input R ) , it pays to expand services because part of the added costs of more services can be passed on to customers via a higher price l? In response to an exogenous rise in stores will increase R by taking out more local ads that reduce customer search costs and supplying more in-store services that reduce the time needed to complete a shopping trip. Notice that in the Ehrlich-Fisher model the consumer is passive, but in an inventory model the shopper modifies his or her behavior to minimize the sum of shopping and inventory ~ 0 s t s . IIn~ a complete model, the implicit buyer cost, 7 , is jointly determined by actions taken by both the customer and the retail firm.

4.5 Pricing by a Monopolistically Competitive Store Each retail firm with its unique location has a limited amount of market power. The sales realized by a store depends on the price it sets P, its service level R , and a vector of exogenous variables Z , whose elements describe the socioeconomic characteristics of the neighborhood constituting its market area as well as the prices and service levels of competing stores. A price cut can expand sales by increasing either the number of customer transactions N or the average basket size q. If E is the price elasticity of the firm’s sales demand function, the store sets price P so that the marginal revenue is equal to the full marginal cost (FMC) which is the sum of (1) the wholesale cost P,; (2) the direct marginal handling or operating cost C,; and (3) the marginal transaction cost T ~Is .

14. The retailer is passive in the inventory model. The two models could be combined to allow for the joint minimization of the full price. The fixed component of my shopping-trip cost function could be expressed in a form analogous to Ehrlich and Fisher, namely, so = Vr,, where r, is the time input needed to search for the right store and to assemble goods at the store. The customer decides on the basket size and the choice of a store; the retail firm sets the price P and the service level R. 15. Profits for the retail firm are given by 1~

= ( P - Pw)X

-

C ( X , N) - P,R

Holding the service level R constant, the first-order condition for a maximum is

(dddP)

=

X

+ (P - P J X ,

- C,Xp - C,N,

=

0

Divide by the demand response, ( d X / d P ) = X , < 0, and define the marginal transaction cost as follows: TN =

cN(Np/xp),

If a price cut attracts no new traffic, N, = 0 and hence T~ = 0. Alternatively, if the basket size is unaffected by a price reduction, ‘TN

=

(cdq).

174

Walter Y. Oi

------ ---- -----_-----

a

D ,

D*

L

Fig. 4.3 Full price as a function of distance

Although this expression assumes that the store handles only one product, it is useful because it directs attention to the concept of the FMC, which exhibits increasing returns. Nearly all stores handle a product line and have to set prices for many related products. The pricing problem is formally identical to that analyzed by R. H. Coase (1946) and M. J. Bailey (1954). The utility of shopping at store A is reduced whenever the price of a product purchased in positive quantity is increased. A decrease in the price of a good that makes up a larger share of the budget has a greater effect on utility and thus attracts more customers, resulting in spillover demands for other goods in the product line. The markup of price over FMC is likely to be smaller for goods that are traffic generators with spillover effects. C. Bliss (1988) treated the retail firm as a multiproduct monopoly whose market power is limited by competition. The optimal spread between price and wholesale cost (operating costs were assumed to be fixed and invariant to sales volume) satisfies the Ramsey rule with smaller spreads for goods with more elastic demands. The surplus is just sufficient to cover the fixed operating cost. The price elasticities that determine the Ramsey prices are obtained from an indirect utility function applicable to a representative consumer who engages in one-stop shopping. To the extent that the same product is sold in several shops (which violates the one-stop shopping assumption but squares with the real world where cigarettes, shampoo, and aspirin can be purchased at a drugstore or supermarket), the pertinent price elasticities must be taken from the residual demand function facing a single spatially differentiated store. The store’s pricing problem is very different from that analyzed in the Bliss model. Setting prices for thousands of items poses a formidable problem. Some supermarket managers allegedly solve this problem by mimicking the price structure of a dominant firm such as the great A&P. The price data collected by Bob R. Holdren (1960) roundly rejects this allegation. For pricing pur-

175

Productivity in Distributive Trades

poses, Holdren claims that goods are placed into four categories: (1) items with externally fixed prices because of consignment selling or resale price maintenance; (2) goods whose prices are unnoticed; (3) goods with wide pricing latitudes because of ignorance, small budget shares, or diversity in product quality; and (4) highly competitive goods which make up his k class and whose prices are important in choosing a store.16The goods in the k class are strong traffic generators with spillover externalities. This latter point is nicely illustrated by Holdren who wrote: “The low margin on cornmeal was a surprise to this writer, but according to the supermarket operators, people who buy cornmeal, buy it relatively frequently and tend to be ‘careful shoppers’ and ‘big eaters’. Thus to attract and hold them, lowering the price of cornmeal is relatively efficacious” (p. 80). Location and distance pose a higher barrier to switching when consumers do not have cars. Additionally, price cuts yield greater returns when customers are like the cornmeal addicts who demand larger baskets on each trip. There is, thus, a strong interaction between a store’s pricing policy and the shopping behavior of consumers.

4.6 Price Competition and the Concentration of Food Stores In 1940 there was one food store for every 78 households, but by 1980 each food store served, on average, 481 households. Expenditures for food per household (in constant dollars) increased by 38 percent, and cars per household nearly doubled. The share of sales captured by chain stores rose from 35.2 percent to 46.7 percent. Sales per store in constant dollars doubled in the decade of the 1940s, doubled again in the 1950s, but remained stable from 1960 to 1970; confer table 4.4. R. Parker (1986) estimated that the average of the four-firm concentration ratios for a sample of 196 cities increased from 45.3 percent in 1954 to 65.8 percent in 1977. Data from the census of retail trade also conform this trend toward increasing concentration. Supermarkets with 50 or more employees accounted for 19.20 percent of all food sales in 1967, which increased to 45.67 percent in 1982. This pattern is also evident in the employment data. The percentage of industrywide employment rose from 19.8 percent to 40.4 percent. This section explores the reasons for these changes in average store size and market concentration. 4.6.1 Scale Economies in Retailing The empirical studies by Hall, Knapp, and Winsten (1961), Douglas (19621, and Bucklin (1972) revealed a positive relation between labor produc16. The preferred store is the one that provides the customer with the highest utility that depends on the vector of full prices and full income. The prices of goods in the k class allegedly have a greater effect on utility and are hence more important in choosing a store. According to Holdren, goods in the k class possess the following characteristics: (1) buyers are aware of the price; (2) price differences across stores are perceptible; (3) the good has a large budget share; (4) demand is predictable; ( 5 ) demand is relatively inelastic; and (6) a price difference will not be confused with a quality difference.

176 Table 4.4

Walter Y. Oi Food Stores- Sales and Related Variables, 1940-1980

No. of stores (in thousands):* Independent Chain Convenience Total

1940

1950

1960

1970

1975

1980

405.0 41.35

375.0 24.70

174.1 34.20

446.35

400.70

240.0 20.05 ... 260.05

209.30

142.73 23.08 25.0 191.80

112.6 18.70 35.0 167.10

36,534 22,316

40,331 36,619

40,080 38,056 3,124 81,260

41,353 40,501 4,870 86,724 367 2166 136 519

...

t

.

.

Sales (in millions of constant 1967 dollars):* Independent 16,563 22,752 Chain 9,034 13,611 ... ... Convenience Total 25,597 36,362 Sales per store (in thousands of constant 1967 dollars): Independent 41 61 Chain 218 530 Convenience ... ... Total 57 91

...

...

...

58,750

76,950

152 1108

232 1071

. . .

...

226

369

28 I 1649 I25 424

No. of households (in thousands)

34,949

42,867

53,021

63,450

71,920

80,390

No. of registered autos:? Privately owned (in thousands) Per household

27,372 0.78

40,191 0.94

61,420 1.16

88,775 1.40

106,077 1.47

120,866 1S O

Disposable income (in constant 1967 dollars):$ In billions of 1967 dollars 183.6 273.9 Per household 5,253 6,390

377.4 7,118

577.0 9,094

668.2 9,291

765.9 9,527

Sources: *Progressive Grocer (April 1983, 48, 66); tMVMA Facts and Figures; $Economic Report of the President (1986, table B.26).

tivity and store size, a relation implying that there are increasing returns to scale. B. Nooteboom (1983) assembled data for Dutch supermarkets with similar service levels in terms of the width of the product line, types of departments, and annual store hours. He initially assumed that the labor input measured in man-hours M was a linear function of annual sales X . where p, is the fixed labor input, and p, = (y, + y2) is the marginal labor requirement that is the sum of the labor time needed to handle another customer plus the time spent in stocking goods. Given a stochastic arrival rate of customers, the ratio of queuing to serving times is kept within narrow limits. Hence, y1 = g , / q is a constant, where g, is the mean waiting time, and the basket size q is a proxy for the mean serving time. Because X = N q , manhours is a linear function of the number of transactions or customer trips N and annual sales X .

177

Productivity in Distributive Trades

M i= p,

+ g,N, + y,X, + ei.

The fixed labor input Po is responsible for the increasing returns. This linear labor requirements function implies that the scale economies with respect to the average basket size q are greater than those with respect to the number of transactions N.” Other inputs such as floor space, equipment, parking, utilities, and advertising appear to be related to N and X in a similar manner. Increasing returns are evident in the U.S. data for food stores, which are shown in table 4.5. Labor productivity measured by sales per employee hour ( X l H ) is higher in larger supermarkets where size is determined by selling area. The effect of size is somewhat weaker for chain stores. Average transaction size, inventory turns, and the capital utilization rate measured by store hours are all positively related to store size. A given relative increase in sales volume X or in the number of weekly transactions N is accompanied by a less than proportionate increase in operating costs. Supermarkets that achieve larger size thus enjoy lower unit operating costs. 4.6.2

Implicit Shopper Costs and Price Competition

According to D. Appel (1972), the supermarket was the institutional innovation that was responsible for the increased efficiency of food distribution. The idea for a self-service, cash-and-carry store was evidently conceived in 1916 when Charles Saunders opened his Piggly Wiggly store in Memphis. The supermarket movement in the east was launched by Michael Cullen who opened his King Kong store at Jamaica, New York, in 1930 and the Big Bear in Elizabeth, New Jersey. These stores followed the Piggly Wiggly model. By locating outside of densely populated areas, they could obtain low rents, which enabled them to acquire large selling areas and parking space. Large sales volumes were generated by cutting prices of nationally advertised brands, and customers were attracted by heavy local advertising. The supermarket offered low prices, but it eliminated free delivery and credit. All transactions were on a cash-and-carry basis. Part of the store’s responsibility for assuring product quality was shifted to the manufacturers of branded goods. I s The success of a low-price strategy obviously depends on high-price elasticities of demand. The emergence of price competition in the 1930s can be explained by a model in which stores compete for customers who reside on a line connecting them. Customers who shop at store A realize a lower full price, P*, < P*, 17. Let 6M = ( d M / M ) denote a logarithmic derivative. The linear function implies that 0 < (GMISq) < (6M/6N) < 1 . D. Schwartzman (1968) recognized this fact and argued that much of the upward trend in labor productivity (through 1963) could be explained by the growth in the average size of transactions. 18. The rapid growth of trademarks in the 1970s has accelerated this shift; confer Landes and Posner (1983) and Pashigian and Bowen (1989). The historical development of the supermarket is more fully discussed by Appel(1972,42-44) and by Blozan (1986, 16).

178

Walter Y. Oi Selected Statistics for Supermarkets by Size and Ownership, 1988

Table 4.5

Independents Item

10-15

20-25

Weekly sales ($) Sales per employee hour ($) Store sales per hour ($) Average transaction size ($) Inventory turns Item stocked Inventory value (in thousands of dollars) Weekly transactions No. of checkouts Employees Employees full-time Employees part-time Ratio part-timeifull-time Store hours (mean) Open 24 hours, 7 days (%) Scanning (9%)

82,968 76.46

190,471 90.47

852.00 12.16

Chains

+

10-15

20-25

35+

37 1,655 91.74

112,365 89.51

176,778 87.82

340,229 92.87

1,619.00 17.02

2,676.00 21.28

937.00 12.39

1,422.00 15.38

2,357.00 19.34

14.2 12,190 250

20.7 17,775 403

17.9 25,932 894

15.0 11,408 340

15.1 18,024 504

14.3 27,151 1,038

6,823 5.0

11,191 7.7

17,465 13.2

9,069 6.2

11,494 7.9

17,592 11.6

16.2 21.2 1.31 96 7 42

31.1 46.2 1.49

46. I 108.9 2.36 138 50 100

14.7 33.9 2.31 122 27 52

27.3 44.0 I .6l 123 17 68

41.3 87.2 1.84

117

26 77

35

144

52 96

Source: Progressive Grocer, selected annual statistical supplements. Note: Store size is measured in thousands of square feet of selling area.

meaning that the implicit shopping cost penalty is less than the price differential; that is, they satisfy the inequality,

+

+

Let S = so s,D = V(t, t,D) denote the implicit cost of a shopping trip, where V is the value of the shopper’s time, to is the time spent at the store, and t,D is the time required to travel to and from the store. Recall that D is the distance to store A ; ( L - D ) is the distance to B . The cost penalty of shopping at A is thus given by,

where

Equation (7) describes a nonlinear transformation from D to Z , which depends on the parameter vector {k, to, t , } . If N consumers with the same value of k are uniformly distributed along the line, thenf(D) = NIL, but their distribution as a function of the cost penalty Z is described by a dome-shaped curve

179

Productivity in Distributive Trades

like g(Z) shown in figure 4.4.The maximum cost penalty of traveling to store A is incurred by the person residing next to store B: (9)

Zm,,

=

Z(L) =

k [ V q T q - fil.

Because Z is negative for those living to the left of the midpoint Dm,g(Z) is symmetrical with a mode at Z = 0 and a lower bound of Z(0) = -Z(L). The traffic attracted by a price cut on the part of store A depends on the height of the frequency distribution evaluated at the size of the price differential, dNJ dP,) = -g(A). Appel identified three factors that raised the price elasticity of demand and hence increased the returns to price cutting: First, the migration from rural to urban places resulted in higher population densities, which translate into proportional upward shifts in g o . Second, rising real incomes and larger families increased the demand for food. A higher demand Q reduces k, which pulls in the bounds, - Z(L) < Z < Z(L) of g(Z). If A’s price advantage remains constant, more customers will shift to A unambiguously increasing its sales, X,. Although the high-price store B loses customers due to a rise in Q , the net effect on its sales is indeterminate.lg Third, higher car ownership rates lower the cost of going to more distant stores. As f, falls, the bounds for g(Z) again move in toward the origin. In the limit as t , approaches zero, g(adegenerates to a spike at Z = 0, and any price reduction below the competing store means that the price-cutter captures the entire contested market. Some high-priced stores disappear, and the market areas of the remaining stores expand. In addition to urbanization, higher incomes, and car ownership rates, the returns to price competition were affected by home inventory costs and the value of time. A decrease in unit home-inventory costs h reduces k thereby increasing the demand facing the low-price store A. However, a higher value of time raises the costs of shopping at both stores. Because (dkldv) > 0, a rise in V flattens g(Z) and hence reduces N,. In an inventory model, a higher cost of time discourages price competition, but this model ignores other responses available to consumers and retailers. A store could broaden its product line so that a shopper can economize on shopping time by purchasing food, drugs, and sundries at one place. These responses are examined in section 4.7.3, below. To sum up, food stores and nearly all retail establishments are getting larger. Reference to table 4.6 reveals that, aside from apparel and eating/ drinking places, we have experienced an absolute decline in the number of retail establishments in every two-digit industry. The decline has taken place in the smaller employment size classes. The number of large retail establishments with 20 or more employees has increased in every two-digit industry.

+

19. If the elasticity of N, with respect to Q lies in the interval, - 1 < (SNJSQ) < 0, an increase in demand per household Q will lead to larger sales, X, = N,Q. The loss of customers is more than offset by higher demands on the part of the remaining customers.

180

Walter Y. Oi

-z(L)

0

A

+z(L)

Fig. 4.4 Distribution of customers by implicit cost penalty

Technological advances outside of the distribution sector are mainly responsible for these trends. The relative prices of cars, refrigerators, and advertising messages have declined. Consumers are prepared to incur higher implicit cost penalties to patronize stores that offer lower prices. By cutting prices and altering the service mix, supermarkets have succeeded in attracting more customers and generating larger sales volumes that can be supplied at lower unit costs because of the economies of large scale and of massed reserves. These developments were responsible for the improvements in labor productivity that were observed through the mid-1970s.

4.7 On the Organization of Production and the Product Line The comer grocer and the giant super belong to the same three-digit industry, SIC 541, but they differ in important respects that affect the relation of outputs to inputs. Attention is directed here to three aspects of this diversity. First, the capital to labor ratio and the rate of capital utilization surely affect labor productivity. Second, the composition of the retail work force has been influenced by the reallocation of distributive functions among the three participants-manufacturers, retailers, and consumers. Finally, I explore the reasons for changes in the output mix over time, and the effect of these changes on labor productivity. 4.7.1

Capital Intensity and Utilization Rate

The measurement of the capital to labor ratio is confounded by variations in the output mix and the quality of capital. The ratio of the book value of assets to employment is higher for larger supers, but selling areas and inventories per full-time equivalent employee are lower. Larger supers are more likely to have delicatessens, bakeries, and fresh fish, which call for less floor space but more capital equipment. Buildings and equipment are newer in big-

181

Productivity in Distributive Trades

Table 4.6

Number of Retail Establishments and Consuming Units, 1963and1982 No. of Establishments

Industry

1963

1982

198211963

No. of consuming units (millions): Families Households Population, total Population, 20 and older

47.5 55.2 189.2 115.3

61.4 83.5 232.3 161

1.293 1.514 1.228 1.396

1,731,055 63,449 32,584 164,595 86,27 1 109,506 126,194 88,918 288,3 15 47,423 232,686

1.130 0.725 0.559 0.569 0.962 0.605 1.154 1.024 1.000 0.91 1 1.037

121,039 10,168 16,199 32,802

0.544 0.682 0.957 0.884

158,064 4,935 13,948 29,041 13,537 2,175 6,765 3,291 70,221 5,288 8,863

2.258 1.994 1.343 1.745 1.328 4.232 1.337 1.598 4.449 1.957 3.618

26,561 393 1,681 615

1.673 5.866 3.410 5.694

Establishments operated entire year (SIC code): Retail trade, total 1,532,291 Building Materials (52) 87,499 General Merchandise (53) 58,264 Food stores (54) 289,073 89,651 Auto dealers (55x) Gas stations (554) 180,879 Apparel (56) 109,392 Furniture (57) 86,832 Eatingldrinking(58) 288,384 Drugstores (591) 52,063 Other (59x) 224,396 Selected 3-digit industries: Groceries (541) 222,442 Meat (542) 14,910 Retail bakeries (546) 16,935 Liquor stores (592) 37,093 Large establishments with 20 or more employees (SIC code): Retail trade, total 70,000 Building materials (52) 2,475 General merchandise (53) 10,383 Food stores (54) 16,644 Auto dealers (55x) 10,193 Gas stations (554) 514 Apparel (56) 5,058 Furniture (57) 2,059 Eating/drinking (58) 15,784 Drugstores (591) 2,702 Other (59x) 2,450 Selected 3-digit industries: Groceries (541) 15,874 Meat (542) 67 Retail bakeries (546) 493 Liquor stores (592) 108

182

Walter Y. Oi

ger stores that make more use of scanners.*OInventory turnover rates are positively related to size resulting in a lower inventory to labor ratio in large supers. If we adjust for differences in the output mix and equipment quality, the ratio of capital to labor is positively related to firm size. The fixed overhead costs of capital account for a larger share of total costs in larger firms, which thus have a stronger incentive to utilize capital more intensively by operating more shifts in manufacturing or establishing longer store hours in retailing. In 1988, the largest supers were open for an average of 138 hours a week, and half of them were always open. The small independents reported an average of 96 hours a week, and only 7 percent were open 24 hours a day, 7 days a week. Size had a weaker effect on store hours for chain stores. The outlets in a given chain evidently adopt similar operating practices, including store hours. Over the last two decades, average store hours have steadily increased-possibly a response to a rise in the ratio of fixed to total costs or to an increase in consumer demands for longer store hours.2i It could be argued that an expansion of store hours could induce a reduction in the price-cost margin if weekday and Sunday sales are viewed as related services. Alternatively, if sales volumes and store hours are positively related, the economies of scale could explain the decline in price-cost margins. Increases in the capital to labor ratio and longer store hours both contribute to improvements in labor productivity measured by the sales to employee hours ratio.

4.7.2 Standardization and the Skill Mix of the Retail Work Force In 1950, one could still be served by a retail clerk or butcher, but selfservice is now the rule, even at convenience stores. Cashiers make change, and stock clerks replenish shelves. We have to punch a computer to find the dog food. National advertising and brand names have replaced trained clerks who used to inform us about products. Store-specific human capital might have been a valuable asset when customers asked for particular clerks whose advice was sought and who would honor implicit warranties in the event that we got a defective or spoiled product. When transactions become impersonal and standardized, there is less to be gained by establishing durable, ongoing relations between customers and clerks who know one another. Barry Blue20. Nearly all of the giant supermarkets defined as those with 35,000 or more square feet of selling area were using scanners in 1986, but only around a third of the small supers with selling areas of 10,000-15,000 square feet had acquired this technology; see Progressive Grocer, pt. 2 (April 1987, 22). In 1981, the average age of the buildings owned by giant supers was 9.8 years compared to 16.6 years for small supers; see Progressive Grocer (April 1982,23). The percentage of capital equipment acquired in the used market was inversely related to firm size for Japanese manufacturing firms; see Oi (1983). 21. Pashigian and Bowen (1989) favor the latter explanation. They find that trademarks and store hours are positively related to female wages but unrelated to male wages. A higher labor force participation rate and a high opportunity cost of female time are, in their view, responsible for longer store hours.

183

Productivity in Distributive Trades

stone (1981) concluded that we have witnessed a retail revolution. Retail services have become impersonal resulting in a deskilling of the work force. The reliance on part-timers is one indicator of the skill mix. In 1975, fulltime employees outnumbered part-time workers in the supermarkets that responded to the Progressive Grocer survey, but by 1988, they made up only 41 percent of all employees at independent supers and 36 percent at the chains. The ratio of part-time to full-time employees (PT/FT) is positively related to size and rose from 1.04 in 1981 to 1.79 in 1988: see table 4.7. Part of the secular trend is due to a shift toward larger supermarkets. Even if size is held constant, the ratio exhibits a positive trend. By 1988, 70.3 percent and 64.8 percent of all employees at the giant independents and chains were on parttime work schedules. If the production function was homothetic, the larger supers should have faced a lower relative wage for part-time employees. The wage ratios shown in table 4.8 exhibit a contrary pattern, being somewhat higher at larger stores. The elasticity of hourly wages with respect to sales was .081 and .079 for part and full-time clerks.22Larger supermarkets have to pay higher wages because their employees have to supply more work effort. The wages reflect the effect of higher customer-arrival rates on labor productivity. More clerks have to be hired at the bigger supers, and the queuing model predicts that these employees are more productive because a smaller fraction of labor time is wasted in idly waiting for customers. More productive employees do indeed command higher wages, and the data indicate that the productivity gains associated with weekly sales volume are relatively greater for part-time employees. In addition to this relation with weekly sales, labor productivity is positively related to sales fluctuations over the diurnal and dayof-the-week cycles. Some 55 percent of all supermarket shopping trips and 60 percent of trips made by employed persons take place in the three days, from Thursday to Saturday. Unemployed individuals make few shopping trips on Sundays. The diurnal cycle of customer arrivals varies with the day of the week and shopper characteristics. Nonworking mothers prefer to shop on weekday mornings; Friday evenings are popular for working individuals.23 Part-time employees are analogous to the standby generators that are activated only during the peak load period. The higher is the fraction of total demand produced during the peak period; the larger is the ratio of standby to fully utilized on-line generators that are operated in both peak and off-peak periods. Although we cannot cleanly identify the peak and off-peak hours at a super-

+

+

22. The log of the hourly wage was regressed on the log of weekly sales for the eight observations in 1984 and 1985. These elasticities imply that a fivefold increase in sales volume (roughly the differential between the giant and small supers) would be accompanied by a 13.9 percent higher wage for a part-time clerk and 13.6 percent for a full-timeclerk. The wage-saleselasticities for the chain stores were .I61 and + .I10 for part- and full-time clerks. The wage premiums associated with a fivefold increase in sales were 29.5 and 19.5 percent for part- and full-time clerks. These results are consistent with the firm-size effect on wages, which were reported by Lester(1967), Mellow(1981), Oi (1983), andBrownandMedoff (1989). 23. These data are reported in Progressive Grocer (April 1989,42).

+

184

Walter Y. Oi Ratio of Part-Time to Full-Time Employees by Store Size, 1981-1988

Table 4.7 Year

< 10

1981 1982 1984 1985 1986 1987 1988 Mean

0.98 0.97 0.92 1.06 1.08 1.11 1.19 1.04

Table 4.8

10-15

15-20

20-25

25-30

1.04

1.12 1.03 1.27 1.24 1.32 1.41 1.42 1.26

1.20 1.15 1.38 1.22 1.40 1.49 1.49 1.33

1.14 1.21 1.36 1.38 1.58 1.29 1.34 1.33

1.11

1.13 1.23 1.22 1.17 1.31 1.17

35+

All Stores

...

...

...

1

1.11 1.43 1.82 2.26 1.78 2.36 1.79

1.06 1.17 1.23 1.33 1.36 1.43 1.26

.oo

1.27 1.22 1.96 2.09 1.70 1.54

Hourly Wages of Part-Time and Full-Time Clerks: 1985

Sales Volume (in millions of dollars) Independents: 2-4 4-8 8-12 > 12 Average Chains: 2-4 4-8 8-12 > 12 Average Regression of log wage on /a: Log sales X* 1985 dummy R2 Regression of log wage on /a: log sales X* 1985 dummy R2

30-35

Part-Time Clerks

Full-Time Clerks

Ratio, Part-Time/ Full-Time

$3.98 $4.31 $4.64 $4.71 $4.18

$4.84 $5.11 $5.35 $5.63 $5.01

0.822 0.843 0.867 0.837 0.834

$4.15 $5.07 $5.49 $5.72 $5.10

$5.45 $6.23 $6.45 $6.69 $6.23

0.761 0.814 0.851 0.855 0.819

0.0810 0.0150 0.8976

0.0791 -0.0136 0.9778

0.0019 0.0286 0.4059

0. I605

0,1099 -0.0196 0.8486

0.0506 -0.0398 0.7576

- 0.0594

0.8347

Nore: Let Yp, = the log of the hourly wage of part-time clerks, Y, = the log of the hourly wage of full-time clerks, and R = ( Y , - Y,,) denote the log of the wage ratio that is equal to the difference in logs. I estimated three equations: Yp, = a , Y,,

R

a, =

a,

+ b,X* + c , D + el + bzX* + c,D + e, + b,X* + c,D + e,

where X* is the logarithm of weekly sales, and D is a dummy variable equal to unity for the 1985 observations. Only the b, and c, parameter estimates are reported. We have the identities, b, = (b, - b,) and c, = (c, - c2). The coefficient of determination for the third regression is not equal to the difference in R’ for the other two columns.

185

Productivity in Distributive Trades

market, two generalizations seem to be warranted: First, working persons tend to shop on Friday evenings and weekends, which are usually the heavy peak hours. A rising labor force participation rate should have increased the ratio of sales in the high peak hours to sales in the low off-peak hours ( X J X L ) . Second, working individuals are more likely to shop at the larger supers whose longer store hours reduce their implicit shopping costs. The analogy to the generation of electric power implies that a rise in (XJXL) increases the ratio of part to full-time employees (PT/FT). The available evidence suggests that the ratio of peak to off-peak sales ( X J X J is positively related to store size and has increased over time in response to the increase in labor force participation rates. This increase in the within-week variability of food sales is responsible for part of the upward trend in the relative demand for part-time employees. Retail trade which once provided full-time, stable jobs for most of its employees has become an industry characterized by low wages and high labor turnover rates. The introduction of scanners, organizational innovations in monitoring and warehousing, and the substitution of advertising for point-ofsale services have reduced the relative demand for stable, full-time, storespecific workers. But this is only part of the story. Shifts by consumers in the allocation of time to market and nonmarket activities and changes in shopping patterns have affected the derived demands for full-time and part-time employees. The productivity of part-time workers with little training has increased in relation to the productivity of full-timers. In order to survive, supermarkets have been obliged to alter the skill mix of the work force and the bundle of services which they provide to customers.

4.7.3 The Product Line and Labor Productivity The supermarket consolidated the sales of groceries, meat, and produce under one roof. As the size of the establishment grew, it expanded the product line. It handled more brands of breakfast cereals and introduced new departments-bakeries, delicatessens, fresh fish, drugs, hardware, fresh flowers, video rentals, and so on. The size distribution of establishments has shifted to the right in nearly all of the two-digit retail trade industries; see table 4.6. In most instances, the pattern is similar to that in groceries; growth is accompanied by an expansion in the breadth of the product line. However, in a few cases, such as gasoline service stations, outlets have narrowed their product lines as they got larger. What are the reasons for these divergent trends? How do changes in the breadth of the product line affect productivity? The implicit cost of a trip to a supermarket is a common cost for all the items in the basket. It is akin to the capital cost in the peak load pricing problem. The trip frequency that minimizes the sum of trip and home-inventory costs is determined by certain critical items just as the capital capacity is determined by demand in the peak period. An increase in the demand for nap-

Walter Y. Oi

186

kins has no effect on the optimum number of shopping trips per If the marginal trip cost is zero for most items, the supermarket would seem to have an advantage over specialty shops. Inventory costs and floor space place limits on the breadth of the product line. A typical supermarket in 1960 did not sell fresh salmon, presumably because it could not supply it at a sufficiently low full price. If the arrival rate of calls for fresh salmon was small, inventory and handling costs would have sharply increased the break-even price. As consumer demands climbed and selling areas expanded, the supermarket could supply salmon at a full price (adjusted for quality and a zero marginal trip cost), below that of a specialty fish market. Rising real incomes expand the range of items that are demanded in sufficient volumes to warrant the inventory and handling costs. The economies of one-stop shopping may prompt a consumer to buy his or her chocolates and fresh fish at the same store.25The product line has moved in the opposite direction in SIC 554, gasoline service stations. In the early 1970s, three-fourths of all stations had service bays for repairs and oil changes compared to fewer than half today. The use of self-service pumps increased from 31 percent of all motorists in 1976 to 78 percent in 1986. Stations are larger and more specialized. They are earning more revenues from pumping gas and fewer from repairs, lube jobs, and sales of tires, batteries, and accessories.26From the viewpoint of shopping trip costs, gasoline and lube jobs are not like soap and bagels. They are more like dry cleaning and haircuts. An optimum trip frequency is not determined by an inventory model. Shopping times are additive and do not exhibit the increasing returns applicable to acquiring more items at a supermarket. Design changes have lengthened the interval between filling tanks and changing oil. Nonprice competition and warranties have shifted part of the repair business to auto dealers. As the arrival rates for repairs and lube jobs fell, gas stations experienced a fall in the utilization rate of mechanics. Those stations that eliminated service bays found that the mechanics who previously pumped gas and repaired cars could be replaced by less skilled employees. Fewer stations offered a full product line (gas and repairs), and roughly half of all drivers chose to produce oil changes at home. In the 1980s, specialists emerged. 24. In the model analyzed by P. 0. Steiner (1957), the case of a firm peak meant that there was excess capacity in the low, off-peak period. Optimum prices in peak and off-peak periods are P , = (a p) and P, = a,where a is the unit operating cost and p is the unit capital cost. Capital is a common input for both periods. Variations in the off-peak demand for electricity have no effect on the choice of the capital input unless the demands in both periods are equal to the capital capacity. 25. If the item is a delicacy that is infrequently purchased, the customer may prefer to go to a specialty shop where the expected full price (possibly including a component for assurance of quality), is lower than that at a supermarket. Stand-alone retail bakeries and delicatessens have declined as a consequence of a rising implicit cost of a shopper’s time. 26. There were 426 cars per station in 1972 and 1,129 in 1986. The figures on average establishment size measured by gallons sold per month are a bit misleading because gasoline sales at convenience stores, whose market share is climbing, are excluded from SIC 554. The data reported in this section were obtained from chap. 3 of the study sponsored by the American Petroleum Institute (1988).

+

187

Productivity in Distributive Trades

Franchise dealers found that they could attract enough arrivals to supply lube jobs at a full price below that attainable via home production. A division of labor was thus achieved by self-service stations and Jiffy Lube dealers who separately pumped gas and changed oil. Variations in the output mix and tied services complicate the problem of measuring productivity. In the case of retailing, Robert Steiner (1978) proposed a vertical measure of productivity wherein inputs at both the manufacturing and retailing levels are aggregated and compared to final outputs at the retail level. This principle can obviously be extended to include the inputs supplied by consumers. A gas station that introduces self-service is substituting the labor services of the customer for hired labor. The consequence is an increase in the BLS measure of labor productivity. The annual growth rates of sales per employee hour reported in table 4.3 were 4.0 percent for gas stations and 0.7 percent for food stores. For stations of comparable size, the output per gas pump is higher at self-service stations. When the customer supplies the labor input, it takes less time to complete a transaction-filling the tank and paying the cashier.*’ The adoption of self-service has clearly led to real efficiency gains. The difference in the rates of technical progress in distributing gasoline versus food can, in part, be explained by the fact that supermarkets have expanded their product lines and introduced new departments that call for larger inputs of labor per dollar of sales. In the context of the Becker (1965) model, the substitution of prepared baked goods for homemade cakes can be interpreted as the outcome of a search for the lowest full price. It is the obverse of the self-service gas station. The product in the vertical measure proposed by Robert Steiner is a cheesecake on a plate or gasoline in the tank. The inputs supplied by both retailer and consumer have to be related to these final products in calculating full prices and in determining labor or TFP. Fewer resources are needed to transform the ingredients into a cake at a supermarket bakery, but the costs of moving this cake onto a dinner plate are higher. The costs of acquiring a prepared cake or salad are higher than the costs of buying packages of flour and cream cheese. A rise in the implicit cost of home labor increases the relative full price of a homemade cake, where the full price is the sum of the transformation and transaction costs. The rate of technical progress for a multiproduct firm as we ordinarily measure it is a weighted average of the rates of technical progress for the component goods in the firm’s product line. Supermarkets in pursuit of higher profits have broadened their product lines into more labor-intensive departments with slower rates of productivity growth. Increases in the implicit costs of shopping 27. A correct production function has to include the input of the customer’s time. In New Jersey and Oregon where self-service is illegal, the customer’s time input entails an implicit cost of waiting while he or she receives full service. In the majority of transactions, the customer incurs the implicit cost of pumping one’s own gasoline, which is probably quite low except in adverse weather. I wish to thank Dr. T. F. Hogarty of the American Petroleum Institute for discussions about the relative efficiency of the two types of gas stations.

188

Walter Y. Oi

and of household labor have enhanced the economies of one-stop shopping and increased the demands for prepared foods. These developments are responsible for the decline in sales per employee hour at food stores that took place in the mid-1970s. Gas stations, on the other hand, are allocating a larger fraction of resources into transferring goods (pumping gas), and moving out of the labor-intensive activities of auto repairs and servicing. Consumers are obtaining the distribution of gasoline and the production of repair services more cheaply by a division of labor between specialized institutions, even when we include the implicit costs of labor provided by customers. 4.8

Concluding Remarks

A production function that relates output to inputs lies at the heart of the received theory of productivity change, which is summarized by Jorgenson (1987) and Griliches (1987). In applying this theory to the distributive trades, one has to recognize at least three differences: First, the consumer supplies an essential input which has to be included as an explicit argument of the production function. Ignoring this fact constitutes a specification error that could bias the measured rate of growth in productivity. Second, producing transactions is similar to repairing machines. Both are characterized by the economies of massed reserves wherein a twofold increase in both the customer-arrival rate and the number of clerks leads to more than a twofold increase in the number of completed transactions. If the frictions resulting from transport and homeinventory costs can be reduced by technical advances outside of the distribution sector, retail firms can achieve larger sizes with their lower unit operating costs. Third, the output of a retail firm is a composite bundle whose composition varies across firms and over time. The relation between an aggregate measure of output (such as sales or value added) and purchased inputs depends on the makeup of this bundle. The output of a supermarket has changed over time, moving toward more embodied labor services in the goods that they handle. Progress in the goods-producing industries is frequently associated with technological innovations. The development of the mechanical cotton picker, of the cranberry paddler, and of the chain saw was responsible for increases in TFP. However, institutional and organizational innovations as well as regulatory changes can have equally strong effects. The tax on chain stores, analyzed by Tom Ross (1986), impeded the spread of an important institutional innovation called the supermarket, which brought together the sales of groceries, meat, and produce under one roof. It mimicked the department store, which assembled a wide variety of goods and provided customers with the economy of one-stop shopping, trained clerks, and a reputation for stocking quality goods. The reputational value of the department store has been eroded by nationally advertised branded goods and specialty franchises that shifted the responsibility for product quality from the retail establishment to the man-

189

Productivity in Distributive Trades

ufacturer. Steiner contends that the fall in the gross margin on toys from 40 percent to 20 percent was largely due to national advertising and mass merchandising by discount outlets.28 The establishment of shopping malls has reduced the value of one firm handling a wide variety of goods. The viability of the traditional department store is threatened by the shopping mall, specialty franchises, and national advertising. I fully expect to observe a leftward shift in the size distribution of general merchandise, apparel, and variety stores. Gasoline service stations are, on average, smaller in those states that have divorcement laws preventing refiners from establishing dual distribution systems. The legislated inefficiency is reflected in higher retail prices and lower labor p r o d u c t i ~ i t yIf. ~these ~ divorcement laws were repealed, the rise in labor productivity could be confused with technical progress rather than to the elimination of a regulatory constraint. The allocation of productive and distributive functions among the manufacturer, retailer, and consumer raises questions about the merits of trying to measure productivity at each level. There are compelling reasons to argue for a measure of final outputs-a cake on a dinner plate or gas in the tank. The division of labor is surely determined by comparative advantage. Technological innovations and changes in factor prices can alter the optimal allocation of resources across sectors. These reallocations are often endogenous and should be incorporated in an overall measure of technical progress in the combined production and distribution of ultimate consumer goods.

References American Petroleum Institute. 1988. Gasoline Marketing in the 1980’s: Structure, Practices, and Public Policy. Washington, D.C.: Temple, Barker, and Sloane. Appel, David, 1972. The Supermarket:Early Development of an Institutional Innovation. Journal ofRetailing 48 (spring): 39-53. Arrow, K., D. Levhari, and E. Sheshinski. 1972. A Production Function for the Repairman Problem. Review of Economic Studies 39 (July): 241-49. Bailey, M. J. 1954. Price and Output Determination by a Firm Selling Related Products. American Economic Review 44 (March):82-93. Barron, J., and J. Umbeck, 1984. The Effects of Different Contractual Arrangements: The Case of Retail Gasoline Markets. Journal ofLaw and Economics 27 (October): 313-28. 28. The spread of television reduced the price of advertising messages which translated into a lower cost of informing consumers about particular goods identified by brand names and trademarks. The cost savings were passed on to consumers via lower retail prices that could not have been achieved if resale price maintenance laws had been strictly enforced. 29. Dual distribution is the practice where a refiner simultaneously distributes its product through vertically integrated outlets operated by salaried managers and independent outlets whose prices and operating practices are not controlled by the refiner. The price effects of divorcement laws have been studied by Barron and Umbeck (1984) and Hogarty (1986).

190

Walter Y. Oi

Barger, Harold. 1955. Distribution's Place in the American Economy since 1869. Princeton, N.J.: Princeton Univ. Press. Becker, G. S . 1965. A Theory of the Allocation of Time. Economic Journal 75 (September): 493-517. Bliss, C. 1988. A Theory of Retail Pricing. Journal oflndustrial Economics 36 (June): 375-92. Blozan, William, Jr. 1986. Retail Dealing as Competition for Customer Store Choice: Theory and Evidence. Ph.D. diss., Univ. of Rochester. Bluestone, Barry. 1981. The Retail Revolution. Boston: Auburn. Brown, C., and J. Medoff. 1989. The Employer Size-Wage Effect. Journal of Political Economy 97 (October): 1027-59. Bucklin, Louis P. 1972. Competition and Evolution in the Distributive Trades. Englewood Cliffs, N.J.: Prentice-Hall. Coase, R . H. 1946. Monopoly Pricing with Interrelated Costs and Demands. Economica, n.s. 13 (November): 78-294. DeVaney, Arthur, 1976. Uncertainty, Waiting Time and Capacity Utilization: A Stochastic Theory of Product Quality. Journal of Political Economy 84 (June): 523-42. Douglas, Edna. 1962. Size of Firm and the Structure of Costs in Retailing. Journal of Business 35 (April): 158-90. Douglas, George W., and James C. Miller 111. 1974. Economic Regulations of Domestic Air Transport. Washington, D.C.: Brookings Institution. Ehrlich, I . , and L. Fisher. 1982. The Derived Demand for Advertising. American Economic Review 72 (June): 366-88. Griliches, Zvi. 1987. Productivity: Measurement Problems. In The New Palgrave: A Dictionary of Economics, vol. 3, ed. J. Eatwell, M. Milgate, and P. Newman, 1010-13. New York: MacMillan. Gross, Donald, and Carl M. Harris. 1974. Fundamentals of Queuing Theory. New York: Wiley. Harris, C. 1974. Fundamentals of Queuing Theory. New York: Wiley. Hall, H., J. Knapp, and C. Winsten. 1961. Distribution in Great Britain and North America. London: Oxford Univ. Press. Heall, G. 1980. Spatial Structure in the Retail Trade: A Study in Production Differentiation with Increasing Returns. Bell Journal of Economics 11, no. 2: 565-83. Hogarty, T. F. 1986. Dual Distribution: Theory and Evidence. Washington, D.C.: American Petroleum Institute. Holdren, B. R. 1960. The Structure of a Retail Market and the Market Behavior of Retail Units. Englewood Cliffs, N.J.: Prentice-Hall. Hutt, W. H. 1939. The Theory of Idle Resources. London: Cape. Reprint. Indianapolis: Liberty, 1977. Ingene, C. A., and R. F. Lusch. 1981. A Model of Retail Structure. Research in Marketing 5: 101-64. Jorgenson, Dale W. 1987. Production Functions. In The New Palgrave: A Dictionary of Economics, vol. 3, ed. J. Eatwell, M. Milgate, and P. Newman, 1002-7. New York: MacMillan. Kunze, K. 1985. Hours at Work Increase Relative to Hours Paid. Monthly Labor Review 108 (June): 44-46. Landes, W. M., and R. A. Posner. 1987. Trademark Law: An Economic Perspective. Journal of Law and Economics 30 (October): 265-310. Lester, R. A. 1967. Pay Differentials by Size of Establishment. Industrial Relations 7 (October): 57-67. Levhari, D., and E. Sheshinski. 1970. A Micro Economic Production Function. Economerrica 38 (May): 559-73. Lewis, W. Arthur. 1970. Competition in Retail Trade. In Economics of Overhead Costs, 116-156. New York: Kelley.

191

Productivity in Distributive Trades

Mellow, W. 1982. Employer Size and Wages. Review of Economics and Statistics 64 (August): 495-501. Mulligan, J. G. 1983. The Economies of Massed Reserves. American Economic Review 15: 725-34. Nooteboom, B. 1983. Productivity Growth in the Grocery Trade. Applied Economics 15: 649-64. Oi, Walter Y. 1983. Heterogeneous Firms and the Organization of Production. Economichquiry 21, no. 2: 147-71. Parker, R. 1986. Concentration, Integration and Diversification in the Grocery Retailing Industry. Washington, D.C. : Bureau of Economics, Federal Trade Commission. Pashigian, B. Peter, and B. Bowen. 1989. The Power of the Purse: The Cost of Time and the Growth of Trademarks and Store Services. Univ. of Chicago. October. Photocopy. Progressive Grocer. 1982. April, 25-26. Ratchford, Brian T., and James R. Brown. 1985. A Study of Productivity Changes in Food Retailing. Marketing Science 4, no. 4: 292-3 1 1 . Robinson, E. A. G. 1958. The Structure of Competitive Industry. Chicago: Univ. of Chicago Press. Ross, T. W. 1986. Store Wars: The Chain Tax Movement. Journal of Law and Economics 29 (April): 125-37. Schwartzman, D. 1968. The Growth of Sales per Man-Hour in Retail Trade, 192963. In Production and Productivity in the Service Industries. ed. Victor Fuchs, 20129. New York: NBER. Smith, Henry. 1948. Retail Distribution. 2d ed. Oxford: Oxford Univ. Press. Steiner, P. 0. 1957. Peak Loads and Efficient Pricing. Quarterly Journal of Economics 71 (November): 585-610. Steiner, Robert L. 1978. Marketing Productivity in Consumer Goods Industries: A Vertical Perspective. Journal of Marketing (January), 60-78. Syrquin, M. 1972. Returns to Scale and Substitutability in the Repairman’s Problem. Econometrica 40:937-41. Wallis, John J., and Douglass C. North. 1986. Measuring the Transaction Sector in the American Economy, 1870-1970. In Long-Term Factors in American Economic Growth, ed. S . L. Engerman and R. E. Gallman, 95-161. NBER Studies in Income and Wealth, vol. 51. Chicago: Univ. of Chicago Press. Wessels, Walter J. 1980. Minimum Wages, Fringe Benefits, and Working Conditions. Washington, D.C.: American Enterprise Institute. Whitin, T. M. 1952. Inventory Control in Theory and Practice. Quarterly Journal of Economics 67: 505-21. Williamson, 0. E. 1979. Transaction-Cost Economics: The Governance of Contractual Relations. Journal of Law and Economics 22, no. 2 (October): 233-61.

Comment

Sherwin Rosen

The empirical economist Colin Clark was among the first to track how the share of service-sector employment and output expand as economies grow and develop. Yet economists, by and large, have ignored services. Not only is there a remarkably small literature on the positive aspects of their analysis, Sherwin Rosen is the Edwin A. and Betty L. Bergman Professor at the University of Chicago and a research associate of the National Bureau of Economic Research.

192

Walter Y. Oi

but the accounting expedient of measuring output by inputs hardly has been addressed. The wealthier we become, the more we are doomed to suffer the sham of smaller measured productivity growth compared to more goodsintensive economies-this, in spite of the obvious and important innovations in service sector technology and economic organization that continue to occur over time. Walter Y. Oi’s paper outlines a systematic way of thinking about the economics of services that is essential to research in this field. He presents a superb minicourse on the economics of retail trade. The essential element of the models reviewed is that distributional services are jointly produced by stores and customers. The economics of the retail service sector cannot be understood without recognizing that shoppers’ time is a key input in the production of distributional services. This point is established in the model showing how a customer divides annual purchases into basket sizes and number of trips to the store. The solution balances the costs of an additional trip against the incremental homeinventory costs of goods. By analogy to the economics of the household, Oi shows that the full price of an item is its retail price plus a term that reflects the consumer’s time and money costs of shopping, and the cost of holding goods at home rather than in the store. An extension allows retailing services to be produced more intensively by varying advertising and brand recognition embodied in goods and salespersons’ efforts to provide customer services as substitutes for customers’ time and effort. The economies of scale implicit in inventory holdings are bounded by spatial monopolistic competition among retail establishments. Nevertheless, they imply systematic changes in distributional service productivity as these bounds are changed over time. The full-price formula organizes these elements as consisting of changes in the value of time, transportation costs, residential density, urbanization, and inventory holding costs. I would emphasize the changing composition of families in this, particularly the increasing labor force participation of women. The substitution of market for home production and the resulting increase in the value of women’s time has markedly changed the production of retail services. The paper stresses an unfamiliar “economy of massed reserves,” as the key ingredient in retailing. This concept comes from queueing theory, where it is shown that an increase in the number of clerks and customers, although holding their ratio constant, reduces queueing time in a multi-server facility. Yet that experiment is partial equilibrium because a customer would consider expected crowding and queue length in choosing a store: those things are another aspect of price. The equalization of full prices at the margin would cause queue lengths to adjust to use up any economies of massed reserves in a market equilibrium. In any event Rothschild, Arrow, Levhari and others have shown that these economies are not very large to begin with. It is the simpler point that there are scale economies of inventory holdings that are key to understanding this sector. A store and its sales force are an inventory of goods waiting for customers. One can think of alternative ar-

193

Productivity in Distributive Trades

rangements in which customers wait for goods rather than the other way around. Ready examples are the queues for consumer goods observed in the former Soviet Union. A more interesting example for measurement is how to treat store hours. When gasoline service stations were open for only a few hours per day during the energy crisis in the 1970s, long lines of autos queueing for service were substituted for the waiting services of station operators and their employees. Measured productivity of service stations would have registered enormous gains over that period, in spite of the fact that the amount of services rendered fell dramatically. Only a scheme in which customer queueing time is subtracted from the value of output would recognize this point. Similarly, increased grocery and convenience store hours in recent years results in a drop in measured productivity, though the change in the value of customers’ time and consumption patterns that provoked these changes no doubt has increased true productivity. Services in general are subject to the problem that as the wage rate increases with economic growth the provision of service-intensive distribution becomes very expensive. This causes substitution to less service-intensive methods at the distribution point, such as self-service, and to more service-intensive methods at the production point, such as advertising and packaging. To that extent what we observe is a movement along some grand production function that should not be confused with productivity change. But to some other extent these observed substitutions are due to technical changes that should be properly counted as productivity improvements. For instance, the rising price of women’s time and the smaller size of families increased the extent of preparedness in grocery stores, which range from old-fashioned raw ingredients for home-cooked meals to frozen, precooked items and to completely prepared gourmet meals. Microwave technology and the like are important in this. The paper stops short of discussing how to parcel these things out between productivity and (full-) price adjustments. Only a little thought is required to show that this will be a very hard problem to solve. How should the increasing range of goods available through the distribution system be treated in all of this? Should the presence of kiwifruit and plantains in grocery stores today be attributed to improvements in food production or to improvements in retail services? How about the Colombian cut-flower business or the availability of Chilean fresh fruit and strawberries in the winter months? Do these things get counted as productivity improvements in the transportation sector rather than the distributional sector? Before initiating a debate about how to allocate them in our accounting schemes, it must give one greater pause to learn that they hardly count at all in our statistics today. My guess is that the most workable general approach to measuring distributional services is to treat them as intermediate products in the consumer production of utility through the economics of home production. To be sure, problems abound in this approach, and it is obvious that many of them will not have easy answers. Yet in setting up the essential nature of the problem, Walter Oi’s paper will serve as an important element in solving it.

This Page Intentionally Left Blank

5

The Real Output of the Stock Exchange Timothy F. Bresnahan, Paul Milgrom, and Jonathan Paul

A stock analyst, unhappy with his job, went to lunch with an officer of a competing pension fund. The analyst made his case carefully, explaining which of his material was being ignored, and what lesser material from others was guiding actual trading. The officer, who had no intention of hiring the analyst, picked up the check. Who paid for the lunch? Over the last three decades, trading on stock exchanges has been growing at a growing rate. In 1960, 958 million shares of stock were traded on the New York Stock Exchange (NYSE). By 1970, this had roughly tripled to 3,124 million. By 1987, the volume had grown another 15-fold to 48,144 million shares, and the NYSE was planning for a potential billion-share day by the early 1990s.’ This pattern of accelerating growth is even stronger on the newer exchanges on which the shares of smaller firms are traded. The volume on NASDAQ grew from 1,390 million shares traded in 1975 to 37,890 million in 1987. The acceleration is even more marked if we consider the recent growth in stocklike instruments such as options. A trade of a share of stock is not a constant u n k 2 In figure 5.1, we show the growth in the deflated dollar value of trades on U.S. securities markets. The line labeled “stocks” is the total market value of trades on all the stock Timothy F. Bresnahan is an associate professor of economics at Stanford University and at the National Bureau of Economic Research. Paul Milgrom is a professor of economics at Stanford University. Jonathan Paul is an assistant professor at the University of Michigan. We thank Bengt Holmstrom, Zvi Griliches and conference participants for useful comments. 1 . For recent stock-exchange data on the NYSE and the NASDAQ, see the 1989 Statistical Abstract ofthe United States, tables 830 and 831. The 1960 number is taken from the 1961 U.S. Securities and Exchange Commission AnnualReport, table 9, p. 219. 2. Stock splitting behavior tends to keep the nominal value of the average share traded roughly constant in the intermediate run.

195

196

T. F. Bresnahan, P. Milgrom, and J. Paul

....................................................................................................................................................

exchanges registered with the Securities and Exchange Commi~sion.~ The line labeled “stocks, etc.” includes nonstock securities; warrants, rights, equity options (whether traded or exercised) and trades in nonequity options. The deflator for the figure is an extremely conservative one, the value of the Standard and Poor’s 500 portfolio over time.4 Clearly, the figure shows a dramatic change in the rate of growth of trading, partly driven by the expansion in nonstock but equitylike securities. Overall, the rate of growth of trades in the figure is 8 percent per annum, but the later periods clearly show more rapid growth. If we were to use a less conservative deflator for the figure, such as the GNP deflator, it would show an even more dramatic acceleration. The growth in trading was accompanied by rapid growth in inputs at stock exchange member firms. Figure 5.2 displays selected inputs into NYSE member firms for the period beginning in 1971. s The solid line shows total expenditures in 1987 dollars, using the GNP deflator. The line with asterisks shows registered representatives, that is, broker personnel. (This series is not available for 1984). This period of rapid trading growth was also a period of rapid growth in the resources consumed by the sector. These resources primarily 3. See, e.g., table 18A of the 1986 U.S. Securities and Exchange Commission AnnualReporf. col. 2 , and corresponding tables in earlier annual reports. 4. The advantage of this deflator is that it standardizes the unit of traded stock to remove the problems associated with stock splitting, etc. It overdeflates because the real rate of interest is built into stock returns in the long run. 5. For 1971-83, the data come from the annual NYSE Fact Book, later, from the Securities Industry Association, Securities Industry Yearbook.

197

Real Output of the Stock Exchange 90 80 70 60

%. a8

50 *s; 40 30

4

$ s

g0

20

5i

10

produce intermediate outputs; trading services, knowledge and analysis to support trading decisions, and sales efforts. What is the real output of the stock exchange? Has it grown as dramatically as figure 5.1 suggests? These are the issues addressed in this paper. Our inquiry focuses on several related analytical issues arising in the nature of the changes in stock exchange activity in recent years. Baumol (1965) lists five ways in which the operations of securities markets produce a social output. For our purposes, we classify these into three administrative and two informational outputs. The administrative outputs are access to capital, liquidity, and low administration costs. The existence of securities markets permits any particular enterprise to be owned by many small capitalists, each diversified. This permits the sharing of risk. Further, the capitalists can have a shorter time horizon than the enterprise; the existence of a price at which the individual can sell shares means that the investment is liquid. The additional liquidity may reduce risk as well as lower investor costs. Finally, when compared with other forms of financial intermediation, the securities exchanges can involve considerably less administrative effort for the transfer of funds. The informational outputs of the exchange are twofold. The importance of each output depends on the belief that securities prices reflect fundamentals, that is, the expected future earnings stream of the enterprise. First, information about the firm's prospects could be useful to investors in their effort to evaluate the wisdom of management's plans for the firm. For the single inves-

198

T. F. Bresnahan, P. Milgrom, and J. Paul

tor, whether individual or institution, that information is potentially useful in deciding whether (and at what level) to continue to invest in the firm. For a particular firm’s investors taken as a group, it is useful in selecting, evaluating, and compensating managers. Second, because stock prices are informative, they offer guidance to business management-information on the current cost of capital which is so important in determining the level of investment which it is appropriate for the firm to undertake. Thus, the securities markets could usefully transfer information out of the firm to investors, or reflect information (presumably known by some investor) that management would not otherwise know. The dramatic increase in trading volume on the stock exchanges might have represented an increased service along any or all of these five dimensions. The origins and form of the increase in trading suggest the importance of the two informational dimensions. The increase is happening worldwide-not just in the United States. In the United States, it is driven by very rapid technological change based on the use of telecommunications equipment and computers and by the 1975 deregulation of stock brokerage. The technological effects have been much the same outside the United States, but deregulation has played a much smaller role until recently. The new technologies and deregulation have made trading cheaper. The technologies let individual traders in brokerage houses more closely monitor and react to events in and affecting the markets. Today, they have Quotron terminals and Unix workstations to follow many different sources of information and act on them. The deregulation and technology let investors outside brokerage houses trade more cheaply at the margin. The exchanges themselves have accommodated the traders’ desire for rapid response with electronic market infrastructure. The NASDAQ small-order execution system and the NYSE direct order transmission system are examples that permit reasonably direct computerized trading. One effect of these lower costs is that more traders in more countries participate in globalized markets. This creates an opportunity to execute very large trades quickly and without much effect on price-increased liquidity in the short term. These changes create opportunities for traders to get and use information about companies and about the market. When a trader can easily take and later liquidate a large position, there are opportunities to earn an expected profit on even a small information advantage. As a result, traders pursue ever smaller arbitrage opportunities, better analysis of public information, and (illegal as well as legal) private information. One response to this increased liquidity is the systematic pursuit of arbitrage opportunities. Wall Street “quants” perform computer intensive research not unlike that of professors of finance. The quants’ goals, however, are to find systematic departures from theory and then to trade on them. Increasing liquidity means that smaller and smaller systematic departures can be ex-

199

Real Output of the Stock Exchange

ploited. Competition among potential traders means that, over time, only smaller and smaller departures persist. Similar opportunities exist on the fundamental side of securities research and trading. A trader who can anticipate the contents and implications of a public announcement about a company can usually make a profit. Increased liquidity and lowered transactions costs mean that it is increasingly valuable to expend resources on anticipation of public announcements. The analyst’s research activity is the more valuable, the more correct it is and the more unique it is; something known to everyone is already reflected in the stock price. As a result, competition among research analysts and traders leads to the pursuit of a priori less and less promising fundamental analysis. The marginal fundamental research project always earns zero expected return on the costs of gathering and processing the information about the company. As with the quantitativekbitrage research efforts, the competition among analysts and traders can take two forms. The analyst can attempt to get to a better answer than analysts at other trading houses. This typically takes either betterinformed or smarter analysts; the former are expensive, the latter subject to rapidly diminishing returns. The other form of competition involves getting to the answer more rapidly and trading on it earlier. This form of competition is a very effective destroyer of rents. But it suggests a substantial increase in the speed with which information is impounded in security prices. The nature of the changes in securities markets suggests two directions of increase in the real output of the exchange: The first is the direct effects of lowered trading costs on the administrative outputs of the exchanges. The second is the indirect effects of increased informativeness of the prices of securities. It appears unlikely that either effect represents a large increase in the real output of the securities exchanges. Almost all of our formal argument is concerned with the informativeness effects, but we will briefly review the administrative effects here. Trading costs are not primarily costs of owning publicly traded companies or costs of the existence of publicly held companies. A buy-and-hold lifecycle investor, for example, gains only incrementally from the decline in trading costs over the last decade and a half. Thus enterprises can be owned by a broad range of well-diversified investors either before or after the trading cost decline. The administrative effects seem likely to be second order.6 The value of quicker incorporation of information in stock prices is a more 6. The increased liquidity of the exchanges does provide some benefits, such as opportunities

to rebalance portfolios at lower cost. In an economy with identical stockholders, equilibrium

portfolio balancing involves no trades. But in an economy with stockholders in diverse tax circumstances, the lowered trading costs may permit more effective use of the securities markets to engage in tax-lowering trades. These, depending on one’s view of the tax code, either usefully broaden the base of capital taxation or destroy useful attempts to tax favor particular types of capital.

200

T. F. Bresnahan, P. Milgrom, and J. Paul

difficult question. We ultimately conclude that this, too, likely represents an unimportant improvement in the output of the stock exchanges. But to reach this conclusion we need a more careful treatment of the potential social gains to informative stock prices. If outside analysts research companies more, and if much of the information they discover is reflected in the price of the stock, who finds this information useful and how? We will provide models of both of the two information flows discussed above, that to investors and that to management. The question of the value of information in stock prices is a difficult one in no small part because of the plausibility of the hypothesis that the stock market should have grown more informative over the last decade. Many firms in the United States, facing increased worldwide competition, have felt themselves to be in a more difficult environment. It is a cliche of the 1980s that firms need to refocus on decision making for the long term and on competitiveness. These are times in which it is quite plausible that improved mechanisms for aligning management and shareholder interests on the long run could be particularly valuable. It is also plausible that, in apparently unsettled times, the stock price could communicate more things to management that it does not know.’ In what follows, we provide separate but similar analyses of the two informational hypotheses: (1) more informative securities prices better align investor and management incentives; and ( 2 ) better information in securities prices inform management about investment prospects. In both cases, we assume that the stock grows more informative because of outsider traders. These are self-interested investors who gather information about companies and trade in their securities in the hopes of private gain. We use a standard model of information gathering and trading in the stock market, that of Grossman and Stiglitz (1980). In their model, the incentive to gather information about companies arises because informed investors can trade on it. Because they do trade on the information, equilibrium price is correlated with the information; by this mechanism, the securities market can grow more informative. Rational expectations by other investors limit the size of their trading gains, however. Information that traders gather and trade on is reflected partially in stock prices, partially in expected profits to the informed trader. The more investors know something, or the larger position informed investors take, the more the information tends to be revealed by the stock price. In equilibrium, the number of traders who become informed is determined by the cost of gathering the information and the return. At a market equilibrium, so many traders become informed that they drive the return to information (suitably adjusted for risk) down to the cost. This model provides 7. Of course, it is also a clichC of this period that the stock market does none of these things, but rather the reverse, by focusing companies on short-run returns. This is a point to which we shall return.

201

Real Output of the Stock Exchange

us with a way to understand the part of the problem concerning information gathering by outsider traders. There are a number of ways in which more rapid or more complete reflection of information about companies in stock prices could be a socially valuable output: (1) stock market prices are often used in compensation contracts for high company officials; better information might make this principal-agent problem go better; ( 2 ) the stock market serves to guide investment. Better information might guide it to better uses; and (3) takeovers, leveraged buy outs, and so on, might provide discipline in management that would be more effective with better public information. By building models of (1) and (2), we argue that any beneficial effects are likely be small. Our arguments will clearly not apply to takeovers. Theories of corporate officers’ Compensation suggest that use of the stock price can align officers’ and stockholders’ interests. There may be substantial costs of waiting to compensate officers on the basis of how their decisions actually work out in the future. Instead, the argument goes, the forwardlooking character of the stock price lets the compensation be done closer to the decision point. This leads to an interesting puzzle. Many managers appear to believe that compensating them with stock tends to focus them on the present, not the future. Rather than encouraging high-payoff investments in R&D or other long-lived assets, the stock market encourages managers to show immediate profits. Further, this tendency has been getting stronger over time. As the managers see it, the markets are systematically shortsighted. Their irrationality translates into an irrational compensation scheme. The managers, it turns out, are very likely right about everything but the irrationality. In the next section, we present a simple model of managerial compensation, a model based on the price established in a fully rational stock market. In the compensation model, managers’ and stockholders’ goals are not necessarily identical. Shareholders care only about their return; managers care only about their compensation and the effort they expend to get it. The incentive contract offered the managers uses the stock price to compensate them. To the extent the stock price reflects the future value of the firm, this tends to make the gap between shareholder and manager interests smaller. Managerial attention might, for example, be directed either to R&D or to making the quarterly earnings report look right. The first activity is high payoff, difficult to evaluate, and uncertain; the second is the reverse. A good incentive contract rewards the first and ignore the second. Our model determines whether the self-interested information-gathering and trading behavior of stock-market participants, who might gather information about either managerial activity and trade on it, leads to such a contract. For the purpose of compensating the managers, shareholders would like the stock price to reflect the R&D management activity more, the earnings report less. But investor research may have the reverse effect. The stock market equilibrium does not systematically tend to invest in information about the more

202

T. F. Bresnahan, P. Milgrom, and J. Paul

valuable activity. Although we show this result precisely below, it is easy to anticipate its intuition here. From the point of view of the compensation contract, one would like the stock price to reflect managers’ value added. But that is not what traders care about; they want to know about the value of the firm. In both gathering information and trading on it, their incentives lead them to the wrong activity. Our analysis of the way the stock market guides investment is similar. The role of information is different here. There is information about the firms’ prospects not known to the firms’ officers or directors. The market, by displaying a large q (Tobin 1969; Hayashi 1982), can signal the need for an expansion of the firms’ capital stock. Our analysis of this situation closely parallels the previous one. Once again, we distinguish between two different pieces of information to gather and trade on. One of these concerns the value of the firms’ existing projects; the other, the value of the incremental investment project. To guide investment correctly, the market would have to emphasize the latter information. Once again, the reverse is true; for the case of a small incremental project, stochastically independent of existing projects, the stock price reflects only the value of existing projects. Once again, the intuition is clear. Traders care about the value of the firm, which may be determined overwhelmingly by existing projects. They are not necessarily interested in the value added of the incremental project.

5.1 Real-Side Decisions, Managerial Agency, and the Long-Run Value of the Firm The first value-of-information issue can be addressed in a basic principalagent model of relationships between the owners of a firm and its stockholders. The classic approach to this problem combined three elements.* First, managers’ and shareholders’ incentives are not aligned a priori; managerial effort needs to be elicited by incentives under moral hazard. Second, the board of directors is sophisticated in setting the managerial compensation scheme but is limited in what it can observe; compensation is based on observable indicators of the efects of managers’ actions, not on the actions themselves. Although the board cannot observe the manages’ actions, it correctly anticipates the effect of incentives on managerial actions. Our model also has rational expectations in another sense. The stock price is equal to the present value of per-share earnings, adjusted for risk and deducting any managerial compensation. This equation is not an identity that managers take as given. Rather, it is an equilibrium condition. 8. Cf. Wilson (1969). Spence and Zeckhauser (1971), and Ross (1973). Our modeling details follow Holmstrom (1979).

203

Real Output of the Stock Exchange

The value of stock market information for managerial compensation can be addressed in this framework with two simple changes. First, we give the stock market itself an explicit informational role. We follow Grossman and Stiglitz (1980),Hellwig (1980), and Admati and Pfleiderer (1987) by introducing outside traders who can research the firm and trade on the results of their research. We use the standard model in which outside traders’ information is partially revealed by the stock price. If it were fully revealed, there would be no incentive for research. To capture the long-rudshort-run distinction of our introduction, we also model the outsiders as capable of researching any of several different aspects of the firm. This also leads immediately to our other novel feature. We let managers choose a vector of activities, not a single effort level. Thus, the model has a rich enough logical structure to permit statements like “outsiders research short-run earnings, not long-run prospects, so managers direct their efforts to the short run.” We make specific assumptions about functional forms and distributions, following the literature. In particular, both shareholders and managers have constant absolute risk aversion preferences about wealth, and managers’ utility is separable in wealth and effort. The joint distribution of the future value of the firm and all signals observed by outside traders is normal. Ultimately, the firm will have earnings

v = f(x) + cioi, where x is a vector of choice variables for management, and 0 is a random term outside anyone’s control. These revenues are discounted to the present. For our application, it is natural to think off(x) + 0 as being calculated from the sum of present earnings, y,, and the present value of future earnings, y,. Both y, and y, depend on x, perhaps differently, and the sum of their random components is 0. The managedagent picks x, and bears some private effort costs, c(x), in doing so. So far, we closely follow the usual agent-theoretical models of managerial compensation, except that x is a vector. The problem for the directors/shareholders, then, is be slightly more complex than usual. They need to elicit not only overall managerial effort, but the right mix of the different efforts, x . ~ We follow the literature by assuming that compensation cannot be based on x, which is fundamentally unobservable, or on !L which is observed too late. Instead, compensation is based on the current stock price, I? The intuition held by both economists and practitioners has been that P reflects all publicly available information about V Thus, under a fundamentals or efficientmarkets theory of how P is set, management’s incentives are aligned with shareholders. 9. The effort variables, x, are naturally unitless, and we have written bothflx) and c(x) as arbitrary functions. Clearly, there is an arbitrary normalization here. We could writeAx*) = Z, xf,c(xf), without any loss of generality.

204

T. F. Bresnahan, P. Milgrom, and J. Paul

The net compensation of the manager is given by:

(2)

T(P) - c(x),

where T(P) is a transfer or compensation function. We assume T is linear, T ( P ) = to t , !?lo In this framework, x is subject to moral hazard. A selfinterested manager will set xito maximize expected monetary income minus effort costs, adjusted for risk. In certainty-equivalent terms, this is

+

u,,,=

(3)

max to + t , E[P I XI X

a

-

2

t: u;.

The last term is the risk premium in the manager’s compensation. Because higher x generally leads to a higher mean price, (3) does represent an incentive contract. As long as the managers’ coefficient of absolute risk aversion, a , is positive, the solution t , = 1 and to = - E[U is not optimal, because it requires the managers to bear too much risk. We assume that directors, in setting T(P), act in the long-run interests of all shareholders. They will pick the function T( .) to maximize the appropriately risk-adjusted terminal value of the firm. We will return to the details of this problem. In the standard model with exogenous information, x is set by the managers according to (3). The stock price is set by (4)

P

=

E[V 1 x] - T(P) - (shareholders’ risk premium):

because constant shareholder risk aversion implies linearity in E [ U and rational expectations implies the value of the firm is calculated at the managers’ equilibrium action. Finally, to and t , are set by the board to maximize shareholder welfare subject to (3). Except in some special cases (certainty, riskneutral manager), the equilibrium does not attain the first best. Perhaps better information from an active set of researcheriinvestors would help.

5.2 Investor Preferences and Stock-Market Equilibrium We let the informativeness of P about V be endogenous. Individual or institutional investors can research different aspects of V and then trade on that information. We assume that in performing these activities, investors are selfinterested. They trade on their research information only to the extent it gives them a risk-adjusted return. They only do research that leads to profitable trades. We will follow Grossman and Stiglitz (1980) in modeling one effect of 10. This can be justified as corresponding to an optimal choice when the model is regarded as the reduced form of a more detailed, dynamic formulation. See Holmstrom and Milgrom (1987). 11. There are no investment bankers or raiders on boards forcing shortsightedness in our model.

205

Real Output of the Stock Exchange

this activity. The more informed investors trade, the more P reflects their research efforts. If other investors have rational expectations, they use the information in P This limits the private value of research. In our treatment, the increased informativeness of P has a second spillout; the board of directors can use it in a compensation contract. The research possibilities are about different aspects of V We assume that investors are grouped in types, i, each of which can research a particular aspect.’* For clarity, we link each research aspect to a particular managerial effort level, x,. In particular, we assume an informed investor of type i observes,

(5)

y; = x;

+ 0,.

This is an imperfect signal of the effort x , because it contains noise. It is also an incomplete signal of the ultimate value of the firm. Because

v = A x ) + c, o,, any particular research project bears on only a particular aspect, such as the current or future earnings of the firm. Our assumption that there are multiple nontrivial signals and that any particular investor sees at most one of these means that each informed investor is still somewhat uncertain about V Thus no risk-averse investor takes infinitely large positions in the security. We will investigate two different problems: a short-run problem in which the amount of research being done is fixed and a long-run problem in which research is endogenous. We assume that investors are risk averse and that they have constant absolute risk aversion. An investor’s utility function over ex post wealth is U ( W ) = -exp( -aW) where a is the coefficient of (absolute) risk aversion. This specific form of preferences is helpful because it, together with competitive rational expectations and normally distributed errors, implies a linear assetdemand equation. An investor who observes signal yi demands stock according to

The numerator in (7) is the investor’s expected return on a share of stock. A positive signal leads informed investors to take a long position, a negative signal to short. The denominator is risk aversion times the variance of the 12. This is an assumption of convenience. Models in which investors choose research projects, possibly pursuing several, have been investigated by Admati and Pfleiderer (1987). These models quickly become extremely technically complex. Price-taking investors who research multiple projects would add little to our treatment. Takeover investors who investigate everything and mount a raid are a different matter.

T. F. Bresnahan, P. Milgrom, and J. Paul

206

investor’s risky return. Investors with better research (i.e., smaller variances around ultimate value) take larger positions. I 3 Why are both numerator and denominator in (7) written as depending on P as well? In rational-expectations equilibrium, P reveals some of the information about y,. Suppose that the results of one particular kind of analysis, for example, y,, suggests low future stock value. Researcheriinvestors who see yI dump the stock, and its price will therefore be lower: As a result, P incorporates some of the information in the y , . This dependence could be exploited by an uninformed investor who demands

The dependence between the price of the stock and researchers’ information is also the reason the board can use P as an input in managerial compensation, as we shall see. To investigate the informativeness of the stock price as a signal of the research’s contents, we now examine the equilibrium price equation. Each informed investor of type i behaves according to (7); let there be A, such investors. (For the moment, we assume A is exogenous.) Similarly, let there be A, uninformed investors, and let the demand by noise traders be z. Normalize by 1 - A,, = C,A,. Then price solves the zero net supply equation: 0 = Auq,(P) + 2, A,q,(R Y,)

(9)

+ z.

Because (9) is linear, P is a linear function of the y , and z , and thus normal. The variance of P and its covariance with they, are determined by (7) - (9). This relationship is simple to understand if we write out (7) and (8) explicitly. Let pi.,be the coefficient of P when V is regressed on P and y, , and let & be the coefficient of y,. Write the manager’s compensation T(P) as to + t,P Then,

Because an informed investor of type i can observe both y, and R she can condition on both. Both are valuable information; P contains information about other traders’ information. The relation between signal y , and trading behavior depends on two things; @, which measures 6E[V I *]/6y,, and a 2 ( V -), which measures the risk borne by a type-i trader. Similarly,

I

13. Following Grossman and Stiglitz (1980). we write the problem as if the stock of the firm at hand were the only risky asset in the economy. Admati and Pfleiderer (1987) treat the case of an investor demanding a portfolio of different securities. Although the definition of the riskiness of a further investment in a particular firm is changed by the portfolio treatment, the fundamental parts do not seem to be altered.

207

Real Output of the Stock Exchange

(11)

9" =

E[VI

+ pi P

-

P - to - t,P

I

a u*(V P )

where pvpis the coefficient of P in a regression for V with no other regressors. Uninformed traders can condition only on the information revealed in 19 They are always at greater risk than informed traders, because u2(VI P ) is necessarily larger than u2(V 1 yi, P). The distribution of P in equilibrium is then determined by

where (13)

*

=

hi (pii - t, - 1) au2 (V I yi, P )

- t , - 1) + X,(pi . a u2(V 1 P )

5.3 The Alignment of Shareholder and Manager Interest Knowing that research and investment leads to cov(y,,P) > 0, and wanting to reward xi,a value-maximizing board of directors will set managerial compensation to depend on 19 They anticipate that managers will supply effort in order to earn stock-price-based income. The board's problem is to maximize shareholder well-being subject to ( 3 ) and subject to Urn2 U,, the manager's opportunity salary. This problem takes the form max A,V~ ro.

11.

x s.t. ( 3 )

+ Cihivj

urn u,.

That completes our statement of the model. The ingredients of the model, then, are threefold. Managers are selfinterested and need an incentive contract to align their interests with shareholders. The functionf(x) - c(x) need not be symmetric, so some activities can be more valuable than others. The management incentive contract should focus attention on those activities. The second ingredient is that investors are self-interested. They research those activities of the firm on which they can hope to make profits as informed outsider traders in informationally efficient markets. The third ingredient is the board, which maximizes shareholder welfare.

5.4 Exogenous Number of Informed 'Ikaders The problem would become easier if the stock price came to incorporate more information about the more important managerial efforts, x,. In (12), we see that P is a linear function of the yi, with coefficients depending on the

208

T. F. Bresnahan, P. Milgrom, and J. Paul

number of informed traders of type i , A,, and on the equilibrium behavior of informed and uninformed traders. The coefficient of y, in the price equation (12) is proportional to its coefficient in type-i investors’ stock demand equation (10). Thus, we see that the stock price coefficient of y, depends on the number of type-i investors and on the strengths of their tendency to trade on useful information. The board’s problem is made easier if investors have a systematic tendency to (1) become informed about or ( 2 ) trade on information about the high-value managerial activities. We divide our discussion of the results of the model into two parts for convenience: The first part takes A as exogenous, though informed investors’ trading behavior is endogenous. Thus, the results in this part answer the question, Does informed investors’ trading behavior make the stock price a better instrument for the board’s managerial compensation problem? This is not such an unlikely prospect, because the informed trader of type i observes a signal correlated with managerial activity x,. We now show that informed investors’ trading behavior has no particular tendency to favor managerial activities for which f ’ , (x,) - c’,($ is large. To make the results simple to state, we first treat a special case and then the more general problem. In the special case, the information structure of the problem is symmetric in that there are an equal number of informed traders of each type and in that each @, has the same variance and all pairs of 0’s have the same covariance.

PROPOSITION I (symmetric information structure, exogenous A). Suppose all of the A, are the same, and that cov(O,, . . . ,On) = cov(O,,, . . . ,@,J for any permutation of the activities. Then the behavior of informed traders rewards management for all activities equally, regardless of their relative value. Proof. Equations ( 5 ) , (lo), (1 I ) , and (12) determine the joint distribution of = a*(V 1 y,, P ) for any pair of i , j in any solution. The manager’s marginal incentive to perform activity i is

t Y, r! and Z . Under symmetry, PLpr= P’, and a 2 ( V I y,, P )

this is set equal to 6c,($)/6xiby the manager. The second term in (15) is the contribution of informed trading to the managers’ marginal incentive. Under symmetry, it is the same for all i . A more general result follows the same logic.

PROPOSITION 2 (arbitrary information structure, exogenous A). Equilibrium outsider trading does not lead a stock-price-based compensation contract to differentially reward high-value activities (those with G flax, - GcJGx, high) or those where managerial effort is elastic. Instead, the equilibrium marginal

209

Real Output of the Stock Exchange

incentivefor managers to pei$orm activity xc versus activity x, is entirely determined by A and the joint distribution of 0. Proof. The joint distribution of F! X C: and Z , now possibly assymmetric, is determined by the same equations as in proposition 1. Thus, the relative size of p",p versus p", and u2(V I y,, P ) versus u2 (V ly,, P ) is entirely determined by the joint distribution of 0,C: and Z , and by A. It does not depend onf,(x,) versusf,(x,) or SC($/SX, versus SC(@SX,.

An even sharper contrast can be seen by comparing the incentives determined by the stock market versus the optimal incentives that the board would desire. Holmstrom and Milgrom (1990) consider the problem of multitask performance incentives. They show (under weak restrictions) that the optimal incentive scheme rewards the activities at the margin according to

where [ ] is used to indicate the vector or matrix whose typical element is inside the brackets. The stark differences are immediate. The optimal incentive depends (of course!) on f 'i.The equilibrium stock-price-based compensation scheme does not. Keeping the level of 2) fixed at the equilibrium point, the slopesi can be changed arbitrarily and the stock market's behavior is unchanged. In particular, we can make an activity such as massaging the earnings report arbitrarily worthless,f', near zero, and the stock market will continue to emphasize it. The optimal incentive depends on the supply elasticities of effort (S*c/Sx,Sx,), and these are lacking in the stock-market equilibrium as well. Only the distribution of 0 is common to the two problems, and it plays very different roles. In (12), we see that traders particularly emphasize the high-variance 0's. In the optimal contract, high-variance signals are emphasized less. Why do these results hold? The intuition is simple. Our investors trade on information to the extent that it gives them a capital gain. In equilibrium, this comes at the expense of uninformed traders, traders who observe a different signal, or noise traders. To make a capital gain, a trader wants to know about the future value of the firm, f ( $ , not managerial value added, f . The private value of information for trading is entirely determined by who knows it, or who knows variables correlated with it. It is unrelated to the value of the information in compensation of managers. In a related paper (Paul 1990), one of us takes up the general question of the equilibrium information content of l? With a general joint distribution for 0 and C: on which signals do investors tend to trade most heavily? For an important class of distributions of 0 and C: equilibrium informed outsider trading systematically weights the noisiest signals the most heavily. (See propositions 2 and 3 in Paul 1990). As Paul (1990) points out, this is exactly the reverse order from what the board would like in the case wheref(3) - C(3) is

210

T. F. Bresnahan, P. Milgrom, and J. Paul

maximized at x, = x,. Instead of putting weight, as the optimal linear compensation package would, on the least noisily signaled effort, in Paul’s results the investors do the reverse. At this juncture it should be clear that there is no systematic tendency for trading by those investors holding that information the board would like to see reflected in the price. If anything, rather the reverse seems to be true. Outsider trading is guided by the logic of the stock market itself, not by the relationship between the stock market and the firm’s decision problem.

5.5 Investor Research Decisions Yet propositions 1 and 2 refer to an environment in which investors’ research is exogenous. The informativeness of P is endogenous, but the information available to be reflected in P is not. Investors cannot pursue research opportunities focused on particularly valuable x,. Would a model with endogenous research effort have more favorable implications for the role of stockmarket compensation in management decision making? The simple answer is no. And the simple reason follows directly from the Grossman-Stiglitz equilibrium condition for endogenous research. Grossman and Stiglitz (1980) assume that traders invest in becoming informed in signal i until their utility gain from trading on y, is just equal to the cost of the signal. The cost of effort to the manager and the value to the firm appear nowhere in this calculation. Let the research costs of aspect i be R,. Let the certainty equivalent of the return to knowing 0,be V,. Let the certainty-equivalent return of an uninformed investor be Vu.Then A,, the mass of informed investors of type i, rises until

(17)

V, = Vu

+ R,.

From Admati and Pfleiderer (1987),proposition 3.2, we know that Vt - Vu is equal to

It is immediate from (18) and (12) that the return to purchasing signal i for the marginal investor is proportional to the tendency of type-i investors to trade on their information, other things equal. (The other things here include p”,, so this may be overstrong. Yet pyipand u2(VI yi,P ) should move in opposite directions in the cross section of signals.) In particular, the return to the marginal investor is exactly proportional to the strength of a trader’s response to a given forecast. (See Admati and Pfleiderer 1987, equation 3.4.) It therefore should not be surprising that the marginal investor’s behavior will not be particularly directed toward the high-value investments. The following result is immediate by symmetry:

211

Real Output of the Stock Exchange

PROFWITION 3 Under the symmetry conditions of proposition 1 , and assuming RL= R,, the mass if informed investors of each type, A,, is the same for all i. Thus, there is no tendency for endogenous information tofavor the high-value research activities. There seems little room left; the use of a stock-price-based compensation contract does not solve a multiple-decision moral-hazard problem. Although it may possibly be good at eliciting overall effort from managers, it has very little role in focusing managers on those particular activities that create shareholder wealth over the long run. This result may help explain the CEOcompensation puzzle. Jensen and Murphy (1989) try to measure the empirical relationship between CEO compensation and growth in shareholder wealth. They find a weak relationship, suggesting CEOs are nor compensated primarily according to rl Why might this be? Suppose the long career path to becoming a CEO weeds out all candidates who disvalue overall effort, leaving only the driven. Then our results show that there is no value to using the stock market in focusing the CEOs efforts; other mechanisms must be employed for that.

5.6 The Value of Externally Generated Information in Managerial Decision Making In the second theory of valuable information in the stock price, it is the firm’s managers and board who learn from it. Part of this theory is similar to that of the last section. Outsider traders research the firm and discover profitable new opportunities. To the extent that they trade on this research, they bid up the price of its shares. Yet part of this theory is different as well; in it, the firms’ managers do not know about the profitable opportunities but instead learn about them from observing the share price. This story is usually told in a slightly different way, in which it is thejrm’s cost ofcapital that is signaled to the managers and the board through its share price. Suppose outsider traders, basing their decisions on private research about fundamentals, drive up the share price. The resulting low cost of capital (rightly) means the firm should acquire additional capital and expand. Baumol (1965) points out that the informational story does not critically turn on the true marginal opportunity cost of capital to the firm. The new project could be financed out of retained earnings, so that there is no relation between the share price and marginal opportunity cost of capital. This does not affect the argument that the stock price might signal valuable opportunities to the managers and board. Indeed, the main point here is not about investment of new capital at all, but instead about the use of the stock price by insiders to gain useful information about the firm’s prospects. A similar role for the stock price is central to q-theory models of investment (Tobin 1969; Hayashi 1982). It is easy to see, however, that there is no relationship between the validity of the q-theory investment model and the valid-

212

T. F. Bresnahan, P. Milgrom, and J. Paul

ity of the valuable-information theory. In the q model, the origin of the information about new projects is unimportant. I4 Management can research the profitable new opportunities and tell analysts about them. Alternatively, people outside the company can have done the research, and management can have inferred the opportunity from the high stock price. In either event, q > 1 and the firm should expand. The usefulness of q-theoretical models in econometric studies of investment, therefore, has nothing to do with the direction of information flow. The value of increased stock-price informativeness, however, does turn critically on the direction of information flow. There is little informational value for managers in having the stock price quickly reflect things they have just told analysts. If outside analysts learn something about new projects that managers do not know and trade on it to the point it is reflected in the share price, then there is an informational gain. Managers, seeing opportunities reflected in the share price, can move to fund the projects.I5 This section has a very simple model of this potential information flow. The model suggests that the flow cannot be very important. The key assumption of the model is that the stock price reflects two kinds of outsider trading and research. The outsiders can research and trade on information about the firm’s incremental projects. But they can also research and trade on information about the firms’ existing projects. Securities prices confuse these two effects and thus provide a poor signal of the value of the incremental projects. We can reuse much of the notation and many of the ideas of the previous section. We continue to assume that the ultimate value of the firm is determined as where is a vector of choice variables for management, and 0 is a random term outside anyone’s control. But for our purposes in this section, the interpretation of the variables changes somewhat. First, management and the board together pick t, and we assume here that they have no agency conflicts. All 3 are already decided except xn, the new project. Further, we think of x, as a single golno-go decision; undertake the project or not. The information structure is as follows: Management knows all of theJ(xi) except f,(x,). The value of the incremental project, f,(x,) + 0,- c, is a normal random variable. Management knows none of the nor can they be

oi,

14. Indeed, Hayashi’s formal model (1982) has no private information at all. All investors are equally informed, and the stock price reflects fundamentals. This is all that is needed for a qtheory model. 15. As far as we can detect, few managers claim to be the recipients of such signals, and few investors the senders. Yet they would not; managers do not lightly admit that outsiders know anything useful about their companies, and successful outsider traders earn more if they keep their own counsel. The exception is investors who have taken large positions in firms and who use takeover threats, seats on the board, and other large-scale mechanisms to compel policy changes. We shall return to this point.

213

Real Output of the Stock Exchange

learned by management. Outside traders can research any or all of the Y, = + Oi,including Yn = f,(x,) + 0,. What management and the board would like to learn is the expected value of the new project, f,(x,,) - c. If they could directly research the signal Y", they would have a noisy (because of 0,)signal of this. It is clearly (privately) cheaper to let the outsider traders do the research and to use the stock price as a signal. To investigate the information management can glean from this strategy, we investigate the distribution of the price of the firm's stock under the assumption that the incremental project is going to go fonvard.I6 The trading, research, and other activities of outsider traders go forward exactly as in section 5.1, above. In particular, the equilibrium distribution of the stock price is determined by (12):

f;(x,)

The implications of this are immediate and straightforward. The stock price, as a signal for Y,, is noisy, potentially very noisy. Consider, for example, the case in which there are nine existing projects and the distribution of the Y's and V is symmetric in the sense of proposition 1, and that all the Y's are independent. Then P is an equally weighted average of ten signals, only one of which is of interest (and also a function of the pure noise z). There is once again no necessary relationship between the pattern of information collection that would support the real-side decision making of the firm and the pattern that self-interested outsider traders would consider. The real-side decision needs information about the marginal project; outsider traders care about the totality of the firms' projects. Of course, if the signals Yi are all highly correlated, the stock price will be a good signal for Y,. In this case, the distribution of the marginal project and all the inframarginal projects are much the same, and P is highly correlated with the signal. Yet there is something very odd about this example, in which the managers do not know the value of the incremental project despite its high correlation with the inframarginal projects. For a firm that is growing rapidly, the value of the incremental project can dominate in overall firm valuation. Then the financial markets may well guide investment. A variety of financial institutions reflect this. Venture capitalists certainly decide much about the allocation of capital to young firms. For somewhat older firms, the initial public offering process can involve an information flow to managers from investment bankers and institutional investors. Yet it is difficult to see how the same information flow could be important for 16. We assume, for purposes of the discussion, that there are no tricky gaming issues between management and the outsider traders. Suppose, for example, that management will withdraw the project if the stock-market reaction is adequately adverse. Then the value of the security reflects this prospect, and (12) does not hold. Considerations such as this can only complicate management's inference problem.

214

T. F. Bresnahan, P. Milgrom, and J. Paul

a mature firm or how capitalizing information into the stock price in minutes instead of days can add much value.’’ We can state the general point easily in Hayashi’s language (1982). The board and the managers would like to know marginal q, the incremental value to the firm of doing the additional project. Outsider traders’ profits depend on average q, not marginal q. Thus the signal-to-noise ratio in P can be arbitrarily small, as in the case of an incremental project that is one nth of the firm and distributed independently of the inframarginal projects. 5.7

Conclusion

How rapidly has the real output of stock exchanges grown? Is it anything like as rapid as figure 5.1 suggests? Are the rapidly growing costs of having stock exchanges (fig. 5.2) merely the flow of resources into a sector making a booming social contribution? On balance, our analysis is not encouraging to a positive view on these issues. The gaps between social and private incentives to trade seem likely to be important in the modem era. Neither of the two informational theories of the real output of the stock exchange suggests a large value for the increased activity of outsider traders. In both cases, the information that is valuable for making a real decisioncompensating managers or deciding on a new project-bears no relation to the information impounded in prices by the activities of traders. The appropriate real decisions depend on value added-the value of the managers’ decisions or of the new projects. The research and trading decisions of outsider traders are focused on the value of the firm, not on value added. One of the best ways to become a clichC is to be a truth. Researching a firm’s quarterly earnings report in the days after the end of the quarter but before the report is released may well be an excellent way to make a capital gain. At the same time, it could have no value in guiding management and could advance the date at which the board has a reasonable assessment of management performance by only a few days. Increased liquidity increases the resources devoted to this sort of rent seeking, without improving any real investment decisions. Our analysis has not addressed takeovers, mergers, or the market for corporate control more generally. Traders in our analysis seek trading profits, not changes in control of the company. It is clear that a large trader seeking new management, a seat on the board, or other changes in the governance of the enterprise links real decisions and financial markets in a very direct way.’* It is an interesting and difficult question whether recent trends on the stock ex17. We have also ignored any potential value of information about the firm to third parties, such as the owners or managers of other firms. 18. See Shleifer and Vishny (1986, 1989). They model the role a low stock price might have in attracting takeovers, and the managers’ resulting desire to pick projects the stock-market “likes .”

215

Real Output of the Stock Exchange

changes help or hinder this process. On the one hand, increased liquidity permits outsiders with valuable changes in corporate governance in mind to move more quickly. Yet such actors are typically held back by regulatory restrictions, which do not bind the arbitrageurs whose free riding is also enhanced by liquidity.

References Admati, A,, and P. Pfleiderer. 1987. Viable Allocations of Information in Financial Markets. Journal of Economic Theory 43:76-116. Baumol, William J. 1965. The Stock Market and Economic Eficiency. New York: Fordham Univ. Press. Bresnahan, Timothy F. 1986. Measuring the Spillovers from Technical Advance: Mainframe Computers in Financial Services. American Economic Review 76, no. 4 (September): 747-55. Grossman, Sanford, and Joseph E. Stiglitz. 1980. On the Impossibility of Informationally Efficient Markets. American Economic Review 70 (June): 393-408. Hayashi, F. 1982. Tobin’s Marginal q and Average q: A Neoclassical Interpretation. Econometrica, 50, no. 1 (January): 213-24. Hellwig, M. F. 1980. On the Aggregation of Information in Competitive Markets. Journal of Economic Theory 22:477-98. Holmstrom, Bengt. 1979. Moral Hazard and Observability. Bell Journal of Economics 10:74-91. Holmstrom, Bengt, and Paul Milgrom. 1987. Aggregation and Linearity in the Provision of Intertemporal Incentives. Econornetrica 55, no. 2 (March): 303-28. . 1990. Multi-task Principal-Agent Analysis: Incentive Contracts, Asset Ownership and Job Design. SITE Working Paper No. 6, Department of Economics, Stanford Univ. Jensen, Michael C., and K. Murphy. 1988. Performance Pay and Top-Management Incentives. Harvard Business School. Mimeographed. New York Stock Exchange. 1983, 1989. FactBook. New York. Paul, Jonathan. 1990. Do Stock Markets Aggregate Information Efficiently for Principal-Agent Problems? Department of Economics, Stanford Univ. Mimeographed. Ross, Stephen. 1973. The Economic Theory of Agency: The Principal’s Problem. American Economic Review 63: 134-39. Securities Industry Association. 1989. Securities Industry Yearbook, 1988-89. New York. Shleifer, Andrei, and Robert W. Vishny. 1986. Large Shareholders and Corporate Control. Journal of Political Economy 94, no. 3 (June): 461-88. . 1990. Equilibrium Short Horizons of Investors and Firms. American EconomicReview 80, no. 2 (May): 148-53. Spence, A. Michael, and R. Zeckhauser. 1971. Insurance, Information and Individual Action. American Economic Review 61 (1971): 380-87. Tobin, James. 1969. A General Equilibrium Approach to Monetary Theory. Journal of Money, Credit, and Banking 1: 15-29. U.S. Bureau of the Census. 1989. Statistical Abstract of the United States. 109th ed. Washington, D.C.: Government Printing Office.

216

T. F. Bresnahan, P. Milgrom, and J. Paul

U.S. Securities and Exchange Commission. 1961-86. Annual Report. 27th-52d ed. Washington, D.C.: Government Printing Office. Wilson, R. 1969. The Structure of Incentives for Decentralization. In La Dkcision, Agregation et dynamique des ordres de prkfkrence, 309-15. Paris: Centre Nationale de la Recherche Scientifique.

6

User Costs, Shadow Prices, and the Real Output of Banks Dennis J. Fixler and Kimberly D. Zieschang

“The distinctive function of the banker-says Ricardo, begins as soon as he uses the money of others.”’ Indeed this aspect of banking lies at the foundation of the difficulties encountered in measuring bank output. Broadly speaking, such output consists of transactions (payments) services and the portfolio management services that banks provide to depositors while acting as their intermediary. There is no consensus in the banking literature on how to measure these services. In this paper, we focus on the measurement of bank financial services arising from deposit products, securities, loans, and other financial services such as corporate payments services and trust services. Two measurement questions immediately arise: (1) Is the financial services output represented by the volume of transactions or the volume of money in the various products? and (2) Which products should be considered part of the output set? It is now generally recognized that the answer to the first question is that both dimensions are important.2 The second question chiefly concerns the treatment of deposits. Because deposits are an input into the acquisition of earning assets, many argue that they should be treated as such. However, some argue that people purchase deposit accounts for the services of record keeping and safeDennis J. Fixler is economist, Division of Rice and Index Number Research, Bureau of Labor Statistics. Kimberly D. Zieschang is chief, Division of Price and Index Number Research, Bureau of Labor Statistics. The authors would like to thank Allen Berger, Diana Hancock, David Humphrey, Jeffrey Smith, Jack Triplett, and Frank Wykoff for their comments. They would also like to thank Hank Leddon of the Board of Governors of the Federal Reserve System for assistance with the call report data. The authors are solely responsible for all remaining errors. The views expressed are those of the authors and do not necessarily reflect the views of the colleagues cited above, Bureau of Labor Statistics or Department of Labor policy, or the views of other staff members of those agencies. 1. AsquotedinBagehot(l915, 21). 2. See Benston, Hanweck, and Humphrey (1982, 1985).

219

220

Dennis J. Fixler and Kimberly D. Zieschang

keeping and that these services make deposit products outputs. A complicating feature of the argument is that some of the services purchased by depositors are typically not paid for explicitly. A bank recovers the cost of these services by setting a loan rate greater than the deposit rate. The absence of an explicit price for many of the financial services attached to deposit products also complicates the measurement of financial services in the national income and product accounts. Specifically, the absence of an explicit payment has made it impossible to determine the value of these services, and so national income accountants must impute their value. Although the proper imputation has been the subject of a long debate, it has recently become more topical with the rise in the international trade in financial service^.^ Fixler (1988) sets out how the financial model we use addresses output measurement and the preponderance of implicit prices. Briefly, all financial services are a part of the output set. Financial services are assumed to attach to each dollar in a financial product at a point in time. Variables such as the number of transactions are viewed as quality variables of the attached financial services bundle. The price of the attached bundle of financial services is characterized as the user cost of money associated with the product, a concept developed by Donovan (1978) and Bamett (1978, 1980) and applied to financial firms in Hancock (1985b, 1986). The fundamental components of a user cost price are an interest rate, a capital gains rate, and an opportunity cost of money or benchmark rate. Given the user cost prices of financial products, banks maximize variable economic profit conditioned on physical capital, labor, purchased materials and services, and technology. The research we report in this paper is part of a program to develop a conceptual framework for a financial services component for the producer price index. Given the financial firm model for price measurement developed in Fixler (1988), this program addresses four questions defining areas of research relevant to the construction and applicability of a financial services output price index: (1) For what nominal sales aggregate could such an index be used as a deflator, and how does the financial firm model relate to concepts underlying the national income accounts for nominal sales of financial services? ( 2 ) On the basis of available data and given the current national income accounting conventions, what would the implications of such a price index be for financial services price and output over a recent period? (3) How would available quality of service attributes be incorporated into the user cost of money measurement framework? and (4) How good are the accounting rules of thumb as estimators of the opportunity cost of funds compared with a structural, econometric estimate, and what are the implications for the resulting financial services output measure? 3. In fact, the United Nations Statistical Office (UNSO) is in the process of revising their imputation of financial services. Our discussion of the UNSO approach below refers to the proposed revision.

221

User Costs and the Real Output of Banks

Fixler and Zieschang (1 99 1) addressed question 1. After reviewing past and present national income accounting treatments for financial services, it was shown that the financial firm model rationalizes the accounting methodology used by the Bureau of Economic Analysis (BEA) and an alternative methodology proposed by the United Nations Statistical Office (UNSO). Each accounting framework was shown to impute uncharged financial services on the basis of its own assumption about the opportunity cost of money. In the BEA framework, the opportunity cost rate was shown to be the interest rate charged on loans; in the UNSO framework the opportunity cost rate was shown to be a simple average of the rate charged on loans and paid on deposits. Because the construction of a financial services output price index compatible with each accounting methodology depends on the assumed opportunity cost, an empirical assessment of these assumptions was deemed necessary. To address question 2, Fixler and Zieschang (1990) expanded on work begun in 1989 to compute price and quantity indexes for large banks based on data from the Federal Deposit Insurance Corporation (FDIC) for the years 1984-88. The paper also developed independent estimates of the capital gains component of the user cost prices. Use was made of results from Fixler and Zieschang (1991) on the opportunity cost of money assumptions underlying the BEA and UNSO accounting schemes to compute the user cost prices of monetary goods and their associated sales weights in Tornqvist price index formulas. The indexes considered were therefore designed to be compatible with the existing and proposed accounting treatments of financial services. The patterns and levels of price change found under the two imputation schemes were generally similar, and the output growth for the period was found to be approximately 40 percent, with the BEA opportunity cost rate. To address question 3, Fixler and Zieschang (1992) conducted a study of the use of bank branches as correlates of or proxies for the quality of financial services delivered, again making use of FDIC data. In that paper, a technique was introduced for using hedonic estimates of the prices of characteristics (the numbers of six types of branches operated by a bank) to construct an exact quality adjustment to a superlative productivity index. The technique was inspired by earlier theoretical work on quality-adjusted superlative price indexes by Zieschang (1985, 1988) and adapted to the financial services price measurement context by Fixler (1988). The quality-adjustment method was then demonstrated by constructing depositor services and labor quality modifiers for multifactor productivity indexes for large banks. The adjustment was modest given the narrow focus on a single product and a single input, increasing the productivity index level by approximately 0.6 percent at the end of the 1984-88 period. In the present paper, which addresses question 4, we characterize a bank’s production of portfolio and payment services in a distance function framework. From standard duality results, we derive the opportunity cost of money for the bank as a shadow price. Econometric estimates of the distance function

222

Dennis J. Fixler and Kimberly D. Zieschang

and the attending product shares are then used to estimate the shadow value of the opportunity cost of capital for the bank. Using FDIC data for approximately 480 banks with assets over $300 million in the years 1984-1988, we find that in any given year the resulting value of the econometric opportunity cost rate differs noticeably from the opportunity cost rates underlying the BEA and UNSO frameworks. Our estimated value of the opportunity cost rate also significantly differs from the 90-day Treasury Bill rate in three of the five years considered. The Treasury Bill rate is another commonsense value of the opportunity cost rate inasmuch as it is a short-term, risk-free rate readily available to banks. Using the estimated value of the opportunity cost rate, we construct price and quantity indexes. Fortunately for price- and quantity-index construction, we find that the Tomqvist output quantity index and its associated implicit price index are not sensitive to the opportunity cost estimate used; output growth was found to be approximately 40 percent; prices declined by approximately 4 percent. However, the insensitivity of index numbers to the opportunity cost estimate may not carry over to the imputations in the national income accounts. In Fixler and Zieschang (1991) we found that the level of the opportunity cost rate affects the imputed value of sector sales and thereby may have a significant effect on the division of gross sales of financial services between intermediate and final consumers. The remainder of the paper is organized in the following way: Section 6.1 provides a detailed discussion of the output and price measurement concepts underlying the financial firm model. Section 6.2 sets out the distance function and section 6.3 its econometric estimation. Section 6.4 describes our data and our results. Section 6.5 concludes.

6.1 Bank Output and Prices Financial services output primarily stems from the role of financial institutions as “users of the money of others.” This role is greatly affected by two features of their environment: the existence of imperfect capital markets and the set of regulations designed to minimize the probability of bank failure and control the money supply. Of special importance to the latter is the fractional reserve system. By capital market imperfections we mean the information asymmetries between lenders (depositors) and borrowers and the existence of substantial transaction costs for depositors to discover information about potential borrowers and to specify the loan contract. It is a bank’s ability to reduce both the informational asymmetry and the attending transaction costs that is crucial to its role as an intermediary. Goodhart (1989) argues that, because banks provide information and lower transactions costs, as well as hold inventories of financial instruments, they act as market makers for money. But a bank’s behavior as an intermediary is substantially limited by bank regulation. Until recently, banks could not pay depositors any interest on their

223

User Costs and the Real Output of Banks

checkable deposits. Deposits cannot be used to acquire equity nor can they be used to underwrite new issues. The imposition of capital requirements limits the returns to a bank’s stockholders. Perhaps the most significant restriction is the reserve requirement on deposits. This “tax” further limits the return that banks earn on their deposits, and, as shown by Barnett, Hinich, and Weber (1986), it is substantial: approximately $10 billion in the early to mid-1980s. The fact that the required reserve rate is less than 100 percent complicates modeling the role of deposits in a bank. In a world with a 100-percent reserve requirement, banks could not lend out deposits and therefore would charge explicitly for the financial services, such as safekeeping and record keeping, provided to depositor^.^ But, because the reserve requirement is substantially less than 100 percent (approximately 12 percent for large bank demand deposit accounts and 3 percent for nonpersonal time deposit accounts), banks can lend most of their deposits. Thus a bank provides an intermediary portfolio service to depositors, in addition to record keeping and safekeeping, that produces a profit that can in turn finance interest payments to depositors or subsidize the costs of the services provided to them. Deposits are therefore simultaneously an input into the loan process and an output, in the sense that they are purchased as a final product providing financial services. Viewed in this way it is not surprising that the classification of deposits as an input or an output has sparked so much debate in the bank literature. It is also clear why it is difficult to price the financial services sold to depositors. The financial firm model we use focuses on banks as producers of financial services. The financial services are attached to each dollar in the various financial products offered by a bank and are therefore measured in monetary units. All financial products are viewed as providing financial services and are produced by employing (physical) capital, labor, and purchased materials and services. But to simply say that all financial services are output is not sufficient to model bank behavior. One must also be concerned with the role of the financial products in the operations of banks; principally, the role of deposits as the raw material used to make loans and acquire other earning assets. To capture this intermediary aspect of bank behavior, it is necessary to assign a financial input or output status to each financial p r ~ d u c t It. ~is important to keep in mind that this status reflects only the role of the product in the financial operations of the bank; the financial services output is the output being measured. A product is a financial output when its economic return is positive 4. Intermediation would also not take place if the interest rate paid to depositors were equal to the loan rate and the reserve requirement were less than I00 percent. 5. There is a considerable debate in the literature (and in this conference) about the inputoutput status of deposits. In our framework, the financial services attached to deposit products (inclusive of the intermediary service) are always considered an output. At the same time, deposits are allowed to act as an input to loan production; in fact, as shown below, the user-cost method explicitly accounts for the net interest earned on the deposit.

224

Dennis J. Fixler and Kimberly D. Zieschang

and a financial input when its economic return is negative. As is explained below, these designations are not permanent in the financial firm model, a flexibility that allows bank behavior to adjust to financial market conditions. Financial products are essentially a bundle of financial services specified by a contract. The contract between depositors and the bank is standard. Fixed nominal values are deposited, and the bank promises to make them available on demand. That is, the provision of services like record keeping and safekeeping do not impinge on liquidity. Product characteristics, such as allowed checks per month, determine the quality of the financial services provided. Measures of activity, such as number of accounts and transactions per account, can also be viewed as quality factors in this framework. Loan contracts are more variable. Banks not only vary the loan contract by type of borrower, for example, between commercial and noncommercial ones, but also within a type according to credit risk and perhaps size of loan. Another feature of loan contracts is that the interest rate charged by the bank may not reflect the actual cost of the loan. Banks may require a borrower to keep a compensating balance in a deposit account or bundle the loan with some payment services, for example, becoming the issuer of dividend checks. Because these arrangements involve an implicit price for the loan service provided, we encounter the same problem as the one discussed above for deposit services. To model fully the effect of such features as compensating balances and like factors affecting the implicit payment for services, we ideally want to augment the financial service bundle attached to each dollar in a product by a vector of product characteristics that captures the particulars of the contract between the bank and the depositor or borrower.6 Such detailed characteristics information, however, is unavailable in our data set. But the data set does contain the number of bank branches, and we use this variable to ascertain the importance of the convenience of service characteristic. Our financial services prices are the user cost of money rates per dollar in a financial product. The user cost of a financial product is an appropriate characterization of the financial services price because it measures the economic return to the bank for providing the financial service. The form of a product’s user cost depends on its asset/liability status. The user cost for the ith asset financial product in period t for a particular bank is given by

where p is the bank’s opportunity cost of capital and hot is the holding revenue rate obtained from the ith asset, which is given by h;,

=

interest rate received

6. Fixler (1988) shows how a full treatment of product characteristics would be incorporated in the quantity index given later. That analysis further shows how, by adjusting changes in the price of the financial services for changes in product characteristics, the price index effectively synthesizes the nominal and activity-based characterizations of banking services.

225

User Costs and the Real Output of Banks

+

capital gain rate - provision for loan losses. The user cost for the ith liability financial product is given by

where h,, denotes the holding cost rate of the ith liability product and is given by h;, = interest rate paid - service fees + p X reserve requirement.’ The last term represents the reserve tax. The sign of the user cost allows one to distinguish products as finuncial inputs and outputs. If the user cost is negative, then the product is a financial output, because it contributes to revenue, and, if the user cost is positive, then the product is a financial input. The user-cost approach’s endogenous categorization of products as financial inputs and outputs is significant because of the extensive debate in the bank literature about the proper status of financial products. The user cost expressions include the return to intermediation, the imputation for the uncharged-for financial service, as part of the price for the financial service. To see this, suppose that the holding cost for a deposit product was simply the interest rate paid less the per dollar service fee charged and that the opportunity cost of money was simply the loan rate. Using (2) above the user cost for the deposit product can be written as - [(loan rate - interest rate paid) + service fee]. If the loan rate were equal to the interest rate paid, then the value of the services provided would simply be the explicit service fee charged. If instead the loan rate were greater than the interest paid to depositors-the usual case-then the value of the financial services would be the implicit payment, represented by the difference in the interest rates, plus the explicit service charge. A similar interpretation applies to the user cost of asset products. The above user cost expressions reveal the importance of p in determining the price of the financial services bundle: it represents the opportunity cost of money to the bank from the perspective of the next best alternative use, in contrast to the typical opportunity cost of capital in the finance literature, which focuses on the cost of capital as determined from its sources.8 The unobservability of p is a hurdle that must be overcome in implementing the financial firm model. As mentioned at the outset there are some commonsense candidates for p that are used by BEA and UNSO. In this paper we derive p from a model of the bank’s technology and estimate it using FDIC data. Our derivation of p relies on the parts of the bank technology concerned with making loans and purchasing securities. This use perspective reflects the 7. Bamett (1980) and Hancock (1985b) derive these equations. To simplify the analysis, we ignore the ramifications of discounting. 8. See, e . g . , Van Home (1983).

226

Dennis J. Fixler and Kimberly D. Zieschang

portfolio behavior of the bank, in particular the uses of the money obtained through deposits or otherwise borrowed. In banking, such portfolio decisions entail duration matching whereby short-term sources of money are matched with short-term uses, and so on, and balancing the volume of interestsensitive assets with the volume of interest-sensitive liabilities to minimize the effect of interest rate risk on net worth.9 Bank models are unavoidably limited by the difficulties inherent in modeling the nuances of portfolio management, especially when dealing with large, complex, multiproduct banks such as the ones in our sample. We capture the portfolio management features, to some extent, by separately considering four loan categories and two security categories. The loan categories are loans secured by real estate, commercial and industrial loans, loans to individuals and credit card loans, and a catchall category-all other loans and leases. The security categories are securities backed by the U.S. government and a catchall, all other securities. The all-other-securities category includes such items as state and local government securities, federal funds sold, repurchase agreements, and foreign securities. Loans are typically long-term with fixed interest rates and no secondary market in which to trade them; securities have a well-developed secondary market that permit trades to counter interest rate risk. By isolating these products in our estimation of p, we allow for productspecific portfolio considerations to affect our estimate of p. We seek a representative p that can serve in the construction of industry output and price indexes. To be acceptable, the representative p should not only reflect bank attitudes toward risk and maturity but also serve as an industry aggregate of the bank-specific portfolio assembly process. The value of p for each bank is taken as some proportion of its return on assets, and that proportion is assumed to be the same for all banks. Thus by estimating the proportion, we obtain an industry representative p, and by applying the proportion to each bank’s return on assets we obtain the distribution of bank specifics p’s. 6.2

The Bank Production Model

Like Hancock (1985b), we view banking firms as transforming the nonfinancial inputs capital, labor, and purchased materials and services into financial products. Our list of nonfinancial inputs is x, = number of employees and officers; x2 = premises and fixes assets, in dollars; and xg = purchased materials and services, in dollars. The bank produces the following financial products: y , = loans secured by real estate; y 2 = commercial and industrial loans; y3 = loans to individuals, including credit cards; y4 = other loans, and leases; y s = federal funds purchased, and federal government securities and federal 9. Duration refers to the average time needed to recover the initial investment. It is in effect a measure of interest rate risk.

227

User Costs and the Real Output of Banks

agency obligations; y , = obligations of states and political subdivisions, and foreign and other securities; y , = fiduciary activities, fees, and other noninterest income; and y s = interest and non-interest-bearing domestic and foreign deposits, federal funds sold, and Treasury demand notes.Io All are measured in dollars. These y i will be designated financial inputs or outputs according to the sign of their user cost. Ideally, we would also include a vector of characteristics, say given by the letter a , describing the services provided. Our data, however, only allows us to include branching as a proxy for service attributes such as convenience. The FDIC call report data contains the number of six different types of branches operated by banks. In Fixler and Zieschang (1992) we examined the use of this branch information as quality indicators. There, we made use of hedonic equations relating service charge rates and the average salary of officers and employees to the branching variables to compute a quality-corrected multifactor productivity index, using a method of incorporating hedonic estimates into exact and superlative index numbers. We found certain branch variables to be significant in explaining cross-sectional interest-rate variation. We also found that branching had a mildly positive effect on banking industry output and productivity over time. Consequently, we include a branching variable in the current structural estimation context. We characterize the bank's production technology as D (x, a , y)

=

1,

where the function D is the output distance function, defined as (3)

D (x, a, y) = [max (8 : (x,a, 8 y) E T } I - l ,

and T is the banking firm's technology set. D thus represents the reciprocal of the factor 0 that scales the output vector y = (yl, y 2 , y 3 ,y4, y s , y,, y,, y,) with characteristic a so that 8y is just producible with inputs x = (xl,x,, x3), where a = number of branches in the bank's domestic offices. It can be used to form the more familiar joint-production function

f ( x , a , y) = D (x, a, y)

- 1 = 0.

The distance function appeared in economics in the early 1950s in works by Debreu (1951), Shephard (1953, 1970), and Malmquist (1953). It has seen extensive, if often implicit, application in the economics and operations research literature on measuring technical efficiency beginning with Farrell (1957)." Malmquist (1953) and Moorsteen (1961) related the distance func10. In previous work, we found a difference in the classification of large certificates of deposit, domestic interest-bearing deposits, foreign interest-bearing deposits and non-interest-bearing deposits. The structural model developed below results in a system of asset share equations. Asset detail was therefore deemed more important than liability detail, and depositlike funds were combined to reduce the number of unknown parameters in the model. 11. For recent studies of cost efficiency in banking, see Femer and Love11 (19%) and Berger and Humphrey (chap. 7, this vol.).

228

Dennis J. Fixler and Kimberly D. Zieschang

tion to the theory of economic quantity indexes, and Caves, Christensen, and Diewert (1982) have related it to indexes of multifactor productivity. In general, D is linear homogeneous in y and nonincreasing in a scalar multiple of x. Other properties often assumed for D include convexity and increasing monotonicity in y, and quasi concavity and decreasing monotonicity in x. Shephard (1970) showed that when D is convex and increasing in y, it is dual to the revenue function, defined T

(x, a, p)

max, { pry : (x, a, y) E T ) ;

=

that is, D and T can be derived from one another as D ( x , a , y)

=

max, {p’y : ~ ( xa,, p) 5 1)

and ~ ( xa,, p)

=

max, { p’y : D ( x , a, y)

5

11,

where p is a vector of known nonzero, nonnegative prices. If one of the outputs y is actually an input, with Vy, D < 0 and Vp,n < 0, then T can be reinterpreted as a restricted profit function (see, e.g., McFadden 1978). Hancock (1985a, 1985b, 1986) used the restricted profit function in her studies of bank technology, focusing on the interest-rate and substitution elasticities of financial products. She estimated the holding revenue (cost) components of the user cost prices of financial products from interest rates, realized capital gains, insurance fees, and loss-provision data. The remaining component, the opportunity cost rate of money, was determined by reasoning that the opportunity cost must not be any higher than the maximum rate at which no bank in her sample would earn an economic loss. In this study, we are interested in developing and evaluating methods for price measurement and deflation of bank revenue to obtain a measure of bank output. Accordingly, we want to estimate the shadow prices of monetary goods, and we approach the problem of modeling bank technology from the primal, instead of the dual, side. We pose a shadow price problem because we want to infer the opportunity cost of money, a key component of the user cost expressions, directly from bank behavior using econometric methods. Iz The system of output shadow price equations (up to a proportional constant) is given by the gradient of D with respect to y, assuming that the distance function is differentiable in y. Formally, the shadow price vector p* is given by p* = V P ( x , y), which is the obverse of the better-known ShephardHotelling lemma yielding the vector of revenue-maximizing outputs as y* = V,T(X, p). We estimate the (conditional) distance function and its gradients, from which we obtain the shadow prices of loans and leases, y , , . . . , y,, 12. Our econometric approach implements the Fire and Zieschang (1991) suggestion that the shadow price equations can be useful for determining the prices of nonmarketed commodities produced by nonprofit organizations, or in any situation where market prices are either absent or not believed to represent marginal revenue.

229

User Costs and the Real Output of Banks

securities, y , and y6, other services, y,, and deposits, y,. Using the expressions from Barnett (1980), Hancock (1985b), and Fixler (1988) to express the user cost of money prices of financial goods, and with knowledge of loan, security, and deposit interest rates, the securities appreciation rate, and the rate of provision for loan losses, we determine the opportunity cost rate, p. We compare this econometric estimate against the estimates implicit in the current and proposed national income accounting imputation methods, and other commonsense rates that might be used as opportunity cost estimates. Our econometric model uses what we will call a conditional distance function. This distance function is conditioned on the level of deposits, y,, and is defined by (4)

where 3, refers to all elements of the output vector y except deposits, y , . The deposits conditional distance function Dc is linear homogeneous in 3, by definition. Inasmuch as we use accounting data, we use this function because deposits and other liabilities are accounting inputs, a use of funds, even though they may be a source of financial services output. We show below that our representation of the production technology yields a system in which the accounting shares of individual asset and fee income in total asset and fee income are functions of the shares of individual assets in the asset portfolio and other variables. We hold that, from an econometric point of view, the gross revenue share system generated by the deposits conditional distance function in equation (4) is better posed than the net revenue share system that would be generated by the unconditional distance function in equation (3). I 3 6.2.1 Modeling the Opportunity Cost Rate As discussed in section 6.1 above, the assumption of a constant opportunity cost rate across banks may be too restrictive in our sample of banks from the FDIC data set, which is heterogeneous and large by comparison with the Federal Reserve functional cost analysis survey data set used by Hancock. We therefore model the opportunity cost rate as a constant proportion of a bank’s return on assets. This specification is appealing because banks with unusually high asset yields are likely to have concentrations of assets in relatively risky categories. Setting the opportunity cost rate as a fraction of return on assets 13. We show below that the deposits-conditional output distance function generates a system of equations that relate the shares of asset receipts in total asset income to the corresponding asset portfolio shares, and the vector of arguments of the conditional distance function. By analogy, it can also be shown that the “unconditional” distance function generates a system relating the “shares” of (positive) asset receipts and (negative) deposit payments in net asset income, to the corresponding net asset portfolio shares and the vector of arguments of the distance function. We consider this latter system ill posed because the net asset income shares are not bounded between zero and one and are likely to be very sensitive to the random variation in the interest rates that are effectively the endogenous variables in the system, particularly disturbances that happen to drive net asset income near zero. We examined the net asset shares in our data and can confirm they are very noisy with numerically large extreme values.

230

Dennis J. Fixler and Kimberly D. Zieschang

therefore takes into account variations in attitudes toward risk by management.

6.3 Using the Distance Function to Estimate p Because we are interested in shadow prices and our data contain some price information, we wish to econometrically estimate the distance function with a system of share equations. Fiire, Fukuyama, and Primont (1988) estimate a distance function along with a system of share equations in which the lefthand-side variables are computed from known price and quantity data. We cannot use this approach without modification, because the unknown opportunity cost rate is a determinant of the left-hand side of the Fare, Fukuyama, and Primont share system for our We therefore develop an alternative set of share equations. Recall that the first six products in the bank's production function are products associated with monetary asset stocks, the seventh is fee and services income, and the eighth is deposits and other liabilities. For i = 1, . . . , 6, the ith product share is

where h, = the holding revenue rate on the ith asset; p = the opportunity cost rate of funds; R, = Xf= h,y, + R , = total asset holding revenue and service charges; R, = income from services produced other than those associated with asset/liability products; A = Cp= y , = total assets; and w, = Vln,,EnD,(x,

a, y,; 3,). This can be restated with a change of variables as (5)

W $=

4 . s,

+ (1

-

4 ) . w,,

where w,= h,y,/R,, the holding revenue share of the ith product in asset income; st = y,/A, the asset portfolio share of the ith product; = p/rTA;and r,, = R,/A, the total rate of return on assets, including (nondeposit) service charge income. For the seventh output, other services, we have

+

(6)

w7 = ( 1 - +I

*

w7,

where w, = Vlny,lnDc(x, a, y,; f,). From the system of equations (5) and (6) we can form an econometric model of bank technology. Assuming the distance function D,is translog, the economic shares w are 14. Other recent studies on estimating distance functions include Fire, Grosskopf, Lindgren, and Roos (1989), Fire, Grosskopf, Lovell, and Yaisawarng (forthcoming), and Lovell and Zieschang (1992). These papers approach the estimation problem in one of two ways: direct fitting of the distance function to the data by linear programming methods, or estimation of a system of share equations.

User Costs and the Real Output of Banks

231 (7)

w~ =

at

+ yya.tlnU +

;='I

yyyglnY] +

':=I

yyx,rklmk,

. . . , 7. From the homogeneity of the distance function: 2:= ,a,= 1 , X : ; l = , yy a , , - 0 , X . 7 = I y y y , J = 01, j, =. . . , 8 , a n d C ~ = l ~ y x r k = 0 1, ,k2=, i = 1,

3. Substituting (7) into (5) and (6)and appending an error term, we have wi =

(8)

4 si

+ pi + $ya,$nu + C/8=1$yy,,lnyj

+ X;=l$yx,iklmk + E), i

=

1, . . . , 6;

and (9)

w7

=

P 7

'!',,$nu + X ~ =JlYy.7j I b', +

%=l(Gryr.7k1mk

+

E7,

where IJ.~ = ( 1 - +) * ai;$p,i = (1 - 4) . yYa,,;J?,,/ = (1 - 4) * Yyy,i/; $,.x.ik = (1 - 4) * yyx,ik; and where, again, 4 is the ratio of the opportunity cost rate to total return on assets for the industry. From the earlier homogeneity conditions: Cy=lp,i = 1 - 4; Z~=l$yo,i = 0; X7=1$yJij = 0; Z~=I$yx,ik = 0; and X ; = I ~= i 0. We estimate 4 as a constant parameter that corresponds to our maintained assumption that the opportunity cost rate is a constant fraction of the total rate of return on assets. 6.4

Data and Results

Our data set is a subset of the FDIC reports of income and condition. The FDIC data consist of quarterly balance sheet and income statement call reports for the approximately 13,000 commercial banks that are covered by deposit insurance, and we hereafter refer to it as the call reports file. We consider only the banks that have international operations or assets over $300 million (FDIC classes FFIEC03 1 and FFIEC032). These banks file more detailed quarterly reports than other reporting banks and cover more than half the deposits of all banks in the United States. We further filtered the set of banks considered by applying the following criteria: banks had to have positive net assets, positive total liabilities and positive total assets, and positive net income from interest and noninterest sources. This yielded a sample of more than 400 banks in each quarter for the period from 1984(1) to 1988(4). We aggregated the report of condition data into annual averages for the years 1984-88. The reports of income contain the annual income and expense flows that match our annual average asset and liability data from the reports of condition. They are annually cumulative, and we used the reports for the last quarter within each year for which each bank reported data. In most cases, this was quarter 4, but some institutions disappeared during each year, and the associated quarters of data used for these banks ranged from 1 to 3. The account classes were initially aggregated to the lowest level possible for which comparable stock and flow information could be computed from the balance sheet and income data and then further aggregated to the classes defining our output variables y I , . . . , ys. Table 6.1 con-

232

Dennis J. Fixler and Kimberly D. Zieschang

Table 6.1 Aggregate Output Class

Components of Financial Product Aggregates Report of Income Code

Description

401 I 4012 4050* 4019 4024 4026 4056 4057 4058 4059 4100 4065

Secured by real estate Commercial & industrial Loans to individuals To depository institutions To farmers Acceptances of other banks To foreign governments Nonsecurity obligations of states All other loans in domestic offices Loans in foreign offices & edge & agreement corporations Balances due from depository institutions Leases

-

Loans & leases:

Yl Y2 4'3 4'4

Securities:

4020 4027 4066 Y4 4067 4068 4069 Directly charged services: Yl 4070 4075 4076 4078 Deposits & other liabilities: 4'8 NINT* 4174 Y5

4176 NFNT*

4172 4180 41 85

Federal funds sold & repurchase agreements U.S. Treasury securities and agency obligations Securities issued by states & political subsivisions Other domestic securities (debt & equity) Foreign securities (debt & equity) Securities in trading accounts Fiduciary activities Trading gains & fees from foreign exchange transactions Other foreign transactions gains Other noninterest income Non-interest-bearing deposits in domestic offices Time certificates of deposit larger than $100,000 in domestic offices All other deposits Non-interest-bedring deposits in foreign Offices All other deposits in foreign offices Federal funds purchased & repurchase agreements Demand notes with the U.S. Treasury

Nores: N I N T and NFNT are placeholders for report of condition accounts that do not appear on the report of income because they do not earn interest. Code 4050 is the sum of report of condition codes 4054 and 4055, which segregate credit card income from other income. The reports of condition contain only account values for the sum of credit card and other loans to individuals. *Created code.

tains the detail of this product aggregation scheme. In table 6.2 we report means of the shares of assets and charged services in asset and service charge income w # ,the asset portfolio shares, s,, and sample sizes.

6.4.1. Construction of the Holding Revenues and Holding Costs Before describing the econometric aspects of estimating p, we describe the construction of the holding revenues of assets and the holding costs of liabili-

233

User Costs and the Real Output of Banks Sample Sizes and Asset-Weighted Means Holding Revenue and Portfolio Shares

Table 6.2

Variable

All Data

N

482

Data without Zeros

Data without Zeros and Influentials

I 984 464

424

WI w2 w3 w4 w5 W6 w7

0.113 0.186 0.087 0.367 0.102 0.040 0.078

0.115 0.186 0.087 0.367 0.102 0.040 0.078

0.119 0.183 0.093 0.364 0.102 0.037 0.077

SI

0. I27 0. I84 0.082 0.442 0.097 0.057

0.127 0. I84 0.082 0.441 0.098 0.058

0.134 0.182 0.087 0.432 0.099 0.057

s2 s3 s4 s5 S6

I985

N

473

455

410 0.122

w1 w2 w3 w4 w5 W6 w7

0.115 0. I62 0.095 0.333 0.116 0.061 0.089

0.115 0.162 0.095 0.332 0.117 0.061 0.089

0.101 0.327 0.114 0.057 0.087

SI s2

0.131 0.180 0.090 0.419 0.102 0.068

0. I32 0.180 0.090 0.418 0.102 0.068

0.139 0.184 0.094 0.408 0.101 0.064

s3 s4 s5 S6

0.164

1986 N

472

45 1

415

w1 w2 w3 w4 w5 W6 w7

0.126 0. I49 0.099 0.290 0.111 0.074 0. I07

0.126 0. I49 0. loo 0.291 0.111 0.073 0. I07

0.131 0.152 0. I06 0.285 0.110 0.068 0. I06

5.1

0.141 0. I76 0.092

0.141 0. I76 0.092

0. I47 0.180 0.097

s2 s3

(continued)

234

Dennis J. Fixler and Kimberly D. Zieschang

Table 6.2

(continued) Variable

s4

s5 S6

All Data

0.394 0.102 0.085

Data without Zeros

0.395 0.102 0.084

Data without Zeros and Influentials

0.384 0.103 0.079

I987 N

464

439

402

WI w2 w3 w4 w5 W6 w7

0.153 0.169 0.107 0.215 0.053 0.055 0.141

0.153 0.167 0.108 0.214 0.053 0.056 0.141

0.157 0.168 0.113 0.210 0.058 0.051 0.138

s1 s2 s3 s4 s5 S6

0.161 0.176 0.090 0.369 0.110 0.081

0.162 0.175 0.090 0.369 0.1 10 0.082

0.169 0.180 0.094 0.358 0.112 0.074

I988 N

453

426

385

w1 w2 w3 w4 w5 W6 w7

0.155 0.161 0.095 0.304 0.087 0.068 0.116

0.156 0.160 0.095 0.305 0.087 0.068 0.117

0.167 0.166 0.104 0.274 0.085 0.059 0.111

s1 s2 s3 s4 s5 S6

0.179 0.178 0.092 0.344 0.115 0.080

0.180 0.177 0.090 0.344 0.116 0.080

0.191 0.183 0.098 0.331 0. I15 0.070

Note: W is used for asset-weighted means holding revenue; S for portfolio shares.

ties. The holding cost and revenue components of the various product user costs were constructed by item. As given earlier, the complete expression for the holding revenue of the ith asset is given by

h: = interest rate received

+ capital gain rate

- provision for loan losses

235

User Costs and the Real Output of Banks

and the expression for the holding cost of the ith liability is given by h: = interest rate paid - service fees

+px

reserve requirement.15

To calculate the interest rates used in the analysis, fourth-quarter income for a particular asset or liability product for each bank in the sample is divided by the annual average of the corresponding aggregate balance sheet item. Asset detail on loan and lease loss provisions was not available in our FDIC data. We therefore allocated the available aggregate loan and lease loss provisions for each bank proportionately (and admittedly somewhat arbitrarily) to the portfolio share of each loan item in the loan and lease portfolio. Because loss provisions reduce taxable corporate income, we multiplied loss provisions by one minus the marginal tax rate (see n. 21 for the tax rates).I6 Deposit service charges per dollar are estimated by the ratio of total service charge income to the annual average of interest and non-interest-bearing deposits in domestic branches, again owing to a lack of account detail in the FDIC service charge data. This amount is then subtracted from the interest rates for the deposit products. We set the capital gains term equal to zero in all holding revenue expressions for assets that are not marketable. This leaves the following security categories for which a capital gain term is relevant: assets held in trading accounts; U.S. Treasury securities; U.S. government agency and corporate obligations; state and local securities; other domestic securities (mainly mortgage related and Federal Reserve stock); and foreign securities. Our assumption about marketable assets does not take into account the recent rise in loan sales by banks. The deregulation of banks and the attending rise in competition has forced banks into a position of diversifying portfolios by selling loan assets in their entirety or in parts. Consequently, secondary 15. The holding cost for a liability product should also include a deposit insurance premium assessed by the FDIC. Because all banks are assessed the annual premium of %2 of 1 percent of total domestic deposits, the exclusion of this term does not qualitatively affect our analysis. The premium term would have to be included if the premium becomes dependent on bank risk-a suggestion that is often voiced in the face of the rising number of bank failures. 16. Our approach to the tax deductibility of loan loss reserves does not take into account the changes in such deductions that were a part of the 1986 Tax Reform Act. One of the provisions of the act limited loan loss reserves to actual charge offs, although banks can add to their reserves as much as they wish. This provision applied chiefly to banks with assets over 500 million. Furthermore, the act required banks to recapture exisring bad debts reserves into taxable income at a set schedule. Financially troubled banks were relieved of this provision until they were in better condition. Because these changes occurred within the period examined, we chose to incorporate a uniform treatment of loan loss reserves. 17. Although foreign securities earn capital gains, we did not consider these capital gains because the data were unavailable. Assets in trading accounts are typically held for only a short period of time, so capital gains income is likely to be fully realized. Realized gains in trading accounts from the call reports is thus used for the capital gains term in our user cost estimates for trading accounts. The capital gain for commercial real estate loans is set equal to zero because these loans are not typically traded, although an informal secondary market is beginning to form among the money center banks. Residential mortgages, on the other hand, are routinely securitized and sold.

236

Dennis J. Fixler and Kimberly D. Zieschang

markets for commercial and industrial loans, albeit informal, have arisen although the volume of such transactions is small.18 Recently, the financing of leveraged buy outs has been inextricably tied to the ability of the lead lending banks to sell pieces of loans. At some banks, the salability of a loan has become as important as the borrower's credit worthiness.'g We approximate the effect of capital gains for Treasury securities and U.S. government agency and corporate obligations by subtracting the computed sample average interest rate on Treasury securities and agency obligations from the average annual total return data, which are the sum of the market interest rate and the rate of capital appreciation, obtained from the MerrillLynch government master bond index.20 This index includes various maturities of Treasury securities and U.S. agency securities. The capital gains term for government securities other than Treasury securities was approximated by subtracting the computed sample average interest rate on government securities from the total return on these securities obtained from the Merrill-Lynch mortgage master index. Although this set of securities is not solely composed of mortgage-backed securities, all securities in this category were imputed with the mortgage master total return rate. The capital gains on state and local securities are difficult to measure because the major indexes are not total return indexes, but rather simple averages of dealer estimates of what the coupon rate would have to be for a particular issue released on the day of the survey and sold at par. To measure changes in the total return to holding state and municipal securities we examined the Lipper index for the performance of a collection of tax-exempt mutual funds.21A potential problem with this measure is the variability in management performance inherent in a cross-sectional sample of tax-exempt 18. The following information was provided by Chris Bumcrot of Loan Pricing Corp. and gleaned from several issues of Loan Pricing Report, a publication of Loan Pricing Corp. Information about bank practices was also provided by Steve Woods and Nori Marshall, both of Bank of America. 19. Another motivation for the sale of loans is the new risk-adjusted capital reserve requirement schedule that goes into full effect in 1992. Under these requirements, in 1992 a bank would have to set aside $8, $4 of which must be in stockholder's equity, for every $100 of private loans (nongovernment supported or related). These requirements substantially reduce the holding revenue for a loan and thereby encourage loan sales. 20. We thank Chet Ragavan of Merrill-Lynch's Fixed Income Research Department for providing the total return data for the Menill-Lynch government master bond index and the mortgage master index. 21. We are grateful to Julie Friedlander of Lipper Analytical Securities Corp. for supplying the quarterly total return on the Lipper General Municipal Bond Funds. We impute a tax-equivalent return for state and municipal securities by setting the federal tax rate at 46 percent for the years 1984-86, 40 percent in 1987 and 34 percent in 1988. These are the maximum statutory rates, which are applicable for the large banks in our sample. Our imputation does not take into account the percentage of interest payments disallowed. Before 1983 all of the interest incurred by a bank in acquiring tax-exempt securities could be deducted. In 1983, the tax law was changed to disallow 20 percent of the interest incurred for tax-exempt securities acquired after 1982. The 1986 tax law changed the disallowance to 100 percent of the interest incurred for securities acquired after August 6, 1986.

237

User Costs and the Real Output of Banks

mutual funds. However, it is reasonable to suppose that a bank’s management of its tax-exempt portfolio is similar to that of firms specializing in such management, otherwise the bank may be better off contracting out for the management of its tax-exempt portfolio. Because the interest on state and municipal bonds is tax-exempt, we incorporate the interest earned on these securities on a tax-equivalent basis.

6.4.2 Model Estimation and Results Our econometric model relates the asset shares in asset holding and service fee revenue, wi,to the asset portfolio share, s,, the log of the branching characteristics variable a, the logs of outputs (y,, . . . , y 8 ) , and the logs of inputs (x,, x,, xJ, as given in equations (8) and (9). Before estimating the model we checked for influential observations by running a sequence of regression diagnostics on an aggregate loan equation with identical functional form to the equations in our more detailed system. We isolated and deleted observations that were tagged as influential via a heuristic test on the size of the diagonal element of the “hat” matrix corresponding to the observation, using the Belsley, Kuh, and Welsch (1980) recommended cutoff. This test is oriented toward finding observations with unusual leverage on the parameter estimates because of the values of their exogenous variables. An examination of this list of banks revealed that a number were in one of two categories: The first category is banks that were or later became troubled, including Continental Illinois in 1984, and a collection of Texas, California, and Florida Banks in each of the five years. The second category included U.S. subsidiaries of foreign banks, particularly Japanese banks located in New York and California. A third, small category included what were apparently savings and loans that had joined the Federal Reserve System. An examination of the loan holding revenue share for the set of “hat” influential banks also revealed some extreme values, including values less than zero and greater than one. Although our inclusion of provisions for loan losses as part of the loan return means that these values are not ruled out, they were rare outliers in our sample. All told, the diagnostic filter reduced our sample size by approximately 10 percent. A handful of observations were also deleted because of excessively large studentized residuals; that is, they had excessive influence on the model fit. Our sample size was reduced somewhat further by the fact that some banks in the sample had no assets at all in certain categories, making them impossible for our translog model to handle.,, This resulted in a further reduction in sample size of about 5 percent. We present statistics for selected variables from the full and edited samples in table 6.2. We estimated the system of equations (8) for each of the years 1984-88 by 22. Using the generalized quadratic functional form recently discovered by Diewert (1992) may offer a solution to this problem.

238

Dennis J. Fixler and Kimberly D. Zieschang

the iterative seemingly unrelated regressions method, dropping without loss of generality equation (9) because of the singularity of the covariance matrix of the disturbances arising from the fact that X; = ,E, = 0. The parameter of greatest interest to us is 4, the coefficient of the portfolio shares sI. Without exception, $I is estimated very precisely, and exhibits substantial stability over the period, rising from 1984 to 1985, a very profitable year for this group of banks, then declining through 1986 to a low in 1987, a year characterized by a profit squeeze, and rebounding in 1988.23In general, we expected the distance function derived economic shares to be positively related to the log of own output, because this is related to the convexity of the translog functional form we used. Our expectations were met for all the loan categories except other loans and leases, y,, whose own elasticities were positive but insignificant in 1984, negative and significant in 1985, negative and insignificant in 1986, positive and significant in 1987, and negative and insignificant in 1988. The securities share equations in various years also displayed intermittent negative own output elasticities. We did not enforce convexity on the distance function parameters. In light of the fact that the portfolio shares clearly swamped the distance function arguments as explanatory variables for the accounting revenue shares, it is very unlikely that imposing nonlinear restrictions on the coefficients of these variables (convexity constraints) would have appreciably changed the results for 4. Our estimated aggregate opportunity cost rate is the asset-weighted average of the opportunity cost rates of the banks in our full sample, computed as p = c$ * fTA,where fTAis the asset-weighted average rate of total return on assets for the sample, and c$ is the estimate of p/rTAfrom our econometric model. In table 6.3 we present a comparison of our econometric estimate of the opportunity cost rate with several plausible, and relatively easily obtained, alternatives. These are (1) the 90-day Treasury Bill rate, r,; ( 2 ) the rate of return on assets, FA; (3) the required rate to cover the interest cost of liabilities, rREQ; and (4) the opportunity cost rate generating the proposed UNSO financial services imputation. The asset-weighted sample mean total return on assets is computed as

where n indexes banks, and N is the sample size. Rate ( 2 ) ,the asset-weighted sample mean holding return on assets, is computed as 23. Our standard error estimates are not corrected for the studentized residual and DFFITS sample trimming techniques (defined in Belsley, Kuh, and Welsch, 1980) we applied to eliminate outliers and increase the resistance of our parameter estimates to influential observations. These filters truncate the dependent variables of our system, and the precision of our estimates is therefore somewhat overstated. However, very few observations were affected by these filters, and we would expect the bias in our standard error estimates from this source to be low.

239

User Costs and the Real Output of Banks Opportunity Cost Rates Asset-Weighted Means

Table 6.3 Variable Description rm

rREp 'A ~ T A

9 P

poNso

90-day Treasury Required rate Asset rate Total asset rate P/r,

Econometric opportunity cost UNSO opportunity cost

1984

1985

1986

,0952 .0818 .I123 .I223 ,7628 ,0933 ,0970

.0747 ,0686 ,1056 ,1163 ,8214 .0961

,0597 ,0557 ,0895 ,1007 ,6950 .0700 .0726

.0871

1987

1988

,0578

.0667 ,0615

.0558

,0696 ,0824 ,6065 ,0500

,0627

,0910

,1039 ,7385 ,0767 ,0762

Rate (4), the asset-weighted sample mean required rate of return on assets, is computed as

where k,, is the ratio of reserves-currency and coin and deposits at Federal Reserve banks-to deposit and other liabilities for the nth bank.24 Fixler and Zieschang (1991) showed that the imputation scheme used by the BEA for allocating uncharged services provided by banks to customers in the business and final consumption sectors implicitly takes p = f A .They also argue that the proposed UNSO imputation method implicitly assumes that p = [fR,,, + f A ] / 2 . To judge which of these alternatives is closest to our estimated p, we consider the average absolute difference between our estimate of p and these alternatives. We find that our estimated value of p is closest on average to the p given by the UNSO approach, the average absolute difference being 57 basis points, where a basis point is one one-hundredth of an interest rate percentage point. In contrast, the average difference between the 90-day Treasury Bill rate and p is 103 basis points; between the fREe and p it is 149 basis points; and between the BEA-determined p and our p the average absolute difference is 164 basis points. Thus our analysis suggests that the UNSO approach provides a reasonable rule of thumb for the calculation of p.25 Because it is a determinant of the weights by which quantity relatives are 24. k,,is not the same as the legally required reserve ratio for deposits because for many banks the currency and coin holdings are amounts necessary for business, and some banks may hold excess reserves on deposit at the Federal Reserve. 25. It is also worth noting from table 6.3 that the simple average difference between rm and p is positive, with rm exceeding p by substantial amounts in three of the five years, and staying numerically close in the two years when it fell below p. Because rm is a rate on short-duration, risk-free assets, this evidence suggests that our estimate of p reflects the higher risk and (presumably) generally longer average duration of the asset portfolios held by U.S. banks over this period.

240

Dennis J. Fixler and Kimberly D. Zieschang

averaged to form an index number, p is important to the calculation of output quantity indexes. If the overall (aggregate) distance function is translog, as in (l), then, following Caves, Christensen, and Diewert (1982), we can compute the following, exact, period-to-period index number: D(xt--l, at-’, y‘) Q = [D(xf-l, a f - l , y‘-l) -

n 8 *=I

D(x‘, a‘, y‘) D(x‘, a‘, yt-1)

“c”:-’Y:-’ r-1 I

I 8

, Ip:- ‘Y:=

I

P:Yf

l2

I

lP:YJ

where p: = h: - p for assets, and p : = p - h: for liabilities. These expressions are the negative of the user cost expressions in (1) and (2); to enable the y , to be always positive, negative user costs are employed. It follows that a negative price means that the product is a financial input and that a positive price means that the product is a financial output. We find that the financial input-output status of the financial products is fairly constant over the examined period, regardless of the estimate of the opportunity cost rate. For example, with our econometrically estimated p, there were a total of four financial input-output status switches: y , was an input in 1984 and switched to an output in 1985; y s switched from an output in 1986 to an input in 1987 and remained an input in 1988; and y , switched from an output in 1986 to an input in 1987 and back again to an output in 1988. The other financial products were always financial outputs. Interestingly, whenever a product became a financial input, the attending price was close to zero. For example, the price of y , was 4 cents per dollar in 1986, declining to -0.4 cents per dollar in 1987. This small value likely results from the value of the uncharged-for financial services nearly offsetting the explicit charges. The other values of p produced approximately the same number of financial input-output status switches, and these changes were concentrated in y s and y,. The cumulative, chained values of the index (10) and the associated implicit price index, computed under alternative opportunity cost estimates, are presented for the years 1984-88 in table 6.4. We found that the aggregate output of banks increased by approximately 40 percent over the period. In addition, it can be seen that our price and output measures are insensitive to the variations in the level of p represented by the alternatives we consider. One should keep in mind that our yearly sample of banks contains the large banks; that is, we consider more than 400 of the top banks measured in total assets from a total of (approximately) 13,000 banks. In terms of deposits, our sample of banks accounted for 54 percent of total deposits in 1988, and similar magnitudes were encountered in the other years. Because more than 12,000 banks accounted for the remainder of the deposits, it is likely that some of these banks were quite small and that some even experienced negative growth. Coupling this fact with the decline in the number of banks over the period, approximately 9 percent, it is quite likely that industry growth was smaller than 40

User Costs and the Real Output of Banks

241

Price and Quantity Indexes under Alternative Opportunity Cost Rates

Table 6.4 Variable Description

1984

1985

1986

1987

1988

Aggregate Output Quantity Indexes r, rRtP rA P

puNs0

90-day Treasury Required rate Asset rate Econometric opportunity cost UNSO opportunity cost

100.00 100.00 100.00 100.00

110.40 110.61 109.83 110.16

123.19 123.46 122.24 122.80

133.36 133.59 132.85 133.01

140.70 140.92 140.31 140.26

100.00

110.24

122.87

133.24

140.63

Aggregate (Implicit) Output Price Indexes r, rRE, rA P

puNs0

90-day Treasury Required rate Asset rate Econometric opportunity cost UNSO opportunity cost

100.00 100.00 100.00 100.00

110.20 109.99 110.77 110.44

102.60 102.37 103.39 102.92

76.25 76.12 76.55 76.46

96.16 96.02 96.43 96.47

100.00

110.37

102.86

76.33

96.21

percent. On the other hand, our output growth result compares well with the output growth computed in Berger and Humphrey (chap. 7, this vol.) for the same banks during the same period. Of note in our output trend is the substantial, double-digit growth in the years 1984 and 1985. This may be explained by the relatively large interest rate margins (measured as r, - r,,,) of these years. The interest rate margin was 370 basis points in 1985 and fell to 138 basis points in 1987. Moreover, in the years beginning with 1985 there were increases in loan loss reserves, largely for loans to less developed countries and energy-related loans in the Southwest. These developments tended to push the return on assets down. We see correspondingly that the implicit output price indexes began to fall in 1986. 6.5

Conclusion

Hancock’s (1985a, 1985b, 1986) studies of commercial banking showed how financial firms could be analyzed within a traditional neoclassical production model, in concert with the user cost of money innovation of Donovan (1978) and Barnett (1978). Fixler (1988) has shown that Hancock’s translog restricted profit function, in concert with her constructed user cost prices, underlies a practical framework for price and output measurement for financial firms based on the Tornqvist (or other superlative) index number. The chief issue in the user cost framework, as in other applications in investment and durable consumption goods, is obtaining reliable estimates of the components of the user cost price. Our focus here has been on the opportunity cost

242

Dennis J. Fixler and Kimberly D. Zieschang

rate, the single truly unobservable item in the user cost formula. We approach the estimation of the opportunity cost rate as a shadow price problem, and in contrast with Hancock, characterize technology in terms of its production or distance function instead of its restricted profit function. We estimate the relevant parameters of the distance function, under the assumption that the opportunity cost rate is an unknown, constant fraction of the rate of total return on assets. We find that our econometric estimate of the opportunity cost rate is in the same range as several rule-of-thumb estimates, including the return on assets, the required rate of return on assets, and the 90-day Treasury Bill rate. However, it is tightly estimated and significantly different from all of them. Using our estimate of the opportunity cost rate, we construct Tornqvist quantity indexes for the years 1984-88. Over the period output grew 40 percent with double-digit output growth in the years 1984 and 1985. At first blush one might suppose that our output index is sensitive to the opportunity cost of money used. In fact, our superlative quantity index is shown to be insensitive to the variations in the opportunity cost rate given by the rule-of-thumb estimates. The importance of this finding to the production of price and (gross) output quantity-index numbers is that it establishes the commonsense rules of thumb as cheaper substitutes for the more expensive econometrically estimated opportunity cost of money.

References Bagehot, W. 1915. Lombard Street. 14 ed. London: Murray. Bamett, W. 1978. The User Cost of Money. Economics Letters 1:145-49. . 1980. Economic Monetary Aggregates. Journal of Econometrics 14:1 1-48. Barnett, W., M. Hinich, and W. Weber. 1986. The Regulatory Wedge between the Demand-Side and Supply-side Aggregation-Theoretic Monetary Aggregates. Journal of Econometrics 33: 165-85. Belsely, D. A., E. Kuh, and R. E. Welsch. 1980. RegressionDiagnostics. New York: Wiley. Benston, G., G. Hanweck, and D. Humphrey. 1982. Scale Economies in Banking. Journal of Money, Credit, and Banking 14, p. I (4):435-56. . 1985. Operating Costs in Commercial Banking. In Dynamics ofBanking, ed. by Thomas M. Havrilesky, Robert L. Schweitzer, & John T. Boorman. Arlington Heights, 111.: Davidson. Caves, D. W., L. R. Christensen, and W. E. Diewert. 1982. The Economic Theory of Index Numbers and the Measurement of Input, Output, and Productivity. Econometrica 50:1393-1414. Debreu, G. 1951 . The Coefficient of Resource Utilization. Econometrica 19:273-92. Diewert, W. E. 1976. Exact and Superlative Index Numbers. Journal of Econometrics 4~115-45. . 1992. Fisher Ideal Output, Input, and Productivity Indexes Revisited. Journal of Productivity Analysis 3, no. 3 (September). Donovan, D. J. 1978. Modeling the Demand for Liquid Assets: An Application to Canada. International Monetary Fund Staff Papers 25(4): 676-704.

243

User Costs and the Real Output of Banks

Fare, R., H. Fukuyama, and D. Primont. 1988. Estimating Returns to Scale via Shephard’s Input Distance Function. Unpublished manuscript. Southern Illinois Univ., Carbondale. Fare, R., S . Grosskopf, B. Lindgren, and P. Roos. 1989. Productivity Developments in Swedish Hospitals: A Malmquist Output Index Approach. Discussion paper no. 89-3. Department of Economicst Southern Illinois Univ., Carbondale. Fiire, R., S. Grosskopf, C. A. K. Lovell, and S. Yaisawarng. Forthcoming. Derivation of Virtual Prices for Undesirable Outputs: A Distance Function Approach. Review of Economics and Statistics 75. (Expected May 1993.) Fiire, R., and K. Zieschang. 1991. Determining Output Shadow Prices for a Cost Constrained Technology. Journal of Economics 54(2): 143-55. Farrell, M. J. 1957. The Measurement of Productive Efficiency. Journal of the Royal Statistical Society ser. A, 120:253-81. Ferrier, G. D., and C. A. K. Lovell. 1990. Measuring Cost Efficiency in Banking: Econometric and Linear Programming Evidence. Journal of Econometrics 46:22945. Fixler, D. 1988. A Commercial Bank Output Price Index. BLS working paper no. 179. Fixler, D., and K. Zieschang. 1990. Output and Price Measurement for Commercial Banking. Unpublished manuscript. Bureau of Labor Statistics, Washington, D.C., February. . 1991. Measuring the Nominal Value of Financial Services in the National Income Accounts. Economic Inquiry 2953-68. . 1992. Incorporating Ancillary Measures of Process and Quality Change into a Superlative Productivity Index. Journal of ProductiviQ Analysis 2:245-67. Goodhart, C.A.E. 1989. Money, Information and Uncertainty. Cambridge, Mass.: MIT Press. Hancock, D. 1985a. Bank Profitability, Interest Rates, and Monetary Policy. Journal of Money, Credit, and Banking 17(2): 189-202. . 1985b. The Financial Firm: Production with Monetary and Non-Monetary Goods. Journal of Political Economy 93:859-80. . 1986. A Model of the Financial Firm with Imperfect Asset and Deposit Elasticities. Journal of Banking and Finance 10:37-54. Lovell, C. A. K., and K . Zieschang. 1992. A DEA Approach to the Problem of New and Disappearing Commodities in the Construction of Price Indexes. In Data Envelopment Analysis: Theory, Methodology, and Applications, ed. A. Charnes, W. W. Cooper, A . Y. Lewin, and L. Seiford. Forthcoming. McFadden, D. 1978. Cost, Revenue, and Profit Functions. In Production Economics: A Dual Approach to Theory and Applications. New York: North Holland. Malmquist, S. 1953. Index Numbers and Indifference Surfaces. Trabajos de Estadistica 4:209-42. Moorsteen, R. H. 1961. On Measuring Productive Potential and Relative Efficiency. Quarterly Journal of Economics 75:45 1-67. Shephard, R. W. 1953. Cost and Production Functions. Princeton, N.J.: Princeton Univ. Press. . 1970. Theory of Cost and Production Functions. Princeton, N.J.: Princeton Univ. Press. Van Home, J. C. 1983. Financial Management and Policy. 6th ed. Englewood Cliffs, N.J.: Prentice-Hall. Zieschang, K. D. 1985. Output Price Measurement when Output Characteristics Are Endogenous. BLS working paper no. 150. . 1988. Characteristics Approach to the Problem of New and Disappearing Goods in Price Indexes. BLS working paper no. 183.

This Page Intentionally Left Blank

7

Measurement and Efficiency Issues in Commercial Banking Allen N. Berger and David B. Humphrey

Commercial banking is a very difficult service industry in which to measure output, technical change, or productivity growth. First, there is disagreement over which services banks produce and over how to measure them. In addition, banking services are often priced implicitly through below-market interest rates on deposit balances, making observed revenue flows inaccurate guides to choosing the important outputs to include in the analysis. Banking also remains a highly regulated industry in which substantial inefficiencies have been shown to exist. As a result, technical improvements that increase the productivity of the most efficient firms may not be well reflected in the industry as a whole. A further complication is that the deposit side of banking underwent substantial deregulation in the 1980s, including the lifting of effective interest rate ceilings on certain deposits and the creation of new types of accounts. The deregulation directly raised banking costs and shifted the optimal mix between the provision of services and the payment of interest to depositors. Measurement of cost changes and productivity gains must take these factors into account, including the possibility of a period of significant disequilibrium as banks attempted to adjust to deposit deregulation. Despite these difficulties, it is important to analyze the banking industry, as it constitutes almost 20 percent of the U.S. finance, insurance, and real estate service sector of the national income accounts (net of owner-occupied housing). In addition, the externalities that banking generates through its roles as Allen N. Berger is senior economist at the Board of Governors of the Federal Reserve System, and David B. Humphrey is the Fannie Wilson Smith Eminent Scholar in Banking at Florida State University. The opinions expressed do not necessarily reflect those of the Board of Governors or its staff. The authors thank Frank Wykoff for his discussion of the paper. They also thank Tim Bresnahan, Dennis Fixler, Zvi Griliches, Diana Hancock, Stacie Humphrey, John Leusner, Jack Triplett, Kim Zieschang, and the other conference participants for helpful comments and suggestions and Alex Wolman for outstanding research assistance.

246

Alien N. Berger and David B. Humphrey

the nation’s primary financial intermediary and conduit for monetary policy are considered to be important enough to require substantial government protection and supervision. Correspondingly, this paper attempts to meet the challenges mentioned above in measuring output and efficiency in U.S. banking in the 1980s. It is hoped that some of the methods used here will also be applicable to the study of other service sectors. Section 7.1 analyzes some of the problems in defining and measuring bank output. Three methods of choosing which banking functions represent important outputs are evaluated: the asset, user cost, and value-added approaches. The value-added approach, which identifies the major categories of deposits and loans as the important bank outputs, is determined to be the most satisfactory for our purposes. Note that, although deposits are specified as outputs because of their associated service output to depositors, we also specify them as having input characteristics, since they provide much of the supply of funds to be invested in creating loan output. Sections 7.2 and 7.3 examine inefficiency and technical change in banking. The striking degree of cost dispersion in banking, where some firms have average costs several times higher than others with similar scale and product mix, suggests that the standard assumption of equal efficiency underlying most analyses of technical change is invalid for banking. Cost function studies of technical change (e.g., Hunter and Timme 1986) or productivity measures that relate total industry output to inputs (e.g., the Bureau of Labor Statistics [BLS] labor productivity measure) may confuse changes in the minimum cost technology with changes in the deviations from that technology (i.e., inefficiency). We separate these elements here by estimating the change over time in both a cost frontier and the dispersion of industry costs from the frontier. Three methods of estimating a frontier are analyzed, and the thick-frontier method is chosen as most appropriate for the highly dispersed banking data. This method is applied to all 14,000 U.S. banks for the years 1980, 1984, and 1988, which roughly correspond to periods of pre-, mid-, and postderegulation of the deposit side of banking, respectively. Shifts in the cost frontier over time are used to examine the effects of deregulation and technical change. Changes over time in the dispersion from the frontier are also evaluated. In this way, the standard approach to measuring productivity change is decomposed into two parts: the frontier shift and the change in dispersion from the frontier. It is found that the interest rate deregulation of the 1980s and the banking industry’s response to it on balance increased costs in banking, but much of this increase benefited bank depositors through higher interest payments without a corresponding decrease in the provision of deposit services.

7.1 Defining and Measuring Bank Output There is long-standing disagreement over exactly what it is that banks produce. Three alternative methods of choosing bank outputs are analyzed here,

247

Measurement and Efficiency in Banking

the asset, user cost, and value-added approaches. It is argued that the valueadded approach, which defines outputs as those activities that have substantial value added (i.e., large expenditures on labor and physical capital), is best for accurately estimating changes in bank technology and efficiency over time. 7.1.1 The Asset Approach to Defining Bank Output Virtually all observers would agree that bank liabilities have some characteristics of inputs, because they provide the raw material of investable funds, and that bank assets have some characteristics of outputs as they are ultimate uses of funds that generate the bulk of the direct revenue that banks earn. Under the asset approach, banks are considered only as financial intermediaries between liability holders and those who receive bank funds. Loans and other assets are considered to be bank outputs; deposits and other liabilities are inputs to the intermediation process (see Sealey and Lindley 1977). For some large banks that primarily purchase their funds (with interest payments) from other banks and large depositors and turn these funds into loans, this is an adequate description of bank output. However, most banks do much more than purchase their funds-they also provide substantial services to depositors, but these services are not counted as output in the asset approach. Mamalakis (1987) makes the useful distinction between the funds intermediation and deposit services of banks, of which the asset approach considers only the former. Intermediation services transform balance-sheet liabilities into assets and pay out and receive interest to cover the time value of the funds used in this capacity. Although some large banks tend to specialize in this function, most banks raise a substantial portion of their funds through produced deposits and provide liquidity, payments, and safekeeping services (as well as interest payments) to depositors to obtain these funds. For some purposes, the asset approach is the most appropriate. For instance, in a study of loan costs or profitability, a reduced-form model in which the costs and different methods of raising funds are taken to be exogenous may be best. However, any study of banking output as a whole needs to consider a structural form in which the investable funds are an intermediate output of raising deposits, and the services are provided to depositors as partial payment to obtain these funds. The reduced-form asset approach excludes the important differences in service output that occur when the funds are raised via produced deposits versus purchased funds. Moreover, under current institutional arrangements, application of the asset approach to measure banking output often leads to contradictions. For example, consider a bank that produces deposits and sells virtually all its funds to a second bank, which makes commercial loans with these funds. If the two banks merge, there is no change in total banking output, ceteris paribus. However, under the asset approach, if both commercial and interbank loans are considered to be outputs, then measured output would be diminished by the merger because there would be no more interbank lending. If only commercial loans are considered to be out-

248

Allen N. Berger and David B. Humphrey

puts, then the bank that sells funds has no measured output, despite its production of deposit services and the fact that the second bank values the funds purchased. 7 . 1 . 2 The User Cost Approach

The user cost approach determines whether a financial product is an input or an output on the basis of its net contribution to bank revenue. If the financial returns on an asset exceed the opportunity cost of funds or if the financial costs of a liability are less than the opportunity cost, then the instrument is considered to be an financial output. Otherwise, it is considered to be a financial input. Hancock (1985a, 1985b) first applied the user cost approach to banking and Fixler and Zieschang (1990; chap. 6, this vol.) used it to determine the weights applied to bank asset and liability categories to derive indexes of bank output and prices.' The user cost approach determines whether an asset or liability category contributes to the financial output of a bank. The operating costs involved in producing nonfinancial services associated with the asset or liability are not explicitly considered. However, under relatively standard assumptions, these operating costs (inclusive of a normal return on capital) are simply the dual of the user cost approach and are included implicitly. An optimizing bank earns (in financial revenue less operating costs) exactly its opportunity cost of funds at the margin on each asset and pays (in financial costs plus operating costs) exactly its opportunity cost at the margin on every liability.* Thus, to the extent that the user cost approach accurately measures marginal financial revenues and opportunity costs, its allocation is largely on the basis of excluded operating costs, which is almost the same as the basis of the value-added approach described below. However, there are some difficulties in measuring financial revenues and marginal opportunity costs that make the user cost approach to distinguishing outputs from inputs subject to significant measurement error and sensitive to changes in the data over time. A problem with measuring the financial flows associated with balance-sheet items, particularly loans and demand deposits, is that there is some commingling of implicit revenues that cannot be easily disentangled. As discussed further below, banks frequently use compensating balances or pay belowmarket rates on deposits as a method of charging for bank services. Borrowers are often required to hold part of their loan funds as idle demand deposit balances, which means that some of a bank's earnings on a loan are implicit and 1. The user cost approach was pioneered by Donovan (1978) and Bamett (1980) in developing money supply indexes. 2. The underlying assumptions used here are fairly common in banking. If costs and revenues are separable across asset and liability categories and a bank holds securities that it perceives to be in infinite supply (e.g., Treasury Bills), then the quantities of asset and liability categories are adjusted until the marginal revenue less operating cost on every asset and the marginal revenue paid plus operating cost on every liability equal the security rate less its marginal operating cost. See Klein (1971) and Hannan and Berger (1991).

249

Measurement and Efficiency in Banking

are earned by paying less than the opportunity cost of funds on deposits. Further implicit earnings accrue to the bank on a loan when additional balances are kept with the bank for liquidity, clearing, or timing purposes associated with spending the loan receipts. If the ratio of the compensating and conjunctive balances to loans were known, the implicit earnings could be allocated to loans in much the same way that the implicit losses on deposits from reserve requirements are calculated. However, this ratio is not known or estimated and these implicit revenues are instead allocated entirely to deposits. As a result, there is a bias toward finding loans to be inputs or to have a smaller output weight and toward finding demand deposits (where the balances are held) to be an output or to have a higher output weight. Another difficulty is in adjusting opportunity costs for the important characteristics of bank assets and liabilities, including differences in credit risk, liquidity, and duration (maturity). Banks earn substantially higher rates for riskier, less liquid, and longer-term assets and pay substantially higher rates for deposits and other liabilities that are uninsured, have fewer liquidity features, and have longer terms to maturity. Theory requires that each dollar of bank liabilities or assets have the same marginal opportunity cost only after adjustment for these important characteristics. Therefore, the opportunity cost must be adjusted for each category or, equivalently, the financial return or cost of each category must be adjusted before applying a common opportunity cost.3 In practice, these adjustments are difficult to make for every category, although there have been some attempts to do SO.^ When such adjustments are not made or fall short, the determination of outputs from inputs and the weights derived for an index of bank output is biased. The bias toward finding an asset to be an output or have a higher output weight is greater, the longer the maturity, the less the liquidity, and the greater the credit risk, because these characteristics increase the unadjusted rate earned on an asset but are not reflected in the opportunity cost as currently m e a ~ u r e dThus, .~ the matching of liability and asset durations to reduce interest rate risk, the holding of assets and liabilities with varying liquidity features, and the making of loans with different credit risks are all commonplace in banking but may not be well reflected in the application of the user cost approach. A final difficulty is the apparent sensitivity for turning outputs into inputs and vice versa with slight changes in the data or assumptions. When Fixler and Zieschang (1990) switch the assumed opportunity cost for all balance3. As well, ex ante rates and opportunity costs are called for, but only ex post values are observed. 4. E.g., Hancock (1985a. 198%) corrected for credit risks on loans by subtracting off historical average loan losses for each bank. Also, Fixler and Zieschang (chap. 6, this vol.) calculated opportunity costs that differ by bank, which may be viewed as a rough method of accounting for differences in risk, liquidity, and duration across institutions. 5 . The biases go in the opposite direction for liabilities, because banks pay higher rates on longer maturity, less liquid, or riskier liabilities, and these higher rates are subtracted from a constant, unadjusted opportunity cost.

250

Allen N. Berger and David B. Humphrey

sheet items between the average interest rate on loans (the Bureau of Economic Analysis [BEA] method) and the average of interest rates on both loans and deposits (the United Nations Statistical Office [UNSO] method), a number of switches in sign of user cost occur.6 In addition, nearly half of their financial categories (mostly the smaller categories) switch between inputs and outputs over a five-year period, even without changing the method of computing opportunity cost. One would expect banking technology to remain sufficiently constant that the determination of inputs and outputs should not change so often. 7.1.3 The Value-Added Approach The value-added approach differs from the asset and user cost approaches in that it considers all liability and asset categories to have some output characteristics rather than distinguishing inputs from outputs in a mutually exclusive way. The categories having substantial value added, as judged using an external source of operating cost allocations, are employed as the important outputs. Others are treated as representing mainly either unimportant outputs, intermediate products, or inputs, depending on the specifics of the category. A significant difference from the user cost approach is that the value-added approach explicitly uses operating cost data rather than determining these costs implicitly as that part of the return or cost not accounted for by the difference between measured financial flows and marginal opportunity costs. The application of the value-added approach here and in other recent cost studies of the banking industry (e.g., Berger, Hanweck, and Humphrey, 1987) identifies the major categories of produced deposits (demand, time and savings) and loans (real estate, commercial, installment) as important outputs, because they are responsible for the great majority of value added. Purchased funds (federal funds purchased, large CDs, foreign deposits, other liabilities for borrowed money) are treated as financial inputs to the intermediation process, because they require very small amounts of physical inputs (labor and capital). On the asset side, government securities and other nonloan investments are considered to be unimportant outputs, because their value added requirements are also very low.’ Table 7.1 shows the distribution of expenses for labor (salaries and fringe benefits) and capital (occupancy and furniture and equipment expenses) for the largest size class of banks reported in the Federal Reserve’s Functional Cost Analysis (FCA) program for 1980, 1984, and 1988.* In 1988, the two 6. Fixler and Zieschang (chap. 6, this vol.) using a distance function approach, estimate opportunity costs that differ noticeably from the BEA and UNSO opportunity costs. However, the output and price indexes formed using the different opportunity costs were not very sensitive to these differences. 7. Government securities also often play an input role when they serve as required collateral on government deposits. 8. The FCA is a cost-allocationiaccounting system that assigns direct and joint costs to a number of banking functions based on expert information and accounting rules of thumb. The FCA sample includes about 400-600 banks each year and is inclusive of all bank sizes except the largest.

251

Measurement and Efficiency in Banking Distribution of Bank Value Added in 1980, 1984, and 1988 (%)

Table 7.1

Deposits

Loans

Year

Demand Deposits

Time & Savings

Real Estate

Commercial & Industrial

Installment

Total

1980 1984 1988

37 31 36

10 14 12

3 4 4

11 13 14

10 12 12

71 80 78

Source: Board of Governors of the Federal Reserve System, FCA data. Note: Data refer to banks with $200 million to $1 billion in deposits, the largest-size class in the FCA data.

major deposit functions shown absorbed 48 percent of bank value added; three major loan functions absorbed 30 percent, for a total of 78 p e r ~ e n t Similar .~ results are shown for the two earlier periods.’O The outputs identified using value added are similar to those used in the BLS measure of bank labor productivity, which uses a set of aggregate transaction flow data on major deposit and loan services, such as the number of checks written for demand deposits, the number of savings deposits and withdrawals for time and savings accounts, and the number of new loans for real estate, commercial, and installment loans (BLS 1989). Unfortunately, these flow data are not available for all banks. In the analysis below, the deflated values of deposit and loan balances are used as outputs for individual banks. The presumption is that these real dollar balances are proportionate to the underlying transactions and account maintenance service flows for the deposit categories and the transactions, credit evaluation, and monitoring service flows for the loan categories.II Note that, although real deposit balances are used to indicate bank service output, the interest costs on these deposits, which are associated with the role of deposits as providing the input of loanable funds, are specified as well. In the existing literature, deposits are generally treated as either an input or an output, but both characteristics are represented here. I 2 Despite the differences between the value-added and user cost approaches, the two methods do give similar results, in at least some cases. When we 9. Other bank functions, in declining order of importance, are trust (8 percent), credit cards ( 5 percent), and other data services (4 percent). The remaining 5 percent includes nonbanking activities (e.g., insurance), nondeposit funds, and safe deposit. 10. If the FCA data were not available, the value-added approach could be essentially replicated for any sample of banks by applying a statistical cost function to call report data. The coefficients of a regression of labor and capital expenses on the dollar volumes of assets and liabilities can substitute for the percentage of value added to determine the important bank outputs and their weights in an output index. 11. In support of this presumption, Humphrey (1992) showed that a cost-share weighted average of the deflated deposit and loan balances used here yields approximately the same growth rate as does the BLS index of bank transactions for the 1980s. 12. Comments by Frank Wykoff and Jack Triplett have helped us clarify our position on this issue, which is essentially an application of Mamalakis (1987).

252

Allen N. Berger and David B. Humphrey

apply our value-added weights to the same group of banks for the same time period (1984-88) as in Fixler and Zieschang (chap. 6 , this vol.), we obtain a 7.6 percent annual growth rate, similar to their rate of about 8.8 percent. 7.1.4 Implicit Revenues versus Explicit Revenues in Banking Much of the disagreement surrounding the choice of bank outputs can be traced to the fact that bank services are not priced in the same manner as services provided by other industries. In many cases, the pricing is implicit for institutional and regulatory reasons.'? On the loan side, most of the revenue is explicit interest and fees. However, as discussed above, some implicit revenue is raised by business borrowers holding additional idle demand balances with the bank. On the deposit side, revenues from the compensating component of deposit balances, defined as the bank's earnings owing to payment of below-market interest rates, dominate explicit revenues. l 4 As shown below, these implicit revenues currently account for over 80 percent of the revenue raised on deposits, although an unknown (but small) part of this figure is implicit revenue for loans. This suggests that in banking, unlike other industries, explicit revenues are an unreliable guide to determining outputs or service flows. If banks paid market rates on all deposits and charged explicit fees for all deposit services, then this large explicit revenue flow would be convincing evidence that deposits provide substantial service output. Thus, much of the controversy regarding the treatment of deposits as an input or an output arises because the explicit revenues on deposits are relatively small. Another problem with the use of revenue data is that the proportion of revenue that is explicit is not constant. The deregulation of bank deposits caused a significant increase over the 1980s in the proportion of revenues that were explicit, from 5 percent in 1980 to 1 1 percent in 1984 to 18 percent in 1988. Table 7.2 illustrates these points, showing estimated breakdowns of deposit revenues and costs for 1988. The implicit revenue from deposit balances is computed as follows:

(1)

Implicit revenue = (1

-

r,/r,,)[balance,

. (1

- RR,)] r,,

where r, = the average interest rate paid on balance j ; rFF= the federal funds rate (a market rate); balance, = the value of thejth deposit balance; RR, = the 13. E.g., idle deposit balances for loan borrowers are negotiated on a case-by-case basis and provide a way to adjust the loan price without altering a very visible and comparable interest rate. For depositors, regulation forbids interest on demand deposits and formerly put ceilings on other deposit rates as well. The use of indirect pricing, such as minimum balance requirements at below market interest rates, also makes comparison shopping difficult. 14. A below-market interest rate on a deposit is equivalent to a zero-rate compensating balance on part of the deposits and a market rate on the remainder. E.g., if the actual rate paid is twothirds of the market rate, then the implied zero-rate compensating balance is one-third of the total deposit balance. For a demand deposit, which has a zero interest rate, the entire balance is compensating.

253

Measurement and Efficiency in Banking

Table 7.2

Deposit Revenues and Costs for All U.S. Banks, 1988

Source of Revenues and Costs Revenues: Value of compensating balances (implicit revenues) Demand deposits Time & savings deposits Other deposits Total implicit revenues Explicit revenues from fees on deposits Total deposit revenue Allocated operating costs: Demand deposits Time & savings deposits Other deposits Total operating costs allocated to deposits

Value (in billions of 1988 dollars)

26.7 14.5 0.7 -

41.9 9.4 51.3

20.5 21.4 5.5 -

47.4

Source: Revenues are calculated from the call report and costs are allocated from the FCA data. See Berger and Humphrey (1990, appendix table AIA) for more details.

reserve requirement on balance j ; and rTB= the 90-day Treasury Bill rate, a standard-earnings credit rate applied to compensating balances. The first term in (1) compares the rate paid on deposits (r,) with the market rate (rFF)and determines the proportion of balance, that is purely compensating. This compensating component is then adjusted for nonearning required reserves (1 RR,) and evaluated using a standard earnings credit rate (rTB),giving the implicit revenue flow. The top half of table 7.2 shows the estimated implicit and explicit revenues for deposits for all banks in the United States in 1988. Implicit revenues ($41.9 billion) account for 82 percent of the $5 1.3 billion in total deposit revenues. About two-thirds of the implicit revenue is generated from demand deposits; one-third is generated by time and savings deposits. The bottom half of the table shows the allocation of operating costs to the deposit categories using ratios of FCA costs for that year. The overall cost estimate of $47.4 billion is just a little below the estimated revenues of $5 1.3 billion. Note that the slightly higher revenues than costs may be expected because some of the revenues from demand deposits are actually implicit revenues on loans. Two final conclusions are suggested by these data: First, the finding of a large amount of total (implicit plus explicit) revenue on both demand deposits and time and savings deposits supports the finding under the value-added approach that both types of deposits have output characteristics. Under the asset approach, neither of the deposit types is an output, and under the user cost approach as applied to date, usually only demand deposits are outputs.1s Sec15. An exception is Fixler and Zieschang (chap. 6, this vol.), who do not distinguish among deposit categories, but rather include all deposits and some other purchased funds in a single output category.

254

Allen N. Berger and David B. Humphrey

ond, if one uses procedures applied to other industries to measure gross output in banking-namely, looking only at the explicit revenue flows-then deposit output is understated by about 80 percent. Moreover, the shifts over time from implicit to explicit pricing can give the false impression that the level of bank output is increasing, even if total revenues (implicit plus explicit) and total output may not have changed.

7.2 Inefficiency and Cost Dispersion in Banking If all banks are approximately equally efficient, as is assumed in most bank cost studies, then it is appropriate to examine technical change and productivity growth over time using the data from either all banks or from a representative sample. However, if banks are not close to being equally efficient, then cost function measures of technical change, such as in Hunter and Timme (1986), or measures of average productivity change, such as the BLS labor productivity index, may confuse shifts in the minimum-cost technology with changes in the dispersion of bank costs away from the minimum-cost technology. We try to separate these elements by forming a thick-frontier cost function for relatively low-cost banks. In this section, we examine inefficiency and cost dispersion away from this frontier; in the following section, we examine changes in this frontier over time. Banking costs show a striking degree of dispersion. In many cases, banks have costs that are several times higher than other banks with similar scale and product mix. This cost dispersion could be due to many factors, including simple inefficiency. Here, we estimate a thick-frontier cost function using data from banks in the lowest average cost quartile, compare it to a cost function for banks in the highest average cost quartile, and then decompose the difference. The residual that cannot be explained with the available variables is assumed to be a reasonable representation of inefficiency. Some evidence cited below supports this view, specifically (1) high-cost banks experienced much greater failure rates than low-cost banks, (2) the set of banks that were low cost were stable over 1980-88, and (3) low-cost banks consistently had the highest profits. 7.2.1

Cost Dispersion in Banking

Figure 7.1 shows for banks in branching and unit banking states the variation in average operating plus interest cost per dollar of assets by bank size class for 1988.16 AC,,, shows the minimum cost per dollar intermediated; AC,,, AC,,, and AC,,,, are the average costs for the low-cost quartile, highcost quartile, and overall mean, respectively.” The sample was divided into 16. Branching and unit banking states are treated separately here (as in other studies) because of the significantly different regulatory and competitive environments. 17. There were too few large banks in unit states in 1988 to form quartiles for the top two size classes, so only the mean is shown for these classes in figure 7.1B.

255

Measurement and Efficiency in Banking

.

A: Branch Banking Slates/l988 I

' L ,

B: Unit Banking Stated1988

.13 .12 .ll -

''

ACa4

\ \

Fig. 7.1 Average costs by size class and cost quartile

256

Allen N. Berger and David B. Humphrey

size classes prior to forming the cost quartiles to ensure reasonable representation of all sizes of banks across quartiles and to limit the problem of dividing up the quartiles on the basis of a function of the dependent variables in the cost regressions below. For both branching and unit states, the data show large variations in average costs between the lowest-cost quartile (AC,,) and the highest-cost quartile (AC,,) for banks within the same size class where size and product mix variations are relatively small. To illustrate how important this dispersion is in light of the highly leveraged nature of banking, consider a typical bank, which has a 6 percent capitaliasset ratio, earns 1 cent per dollar of assets, and has a return on equity (ROE) of 16.7 percent. An increase in costs of 3 cents, the typical difference between AC,, and AC,,, will result in an ROE of -33.3 percent and wipe out equity capital in 3 years, all else equal. The marked cost dispersion also appears to dominate the relatively smallscale and product mix economies in banking. For the 10,961 banks in branching states, the costs for the highest-cost quartile (Q4) are 36 percent higher on average than for the lowest (Ql); the maximum difference in costs across size classes (taken from the ACME*, curve) is only 10 percent.Is The size of these cost differentials strongly suggests that banks are not close to equally efficient, as is assumed in conventional banking studies. 7.2.2 Frontier Approaches to Measuring Inefficiency With the exception of engineering-based analyses, production technologies are essentially unknown. As a result, inefficiencies must be measured relative to some cost or production “frontier” that is estimated from the data. Accordingly, measures of inefficiency reflect deviations of costs or input usage away from some minimal levels found in the data, rather than a true technologybased minimum. The difference among techniques found in the efficiency literature largely reflect differing maintained assumptions involved in estimating the location of the efficient or best-practice frontier. The major difficulty in estimating a frontier cost or production function is in disentangling inefficiencies from random measurement error and luck. The econometric approach (e.g., Ferrier and Love11 1990) estimates a frontier cost function where the (composed) error term includes both inefficiency and random error, which are assumed to be orthogonal to the regressors. The two components are separated by assuming that the inefficiencies are drawn from a half-normal distribution and the random errors are drawn from a normal distribution. Unfortunately, the location of the frontier is highly dependent on the actual shapes of the two distributions. As pointed out by Greene (1990) and Stevenson (1980), the half normal is rather inflexible and embodies an assumption that most observations are clustered near full efficiency, with 18. For the 1,844 banks in unit states, Q4 costs are 30 percent higher on average than Q l ; the maximum difference across size classes is 20 percent.

257

Measurement and Efficiency in Banking

higher degrees of inefficiency being decreasingly likely. This runs counter to the observed bank cost data (fig. 7. l), which suggest a relatively thick-tailed, unskewed distribution of costs. The data envelopment analysis (DEA) approach (e.g., Aly, Grabowski, Pasurka, and Rangan 1990) avoids distributional assumptions by using linear programming techniques to estimate frontiers that connect the input requirements of the efficient firms. Unfortunately, it does so through the ad hoc assumption that there is no random error-all variation not in the inputs is treated as reflecting inefficiency. If random error does exist, it can have a large cumulative effect on aggregate inefficiency because this measure is determined by comparing the few fully efficient firms on the frontier with all other firms not on the frontier. As indicated in figure 7.1, the lowest-cost observations (AC,,,) have costs far below both the mean (ACMEAN) and the average of the lowest-cost quartile (AC,,), indicating that a substantial degree of random error may be present. This paper views the measurement of inefficiencies from a different perspective and uses a set of ad hoc assumptions that are somewhat more intuitive and better justified by our data. Instead of trying to estimate a precise cost or production frontier edge, we estimate a “thick-frontier’’ cost function for the lowest average cost quartile of banks, where it may be reasonably assumed that banks are of greater than average efficiency. A cost function is also estimated for the highest-cost quartile, in which banks are presumably of less than average efficiency. Differences between these two cost functions are then divided between market factors (e.g., scale, product mix, branches) that are not easily attributable to inefficiency, and a residual, which we assume reasonably represents inefficiency. This inefficiency is then decomposed into several components. In the figures, these differences are roughly represented by the difference between the AC,, and AC,, lines. The exact maintained assumptions here are that the error terms within the lowest- and highest-cost quartiles reflect only randomly distributed measurement error and luck and that the differences between the lowest- and highest-cost quartiles reflect only market factors and inefficiencies. A benefit of the thick-frontier approach is that it requires less specificity in the maintained statistical assumptions, and therefore is less likely to be substantially violated by the data. First, the assumption that the inefficiencies are uncorrelated with the regressors, maintained in the econometric approach, is not needed. Second, our assumption that the error terms for the quartiles satisfy standard regression properties seems no worse than (a) the econometric approach assumption that inefficiencies are from an arbitrary (half-normal) distribution, or (b)the DEA assumption that random error is zero. Third, even if the error terms within quartiles represent inefficiencies, rather than only random error as maintained, the thick-frontier approach remains a valid comparison of the average inefficiencies of high- and low-cost firms. Finally, as discussed below, the cost quartiles are quite stable over time and are inversely

258

Allen N. Berger and David B. Humphrey

related to long-term profits, both of which are consistent with the cost differences between quartiles reflecting long-term inefficiencies.I9 7.2.3 Specification of the Thick Frontier A separate equation is specified for each of three types of costs: physical operating costs, interest costs on produced deposits, and interest costs on purchased funds (which sum to total costs). This permits the use of known, exact prior information on which types of bank outputs affect which types of costs and also allows us to draw separate conclusions about inefficiencies in each of these three cost areas. As discussed in section 7.1, the specified bank outputs are two types of produced deposits-demand and time and savings deposits (DD and TS)-and three types of loans-real estate, commercial and industrial, and installment loans (RE, CI, and IN). Inputs are labor ( L ) , physical capital ( K ) , and purchased funds (PF). There is one translog cost equation for each of the three types of cost and an input share equation for operating expenses: 5

lnOC

=

a1

+ C P; lnYi + -2l C5 C i= I

j = ]

lnYilnYj

j=l

n= I

I ~ B

m=l

i= 1

L

+ A;

I=

I

where OC = operating costs owing to (1) labor and (2) physical capital and other expenses; SOC, = share of operating costs paid to input 1 (labor); ID = interest on deposits (demand and retail time and savings deposits); IPF = interest on purchased funds (federal funds purchased, large CDs, foreign de19. For a more extensive discussion of the differences among the frontier approaches, see Berger and Humphrey (1991).

259

Measurement and Efficiency in Banking

posits, and other liabilities for borrowed money); Yi = real dollar amount of output i, (1) demand deposits, (2) time and savings deposits, (3) real estate loans, (4) commercial and industrial loans, and (5) installment loans; OA = other assets; B = number of banking offices; and w,,, = price of input m, ( 1 ) labor and (2) capital. Coefficients are indicated by a,p, A, T, y, and p; error terms are indicated by E . The superscripts on the coefficients and error terms signify the equation numbers and the$ subscripts on equation (5) refer to size class and quartile.,O All dollar-value data are in real terms (using the GNP deflator), as are all the cross-year comparisons shown below. Note that the characteristics of deposits as both inputs (interest costs, ID) and outputs (real dollar values Y ,, Y,, reflecting transactions and account maintenance service flows) are included in the model.

7.2.4 Bank Inefficiency Measures and Empirical Results Decomposition of diyerences between the highest- and lowest-cost quartiles. The model shown in equations (2)-(5) was estimated by iterative seemingly unrelated regression (ITSUR) for banks in the lowest-cost quartile and for banks in the highest-cost quartile (performed separately for banks in branching and unit banking states). For each size class, the proportionate difference in unit costs between high-cost and low-cost banks to be decomposed is

eQ1

where ACQi _= eQ1(XQi)ITAQi, is the predicted cost function using the parameter estimates of equations (2)-(5) obtained using the Qi data, XQ'is the vector of mean outputs and other regressors for the size class for the ith quartile, and T A Q ~is the mean total assets for the size class for the ith quartile (size class scripts are suppressed for expositional ease). Thus, Diff is the proportional increase in predicted unit costs of Q4 data relative to the Ql data, evaluated at the mean of each size class. Differences in output levels and mix, branch offices, other assets, input prices, and purchased funds levels are not necessarily the result of inefficiencies. These are attributed to exogenous differences in the local markets in which banks operate. Therefore, the part of Diff owing to these data differences is referred to as the market component, or (7) 20. The purchased funds equation (5) is restricted so that there are no scale or product mix effects within a size class-quartile pairing. This corresponds to a national market for these funds in which every bank has virtually the same opportunities. However, banks in different quartiles may pay different average rates and have different efficiencies because they take different positions in the market with respect to maturity structure or funds type or because they respond differently to changes in market conditions. This restriction improved estimation performance considerably. Additional details of the model are in Berger and Humphrey (1991).

260

Allen N. Berger and David B. Humphrey

where ACQ4* = eQ1(XQ4)/TAQ4. Equation (7) differs from ( 6 ) in that the predicted cost for Q4 data is evaluated using the efficient technology (estimated from the Q 1 thick-frontier cost function), rather than the inefficient technology (estimated from Q4 data). Embedded in the computation of &'(XQ4) is the assumption that an efficient firm would pay the average interest rate on purchased funds actually paid by Q l firms. Thus, Market captures the effects on costs of differences in the levels of the data (XQ" versus XQ'), but not in the cost function, because costs are evaluated using only the parameters from the efficient cost function (&I). The remaining differences in average costs that cannot be attributed to output levels and mix, branch offices, other assets, input prices, and purchased funds levels are assumed to be owing to inefficiencies:

(8)

Ineff = [ACQ4 - A(?Q4*]/ACQ' = D i f - Market.

Zneff captures only the difference in the estimated cost functions that are taken to represent inefficiency, holding the data constant at Q4. Included in Zneff are financial inefficiencies in the payment of produced deposit and purchased funds interest, as well as operating inefficiencies in the use of physical labor and capital. 21 Inefficiencies can be decomposed into several sources by examining the differences in predicted costs attributable to each cost equation separately. For example, the proportion of Ineff owing to operating cost inefficiencies is given by

(9) where A e E and AeEF in the numerator indicate the same predicted average costs as in the denominator, except that only operating costs are included. The inefficiencies owing to interest on deposits and purchased funds are computed in similar fashion.** Market factors and bank ineficiency. Table I . 3 shows the value of Diff and its decomposition for banks in both branching and unit banking states for 1980, 1984, and 1988. The table includes computations for the overall mean of the data and for the mean exclusive of banks in the largest size class (over $10 billion in assets), which are not well matched in size and can be distorting. The differences in predicted costs range from 19 percent to 44 percent for 21. The inefficiency measure also reflects cost diffeiences among banks not specified as market factors-quality differences, left-out variables, and measurement errors. These additional effects are believed to be small. Banking output is quite homogenous across banks within a size class, so quality differentials are negligible. The problem of having a limited number of regional prices for the capital input is of greater concern, but this is mitigated by the fact that capital has only about a 15 percent share in total costs. 22. Data and variable definitions are given in more detail in appendix table 7A. 1. The results are virtually unchanged when the large banks (assets over $1 billion) are dropped, but the robustness of the large bank results cannot be verified because there are too few of them per quartile to estimate a separate cost function with confidence.

261

Measurement and Efficiency in Banking

Table 7.3

Year

Decomposition of Costs between Highest- and Lowest-Cost Quartiles 1980-1988 (%) Difference in Predicted Average Costs (Dim

Total Market Factors

Total Inefficiencies

(Market) (2)

(IneKI (3)

(1)

Branch Banking States 1980: Overall mean Mean < $10 billion 1984: Overall mean Mean < $10 billion 1988: Overall mean Mean < $10 billion

26.1 30.3

9.0 5.7

17.1 24.7

26.4 28.4

2.9 3.6

23.6 24.8

43.8 35.1

2.2 4.2

41.7 30.9

31.9 30.9

7.0 4.6

24.8 26.3

19.2 21.7

0.1 1.5

19.1 20.2

19.5 27.0

-6.2 0.8

25.7 26.2

Unit Banking States 1980: Overall mean Mean < $10 billion 1984: Overall mean Mean < $10 billion 1988: Overall mean Mean < $10 billion

~

Note: Columns (2) and (3) sum to column ( I ) . See Berger and Humphrey (1990, appendix tables A2A, A2B, and A2C) for size class detail.

all banks over all time periods (col. l), similar to the raw data in figure 7.1. When this difference is decomposed into market factors (col. 2 ) and a residual reflecting inefficiencies (col. 3), the inefficiencies clearly dominate. Also, when the results are disaggregated by size class (not shown), the smallest firms show the greatest inefficiencies, consistent with figure 7.1 .23 Decomposing ineficiency into operating and Jinancial components. Table 7.4 shows the decomposition of inefficiencies (Znef) for the same three years. For banks other than those in the largest size classes, operating cost inefficiencies (col. 1) are generally substantially greater than either of the financial (interest cost) inefficiencies (cols. 2 and 3). For the largest banks, purchased funds inefficiencies generally are largest and significantly affect the figures shown for the overall mean. This is because purchased funds are intensively used by large banks (so their weight is higher) and because the rates on these funds are quite volatile and banks differ in their speeds of adjustment to relative rate changes among purchased funds categories. During this period, there were a number of significant regulatory changes that ( a ) removed interest rate ceilings on savings and small time deposits

262 Table 7.4

Allen N. Berger and David B. Humphrey Decomposition of Inefficiencies between Highest- and Lowest-Cost Quartiles, 1980-1988 (%)

Year

Operating Cost (IneffJ (1)

Produced Deposit Interest (Ine&;,) (2)

Purchased Funds Interest (IneffJ (3)

Brunch Bunking States

1980: Overall mean Mean < $10 billion 1984: Overall mean Mean < $10 billion 1988: Overall mean Mean < $10 billion

9.9 13.9

3 .O 5.0

4.2 5.8

12.7 16.4

2.8 3.9

8.1 4.5

22.6 25.9

1.1 1.2

18.0 3.8

Unit Banking States

1980: Overall mean Mean < $10 billion 1984: Overall mean Mean < $10 billion 1988: Overall mean Mean < $10 billion

14.6 17.7

2.6 3.6

1.6 5 .O

14.0 15.7

-0.2 -0.1

5.3 4.6

21.3 25.4

- 0.3 -0.1

4.7 0.9

Notes: Percentages add up to total inefficiencies. Columns (1). (2). and (3) sum to column (3) in table 7.3. See Berger and Humphrey (1990, appendix tables A3A, A3B, and A3C) for size class detail.

(starting in 1981 and completed in 1986); and (b) permitted banks to offer checkable consumer accounts that paid an uncontrolled interest rate (starting in 1981 and expanded in 1982). From this perspective, 1980 may be viewed as a prederegulation period, 1984 as a mid-deregulation period, and 1988 as a postderegulation period by which time the adjustments to deregulation may or may not have been completed. The usual expectation is that deregulation reduces inefficiency in the long run, but there is some question as to how long that process takes. As seen in table 7.4, operating cost inefficiencies, the main source of inefficiencies and the one expected to be most affected by deregulation, remained approximately constant from 1980 to 1984 and then increased significantly from 1984 to 1988. This pattern was particularly pronounced for larger banks. Thus, operating cost dispersion has increased, and it appears that the adjustment process to the new less regulated equilibrium may not yet be completed. Prior to the 1980s, banks substituted operating expenses (more convenient offices and free deposit services) for their inability to pay market rates on all deposits. After interest ceilings were raised and many zero-interest consumer demand bal-

263

Measurement and Efficiency in Banking

ances shifted to interest-earning checking accounts, the substitution of operating costs for interest expenses was reversed. The optimal mix between the provision of banking services and the payment of interest to depositors shifted in favor of the latter, but movement to the new equilibrium mix took time because it required closing branches and other capital changes, staff reorganizations, and so on. Some additional data on the changes in real deposits per branch office tend to support this explanation of the time pattern of inefficiencies. From 1980 to 1984, real deposits per branch office grew 3.8 percent for banks in branching states and then increased by 13.2 percent from 1984 to 1988. This difference is even more pronounced for the larger size classes. Apparently, it took several years for banks to arrange branch closings and mergers to reduce the service/ interest ratio toward its new equilibrium and large banks on average moved at a faster pace. One reason for the delay is that many banks likely had new branches in the planning and building pipeline before determining the full effects of deregulation. Consistent with this explanation, the total number of branches nationwide continued to grow, but at a decreasing rate over the 1980s, even as many large banks were closing b r a n c h e ~ . ~ ~ Technical versus allocative ineficiency. Operating inefficiencies may be further decomposed into their technical and allocative components, which derive from proportionate overuse and incorrect mix, respectively, of the physical labor and capital inputs. Using the methodology of Kopp and Diewert (1982) and Zieschang (1983), the main result (not shown) is that almost all of the operating cost inefficiencies are in the technical category, with less than 10 percent owing to allocative inefficiencies for all three years analyzed.

7.2.5 The Relationship between Cost Dispersion and Bank Failure The importance of cost dispersion or inefficiency in banking ultimately depends on whether banks identified as high cost have difficulty competing. This issue is examined by determining the relationship between costs and bank failures, the premiere measure of competitiveness. Over the nine years from 1981 to 1989, 1,074 banks failed, a substantial increase over previous postwar decades when typically fewer than ten banks per year failed. Without question, the deregulation of deposits in the early 1980s played a part in raising failure rates, raising costs directly through the removal of interest ceilings on deposits and increasing the competition among banks. Although it is not possible to say that high costs by themselves caused any banks to fail, the analy23. See our working paper (Berger and Humphrey 1990, appendix tables A2A, A2B, A2C) for this disaggregation. 24. Total banking offices grew 3.2 percent from 1980 to 1981, 1.8 percent from 1984 to 1985, and 1.6 percent from 1987 to 1988. At the same time, many large banks cut branch operations and staff severely. As examples, Bank of America cut branches by 27 percent (about 350 offices) and staff by 34 percent; Manufacturers Hanover reduced staff by 24 percent. Despite these and many similar cuts, a study for the American Bankers Association (Booz-Allen and Hamilton 1987) reported that as of 1986 about half of all branches remained unprofitable and required further cost cutting.

264

Allen N. Berger and David B. Humphrey

sis below suggests that having relatively high costs is a consistent associated factor. For each of the three years 1980, 1984, and 1988, all banks have been ranked into four average cost quartiles (as noted above). From these rankings, the cost quartile position for each bank that failed in a subsequent year was determined. The summary results are presented in table 7.5. Of the 768 banks that failed over the nine years, 1981-89, and had been started up by 1980 and had complete call report data for that year, 41 percent were in the highest-cost quartile (Q4) during 1980. This is more than three times as many as the 13 percent that were in the lowest-cost quartile (Ql). Of the 748 that failed after 1984 but existed and had complete data for 1984, 57 percent were from Q4, more than eight times as many as the 7 percent that were in Q l . Finally, of the 178 banks that failed in 1989 and had complete data in 1988, 66 percent were ranked as having the highest costs, almost 15 times as many as were ranked in the lowest-cost quartile. As these results indicate, high-cost banks incur an appreciably greater probability of failure, and this probability increases as the time of failure nears. There are several possible reasons for this positive relationship between high costs and bank failure: First, the high leverage-low spread nature of banking means that a relatively small increase in costs can wipe out earnings and financial capital relatively quickly. Second, high costs tend to be symptomatic of poor management in general. Firms that control costs poorly also tend to have poorly conceived loan policies that contribute to a high failure probability. Third, high costs reduce expected rates of return on equity, ceteris paribus, which may induce a high-cost bank to increase expected return by undertaking more risky activities (i.e., shift to a point further out on its riskexpected return possibilities frontier), increasing the probability of failure. This may be accentuated by the moral hazard aspects of FDIC insurance. When either of the latter two explanations hold, costs contribute to the failure, but the reported reason may be fraud or a high-risk loan portfolio. 7.2.6 The Stability and Relationship to Profits of Low-Cost and High-Cost Banks With the exception of the small effect of the market factors, the differences in costs between the high- and low-cost banks have largely been attributed here to an inefficiency residual. However, it is also possible that these cost differences may reflect short-term differences in luck or omitted variables such as product quality or risk. In this section, we investigate these alternatives by examining the stability of the cost quartiles and their relationship with longterm profits.25 25. Another potential explanation of the cost differences between quartiles, differences in monopsony power. may be ruled out as unimportant. The component of the market factors owing to differences in input prices for capital and labor is trivial. Also, although some previous research suggests that banks exercise monopsony power in setting deposit interest rates (see Berger and

265

Measurement and Efficiency in Banking

Table 7.5

Cost Quartile Ranking of Banks That Failed over 1981-1989 (%) Failure Percentage by Quartile Cost Quartile

1980

1984

1988

Q4 (highest cost)

41.4 24.1 20.6 13.3

57.4 24.5 11.2 7.0

65.7 21.3 8.4 4.5

768

748

178

43 Q2 QI (lowest cost) No. of failed banks operating in year of quartile ranking

Table 7.6

Stability and Relation to Profits of Cost Quartiles: Correspondenceof Low-Cost and High-Cost Banks for three Single Years with Cost and Profit Quartiles Formed Using Data from 1980,1984, 1988 Combined Banks in Cost Q1 for a Single Year (%)

Q1

42

Branching states Unit states

76 75

20 20

Branchingstates Unit states

49 49

27 30

Banks in Cost Q4 for a Single Year (%)

42

43

Q4

No. of Banks

Long-term Cost Quartiles 4 1 0 4 1 0

3 4

21 21

76 74

5,403 1.625

Long-term Profit Quartiles 15 9 11 14 7 14

18 14

26 24

45 48

5,403 1,625

43

44

Q1

The upper half of table 7.6 examines the stability or consistency over time of the lowest- and highest-cost quartiles (Ql and Q4, respectively) by showing the correspondence between the quartiles for the three individual years separately and quartiles formed by combining the cost data from all three years together. To avoid the problems of entry, exit, merger, and altered branching laws, we focus on only those banks that (1) existed in all three years and (2) were in states that did not switch between unit and branching status. As shown, 76 percent of banks we rank in Ql using a single year’s data also had costs in the lowest quartile for the three years combined. That is, 76 percent of the 1980, 1984, and 1988 Q l banks were also in Ql for the combined data set. Further, 96 percent of Ql banks had costs over time below the median and only 1 percent had costs in the highest quartile using all three years’ data. Similarly, 76 percent of the banks we rank in Q4 had costs in the highest quartile for the three years combined, 97 percent remained above the median, Hannan 1989), such an effect on total costs would be small, because differences in predicted deposit interest costs constitute only about 12 percent of the total predicted cost difference.

266

Allen N. Berger and David B. Humphrey

and less than half of 1 percent were in the lowest quartile for the three years combined. These stability results suggest that the differences in costs between quartiles do in fact represent long-term differences in firm-specific efficiencies, rather than short-term differences in luck or measurement error, because the latter explanations would imply little stability in the quartiles over time. Moreover, our efficiency comparisons between banks in the lowest- and highest-cost quartiles would only in very rare circumstances involve a misordering of efficiencies.26 The lower half of table 7.6 examines the relationship between the cost quartiles for individual years and long-term profits by computing profit quartiles that average net income per dollar of assets for the three years combined. The data show that the costs are strongly negatively related to long-term profitsbeing in the lowest-cost quartile (Ql) in a single year makes it from five to seven times as likely that a bank will be in the highest long-term profit quartile (Ql) than in the lowest profit quartile (Q4). Similarly, being in the highestcost quartile for a single year makes it from three to four times as likely that a bank will be in the lowest rather than highest profit quartile.*’ These data are consistent with inefficiency being the dominating explanation of the difference in costs between quartiles but are not consistent with the omitted effects of product quality differences or bank risk being dominant. If high-cost banks simply spent more on service quality and were reimbursed on the revenue side, then the cost quartiles would not be highly related to profits. Similarly, if high-cost banks simply chose high risk-high expected return financial strategies and had high costs because they paid high-risk premia for their funds, then costs would be positively, rather than negatively, related to profits on average. 28

7.3 Shifts in the Thick Frontier over Time Overall unit costs in an industry can change over time because of (1) technical innovation (reflected primarily by shifts in the minimum cost frontier); 26. An alternative method of examining stability, which yielded similar results (not shown), was also employed. It was found that being in either Q l or Q4 in 1980 made it three to five times more likely that the bank would again be in that quartile in 1988, rather than migrating to the other extreme. Our use of cost stability over time to identify the presence of inefficiencies from luck is analogous to Gordon’s (1965) use of cost stability over product lines to identify managerial inefficiency in airlines. 27. There is also evidence that the strong, negative relationship between costs and profits held in the 1970s as well as in the 1980s. See Kwast and Rose (1983). 28. One caveat to this analysis is that average total costs, the basis of the quartile rankings, is functionally related to the dependent variables in the cost regressions and this could bias the slope coefficients. That is, the quartiles are based on (OC + ID + IPF)/TA, while InOC, InlD, and lnlPF are the dependent variables. However, for several reasons discussed in Berger and Humphrey (1991), this does not appear to present a serious problem. The most important reason is that the quartiles were formed separately by 13 size classes, which removes the great majority of the relationship between the cost variables and the quartiles, because the overwhelming determinant of InOC, lnlD, and l d P F is bank size. E.g., the smallest bank in the largest size class (over $10 billion in assets) is more than I ,OOO times larger than the largest bank in the smallest size class (less than $10 million in assets).

267

Measurement and Efficiency in Banking

(2) changes in average efficiency or the variability of market conditions (reflected primarily by changes in the dispersion of costs from the frontier); and (3) the effects of regulatory changes and other disequilibrium phenomena (reflected in both shifts in the frontier and changes in the dispersion from it). The dispersion analysis of the previous section showed ( a ) that banking costs were substantially dispersed from the frontier; (b) that this dispersion was largely dominated by efficiency differences; and ( c ) that the dispersion increased considerably between 1984 and 1988. Results (a) and (b)suggest that efficiency was important and result ( c ) suggests that there may have been some disequilibrium effects from deregulation remaining in 1988. This section focuses on frontier shifts over time in order to determine the effects of technical change and deregulation as well as the associated disequilibrium. By confining attention to the thick frontier, the effects of technical change and regulation can be determined without the confounding influence of changes in the dispersion of costs from the frontier. 7.3.1 The Effect of Deregulation on Deposit Costs and Revenues The deregulation of bank interest rates and the development of interestearning consumer checking accounts in the early 1980s had a direct effect of transferring moneys from banks to consumers by increasing deposit interest rates.29In effect, banks had some legally enforced monopsony power over retail depositors, primarily on their checking accounts, which was eliminated. The direct effect on interest expenses of banks is independent of bank efficiency, because insured banks in the same market must pay approximately the same (service-adjusted) deposit rates whether they are efficient or not. Table 7.7 shows how deposit rate deregulation has decreased bank monopsony power over depositors. The first row of the table shows total (implicit and explicit) deposit revenues for 1980, 1984, and 1988, calculated as in equation 1 above. The implicit revenues on deposits declined over time as deposit rates moved closer to market rates, but explicit revenues were not raised sufficiently to cover the implicit revenue reductions. Although deposit revenues were cut almost in half in real terms from 1980 to 1988, deposit operating costs increased, virtually eliminating deposits as a profit center (as estimated profits fell from $61.2 billion to $3.9 billion). Thus, deregulation caused banks to pay out substantially more interest, and this was not offset either by increases in explicit fees or by reductions in operating costs. 7.3.2 Shifts in the Thick Frontier and Their Decomposition The shifts in the thick frontier over time are computed and decomposed in much the same way as the difference between the thick frontier and the highest-cost quartile were computed and decomposed in section 7.2. The 29. The deregulation of deposit rates was largely sparked by two events: (1) the unexpected inflation of the late 1970s and early 1980s, and (2) the growth of competition for bank deposits from money market mutual funds, which have no interest rate restrictions.

268 Table 7.7

Allen N. Berger and David B. Humphrey The Decline in Bank Deposits as a Profit Center (billions of 1988 dollars)

Implicit & explicit revenues Allocated operating costs Net contribution to profits

1980

1984

1988

98.9 31.7 61.2

65.9 42.5 23.4

51.3 41.4 3.9

Nore: All figures in 1988 dollars. See Berger and Humphrey (1990, appendix tables A 1 A, A 1 B, AlC) for more details.

unadjusted shift in the frontier from year t to t + 4 (analogous to the difference in average costs between quartiles DzT) is given by (10)

UShift

=

es

[A&+' - A e ] / A O ,

where A& = &(X.)/TAs, is the predicted cost function using Q l data for year s, X . is the mean argument vector for Q l for years, and T k is the mean total assets for Ql for years (Ql scripts are suppressed for convenience in this section). Next, the frontier shift is corrected for exogenous differences in the data that may be owing to changes in market factors (i.e., input prices, output levels and mix, branch offices, and purchased funds levels). Holding these factors in X constant at their time t-values gives the adjusted frontier shift: (1 1)

Ash$

=

[ A t r * - A&] / A t ' ,

where A t ' * = &+' (X')ITA'. A further adjustment must be made for an additional market factor not explicitly included in the cross-section analyses: aggregate interest rates. The large swings in aggregate rates during the 1980s undoubtedly had large effects on bank interest costs. The cost function parameters in in equation (1 1) implicitly include the aggregate interest rate prevailing at time t + 4 and apply it to data vectorXr. To subtract out this effect, it is assumed that in the absence of technical change and regulation, purchased funds interest rates would move in lockstep with the 90-day Treasury Bill rate, .,i This corresponds with the national nature of the purchased funds (PF) market and the fact that ,i is an appropriate opportunity cost of funds for most participants. For time and savings deposits (TS), rates normally do not move as freely with market rates, because they have a service component and appear to provide depositors with some implicit insurance against swings in market rates.30It is assumed that in the absence of technical change and deregulation, the annual average TS interest rate (i,) would move proportionately with.,i Making these two adjustments to Ash$ gives the final shift in the frontier to be decomposed: (12)

FShift = Ash$ -

+ ((i;;4/i;B)

{[(ik;' ikS

- ikB) -

i;J

*

. (PF'/TA') (TS'/TA')]IAC'}.

30. See Hannan and Berger (1990) for an examination of the rigidity of deposit interest rates.

269

Measurement and Efficiency in Banking

The shifts in the thick frontier over 1980-84 and over 1984-88 are shown by bank size class in table 7.8. Columns (1) and (2) show a striking contrastthe unadjusted frontier shifts (UShif) indicate increases in costs for all but the very largest size class of efficient banks in the 1980-84 interval, followed by Table 7.8

Shifts in the Thick Cost Frontier: 1980-1984 and 1984-1988 (total percentage change over 4 years, in 1988 dollars) ~

Unadjusted Frontier Shift CrShift (%) Asset Size Class (millions of dollars)

0-10 10-25 25-50 50-75 75-100 100-200 2W300 300-500 500- 1000 100~2000 2000-5000 5000-10,000 Over 10,000 Overall mean* Mean < 10.000*

0-10 10-25 25-50 50-75 75-100 100-200 200-300 300-500 500-1000 1000-2000 2000-5000 5 W 10,000 Over 10,000

Overall mean* Mean < 1O,OOO*

1980-84 (1)

1984-88 (2)

Shift Adjusted for Market Factors and Interest Rates FSh$t (%)

1980-84 (3)

Ejicient Banks in Branch Banking States 32.0 - 14.6 33.5 25.2 - 18.0 30.2 - 18.7 26.7 22.3 19.8 - 18.5 22.9 - 18.4 21.1 18.9 17.4 - 18.0 18.4 - 19.6 15.8 13.5 15.5 - 16.9 15.6 11.5 - 16.2 11.3 9.7 - 19.1 11.1 5.1 - 19.3 5.6 4.7 - 18.5 2.4 - 3.8 - 22.6 3.3

7.3 12.2

- 19.6 - 18.4

ESficient Banks in Unir Banking States 33.9 - 15.0 - 20.0 26.1 - 20.0 22.5 18.0 -21.4 18.5 -21.5 14.0 - 22.2 12.7 - 22. I 11.9 - 21.6 11.9 - 28.1 - 20.9 3.5 - 2.8 - 19.6 $ 16.7 - 12.4 $

12.4 15.2

-21.4 $

1984-88 (4) 6.3 4.7 3.1 2.1 2.4 0.7 1

.o

0.4 - 0.9 - 3.7 -1.1 4.6 0.6

10.0 13.0

0.7 0.8

32.6 26.8 22.6 19.7 18.9 15.1

4.1 3.7 2.1 0.3 - 0.6 - 1.7 - 3.2 - 4.2 -7.1 - 9.5 - 5.9

11.1

12.2 9.6 -5.6 0.3 9.5 -5.1 12.6 14.6

$ $ -

1.7 $

*The mean values reported are asset share weighted sums of the size classes indicated. $There were too few large unit banks to form quartiles in 1988 for the top 2 size classes. Thus the mean values reflect the first 11 size classes.

270

Allen N. Berger and David B. Humphrey

decreases in costs for all sizes from 1984 to 1988. Over 1980-84, small banks experienced more severe cost increases than large banks. This is largely because small banks tend to fund more with produced deposits, for which the average rate paid increased from 4.7 percent to 7.0 percent (because of deregulation); large banks more often use purchased funds, for which the average rate paid decreased from 1 1.6 percent to 8.7 percent. This suggests that deregulation hurt smaller banks more than large banks over this time period. In contrast, the unadjusted frontier shift over 1984-88 affected banks more equally because interest rates on produced deposits and purchased funds both fell over this time period. Columns ( 3 ) and (4) of table 7.8 show the same frontier shifts after adjusting for market factors and aggregate interest rates (FShijft).The net effect of these two adjustments is to offset one another for 1980-84, so FShijr and USh@ are about the same. In contrast, for 1984-88, the removal of local market factors was swamped by the continuing fall in market rates, so that the fall in costs is essentially eliminated once all adjustments are made. The shifts in the adjusted frontier (FShift) can be further decomposed by cost type into operating costs, interest on produced deposits, and interest on purchased funds, each of which has a different economic interpretation. The change in operating costs over time among the efficient firms is interpreted as representing net technical change, which includes pure technical change, any changes in the level of service produced per dollar of deposits and loans as a result of deregulation, and any disequilibrium effects on the operating costs of the best practice banks. The former two are expected to reduce costs because technical changes generally reduce costs and because banks would be expected to provide less service per dollar of deposits after deregulation (e.g., by increasing the level of real deposits per branch office) in an optimal tradeoff with the higher deposit interest rates. The disequilibrium effects are expected to raise costs as banks incur short-term costs in adjusting to their new long-term equilibrium by shutting branch offices, and so on. The change in interest on deposits net of aggregate interest rate changes is interpreted as the direct effect of the deregulation of deposit interest rates. As mentioned above, these rates went up over 1980-84, even while market rates were decreasing. The change in interest on purchased funds net of aggregate interest rate changes is a residual reflecting maturity, liquidity, or credit risk changes in purchased funds rates not captured by changes in the Treasury Bill rate.3' These separate effects are shown in table 7.9 for all sizes of banks. The percentage shift in each type of cost is weighted by its share of total costs, so that the figures sum to FSh$ in table 7.8. Column (1) shows that operating costs per dollar of assets increased as deregulation began in the 1980-84 interval, particularly for banks in the smallest size classes. Column (2) indicates a slightly better performance in the 1984-88 interval, with the smaller banks 31. This does not rule out the possibility that technology can reduce interest costs by improving the monitoring of market conditions, etc. However, such effects are swamped by aggregate interest rate and deregulation effects on deposits and purchased funds.

271

Measurement and Efficiency in Banking

Table 7.9

Net Technical Change and Deregulation Effects: 1980-1984 and 1984-1988 (total percentage change over 4 years, in 1988 dollars) Operating Cost Net Technical Change (%)

Asset Size Class (millions of dollars)

1980-84

0-10

8.0 7.2 5.2 3.7 2.7 1.9 1.3 0.6 0.2 0.3 -0.2 2.8 3.6

10-25 25-50 5c75 75-100 1w200 200-300 300-500 500-1000 1000-2000 2000-5000 5000- 10,000 Over 10,000 Overall mean Mean < 10.000 c10 I0-25 25-50 50-75 75- 100 100-200 200-300 300-500 5W1000 1000-2000 2000-5000 5000-10,OOO Over 10,OOO Overall mean Mean < 10,000

(%)

1984-88

1.2 0.1 1.1 1.1 1.5 I .4 2.3 -0.2 -0.7 -1.1 -9.1 -3.1

1 .o

0.2 0.5 0.4 0.4 1.2 0.4 1.2 2.0 1.7 1.3 4.2 7.8 3.7

-0.5 - 0.2

- 1.6 -1.0

3.1 2.9

EfJicient Banks in Unit Banking States - 3.4 20.6 7.4 - 2.4 5.8 18.1 - 2.3 4.0 15.5 2.3 13.8 -2.1 12.5 - 1.9 1.1 - 2.0 10.9 0.2 7.9 -2.1 - 1.4 - 1.8 7.2 -3.9 - 1.8 6.6 -4.2 0.1 -11.3 0.2 0.0 -0.9 - 12.2 -0.6 * * -1.1 * - 1.6 *

1.5 0.9 1.7 2.1 3.1 2.0 2.8 4.9 3.2 - 3.2 0.6 10.9 -2.3

0.1 0.4 0.4 0.2 0.3 0.1 0.3 1.5 -1.1 1.6 7.1

2.3 1.7

1980-84 (3)

1984-88 (4)

Interest on Purchased Funds Residual Effect (%) 1980-84 (5)

(1)

1984-88 (2)

Interest on Deposits Deregulation Effect

EDcienr Banks in Branching States 2.6 24.2 3.5 2.9 22.9 1.2 2.0 20.5 0.7 1.4 18.1 0.3 0.8 16.9 0.4 0.1 15.1 0.1 - 0.3 13.4 0.1 - 1.2 12.8 -0.3 - 2.2 11.4 -0.4 -4.2 10.4 -0.9 6.9 -1.0 -4.3 0.0 -3.3 8.7 - 1.9 2.8 - 1.3

-1.9 -1.9

9.4 12.3

10.5 7.8 5.5 3.7 3.3 2.3 0.4 0.1 -0.3 - 2.6 -0.2 - 0.9 2.0 2.4

-0.9

*

8.5 9.6

-

1.9

*

2.1 2.6

(6)

*

* 1.o *

Note: Columns (l), (3), and (5) sum to column (3) in table 7.8 and columns (2), (4), and (6) sum column (4) in table 7.8 by construction. *see t in table 7.8.

showing a lesser cost increase; the larger banks had cost decreases that offset the increases in the earlier interval. Under normal circumstances (i.e., equilibrium growth), these results, particularly those for smaller banks, would be quite unusual, because technical progress is rarely negative for such a long interval of time. However, the 1980s were not normal circumstances and the

272

Allen N. Berger and David B. Humphrey

results may be interpreted as representing substantial disequilibrium brought about by deregulation. As mentioned above, deregulation altered the balance previously obtained where capital (mainly extra branches) and labor partially compensated depositors for artificially low rates paid on deposits. When rates were deregulated, higher deposit rates were established faster than capital and labor were reduced. When the new equilibrium is reached, average costs are expected to be lower because of pure technical innovation and a lower level of service per dollar of deposits as banks eliminate the extra service that was substituted for legally prohibited interest payments.32 The evidence cited above on the large increase in real deposits per branch office in the 1984-88 interval suggests that banks were still overbranched and overstaffed and were bearing the transition costs of closing some branches. In addition, to the extent that banks had excess capacity as a result of the temporary disequilibrium, the increase in average operating costs may be overstated as a measure of negative technical change (see Berndt and Fuss 1986).” Columns (3) and (4) of table 7.9 show that deregulation increased deposit interest costs from 1980-84, when most of the initial rate increases occurred, but had little effect from 1984-88. As mentioned above, the interest cost increases from deregulation are much greater for smaller banks, who tend to secure a higher proportion of their funds from produced deposits. Columns (5) and (6) show the residual purchased funds effects, which are small as expected, except for the largest banks which use these funds intensively. Table 7.10 expresses these effects in annual rates of change for all sizes of Q l banks together and allows for examination of the net effects over the entire eight-year interval. The first two columns simply reexpress the effects in table 7.9 in annual terms; the final column represents a new computation of the frontier shift between 1980 and 1988 using 1980 as a base year. These results suggest that slight technical progress over the last four years offsets the disequilibrium cost increases of the first four years. For banks in branching states, there is a small net annualized reduction in operating costs over the entire eight-year period. For banks in unit banking states, where there was less latitude for cost savings through branch office closings, there was no net measured progress. When this information is combined with the findings given earlier that (1) measured inefficiencies were still relatively high in 1988; (2) smaller banks still had higher operating costs in 1988 than in 1984; and (3) the growth of banking offices was still decelerating in 1988, it suggests that the banking industry had still not reached its new post-deregulation equilibrium by 1988. The effect of deregulation on produced deposit interest costs is 32. An exception to this would be a technical innovation that both increased costs and increased the quality of service. The advent of the automated teller machine (ATM) may fit this category. Undoubtedly, ATMs increased the quality of bank service. There is also some evidence (see Berger 1985) that ATMs may have raised bank costs, especially during the 1980-84 interval when many machines were not at mature volume. 33. Formally, the parameters of the operating cost portion of the cost function in ( I 1) estimated for 1984 may overstate costs owing to excess capacity at that time, yielding an increased operating cost component of FShifr.

273

Measurement and Efficiency in Banking

Table 7.10

Cost Effects of Technical Change and Deregulation: Annual Growth Rates (at the mean for the low-cost banks) (%)

Branch banking states: Net technical change (oc) Deregulation (ID) Residual (IPF) Unit banking states: Net technical change (OC) Deregulation (ID) Residual (IPF) BLS bank labor productivity

1980-84

1984-88

1980-88

0.6 2.3 -0.4

-0.5 -0.1 0.8

-0.5 0.6 0.2

0.5 2.1 0.5 3.0

-0.2* -0.5* 0.2* 3.8%

0 .I * 0.7* 0.4* 3.3$

Source: Computed from table 7.9 plus a similar run comparing 1980 and 1988 data. *See $ in table 7.8. SBLS index available only through 1987.

on net an increase, but it virtually all occurred in the first time interval, and its cumulative effect is dying out. Technical change and deregulation seem to have had little or no effect on purchased funds costs, as expected. Examining the final 1980-88 column, it appears that on net, adjusted real banking costs have increased slightly for banks in branching states, as operating cost improvements have not quite offset additional interest payments. For unit states, the lesser possibilities for branch office closings to save on operating costs have left them with a greater increase in real costs over the 1980-88 time interval.

7.3.3 Comparison with the BLS Measure of Bank Labor Productivity It is instructive to compare our results with the BLS measure of bank labor productivity shown in the bottom row of table 7.10. The BLS measure relates bank output as a physical flow (numbers of transactions processed and new loans made) to a single input (labor). It would be expected that the BLS measure would be approximately equal and of opposite sign to our net technical change figures. What is observed, however, is quite different. The BLS finds productivity growth of 3 percent or better per annum for both the 1980-84 and 1984-88 time intervals; we observe a slight increase followed by a slight decrease in costs per unit of output. One important reason for these seemingly incongruent results is that the BLS measure is based only on labor input; our measure is based on real operating costs, which implicitly include all physical inputs and adjustments in their proportion^.^^ When markets are in equilibrium, the change in labor in34. Another potential reason for these differing results is that the BLS measure is inclusive of all banks, not just those on the thick cost frontier as used here. Although it is possible that the hanks in the three highest-cost quartiles reduced operating costs by more than enough to offset the increased operating costs of the lowest-cost quartile, the data suggest that, if anything, the highercost banks fared even worse than the low-cost banks over the 1980s. Thus, this possibility can be discounted as an important explanation of the differing results.

274

Allen N. Berger and David B. Humphrey

put is a good proxy for the proportional change in all inputs or total operating costs. During the disequilibrium of the 1980s, however, such relationships may not hold. Costs likely increased relative to employment because the expenses of liquidating branches and of installing automated teller machines have often been paid to nonbank capital and labor sources, rather than to employees of the bank (the measured labor input). In addition, to the extent that employment grew more slowly or decreased as a result of branch closings and other pressures from deregulation, any reductions have likely been mostly confined to low-cost employees serving in branch offices, which would decrease measured employment more than proportionate with true valueweighted labor or total operating costs.35 An analysis of some raw data tends to confirm these hypotheses. Over the 1980-84 interval, the ratio of bank employment to real operating costs for all banks together fell at a 3.3 percent annual rate. This alone is more than enough to explain the 3.0 percent growth in the BLS productivity index when in fact, productivity in terms of all factors may have fallen. It also can account for nearly the entire difference between the BLS 3.0 percent growth and our 0.5 percent to 0.6 percent increase in operating costs per unit of output. Over the 1984-88 interval, the employment-cost ratio fell at a 2.1 percent annual rate, large enough to explain most of the 3.8 percent measured BLS productivity gains and most of the deviation from our measures. As expected, the fall in labor cost share, 1.0 percent per annum for both time intervals, was smaller than the fall in the employment-cost ratio, consistent with the hypothesis that the proportion of lower-paid employees has decreased over time. This evidence suggests that the BLS measure may be misleading during periods of significant disequilibrium.

7.4 Conclusions Commercial banking is one of the most difficult service industries in which to measure output, technical change, or productivity growth. The problem of choosing which banking functions constitute the important outputs is difficult because many banking revenues are implicit and commingled, so that the flow of explicit revenues is an unreliable guide to the flow of banking services. However, the value-added approach, in which the flows of physical labor and capital inputs are matched to banking functions, identifies the important bank outputs as being the major deposit and loan categories. It is also difficult to measure technical change and productivity growth because of the confounding effects of changes in inefficiency over time and the deregulation of the deposit side of banking. If inefficiency is not taken into 35. Our use of operating costs implicitly values labor (and capital) at their appropriate marginal value product weights to the extent that different prices paid to different workers accurately reflect their productive values.

275

Measurement and Efficiency in Banking

account, then measures of technical change or productivity growth may confuse shifts in the minimum-cost technology with changes in the deviations from that technology. In addition, higher deposit interest rates were quickly adopted as a result of deregulation, but the offsetting reductions in depositor services (such as reducing branching convenience) have been relatively slow. These different factors are accounted for here by estimating multiple-equation thick-frontier cost functions for each of three years, 1980, 1984, and 1988, which roughly correspond to pre-, mid-, and postderegulation periods. Cost dispersion and inefficiency are analyzed for each of these years and the shifts between years are decomposed into operating and financial cost categories. The major findings are as follows: Most of the dispersion in bank costs appears to represent inefficiencies, rather than market factors, such as differences in input prices, scale of operations, or product mix. Except for the very largest banks, the inefficiencies are mainly operational in nature, involving overuse of physical labor and capital inputs, rather than financial, involving excessive interest costs. As well, the set of low-cost banks is seen to remain quite stable over time and to have the highest profits and lowest probabilities of failure during the 1980s, indicating that cost differences are not simply owing to luck and that they are important to bank performance. In addition, operating cost dispersion and inefficiency rose substantially over the period, particularly from 1984 to 1988, suggesting a less than complete adjustment to the new, less regulated equilibrium. The shift over time in the thick-frontier cost function, after adjustment for changes in market factors and aggregate interest rates, shows important changes in both operating and interest costs resulting from deregulation. First, operating costs for the low-cost banks rose over the 1980-84 interval and then fell over 1984-88. However, the process was uneven, with larger banks able to close and restructure branch operations and otherwise reduce costs; smaller banks continued to have increasing operating costs over 1984-88. Had the progress to the new post-deregulation equilibrium been substantially complete by 1988, one would have expected both technical progress and a shift toward supplying fewer services per dollar of deposits to have resulted in considerable net technical change. However, the overall change is quite small and uneven. Combining these findings with the increase in cost dispersion and the increased real deposits per branch office suggests that progress toward the postderegulation equilibrium remained incomplete by 1988, especially for smaller and less efficient banks. Second, deregulation removed a substantial source of monopsony power over depositors for banks, raising interest costs significantly and virtually eliminating deposits as an independent profit center. Even by 1988, several years after deregulation, this increase in deposit interest costs generally was not offset by decreases in operating costs, except for relatively large and relatively efficient banks. Given the strong empirical association between high

276

Allen N. Berger and David B. Humphrey

costs and bank failures, it is likely that this loss of monopsony power contributed to the dramatic increase in bank failures in the 1980s. Finally, our results contrast sharply with those of the BLS labor productivity index for banking, which shows productivity rising at a 3 percent or more per annum through the 1980s. The major reason for this difference appears to be the use of bank employment as the single factor by BLS. Overall, deregulation appears to have resulted in little, if any, net technical change or productivity growth in banking in the 1980s. However, offsetting this lack of progress are the benefits of deregulation to consumers, which are not reflected in measured bank output. Consumers obtained a higher return on deposits without a fully offsetting reduction in branch office convenience or higher service fees. Thus, part of the cost increases from deregulation could alternatively be interpreted as increases in output quality, suggesting that the true combined effect of technical progress and deregulation is more favorable than that measured here. In any event, we have identified why measured technical change has been so slow in the 1980s-the reason is banking deregulation and the less-than-cost-minimizing response to it by the banking industry.

277

Measurement and Efficiency in Banking

Appendix Table 7A.1

Summary of Data (all banks, 1988) Branch Banking States (10,961 banks) Mean

Cost variables: OC Operating costs (% of assets)* SOC, Labor share of operating costs (%) ID Interest on produced deposits (% of assets)* IPF Interest on purchased funds (% of assets)* TC Total operating plus interest costs (W of assets)* Output variables: DD Demand deposits (% of assets)* TS Retail time & savings deposits (% of assets)* RE Real estate loans (% of assets)* Commercial & industrial loans (% of CI assets)* Installment loans (% of assets)* IN Other variables: B Number of banking offices OA Other (nonloan) assets (% of assets)* Total assets $OOO,OOO,1988 dollars (not TA used in regressions) Price of labor, $OOO per year, 1988 dollars w1 Price of physical capital, 1988 dollars w2 (assumed to be proportionate to the replacement cost of office space in the region, taken from F. W. Dodge)

Standard Deviation

Unit Banking States (1,844 banks) Mean

Standard Deviation

3.49 48.93 4.03

1.75 8.54 0.87

3.58 47.81 4.01

1.51 9.29 0.81

0.89 8.41

0.86 1.76

0.68 8.27

0.69 1.44

13.57 63.88

6.68 13.00

14.51 65.03

11.61

24.15 17.58

12.82 10.38

21.68 17.79

12.57 10.60

11.28

9.36

10.10

7.13

5.05 46.98 223.04

20.33 15.12 20.61

1.59 50.43 104.39

1.06 14.09 23.80

25.65 81.87

7.21 9.98

25.48 72.80

5.24 1.93

6.73

Source: Reports of condition and income (call reports), except as noted. The flow figures are the annual totals from the December 1988 call; the stock figures are averages from the December 1987, June 1988, and December 1988 calls (to avoid biases from growth or decline over the year). *Numbers are expressed relative to assets for exposition only. Regressions are based on raw data in $OOO.

References Aly, Hassan Y., Richard Grabowski, Carl Pasurka, and Nanda Rangan. 1990. Technical, Scale, and Allocative Efficiencies in U.S. Banking: An Empirical Investigation. Review of Economics and Statistics 12 (May): 21 1-19. Bamett, William A. 1980. Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory. Journal of Econometrics 14 (September): 1 1-48.

278

Allen N. Berger and David B. Humphrey

Berger, Allen N. 1985. The Economics of Electronic Funds Transfers. Outline. Board of Governors of the Federal Reserve System, Washington, D.C., October. Berger, Allen N . , and Timothy H. Hannan. 1989. The Price-Concentration Relationship in Banking. Review of Economics and Statistics 71 (May): 291-99. Berger, Allen N., Gerald A. Hanweck, and David B. Humphrey. 1987. Competitive Viability in Banking: Scale, Scope, and Product Mix Economies. Journal of Monetary Economics 20 (December): 501-20. Berger, Allen N., and David B. Humphrey. 1990. Measurement and Efficiency Issues in Commercial Banking. Finance and Economic Discussion Series, Working paper no. 151. Board of Governors of the Federal Reserve System, Washington, D.C., December. . 1991. The Dominance of Inefficiencies over Scale and Product Mix Economies in Banking. Journal of Monetary Economics 28 (August): 117-48. Berndt, Ernst, and Melvyn A. Fuss. 1986. Productivity Measurement with Adjustments for Variations in Capacity Utilization and Other Forms of Temporary Equilibrium. Journal of Econometrics 33 (OctobedNovember): 7-29. Board of Governors of the Federal Reserve System. Various years. Reports of Condition and Income (call reports), and Functional Cost Analysis. National Average Report for Commercial Banks, Washington, D.C. Booz-Allen and Hamilton. 1987. Managing Delivery System Economics. Bank Branch Profitability Study for the American Bankers Association, October. Bureau of Labor Statistics. 1989. Productivity Measures for Selected Industries and Government Services. U.S. Department of Labor, Bulletin no. 2322, Washington, D.C., February. Donovan, Dona1 J. 1978. Modeling the Demand for Liquid Assets: An Application to Canada. International Monetary Fund Staff Papers 25 (December): 676-704. F. W. Dodge Division. 1980-88. Dodge Construction Potentials Bulletin. Summary of Construction Contracts for New Addition and Major Alteration Projects. New York: McGraw-Hill. Ferrier, Gary D., and C. A. Knox Lovell. 1990. Measuring Cost Efficiency in Banking: Econometric and Linear Programming Evidence. Journal of Econometrics 46 (October/November): 229-45. Fixler, Dennis J., and Kimberly D. Zieschang. 1990. Output and Price Measurement in Commercial Banking. Unpublished manuscript. Bureau of Labor Statistics, Washington, D.C., February 7. Gordon, Robert J. 1965. Airline Costs and Managerial Efficiency. In Transportation Economics, 61-94. New York: Columbia Univ. Press. Greene, William H. 1990. A Gamma Distributed Stochastic Frontier Model. Journal of Econometrics 46 (October/November): 141-63. Hancock, Diana. 1985a. Bank Profitability, Interest Rates, and Monetary Policy. Journal of Money, Credit, and Banking 14 (May): 179-92. . 1985b. The Financial Firm: Production with Monetary and Nonmonetary Goods. Journal of Political Economy 93 (October): 859-80. Hannan, Timothy H., and Allen N. Berger. 1991. The Rigidity of Prices: Evidence from the Banking Industry. American Economic Review 81 (September): 938-45. Humphrey, David B. 1992. Cost and Technical Change: Effects of Bank Deregulation. Journal of Productiviw Analysis. Forthcoming. Hunter, William C., and Stephen G. Timme. 1986. Technical Change, Organizational Form, and the Structure of Bank Production. Journal of Money, Credit, and Banking 18 (May): 152-66. Klein, Michael A. 1971. A Theory of the Banking Firm. Journal of Money, Credit, and Banking 3 (May): 261-75. Kopp, Raymond J., and W. Erwin Diewert. 1982. The Decomposition of Frontier Cost

279

Measurement and Efficiency in Banking

Function Deviations into Measures of Technical and Allocative Inefficiency. Journal ofEconometrics 19 (August): 319-31. Kwast, Myron L., and John T. Rose. 1983. Profitability Differences among Large Commercial Banks During the 1970s. Magazine of Bank Administration 59 (Sep-

tember): 54-62. Mamalakis, Markos J. 1987. The Treatment of Interest and Financial Intermediaries in the National Account: The Old “Bundle” versus the New “Unbundle” Approach. Review of Income and Wealth 33 (June): 169-92. Sealey, Calvin, and James Lindley. 1977. Inputs, Outputs, and a Theory of Production and Cost at Depository Financial Institutions. Journal of Finance 32 (September): 1251-66. Stevenson, Rodney E. 1980. Likelihood Functions for Generalized StochasticFrontier Estimation. Journal of Econometrics 13 (May): 58-66. Zieschang, Kimberly D. 1983. A Note on the Decomposition of Cost Inefficiency into Technical and Allocative Components. Journal of Econometrics 23 (December): 401-5.

Comment

Frank C. Wykoff

Allen N. Berger and David B. Humphrey (BH) report two empirical regularities in commercial banking activity during the 1980s: (1) the variance in average costs among banks in the United States was large and persistent-average costs of the highest-cost quartile of banks exceeded average costs of the lowest-cost quartile by 30 percent to 50 percent; and (2) these large unit cost differences are not related to bank size, branching, or other observed causal variables but seem to be associated with profitability and failure rates. BH measure and interpret these unit cost differences using two data bases( 1 ) functional cost analysis (FCA), consisting of a large but varying nonrandom sample of banks who voluntarily report to the FED on the allocation of costs to different activities, such as deposits and loans, and (2) call reports, in which virtually all American banks, as required by law, report to bank regulators book values of capital, costs of funds, rates of return, and other financial statistics. The FCA data attributes value added to sources-two categories of deposits, three categories of assets, and other sources. BH find that 48 percent of value added comes from deposit accounts, 30 percent from loans and 22 percent from other activities. Largely on the basis of this evidence, BH define output to consist of the two deposit categories plus the three loan categories. BH then use call report data to estimate translog cost functions for bank output, with suitable normalizations, from input prices, costs, and levels of outputs for the quartile of banks with the lowest average costs-BH call this quartile, the most efficient banks. Frank C. Wykoff is Eldon Smith Professor of Economics at Pomona College and the Clarernont Graduate School and editor in chief of Economic Inquiry.

Allen N. Berger and David B. Humphrey

280

Figure 7C. 1 illustrates their econometric methodology for estimating the cost function. By estimating unit costs as a function of variations in input prices and levels of outputs, they trace out the unit isoquant, q = 1, of the average efficient firm by rotating isocost curves like 11. If, given input prices, the average efficient bank were a cost minimizer, then it would produce at a point, such as points a and e, on the isoquant tangent to the isocost. Thus, both the mix of inputs and the level of costs would be optimal. BH compare the costs incurred by the most costly quartile, the least efficient banks, to the estimated unit isoquant of the efficient banks. The “least efficient” quartile of banks are operating beyond the frontier isoquant, q = 1, producing at points like b and c. The distances from b to a and from c to e, along rays from the origin, constitute measures of inefficiency. BH also trace growth of productivity of the average efficient banks over the 1980s and decompose their measure between technical change and efficiency causes. Thus, although much of the paper focuses on efficiency issues, the authors also study productivity growth. How important are the BH results, how can they be explained, and what do their results have to do with output measurement and productivity growth per se? In my judgment, this paper presents very important empirical evidence

K

I

Capital

Too

Much Capital

Frontier lsoquant

q=l bc {a Too Much Labor Fig. 7C.1 Cost differences among banks

’

L Labor

281

Measurement and Efficiency in Banking

based on comprehensive data on the nation’s banking system. The key result, that many banks have been operating in the 1980s at the margin-with excessive costs and low profits, does not auger well for the banking industry should the economy slip into recession. Many major banks could face serious financial stresses, and some could fail. On the brighter side, this same result implies plenty of room for improvement. Many banks could trim fat and tighten their belts. Such a tightening would produce a one-shot jump in productivity growth for the banking system as a whole. Perhaps the BH paper will spur banks and their regulators to move forward with this belt tightening before a recession forces them to do so in a crisis context. To correct a flawed system, though, it is helpful to know the root causes of the flaws. In this context, we may wonder why do average cost differences, unrelated to observed explanatory variables, persist in the banking system for so long? I have been around Chicago economists long enough to see flares when told of persistent inefficiencies, and so I must ask whether these inefficiencies represent unexploited rent-seeking opportunities? What do BH mean by the words “efficient” and “inefficient”? BH define efficient firms as those with relatively low average costs and inefficient firms as those with relatively high average costs. This is not necessarily X inefficiency nor inefficiency in the broad sense of an economy operating inside its production possibilities frontier. BH’s finding of persistent inefficiencies is a puzzle calling for some explanation. I suggest four possible explanations of their results: 1. Rising marginal costs. BH’s observation, that unit costs vary among banks, may not be very interesting per se. Supply curves slope upward precisely because each increment in output pulls in marginal resources, including labor and management, that are less well suited to the purpose than previous increments. Persistent cost differences may simply reflect the reality that different qualities of resources are needed to satisfy the entire market-the industry is operating on the upward sloping portion of the marginal cost curve. (This results in producer surpluses for superior firms.) Were inefficiencies to exist, in the sense that resources outside banking could earn more by leaving their present pursuits and produce banking services, then these resources would, unless restrained by regulation, move in and capture the unexploited rents. But BH do not show that inefficiencies exist in this sense. That is, they do not show that unexploited rent-seeking opportunities exist in banking. They show only that unit costs, and profitability, differ among banks. 2. Regulatory barriers to entry. Regulation can protect inefficient firms by creating barriers to entry. An Averch-Johnson regulator, for instance, would cause firms to overcapitalize. Like airlines, who before deregulation flew too 1. See Averch and Johnson (1962) and Diewert (1981) for an econometric model justifying the use of BH-type methods for estimating a variable cost function under Averch-Johnson-typeregulation.

282

Allen N. Berger and David B. Humphrey

many planes on too many routes, BH banks might have too much capital so that their input mix is suboptimal. This would place them at a point like f in figure 7C. 1-employing too much capital. BH empirically test for and reject the possibility of the wrong input mix, but they might be wrong about this. Bank regulation, however, is probably not Averch-Johnson, taking instead some form of branch and geographical restriction. This could result in many banks operating at high average costs. If this were true, however, then BH have an econometric problem. The assumption of cost minimization needed for their econometric technique to trace out the unit isoquant would be false. Furthermore, BH do not model regulation in order to explain the average cost variance nor to justify estimation of a cost function. As far as I know, the only econometric model developed for regulated firms that justifies BH methods for estimating a variable average cost curve is by Diewert (1981). 3. Capitalization of land values. Consider two banks that are the same size, have the same output mix, use the same technology and input mix, yet bank A’s costs, except interest costs, are all larger than bank B’s. Why? A is a New York bank, and B is a South Dakota bank. Higher land values in N.Y. have become capitalized into all input prices. Tellers cost more; paper products cost more; deliveries cost more. Even lunch costs more. Only money, traded in a world market, costs the same. Banks A and B are both efficient but average costs at bank A, point b in figure 7C. 1, exceeds average costs at bank B, point a in figure 7C.1, despite the same input mix. As far as I can tell, BH’s data are consistent with this story. Because BH do control in their regressions for both rental price and labor cost variations among banks, they may already have largely captured this effect. Furthermore, they also found substantial cost and profit differences among banks in the same large cities. 4 . Different product mix. Is it possible that bank customers and the services they demand differ across banks? Is it further possible that these service differences cannot be detected in data on financial instruments? Perhaps. Deposits are only representations of the actual underlying flow of services provided by banks to deposit customers. Consider for example, a $2,000 checking account of two different customers in two different banks. The bank A customer may require more labor and capital services than the bank B customer, and this higher level of services may impose higher unit costs on bank A than on B. The bank A niche may be its appeal to a clientele who differ from bank B clientele. The flow of services that accrue to deposit customers are varied, complex, and subtle-visits to the teller, withdrawals, time saved in a complex variety of check and credit card transactions, access to one’s funds at various locations and various periods during the week. Larger or smaller variances of holdings may differ across customers. All those services associated with deposit accounts accrue to different customers, who with apparently identical deposit accounts impose different cost of services on different banks. Security costs, for example, may

283

Measurement and Efficiency in Banking

well differ significantly within a given metropolitan region. This would imply cost differences but not necessarily differences in deposit accounts. Loans, similarly, are only representations of the actual underlying flow of services provided by banks. How much analysis must go into assessing oil exploration loans in Texas as opposed to housing loans in Maine or shipping loans in Long Beach? Providing credit to Latin American nations may have different costs than a line of credit to IBM. Pooling risks may be easier in diverse California than in homogeneous Nebraska. Some of these service differences are very difficult, if not impossible, to detect from data on the various pieces of paper produced by banks-financial instruments. BH have only data on financial instruments, assets, and liabilities, not data on the ultimate services that flow to customers from these instruments. This fourth potential explanation brings me to the core of my comment on the state of knowledge about banking output and productivity growth measurement. Section 7.1 of the BH paper contains a discussion of the appropriate treatment of bank liabilities. Should deposit accounts be treated as output, input, or what? This question is not resolved by BH nor by anyone in the literature. Fixler and Zieschang (1991, and chap. 6, this vol.), for instance, have decided to treat deposits as inputs when net financial flows accrue to depositors and outputs when net financial flows accrue to banks. As noted BH treat deposits as outputs whereas others have treated them as inputs. Some components of liabilities are viewed by BH as inputs, such as federal funds purchased and large CDs. Interest paid on core deposits are treated as a cost. I believe the BH discussion in which they explain differences in approaches, like many others, skims over the fundamental questions, and, unless one focuses the debate on these fundamental questions, disagreements over how to treat deposit accounts and over the key question concerning the output of banking will remain unresolved. It may turn out that BH are absolutely correct in their empirical choices, but in my view we need better explicit conceptual reasons for our choices. To focus the debate, consider table 7C. 1, which shows five possible assumptions about the roles of deposits in the productivity framework, examples of economists who have assumed each role, and questions that must be answered by those making each assumption. Are deposit accounts (liabilities) only inputs that banks use to produce output on the asset side of the balance sheet? Economics has a long history, traceable to both Karl Marx and Adam Smith, doubting the productivity of banks and bankers, so the view that banks do not provide productive services, especially to depositors, is widespread. See, for example, Fixler and Zieschang’s (1991) critique of the United Nations system of national accounts. Advocates of the view that banks provide no output to depositors must explain then why people open bank accounts, store money in the banking system, write checks, deposit money, withdraw cash, carry bank cards, check guarantee cards, and so forth. This is

284

Allen N. Berger and David B. Humphrey

Table 7Cl. Assumption

What Are Deposits? Five Possible Assumptionswith References and Questions for Each Reference

Inputs

Sealey and Lindley 1977

outputs

Berger and Humphrey (chap. 7)

Both

Arndt 1984; Triplett (comment on chap. 6 and 7); Wykoff 1991

Either

Fixler and Zieschang (chap. 6, this vol.); Hancock 1985; Bamett 1980

Neither

Wykoff (this comment)

Question If deposit customers do not receive outputs from the banks, then why do they spend time and effort to travel to banks to give them these free inputs? Why are these bank outputs so cheap and why have their nominal prices been comparatively stable, thus falling even in real terms, over the years? How do you measure the price of the outputs and inputs for purposes of partitioning prices and quantities of output growth? Advocates must answer questions under both “outputs” and “inputs,” above, because one or the other is assumed for each liability at each moment in time. If deposits are not output, then what is output and what are deposits if they are neither?

a substantial amount of activity to undertake without compensation. Do they do so in order to voluntarily provide input for bankers without receiving compensating value? If so, how do capitalist bankers force customers to provide them with these inputs without charge? Does nothing but trivial yield accrue to the depositor? BH and many others treat deposits as output. This position, too, requires its supporters to answer a difficult question, If deposits are outputs only, then why have the explicit nominal prices of these products been so low and so unchanging? Through the volatile period of price level instability from 1965 to 1981 bank deposit fees and charges were flat. Even throughout deregulation under the Decontrol Act of 1981, fees, charges, and rates on deposit accounts have been remarkably low and inflexible. Even though financial institutions are going through difficult times during the 1980s and early 1990s-the Third World loan crisis, bank failures, the oil and real estate collapse, and the savings and loan scandals-explicit charges to customers have not changed very much. If deposits are major outputs, then banks are giving away their products at very low and very stable nominal prices. Rate changes have not even accompanied large fluctuations in the inflation rate! Even the staunchest defender of capitalists do not suggest that they give away their products without compensation. BH, in table 7.7 provide a potential answer to this question by pointing out that profitability of deposit accounts has declined. This implies that banks have been forced by competitiveness to limit charges on deposit accounts.

285

Measurement and Efficiency in Banking

Arndt (1984), Triplett (comment on chap. 6 and 7), and I elsewhere (1991) have argued that deposits are inputs and outputs simultaneously.* Banks receive valuable inputs, cash, that they use as loans. They are willing to pay for their inputs. Depositors, simultaneously, receive outputs from banks that accrue on the accounts. They would willingly pay for these accounts. However, because both the bank and the depositor receive benefits, the gains largely offset one another and no flow of payments from either party to the other occurs. Thus we have a very odd kind of barter transaction involving money in which one type of money is trade for another. The bank receives an input, money, and pays for it with an output, a deposit account. The customer receives an output, the deposit account, and pays for it by providing an input, money. These barter exchanges are not always of exactly equal value to each party in the trade. When the input value exceeds the output value, then added compensation is demanded by the deposit customer. Thus bank rates on deposits exceed any fees and a tiny net cash flow accrues to depositors. When output value exceeds input value, the net cash flow in the form of fees and charges accrues to banks. Advocates of this approach, too, must resolve a tricky issue-how does one measure the gross price of the trade-in-kind exchange? All we have data on is those tiny net flows of explicit fees and explicit interest charges. Fixler and Zieschang, following Hancock (1985) and Barnett (1980) treat deposits as either inputs or outputs depending on the minuscule net flow of deposit rates and fees. It seems to me that this view is wrong, even nonsensical. They have to answer both sets of questions raised for the input only and output only crowds, When the deposits are outputs, why are they so cheap? When they are inputs, why do people provide them to banks? My position on this debate requires that I explain what the outputs are and, if deposits are not output, then what are they? These are tough questions that begin to bring us even deeper toward the heart of the matter. In my view, deposits are neither outputs nor inputs. Deposits, in my view, are financial instruments associated with a flow of a wide variety of complex and subtle services received by deposit customers. Deposits are also intermediate goods created in the bank production process partly to provide these services and partly to generate other financial instruments that, in turn, generate final product services. To determine exactly what these service flows are gets to the very heart of an issue not mentioned so far in discussions of bank productivity growth2. Triplett (1990) provides a cogent explanation and several examples of other barter transactions that involve trades in kind. 3. Mamalakis (1987) presents a heuristic discussion of services provided by financial institutions that focus on, among other things, the time dimension of loans. This is a valuable point of departure toward identifying the different social services provided by banks.

286

Allen N. Berger and David B. Humphrey

namely, what do banks do? As a service-sector firm, a bank must provide services. I do not claim to have fully resolved exactly what banks do nor exactly what the services are, but it does seem clear that financial instruments, per se, are not services. But what the services are, and how one measures the quantities and prices of these services is not fully resolved in the economics literature i t ~ e l fNonetheless, .~ I would like to suggest the direction we should look for answers, because I do believe we are closer to the answers than suggested by the present treatment of banking in the productivity literature. In recent years, the comparatively new transactions-cost approach to analyzing markets has challenged the standard neoclassical approach that underlies virtually all productivity growth analysis. The central model of neoclassical theory views competitive market demand and supply schedules as sums of separable decisions of producers and consumers who maximize constrained objective functions. The transactions-cost approach, based on ideas of Ronald Coase in the 1930s and 1940s, focuses on differences among and unique features of various markets that cause these markets to be organized in different ways. The stock market resembles a Walrasian auction market, but one that must be set up by a stock exchange. The New York Stock Exchange, for example, reaps a return and other benefits for providing the service of creating the market. Sunday flea markets held in drive-in movie theaters resemble ArrowDebreu barter exchange markets, but here also market organizers receive a return for creating the market. Other markets are organized differently. Steel workers negotiate compensation packages. As discussed by Walter Oi (chap. 4, this vol.), retail store arrangements are very complex. The market for doctors is set somehow in college chemistry departments, and as Coase (1988) points out those who set up shopping malls bring buyers and sellers together and receive significant compensation for this service. I would argue that the services provided by banks are better understood in the context of a transactions cost model than a neoclassical model.5 Whereas in the neoclassical model firms and markets exist in which trades occur, the transactions-costs approach argues that markets must be made. One essential function of banks is that they make markets in money. This means they quote a price, absorb imbalances during trading, assure immediacy, insure traders against minor stochastic fluctuations in available supplies and demands, and banks do very much more. They operate the payments system. They transform maturities so as to reconcile the market for loans. They assess risks and label customers as worthy of various levels of credit. They provide investment advice; manage portfolios; provide safekeeping for funds; insure against theft; make trades more convenient, and provide payment services. The essence of this wide variety of service activities is inherent characteristics of uncertainty, spatial separation, costliness of private information and 4. See Santornero (1984) for a summary of modeling efforts of banks as firms. 5 . Goodhart (1989) contains an excellent discussion of the role of banks and banking as viewed

in the transactions cost literature as opposed to the Arrow-Debreu approach.

287

Measurement and Efficiency in Banking

the costs of time. Until and unless we can model exactly what services banks provide and how banks provide these services to facilitate the exchange process in various markets, we are not going to know how to measure the services even if we had unlimited access to data. Until w e know what the services are, we cannot tell statistical agencies what data is missing and what to collect. In short, we are, in my judgment, a long way from having viable measures of output in banking.

References Arndt, H. W. 1984. Measuring Trade in Financial Services. Banca Nazionale del Lavorro Quarterly Review 149: 197-2 13. Averch, H., and L. Johnson. 1962. Behavior of the Firm under Regulatory Constraint. American Economic Review December, 1053-69. Bamett, William A. 1980. Economic Monetary Aggregates. Journal of Econometrics 14:11-48. Coase, Ronald H. 1988. The Firm, the Market, and the Law. Chicago: Univ. of Chicago Press. Diewert, W. Erwin. 1981. The Theory of Total Factor Productivity Measurement in Regulated Industries. In Thomas G. Cowing and Rodney E. Stevenson. Productivity Measurement in Regulated Industries, ed. New York: Academic Press. Fixler, Dennis J., and Kimberly D. Zieschang. 1991. Measuring the Nominal Value of Financial Services in the National Income Accounts. Economic Inquiry 29 (January): 153-68. Goodhart, C. A. E. 1989. Money, Information and Uncertainty. Cambridge, Mass.: MIT Press. Hancock, Diana. 1985. The Financial Firm: Production with Monetary and Nonmonetary Goods. Journal of Political Economy 93, no. 5 : 859-80. Mamalakis, Markos J. 1987. The Treatment of Interest and Financial Intermediaries in the National Accounts: the Old “Bundle” Versus the New “Unbundle” Approach. Review of Income and Wealth 33 (June): 169-92. Santomero, Anthony. 1984. Modeling the Banking Firm: A Survey. Journal of Money, Credit, and Banking 16, no. 4, pt. 3576-602. Sealey, C. W., and James Lindley. 1977. Inputs, Outputs, and a Theory of Production and Cost at Depository Financial Institutions. Journal of Finance 32, no. 4:125166. Wykoff, Frank C. 1991. Commercial Banking Productivity Growth: Evidence from Large Bank Balance Sheets. Working paper, Claremont Graduate School.

Comment

Jack E. Triplett

I have elsewhere remarked that progress in the measurement of banking has been inhibited by two major unresolved questions: (1) What are the outputs? Jack E. Triplett is chief economist of the Bureau of Economic Analysis, U.S. Department of Commerce.

288

Allen N. Berger and David B. Humphrey

and (2) What are the inputs? Because these questions correspond exactly to the issues that are displayed in the two papers and discussion in part IIIA of this volume, it may be useful first to summarize approaches to banking output that are found in the literature.’ Comments on the two papers appear in sections 7C.3 and 7C.4.

7C. 1 The ’kaditional National Accounts Approach The oldest measure of banking output is the one contained in the national accounts of most countries. In national accounts, the banking output measure is determined largely as a consequence of the treatment of interest flows. Production originating in a firm (value added) is defined to include net interest payments (interest paid minus interest received), so that the value added of financial firms’ borrowing and lending activities is (1)

VA

= Ci,D,

- Zr,L,,

where the first term records the firm’s deposits (or other financial liabilities) and interest rates paid and the second loans (or other financial assets) and interest rates received. The result is, obviously, normally negative. Because interest earnings enter negatively into equation (l), the major source of bank revenue (income from lending activity) is excluded definitionally from the measure of banking output. Gorman (1969) colorfully remarks that the national accounts treatment of interest flows-unless adjustedleaves the “commercial bank . . . portrayed as a leech on the income stream.” To avoid a clearly nonsensical output measure, banks are assumed in national accounts to provide unpriced or free services to depositors (such as check cashing for which no explicit charges are made) that are equal in value to the entire net proceeds from banks’ lending operations. In some formulations, borrowers are also deemed to receive free services (bookkeeping, credit ratings, and the like). In either case, an imputation for banking output takes the form: where f, and S, are the implicit fee and (unobserved) quantity of unpriced service u , and the other symbols are defined as in equation (1). The total output of the banking industry includes the imputed value of unpriced services, as defined in equation (2), plus the value of services for which an explicit charge is levied (not only certified checks and so forth-a very small part of bank revenue-but also in principle the panoply of financial and fiduciary services that characterize a modem bank). In the United Nations’ (but not in the American) implementation, an additional step assures that most of banking output is excluded from GDP and from international transactions. 1. This material is condensed from Triplett (1991).

289

Measurement and Efficiency in Banking

The national accounts approach to bmking was introduced by Yntema (1947); see also United Nations (1968). Criticisms of the national accounts approach to banking output are quite old. Equation (2) implies that banks act as agents for their depositors (or perhaps for both depositors and borrowers); there is little evidence confirming such a model of bank behavior, or the idea that banks convert their entire earnings into unpriced services. More fundamental is criticism of equation (1) and its exclusion of loan revenue from bank output. Warburton (1958) asserted that a bank’s sources of revenue (interest earnings from loans) are as good an indicator of what banks produce and sell as are the revenues of a coal mine or a laundry, and proposed an alternative services approach that would recognize lending activity as the primary bank output. The services approach has been advocated recently by Sunga (1984), Ruggles (1983), and others. The exclusion of banks’ provision of finance to borrowers from the national accounts measure of banking output is a serious defect for any analytic purpose.

7C.2 The View from the Finance Literature Another approach that emphasizes bank deposits occurs in the macroeconomic literature of money and banking, and finance. In this literature, the major concern is the bank’s role as a portfolio manager, so the banking firm is usually modeled as a seller of deposits (Fama 1980; Pesek 1970; Saving 1977; and Towey 1974)-which is equivalent, of course, to depicting banks as suppliers of money. The traditional money and banking view of banks even has some remote connection to the banking measurement used in national accounts. Baltensperger (1980) and Niehans and Hewson (1976) point out that the traditional finance-macro approach, because it concentrates on portfolio management, neglects the real side of the economy and also neglects the fact that banks function as distributors of funds. To model banks as distributors of funds, it is necessary to think of them as purchasing funds from depositors and offering interest and bartered depositor services as payment for the use of depositors’ funds. The traditional money and banking paradigm-banks selling liquid securities to depositors-is not inappropriate for its own purposes, but it is unenlightening as a paradigm for analyzing bank production and productivity.

7C.3 Bank Production Function Approaches Models of real banking activity and measures of bank output have been developed in the bank regulation literature. To determine whether economies of scale or economies of scope exist in banking, researchers have estimated explicit multioutput production or cost functions, where various bank finan-

290

Allen N. Berger and David B. Humphrey

cia1 outputs and inputs and the usual capital, labor, and materials inputs are specified, Hancock (1991) provides comprehensive references to bank production and cost function studies. Though obtaining a valid measure of output is crucial for modeling bank production and costs, a variety of approaches have been followed, and a consensus on conceptual questions has not yet emerged. One approach-inexplicably known as the production, or sometimes the value-added, approach (but better termed the activity approach)-takes any bank activity that absorbs real resources as a bank output. Benston, Hanweck and Humphrey (1982) remark, “Output should be measured in terms of what banks do that cause operating expenses to be incurred.” In their paper, Allen N. Berger and David B. Humphrey follow a modified activity approach. They define bank outputs “as those activities which have . . . large expenditures on labor and physical capital . . .”; however, they also acknowledge “input characteristics” of deposits and set up their empirical work to incorporate aspects of deposits as both outputs of banking and as banking inputs. U.S. measures of banking labor productivity (Dean and Kunze, chap. 2. this vol.) adopt the activity approach-bank output includes counts of loan and deposit activities (such as loan applications processed and checks cleared). Critics contend that the cost criterion followed in the activity approach does not adequately serve to distinguish financial inputs from financial outputs. Obtaining any financial input incurs some labor and capital costs (processing certificates of deposit, e.g.). In the empirical work, however, the bank deposits that are usually identified as outputs under the activity approach are precisely the ones (demand deposits) where depositor compensation contains large elements of bartered services; those bartered services are clearly produced by the bank and should be included in any comprehensive measure of bank output. In a second approach, the researcher distinguishes a priori between those banking activities that are properly considered the outputs of a bank and others that are deemed financial inputs. For example, Mester (1987) assumes, of savings and loan institutions, that “output is best measured by the dollar value of earning assets of the firm, with inputs being labor, capital, and deposits.” Three outputs (two types of loans, plus other assets) and three deposit inputs (passbook, NOW accounts, and certificates) were specified. Because only bank assets, and not bank liabilities, are specified as outputs, this approach is usually termed the asset approach to defining bank output (though sometimes it is also referred to as the intermediation approach). Bank deposits are regarded as financial inputs to banks, a necessary source of finance that permits them to sell finance to others. The asset approach implies that banks buy funds and sell funds, much the same as any other specialized merchant. It is equivalent to the services approach in the national accounts literature (see sec. 7C.1). A criticism of the assets approach is that its grouping of inputs and outputs

291

Measurement and Efficiency in Banking

is arbitrary. The choices made by some researchers are disputed by others, and the approach admits no mechanisms for resolving such debates. As it has usually been implemented, the asset approach fails to acknowledge the substantial bank production of services that are bartered to depositors as part of the compensation for the use of their funds, a flaw it shares with the parallel services approach in the national accounts literature. A third approach resolves the issues empirically. Appealing to Bamett’s (1980) notion of the “user cost of money,” Hancock (1985, 1991) permits any particular banking activity to be an input or an output according to the sign of its derivative in a bank profit function, which she estimates empirically. In Hancock’s findings, loans are bank outputs (which is consistent with both activity and asset approaches-and, of course, inconsistent with the nationalaccounts approach); time deposits are inputs, but demand deposits are outputs. Fixler and Zieschang follow Hancock’s approach in their paper and obtain similar empirical results, including the finding that demand deposits are bank outputs. A major advantage of the user-cost approach is that it permits statistical tests of the hypotheses maintained in other approaches. Note, however, a potential bias to the empirical results for deposits. Time deposits are typically paid for in strictly monetary terms, so the user cost measure is adequately represented when the nominal cost of deposits is employed in the estimating equation. Demand depositors, on the other hand, receive a large portion of their return in unpriced services. Banks’ user costs of demand deposits are accordingly understated when the value of these bartered services is omitted, which biases the estimated sign of demand deposits in the profit function.2 The bias can readily be seen in Hancock’s (1991, 31-32) expression for the real user cost of a particular deposit type, which (slightly simplified) is

(3)

U,

= -

1

+ (1 + r, + d, + Rk, - s,)/(l + R ) ,

where the variables are defined as follows: U , = real user cost per dollar of type i deposits; r, = interest rate paid to depositors; d, = deposit insurance rate for the type i deposits; R = discount rate; k, = reserve requirement for type i deposits; and s, = actual service charges earned on type i deposits. Equation (3) implicitly takes the bank’s acquisition cost for funds to consist only of direct interest payments, r,. For demand deposits, NOW accounts, and similar sources of funds, nominal interest payments account for only a portion of acquisition cost. On conventional checking accounts, for example, r, = 0, and the entire bank acquisition cost is made up of services for which no explicit charge is made. The value of these services, or the cost of producing them, is omitted from equation (3); if the value of free checks and the like were added in to the numerator of equation (3), the effect must obviously 2. I am indebted to Diana Hancock for helpful comments on the analysis in the following paragraphs.

292

Allen N. Berger and David B. Humphrey

increase the estimated value of U ,, which would make it more likely that demand deposits would be classified as financial inputs (for a financial input, u, > 0). Nominal interest rates are complete measures of compensation for purchased funds and are nearly complete for certificates of deposit and other simple time deposits. In these cases, the estimate of real user cost is positive, because rt is appropriately measured. These deposits are accordingly classified as financial inputs by Hancock (1985, 1991; and also by Fixler and Zieschang, whose approach is similar). If banks adjust service schedules and interest rates on the various accounts they offer so as to equalize the cost of funds at the margin, this implies that the user cost of funds from all sources would be equal; this is, of course, a testable hypothesis, but the hypothesis cannot be tested with data that fail to incorporate a major portion (unpriced services) of banks’ acquisition cost of certain funds. Adding an imputation for the value of unpriced depositor services to the nominal cost of demand deposits would correct the bias, and, one expects, move the estimates in the direction of making demand deposits financial inputs to the bank.3 The omission of unpriced depositor services from the bank deposit user cost measure could also account, in part, for the puzzling sign reversals in Fixler and Zieschang’s findings for deposits. The greater is the proportion of direct interest in total depositor compensation, the more likely are demand deposits to emerge as bank financial inputs. Presumably, deregulation of deposits increased the proportion of explicit payments in total depositor compensation.

7C.4 Conclusions and Research Directions In the three literatures on measuring banking activity summarized in sections 7C. 1-7C.3, the fundamental difficulty arises in the treatment of demand deposits. The underlying cause of the difficulty is the fact that banks compensate depositors at least in part with bartered services, and data on prices and quantities of those bartered services are not available. When deposits are treated as bank output (activity and user-cost approaches, in part), the logic must be that a count of the volume of deposits serves as a proxy for unpriced services produced by the bank and provided to depositors as compensation for the use of their funds. But by thus obtaining an imperfect proxy for the unobserved portion of bank output, the researcher understates a major part of the bank’s cost of funds (though not necessarily 3 . The omission of unpriced services from Barnett’s (1980) formulation of user costs was noted by Offenbacher (1980, 5 3 , who wrote: “Barnett follows the vast majority of money demand studies by assuming that it is useful to treat regulated own rates of return [to deposit holders] as the true rates. , . . It may be more useful to assume that [interest rates ceilings on bank deposits] are almost totally ineffective . . . [and] banks completely evade the ceilings and pay a competitive rate of return on deposits.” Evasion of interest rate ceilings (then set at zero for demand deposits) took the form of provision of varying quantities of depositor services.

293

Measurement and Efficiency in Banking

understating total bank costs) and distorts cost of funds comparisons between banks that use purchased funds, compared with those that obtain funds from traditional deposits. When deposits are treated solely as financial inputs, on the other hand (the asset approach), the substantial part of bank output made up of unpriced services produced by the bank is omitted. The cost of financial inputs is likewise understated by the portion of depositor compensation that takes the form of unpriced services. The same problem arises with respect to the services approach in the national accounts literature: it would correct the conceptual incongruity in the national accounts definition of bank output (its omission of loan activity from the output measure) at the cost of excluding unpriced services that are imputed (if inadequately) in the present measure. The national accounts measure of banking contains, of course, an estimate of the value of unpriced depositor services, but not a defensible one. The national accounts estimate of depositor services is clearly too large, because it, in effect, assigns the loan rate as the opportunity cost forgone by depositor~.~ All approaches to banking thus suffer from the absence of data on bartered banking transactions. No approach satisfactorily deals with demand deposits in the absence of such data, and no approach gets around the basic data deficiency. Once the barter nature of banks’ transactions with depositors is recognized, then it becomes clear that one must separate conceptually depositor services (the bank output) from the deposits themselves, which function as purchased financial inputs to the bank. The value of free checks, automatic-tellermachine usage, and so forth must be added to banks’ output. Simultaneously, the same values must be added to the cost of banks’ purchased financial inputs. From the depositor’s perspective, the value of unpriced services is simultaneously income and outlay on banking services. Obtaining values for unpriced depositor services is a formidable problem. It seems natural to view depositor compensation as consisting of a bundle of interest and unpriced services, much as labor compensation is made up of direct wages plus benefits. One method, applicable in regulated and unregulated environments alike, is to assume that the full value of the bundle is equal for all types of accounts-that banks equalize at the margin the cost of funds 4. Fixler and Zieschang (1991) maintain that, under certain circumstances, the user-cost approach they follow can rationalize the idea that banks pay out their earnings to depositors, and this seems to offer support for the traditional national accounts treatment of banking. Their demonstration is indeed helpful in assessing the plausibility of the agency model of bank behavior (that is, the assumption embodied in eq. [2], sec. 7C. I). However, the essential part of the national accounts approach to banking is its treatment of loans as negative contributions to bank output, in equation (1). This treatment of loans is shared by no other approach to banking (including that of Fixler and Zieschang, this vol.). The negative contribution of loans to equation (1) gives rise to the corresponding necessity for inserting a negative sign before the bracketed quantity in equation (2). That negative sign in equation (2)-and not the sensible mathematics that eliminatesit-is essential to the logic of the national accounts approach to the output of financial firms.

294

Allen N. Berger and David B. Humphrey

from different sources (this implies that deposits are indeed financial inputs to the bank), or that depositors value equally at the margin a dollar's worth of interest and a dollar's worth of unpriced service^.^ This assumption implies that (4) where i, is the interest rate paid on some account with minimal services (a certificate of deposit, perhaps, or purchased funds), i, is the explicit interest (if any) paid on the jth type of account, and (Efu, S,) designates the quantity of unpriced services earned on the jth account. If alternative mixes of interest and services are observed on various accounts, which is true under deregulation, a hedonic function (Griliches 1971) might be used to estimate the unpriced components of depositor compensation (Triplett 1991). This approach is a generalization of equation (4). It requires both schedules of direct interest payments and of uncharged services (the quantity vector S , in equation (4), which would be used in combination to estimate the implicit price vectorf",). Data for implementing a hedonic approach have yet to be assembled, but it is in principle little more difficult than any other hedonic investigation. Beyond this, the heterogeneity of bank loans has not been addressed satisfactorily in empirical estimates. Irrespective of their approach to banking output, banking production function studies frequently consider whether bank output activity is best specified by the count of the numbers of loans (or deposits) of different types, or by their respective monetary volumes. The issue arises, of course, because loans are not a homogeneous commodity: They differ in size and also in other characteristics (riskiness, e.g., or compensating balance requirements). Compensating balance requirements imply that the nominal quantity of loans overstates, and the nominal interest rate understates, the true magnitudes of the loan transaction. Moreover, because banks have extended their financial activities beyond the traditional deposit-taking and lending roles, banking output measures must incorporate these nontraditional activities; some of them (brokerage, selling insurance, executing hedging arrangements) are areas where defining or measuring the output of the activity, or its price, pose conceptual problems comparable in difficulty to the ones confronted in traditional banking. A perhaps more fundamental question also remains. When banks sell finance (or rent loanable funds) to borrowers, what is the nature of the services that finance provides? The ultimate test for the empirical validity of a measure of bank output is to find some effect on, say, the production process and pro5 . Presumably it is after tax returns that are equated by depositors. Interest income is taxable; implicit unpriced services income is not. In Triplett (1991) I argued that the relevant bank marginal cost might differ from the direct cost of producing services if the method of depositor compensation affects the costs of reserves. Neither of these complications needs to be considered here.

295

Measurement and Efficiency in Banking

ductivity of business borrowers, for whom banking output is an intermediate input.

References Baltensperger, Emst. 1980. Alternative Approaches to the Theory of the Banking Firm. Journal of Monetary Economics 6: 1-37. Bamett, William A. 1980. Economic Monetary Aggregates: An Application of Index Number and Aggregation Theory. Journal of Econometrics 14 (September): 11-59 (Annals of Applied Econometrics 1980-83. A Supplement to the Journal of Econometrics). Benston, George J., Gerald A. Hanweck, and David B. Humphrey. 1982. Scale Economies in Banking: A Restructuring and Reassessment. Journal of Money, Credit, and Banking 14, pt. 1 (November): 435-50. Fama, Eugene F. 1980. Banking in the Theory of Finance. Journal of Monetary Economics 6:39-57. Fixler, Dennis J., and Kimberly D. Zieschang. 1991. Measuring the Nominal Value of Financial Services in the National Income Accounts. Economic Inquiry 29 (January): 53-68. Gorman, John A. 1969. Alternative Measures of the Real Output and Productivity of Commercial Banks.” In Production and Productivity in the Service Industries, ed., Victor R. Fuchs, 155-89. NBER Studies in Income and Wealth, vol. 34. Irvingtonon-Hudson, N.Y.: Columbia Univ. Press. Griliches, Zvi, ed. 1971. Price Indexes and Quality Change: Studies in New Methods of Measurement. Cambridge, Mass.: Harvard Univ. Press. Hancock, Diana. 1985. The Financial Firm: Production with Monetary and Nonmonetary Goods. Journal of Political Economy 93:859-80. . 1991. A Theory of Production for the Financial Firm. Boston: Kluwer Academic. Mester, Loretta J. 1987. A Multiproduct Cost Study of Savings and Loans. The Journal of Finance 42 (June): 423-45. Niehans, Jurg, and John Hewson. 1976. The Eurodollar Market and Monetary Theory. Journal of Money, Credit, and Banking 8 (February): 1-27. Offenbacher, Edward K. 1980. Economic Monetary Aggregates-Comment. Journal of Econometrics 14 (September): 11-59 (Annals of Applied Econometrics 1980-83: A Supplement to the Journal of Econometrics). Pesek, Boris P. 1970. Bank’s Supply Function and the Equilibrium Quantity of Money. Canadian Journal of Economics 3 (August): 357-85. Ruggles, Richard. 1983. The United States National Income Accounts, 1947-1977: Their Conceptual Basis and Evolution. In The U.S.National Income and Product Accounts: Selected Topics, ed., Murray F. Foss, 15-96. NBER Studies in Income and Wealth, vol. 47. Chicago: Univ. of Chicago Press. Saving, Thomas R. 1977. A Theory of the Money Supply with Competitive Banking. Journal of Monetary Economics 3:289-303. Sunga, Preetom S. 1984. An Alternative to the Current Treatment of Interest as Transfer in the United Nations and Canadian Systems of National Accounts. Review of Income and Wealth 30:385-402. Towey, Richard E. 1974. Money Creation and the Theory of the Banking Firm. The Journal of Finance 29 (March): 57-72.

296

Allen N. Berger and David B. Humphrey

Triplett, Jack E. 1991. Me~suringthe Output of Banks: What Do Banks Do? BEA discussion paper no. 53. Washington, D.C.: Department of Commerce. United Nations. 1968. A System of National Accounts. Studies in Methods, series F, no. 2. New York: United Nations. Warburton, Clark. 1958. Financial Intermediaries. In A Critique of the United States Income and Product Accounts, 509-16. NBER Studies in Income and Wealth, vol. 22. Princeton, N.J.: Princeton Univ. Press. Yntema, Dwight B. 1947. National Income Originating in Financial Intermediaries. In NBER Studies in Income and Wealth, vol. 10. New York: NBER.

COmI’IleIlt

Diana Hancock

There are several issues that must be resolved before the existing literature on output aggregation can be applied to banking. Of primary importance is a methodology for classifying and measuring financial services. Although it is agreed that banking firms produce heterogeneous services, there has been little consensus on the measurement of their outputs and inputs. The outputs used by various researchers include total assets, earning assets, loans, total deposits, produced deposits, demand deposits in dollar terms, the number of deposit and loan accounts, gross operating income, and combinations of these measures. The central questions in what can be termed “the classification problem” are (1) Which balance sheet items produce services that are net outputs, and which ones are net inputs? In particular, are demand deposit services net outputs, or are these services intermediate inputs? and (2) How does one measure the outputs and inputs, or put prices on them? The measurement of price is dual to the question, What units is output measured in? One can be obtained from the other if the necessary conditions for producer equilibrium are satisfied. Another way of posing the problem is whether stock or flow variables measure the relevant concept of bank output and input. Even with appropriate prices and quantities for financial services determined, the following topics need to be addressed before exact aggregate output indexes for banking can be constructed. First, tests for whether the necessary separability restrictions hold to construct each output subaggregate need to be performed.’ Second, if all prices move proportionately, then a Hicksian aggregation scheme can be used to aggregate over the firm’s joint output supplies. This proportionality assumption is unlikely to hold for financial service prices, due in part to regulation, and hence aggregation over outDiana Hancock is an economist in the Division of Monetary Affairs, Board of Governors of the Federal Reserve System, Washington, D.C. 1. A subaggregate refers to an index containing fewer than all the prices or quantities used in production.

297

Measurement and Efficiency in Banking

puts is possible only if outputs are separable from inputs in the financial firm’s technology.2 Third, the functional form for the aggregator function, and whether it is linear homogeneous, determines the appropriate nonparametric approximation, or index number, which corresponds to the economic output quantity index in banking. The papers by Dennis J. Fixler and Kimberly D. Zieschang and Allen N. Berger and David B. Humphrey both use nonparametric approaches to classify financial services as inputs or outputs. Fixler and Zieschang employ the user-cost approach to determine whether a balance sheet item is a net output of a bank. In contrast, Berger and Humphrey use a value-added approach. The user-cost approach tackles the classification problem by deriving complete rental prices for each balance-sheet item. The user cost of a financial service, or price, is the net effective cost per dollar of holding the asset or liability on the balance sheet over period t . These prices depend on the opportunity cost of capital as well as interest rates, capital gains, reserve requirements, and insurance premiums. In continuous time, the user cost for each asset is the difference between the bank’s opportunity cost of capital and its holding revenue rate. If the holding revenue is not sufficient to cover the opportunity cost of capital, then the balance-sheet item contributes to the financial institution’s costs, and the financial product is a net input. If, however, holding revenues are greater than the opportunity cost of capital, then the firm’s production of this service contributes to revenue, and the service is a financial output. The user cost for each liability incorporates the implicit revenue from deposit balances and takes into account reserve requirement^.^ If holding costs are greater than the opportunity cost of capital, then holding the liability on the balance sheet contributes to costs, and the liability is classified as an input. Estimation of the opportunity cost of capital is important because it influences the prices, the classification of inputs and outputs, and hence the revenues and cost earned from the production of financial services. The paper by Fixler and Zieschang obtains an estimator for the opportunity cost of capital that comes from the specification of the technology producing intermediation 2. See William Bamett, The Microeconomic Theory of Monetary Aggregation, in New Approaches to Munetary Economics,ed. William Bamett and Kenneth Singleton (Cambridge: Cambridge Univ. Press, 1987). 124, for a discussion of this problem in the context of money aggregation. 3 . Berger and Humphrey calculate the implicit revenues from deposit balance j as implicit revenue = (1 - rl/rflj [DB, (1 - k,)]r,,; where r, is the average interest rate paid on depositj, rf is a market rate such as the federal funds rate, DB, is the dollar balance of deposit j , k, is the reserve requirement rate, and r,, is the 90-day Treasury Bill rate. Rearranging terms the implicit revenue per dollar of deposit balance j is (rr, - k,rTa - rl(rrJrflj + k,rJrTJrJ). These implicit revenues are included in the user cost calculation with the assumption that the appropriate opportunity cost of capital is both the 90-day Treasury Bill rate and a proxy for the market rate. See Diana Hancock, The Financial Firm: Production with Monetary and Nonmonetary Goods, Journal of Political Economy 93(1985j: 859-80 for a derivation of user cost formulas for asset and liability items for banks.

298

Allen N. Berger and David B. Humphrey

services, and the assumption of profit maximizing behavior. The authors assume that the value of the opportunity cost of capital for each bank is equal to some proportion of its return on assets, and this proportion is the same for all banks. By adopting this approach, when the industry representative opportunity cost of capital is estimated, a distribution of bank-specific opportunity costs is also obtained because each bank has a different return on By developing a theoretically appropriate opportunity cost of capital for banking, the authors have helped to answer the question of how inputs and outputs for financial services are classified using the user-cost approach.s Fixler and Zieschang estimate a linear homogeneous conditional distance function to obtain an estimator for the opportunity cost of capital. A Malmquist economic output index is the ratio of this conditional distance function for two time periods. The assumption of the linear homogeneity is important because otherwise the output quantity aggregate depends on the choice of the reference quantity used to condition the distance function.6The translog specification for the distance function used in their paper can produce a secondorder approximation to any distance function. An appropriate nonparametric approximation to the Malmquist quantity index is the Tornqvist Divisia index.’ This index is chained, and measures changes relative to the previous period rather than a base period. It remains suitable even when the technology is changing over time, and the aggregator function is shifting. This feature is crucial in the measurement of banking output in the 1980s. Berger and Humphrey state, “The shift over time in the thick frontier cost function, after adjustment for changes in market factors and aggregate interest rates, shows important changes in both operating and interest costs resulting from deregulation.” Hence, a useful measure of aggregate bank output needs to be flexible enough to allow financial institutions to respond to their external environment and technological changes over time. An extension of this approach is to test whether separability conditions for output subaggregates hold. It may be possible to construct aggregates which only use financial service data. The value-added approach assumes that the firm’s technology can be written,

4. The different returns on assets reflect differences in risk, liquidity, and duration across institutions that affect their opportunity cost of capital. 5. Market proxies, such as the 90-day Treasury Bill rate, are provided for the opportunity cost of capital. 6. Fixler and Zieschang use the level of deposits as the reference quantity in their estimation of the technology for banking firms. Quality variables are incorporated in the conditional distance function, too. 7. W. E. Diewert, Exact and Superlative Index Numbers, Journal of Econometrics 4(1976): 115-45, has shown that the discrete Divisia index is exact for the Malmquist quantity index even if the distance function is a nonhomogeneous translog if the reference level is chosen appropriately. Fixler and Zieschang calculate a Tornqvist index of real bank output.

299

Measurement and Efficiency in Banking

where Q, is an exact output aggregate for period t , and the firm’s input vector has been partitioned such that x,, is the vector of quantities of primary input (such as labor and capital) and x2,is a vector of intermediate inputs. The factor price vector is partitioned in a corresponding manner with w, = (wl,, w2). The firm’s variable profit function conditional on xi,is n, = n(xir, wz,. p,), where p, is the output price vector. The true index of real value added is

which depends on the reference prices w;,p* . The need to select the reference prices become unnecessary if and only if g is separable so that

+

In this case nr = ni(x,,)n2(w2, p) and n,orl= ni(xl,o>/ni(x,,i). If has a In translog functional form, then a discrete Divisia index is exact for nrori. continuous time, the Divisia index is always exact for +(xl),which is value added. Berger and Humphrey extend the value-added approach to classify financial services as inputs or outputs. The primary input costs, salaries and fringe benefits, occupancy, furniture and equipment expenses are allocated ex ante to specific balance-sheet items such as real estate loans, and demand deposits using an external source of operating cost allocations.8 Outputs are defined as those services that are responsible for the largest amount of operating costs. This approach assumes that the firm’s technology can be written (4)

Q, = h(xIi,, . . . x,~,~x~),

where the primary input vector x,,has been partitioned into n separate banking functions. Value added is calculated for each financial service, and the index of real value added is

8. Berger and Humphrey use functional cost analysis data. This is a cost accounting system, developed by the Federal Reserve, that assigns direct and joint costs to specific banking functions, such as demand deposits. This system is based on expert information, participant surveys, and accounting rules of thumb.

300

Allen N. Berger and David B. Humphrey

Underlying this representation of the technology is the assumption that the transformation function is nonjoint in inputs, or that there exist individual subproduction functions for each financial s e r v i ~ e . ~ Tests need to be performed using banking data to determine whether the necessary separability conditions hold to construct either a general true index of value added, or an index of product specific value added for banking. The latter index would require additional testing on the structure of the technology for the banking firm. The resulting economic quantity aggregates can be approximated using index number theory. Berger and Humphrey investigate technical change, or shifts in the production technology for banking output over time. Shifts in the minimum-cost technology are distinguished from changes in the dispersion of bank costs away from the minimum technology. The dispersion is decomposed into inefficiency components and market factor components. They find that, if inefficiency is not taken account of, then measures of technical change may be biased in periods of disequilibrium.'0This result is important because it indicates that measurement of technical progress requires estimation of the firm's technology. The rate of technological change may not be able to be measured exactly in banking using input and output indexes. In conclusion, the literature on aggregation and index number theory can be used to construct economic measures of banking output. Examination of the production technology is essential to test whether the necessary separability conditions hold, if jointness in production is statistically important, and to study technical change and productivity. This examination may also help determine whether deposit services are intermediate inputs or outputs. 9. Berger and Humphrey do not impose this structure on their estimating system once outputs and inputs have been classified using the value-added approach. Z. T. Adar, T. Agmon, and Y.E. Orgler, Output Mix and Jointness in Production in the Banking Firm, Journal of Money, Credit, and Bunking 7( 1975): 235-43, argue that interdependence may arise from the joint use of certain inputs by many banking products. Jointness in production is evident in the joint use of information by different departments. An example is the use of depositor information when evaluating a loan application. Jointness in production, also called economies of scope, has been found to be statistically significant in some but not all studies of financial service production. 10. Berger and Humphrey argue that much of the disequilibrium in the 1980s was caused by deregulation, and the less than cost-minimizing response to it by banks.

8

The Output of the Education Sector Dale W. Jorgenson and Barbara M. Fraumeni

In recent years educational expenditures have averaged around 7 percent of the national product. This percentage doubled between 1950 and 1970 and has remained stable since then. Obviously, education is a very important economic activity. Excellent statistics exist on all aspects of education, except the one most fundamental from the economic point of view, namely, the output of the educational system. We need a measure of output to put the education industry on par with other industries producing goods and services. The purpose of this paper is to present a new approach to measuring the output of the education sector. Our point of departure is that, although education is a service industry, its product is investment in human capital. The effects of formal schooling on income endure throughout the lifetime of an educated individual. Accordingly, we employ the effect of education on an individual’s lifetime income as a measure of educational output. A second important idea is that the benefits of schooling are not limited to time spent at work. Education also enhances the value of activities outside the labor market, such as parenting and the enjoyment of leisure time. Our estimates of the output of the education sector incorporate the value of time spent outside the labor market. Beginning with the seminal contributions of Becker (1964), Mincer (1974), and Schultz (1961), economists have found it useful to characterize the benefits of education by means of the notion of investment in human capital.’ This Dale W. Jorgenson is Fredenc Eaton Abbe Professor of Economics at Harvard University and director of the Program on Technology and Economic Policy of the Kennedy School of Government at Harvard. Barbara M. Fraumeni is associate professor of economics at Northeastern University and a research fellow of the Program on Technology and Economic Policy at Harvard University. 1 , Rates of return to investment in human capital are discussed by Becker (1975) and Mincer (1974). Welch (1979) presents estimates of relative rates of return for different age cohorts of the U.S. population. Murphy and Welch (1989) give estimates of rates of return for higher education. Surveys of different aspects of the literature are provided by Griliches (1977) and Rosen (1977).

303

304

Dale W. Jorgenson and Barbara M. Fraumeni

idea captures the fact that investment in human beings, like investment in tangible forms of capital such as buildings and industrial equipment, generates a stream of future benefits. Education is regarded as an investment in human capital because benefits accrue to an educated individual over a lifetime of activities. Of course, investment in education is only one of many forms of investment in human capital. Important investments are made by families in the rearing of their children and by employers and workers in onthe-job training. The most common approach to compiling data on education investment is to measure the inputs, rather than the output, of the educational system.2 Data on the expenditures of educational institutions for teachers and other personnel, buildings and equipment, and materials can be compiled from accounting records. This information can be supplemented by estimates of the value of time spent by students (and their parents) as part of the educational process. Costs of schooling and the value of the time spent by students can be used to measure the flow of resources into schools and universities. Although the costs of education are highly significant in economic terms, the cost-based approach to measurement of educational investment ignores a fundamental feature of the process of education, the lengthy gestation period between the application of educational inputs-mainly the services of teachers and the time of their students-and the emergence of human capital embodied in the graduates of educational institutions. Furthermore, some of the benefits of investment in education, such as greater earning power, are reflected in transactions in the labor market; others-such as better parenting and more rewarding enjoyment of leisure-remain ~ n r e c o r d e d . ~ In measuring the output of the educational system our first step is to compile data on the economic value of market labor activities. In section 8.1 we show that the constant dollar value of time spent working has doubled in the postwar United States. The growth of this value has been greater-or the decline has been less-for women than for men at all levels of educational attainment and reflects the rapid increase in labor force participation by women relative to men. The proportional increase in the value of market labor time has been greatest for college-educated men and women and corresponds to the substantial growth in levels of educational attainment. Our second step in measuring the output of the education sector is to estimate the value of nonmarket labor activities. These activities include both time spent in investment in education and time spent in the consumption of leisure. We infer rates of compensation for nonmarket activities from market wage rates. The value of nonmarket activities, measured in this way, exceeds 2. In this context we employ the notion of output as the economic value produced within the educational sector. Outputs of the educational system can also be defined in terms of measures of educational achievement, such as performance on standardized tests. This definition is the basis for the literature on educational production functions reviewed by Hanushek (1986, 1989). 3 . Nonmarket benefits of education are discussed by Haveman and Wolfe (1984) and Michael (1982).

305

Output of the Education Sector

the value of market activities, primarily because nonmarket time exceeds time in the labor market. However, the value of nonmarket labor activities has grown more slowly. The expansion of the value of nonmarket time has been more rapid for men than for women. We discuss these findings at greater length in section 8.1. In section 8.2 we estimate lifetime labor incomes for all individuals in the U.S. population. These incomes include the value of both market and nonmarket labor time. We then estimate the effect of increases in educational attainment on the lifetime incomes of all individuals enrolled in school. We find that investment in education, measured in this way, is greater in magnitude than the value of working time for all individuals in the labor force. Furthermore, the growth of investment in education has exceeded the growth of market labor activities. Investment in education has increased much more rapidly for women than for men, especially at the college level. We present the conclusions of our study in section 8.3. One of the most attractive aspects of cost-based estimates of investment in education from the accounting viewpoint is that these estimates can be derived primarily or even entirely from data on market transactions. Unfortunately, it is precisely this feature that leads to the undeserved neglect of nonmarket activities. The lifetime-income approach presented in this paper easily encompasses the value of time spent outside the labor market. When applied to education, this approach yields far greater estimates of the output of the education sector than do those approaches based on costs of inputs.

8.1 Market and Nonmarket Labor Incomes In order to measure investment in human capital as an output of the educational system we have constructed a new data base for measuring lifetime labor incomes for all individuals in the U.S. population. This data base includes demographic accounts for the population in each year, cross-classified by sex, age, and year of highest educational attainment. Our demographic accounts include data on the number of individuals enrolled in formal schooling and on the number employed. These demographic accounts are based on annual population data from the U.S. Bureau of the C e n ~ u s . ~ Table 8.1 presents our estimates of numbers of students between 5 and 34 years old enrolled in school, cross-classified by sex and level of e d ~ c a t i o n . ~ Enrollments in grades 1-8 and high school peaked during the late 1960s or the 1970s and have gradually drifted downward through 1986, the last year for which our data are available. Enrollments in college flattened in the 1980s for both men and women and have begun to decline. Enrollments in primary 4. See, e.g., Bureau of the Census (1985). We employ a system of demographic accounts for the United States constructed by Land and McMillen (1981). Demographic accounting is discussed by Stone (1981). 5. See, e.g., National Center for Education Statistics (1988). A compendium of educational statistics is given by O'Neill and Sepielli (1985).

306

Dale W. Jorgenson and Barbara M. Fraumeni

schools have increased over the period 1947-86 as a whole; enrollments in secondary schools have nearly doubled. Enrollments in higher education have risen very dramatically, especially for women. To measure lifetime labor incomes for all individuals in the U.S. population we begin with a data base on market activities constructed by Gollop and Jorgenson (1980, 1983). We derive estimates of hours worked and labor compensation for each sex by 61 age groups and 18 education groups for a total of 2,196 groups for each year. Table 8.2 presents our estimates of the value of time spent working, cross-classified by sex and educational attainment, for all individuals in the U.S. economy from 1948 to 1987. In this table we give estimates of the value of labor time in current prices. The corresponding estimates in constant prices are given for 1949-87 in table 8.3. Labor time in constant prices is a quantity-index number, defined in terms of annual hours worked for individuals cross-classified by age, sex, and educational attainment. To construct a quantity index of labor time, we weight these hours worked by average compensation per hour. We assume that labor time can be expressed as a translog function of its 2,196 components. The growth rate of the corresponding quantity index is a weighted average of growth rates of these components. The weights are given by the shares of each component in the value of market labor time. A quantity index of labor input is unaffected by inflation in rates of labor compensation; the current market value obviously reflects this inflation. The current dollar value of market labor activities has increased 17-fold over the postwar period. The proportional increases were greatest for collegeeducated workers-almost 40 times for men and 65 times for women. The proportional increase for women exceeds that for men for all levels of educational attainment. For the population as a whole the growth of labor compensation is due to a rise in employment and very substantial increases in rates of labor compensation per hour worked. The contrasting trends for men and women are due to a modest rise in employment for men and much greater increase in employment for women. Hours worked per employed person have declined for both sexes. The constant dollar value of market labor activities has more than doubled over the postwar period. However, the quantity index for workers with eight or fewer years of educational attainment has declined substantially. For highschool-educated workers quantity peaks in 1979 for males and rises throughout the period for females. Finally, working time in constant prices increases by more than four and a half times for college-educated males and almost seven times for college-educated females. The constant dollar value of working time for males with a college education exceeds that for high-schooleducated males, beginning in 1980; the value for college-educated females exceeds that for females with a high school education at the end of the period in 1987.

307

Output of the Education Sector

Turning next to the task of evaluating labor time spent in nonmarket activities, we consider activities, such as formal schooling, that enter into investment in human capital and activities that result in consumption. The importance of evaluating time spent in nonmarket activities is widely recognized.6 For example, Nordhaus and Tobin (1972) have incorporated measures of the value of these activities into their measure of economic welfare. Kendrick (1976) and Eisner (1989) have also imputed values for time spent outside the labor market. Five types of nonmarket activities are commonly distinguished in studies of time allocation-household work, human capital investment, travel, leisure, and maintenance-the satisfaction of physical needs such as eating and sleeping. ’ We allocate the total time available for all individuals in the population among maintenance, work, school, and household production and leisure. Studies of time allocation show that maintenance time per capita has changed very little during the postwar period. We estimate that time spent in maintenance is ten hours per day per person and exclude this time from our measure of the value of nonmarket activities. We estimate the time spent in formal education for all individuals enrolled in school and allocate this time to investment. Finally, we allocate the time that is not spent on maintenance, work, or school to consumption. We impute rates of labor compensation for nonmarket activities from wage rates for employed individuals with the same age, sex, and educational attainment. Market wage rates are reduced by taxes on labor incomes estimated by Jorgenson and Yun (1990). Table 8.4 gives the value of nonmarket activities in current prices, crossclassified by sex and educational attainment, for all individuals in the U.S. population for the period 1948-1987. The corresponding estimates in constant prices are given for 1949-87 in table 8.5. As before, nonmarket time in constant prices is a quantity-index number, defined in terms of hours of nonmarket time for all 2,196 categories of workers. Although nonmarket time in current prices reflects inflation in imputed rates of compensation, the quantityindex number is unaffected by inflation. The value of nonmarket activities in either current or constant prices exceeds the value of market activities by a factor of two. This is due to the fact that nonmarket time, as we measure it, is greater than time spent at work. For the population as a whole the growth of the value of nonmarket time is roughly comparable to the growth of the value of work time; however, the distribution of this growth is considerably different. Because each individual has a fixed time budget of 14 hours per day, allocated between market and nonmarket 6 . An economic theory of time allocation is presented by Becker (1965). Detailed references to the literature are given by Murphy (1980). Gates and Murphy (1982) present time use accounts for the United States for 1975-76, based on data collected by the Survey Research Center of the University of Michigan. A survey of time allocation is given by Juster and Stafford (1991). 7. See, e.g., Gates and Murphy (1982) and Juster, Courant, and Dow (1981).

308

Dale W. Jorgenson and Barbara M. Fraumeni

activities, the general pattern for nonmarket time is a mirror image of that for work time. For both men and women the value of nonmarket activities has grown considerably more slowly than the value of time spent working. Given increased rates of labor force participation for women, the value of work time has grown more rapidly for women than for men. With fixed time budgets for both men and women, the value of nonmarket time has increased faster for men. For example, the value of nonmarket time for college-educated men has increased by 42 times, whereas the value for college-educated women has grown by a factor of 38. The relative increase in the value of nonmarket time is greater for individuals of both sexes with higher education than for individuals with only secondary education. This increase is greater for individuals with secondary education than for those with only primary education. These trends reflect increases in levels of educational attainment for both men and women. Our final step in measuring lifetime labor incomes for all individuals in the U.S. population is to project incomes for future years, discount these incomes back to the present, and weight income for each individual by the probability of survival.* We obtain these probabilities by sex from life tables published by the National Center for Health statistic^.^ We combine estimates of lifetime labor incomes by sex, age, and educational attainment with demographic accounts for the numbers of individuals to obtain estimates of human capital, investment in this capital, and the flow of human capital services. The value of the services of human capital is, of course, equal to the sum of the values of market and nonmarket time presented in tables 8.2-8.5 above. In estimating lifetime labor incomes we distinguish among five stages of the life cycle. We assume that all individuals 75 or older are retired, so that the value of current labor time is set equal to zero. Lifetime labor income for these individuals is zero. l o We assume that individuals between 35 and 74 may work but do not attend school. Lifetime labor income is the discounted sum of future labor incomes through age 74, so that the level of educational attainment is held constant. We project future labor incomes for a person of given sex and educational attainment by taking these incomes equal to the current average for all individuals with the same age, sex, and educational attainment, increased by 1.32 percent per year to reflect future increases in real incomes." For example, we project future labor incomes for a male with a high school education at age 35 by first considering current labor incomes for males with 8. Estimates of lifetime labor incomes for men based on market labor activities are presented by Weisbrod (1961), Miller (1965). and Graham and Webb (1979). 9. See National Center for Health Statistics (various annual issues). 10. The proportion of the U.S. population 75 and over has risen from 2.4 percent in 1948 to 5 percent in 1987, so that omissions of lifetime labor incomes for this part of the population imparts a small but slowly increasing bias to our estimates of human wealth for the population as a whole. 11. Our estimate of the growth rate of real incomes is based on the rate of Harrod-neutral productivity growth for the United States estimated by Jorgenson and Yun (1990).

309

Output of the Education Sector

high school education at ages 35,36, and so on, up to age 74. We increase the labor income for a 36-year-old individual by 1.32 percent to reflect increase in real income. We increase labor income for an individual aged 37 by a further 1.32 percent, and so on. We then multiply labor incomes foe ages 35-74 by the probabilities that the individual will survive to each of these ages, given that he or she has already reached the age of 35. Finally, we discount the labor incomes at 4.58 percent per year back to the present.'* For individuals between 14 and 34, we assume that an individual may work as well as enroll in school. For an individual of a given age and sex enrolled in the highest level of formal schooling, which is the 17th year of school or higher, lifetime labor income is the discounted value of labor incomes for a person with 17 years or more of education. For an individual enrolled in the 16th year of school, lifetime labor income includes the discounted value of labor incomes for a person with 17 years of formal education or more, multiplied by the probability of enrolling in the 17th year of school, given enrollment in the 16th year. This income includes the time not spent in school during the 17th year. It also includes the discounted value of labor incomes for a person with 16 years of education, multiplied by one minus this probability, which is the likelihood of terminating formal schooling at 16 years. By working backward from the lifetime incomes of individuals with the highest level of education we can derive the lifetime labor incomes for all individuals enrolled in school. At each level of formal education this is the lifetime labor income of an individual who terminates formal schooling at the end of the current level, multiplied by the probability of terminating at that level, plus the lifetime income of an individual with the next higher level of formal education, multiplied by one minus this probability, which is the likelihood of completing an additional year of schooling. In addition, lifetime labor income for each individual enrolled in school includes the value of time not spent in school. Individuals between 5 and 13 years old are not permitted to participate in the labor market, so that the value of time not spent in school is set equal to zero. However, lifetime labor incomes for these individuals are affected by formal schooling and are calculated in the same way as for individuals between 14 and 35 who are enrolled in school. Because the probabilities of continuing in school are very close to unity for people below the minimum age for leaving school, differences in lifetime labor incomes by age primarily reflect greater discounting of future labor incomes for younger individuals. For people younger than 5 years old lifetime labor incomes are well defined but are not affected by school enrollment. A summary of our methodology in algebraic form is presented in the appendix. 12. Our estimate of the discount rate is based on the long run rate of return for the private sector of the U.S. economy estimated by Jorgenson and Yun (1990).

310

Dale W. Jorgenson and Barbara M. Fraumeni

8.2 Investment in Education To estimate investment in education we employ data on lifetime labor incomes, cross-classified by sex, single age, and single grade of highest educational attainment. We use increments in lifetime labor incomes and estimates of the number of individuals enrolled in school presented in table 8.1 above to measure the value of investment in education.I3 At this point our approach to measuring investment in education incorporates the crucial time dimension of the educational process. Lifetime incomes reflect the effect of educational attainment on the values of future market and nonmarket labor activities over the whole lifetime of an educated individual. These values are discounted back to the present in order to reflect the time value of money. The gestation periods between educational outlays and the final emergence of human capital embodied in the graduates of educational institutions are very lengthy-8 years for individuals completing primary education, 12 years for secondary education, and 16 or more years for higher education. These long gestation periods imply that educational investment must reflect the increase in the value of previous investments in education that are due to the time value of money as well as to the current outlays of educational institutions. In measuring investment in education we focus on increments in life'time labor incomes that are due to increases in educational attainment. These increments incorporate the time value of money for investments in education in earlier time periods. Of course, increments in lifetime labor incomes, as we define them, incorporate the effects of enhanced earning power on the values of both work time and nonmarket time. In table 8.6 we present estimates of the value of educational investment in current prices for 1947-86. We give the corresponding estimates in constant prices for 1948-86 in table 8.7. Our most remarkable finding is that the value of investment in education is considerably greater in magnitude than the value of time spent at work, presented in table 8.2 above. The value of investment in education, as we measure it, accrues in the form of increments to the lifetime incomes of individuals enrolled in school. This value is greater than the value of the time spent at work by the whole labor force. However, the growth in the value of educational investment is almost 21 times the initial level, whereas the increase in the value of work is only 17 times the initial level. This growth reflects the investment associated with rising levels of educational attainment. The growth of investment in education is greater in relative terms for women than for men. Although the value of market activities for collegeeducated women has increased 65 times, the value of investment in higher education for women has grown by a factor of 74. The corresponding growth in the value of market activities for college-educated men is 40 times the ini13. Details are discussed in Jorgenson and Fraumeni (1989).

311

Output of the Education Sector

tial level, and investment in higher education for men has increased by 51 times. The massive rise in investment in education by women is associated with the costs of substantially higher levels of educational attainment. These costs have preceded the entry of more highly educated women into the labor force. Our estimates of investment in education incorporate a number of critical assumptions. We have assumed that the future growth of real incomes is constant at 1.32 percent per year. We have discounted future incomes by 4.58 percent per year to reflect the time value of money. Finally, we have estimated the value of nonmarket labor activities by subtracting time spent in market activities from a total time budget of 14 hours per day for both men and women. We obtain this time budget by allocating 10 hour per day to maintenance for each individual. In order to assess the sensitivity of our estimates to these assumptions, we present a series of alternative estimates of investment in education in table 8.8. In giving investment in education in current prices, we assume in the first panel of table 8.8, that real incomes grow at 2 percent per year and future incomes are discounted at 4 percent per year. We have used these assumptions in earlier estimates of investment in human capital, for example, in Jorgenson and Fraumeni (1989). Because the difference between the discount factor and the growth rate of real income is reduced from 3.26 for the estimates given in table 8.6 to only 2 percent for those in the first panel of table 8.8, we expect the resulting values of investment in education to be substantially larger. The differences decline from 43 percent in 1947 to 33 percent in 1986. These differences are greatest for primary education and reflect the longer gestation period between the investments and the resulting future incomes. To consider the effect of an increase in the difference between the discount factor and the growth rate of real income, we present investment in education in the second panel of table 8.8 under the assumptions that real incomes grow at only 1 percent per year and future incomes are discounted at 6 percent per year. The difference between the discount factor and the growth rate is 5 percent by contrast with 3.26 percent in table 8.6. We anticipate a substantial reduction in the value of investment in education. The difference declines from 36 percent in 1947 to 29 percent in 1986. As in the first panel of table 8.8, estimates of investment in primary education are more strongly affected by this change in assumptions. Although our estimates of investment in education are affected by these assumptions, the qualitative features of the estimates remain the same. An important feature of our estimates of investment in education is that they incorporate the values of both market and nonmarket labor activities. Whereas hours worked in the labor market can be measured directly, hours allocated to nonmarket activities depend on our assumption about the total time available. In the third panel of table 8.8 we reduce our estimate of maintenance time from ten to eight hours per day and thereby increase the time allocated to

312

Dale W. Jorgenson and Barbara M. Fraumeni

nonmarket activities by two hours per day. In the fourth panel of table 8.8 we increase the estimate of daily maintenance to twelve hours, reducing our estimate of nonmarket time by two hours. These alternative assumptions produce relatively modest changes in our estimates of investment in education. As before, the qualitative features of the estimates are unaffected. Investment in education in constant prices is a quantity-index number, based on the school enrollments presented in table 8.1 above. The numbers of individuals in school are weighted by increments in lifetime labor incomes, cross-classified by age, sex, and level of schooling. Investment closely parallels school enrollments for each level of education. However, there are important differences for different levels of schooling. To analyze these differences in greater detail we present investment in education per student in current prices for 1947-86 in table 8.9 and constant prices for 1948-86 in table 8.10. These estimates make it possible to separate trends in the number of students from trends in per-capita levels of educational investment. The value of educational investment per student is far greater than per capita income from market activities. This difference reflects the fact that investment in education includes the effect of formal schooling on the value of nonmarket as well as market activities.I4 For most of the period the values of investment for men and women are similar at all levels of education, despite differences in labor compensation between the sexes. For men the value of investment per student in higher education considerably exceeds that for secondary education, which exceeds in turn the value for primary education. These relationships also hold for women for most of the period. They reflect the lower differentials between wages of workers with secondary and primary education and the greater importance of time discounting for investments in primary education. Investment per student in constant prices increases steadily throughout the period, reflecting the rising enrollments in secondary and higher education for both men and women. Although for men the values of investments in primary and secondary education are relatively constant throughout the period, the value of investment in higher education rises steadily. For women the value of investment in primary education increases, the value in secondary education rises and then falls, and the value in higher education remains almost the same throughout the period. The values of investment in primary and secondary education are higher for women than for men throughout the period, and the value of investment in higher education is greater for women than for men until 1979. We have emphasized that our estimates of investment in education incorporate the value of nonrnarket labor activities. Estimates implicit in the rate of return calculations reported, for example, by Becker (1964) and Mincer (1974) exclude the value of nonmarket time. In order to bring out the signifi14. Kroch and Sjoblom (1986) give estimates of investment in education based on lifetime labor incomes from market activities for men and women.

313

Output of the Education Sector

cance of nonmarket time, we find it useful to consider estimates based on market time alone. To do so requires that we re-estimate lifetime incomes for all individuals in the U.S. population. For this purpose we include the values of work time given in table 8.2 above but exclude the values of nonmarket time presented in table 8.4. Investment in education including only market time is given as a percentage of investment also including nonmarket time in table 8.11. This percentage rises rapidly over the period 1948-52, reflecting increases in labor-force participation during this period. Since 1952 the percentage has varied around 40 percent of the estimates we present in table 8.6 and is higher for men than for women at every level of education. This percentage is rising for women and falling for men. We conclude that the magnitude of this bias is changing for both men and women. In order to capture trends accurately, both market and nonmarket activities must be included in estimates of the value of investment in education. Excluding nonmarket activities from these estimates produces a much more substantial downward bias for women than for men. Human wealth is the sum of lifetime labor incomes for all individuals in the U.S. population. Table 8.12 presents estimates of human wealth in current prices by sex and level of educational attainment for 1947-86. The corresponding estimates in constant prices are given for 1948-86 in table 8.13. These estimates are obtained by multiplying lifetime labor incomes by numbers of individuals in the population, cross-classified by sex, age, and education. The totals presented in tables 8.12 and 8.13 are obtained by summing over age groups. The value of human wealth reflects the value of market and nonmarket activities given in tables 8.2-8.5 above. However, our estimates of human wealth incorporate not only investment in education but also all forms of investment in human capital including, for example, investments in child rearing and the value of new individuals added to the population. In table 8.14 we present the average values of human wealth per person in current prices for individuals cross-classified by sex and educational attainment for 1947-86. We give the average values in constant prices for 1948-86 in table 8.15. These values have increased slightly for primary and higher education throughout the postwar period, but the relative values for men and women have remained fairly stable. By contrast human wealth per person in constant prices for secondary education has declined slightly for both men and women. Growth in human wealth for the population as a whole results from the increase in the population, the rise in average levels of educational attainment, and the growth in rates of labor compensation. Growth in compensation rates is by far the most important component of the increase in human wealth. Our estimates of the value of human wealth, like our estimates of investment in education, are based on lifetime labor incomes that include both market and nonmarket activities. In table 8.16 we present measures of human wealth that exclude nonmarket time as a percentage of the estimates given in table 8.14. For the population as a whole the percentage of human wealth

314

Dale W. Jorgenson and Barbara M. Fraumeni

based on market labor activities alone is fairly stable, varying from 29.5 percent in 1947 to 32.5 percent from 1979 to 1986. However, this percentage has fallen slightly for men from the values of the 1960s. By contrast the percentage has grown very rapidly for women. The omission of nonmarket activities produces a downward bias for women that greatly exceeds the downward bias for men.

8.3 Conclusion Our new estimates of investment in education help to bring the role of human capital in the process of economic growth into proper perspective.15 Economic growth is measured through increments in the national product, as recorded in the U.S. National Income and Product Accounts.16 These accounts are compiled by the Bureau of Economic Analysis of the U.S. Department of Commerce. The accumulation of human and nonhuman capital accounts for the predominant share of economic growth. Although both human and nonhuman capital accumulation are important sources of economic growth, the information required to measure investment in human capital is not available in standard data sources like the U.S. national accounts. For example, the Bureau of Economic Analysis publishes a great deal of valuable information on investment in nonhuman capital.Is The national accounts provide nothing on investment in human ~apita1.I~ The primary reason for this fact is that the accounts are limited to market transactions. Although there have been numerous attempts to augment the U.S. national accounts to incorporate human capital, none of them measures investment in education as an output of the education sector.*O Investment in education, which is a major portion of investment in human capital, is produced almost entirely outside the business sector of the econ15. Jorgenson, Gollop, and Fraumeni (1987), especially chap. 8, present estimates of the contribution of education to U.S. economic growth. In Jorgenson and Fraumeni (1989) we give a complete set of U.S. national accounts, incorporating the estimates of market and nonmarket labor time, investment in education, and human wealth given above. Surveys of the contribution of education to economic growth are presented by Dean (1984), Mincer (1984), and Murnane (1988). 16. See, e.g., Bureau of Economic Analysis (1986). 17. See Jorgenson, Gollop, and Fraumeni (1987), especially chaps. 1 and 9. 18. See Bureau of Economic Analysis (1987). which gives investment and capital stocks for 61 industries broken down by 72 categories of physical assets. 19. Gates (1982) provides time-series estimates of education and training costs for 1965-79. The compendium edited by Peskin (1982) includes other studies of nonmarket activities at the Bureau of Economic Analysis. Unfortunately, the bureau has discontinued this line of investigation. 20. The cost-based approach to measuring investment in human capital was originated by Machlup (1962) and Schultz (1961). Campbell and Peskin (1979) and Eisner (1988) survey augmented accounting systems, including those containing cost-based estimates of investment in human capital. Kendrick’s (1976) accounting system is also discussed by Engerman and Rosen ( 1980).

315

Output of the Education Sector

omy.zl Transmission of education from schools and universities to their students involves increases in educational attainment that are not evaluated in the marketplace, at least not initially. However, the economic value of these increases can be traced through their impact on the lifetime incomes of individuals enrolled in school. Fortunately, participation in schooling is recorded in enrollment statistics. Furthermore, levels of educational attainment are routinely collected for individuals as part of the census of population. We have emphasized the critical importance of including both market and nonmarket incomes in estimating the value of investment in education. In section A of table 8.17 we present a comparison between our estimates of the value of nonmarket activities and the well-known estimates of Nordhaus and Tobin (1972). Their estimates are derived from rates of labor compensation before taxes; our estimates employ after-tax wage rates. The use of before-tax wage rates imparts a substantial upward bias to the estimates of Nordhaus and Tobin; however, the trend in these estimates is nearly identical to that in the estimates we have presented in table 8.4. We have pointed out that existing estimates of the value of human wealth are based on the costs of education. Estimates of this type have been constructed by Kendrick (1976) for an augmented system of U.S. national accounts. We present a comparison of our estimates with those of Kendrick for the period 1948-69 in section B of table 8.17. The ratio of our estimates in current prices to Kendrick’s varies from 17.47 to 18.75 with very little trend from 1948 to 1969. The corresponding ratio for the two constant-price estimates declines from 20.31 in 1948 to 14.29 in 1969. We conclude that Kendrick’s cost-based estimates differ from our lifetime labor income-based estimates by more than an order of magnitude.22The trends in the two sets of estimates are broadly similar, but far from identical. It is important to note that Kendrick’s cost-based estimates of human capital include the accumulated costs of rearing within the family as well as the costs of formal schooling. However, our lifetime income-based estimates include all sources of lifetime labor income, including investment in education, the value of rearing-which is partly offset by depreciation of human capital with aging-and the lifetime incomes of individuals added to the population, prior to any investment in education or rearing. Nonetheless, the disparities between the two sets of estimates of human capital are very striking. These disparities provide a graphic demonstration of the conceptual differences between the cost-based approach and the income-based approach to the measurement of investment in human capital. 21. The educational sector is discussed from the economic point of view in the collection of papers by Froomkin, Jamison, and Radner (1976). 22. Graham and Webb (1979) compare Kendrick’s estimate of human wealth for 1969 with estimates based on lifetime labor incomes for males, excluding the value of nonmarket activities. Kroch and Sjoblom (1986) compare their estimates of human capital accumulated through education, on the basis of lifetime labor incomes from market activities for men and women, with Kendrick’sestimates, based on costs of education and training.

316

Dale W. Jorgenson and Barbara M. Fraumeni

Although cost-based estimates of investment in education reflect the current flow of resources into educational institutions, they do not capture the crucial time dimension of educational investment. There is a lengthy gestation period between the current outlays of educational institutions and the emergence of human capital embodied in their graduates. A very substantial proportion of educational investment is attributable to the time value of money, applied to previous investments in the education of individuals who are still enrolled in school. This feature of investment in education is entirely disregarded in estimates limited to current educational outlays. The availability of estimates of the output of the education sector has created an opportunity for important new research on educational productivity. By combining cost-based estimates of educational inputs with our estimates of educational output, it is possible to measure the productivity of the educational sector. A productivity measure for this sector requires estimates of capital, labor, and intermediate inputs in current and constant prices like those compiled by Jorgenson, Gollop, and Fraumeni (1987) for all the other industries that make up the U.S. economy. An important issue that remains to be resolved is the appropriate valuation of the time spent in educational institutions by students. This time is an important input into the educational sector. We conclude that the time scale for measuring human capital formation is given by the life span of an educated individual. The appropriate value of investment in education is given by its effect on the individual’s lifetime income. The relevant concept of income must not be limited to market activities alone, because many of the benefits of education accrue in the form of enhanced value to nonmarket activities. Our estimates of investment in education incorporate the effect of higher educational attainment on the value of nonmarket activities such as parenting or enjoyment of leisure as well as the effect of increased education on earning power in the labor market. Our estimates of investment in education are based on very detailed information on the value of working time. However, we have based our estimates of the value of nonmarket labor time on market wage rates. The valuation of nonmarket activities could be refined considerably, especially for individuals not in the labor force. An alternative approach is to infer the value of nonmarket time from labor supply behavior. Second, we have estimated the value of increments in lifetime incomes as a result of increases in educational attainment by comparing the incomes of individuals of the same age and sex with different levels of education. An important further refinement would base estimates of differences on lifetime incomes on the determinants of educational attainment for a given individual. These limitations of our existing estimates suggest opportunities for significant new research on the benefits of education. Finally, another important source of new research opportunities is the extension of our methods to encompass other forms of investment in human capital. We have already mentioned three extensions of this type. First, fertil-

317

Output of the Education Sector

ity behavior is influenced by the lifetime incomes of children added to the population and by the effects of childbearing on the lifetime incomes of parents. Second, investment in child rearing is an important component of investment in human capital and can be measured on the basis of its effect on lifetime incomes of children. Third, the value of on-the-job training can be appraised by employers and workers in terms of its effect on lifetime labor incomes.23

Appendix In this appendix we outline the methodology for measuring the output of the educational sector in algebraic form. To represent the use of time and the corresponding labor income we require the following notation: y

=

1947, 1948, . . ., 1987-calendaryear. s = 1, 2-sex,

a

=

male or female.

0, I , . . ., 74, 75, 75+-age.

e = 1, 2 , . . ., 18-educational attainment, none or less than grade one, grade one, . . ., five years of college or more. The variables required for estimates of the output of the educational sector are denoted as follows: com-hourly

compensation, net of taxes on labor income.

empr-employment

rate.

life-lifetime

labor income per capita.

mhrs-market

labor time per capita.

mi-lifetime

market labor income per capita.

nmhrs-nonmarket nmi-lifetime

labor time per capita.

nonmarket labor income per capita.

senr-school enrollment rate, the probability that an individual with 1. educational attainment e is enrolled in educational level e

+

shrs-school hours per capita; enrolled individuals are assumed to be in school 1300 hours per year. si-investment

in education per capita.

23. A survey of recent research on the prevalence and impact of on-the-job training is presented by Mincer (1989a). Mincer (1989b) presents estimates of the annual costs of training in the United States for 1958, 1976, and 1987. For 1976 these costs amount to half the costs of formal schooling.

318

Dale W. Jorgenson and Barbara M. Fraumeni

sr-probability tax-average

of survival, specific to the year of birth.

tax rate on labor income.

taxam-average whrs-annual ymi-annual

marginal tax rate on labor income. market hours worked per person employed.

market income per capita, net of tax on labor compensation.

ynmi-annual nonmarket income per capita, net of tax on labor compensation, where the tax is calculated at the average marginal rate. Our first set of equations provides estimates of annual hours of market and nonmarket components of labor time. The first equation gives school hours per capita: shrs,,,9,,,e= senry,s,a,e * 1300. The second equation gives market hours per capita: mhrs>s.n.e= whrsy,s,ae* emPry,s,a,e. Our third through eighth equations give nonmarket hours per capita for each of the five stages of the life cycle described in section 8.2. Stage 1 includes ages 0-4; stage 2 includes ages 5-13; stage 3 includes ages 14-34; stage 4 includes ages 35-74; stage 5 includes ages 75 and over. Maintenance time per capita is 10 hours per day, leaving 14 hours per day to be allocated between market and nonmarket time. The first stage is no school and no work: nmhrsy,r,a,r = 14 * 7

* 52.

The second stage is school but no work: nmhrsv,s,y,,= 14 * 7

* 52

- shrs,.,,,,c.

The third stage is school and work: nmhrsv,,T,,,,= 14 * 7

* 52 - shrs,,.,,, - mhrsy,s,a,e.

The fourth stage is work but no school: nmhrs,,,,,, = 14 * 7

* 52

- mhrs,,s,o,e.

The fifth and final stage is retirement or no school or work: nmhrs,,,,,, = 14 * 7

* 52.

Our second set of equations provides estimates of market labor income. The first equation gives annual market labor income per capita: Ymiws,a,e = mhrs>s,a,e*

~0my.s.a.e.

The second equation gives annual nonmarket labor income per capita: ynmi,,,,,, = nmhrsy,,v,,,, * comy,s,,s.a,, * (1

+ tax,) * (1 - taxamJ.

319

Output of the Education Sector

Our third through eighth equations give lifetime market labor income per capita at the five stages of the life cycle described in section 8.2. Lifetime incomes are calculated by a backward recursion, starting with age 74, which is the oldest age before retirement. Future incomes are discounted back to the current age of the individual. The first stage is no school and no work:

The second stage is school but no work:

The fourth stage is work but no school:

The fifth and final stage is retirement or no school or work: mi?,s,a.c = 0 Our third set of equations gives estimates of nonmarket labor income. The first through fifth equations give lifetime nonmarket labor income for the five stages of the life cycle described in section 8.2. The first stage is no school or work:

The second stage is school but no work:

320

Dale W. Jorgenson and Barbara M. Fraumeni

The fourth stage is work 5ut no school:

nmi,,.,

=

Ynmil+lruc + SrIsatI

* nmi,s,+,, *

1.0132 ~

1.0458’

The fifth and final stage is retirement or no school or work:

nmi>sop = 0 Total lifetime labor income per capita, including market and nonmarket components is M e > 5 o e= m i , A o c+ n m i , A a e

Investment in education per capita is:

References Becker, G. S. 1965. A Theory of the Allocation of Time. EconomicJournal75:493517. . 1975. Human Capital. 2d ed. New York: Columbia Univ. Press. Bureau of the Census. 1985. Census of Population and Housing, 1980: One Percent Sample Computer Tape. Washington, D.C.: Department of Commerce. Bureau of Economic Analysis. 1986. The National Income and Product Accounts of the United States, 1929-1982, Statistical Tables. Washington, D.C.: Government Printing Office. . 1987. Fixed Reproducible Tangible Wealth in the United States, 1925-85. Washington, D.C.: Government Printing Office. Campbell, B., and J. Peskin. 1979. Expanding Economic Accounts and Measuring Economic Welfare:A Review of Proposals. Washington, D.C.: Department of Commerce, Bureau of Economic Analysis. Dean, E., ed. 1984. Education and Economic Productivity. Cambridge, Mass.: Ballinger. Eisner, R. 1988. Extended Accounts for National Income and Product. Journal of Economic Literature 24: 1611-84. . 1989. The Total Incomes System of Accounts. Chicago: Univ. o f Chicago Press. Engerman, S., and S. Rosen. 1980. New Books on the Measurement of Capital. In The Measurement of Capital, ed. D. Usher, 153-70. Chicago: Univ. of Chicago Press. Froomkin, J. T.,D. T. Jamison, and R. Radner, eds. 1976. Education as an Industry. Cambridge: Ballinger. Gates, J. A. 1982. Education and Training Costs: A Measurement Framework and Estimates for 1965-79. In Measuring Nonrnarket Activity, ed. J . Peskin, 107-35. Washington, D.C.: Government Printing Office. Gates, J. A , , and M. Murphy. 1982. The Use of Time: A Classification Scheme and

321

Output of the Education Sector

Estimates for 1975-76. In Measuring Nonmarket Activity, ed. J. Peskin. 3-22. Washington, D.C.: Government Printing Office. Gollop, F. M., and D. W. Jorgenson. 1980. U.S. Productivity Growth by Industry, 1947-73. In New Developments in Productivity Measurement and Analysis, ed. J. W. Kendrick and B. N. Vaccara, 17-136. Chicago: Univ. of Chicago Press. . 1983. Sectoral Measures of Labor Cost for the United States, 1948-1979. In The Measurement of Labor Cost, ed. J. E. Triplett, 185-235, 503-20. Chicago: Univ. of Chicago Press. Graham, J. W., and R. H. Webb. 1979. Stocks and Depreciation of Human Capital: New Evidence from a Present-Value Perspective. Review of Income and Wealth 25:209-24. Griliches, Z . 1977. Estimating the Returns to Schooling: Some Econometric Problems. Econometrica 45:l-22. Hanushek, E. A. 1986. The Economics of Schooling. Journal of Economic Literature 24~1141-78. . 1989. The Impact of Differential Expenditures on School Performance. Educational Researcher 18:45-51. Haveman, R. H., and B. L. Wolfe. 1984. Schooling and Economic Well-Being: The Role of Nonmarket Effects. Journal of Human Resources 19:377-407. Jorgenson, D. W., F. M. Gollop, and B. Fraumeni. 1987. Productivity and U.S. Economic Growth. Cambridge: Harvard Univ. Press. Jorgenson, D. W., and B. M. Fraumeni. 1989. The Accumulation of Human and Nonhuman Capital, 1948-1984. In The Measurement of Saving, Investment, and Wealth, ed. R. E. Lipsey and H. S. Tice, 227-82. Chicago: Univ. of Chicago Press. Jorgenson, D. W., and K.-Y. Yun. 1990. Tax Reform and U.S. Economic Growth. Journal of Political Economy 98:s 151-193. Juster, F. T., P. N. Courant, and G. K. Dow. 1981. The Theory and Measurement of Well-Being: A Suggested Framework for Accounting and Analysis. In Social Accounting Systems, ed. F. T. Juster and K. C. Land, 23-94. New York: Academic Press. Juster, F. T., and K. C. Land, 1981. SocialAccounting Systems. New York: Academic Press. Juster, F. T., and F. P. Stafford. 1991. The Allocation of Time: Empirical Findings, Behavioral Models, and Problems of Measurement. Journal of Economic Literature 29: 47 1-522. Kendrick, J. W. 1976. The Formation and Stocks of Total Capital. New York: Columbia Univ. Press. Kroch, E., and K. Sjoblom. 1986. Education and the National Wealth of the United States. Review of Income and Wealth 32:87-106. Land, K. C., and M. M. McMillen. 1981. Demographic Accounts and the Study of Social Change, with Applications to Post-World War I1 United States. In Social Accounting Systems, ed. F. T. Juster and K. C. Land, 242-306. New York: Academic Press. Machlup, F. 1962. The Production and Distribution of Knowledge in the United States. Princeton: Princeton Univ. Press. Michael, R. T. 1982. Measuring Non-monetary Benefits of Education: A Survey. In Financing Education: Overcoming Ineficiency and Inequity, ed. W. W. McMahon and T. G. Geske, 119-49. Urbana: Univ. of Illinois Press. Miller, H. P. 1965. Lifetime Income and Economic Growth. American Economic Review 55:834-44. Mincer, J, 1974. Schooling, Experience, and Earnings. New York: Columbia Univ. Press.

322

Dale W. Jorgenson and Barbara M. Fraurneni

. 1984. Human Capital and Economic Growth. Economics of Education Review 3:195-205. . 1989a. Human Capital and the Labor Market. Educational Researcher 18:2734. . 1989b. Job Training: Costs, Returns, and Wage Profiles. New York: Columbia Univ., Department of Economics. Murnane, R. 1988. Education and the Productivity of the Work Force: Looking Ahead. In American Living Standards: Threats and Challenges, ed. R. E. Litan, R. Z. Lawrence, and C. L. Schultze, 215-45. Washington, D.C.: Brookings Institution. Murphy, K., and F. Welch. 1989. Wage Premiums for College Graduates: Recent Growth and Possible Explanations. Educational Researcher 18:27-34. Murphy, M. 1980. The Measurement and Valuation of Household Nonmarket Time. Washington, D.C.: Department of Commerce, Bureau of Economic Analysis. National Center for Education Statistics. 1988. Digest of Education Statistics. Washington, D.C.: Department of Education. National Center for Health Statistics. Various annual issues. Vital Statistics of the United States. Washington, D.C.: Department of Health, Education and Welfare, Public Health Service. Nordhaus, W. D., and J. Tobin. 1972. Economic Research: Retrospect and Prospect. Vol. 5 , Economic Growth. New York: National Bureau of Economic Research. O’Neill, D. M., and P. Sepielli. 1985. Education in the United States: 1940-1983. Washington, D.C.: Government Printing Office. Peskin, J., 1982. Measuring Nonmarket Economic Activity. Washington, D.C.: Government Printing Office. Rosen, S. 1977. Human Capital: A Survey of Empirical Research. In Research in Labor Economics, Vol. I , ed. R. G. Ehrenberg, 3-39. Greenwich, Conn.: JAI Press. Schultz, T. W. 1961. Investment in Human Capital. American Economic Review 51:l17. Stone, R. 1981. The Relationship of Demographic Accounts to National Income and Product Accounts. In Social Accounting Systems, ed. F. T. Juster and K. C. Land, 307-76. New York: Academic Press. Weisbrod, B. A. 1961. The Valuation of Human Capital. Journal of Political Economy 69:425-36. Welch, F. 1979. Effect of Cohort Size on Earnings: The Baby Boom Babies’ Financial Bust. Journal of Political Economy 87:S65-98.

323

Output of the Education Sector School Enrollment by Sex and Level, United States, 1947-86 (thousands)

Table 8.1

Male

Year

Total

Grades 1-8

1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

28,411 28,876 29,581 30,318 30,980 31,721 33,011 34,433 35,791 37,166 38,577 40,028 41,492 43,198 44,643 46,121 47,645 49,140 50,432 51,665 52,894 54,068 55,102 55,907 56,447 56,717 56,736 56,554 56,301 55,996 55,680 55,200 54,437 53,552 52,696 52,648 51,980 51,664 51,037 51,110

9,871 10,120 10,485 10,840 11,120 11,407 11,954 12,545 13,072 13,551 13,954 14,368 14,819 15,382 15,683 15,929 16,203 16,496 16,759 16,991 17,206 17,358 17,421 17,392 17,282 17,048 16,739 16,389 16,037 15,723 15,476 15,202 14,863 14,560 14,098 13,989 13,832 13,722 13,577 13,612

Female

High School College 3,593 3,570 3,555 3,562 3,623 3.712 3,810 3,922 4,055 4,232 4,493 4,756 4,969 5,157 5,442 5,797 6,154 6,475 6,636 6,756 6,901 7,080 7,268 7,434 7,616 7,783 7,908 7,989 8,037 8,048 8,017 7,968 7,843 7,644 7,425 7,361 7,178 7,149 7,064 7,177

1,663 1,694 1,719 1,741 1,702 1,669 1,648 1,644 1,655 1,685 1,730 1,788 1,870 1,999 2,164 2,335 2,498 2,668 2,950 3,271 3,582 3,899 4,218 4,567 4,764 4,957 5,129 5,273 5,401 5,499 5,562 5,517 5,563 5,511 5,879 5,940 6,063 6,058 5,855 5,744

Grades 1-8

High School

College

9,142 9,387 9,731 10,069 10,352 10,639 11,177 11,751 12,259 12,718 13,097 13,497 13,950 14,497 14,767 15,002 15,283 15,580 15,838 16,072 16,276 16,417 16,511 16,450 16,352 16,129 15,833 15,503 15,171 14,878 14,647 14,381 14,062 13,775 13,359 13,308 13,174 13,005 12,869 12,908

3,378 3,341 3,324 3,333 3,403 3,503 3,61 I 3,731 3,873 4,055 4,324 4,579 4,771 4,943 5,238 5,588 5,927 6,229 6,347 6,458 6,607 6,788 6,966 7,130 7,300 7,451 7,562 7,638 7,679 7,677 7,635 7,565 7,427 7,214 7,118 7,088 6,843 6,757 6,673 6,779

764 764 767 773 780 79 1 811 840 877 925 919 1,040 1,113 1,220 1,349 1,470 1,580 1,692 1,902 2,117 2,322 2,526 2,718 2,934 3,133 3,349 3,565 3,762 3,976 4,171 4,343 4,507 4,679 4,848 4,817 4,962 4,890 4,973 4,999 4,890

324

Dale W. Jorgenson and Barbara M. Fraumeni Value of Market Activities by Sex and Educational Attainment, 1948-87 (billions of current dollars)

Table 8.2

Male Year

Total

Grades 1-8*

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 I966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

147.3 151.2 164.5 187.5 201.4 215.2 220.6 229.4 245.6 262.8 274.1 285.4 299.4 311.4 328.8 340.8 365.9 388.4 429.1 460.4 504.2 548.7 602.3 651.4 701.5 797.8 870.1 941.7 1,036.0 1,132.4 1,273.4 1,429.1 1,560.8 1,690.3 1,757.6 1,869.8 2,047.3 2,190.2 2,350.9 2,519.6

44.2 44.1 47.6 51.8 53.6 55.8 54.3 54.4 56.4 58.0 56.9 57.2 58.4 55.9 55.8 55.8 56.6 58.3 62.4 63.1 65.4 67.2 68.1 66.4 66.0 69.1 71.5 65.9 67.8 70.1 74.8 77.6 77.6 76.8 69.2 66.4 69.2 70.8 71.5 72.0

Female

High School College Grades 1-8* 51.6 52.6 56.0 65.2 70.5 76.1 79.1 83.1 89.6 96.6 101.9 107.5 113.8 118.3 125.2 131.1 141.5 151.2 167.5 178.2 195.5 211.7 230.2 247.6 267.3 301.7 323.5 341.4 370.5 399.1 441.6 490.0 518.8 553.4 554. I 561.4 620.7 654.5 686.5 722.5

24.1 25.6 29.2 35.4 40.0 43.9 47.3 49.7 54.3 60.0 65.4 68.7 70.8 76.5 81.7 85.3 92.2 97.1 106.8 118.5 131.0 144.0 160.3 181.5 201.7 236.8 262.7 298.4 335.3 371.3 423.5 485.0 533.8 586.0 617.7 683.3 752.2 809.8 875.6 946.8

7.9 8.2 8.7 9.2 9.5 9.7 9.4 9.8 10.2 10.4 10.5 10.6 11.0 10.5 10.5 10.8 11.3 11.6 12.5 13.2 13.8 14.7 15.4 15.5 15.3 15.8 16.2 16.5 17.3 17.5 19.8 20.2 21.4 21.6 21.1 20.9 21.6 21.5 22.3 22.4

High School College 13.6 14.3 15.1 17.2 18.6 19.8 20.3 21.6 23.5 25.2 26.1 27.3 30.2 32.9 36.3 38.6 43.2 47.5 54.4 58.8 65.6 74.8 85.5 90.8 97.9 111.0 121.0 132. I 145.0 161.6 183.7 204.8 231.4 253.1 268.9 281.6 303.9 320.1 347.6 371.0

*The grades 1-8 column includes persons who have completed less than the first grade

5.9 6.5 7.9 8.7 9.3 9.8 10.2 10.8 11.7 12.6 13.5 14. I 15.3 17.3 19.3 19. I 21 .o 22.7 25.6 28.6 32.8 36.3 42.7 49.5 53.4 63.3 75.3 87.4 100.1 112.9 130.0 151.6 177.8 199.5 226.5 256.2 279.6 313.5 347.4 384.9

325

Output of the Education Sector

Table 8.3

Value of Market Labor Activities by Sex and Educational Attainment, 1949-87 (billions of constant dollars) Male

Year

Total

Grades 1-8*

High School

1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

981.8 1,021.1 1,105.3 1,134.8 1,154.6 1,130.1 1,163.6 1,187.3 1,191.5 1,169.2 1,207.4 1,221.2 1,231.1 1,268.1 1,285.0 1,312.8 1,352.8 1,399.6 1,419.8 1,450.2 1,487.3 1,467.9 1,466.2 1,508.1 1,564.2 1,568.1 1,544.0 1,587.7 1,640.2 1,713.2 1,772.3 1,764.8 1,782.8 1,757.6 1,799.9 1,897.8 1,937.5 1,953.2 2,013.4

308.8 318.1 326.7 324.9 323.1 302.6 301.5 297.5 287.0 267.8 268.6 259.8 239.8 231.4 225.0 214.4 212.1 208.6 198.2 190.2 182.2 164.4 149.1 142.2 135.0 127.7 107.5 103.0 100.2 98.6 93.8 86.1 79.9 69.2 64.3 64.9 63.3 60.1 58.3

336.4 343.8 383.8 399.6 411.8 409.4 423.1 434.3 439.9 436.5 454.6 464.6 470.6 487.3 497.9 510.7 528.2 549.2 554.5 569.1 580.8 571.1 564.6 579.2 593.3 583.4 560.9 568.2 577.3 593.2 605.9 587.0 584.3 554.1 546.9 583.4 590.8 584.0 594.3

Female College Grades 1-8* 163.0 176.8 200.1 212.2 219.9 223.2 232.2 240.3 246.2 248.2 259.8 266.9 280.2 295.3 305.7 318.6 329.8 342.5 362.7 375.2 392.0 393.0 406.7 432.1 462.1 471 .O 484.9 510.5 536.1 570.0 599.8 601.7 616.0 617.7 654.3 693.4 711.1 718.9 744.9

59.9 60.8 61.5 60.0 58.3 54.1 56.3 56.5 55.0 52.6 53.1 52.0 47.9 45.9 45.5 44.4 43.9 43.5 42.2 40.7 40.0 37.3 34.8 33.0 31.8 29.9 27.9 27.0 25.8 26.9 25.2 23.9 22.6 21.1 20.1 19.9 18.7 18.3 17.8

High School College 91.8 93.2 102.6 106.2 108.8 106.9 113.9 119.4 121.8 120.3 125.2 129.0 134.6 142.7 146.9 155.6 165.0 176.9 179.2 185.6 198.5 204.6 204.5 21 1.4 220.6 221.5 220.3 225.8 237.2 249.2 257.5 263.7 268.2 268.9 269.5 279.9 280.3 286.1 294.1

*The grades 1-8 column includes persons who have completed less than the first grade

43.6 50.5 52.5 53.1 53.1 52.0 54.3 56.2 57.0 56.9 58.2 59.6 65.5 71.1 68.7 72.5 76.5 81.1 84.7 90.7 94.7 97.9 106.7 110.2 121.2 134.0 141.8 152.3 162.7 174.5 189.6 202.2 211.7 226.5 244.6 256.1 272.9 284.9 302.8

Dale W. Jorgenson and Barbara M. Fraumeni

326

Value of Nonmarket Activities by Sex and Educational Attainment, 1948-87 (billions of current dollars)

Table 8.4

Male Year 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987

Total 376.4 401.7 415.4 425.3 441.4 465.6 512.7 519.1 546.9 590.8 638.7 644.4 693.3 739.5 778.4 804.2 878.8 953.7 1,038.8 1,127.4 1,218.1 1,317.7 1,517.4 1,676.3 1,788.1 1,937.9 2,147.4 2,365.3 2,559.9 2,725.3 2,993.3 3,294.1 3,629.3 3,930.1 4,372.8 4.706.9 4,942.2 5,346.6 5,774.8 6,536.0

Grades 1-8* 78.0 80.8 80.4 81.5 82.2 84.4 91.4 90.2 93.5 100.0 106.2 102.9 108.3 116.0 119.6 118.4 126.7 132.6 141.6 148.2 154.0 160.7 177.7 190.8 195.0 206.8 219.2 234.6 241.6 243.7 255.7 270.0 271.2 287.8 300.1 309.1 315.6 343.6 358.1 381.4

Female

High School College Grades 1-8* 75.7 84.1 89.4 89.4 92.5 98.2 109.5 113.3 121.3 132.7 146.5 149.8 161.1 170.4 178.1 184.4 202.8 220.9 239.8 260.8 281.6 306.5 357.0 397.9 424.6 463.7 518.6 575.7 622.2 661.8 724.8 795.6 877.5 972.4 1,089.6 1,145.9 1,188.6 1,259.9 1,369.2 1,557.8

37.5 41.8 43.8 45.8 48.9 53.4 60.9 63.3 68.8 76.8 86.4 88.1 92.5 99.5 105.1 109.4 120.1 131.9 146.6 158.4 174.9 190.9 226.3 261 .O 287.3 322.6 373.6 426. I 473.8 519.1 583.8 665.0 764.9 832.6 961.2 1.071.8 1,127.3 1,242.9 1,373.7 1,583.9

68.5 69.3 69.6 70.3 71.7 73.9 78.5 76.5 77.7 80.8 83.2 80.8 85.0 90.7 95.3 96.6 103.5 110.6 118.8 127.5 135.4 143.9 159.7 173.9 178.7 181.8 192.1 201.3 211.4 216.7 228.5 243.3 255.2 271.9 282.6 287.6 300.4 323.7 331.7 356.0

High School College 87.4 93.8 98.7 102.3 107.4 114.0 125.2 127.3 134.1 144.4 155.1 159.0 176.6 188.4 201.1 210.7 232.3 255.0 278.0 305.5 332.8 361.6 413.3 448.6 475.8 511.9 563. I 612.5 661.2 701.1 768.3 837.5 912.7 987.1 1,093.0 1,149.6 1,209.3 1,281.8 1,367.8 1,525.5

*The grades 1-8 column includes persons who have completed less than the first grade.

29.4 32.0 33.6 36.0 38.7 41.8 47.1 48.3 51.4 56.1 61.4 63.8 69.9 74.5 79.3 84.7 93.4 102.6 1 14.1 126.9 139.3 154.1 183.5 204.1 226.8 251.1 280.8 315.2 349.7 382.8 432.2 482.7 541.8 578.2 646.3 742.8 801.1 894.7 974.4 1,131.3

327

Output of the Education Sector

Table 8.5

Value of Nonmarket Labor Activities by Sex and Educational Attainment, 1949-87 (billions of constant dollars) Male

Year

Total

1949 2,438.5 1950 2,457.0 1951 2,434.7 1952 2,456.9 1953 2,484.0 1954 2,551.5 1955 2,570.6 1956 2,599.5 1957 2,647.5 1958 2,713.6 1959 2,730.0 1960 2,784.0 1961 2,849.9 1962 2,888.7 1963 2,942.6 1964 2,988.3 1965 3,030.0 1966 3,070.5 1967 3,136.1 1968 3,195.6 1969 3,250.5 1970 3,351.4 1971 3,447.7 1972 3,513.3 1973 3,563.8 1974 3,659.8 1975 3,782.8 1976 3,855.2 1977 3,921.5 1978 3,975.8 1979 4,040.4 1980 4,153.4 1981 4,240.0 1982 4,372.8 1983 4,468.1 1984 4,503.7 1985 4,578.9 1986 4,663.8 1987 4,731.4

Grades 1-8* 524.2 505.1 489.5 481.9 473.6 480.7 471.5 465.2 465.4 468.8 455.9 452.3 461.3 454.4 442.9 435.1 421.6 409.0 402.1 392.4 382.2 378.2 380.1 374.3 367.4 360.5 364.1 354.6 343.0 330.3 319.1 308.8 305.6 300.1 295.3 292.3 295.2 288.2 274.9

High School 484.4 502.1 489.3 496.1 504.7 526.3 535.2 546.9 563.2 587.2 594.9 613.7 630.7 642.1 659.4 675.6 689.5 700.9 725.2 743.8 766.1 805.6 830.8 840.4 849.2 877.5 916.3 932.1 946.2 955,9 967.6 1,002.3 1,045.6 1,089.6 1,094.6 1,092.1 1,096.5 1,123.7 1,151.5

Female College Grades 1-8* 238.0 241.9 239.2 244.9 253.8 266.9 275.0 284.4 295.2 308.7 314.9 327.1 340.9 354.1 370.9 385.9 403.2 422.6 437.6 459.0 477.5 509.6 542.0 567.9 591.0 633.1 673.6 707.6 743.9 776.1 813.8 875.1 897.0 961.2 1,016.0 1,028.1 1,067.2 1,108.0 1,140.4

508.6 500.1 493.2 488.3 483.0 479.4 470.4 463.4 458.1 451.7 441.5 434.4 434.6 427.9 417.7 408.1 398.7 389.8 381.4 373.2 363.8 355.6 352.5 348.1 341.5 335.3 329.4 321.8 313.8 303.9 295.2 284.8 290.5 282.6 271.8 270.2 268.9 262.2 252.5

High School College 554.5 571.3 579.6 593.0 607.2 624.9 636.5 649.3 666.1 686.3 701.8 723.6 738.4 753.6 773.7 790.6 808.8 824.7 849.1 871.8 890.0 914.8 934.7 949.6 961.5 980.2 1,000.5 1,015.5 1,025.4 1,034.8 1,046. I 1,057.5 1,073.2 1,093.0 1,086.6 1,095.4 1,090.4 1,098.6 1,099.7

*The grades 1-8 column includes persons who have completed less than the first grade.

191.4 194.7 200.2 206.4 212.8 220.2 224.5 229.6 235.6 242.6 248.0 255.4 263.5 272.5 288.5 299.8 31 1.7 325.4 340.6 354.6 370.3 387.8 407.9 433.4 453.2 473.0 498.1 522.4 547.7 573.6 598.0 625.0 628.1 646.3 703.3 725. I 760.0 782.7 812.4

Dale W. Jorgenson and Barbara M. Fraumeni

328 Table 8.6

Investment in Formal Education by Sex and Level of Environment, 1947-86 (billions of current dollars) Male

Female

Year

Total

Grades 1-8

High School

College

Grades 1-8

High School

College

1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 I962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

181.1 214.3 231.6 257.2 293.2 323.4 372.7 390.2 433.8 495.8 570.8 603.5 616.9 650.9 678.3 695.3 738.6 768.2 832.3 903.5 985.3 1,039.4 1,252.5 1,383.7 1,458.4 1,521.7 1,594.5 1,719.2 1,792.7 1,825.5 1,883.6 1,991.9 2,113.1 2,346.6 2,515.9 2,834.9 2,975.4 3.171.2 3,359.3 3,779.0

66.7 71 .O 76.2 84.7 98.5 108.2 123.2 126.4 139.4 158.8 182.8 185.2 177.9 185.8 185.4 184.1 186.4 185.3 193.3 199.8 209.6 208.1 232. I 250.3 252.5 261.8 265.5 279.4 284.5 281.0 278.7 282.3 283.1 311.1 355.0 342.5 360.3 384. I 413.4 461.3

32.0 42.6 44.4 46.9 51.2 55.0 62.1 63.9 69.5 78.4 90.7 93.1 93.3 101.4 107.4 108.1 113.7 118.3 129.4 140.4 151.0 154.1 176.9 201.9 210.0 221.9 237.1 261.5 276.0 282.2 294.7 314.5 330.5 360.9 374.9 421.1 419.2 448.9 487.6 546.5

21.4 30.5 33.6 39.7 43.0 47.0 52.7 53.5 57.9 65. I 73.7 74.6 72.1 77.1 81.4 84.3 90.3 94.1 101.1 107.9 125.9 140.5 169.0 192.4 206.9 234.9 265.2 310.7 342.7 367.6 405.4 464.7 525.3 584.3 601.0 725.1 800.3 885.0 981.2 1,096.6

33.3 35.5 39.6 44.4 53.7 61.1 73.2 79.8 92.2 107.7 125.0 139.8 151.1 157.6 163.8 169.7 181.5 190.3 207.1 228.4 244.7 259. I 323.4 352.6 374.2 375.4 373.1 373.4 360.5 340.5 319.9 307.4 304.2 338.3 383.7 421.6 422. I 433.7 435.2 510.6

18.0 21.7 23. I 24.4 27.4 30.2 35.3 38.2 42.7 48.9 56.4 63.3 72.0 78.4 84.4 86.0 95.5 104.2 118.0 134.2 143.0 152.5 191.4 210.3 227.1 231.6 241.5 256.4 262.8 263.1 263.4 267.6 275.2 304.9 341 .O 373.5 371.2 384.3 382.6 442.6

9.7 13.0 14.8 17.2 19.3 22.0 26.2 28.4 32.2 36.8 42.1 47.3 50.5 50.6 55.8 63.0 71.2 76.0 83.4 92.7 111.0 125.1 159.6 176.3 187.8 196.0 212.1 237.8 266.2 291.2 321.5 355.5 394.8 447.1 480.3 551.1 602.3 635.2 659.4 721.4

329

Output of the Education Sector

Table 8.7

Investment in Formal Education by Sex and Level of Enrollment, 1948-86 (billions of constant dollars) Male

Female

Year

Total

Grades 1-8

High School

College

Grades 1-8

High School

College

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

1,073.9 1,087.6 1,105.7 1,128.2 1,139.5 1,153.0 1,168.0 1,204.2 1,241.4 1,281.3 1,320.2 1,360.6 1,403.2 1,481.5 1,566.8 1,642.8 1,714.2 1,788.9 1,868.9 1,967.9 2,057.8 2,129.5 2,200.6 2,246.1 2,292.1 2,363.0 2,430.7 2,504.4 2,567.2 2,627.7 2,652.7 2,707.1 2.742.6 2,720.7 2,834.9 2,861.7 2,859.4 2,833.5 2,834.6

272.7 278.6 287.3 300.5 302.8 305.3 306.9 316.3 324.8 333.2 339.3 342.2 345.2 351.0 364.9 372.8 380.1 387.6 395.8 403.5 41 1.8 417.5 421.3 422.2 423.9 422.5 418.5 413.6 406.8 398.2 384.6 378.0 368.9 359.2 342.5 337.2 333.5 331.4 331.2

175.0 174.5 173.1 171.7 171.4 171.5 174.0 176.8 180.6 185.6 192.2 202.6 213.9 234.4 246.3 261.0 277.7 298.3 320.2 335.3 346.7 360.9 378.1 382.2 385.8 399.9 412.5 421.4 429.0 434.2 433.3 431.5 428.9 418.8 421.1 402.4 394.7 394.1 394.5

131.9 133.6 134.9 132.0 135.1 138.2 141.5 146.3 151.8 158.0 165.2 172.7 181.0 195.8 213.4 230.7 245.1 256.3 269.0 301.6 336.5 358.1 379.3 401.1 429.9 456.5 484.7 509.6 533.3 556.9 581.2 606.8 623.9 640.4 725.1 767.6 780.5 785.4 788.3

262.8 270.9 281.9 297.9 301.8 305.5 308.0 320.7 331.8 342.4 350.2 354.9 360.8 369.6 386.8 396.0 406.0 417.9 430.4 442.3 452.8 462.7 471.3 476.8 480.8 480.5 476.7 471.6 464.9 457.1 450.6 445.7 437.8 429.1 421.6 416.4 410.6 403.8 399.4

144.4 143.3 141.9 142.0 141.5 142.7 145.2 148.4 152.8 158.1 164.5 175.4 185.2 195.5 201.3 214.4 229.3 246.1 263.1 268.8 276.4 286.1 297.6 306.7 308.5 323.7 337.3 349.6 360.4 369.8 379.0 385.0 389.0 380.5 373.5 365.0 361.7 350.1 349.5

69.6 69.9 70.4 70.3 72.8 75.4 77.9 81.2 85.1 89.5 94.3 99.1 104.3 123.1 142.6 157.9 167.6 175.2 184.1 214.1 235.4 248.6 260.3 266.5 277.0 296.2 319.6 357.1 391.0 427.1 437.8 468.3 497.9 494.5 551.1 572.9 578. I 566.9 569.4

330

Dale W. Jorgenson and Barbara M. Fraumeni

Table 8.8

Investment in Formal Education by Sex and Level of Enrollment, 1947-86 (billions of current dollars) Male

Year

Total

Grades 1-8

High School

Female College

Grades 1-8

Real Income Growth Rate = 2%: Discount Rate 1947 1953 1957 1960 1966 1969 1973 1979 1986

259.0 529.3 811.1 918.0 1,262.7 1,743.2 2,209.6 2,845.9 5,030.2

99.6 186.1 276.3 278.2 295.8 346.9 399.8 420.4 682.6

45.3 86.5 125.9 139.8 190.0 238.9 320.2 443.5 726.6

28.4 68.6 95.6 99.4 138.8 217.0 340.7 676.2 1,387.7

48.4 106.7 182.9 230.0 336.3 475.5 550.0 442.5 744.3

High School =

College

4% 24.3 47.2 75.6 105.0 181.5 258.5 326.5 362.7 582.0

13.0 34.3 54.9 65.6 120.3 206.3 272.5 500.6 907.0

Real Income Growth Rate = 1%; Discount Rate = 6% 1947 1953 1957 1960 1966 1969 1973 1979 1986

116.3 241.9 370.4 427.2 599.9 835.5 1,070.2 1,474.3 2,676.9

40.4 73.0 108.0 111.7 122.1 139.2 157.5 172.6 281.9

20.7 41.0 60.3 68.2 97.0 122.5 164.5 231.9 387.6

15.1 38.2 53.7 56.6 79.5 124.7 195.3 385.6 823.5

20.9 45.9 78.0 98.6 140.8 199.4 229.0 190.3 318.1

12.6 24.9 39.8 55.2 93.0 132.7 167.1 197.7 318.6

6.8 18.9 30.6 36.9 67.6 117.1 156.8 296.2 547.3

20.5 40.2 64.3 89.3 153.0 218.0 274.0 309.5 500.2

11.1 29.9 48.0 57.6 105.6 181.9 240.8 446.6 818.6

15.5 30.4 48.6 67.4 115.4 164.8 208.9 241 .O 385.0

8.3 22.5 36.3 43.5 79.8 137.4 183.5 343.0 624.2

8 Hours of Maintenance per Day 1947 206.2 1953 423.2 1957 648.1 1960 739.1 1966 1,026.7 1969 1,422.0 1973 1,802.5 1979 2,373.4 1986 4,269.2

76.0 139.6 207.0 210.5 226.4 262.4 298.4 315.9 518.5

36.4 70.4 102.8 114.9 159.0 199.7 266.5 368.8 614.8

24.3 59.8 83.5 87.3 122.2 191.2 299.1 590.0 1.238.7

37.9 83.4 142.4 179.6 260.5 368.8 423.7 342.6 578.5

12 Hours of Maintenance per Day 1947 1953 1957 1960 1966 1969 1973 1979 1986

156.0 322. I 493.5 562.7 780.3 1,083.1 1,386.4 1,852.9 3,288.8

57.5 106.7 158.5 161.2 173.3 201.9 232.6 250.3 404.2

27.5 53.7 78.6 88.0 121.8 154.1 207.8 292.2 478.3

18.5 45.7 64.0 67.0 93.7 146.8 231.2 460.7 954.4

28.7 63 .O 107.6 135.5 196.3 278.1 322.4 265.8 442.8

Output of the Education Sector

331 Table 8.9

Investment per Student by Sex and Level of Enrollment, 1947-86 (thousands of current dollars) Male

Female

Year

Total

Grades 1-8

High School

College

Grades 1-8

High School

College

1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

8.6 10.5 11.1 12.1 13.4 14.7 16.8 17.4 18.8 20.9 23.4 24.1 24.0 24.7 24.8 24.4 25.0 25.0 26.1 27.2 28.7 29.2 34.1 36.6 37.9 38.9 39.8 41.9 42.9 43.1 43.9 46.0 48.4 53.7 57.9 64.5 68.2 72.5 77.8 87.1

8.8 10.2 10.7 11.5 12.8 14.0 15.8 16.1 17.2 19.1 21.5 21.5 20.4 21.1 20.7 20.0 19.9 19.4 19.9 20.1 20.8 20.4 22.5 24.1 24.3 25.2 25.6 21.3 28.2 28.3 28.6 29.5 30.0 33.7 37.1 39.8 42.5 45.7 49.5 55.5

11.6 13.1 13.7 14.5 15.9 17.2 19.4 19.8 21.2 23.5 26.5 26.2 25.0 26.0 25.3 23.9 23.7 23.1 23.7 24.2 24.8 24.3 26.6 28.9 29.9 31.3 32.3 34.6 35.8 36.0 37.2 39.7 41.9 46.3 49.3 55.1 57.4 62.2 68.0 75.6

16.4 19.0 20.5 23.9 26.2 27.9 30.5 30.2 31.7 34.4 37.4 36.4 33.8 34.5 34.6 33.3 33.2 32.4 32.8 33.0 35.2 35.7 39.8 42.5 43.3 46.5 50.0 55.8 59.2 61.2 65.2 72.3 79.9 87.1 85.0 92.5 99.9 107.6 118.8 131.8

5.2 6.6 7.2 7.7 8.9 10.0 11.9 12.9 14.4 16.3 18.3 20.1 21.4 21.9 22.7 23 .O 24.2 24.8 26.4 28.6 30.0 31.3 38.6 41.8 44.1 44.0 43.8 44.1 43.0 41.3 39.5 38.5 38.6 43.8 50.8 58.2 58.7 61.4 62.6 74.1

7.7 8-8 9.4 10.0 11.1 12.4 14.4 15.4 16.9 18.9 21.1 22.6 24.3 25.3 26.0 25.6 26.9 27.7 29.6 31.9 33.1 34.1 41.4 43.9 46.7 47.2 46.1 46.1 44.8 43 .O 41.4 40.5 40.6 44.2 50.9 56.4 57.1 58.8 60.5 70.0

15.3 16.4 18.5 21.2 23.8 26.2 30.1 31.6 34.5 37.9 41.6 44.6 45.5 43.5 42.2 41.5 43.3 43.2 44.8 47.1 51.1 52.5 62.5 64.9 64.2 62.9 63.7 66.0 68.2 69.0 70.8 73.6 77.7 83.4 87.0 92.2 98.4 100.9 107.9 113.9

332

Dale W. Jorgenson and Barbara M. Fraumeni

Table 8.10

Investment per Student by Sex and Level of Enrollment, Market and Nonmarket Labor Activities, 1948-86 (thousandsof constant dollars) Male

Female

Year

Total

Grades 1-8

High School

College

Grades 1-8

High School

College

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

52.4 52.2 51.9 51.6 51.8 52.0 52.1 52.2 52.4 52.5 52.7 52.9 53.1 54.2 55.0 55.5 55.8 56.0 56.4 57.3 57.8 57.9 58.2 58.4 58.6 59.0 59.3 60.0 60.6 61.3 61.2 62.1 62.8 62.6 64.5 65.6 65.4 65.6 65.3

39.1 39.0 38.9 39.2 39.2 39.2 39.1 39.0 39.1 39.2 39.3 39.3 39.1 39.2 39.7 39.8 39.8 39.8 39.9 40.0 40.3 40.4 40.6 40.7 40.8 40.8 40.9 40.9 40.9 40.9 40.2 40.1 40.0 39.8 39.8 39.8 39.7 39.7 39.8

53.7 53.7 53.6 53.3 53.5 53.5 53.8 53.9 54.1 54.2 54.2 54.4 54.8 55.2 54.6 54.3 54.3 54.7 55.1 55.1 54.6 54.3 54.2 54.4 54.5 54.5 54.5 54.6 54.7 54.8 54.6 54.7 55.0 55.1 55.1 55.1 54.7 55.0 54.6

82.1 81.6 81.2 80.5 80.3 80.0 79.8 80.0 80.2 80.3 80.6 80.8 81.1 83.2 84.4 84.9 84.4 83.3 82.4 84.3 85.6 84.4 83.8 83.9 85.0 86.1 87.0 88. I 88.8 89.6 90.5 92.3 93.0 90.5 92.5 95.8 94.9 95.1 94.7

48.9 48.9 48.9 49.3 49.5 49.7 49.7 49.9 50. I 50.2 50.3 50.3 50.2 51.3 52.5 52.7 53.0 53.3 53.9 54.3 54.8 55.3 55.9 56.2 56.4 56.4 56.3 56.3 56.4 56.4 56.5 56.6 56.7 56.8 58.2 57.9 58. I 58.1 58.0

58.4 58.4 58.2 57.6 57.9 58.3 58.6 58.9 59.1 59. I 58.8 59.3 59.7 60.1 60.0 60.4 61.0 61.7 62.6 62.2 61.9 61.8 62.2 63.1 62.9 61.8 60.6 59.6 58.8 58.1 57.3 56.8 56.4 56.8 56.4 56.1 55.3 55.3 55.3

88. I 87.5 86.9 86.5 86.5 86.7 86.7 87. I 87.7 88.4 88.9 89.3 89.6 93.1 93.9 96.0 95.3 94. I 93.5 98.6 98.9 97.3 95.8 91.1 88.8 88.9 88.7 91.4 92.7 94.2 90.6 92.1 92.9 89.6 92.2 93.6 91.8 92.8 89.9

333

Output of the Education Sector

Table 8.11

Percentage of Investment Based on Market Activities to Total Educational Investment, 1947-86 Male

Year

Total

Grades 1-8

High School

Female College

Grades 1-8

High School

College

28.6 27.8 27.5 28.9 30.7 28.3 36.0 37.4 40.4

30.1 29.9 29.8 29.0 31.2 25.6 33.1 32.4 33.6

~~

1947 1953 1957 1960 1966 1969 1973 1979 1986

33.7 40.5 38.5 37.7 35.7 31.1 38.7 42.6 41.2

36.4 46.8 44.7 44.5 38.8 32.4 40.6 46.0 41.8

36.2 48.5 45.7 46.9 42.1 39.5

46.1 49.9 46.0

32.3 45.2 42.8 42.6 40.6 35.4 39.6 47.3 43.9

28.8 29.4 29.0 29.2 30.8 27.2 35.8 37.6 40.9

334

Dale W. Jorgenson and Barbara M. Fraumeni

Table 8.12

Human Wealth by Sex and Educational Attainment, 1947-86 (billions of current dollars) Male

Year

Total

Grades 1-8*

1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

15,082 16,081 16,957 18,055 19,178 20,513 22,433 23,176 24,805 27,105 29,570 30,492 32,457 34,672 36,788 38,067 41,145 44,096 48,087 51,788 56,099 60,327 68,923 75,554 80,601 88,245 96,651 106,010 114,568 121,760 133,148 146,260 159,836 171,254 186,814 198,951 2 10,240 225,320 242,7 13 268,567

4,780 5,052 5,323 5,660 5,989 6,403 6,964 7,237 7,779 8,515 9,294 9,527 9,923 10,521 11,042 11,249 1 1,932 12,524 13,402 14,062 14,901 15,606 17,270 18,487 19,271 20,642 21,953 23,271 24,289 24,939 26,300 27,992 29,515 30,877 32,980 34,558 35,859 28,203 40,996 44,683

Female

High School College Grades 1-8* 3,133 3,344 3,532 3,780 3,998 4,270 4,631 4,790 5,131 5,626 6,168 6,383 6,771 7,214 7,573 7,855 8,535 9,198 10,129 10,911 11,900 12,876 14,733 16,544 17,827 19,855 2 1,993 24,296 26,328 28,074 30,733 33,898 36,836 39,3 17 43,442 45,827 47,239 50,255 53,548 58,966

1,382 1,517 1,646 1,832 2,009 2,193 2,443 2,537 2,749 3,066 3,431 3,505 3,611 3,851 4,090 4,273 4,647 5,014 5,560 6,107 6,857 7,555 8,884 10,266 1 1,353 13,011 14,818 16,988 18,964 20,777 23,464 26,888 30,408 33,759 36,355 39,518 43,472 47,162 52,120 58,215

2,842 3,000 3,129 3,271 3,464 3,693 4,056 4,174 4,441 4,803 5,160 5,348 5,824 6,258 6,742 6,967 7,505 8,032 8,668 9,309 9,905 10,511 11,884 12,427 12,836 13,415 14,189 15,051 15,886 16,455 17,517 18,547 19,785 20,492 22,293 23,707 24,641 26,036 27,493 30,229

High School College 2,232 2,388 2,490 2,607 2,743 2,905 3,171 3,234 3,418 3,688 3,978 4,122 4,582 4,969 5,325 5,587 6,160 6,734 7,427 8,132 8,874 9,696 1 1,220 12,271 13,114 14,337 15,794 17,389 18,965 20,330 22,468 24,605 27,024 28,762 3 I ,892 33,724 34,793 37,222 38,993 43,206

*The grades 1-8 column includes persons who have completed less than the first grade

71 1 178 835 902 972 1,046 1,167 1,202 1,285 1,404 1,536 1,604 1,745 1,857 2,014 2,134 2,363 2,591 2,899 3,264 3,658 4,081 4,928 5,558 6,196 6,984 7,901 9,012 10,135 11,182 12,664 14,327 16,265 18,043 19,850 21,615 24,233 26,439 29,559 33,265

335

Output of the Education Sector

Table 8.13

Human Wealth by Sex and Educational Attainment, 1948-86 (billions of constant dollars) Male

Year

Total

Grade 1-8*

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

112,520 114,719 116,858 119,151 121,505 123,879 126,564 129,203 131,854 134,705 137,904 140,923 143,941 147,149 150,478 153,503 156,537 159,346 162,066 164,712 167,288 169,624 172,301 175,200 177,591 179,806 181,960 184,167 186,354 188,420 190,380 192,372 194,591 196,836 198,951 201,208 203,319 205,415 207,234

35,318 35,758 36,184 36,733 37,326 37,942 38,653 39,355 40,021 40,714 41,371 41,934 42,478 43,073 43,405 43,488 43,549 43,491 43,307 43,016 42,621 42,097 41,596 41,341 40,795 40,111 39,351 38,618 37,829 37,047 36,269 35,540 34,970 34,834 34,558 34,346 34,497 34,622 34,505

Female

High School College 22,297 22,756 23,213 23,641 24,075 24,522 25,041 25,557 26,115 26,752 27,575 28,432 29,264 29,883 30,762 3 1,720 32,707 33,685 34,554 35,390 36,287 37,211 38,248 39,044 39,805 40,561 41,282 41,958 42,610 43,168 43,636 44,042 44,363 45,569 45,827 45,299 45,589 45,538 45,995

9,625 10,021 10,410 10,815 11,199 11,572 1 1,962 12,345 12,729 13,120 13,539 13,963 14,388 15,071 15,841 16,604 17,372 18,155 19.1 17 20,177 21,247 22,283 23,408 24,630 25,882 27,181 28,542 29,955 3 1,436 32,959 34,490 36,068 37,704 38,070 39,518 41,646 42,539 43,717 44,464

Grade 1-8* 23,618 23,93 I 24,227 24,621 25,045 25,474 25,959 26,433 26,888 27,347 27,822 28,172 28,562 29,021 29,260 29,346 29,424 29,432 29,353 29,195 28,968 28,677 28,356 28,190 27,816 27,343 26,826 26,326 25,789 25,257 24,731 24,244 23,870 23,799 23,707 23,547 23,480 23,483 23,390

High School College 16,853 17,233 17,591 17,932 18,290 18,663 19,079 19,492 19,926 20,437 2 1,088 21,740 22,376 22,813 23,463 24,166 24,874 25,565 26,177 26,802 27,483 28,149 28,908 29,464 30,003 30,521 3 1,023 3 1,495 3 1,947 32,337 32,661 32,927 33,136 33,732 33,724 33,225 33,441 33,006 33,199

*The grades 1-8 column includes persons who have completed less than the first grade.

5,390 5,578 5,757 5,905 6,045 6,176 6,313 6,451 6,583 6.73 1 6,894 7,056 7,232 7,596 8,008 8,417 8,827 9,241 9,777 10,354 10,935 1 1,490 12,099 12,827 13,572 14,338 15,144 15,983 16,871 17,753 18,663 19,589 20,554 20,838 21,615 23,133 23,758 25,023 25,651

336

Dale W. Jorgenson and Barbara M. Fraumeni

Table 8.14

Human Wealth per Person by Sex and Educational Attainment, 1947-86 (thousands of current dollars) Male

Year

Total

Grades 1-8*

1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

101.1 106.0 110.0 115.2 120.4 126.7 136.4 138.5 145.7 156.5 167.8 170.2 178.3 186.7 194.9 198.6 211.6 223.7 241 .O 256.6 275.0 292.8 331.2 359.4 379.5 411.9 447.9 487.8 523.1 551.9 598.7 652.3 706.8 750.7 810.4 853.7 893.4 948.4 1,011.5 1,108.3

102.3 107.3 112.2 118.4 124.0 131.2 141.3 145.3 154.4 167.3 180.8 184.1 190.6 200.0 208.9 212.9 226.7 239.1 257.6 273.0 292.8 311.2 349.9 380.8 40 1.4 437.1 474.2 513.6 547.2 574.4 618.7 672.3 723.0 769.5 825.9 873.3 911.5 971.9 1,038.0 1,137.5

Female

High School College Grades 1-8*

148.4 154.7 159.7 167.0 173.1 181.2 192.6 195.0 204.3 218.9 234.1 235.9 243.3 251.2 257.6 259.7 273.8 286.4 306.5 321.6 341.9 360.5 401.7 439.0 464.8 508.5 553.6 601.4 641.2 673.3 726.6 790.6 848.6 896.5 969.0 1,012.8 1,048.9 1,103.3 1,173.2 1,276.6

195.7 206.2 215.0 230.4 243.5 256.9 277.0 278.5 292.3 315.8 342.6 339.9 339.8 350.1 355.3 354.2 368.3 380.6 404.5 423.2 452.0 475.0 533.9 589.8 620.7 677.1 734.9 803.0 854.6 893.2 962.5 1,053.4 1,138.8 1,210.1 1,283.9 1,339.0 1,398.1 1,484.6 1,593.4 1,734.0

65.6 68.5 70.8 73.3 76.7 80.7 87.6 89.0 93.6 100.0 106.2 109.0 118.0 125.3 134.0 138.4 149.3 160.3 173.8 188.1 202.2 217.4 249.1 264.3 275.7 292.5 315.1 341 .O 366.9 388.1 421.4 455.0 494.5 520.4 562.0 602.7 634.6 672.4 711.6 785.4

High School College

90.7 94.6 96.2 98.3 101.2 104.8 111.8 111.4 114.9 121.o 127.1 128.0 138.3 145.2 152.0 155.0 165.9 176.2 188.9 201.5 214.1 227.6 256.4 272.8 286.4 307.3 332.6 359.9 386.0 407.2 443.1 478.2 518.4 545.5 597.9 624.0 647.7 687.3 722.7 794.0

*The grades 1-8 column includes persons who have completed less than the first grade

111.8 117.7 121.7 126.9 133.0 139.2 151.5 152.1 158.5 168.9 180.0 183.3 194.4 200.6 208.1 210.6 223.1 234.4 251.5 269.6 287.7 306.2 353.6 381.1 402.5 430.1 462.1 500.7 535.1 561.2 605.1 65 I .9 705.8 747.6 812.1 853.6 887.8 942.2 1,001.8 1,095.3

337

Output of the Education Sector

Table 8.15

Human Wealth per Person by Sex and Educational Attainment, 1948-86 (thousands of constant dollars) Male

Female

Year

Total

Grades 1-8*

High School

College

Grades 1-8*

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986

741.8 743.9 745.8 748.2 750.4 752.9 756.3 759.0 761.4 764.4 769.7 774.3 775.0 779.4 785.0 789.5 794.2 798.6 803.0 807.3 811.8 815.1 819.6 824.9 828.9 833.2 837.3 840.9 844.6 847.3 849.0 850.7 853.0 853.9 853.7 855.1 855.8 856.1 855.2

750.4 753.5 745.6 760.6 764.9 769.9 775.8 781.3 786.3 791.9 799.3 805.3 807.6 814.8 821.4 826.2 831.2 836.1 840.8 845.2 849.9 853.0 856.8 861.1 863.9 866.4 868.5 870.0 871.3 871.6 871.1 870.6 871.5 872.3 873.2 873.0 877.6 876.6 878.4

1,03 1.2 1,028.6 1,025.3 1,023.4 1,021.3 1,019.9 1,019.3 1,017.7 1,015.9 1,015.3 1,019.0 1,021.8 1,018.9 1,016.2 1.016.9 1,017.5 1,018.5 1,019.2 1,018.3 1,016.8 1,016.0 1,014.4 1,014.9 1,017.9 1,019.4 1,020.9 1,021.8 I ,02 1.9 1,02 1.9 1,020.5 1,017.7 1,014.5 1,011.6 1,016.5 1,012.8 1,005.8 1,000.9 997.7 995.8

1,307.4 1,308.7 1,308.9 1,310.9 1.31 1.1 1,312.0 1,312.7 1,312.3 1,310.8 1,310.0 1,312.7 1,313.9 1,308.1 1,309.0 1,313.0 1,315.7 1,318.3 1,320.5 1,324.6 1,330.0 1,335.7 1,339.0 1,344.7 1,346.5 1,346.9 1,348.0 1,349.1 1,350.0 1,351.4 1,35 1.9 1,351.2 1,350.8 1,351.5 1,344.5 1,339.0 1,339.4 1,339.0 1,336.5 1,324.4

539.5 541 .o 542.6 544.8 547.4 550.2 553.7 556.8 559.8 562.7 567.0 570.6 571.9 576.9 581.1 583.9 587.1 590.2 593.2 596.0 599.0 601.0 603.2 605.5 606.4 607.2 607.8 608.0 608.2 607.6 606.6 605.9 606.2

600.0 602.6 606.4 606.4 607.8 607.7

High School College 667.5 665.8 663.5 661.5 659.5 657.9 656.9 655.3 653.4 652.7 654.7 656.1 653.9 651.2 651.1 650.8 650.6 650.1 648.4 646.5 645.2 643.3 642.7 643.4 643.1 642.7 642.1 641.0 639.8 637.8 634.8 631.6 628.4 632.4 624.0 618.5 617.5 611.7 610.0

*The grades 1-8 column includes persons who have completed less than the first grade.

815.0 812.4 809.5 807.4 804.6 801.4 798.4 795.4 791.2 788.3 787.5 785.8 781.2 784.6 790.1 794.3 798.3 801.5 807.6 814.1 820.3 824.2 829.6 833.2 835.9 838.5 841.3 843.8 846.6 848.2 849.1 850.1 851.6 852.6 853.5 847.5 846.6 848.0 844.6

Percentage of Human Wealth Based on Market Labor Activities to Total Human Wealth by Sex and Educational Attainment, 1947-86

Table 8.16

Female

Male Year

Total

Grades 1-8*

High School

College

Grades 1-8*

High School

College

1947 1953 1957 1960 1966 1969 1973 1979 1986

29.5 30.8 31.6 31.9 32.3 32.4 32.4 32.5 32.5

38.4 40.9 41.1 41.1 41 .O 40.9 40.5 40.1 39.0

40.5 41.8 42.2 42.3 42.7 42.7 42.2 41.4 39.2

39.7 43.5 43.9 43.9 43.9 43.9 43.1 42.2 40.1

13.3 14.6 15.2 15.6 16.5 17.0 17.5 18.9 20.6

13.4 14.6 15.2 15.6 16.7 17.3 17.9 19.2 20.5

16.9 19.2 19.3 19.5 20.2 20.5 20.7 21.9 23.3

*The grades 1-8 column includes persons who have completed less than the first grade

Table 8.17

Comparison with Other Results

A. Value of Nonmarket Activities, Selected Years, (billions of current dollars) Current Dollars Year

J-F

Nordhaus-Tobin

Ratio

1954 1958 1965

512.7 638.7 953.7

637.0 794.6 1,096.9

0.805 0.804 0.869

B. Private National Human Wealth, 1948-69 (billions of dollars) Current Dollars

1958 Dollars

Year

J-F

Kendrick

Ratio

J-F

Kendrick

Ratio

1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969

16,081.4 16,957.9 18,055.8 19,178.6 2 0 3 13.8 22,433.3 23,176.5 24,805.7 27, I 05.4 29,570.6 30,492.0 32,457.5 34,672.6 36,788.8 38,068.0 41,145.1 44,096.6 48,087.5 51,788.3 56,099.5 60,327.7 68,923.4

908.8 938.9 991.3 1,097.7 1,172.6 1,236.8 1,294.4 1,364.2 1,462.7 1,576.8 1,682.6 1,786.9 1,901.4 2,012.8 2,137.4 2,273.0 2,423.9 2,594.4 2,818.7 3,049.7 3,344.4 3,699.9

17.70 18.06 18.21 17.47 17.49 18.14 17.91 18.18 18.53 18.75 18.12 18.16 18.24 18.28 17.81 18.10 18.19 18.54 18.37 18.40 18.04 18.63

24,505.0 25,156. I 25,598.1 26,036.4 26,715.6 27,3 10.1 27,911.5 28,494.8 29,190.4 29,837.0 30,492.0 3 1,203.8 3 1,961.6 32,701.2 33,440.1 34,262.1 34,903.3 35,667.5 36,365.5 36,959.7 37,641.8 38,215.0

1,206.3 1,242.9 1,280.5 1,322.2 1,366.9 1,413.3 1,460.0 1,509.9 1,565.6 1,623.7 1,682.6 1,744.7 1,615.1 1,888.4 1,962.5 2,041.9 2,126.8 2,218.8 2,323.4 2,434.0 2,550.1 2,674.4

20.31 20.24 19.99 19.69 19.54 19.32 19.12 18.87 18.64 18.38 18.12 17.88 19.79 17.32 17.04 16.78 16.41 16.08 15.65 15.18 14.76 14.29

339

Output of the Education Sector

Comment

Michael Rothschild

Comparative advantage dictates a focus on theory rather than data. I applaud Dale W. Jorgenson and Barbara M. Fraumeni’s general framework. The most important output of education is human capital. The value of human capital is the discounted value of its services. I quarrel with one important modeling choice Jorgenson and Fraumeni make. The authors assert that human capital raises the productivity of time spent at leisure by the same amount that it does time spent working. Little evidence is offered for this assertion. My empirical work, which consists of introspection, does not resolve the issue. I would like to believe that Ph.D.3 make better lovers; I can believe that education is complementary to such leisure-time activities as operating a VCR; I doubt that within the audience at a football game (or an opera) the quality of the experience varies directly with the market wage. I Noting that the market wage is the opportunity cost of leisure does not resolve this issue. Consider two simple variants of the standard time-allocation problem. Utility derives from goods, g, and leisure or recreation, r, and is calibrated by the utility function U( g, r ) . Goods are bought with wage income paid at the rate of w per unit of time worked. The individual must divide his or her time between working and leisure. Choose units so that the total amount of time available for working and leisure is one. Let h (for hours) be the fraction of time spent working. Thus (1 - h) is the fraction of time spent at leisure. As human capital increases so does the wage per unit time. For simplicity measure human capital in units of wage income. Thus the consumer’s budget constraint is g = hw, or (1)

g

+ (1 - h)w = w ; O l h l l .

If human capital does not augment the ability to enjoy leisure, then the amount of leisure consumed is just the amount of time spent not working or

(2)

r=(l -h)

If human capital does augment the ability to enjoy leisure, then (3)

r

=

(1 - h)f(w),

wherefi) is some increasing function. Jorgenson and Fraumeni focus on the case wherefi) is the identity function. Thus in their world

(4)

r = (1 - h)w.

Michael Rothschild is professor of economics and dean of the division of social sciences at the University of California, San Diego, and a research associate of the National Bureau of Economic Research. 1. This issue is not new; for one earlier discussion see Donald V. T. Bear, The University as a Multi-Product Firm, in Eficiency in Universities: The La Paz Papers, ed. Keith G . Lumsden (New York: Elsevier, 1974), 85 n.6. Bear attributes the distinction made here to comments by Arrow and Brainard.

340

Dale W. Jorgenson and Barbara M. Fraumeni

The time-allocation problem is then a matter of choosing the fraction h, which maximizes U( g, r ) subject to (1) and either ( 2 ) or (4). These two problems share the budget constraint (1). The right-hand side of equation (1) is often called full income. Clearly full income is a linear function of the level of human capital. If human capital doubles so does full income. I understand national income accounting as a kind of cardinal welfare economics. Full income is an accounting concept. The question then is, Under what conditions is full income a reasonable measure of welfare? In order for the Jorgenson-Fraumeni procedure (which is linear) to have a chance, it is necessary to suppose that the utility function, U ( g , r ) , exhibits constant returns to scale; this form of the hypothesis that the marginal utility of income is constant is appropriate for this example. Define (5)

V(w) = max U ( g , r ) ,subject to (1).

Then V(w) is the value of human capital and is a good measure of national income. Consider the particular utility function (6)

U ( g , r)

=

U(gar"-al).

The utility maximizing hours of work, h', is equal to ( 1 - a),and the utility maximizing g* = aw,when either (2) or (4) holds. If (2) describes the leisure technology, total utility is

V,(w)

=

w"K,

where K = a" (1 - a)('-"). National income is not a linear function of w. In the Jorgenson-Fraumeni case total utility is VJw)

=

wK.

This argument generalizes somewhat. As long as the utility function is homothetic, the choice of hours of work is independent of the level of human capital. This pattern is true even if there are taxes on wage income. If 7 is the tax rate, then in the Jorgenson-Fraumeni world the allocation of time problem reduces to the problem of choosing h to maximize U[h(1 - T)W, (1 - h)w]. Because Up,*) is homothetic, the optimal h is independent of w. Because the utility function is homogeneous of degree one, the indirect utility function VJw) always has the form VJF(w)= wK, where K is some constant. If human capital does not increase the efficiency of leisure the allocation of time problem reduces to choosing h to maximize U[h(1 - T ) W , ( 1 - h ) ] .In general the optimal h is be a function of w ; because U ( g , r) is homothetic, hours worked increase as w increases if and only if the elasticity of substitution between g and r is greater than one. Whatever the elasticity of substitution, V(w)is not a linear function of w. It is of some interest to note that V,(w)

341

Output of the Education Sector

can be either concave or convex. In the Cobb-Douglas case we saw that V,(w) is concave. If U ( g , r ) = g + r, then V,(w) = max [ l , w],which is convex. It is natural to ask whether it is possible to test the specification (4).At first sight it would seem that the Jorgenson-Fraumeni hypothesis has the strong implication that, if utility is homothetic, then hours worked are independent of the level of human capital. Unfortunately this prediction of the theory vanishes when the consumer has nonlabor income.

This Page Intentionally Left Blank

9

Measurement of Output and Quality Adjustment in the Day-care Industry Swati Mukerjee and Ann Dryden Witte

This paper studies a growing component of child-care arrangements and focuses on various measures of output and of quality of output of day-care centers. I A methodology for adjusting output measures for changes in quality is developed by the construction of a quality index. This route for measuring quality changes is suggested as an alternative when traditional hedonic techniques are difficult to implement. One such specific case is the day-care market, where defining a market price is both conceptually and empirically difficult. To illustrate the construction, interpretation, and use of the quality-index approach, we utilize both state (Massachusetts) and national (U.S.) data. Because of the unavailability of data at the national level, estimates from state data had to be employed to illustrate the proposed method of quality adjustments. Caution is required in interpreting our quality-adjusted output for the United States for at least two reasons. First, our estimates of the cost of quality improvements are obtained using data for a single state. Second, our measure of quality, although consistent with the literature, is open to question. We present two measures of the output of the day care industry: The first measure is a physical measure of output-the number of children in care. The second is a real dollar measure of output provided to us by Robert Parker of Swati Mukerjee is assistant professor of economics at Bentley College, Waltham, Mass. Ann Dryden Witte is professor of economics at Wellesley College, Wellesley, Mass., and a research associate of the National Bureau of Economic Research. The authors wish to thank Zvi Griliches of Harvard University and National Bureau of Economic Research, Marilyn Manser of the Bureau of Labor Statistics and Robert Coen of Northwestern University for their valuable comments and suggestions. They also thank Sheila Hollowell, Deborah Czuba, and Suzanne Miller of Wellesley College for dedicated research assistance. And last but not least, their thanks go to Claire Loranz of Wellesley College for her unfailing ability to locate always the right data at the right time. 1. According to Sandra Hofferth’s statement at the hearing before the Select Committee on Children, Youth, and Families, from the mid 1960s to the mid-1980s. there was a substantial rise in care at a day-care center or nursery school relative to other forms of care.

343

344

Swati Mukerjee and Ann Dryden Witte

the Bureau of Economic Analysis (BEA). Regardless of the measure considered, the output of the child-care industry, when not adjusted for quality, grew rapidly during the 1970s and 1980s. (See fig. 9.1) The growth rate of output depends critically on the way in which one measures output, with physical output showing a higher rate of growth than constant-dollar measures of output. Adjustments of either measure of output for quality alter both the levels and rates of growth given by either measure. There have been extensive discussions and several studies regarding ways to measure quality of child care. The federal government and the states have sought to regulate the quality of day care. More recently the National Association for the Education of Young Children (NAEYC) has sought to develop standards for the accreditation of child-care programs. The discussions and professional studies on the measurement of the quality of care have yet to be reflected in research seeking to measure the output of child care. In this paper, we seek to combine the work on output measurement in day care with current research on the measurement of quality. We measure quality in a way consistent with state and federal regulations and estimate the valuation of quality using cost functions. We adjust the output of child care to reflect the decline in quality experienced in the 1970s and 1980s. The result is a change in both the level and the growth rate of output, suggesting that adjustment for quality lowers the level of output. This suggests that current national data, at least on children, may overstate both the level and the growth rate of output and highlights the need for some adjustment of the national figures. The structure of the paper is as follows: In section 9.1, we discuss current methods of measuring output in the day-care industry. Availability of data and the construction of a consistent output and employment series at the national level are dealt with in section 9.2. Issues related to the measurement of the quality of care are discussed in section 9.3. In sections 9.4 and 9.5, the methodology for constructing a quality index is developed and applied to adjust national data for quality changes. The final section contains our summary and conclusions.

9.1 Output Measurement One way in which service output may be measured is by examining the value of the inputs. This approach is commonly used in the national accounts to measure government output. However, as more sophisticated approaches become available this methodology is being replaced.* Two other methods of measuring the output of service industries are discussed in the 1iteratu1-e.~ Output may be measured as either D-output or C-output (using the terminology of Bradford, Malt, and Oates 1969). D-output consists of direct services pro2. E.g., BLS has adopted the alternative transactions approach to measuring banking output in lieu of the liquidity approach. See Kendnck (1985), 116. 3. See Searle and Waite (1980). Ross and Burkhead (1974), and Bradford, Malt, and Oates ( 1969).

345

c

Output and Quality in Day Care 2

3

n

--A--Children

c

-(3a al

-.-

BEA Real

1.5 -

a a

w m Od

1-

C

al

74

75

76 77 78 79 80 81 82 83 84 85 86 87 88 Y e a r s

Fig. 9.1 Comparison of children and BEA real dollar output over time in United States (1982 = 100)

duced (e.g., the number of hours a child is cared for). C-output measures the consequences (e.g., the cognitive ability of the child cared for). The latter measure creates the difficulty of not differentiating the output from its consequences (e.g., a child's cognitive ability is also a function of the home environment).4 Measuring D-output, on the other hand, has the advantage of separating the output from an evaluation of its effectiveness. D-output, the direct services produced, has been measured for many service industries. For example, the output of hospitals can be measured as the number of patient days in the hospital, with the caveat that the product has to be a function of the diseases treated. See Newhouse (1970).5 When a firm such as a hospital produces different types of output, an aggregation problem is inevitable. Day-care centers are like hospitals in that both the type of child cared for and the quality of care provided can differ widely. The ultimate output (C-output) of day care has an effect on both the children and the parents. High-quality day care may increase the productivity of parents as they are freed of concern for the welfare of their children.6 The effect 4. See Searle and Waite (1980), 336. 5 . In the case of symphony orchestras (Lange et al. 1985), output measures have been the number of concerts performed or the concert attendance. However, in a study of the sports industry, particularly for cricket, team victories are often used as the measure of output (Schofield 1988). 6. This was suggested informally at the conference by Robert M. Coen of Northwestern University, who was concerned that all current studies focused on the output of day care too narrowly and that the broader perspective was ignored. We agree. In this context Klerman and Leibowitz (1989) have found that higher cost of child care slows the process of women's returning to work.

346

Swati Mukerjee and Ann Dryden Witte

of day care on children has been measured through test scores and such direct measures as the PrescotUSRI Child Observation System. (Stallings and Wilcox 1978). This paper, although recognizing the desirability of measuring the ultimate effect of day care, focuses on the direct output of day-care centers. Our contribution lies in providing and contrasting two different measures of output (dollar and physical) and adjusting the physical measure for changes in the quality of care. Attempts to provide distinct measures of output for the day-care industry began in the mid-1970s. As far as we know, the first attempt to discern the number of children in day care occurred in 1974 (U.S. Department of Commerce 1976). As part of the current population survey, the Census Bureau asked mothers questions concerning day-care arrangements. The census of services reports information for child day-care services beginning in 1977. Prior to 1977, information on day care was incorporated in social service (SIC 83) and could not be examined separately. The census of services distinguishes between for-profit (taxable) and not-for-profit (tax exempt) establishments. According to the 1987 census of services, day-care services are provided in approximately equal amounts by the for-profit and the not-for-profit sectors. The distinction between for-profit and not-for-profit establishments has been dealt with in various ways by researchers.’ Basing its work on the census of services, BEA began producing estimates of the dollar personal consumption expenditures on day care in 1977. In accordance with its standard practice for services, BEA used weekly dollar receipts as its measure of output for profit-making organizations (PMOs) and weekly expenditures as its measure of output for not-for-profit organizations (NPOs). Physical measures of output have been used by researchers studying the nature of production in day-care centers. For example, Ruopp et al. (1979) use the number of full-time equivalent children enrolled at the center.8Mukerjee, Witte, and Hollowell (1990) use the number of hours of child care provided. There is substantial controversy whether the physical or dollar measure pro7. The distinction between PMOs and NPOs has been dealt with in various ways by researchers analyzing the cost of day care. They have categorized centers in various ways. E.g., in Hall’s estimated revenue function (1978), he includes a binary indicating whether a center operates forprofit or as a not-for-profit establishment and another binary indicating whether the center was run by a public or private organization. Ruopp et al. (1979) estimate separate average cost functions for parent-fee and publicly funded centers. In a recent paper, Mukerjee, Witte, and Hollowell (1990) estimate separate cost functions for not-for-profit and for-profit day care centers in Massachusetts. For consistency with current work, we treat PMOS and NPOs as separate entities. To obtain aggregate measures, we simply sum the aggregate level of output for the two types of entities. 8. Hall (1978) claims to estimate cost equations for the day-care centers utilized by participants in the Seattle and Denver income maintenance experiments. His dependent variable is the charge for a 40-hour week of care. Assuming that the relationship between the charge and the quantity of care is proportional, he includes no measure of output among his explanatory variables.

347

Output and Quality in Day Care

vides a superior measure of output. For example, see articles in Inman (1985). The real issue is whether these two measures have different implications regarding the level and growth of output. We will compare the implications of these two measures of output using aggregate data for the United States and a unique data set recently collected in Massachusetts. These data contain detailed information on a random sample of day-care centers. (See app. 2 for a detailed description of the data.) For the Massachusetts sample, physical and dollar measures of output were highly correlated (. 80) for PMOS; for NPOs the correlation between physical and dollar measures was substantially lower (.59). These results suggest that particularly for the large NPO sector the choice of output measure is important and deserves careful consideration.

9.2 The Availability of Data At the national level, it is difficult to obtain a physical measure of output of day care during the 1970s and 1980s. Using a wide range of data sources and extrapolation, we obtain physical measures of output for 1977-87. We use BEA’s measure of real expenditures as our dollar measure of output. (See table 9.1 .) As noted earlier, the two series, physical and dollar, show different rates of growth. As far as we are aware, there has been no previous attempt to obtain a national time series of the physical output level of day-care centers. We attempted to obtain child-care hours, our preferred measure, provided by all day-care centers in the United States but were unable to find the necessary data. We were able to find measures of the number of children in care for selected years from current population survey (CPS) and survey of income and program participation (SIPP) data. There were several obstacles to compiling a consistent series on the number of children in care. The data were drawn from different sources, years, and universes. In 1974 and 1977 the CPS collected the data. In 1974, the data covered all mothers and the three youngest children ages 3-13. In 1977, the gathered data pertained to the two youngest children under five years of employed mothers. Data were available for 1976 from an Abt study of licensed centers (Ruopp et al. 1979), but the adjustments in the final estimates were too numerous for us to make the data comparable. In 1985 and 1986, the data collected by SIPP covered the three youngest children (younger than 15 years) of employed mothers. Adjustments were made to make the universes as comparable as possible. These adjustments are detailed in appendix A. The resulting data points for these years were used to extrapolate for intervening years and to extend the series to include 1987 and 1988. Extrapolation, using the average annual compound rate of growth, yielded the series on children as given in table 9.1. The series is reasonably consistent with other available data on day care.

Aggregate Time Series for Day-care Centers in United States (1974-1988)

Table 9.1

Year 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988

BEA Output (in millions of $)*

... 3,436 3,986 4,270 4,211 4,200 4,242 4,633 5,203 5,724 6,272 6,767 7,339

Annual Average Employment Children (in thousands)

Upper (in thousands)'

Lower (in thousands)'

386 459 546 649 725 810 904 1,010 1,128 1,260 1,407 1,571 1,889 1,959 2,188

104 113 123 135 159 177 187 194 200 216 24 1 26 1 280 303 329

71 77 84 92 108 121 128 133 137 148 165 176 190 206 225

'Nominal figures deflated by price deflator (1982 = 100) obtained from BEA. 'SIC 835, includes day-care centers, nursery schools, and some Head Start programs. $Day-carecenters.

Staff-Child Ratio Upper 0.270 0.246 0.225 0.209 0.219 0.219 0.207 0.193 0.178 0.172 0.171 0.166 0.148 0.155 0.150

Marginal Valuation

Quality Valuation

Lower

Upper

Lower

Upper

Lower

0.184 0.168 0.153 0.142 0.149 0.149 0.141 0.131 0.121 0.117 0.117 0.112 0.101 0.105 0.103

2.934 3,091 3,257 3,400 3,305 3,309 3,413 3,559 3,727 3,800 3,803 3,702 3,977 3,805 3,949

3,652 3,848 4,054 4,232 4,117 4,120 4,250 4,432 4,640 4,732 4,736 4,659 4,985 4,771 4,924

79 1 76 1 132 709 724 723 707 685 662 652 652 615 590 589 593

672 646 622 602 615 614 600 582 562 554 554 52 1 501 501 506

349

Output and Quality in Day Care

Hofferth (1987) reports tb.at the total capacity of licensed day-care centers was 1.01 million children in 1976 and 2.1 million in 1986.9The spaces available in licensed day-care centers approximately doubled over the ten-year period. Our estimates of the number of children in care is below capacity as would be expected (.55 million in 1976 and 1.89 million in 1986). It is interesting that our estimates of the number of children in care is below capacity in both 1976 and 1986, given the prevalence of articles in the press regarding nonavailability of child care. To the extent these are references to day-care centers (as opposed to several other forms of care, e.g., family day care) a possible reason may be a mismatch between the location of available slots and the location of families with child-care needs. Note also that the gap between capacity and the number of children in care narrowed markedly between 1976 and 1986. This could have also contributed to the refrain, “I can’t find day care.” Another reason lends credence to the number of children in care being lower than capacity. Conversations with directors of day-care centers indicate that it is common practice for day-care centers to obtain a license for as high a capacity as possible. Many centers actually choose to run below capacity for professional and other reasons. For example, one center director with whom we spoke had a licensed capacity of 101 and actual enrollments of only between 80 and 90. Most centers we talked with run at about 80 percent of their capacity. This is in line with our estimates, which indicate that in the ten years from 1976 to 1986, capacity utilization has increased from 54 percent to about 90 percent. Our estimates of the number of children in care imply a 246 percent increase in the number of children in care. This may overestimate the rate of increase in the number of children in care, and we suspect that the number of children in care increased by between 100 percent and 246 percent.I0 Now we turn to the data on employment in day-care centers. The employment numbers are the annual average employment figures taken from Bureau of Labor Statistics (BLS) publication, Employment and Wages, Annual Averages, of various years. Employees of day care centers are included in the SIC code 8351, child day-care services. This SIC code contains nursery schools, preschools, and some Head Start centers as well as day-care centers. These employment data cover employees subject to state unemployment insurance (UI) laws and also those subject to unemployment compensation for federal employees (UCFE) program. Use of these data to obtain a series for employment in day-care centers may result in either an over or under estimate of the number of employees in day-care centers. The source of overestimation is clear as these data reflect employment in nursery schools, preschools, and some Head Start centers. 9. The 1976 data were taken from the NDCS, which reported on licensed day-care centers. The 1986 figures come from a survey of each state’s licensing office done by the NAEYC. See Hofferth (1987), 565. 10. A recent report on child care by the U.S. Department of Labor (1988) estimates that the number of children in care increased by 77 percent between 1977 and 1985.

350

Swati Mukerjee and Ann Dryden Witte

However, three sources of underestimation must be set against this source of employment overestimation. SIC 835 excludes employer supported programs, as long as workers are on the employers’ payroll. Hofferth and Phillips (1987) conclude that “employers would not appear to be a major source of expanded child care.” Even among employers that do provide care, the most common way is to help families find and pay for such care. Hofferth and Phillips estimate that in 1985, “120 corporations and 400 hospitals provided child care at or near the work place.” These centers cannot be assumed to be run by the employers. Many lease out to chains like Kinder Care or Bright Horizons as, for example, Beth Israel Hospital in Boston is currently planning to do. Hence this source of bias may not be significant. Secondly, SIC 835 excludes church-supported day-care centers. We may note, however, that “under a 1981 Supreme Court ruling, church-chartered schools are not required to be covered under the system. However, many of these schools continue to cover their employees on a voluntary basis.”” This second source of underestimation therefore appears to be relatively small. The third source of downward bias is the inclusion in SIC 835 of only paid employees, to the exclusion of proprietors and partners who might also actively provide care. Iz As centers get larger, it is likely that this source of bias would be weakened. However, we have no information on which to judge the actual level of bias. In view of the above, we believe that use of BLS employment figures for SIC 8351 probably results in an overestimate of employment in day-care centers. To adjust for this, we multiply employment in SIC 8351 by the ratio of output of day-care centers to the combined output of day-care centers and nursery schools, that was obtained from the BEA. This establishes a lower bound on the employment numbers. The upper bound would be the unadjusted figures (table 9.1).

9.3 The Quality of Output As in the case of most other service industries, the output of day-care centers is not homogeneous. Thus, to measure output adequately, it is essential to adjust simple dollar or physical output measures for differences in quality. How does one measure the quality that day-care services provide? The concept of quality is a difficult one to define and to reduce to measurable indices. Historically, the concept of what constitutes quality shifted from post-World War I1 to the present times. Earlier, the nurturing aspect of care was emphasized to calm prevalent fears regarding perceived social and emotional ill effects of institutional care. In the 1960s, the emphasis shifted to an interest in the development of cognitive skills of children, triggering the beginning of Head Start and similar programs. In the 1970s, the emphasis was placed on a more balanced approach to developing both cognitive and social skills, thus 11. U.S. Department of Labor, (various years), sec. on “Characteristics and Uses of the Data.” 12. Manser (1990), preliminary comments.

351

Output and Quality in Day Care

enhancing both physical and emotional development.I 3 Today, the balanced approach prevails, as can be seen in the varying approaches to quality in accreditation criteria, in federal and state regulations, and in other studies.14 Today, the NAEYC,I5 which accredits child-care programs, defines a highquality early childhood program “as one that meets the needs of and promotes the physical, social, emotional, and cognitive development of the children and adults-parents, staff, and administrators-who are involved in the program. Each day of a child’s life is viewed as leading toward the growth and development of a healthy, intelligent, and contributing member of society.” (NAEYC 1989, 7). Among the integral criteria in the establishment of a highquality early childhood program, as cited by the NAEYC, are the following components: interactions among staff and children, a well-rounded and developmental curriculum, frequent staff-parent interaction, effective administration, sufficient and highly qualified staffing, a nurturing and spacious physical environment, a high degree of health and safety, a well-balanced nutrition and food service program, and continual evaluation of the program to assess its strengths and weaknesses. U.S. day-care centers have been regulated by federal, state, and local bodies. These regulations relate to enrollment policies, staff quality, group size and staff-child ratios .I6 With the suspension of the federal interagency daycare requirements (FIDCR) in 1981, the regulation of child-care services is under the jurisdiction of individual states.” In spite of a gradual tightening of regulations by the states since the mid- 1970s, individual state regulations can and do vary a great deal. In addition, the existence of these regulations does not necessarily imply that they are actually followed or enforced, because monitoring is poor. Is An important component, nevertheless, in both federal and state regulations has been the staff-child ratio, the requirements of which usually vary with the age of the child. Measures like the staff-child ratio, however, can only be an indication of 13. This discussion is taken from Stallings and Wilcox (1978), 103-4. 14. A thorough and up-to-date discussion on research and professional practice regarding the quality of care is given in Hayes, Palmer, and Zaslow (1990). chap. 4, and app. B. The appendix compares the main features of the major sets of standards that have existed and do exist in programs covering early childhood care. These are six in all, two being federal standards; the remaining four are standards set by professional bodies. The Head Start program still exists; the FIDCR was suspended in 1981. The remaining four sets of standards have been developed for voluntary compliance when applicable. These are (1) the accreditation criteria developed by the NAEYC; (2) guidelines developed by the NBCDI (National Black Child Development Institute); (3) standards of the ECERS (Early Childhood Environment Rating Scale), developed by child development scholars at the University of North Carolina; and (4) standards of the CWLA (Child Welfare League of America). 15. According to the GAO report, the NAEYC, “a membership organization of more than 70,000 professionals in the field of child development and early childhood education, provides the only national voluntary accreditation system exclusively for all types of early childhood centers and schools” (GAO 1990,61). 16. See Ruopp et al. (1979), 229. 17. See Hayes, Palmer, and Zaslow (1990), chap. 6, and app. A. 18. See Hayes, Palmer, and Zaslow (1990), chap. 4, sec. on regulations.

352

Swati Mukerjee and Ann Dryden Witte

quality and not be quality per se as considered by the comprehensive national day-care supply study (NDCS). This study’s measures of quality were based on a dual observation system, where daily behavior of both children and care givers was observed in detail. This dual approach was supplemented by several standardized tests. The NDCS study then correlated measurable characteristics of day-care centers, like staff-child ratio and group size, with their observations and test measures of quality. Their conclusion was that out of the measurable characteristics, group size, and caregiver-child ratio were very important, especially for infants and toddler^.'^ This finding is corroborated in subsequent research.*O More recently, a Department of Labor report (U.S. Department of Labor 1988) cited a reader survey from Working Mother magazine (March 1988) indicating that parents feel that “warmth, frequency, and kind of interaction” between child and care giver is important, imparting values consistent with their own. Other quality factors mentioned in the report were the training level of the staff, wages, and stability of staff. Table 9.2 lists various measures of the quality of child care that have been suggested.21Clearly, the quality of child care is best measured by a vector of attributes.22 Combining insights from all the above sources, it seems that an important, if not the most important, element in the measurement of the quality of care provided by centers is the interaction between the staff and children.23The measure of quality that is most often used is the staff-child ratio. However, this ratio captures the interaction only to a limited extent. Because group size also proved important, we used a ratio that reflects more closely this interaction and the group size. We take into account the fact that interaction occurs only when the staff is in contact with the children and that children of different ages require different levels of attention. For example, it requires far more time to provide necessary care for an infant than to provide necessary care for a preschooler. We therefore use weighted children instead of the simple number of children. The ratio we use is the number of staff-class hours divided by a weighted index of the number of children.24This is our preferred measure 19. Findings for infants and toddlers (6 weeks-3 years) showed both group size and care giverchild ratio strongly correlated with quality. In the case of preschoolers, the results were somewhat different. In their case, a smaller group size led to better quality and this result was both “strong and consistent.” However, the relationship of quality to care giver-child ratio was “slight.” Ruopp et al. (1979, xxxvii). 20. See Hayes, Palmer, and Zaslow (1990), 88. 21. We thank Marilyn Manser for use of this table from her preliminary comments to this Paper. 22. This is no different from what the literature has discussed (e.g.. in the case of hospitals). See Newhouse (1970). 23. A way of reducing the large number of quality characteristics would be to use a factoranalysis approach. However, there are well-known problems of interpretation. In this study, on the basis of the existing work done, we formed a judgment on what would be most important in measuring quality. 24. A better measure would be the staff-class hours divided by a weighted average of the number of child-care hours provided. The weights would reflect the differential time requirements of children of different ages.

353

Output and Quality in Day Care

Table 9.2

Definitions of Quality of Child Care Definitions or Measures

Viewpoint Parents

Professional associations

Other

Warmth, frequency, and kind of interaction between the child and the provider Values imparted consistent with those of their own family Presence of a program of developmentally appropriate activities Degree of parental involvement Training and knowledge of staff about child development Ratio of children to providers Size of groups in which children receive care Nutritional value of the meals provided Safety of the physical environment Other policies and practices affecting the health of the children, such as staff hygiene and the handling of medications Stable staff

Source: U.S. Department of Labor (1988). Note: Adapted from Manser (1990).

of quality and is available from our Massachusetts data. The staff-child ratio and our preferred measure of quality provide quite different results for the centers in our Massachusetts data. The staff-child ratio (QUALZ) consistently indicates higher levels of quality, particularly for larger centers, than does the ratio of staff class hours to the weighted number of children (QUALI).This may reflect an increase in the size of administrative staff as centers become larger. Clearly, these two measures of quality produced different results. Specifically, the staff-child ratio indicates that larger centers produce much higher quality care than the ratio of staff-class hours to the weighted number of children would indicate. Turning from the Massachusetts data to the national data, we are unable to obtain our preferred measure of quality, because there is no national information on the number of hours that children are spending at day-care centers. This is a serious gap in the national data. We are left with the staff-child ratio to measure the trends in the quality of care in the United States. We estimate that in 1976, staff-child ratio was between 0.23 to 0.15 or 4.4 to 6.5 children per care giver.25By 1986, we estimate that the staff-child ratio had declined so that there were between 6.8 and 9.9 children per care giver. These numbers are well within reasonable bounds, given federal and state regulations. However, they suggest a decline in quality that is worrisome.26 25. The reciprocals have been calculated for figures correct to 3 decimal places as taken from table 9.1. 26. Because ages of children could not be controlled for, one explanation could be a very high increase in the proportion of older children in day-care centers over the years. This appears unlikely given the increasing labor force participation of women with children younger than five years old.

354

Swati Mukerjee and Ann Dryden Witte

In the next section, we use U.S. staff-child ratios as the only measure of quality available at the national level and adjust national level output for quality changes by using a quality index. We carry out adjustments for output measured in physical as well as in dollar terms.

9.4 Valuing Quality Quality changes may be handled in a variety of ways. (See Armknecht and Ginsberg, chap. 3, this vol.) In the literature, hedonics has usually been the preferred route, where price is regressed on each of the characteristics of the product. The regression parameters then provide implicit prices for characteristics and enable calculation of the value of quality changes. An important prerequisite to using traditional hedonic analysis is the availability of an appropriate price. In the day-care case, it is difficult to decide which price to use. Parents pay one price (often subsidized), and the actual cost of care is quite different, as centers receive various private, state, and federal subsidies. Hedonic analysis relies on a market clearing price that reflects both marginal costs and marginal utilities. Such a price is not often observed for day care. An alternative way of adjusting for quality is based on costs. Armknecht and Ginsberg observe, “When market information is not available, we could ask the provider of the service to estimate the value of the change from their cost data. Changes in provider’s costs could be adjusted for normal profit margins and marked up to the retail level. The resulting price change serves as a proxy for consumer’s valuation of the quality change which cannot be observed directly in the market place. This approach is used most frequently for commodities in the CPI, particularly for new vehicles.” Our approach is in some ways consistent with Armknecht and Ginsberg’s suggestion. We estimate a cost function using accounting data and calculate the marginal cost of quality using the parameter estimates that result. For PMOS it seems reasonable to assume that this marginal cost is a reasonable proxy for consumer valuation. For NPOs this marginal cost may not provide a good proxy for consumer valuation. However, it does provide an estimate of the cost of producing higher quality, and this can be seen as a useful first step toward producing valuations for quality. Conceptually, C = C (Z, Q), where Z is the vector of explanatory variables, and Q is the vector of the quality variables. Then the partial derivatives dCldQ would give, under the assumptions of equilibrium, competition, and no externalities in the day-care market, the vector of the marginal valuations of quality in day care with respect to each quality variable. These marginal valuations can then be used to construct a quality index with which to adjust OUtpUt.2’ 27. Regarding the measurement of quality, the controversy of price-versus-cost approach can be seen in earlier works (e.g., Nicholson 1967 and Gilbert 1961). As Newhouse says, “J. L. Nicholson has criticized Milton Gilbert for using cost rather than price to measure the contribution

355

Output and Quality in Day Care

At the national level, we lack the data needed to construct cost functions, and, therefore, we construct a quality index using parameter estimates from cost functions estimated using Massachusetts data sample.**In an earlier paper (Mukerjee, Witte, and Hollowell 1990) separate cost functions were estimated for PMOSand NPOs assuming a generalized homothetic Cobb-Douglas production technology (Zellner and Revankar 1969)29and our preferred quality measure. This technology has the advantage of allowing returns to scale to vary with output. Table 9.3 contains a definition of all variables included in the cost functions30 and descriptive statistics for these and some additional variables. The empirical results for the cost functions using the preferred quality measure appear quite reasonable3‘ with increases in the price of inputs increasing costs. Furthermore, the production technology utilized by forprofit and not-for-profit centers are significantly different. For the purposes of the present study, cost functions were reestimated separately for PMOS and NPOs using staff-child ratios. (See results in table 9.4.) From the perspective of quality valuation, we are primarily interested in the coefficient on the quality measure, in this case the staff-child ratio. The coefficient on these variables always have t-ratios greater than two-the quality of output consistently has a significant effect on the total costs of operating a dayto welfare of a change in quality. Since cost equals price in our model, this criticism presents no problem to it; one could merely say that the assumptions imply that an increase in quality, quantity held constant, implies an equal increase in both cost and price. Our analysis is really in the same spirit as Nicholson’s by proposing a criterion which relates to the consumer’s preferences as revealed in the market-place; that is, that the decision maker is in equilibrium at the quality level which maximizes quantity bought at a given price . . .” (67). 28. There are two reasons why using estimates from Massachusetts data may overstate the national valuation. First, incomes are in general higher in Massachusetts, people there tend to hold liberal views, and there is a long tradition of concern with social welfare. This is likely to lead to a higher than average valuation to quality, overstating the coefficient on quality thereby. Second, average costs may be higher because Massachusetts is generally acknowledged to be a high-wage state. We suspect that using Massachusetts data to get the marginal cost of quality overstates the national valuation. 29. The theoretical reasons we did not choose a more flexible form like the translog are outlined in Mukerjee, Witte, and Hollowell (1990). The first difficulty is that the large number of parameters to be estimated in a flexible form call into question the precision of estimates in a reasonably sized data set. The second problem is that, in the presence of a large range of observations in the data set, flexible forms may fail to fulfill certain restrictions such as diminishing marginal physical product. These considerations led us to consider the class of homothetic functions as good candidates for selection. However, it is interesting that Kremp and Mairesse (chap. 2, this vol.) reject the null hypothesis that elasticity of substitution is equal to one. In future work we will be using more general forms. An alternative method of estimating the cost function would be to assume a multiproduct production function. This would greatly increase the number of right-hand variables and collinearity. We decided to use a single output production function and adjust the output for quality. 30. Note that in table 9.4 LN before a variable name indicates that a natural logarithm of the variable was included in the cost function. 3 1. The GAO report (1990) for Congress deliberations on the “Smart Start” bill used a CobbDouglas production function and obtained cost estimates with quality measures like the staff-child ratio. However, their results are not comparable with ours as they take into consideration only accredited centers (high quality).

Table 9.3

Descriptive Statistics for Day-care Centers in Massachusetts

NPO Sample

PMO Sample Variable

Mean

Standard Deviation

Mean

Standard Deviation

t-Statistic*

Variable Definition

TOTEXP

1,583.690 1,689.510 6.350 337.970 55.900 0.570

1,425.190 5,174.700 2.760 364.200 250.710 0.520

3,684.320 1,347.780 9.740 243.740 18.400 0.750

5,603.870 1,486.670 8.130 553.770 18.070 0.540

0.360 0.410 3.220 1.080 0.960 1.730

0.110 1.160 2.700 0.500

0.200 4.440 6.180 0.570

0.160 1.150 2.630 0.500

2.700 0.110 2.200

SUBIDUM

0.140 4.380 7.330 0.460

SUBZDUM

0.150

0.360

0.440

0.500

...

0.829 7.732 36.415 2.878 2.659 50.512 4.195 2.439 158.683 19.488 178.512

2.418 9.841 20.434 8.487 8.478 25.899 4.589 2.530 181.874 23.201 199.177

2.694 8.319 38.764 3.846 8.778 62.403 8.653 2.972 321.931 37.889 360.486

6.090 16.583 33.496 0.552 25.490 53.325 13.652 4.135 513.440 52.176 553.604

Total weekly expenditure ($) Total no. of hours children are cared for in a week Hourly price of labor Monthly price of capital (per room) Monthly price of material (per child) Ratio of paid staff classroom hourdweighted children Staffkhild ratio Average education of staff Average experience of staff Dummy variable = 1 if time donated by volunteers/ parents Dummy variable = 1 if received state/private funding Total no. of infants Total no. of toddlers ages 1-2 Total no. of preschoolers Total no. of kindergartners Total no. of school-age children Total enrollment Total no. of full-time staff Total no. of part-time staff Total hours staff spend in classroom per week Total hours staff spend in other duties per week Total hours of all staff per week

CHRS LAB

PCAP PMAT

QUALl QUALZ QUALED

QUALEX

INF TOD

PRE

KIN SCH TOTCHLD

NFT NPT PSCLHRS

PSOTHRS PSHRS

*r-statistic for test of difference in means.

...

2.300 0.236 0.463 0.730 1.864 1.592 2.530 0.850 2.442 11.264 2.518

357

Output and Quality in Day Care

Table 9.4

Cost Function Estimates for Day-care Centers in Massachusetts (using preferred measure of quality) PMO Sample

Variable Constant CHRS LNCHRS LNPLAB LNPCAP LNPMAT LNQUALI LNQUALED LN Q U ALEX SUB I DUM SUBZDUM

R= Adjusted R2 F F'robability>F

N

Coefficient Estimates 9.3990 0.0008 -0.5020 0.7980 0.2160 0.1540 0.3260 - 0.9290 -0.6090 -0.1630 0.1010 0.8204 0.7585 13.2500 o.Ooo1 40

NPO Sample t*

4.2110 2.6090 1.7980 4.2310 2.1400 1.8950 2.1780 2.7560 2.7590 0.9370 0.3740

Coefficient Estimates 4.7430 0.0002 0.2270 0.2520 0.0750 0.1800 0.2520 0.0330 -0.2900 0.2700 0.1870 0.7954 0.7499 17.4920 0.0001 56

I*

4.9720 2.7680 1.9110 2.4360 1.4020 1.8010 2.2600 0.1250 1.6570 1.9630 1.1340

r-Value+ 1.918 2.000 0.908 2.516 1.240 0.200 3.895 2.253 1.130 1.959 0.273

'Absolute values. 'r-value for test of difference of estimates

care center.32 Further, the coefficients on the quality measures are insignificantly different for PMOS and NPOs. To obtain a marginal valuation for quality, we take the derivative of total cost with respect to the measure of quality (QUAL). Given our homothetic Cobb-Douglas cost function the derivative is

where QUAL in this case is measured as the ratio of the number of staff to the number of children, e is the base of natural logarithms, Q is our measure of output (the total number of hours children are cared for at the center, or child hours, CHRS), and X is a vector of all other explanatory variables, listed in table 9.4. This derivative is the marginal cost of providing higher staff-child ratios. If we assume equilibrium, competition, and no externalities in the day-care mar32. As has been pointed out by Zvi Griliches, ed. of this vol., it is possible that our measure of quality is endogenous and that the significance of the quality measure stems from its correlation with the error term.

358

Swati Mukerjee and Ann Dryden Witte

ket,33this marginal cost is equal to the marginal social valuation of increases in the staff-child ratio, the measure of the quality of day care available at the national level. Recall that at the national level we were able to construct a time series for the staff-child ratio for the 1974-88 time period. It was not possible to construct a separate time series for the PMOs and NPOs. Because the Chow test indicated that the PMOs and NPOs have significantly different technologies, two sets of marginal valuations, one using PMO coefficient estimates and means and the other using NPO coefficient estimates and means, were estimated separately using the U.S. staff-child ratios. To determine the U.S. marginal valuation, the marginal valuations for PMOS and NPOs were weighted by the estimated proportion of the two types of Our marginal valuations of quality are open to a number of questions and are, perhaps, best considered illustrative of a methodology rather than policy relevant. The first difficulty is that we use data for a single state to estimate the cost functions. Clearly, it would be better to estimate these functions using data that are nationally representative. The second difficulty is our simplistic measure of quality. Clearly, the quality of day care is best measured using a vector of attributes rather than a single proxy measure as we have done. Finally, it is possible that our measure of quality is endogenous. If an appropriate instrument were available, we could adjust for this potential problem. Unfortunately, there was no reasonable instrument in our data set.

9.5 The Adjustment of Output for Quality Taking the weighted marginal valuations for the United States, calculated as described above, we multiplied this marginal valuation by the level of quality (i.e., staff-child ratio) in the United States in each year from 1974 to 1988. This is the quality value that we used to construct an index of quality valuation. We adjusted our physical measure of output by multiplying the output by its index for the value of quality actually attained in that year. Figure 9.2 compares the quality-adjusted physical measure with its unadjusted counterpart, the number of children. Notice that adjusting for quality has lowered the rate of growth. Quality-adjusted physical output is below the unadjusted out33. These conditions are more likely to hold for profit-making firms rather than nonprofit ones. Thus, we have far more confidence in the case of PMOs rather than NPOs that the marginal cost we calculated are reasonable proxies for marginal valuations. However, we believe that the calculation of marginal cost for NPOs is a useful step on the road to obtaining good proxies for quality valuation. 34. The weights were the proportion of the receipts and revenues of PMOs and NPOs to the total receipts and revenues of all centers. The receipts and revenues for 1977, 1982, 1985, 1986, 1987, and 1988 were obtained from the various censuses and the Service Annual Surveys. Receipts and revenues for the years in between were obtained by extrapolating using the assumption that they grew proportionally to the employment in day-care centers. In our case, we used the lowerbound employment numbers.

Output and Quality in Day Care

359

21 ...... Children

-A- Adj. Children

1.6

*

Oe2 0 74 75 76 77 78 79 ao

a1 a2 a3 a4 Y e a r s

a5 a6 a7 aa

Fig. 9.2 Growth of children and quality-adjustedchildren over time in United States (1982 = 100) 0.3

0.27 0.24 v)

.-c0

0.21

K

0.18

m

.......... ............. '..

...

...........................

....... ....................

0.15 0.12

...... Upper Bound

-Lower Bound

0.03 74 75 76 77 78 79

ao

a1 a2 a3 a4 a5 Y e a r s

a6

a7 aa

Fig. 9.3 Movement of staff-child ratios over time in United States

put measure after 1982 because of the decline in the staff-child ratio (fig. 9.3). The decline in quality, as measured by staff-child ratios over time, has led to an increasing marginal valuation of quality. (See table 9.1 .) This is consistent with the increasing national concern over the quality of day care. Griliches, editor of this volume, has pointed out that we may be overadjusting for quality when we adjust real dollar measures of output using our methodology. This

360

Swati Mukerjee and Ann Dryden Witte

may be true because the real price of day care already reflects changes in quality. To check on this possibility, we correlated the real price index for day care with our quality index.35 The correlation was high, at 0.92. Note both a decline in the estimated real price of day care and the decline in the estimated staff-child ratio indicate a decline in the quality of care at day-care centers. We also compared quality-adjusted children with the unadjusted real dollar value of output (see table 9.5). The two measures were highly correlated at 0.98. Figure 9.4 compares the movements of the quality-adjusted physical measure and the unadjusted real dollar measure over time. The figure indicates that, after 1982, both the real dollar output used by the BEA and the quality-adjusted children measures move fairly closely. This underscores the necessity of adjusting the physical measure for quality changes, bringing it in line with the BEA real dollar measure of output.

9.6 Summary and Conclusions The twin objectives of this paper were, first, to study different available output measures of day-care centers and, second, to propose a methodology for adjusting a given output measure for quality changes. To develop a national measure of quality-adjusted output, we carefully considered the measurement of both output and quality. Output may be measured either in physical or in dollar terms. At the national level, the two measures indicate different rates of growth over time with the difference between the measures increasing over time. To measure the output of day care in physical terms, we suggested the use of the number of hours of child care provided rather than a simple count of the number of children in care. An even better measure of output, we felt, would be the number of hours of care weighted by the age of children in care. However, at the present time, data on the suggested physical measure of output are not available. What is available is information on the number of children in care, but only for selected years. The problem is made more complex by the fact that the survey questions used to obtain information on the number of children in care are not comparable over different years. We had no alternative but to use various adjustments and interpolation and extrapolation to construct a time series for children. This series was compared with the available alternative dollar measure of output: BEA’s series on the real dollar value of day-care expenditures. The most widely used traditional measure of the quality of day care has been the staff-child ratio. We suggest that the ratio of the number of staff-class hours to the age-adjusted number of children is a superior measure of quality because it better proxies staff-child interaction. In the absence of national data 35. Real price is defined as price that has been adjusted for inflation. At the national level, we obtained this in two stages. First we multiplied the number of children by the average number of hours from Massachusetts. The next stage consists of dividing the constant-dollar measure of output by the estimated child hours for the United States.

Output and Quality in Day Care

361 Table 9.5

Comparison of Quality-Adjusted Physical Output Measure with BEA Real Output (1974-1988) Children Adjusted (in thousands) Year

Real BEA Output (in millions of $)

Upper Bound*

Lower Bound'

...

462 528 604 695 793 885 966 1,045 1,128 1,242 1,386 1,459 1,684 1,744 1,961

46 1 528 604 695 793 885 965 1,045 1,128 1,242 1,386 1,455 1,684 1,747 1,968

1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988

...

...

3,436 3,986 4,270 4,21 I 4,200 4,242 4,633 5,203 5,724 6,272 6,767 7,339

*Refers to the upper bound of the staff-child ratio. 'Refers to the lower bound of the staff-child ratio.

dQ o) o

.

U

5

74

75

76

77

78

79

80 81 82 Y e a r s

2 83

84

85

86

87

88

Fig. 9.4 Growth of unadjusted BEA output and quality-adjusted children over time in United States (1982 = 100)

362

Swati Mukerjee and Ann Dryden Witte

on the preferred measure of quality, we constructed a series on the traditional measure of quality for the United States. This required statistics on employment in day-care centers, and we obtained these by adjusting the numbers given in the employment and wages (U.S. Department of Labor) of various years. Dividing the number of children in care by the number of employees, we obtain estimated staff-child ratios for 1974-88. The estimated staff-child ratio declined throughout the period studied. This suggests that there has been a decline in the quality of care provided by day-care centers in the United States during the period considered. To fulfill our second objective, we illustrated a method for adjusting daycare output (whether measured in physical or dollar terms) by any given quality measure. The need for this method grew out of the difficulty of applying traditional hedonic methods because of the presence of both supply- and demand-side subsidies that made the definition of a single price d i f f i c ~ l tAn .~~ alternative to the traditional route, the proposed methodology utilizes cost functions to estimate a marginal cost of quality. Under equilibrium and competitive conditions this marginal cost is the same as the marginal valuation of quality by the consumer. In the not-for-profit sector, however, these conditions may not hold, and, in such instances, these valuations simply give the marginal cost of quality. Separate marginal costs of quality were obtained for notfor-profit and for-profit day-care centers in the state of Massachusetts, these estimates being used to illustrate the methodology. To obtain a marginal valuation of quality for the United States, a weighted average of the separate marginal valuations was obtained with the weights being the proportions of each type of center as reflected in the BEA data for the United States. The national marginal valuation was multiplied by the estimated staff-child ratios in the United States and then indexed to obtain a quality index. To illustrate our methodology, we proceeded to use the quality index to adjust both the simple physical and dollar measures of output in the United States in the 1974-88 time period. The decline in the staff-child ratio during the 1970s means that the quality-adjusted output shows lower rates of growth than the simple output measures. Unadjusted output may overstate both the level and the growth rate of output during our study period. It is possible, however, that the real dollar measure of output may already incorporate a quality adjustment, and so our quality measure may be overadjusting in such a case. Indeed, the quality-adjusted children and the real dollar measure of output are highly correlated, suggesting that both methods of quality adjustment (using real prices to obtain a dollar measure of output or adjusting physical output using a marginal valuation of quality) give similar results. Clearly, additional work in quality adjustment is called for. This is apart from the fact that quality itself 36. Griliches, ed. of this vol., has pointed out that the existence of subsidies may cause difficulties for cost-function estimation. This is certainly true and suggests that an essential next step is to develop models that better reflect the unusual environment of day care.

363

Output and Quality in Day Care

may be more appropriately measured as a vector: the staff-child measure may be challenged as being too naive or too crude. Furthermore, because national data were not available for the estimation of cost functions at the national level, cost functions were obtained using a Massachusetts data set. We are aware that Massachusetts cannot be regarded as typical of the United States. Nevertheless, it serves to illustrate a methodology that may be useful in the future when national level data are available. It also supplies an alternative to traditional hedonics and may be applicable to cases other than the day-care industry.

Appendix A In this appendix, we describe the way in which we obtain the aggregate timeseries data on day care reported in table 9.1. The main problems in constructing this time-series data are the unavailability of data for several years, changes in the definition of variables, and changes in the source of the data. We describe the sources of data and the definition of variables that underlie the data. We also explain the adjustments made to make the available data comparable for different years. Finally, we explain how we interpolated and extrapolated to obtain estimates for years when no primary data were available.

Number of Children Sources 1974-7.5. The data for 1974 and 1975 were obtained from the Current Population Reports, series P-20, no. 298, October 1976. 1976. Data for 1976 were taken from a random sample of day-care centers that are open at least 25 hours per week, have a capacity for 13 or more children, operate at least nine months per year, and have a majority of enrollment that is nonhandicapped. The survey was conducted by Abt Associates and is reported in Coelen, Glantz, and Calore (1979). Because of difficulties in making appropriate adjustments, this source was not used. This source is also supply based, whereas CPS and SIPP data are all demand based. 1977. Data for 1977 were from Current Population Reports, Special studies, P-23, no. 117, June 1982, Bureau of the Census, U.S. Department of Commerce. 1985. The data for 1985 were from the Current Population Reports, Household economic studies, series P-70, no. 9, Bureau of the Census, U.S. Department of Commerce.

364

Swati Mukerjee and Ann Dryden Witte

1986. The data for 1986 were obtained from a press release, U.S. Department of Commerce News, Washington, D.C., Thursday, July 27, 1989, no. CB 89-1 19. Adjustments Made to Make Data Comparable Source of the 1974 Data As far as we know, data for the number of children in nonparental care was first collected as a supplement to the CPS in October 1974 and February 1975. The 1974 survey obtained information on care arrangements for children 713 years old. The 1975 survey included questions for children 3-6 years old. Adjustments to 1974 Data The children included in the two CPS surveys were between 3 and 13 years old, whereas in later years, with the establishment of the SIPP, coverage was extended to include all children younger than 15 years. We expanded the 1974-75 data to include children younger than 3 by using the ratio of infants and toddlers to the total number of children obtained from the Massachusetts data. This may provide an overestimate of the number of children in care during the early years if we consider the implication of Hofferth’s findings (1987): “The most striking trend is the substantial growth in use of group care programs from 1965 to 1982. This growth was steady for children of full-time employed mothers. It was equally dramatic for children under age 3 and for preschoolers ages 3 and 4. Among children of part-time employed mothers, use of centers for both age groups rose substantially between 1965 and 1977, and then declined somewhat between 1977 and 1982” (562). We verified from the Bureau that the coverage has always included the three youngest children and has been the same through succeeding CPSs and SIPPs. Hence, no adjustments were necessary regarding this aspect. Day-care centers were defined by respondents in the CPS and SIPP questionnaire separating nursery schools and preschools from other types of daycare centers. We use the number of children reported by the respondent to be that enrolled in day/group care. The 1974-75 survey reflects child-care arrangements for all mothers, whether or not they participated in the labor force, either on a full-time or a part-time basis. Subsequent SIPP surveys include only employed mothers. Adjustments to 1977 Data The children included were the two youngest children (younger than five years) of employed mothers. We adjusted the number of children to account for the children of mothers who are unemployed and for those not in the labor force by using the 1974-75 CPS data described above. These data give the

365

Output and Quality in Day Care

day-care arrangements for mothers working, unemployed, or not in the labor force. We then adjusted the number of children to include those who were 5 and older by utilizing corresponding ratios from the Massachusetts data sample and by inflating the data accordingly. Adjustments for 1985-86 Beginning in 1985, a special module on child-care arrangements was included as part of the SIPP program. SIPP obtains information on child-care arrangements for the three youngest children (younger than 15) of working parents or guardians. SIPP data are currently available for 1985 and 1986. We adjusted the number of children to account for the children of mothers who are unemployed and for those not in the labor force by utilizing the 1974-75 CPS data described above. These data give the day-care arrangements for mothers working, unemployed, or not in the labor force. Data are currently available from the SIPP child-care module administered in January-April 1985 and September-November 1986. (See app. B of U.S. Department of Commerce 1987; and U.S. Department of Commerce 1990) We used the January-April, 1985 results as the basis for obtaining the number of children in day-care centers in 1985 and the September-November 1986 results to estimate the number of children in day-care centers in 1986.

Methods of Interpolation and Extrapolation Using all data currently available, we were able, even after the extensive adjustments described above, to obtain estimates of the number of children attending day-care centers for only 1974, 1977, 1985, and 1986. We obtained estimates of the number of children in care for all other years by interpolation and extrapolation. Specifically, our estimates of the number of children in day care in 1975 and 1976 were obtained by using the average annual compound rate of growth between 1974 and 1977. Our estimates of the number of children in care for 1978 through 1984 were obtained by using the average annual compound rate of growth between 1976 and 1985. Our estimates of the number of children in care for 1987 and 1988 were obtained by using the average annual compound rate of growth as reflected in the SIPP data for 1977 and 1985.

Number of Employees Source. U.S. Department of Labor (various years). Because SIC 835 included nursery schools, preschools, and some types of Head Start centers, we adjusted the data downward by utilizing unpublished

366

Swati Mukerjee and Ann Dryden Witte

data on output of day-care centers and nursery schools given to us by Robert Parker of BEA . BEA’s Dollar Measure of Output These data were obtained from Robert Parker of the BEA and are unpublished. The estimate of the nominal dollar value of the day-care centers’ output aggregates the receipts of PMOs and the expenditures of NPOs. The data have been adjusted by BEA for misreporting on the tax returns and other coverage errors. The real output was obtained by using the implicit price deflator for day care as developed by BEA.As we understand from Parker of the BEA, this deflator is a composite of input prices assembled by the BEA, being a weighted average of the index of average earnings and the producer price index for industrial commodities less final and related products and power. The actual weights themselves are derived from the 1977 input-output table. Quality Adjustments on Aggregate Data

To obtain a quality index for aggregate data, the marginal valuation of quality was calculated for various staff-child ratios at the national level, using the estimates with QUALZ for all centers in Massachusetts (i.e., combining PMOs and NPOs). These marginal valuations were multiplied by the actual staffchild ratios to yield a value of quality that was utilized in obtaining a quality index. This index was then used to adjust the quantitative measure of output (children).

Appendix B The data are obtained from a random sample of day-care centers in Massachu~ e t t s . ~The ’ dependent variable, total costs, is the annual expenditure of each center summing all labor, capital, supplies, food, transportation, utilities, phone, liability insurance, and other costs. The independent variables are discussed below. The price of labor is the personnel costs, fringe benefits, and payroll taxes divided by total paid staff hours. To arrive at the total capital cost for a center, 37. The data set contains information for centers selected from two sampling frames. The first sampling frame was the licensing lists of the Massachusetts Office for Children. The data for 86 centers from this sampling frame are used. The second frame was the centers used by a random sample of Massachusetts families with children under the age of 13. The data for 27 centers were selected from this sampling frame. The addition of the second set of centers in February 1988 tends to overcome deficiencies in the licensing-list sampling frames (e.g., incomplete and out-ofdate lists). See the affordability report (Marshall et al. 1987) for details.

367

Output and Quality in Day Care

we sum rent or mortgage payments, utilities, maintenance, and repair costs. This figure is divided by the total number of rooms to yield the price of capital. Summing the costs of supplies, equipment, food, phone services, and transportation and then dividing by the number of children gives us the price of material. Day-care centers receive a range of subsidies: state food-program allocations, donations, funds from endowments, supplies brought in by parents, volunteer hours, and Department of Social Services (DSS) funding.38Because these subsidies fall naturally into two groups, two subsidy variables were created. Subsidy 1 is a binary variable equal to one if the center used volunteer hours and zero if the center did not. Subsidy 2 is a binary variable equal to one if the center received state funding items (state food and DSS), financial subsidies such as endowments and loans, or funding from private organizations or the United Way. These subsidies may be expected to lower costs for PMOS, but prima facie it is not possible to say how they may affect NPOs. In the latter case the use of subsidies may actually increase costs-for example, they may be used for staff perquisites. Costs may also rise because of unmeasured improvements in quality. For instance, better and more varied play equipment may be purchased. The quality variables used have been discussed in section 9.3. The next two variables relate to education and experience. To reflect the diverse educational levels of the staff in any day-care center, we construct a variable that indicates the average education of the staff. The experience variable is created by taking the weighted average of the total years of experience possessed by the staff. This reduces the range of staff experience in any one center to a scalar. It is unclear how parameter estimates on these two variables should be interpreted. They may be reflecting technology embodied in labor. In that case, there is the possibility that, as more technology is incorporated into labor, certain aspects of quality, for example, the fostering of creativity in children, may be enhanced. Alternatively, this labor embodied technology may lower costs. The coefficients on these variables may also measure marginal productivity not included in labor price. To the extent these two variables measure quality, we would expect a positive relationship to costs. However, to the degree they reflect cost-reducing technological change or differences in marginal productivity not reflected in labor price, we would expect negative coefficients.

References Bradford, D. F., R. A. Malt, and W. E. Oates, 1969. The Rising Cost of Local Public Services: Some Evidence and Reflections. National Tau Journal 22(2): 185-202. 38. The DSS provides funding for families based on need. This subsidy is given directly to the centers.

368

Swati Mukerjee and Ann Dryden Witte

Coelen, Craig, Frederic Glantz; and Daniel Calore. 1979. Day Care Centers in the US.:A National Profile 1976-1977. Cambridge, Mass.: Abt. General Accounting Office. 1990. Early Childhood Education: What Are the Costs of High Quality Programs. Washington, D.C. Gilbert, M. 1961. Quality Changes and Index Numbers. Economic Development and Cultural Change 9:287-94. Hall, A. 1978. Estimating Cost Equations for Child Care. In Child Care and Public Policy, ed., P. K. Robins and S. Weiner; 157-85. Lexington, Mass.: Lexington. Hayes, C. D., J. Palmer, M. Zaslow, eds. 1990. Who Cares for America’s Children? Child Care Policy for the 1990s. Panel on Child Care Policy, Committee on Child Development Research and Public Policy, Commission on Behavioral and Social Sciences and Education, National Research Council. Washington, D.C.: National Academy Press. Hofferth, S. 1987. American Families in Tomorrow’s Economy. Statement at the hearing before the Select Committee on Children, Youth and Families. House of Representatives, 100th Cong. 1st sess. Hofferth, S . , and D. Phillips. 1987. Child Care in the United States, 1970 to 1995. Journal of Marriage and the Family 49559-71. Inman, R., ed. 1985. Managing the Service Economy: Prospects and Problems. Cambridge: Cambridge Univ. Press. Kendrick, J. W. 1985. Measurement of Output and Productivity in the Service Sector. In Managing the Service Economy: Prospects and Problems, ed. R. Inman, 111-33. Cambridge: Cambridge Univ. Press. Klerman, J., and A. Leibowitz. 1989. Child Care Costs and Return to Work. Revision of paper presented at Western Economic Association Meetings, Rand Corporation, Lake Tahoe, Nev., June 18-22. Lange, M . , J. Bullard, W. Luksetich, and P. Jacobs. 1985. Cost Functions for Symphony Orchestras. Journal of Economic Literature 9(2): 7 1-85. Manser, M. 1990. Conference comments. NBEWCRIW Conference on Output Measurement in the Services Sector, Charleston, S.C., May. Marshall, N. L., and A. D. Witte, L. M. Nichols, and F. Marx. 1987. Report of the Child Care Affordability Study. Center for Research on Women, Wellesley College. Mukerjee, S . , A. D. Witte, and S. Hollowell. 1990. Provision of Child Care: Cost Functions for Profit-Making and Not-for-Profit Day Care Centers. NBER working paper no. 3345; Cambridge, Mass., April. National Academy of Early Childhood Education. 1989. Accreditation Pamphlet. Washington, D.C.: NAEYC. Newhouse, J. P. 1970. Toward a Theory of Nonprofit Institutions: An Economic Model of a Hospital. American Economic Review 6054-74. Nicholson, J. L. 1967. The Measurement of Quality Changes. Economic Journal 7 7 5 12-530. Pauly, M., and M. Redisch. 1973. The Not-for-Profit Hospital as a Physicians’ Cooperative. American Economic Review 6397-100. Ross, J. P. and J. Burkhead. 1974. Productivity in the Local Government Sector. Lexington, Mass.: Heath. Ruopp, R., J. Travers, F. Glantz, and C. Coelen. Children at the Center. Cambridge, Mass.: Abt. Schofield, J. A. 1988. Production Functions in the Sports Industry: An Empirical Analysis of Professional Cricket. Applied Economics 20:177-93. Searle, A. D., and C. A. Waite. 1980. Current Efforts to Measure Productivity in the Public Sector: How Adequate for the National Accounts? In New Developments in Productivity Measurement and Analysis, ed. J. Kendrick and B. Vaccara, 333-56. NBER Studies in Income and Wealth, vol. 44. Chicago: Univ. of Chicago Press.

369

Output and Quality in Day Care

Stallings J . , and M. Wilcox. 1978. Quality of Day Care: Can It Be measured? In Child care andpublic policy, ed. P. K. Robins and S. Weiner. Lexington, Mass.: Lexington. U.S. Department of Commerce. 1976. Daytime Care of Children: October 1974 and February 1975. Washington D.C.: Government Printing Office, Current Population Reports, series P-20, 298. U.S. Department of Commerce. Bureau of the Census. 1987. Who’s Minding the Kids? Child Care Arrangements: Winter 1984-85. Current Population Reports: Household Economic Studies, ser. P-70,110.9. . 1990. Who’s Minding the Kids? Child Care Arrangements: Winter 1986-87. Current Population Reports: Household Economic Studies, ser. P-70,no.20. U.S. Department of Labor. Various years. Employment and Wage: Annual Averages. Washington, D.C.: Bureau of Labor Statistics. . 1988. Child Care: A Workforce Issue. Report of the Secretary’s Task Force. Superintendent of Documents Classification no. L1.2:C43/5. Zellner, A . , and N. Revankar. 1969. Generalized Production Functions. Review of Economic Studies 37:241-50.

This Page Intentionally Left Blank

10

Productivity in the Transportation Sector Robert J. Gordon

If we are ultimately to gain an understanding of the underlying causes of the worldwide slowdown of productivity growth in the 1970s and 1980s, analysts must probe at the microeconomic level of industries, firms, and establishments. The transportation sector has a special appeal for microeconomists, because of its long history of government regulation, and more recently, the laboratory experiment provided by the virtually complete deregulation of domestic air transport and the substantial deregulation of railroads and intercity trucking. The transportation sector is endowed with a unique and largely public data base, as one beneficial side effect of its history of regulation, helping to explain why microeconomists have expended a disproportionate amount of effort studying an industry that in 1987 accounted for only 3.3 percent of total GNP and 5.9 percent of service GNP. As shown in table 10.1, the transportation sector illustrates the same general pattern of post-1973 productivity slowdown as the total economy, only more so.’ The growth rate of average labor productivity (ALP) in the transportation sector exhibited a sharper deceleration during 1973-87 (as compared to Robert J. Gordon is Stanley G . Harris Professor in the Social Sciences at Northwestern University and a research associate of the National Bureau of Economic Research. This research is supported by the National Science Foundation. The author is grateful to Tim Schmidt for outstanding research assistance, to Victor Li and George Williams for their help in the last stages of the research, and to Severin Borenstein, Robert Fogel, Zvi Griliches, John Panzar, Frank Spencer, and Michael Tretheway for helpful suggestions. Richard Cames and Edwin Dean of the Bureau of Labor Statistics, as well as Michael Mohr and Jack Triplett of the Bureau of Economic Analysis, provided information on the sources of several data discrepancies. This paper is dedicated to the Northwestern University Transportation Library, one of the nation’s great resources in transportation economics, without which this research would not have been possible. 1. In what follows the terms “unrevised” and “revised’ refer to industry output data for 197787 published in the NIPA prior to and after January 1991. Table 10.1 links Kendrick’s (1961) estimates for the pre-1948 period with the unrevised NIPA data for the period since 1948; it provides the only long-run view of transportation productivity available to analysts prior to early 1991. Below we shall incorporate the revised NIPA output data for 1977-88.

371

372

Robert J. Gordon

1948-73) than did the nonfarm private economy, with respective slowdowns of 1.87 and 1.51 annual percentage points. The slowdown is even more serious when 1973-87 is compared with 1909-48, yielding a 3.71 point slowdown for transportation that is triple the 1.23 point slowdown for the economy as a whole.2 How could the productivity performance in transportation be so lamentable in an era when deregulation was widely perceived as offering management a myriad of opportunities for pursuing operating efficiencies that were formerly prohibited by regulators? This paper explores two complementary hypotheses: First, the data used in table 10.1 on the growth of ALP in transportation may incorporate a downward bias that is particularly large in the most recent decade. Second, productivity growth in the transportation sector is driven by the pace of labor-saving and energy-saving innovation achieved outside that sector by the manufacturing firms that produce transportation equipment. The ALP data in table 10.1 do not take into account either capital or energy inputs and thus do not rule out the possibility that multifactor productivity (MFP) growth slowed down after 1973 by less than labor productivity or even speeded up. The objectives of this paper are to reconcile conflicting measures of output and employment, to examine aspects of unmeasured changes in the quality of output, to provide improved measures of the quantity and quality of capital input, and to construct a consistent time series of MFP growth for the major transportation subsectors over the entire postwar period. The detailed analysis in this paper is limited to the three most important subsectors, railroads, trucking, and airlines, which constituted 82 percent of nominal transportation output in 1973.3The paper differs substantially from most of the literature on transportation productivity that has emerged in the past decade. With few exceptions, recent studies of air and surface transport have estimated cost functions from panel data sets in which individual carriers are observed over time. Although the use of data for individual carriers allows the effects of firm size, network density, and other cross-section issues to be addressed, these studies are limited by the relatively short sample period of the available data. In contrast, this paper attempts to assess the performance of the transportation sector over the entire postwar interval from 1948 to present, while sacrificing the added richness of data on individual carriers that are available for shorter periods. Because the primary focus of this study is to address the measurement of productivity in national economic statistics, a move along the trade-off curve toward a longer sample period and away from firm-specific observations seems appr~priate.~ 2. Mansfield (1965), using Kendrick’s data, treats the faster rate of productivity growth in transportation than in the aggregate economy as a well-accepted fact of economic history. 3. The remaining subsectors consist of local transit, water transportation, pipelines, and transportation services. 4. For a review of the cost-oriented studies of productivity change, see Winston (1985, 6669). The cost studies for air transport based on individual carrier data, and the sample periods

373

Productivity in the Transportation Sector

Table 10.1 Sector Nonfarm private economy Transportation* Railroads' Trucking'.* Airlines' Local transit'.'

Growth of Output per Hour in Nonfarm Private Economy, lkansportation Sector, and of Output per Employee for Subsectors, 1889-1987 1889-1909

1909-29

1929-48

1948-73

1973-87

1889-1987

2.27

2.18

2.12

2.43

0.92

2.07

2.05 1.88 ... . . . 2.34

3.36 1.58

4.75 2.95 9.70 8.25 3.96

2.20 3.66 3.70 5.33 -3.34

0.33 1.24 -0.28 -0.87 -2.12

2.63 2.40

...

... 3.00

. . .

... 0.70

Sources: 1889-1948-Nonfarm private: Kendrick (1961). table A-XXIII, 338-40. Transportation and subsectors: Kendrick (l96l), tables (3-11, G-111, G-VIII, G-X, and G-IV. 1948-87-Nonfarm Private: Economic Reporr offhe President 1990, table C-46, 346. Transportation and subsectors: NIPA table 6.2 divided by table 6.1 1 for total transportation; divided by table 6.10B for subsectors. *The transportation sector includes minor subsectors not included here, mainly water, pipeline transportation, and transportation services. 'Per employee, not per hour, for all subsectors 1948-87, and for trucking and airlines 1929-48. *Intercity only 1929-48, trucking and warehousing, 1948-87. '1889-1948, local railways and bus lines.

The longer sample period provides another benefit. Many of the earlier studies of productivity suffer from their timing; when data terminate in the period 1980-83, they are inevitably influenced by the idiosyncratic confluence of high energy prices and low aggregate demand prevalent during that period. A study that can include data through the late 1980s benefits from the recovery of the economy to a macroeconomic condition comparable to earlier prosperous years, as well as the partial reversal of the 1974 and 1979-80 oil shocks. Part 10.1 of the paper contains an analysis of measurement issues in the official U.S. government data on output and employment; it shows that the recent revisions of the industry output data in the national income and product accounts (NIPA) (de Leeuw, Mohr, and Parker 1991) resolve some inconsistencies in output data but leave substantial divergences between official agencies in estimates of employment and ALP growth. After a discussion of general conceptual issues in part 10.2, the paper turns to the detailed analysis of the subsectors. Much more attention is devoted to air transportation (part 10.3) than to rail (part 10.4) or trucking (part 10.5). This reflects two important differences: First, because rail and trucking output consists almost entirely of the carriage of freight, these subsectors provide intermediate rather covered, include Caves, Christensen, and Tretheway (CCT) (1981). 1972-77; CCT (1983), 197080; CCT (1984). 1970-81; CCT and Windle (1987), U.S. and foreign airlines, 1970-83; Sickles (1985). 1970-78; and Good, Nadiri, and Sickles (1989), 1977-81. Estimates of MFP growth based on groups of carriers (domestic, international, etc.) are available for 1948-81 in CCT (1985) and for air transportation as a whole in Jorgenson, Gollop, and Fraumeni (1987) and Jorgenson (1990).

374

Robert J. Gordon

than final goods. Hence any mismeasurement of productivity implies an offsetting adjustment in other industries rather than for the economy as a whole. In contrast, much of the output of air transport is sold directly to consumers, and so revisions to existing NIPA measures carry through to total GNP. Second, the quality of capital input in air transportation has changed much more dramatically over the postwar era than in rail or trucking, explaining our attention to alternative measures of capital input for airlines. We also incorporate changes in nonconventional inputs, including purchased services (e.g., those provided by travel agents) and government expenditures on airports, air traffic control, and highways.

10.1 Conflicts in the Official Data 10.1.1 The Discrepancy between NIPA and BLS

The U.S. official statistical system provides two independent measures of ALP in the transportation sector, but no estimates of MFP.5 Accordingly in part 10.1 we take a close look at the official output and employment data that enter into estimates of ALP like those already examined in table 10.1. One set of official ALP measures is provided by the NIPA, which contain estimates of real output and employment for total transportation and seven subsectors (see n. a , to table 10.l), and of hours for total transportation but not the subsectors. Measures of ALP can be constructed for the years since 1948 as the ratio of output to one of several alternative employment series.6 Although in principle the NIPA measure of output is gross product originating, that is, value added, in practice value added is calculated by double deflation only since 1977; prior to 1977 value added is calculated only for rail transport. Output in air and truck transportation is based on deflated gross revenue prior to 1977. Another set of ALP measures is provided by the Bureau of Labor Statistics (BLS) Office of Productivity and Technology over most of the postwar period.’ The data published by the BLS include gross output, employment, and output per employee for five transportation subsectors (the same as NIPA minus water transportation and transportation services, and with some defini5. The BLS publishes MFP series only for the total economy (private and private nonfarm) and for the manufacturing sector (see Mark and Waldorf 1983). MFP estimates are published at the disaggregated level for only four industries: tires and inner tubes, steel, footwear, and motor vehicles and equipment (U.S. Bureau of Labor Statistics 1990). 6. These are full-time and part-time employees (NIPA, table 6.6B), full-time equivalent employees (table 6.7B), and persons engaged (table 6.10B). All NIPA ALP measures in this paper are based on persons engaged. Results would be almost identical for rail and air using full-time equivalent employees, which make up 100 percent of persons engaged for rail and 99 percent for air, but not for trucking, where self-employment is more important. 7. Published BLS indexes begin in 1958; unpublished estimates for air and rail begin in 1947 and for trucking begin in 1954. See the notes to table 10.2. A general introduction to the BLS methodology for the indexes covering the service sector is provided by Dean and Kunze (chap. 2, in this vol.).

375

Productivity in the Transportation Sector

tional differences discussed below). Hours and output per hour are also included for railroads and bus carriers. Output is measured by physical output data reported by regulatory agencies; in the case of railroads raw data on tonmiles are adjusted for changes in the composition of goods carried.8 Data on employment and hours include the self-employed and come from the BLS establishment survey. An important conceptual difference between the NIPA and BLS series is that the BLS incorporates links when definitional changes occur in source data; the NIPA data do not. Below we find that this helps to explain the difference between NIPA and BLS estimates of airline employment. Table 10.2 provides our first detailed look at ALP data for the transportation sector and three subsectors. The NIPA data in the top section of table 10.2 duplicate those in table 10.1 for the three subsectors but differ for the total transportation sector by reporting output per employee rather than output per hour and by excluding the four minor transportation sector^.^ The NIPA output data for 1977-87 are the unrevised series published prior to 1991, presented here in order to highlight the sharp discrepancies between the NIPA and BLS data that in part motivated the recent NIPA revisions (subsequently we examine the revised output data in table 10.4 below). Growth rates are shown for intervals broken in 1958 (the starting year of the published BLS data), 1973, 1979, and 1987. The productivity growth slowdown in the final column compares 1973-87 with 1958-73 (not 1948-73 as in table 10.1). The post-1973 productivity growth slowdown is much larger in table 10.2 than table 10.1, mainly because pre-1973 productivity growth is held down in table 10.1 by the inclusion of local transit (where productivity collapsed, particularly during 1948-58). The BLS data shown in the middle section of table 10.2 tell a very different story from the unrevised NIPA data shown in the top section, particularly for 1979-87 when the growth rate of BLS ALP for total transportation exceeds that of NIPA ALP by 4.43 points per year.Io The BLS slowdown occurs entirely for airlines, and there is virtually no slowdown for railroads and trucking. The bottom section of table 10.2 subtracts each NIPA growth rate from the corresponding BLS rate and shows that the discrepancy was large for all three major subsectors over 1973-87. 8. This adjustment is based on Interstate Commerce Commission data on unit revenue for 200 commodity lines, see Mark (1988, 146-47). This source indicates that a similar adjustment was formerly made for trucking, but that the disaggregated commodity data from the source agency were discontinued at an unspecified date. 9. The “minor” sectors included in the transportation total in table 10.1 but excluded in table 10.2 and subsequent tables are local transit, water transportation, pipelines, and transportation services. 10. The BLS does not publish data for the aggregate transportation industry. In tables 10.2 and table 10.3 we use a quasi-Tornqvist index that takes the shortcut of aggregating over multiyear intervals (using the average shares in the first and last year of each interval), rather than of aggregating each year-to-year change and averaging these. The Tornqvist formula is shown to be one of the class of “superlative” index numbers by Diewert (1976). The same formula is labeled the Tornqvist-Theil-translogindex by Caves, Christensen, and Diewert (1982).

376

Robert J. Gordon

Table 10.2

Unrevised NIPA: Transportation Railroads Trucking Airlines BLS: Transportation* Railroads Trucking Airlines BLS - NIPA: Transportation Railroads Trucking Airlines

Growth of Output per Employee, Unrevised NIPA versus BLS, Selected Intervals, 1948-1987

1948-58

1958-73

1973-79

1979-87

1948-73

1973-87

2.96 2.17 3.59 7.39

4.15 4.65 3.78 3.96

1.06 1.37 0.18 2.72

-0.63 1.14 -0.62 -2.13

3.88 3.66 3.70 5.33

0.11 1.24 -0.28 -0.87

n.a. 1.75 n.a. 8.43

4.44 5.46 2.86 6.64

3.03 1.48 3.15 4.66

3.67 8.17 2.18 3.20

n.a. 4.46 n.a. 6.16

3.50 5.30 2.59 3.83

n.a. -0.42 n.a. 1.04

0.29 0.82 -0.92 2.68

1.97 0.11 2.97 1.94

4.43 7.03 2.80 5.33

n.a. 0.80 n.a. 0.83

3.39 4.06 2.87 4.70

Slowdown, 1973-87 1958-73

-4.04 -3.41 - 4.05 -4.83 -0.94 -0.16 -0.27 -2.81 3.10 3.25 3.79 2.02

Sources: NIPA: Output per employee is calculated as output from table 6.2, as published most recently in the July 1988 Survey ofcurrent Business, divided by persons engaged from table 6.10B. BLS: Output, employees, and output per employee for 1958-63 are from BLS bulletin no. 2296, 134-38, and for 1963-87 are from bulletin no. 2349, 142-46. For railroads and air transportation data for 1948-58 are available in unpublished computer printouts, BLS Office of Productivity and Technology, January 16, 1990. *The transportation aggregate for BLS is obtained by weighting the BLS growth rates of output and of total employment by a quasi-Tomqvist method. Output and employment growth in each subsector is weighted by the NIPA nominal output weight (table 6. I ) for the average of the initial and terminal year within each interval, e.g., the average of 1973 and 1979 weights for the 1973-79 interval. Aggregate transportation in NIPA includes railroad transportation, trucking and warehousing, and air transportation. The BLS aggregate includes railroad traffic (revenue traffic), intercity trucking, and air transportation. n.a. indicates “not available.”

Because ALP is the ratio of output to employment, the discrepancy between the BLS and NIPA data could result from differences in the treatment of output, employment, or some combination of both. A decomposition is provided in table 10.3, which expresses the difference between the BLS and NIPA annual growth rates of output in the top part of the table and of employment in the bottom part. Here we learn, surprisingly, that the puzzle for total transportation after 1973 lies almost entirely in the differing data on employment, albeit this aggregation disguises very large and offsetting differences for output growth in the four subsectors. 10.1.2 The NIPA Output Revisions and Remaining Discrepancies

In earlier versions of this research, beginning with Baily and Gordon (1988), we showed that the slow growth in the unrevised NIPA output series for railroads and airlines relative to the more rapid growth of the BLS output

377

Productivity in the Transportation Sector

Table 10.3

Difference between BLS and Unrevised NIPA Estimates of Output and Employment, Annual Percentage Growth Rates, Selected Intervals, 1958-1987 Slowdown, 1958-73

output: Transportation Railroads Trucking Airlines Employment: Transportation Railroads Trucking Airlines

1973-87 1958-73

1973-79

1979-87

1973-87

0.98 1.16 0.38 2.36

-0.18 -0.19 -0.56 0.76

1.17 6.87 - 2.03 3.42

0.67 3.84 - 1.40 2.28

-0.31 2.68 - 1.78 - 0.08

0.69 0.35 I .30 -0.31

- 2.19

-3.13 -0.16 - 4.83 - 1.91

-2.71 -0.22 -4.27 - 1.60

-3.98 -0.57 -5.57 - 1.29

-0.30 - 3.53 - 1.19

Sources and notes: Same as table 10.2

series could be traced to overdeflation. In particular, the NIPA price deflators for airline output and for consumer expenditures on airline transportation made little or no allowance for discount fares in the 1977-83 period and thus rose much too quickly, causing deflated gross revenues to increase much too slowly. The same problem appears to have plagued the previous NIPA railroad deflators. Responding to this criticism, the revised NIPA industry gross output estimates have shifted from deflated gross revenue to physical volume measures (as well as shifting to double deflation, i.e., subtracting purchased inputs, for trucking and airlines, as was done previously for railroads). The top section of table 10.4 shows that the revised NIPA indexes for 1977-87 now rise faster than the BLS indexes for all three subsectors; previously this was true only for trucking. The revision for railroads is an astonishing 7.5 percent per annum, and for airlines a smaller but substantial figure of 4 percent per annum. Nevertheless, as shown in the middle and bottom sections of table 10.4, the BLS series on ALP in total transportation, as well as for the trucking and airline subsectors, rises faster than the NIPA ALP series, despite more rapid growth of NIPA output. This occurs because the BLS registers slower growth in employment in each sector. Although the difference for railroads is not important, that for trucking and airlines makes a substantial difference. 10.1.3 Sources of Employment Discrepancies By far the most important remaining discrepancy concerns trucking employment. An important definitional difference between NIPA and BLS is that the former includes all trucking (intercity and local), as well as warehousing; BLS includes only a fraction of intercity trucking. Table 10.5 displays the 1979 and 1987 values, and 1987/1979 ratios, for a variety of measures of

378

Robert J. Gordon

Table 10.4

Growth Rates for Revised NIPA, Unrevised NIPA, BLS, and Differences for Output and Output per Employee, for Interval 1917-1987 Unrevised NIPA

output: Transportation Railroads Trucking Airlines Employment: Transportation Railroads Trucking Airlines Output per employee: Transportation Railroad Trucking Airlines

Revised NIPA

BLS

BLSUnrevised NIPA

BLSRevised NIPA

0.52 -4.65 1.18 3.36

3.67 2.80 2.26 7.42

1.46 0.90 -0.59 6.31

0.94 5.55 - 1.77 2.95

-2.21 - 1.90 - 2.85

1.04 -5.76 2.15 4.39

1.04 .5.76 2.15 4.39

-2.74 -6.16 -3.87 2.52

-3.78 -0.40 -6.02 - 1.87

-3.78 -0.40 -6.02 - 1.87

-0.52 1.11 -0.97 - 1.03

2.63 8.56 0.11 3.04

4.20 7.06 3.28 3.79

4.75 5.95 4.26 4.82

-1.11

1.57

- 1.50 3.17 0.75

Sources: Same as table 10.2, except revised NIPA output from de Leeuw, Mohr, and Parker (1991). table 6, 34.

nominal and real output, price indexes, and employment in the trucking industry. The data include both measures for the comprehensive truckingwarehousing universe partially covered by the NIPA, and the intercity subsector covered by the BLS. To summarize our conclusions in advance, we find that the NIPA data correspond closely to independent measures of the trucking universe, but that the BLS data are badly biased by including only a part of the intercity subsector that has experienced a sharply reduced share of output and employment as a result of deregulation. The nominal output data in section 1 of table 10.5 show a close correspondence for the 1987/1979 ratio of, respectively, NIPA nominal output and a related measure called “outlays for highway freight transportation” (which includes both intercity and local transportation). A separate series for intercity class I carriers (line Id) indicates a much slower increase in revenue, resulting from a shift in the composition of intercity freight away from class I carriers. Three price series are shown in section 2, the NIPA implicit deflator, an implicit price series that results when the intercity outlays series in line l c is divided by the output series in line 3b, and a direct measure of revenue per ton-mile for class I intercity freight. The implicit intercity price increases at about the same rate as the NIPA deflator; the direct measure of revenue per ton-mile increases less. Because all three deflators in section 2 refer to intercity freight, they should be viewed as different measures of the same con-

379

Productivity in the Transportation Sector

Table 10.5

Comparison of Data on Nominal and Real Output, Price Indexes, and Employment for Total and Intercity lkucking, 1979 and 1987 1987/ 1979(%)

1979

1987

41.4 142.7 90.2 30.1

65.2 220.3 132.8 35.0

157.5 154.4 147.2 116.3

74.7

99.8

133.6

14.8 11.6

19.9 14.1

134.7 121.6

55.4 608 36.4

65.3 666 35.0

117.9 109.5 96.2

104.3

94.3

90.4

1. Nominal output (in billions of dollars):

a. Revised NIPA (table 6.1) b. Outlays on highway freight c. Outlays on intercity freight d. Operating revenue, class I intercity freight carriers 2. Price indexes: a. Revised NIPA implicit deflator for trucking output, 1982 = I W ( l d 3 a ) b. Intercity outlays per ton mile (in cents) (Ic13b) c. Class I intercity revenue per ton mile (in cents) 3. Real output: a. NIPA output in 1982 dollars (table 6.2) b. Intercity freight ton miles (in billions) c. Implicit real revenue (in billions of 1987 dollars), Class I intercity freight carriers d. BLS output index (1977 = 100) 4. Employment (in thousands) a. NIPA (no. of persons engaged) b. BLS trucking and warehousing employment c. Class I intercity freight carriers d. BLS employment level

1498 1340 575 57 1

1674 1464 519 434

111.7 109.2 93.3 76.0

Sources by line: (la,2a,3a) de Leeuw, Mohr, and Parker (1991), tables 5 and 6, 33-34. (3d) Basic BLS source, same as table 10.2. (lb,lc) Statistical Absrracr, 1989, table 998. (ld,4c) 1979-80, TRINC Associates, linked for 1980-87 to Statistical Abstract. 1990, table 1055, sum of figures given for common carrier general freight, common carrier other than general freight, contract carrier other than general freight, and carriers of household goods. (2b) lc/3b. (2c) Narional Transporfarion Statistics, annual report 1989, U.S. Department of Transportation for 1977-87, 1981 issue for 1969-76, 1972 issue for 1960-68. (3b) Statistical Abstracr, 1989, table 1OOO. (3c) Equals line Id for 1987. For 1979 equals line Id for 1979 times 198711979 ratio from line 2c. (4a) Basic NIPA source, same as table 10.2. (4b) Staristical Abstract, 1989, table 999, totals given for SIC 421,422, and 423. (4d)Source of BLS employment data provided by Edwin Dean (American Trucking Association, 1987 Motor Carrier Annual Report), lists total employment in 1987 as 349,842. To this is added 84,000 leased drivers, as stated in a letter from Dean. 1979 employment equals 1987 employment times the 1979/1987 ratio of the BLS trucking employment index.

cept." We view the final measure in line 2c as superior, as it is a direct measure of revenue yield per ton-mile, rather than an implicit ratio of numerator and denominator that may not cover the same universe. The intercity output series on line 3b rises at about the same rate as the NIPA real output series; a constructed series (line 3c) for the implied real 11. The source listing provided by de Leeuw, Mohr, and Parker (1990, table 3) indicates that the nominal value is based on class I motor carriers and real output is based on a physical measure of ton-mile volume, which could only refer to intercity freight, as ton-miles for local traffic are not available.

380

Robert J. Gordon

revenue of class I intercity carriers based on the implicit price series from line 2c declines somewhat slower than the BLS output series for Class I and I1 intercity carriers (line 3d), as would be consistent with the evidence presented below that the BLS has been measuring a shrinking fraction of the intercity trucking industry. l 2 The employment data display the same ranking of 19871 1979 ratios as the output data, except that the BLS employment series shows even more relative shrinkage, contributing to the relatively favorable performance of the BLS productivity series examined previously in table 10.2. To track down the source of the rapid decline in the BLS employment series, we have attempted to reconstruct the absolute level on which the BLS series is based in 1979 and 1987 (see source notes to table 10.5). If these figures are correct, they imply that coverage by the BLS of the NIPA employment total fell sharply from 38.1 percent in 1979 to 25.9 percent in 1987.13 In our detailed examination of the trucking industry in part 10.5 below, we learn that there was a huge shift in the composition of firms in the intercity trucking industry as a result of deregulation. The BLS, by choosing to cover a portion of the industry that is declining in importance, has misrepresented employment trends in the industry as a whole. This leaves as a mystery why the segment of the industry covered by the BLS exhibits healthy productivity growth over 1979-87; NIPA productivity growth for the trucking industry as a whole is a barely positive 0.7 percent per annum slightly (line 3a divided by 4a).I4 If both the NIPA and BLS productivity data are correct, they imply a slight decline in the absolute level of ALP between 1979 and 1987 for the part of the NIPA trucking universe not covered by the BLS.I5 Because of its much greater coverage, the NIPA output and employment series are preferable to those of the BLS. There remains a potential measurement error in the NIPA output series, because of the possibility of an overly rapid increase in the implicit deflator. The direct measure of class I revenue per ton-mile rises 1.2 percent per annum less than the NIPA deflator. Support12. The intercity freight output series on line 3b comes from a source that allows the relative share of railroad and trucking output to be computed; these shares are almost identical to those in data independently collected by Winston et al. (1990, table 1-1). 13. We were unsuccessful in locating additional independent sources of trucking employment over the full 1979-87 period. In particular the TRINC data used in table 10.5 for 1958-80 are not available after 1983. 14. Despite its tantalizing title, the recent article by Ying (1990) contains only estimated parameters allowing a calculation of the marginal effect of deregulation on trucking productivity, but no data on the level or rate of change of actual productivity. 15. If revenue per employee were the same in the BLS and non-BLS part of the total NIPA trucking universe at the 1987 level of $78,876 reported by the BLS source (American Trucking Association, 1987 Moror Carrier Annual Report, summary table I , col. 7, then the implied 1987 revenue figures are $34.2 billion for BLS, $97.8 billion for non-BLS, and $132.0 billion for the total. Using NIPA real output to extrapolate the total back to a 1979 figure of $124.1 billion real revenue in 1987 dollars for the total, and the BLS output index to obtain a 1979 figure of $37.8 billion for the BLS segment, the implied non-BLS real 1979 revenue is $86.3 billion. Implied non-BLS real revenue per non-BLS employee fell from $93,096 to $78,876, for an implied decline in non-BLS productivity of 15.3 percent.

381

Productivity in the Transportation Sector

ing a slower price increase is the contrast of the 33.6 percent 1979-87 increase of the NIPA deflator with the increases in the prices of inputs, 35.8 percent for labor and 28.6 percent for diesel fuel.l6Output prices should have increased less than input prices if there was an improvement in labor productivity and fuel efficiency; the improvement in fuel efficiency is a solid fact; labor productivity increased even with the NIPA deflator and even more with the alternative deflator.” In part 10.5 we explore the consequences of replacing the NIPA output index with an alternative index based on the deflator in line 2c of table 10.5. In the airline subsector NIPA employment also grows substantially more rapidly than BLS employment, but here the discrepancy is resolved in favor of the BLS series. The most important cause of this difference, also uncovered by Card (1989, table lO.l), is that Federal Express was added to industry output and employment figures in 1986. Because Federal Express carries high-value shipments, it has an extremely low ALP measured as ton-miles per employee, less than one-tenth that of American Airlines in 1989.18Thus the introduction of Federal Express into the statistics introduces a spurious downward shift in the ALP of the airline industry that the BLS handles by linking out Federal Express output and employment. A superior approach, but one with more onerous data requirements, would be to follow Caves, Christensen, and Tretheway (1981, 1983, 1984) by constructing a Tornqvist output index that weights different output components by their revenue shares. Because it recognizes the Federal Express problem and makes two other links to improve comparability, we deem the BLS output and employment data to be superior to those in the NIPA and use them in part 10.3 below.I9 10.1.4 Choice of Series for Further Study Subsequent sections of this paper develop new measures of MFP for the three transportation subsectors. Our desired output concept is gross rather than value added, because we want to include fuel and materials inputs explicitly in the MFP calculation. The BLS output measures have the double advantage that they explicitly measure gross output and are conceptually consistent over the postwar period; the NIPA output series is inconsistent, measuring 16. Labor cost is compensation per full-time equivalent employee, NIPA table 6.4B divided by table 6.7B. The fuel cost is the retail price of diesel fuel, from American Trucking Trends. 17. Average miles per gallon for single-unit trucks increased by 14 percent from 1979 to 1986 (American Trucking Trends 1987, 44). The 1979-87 percentage increase in ALP is 5.5 percent for the NIPA deflator (table 10.5, line 3a/4a) and 16 percent for the alternative deflator. 18. Making the arbitrary assumption that Federal Express shipments travel 700 miles on average, one can calculate from its 1989 annual report an average of 10,233 ton miles per employee, in contrast to American’s 115,716 (ton miles per “average equivalent employee,” from an American Airlines, annual report). 19. According to Richard Carnes of the BLS, the two other links occur in the 1979-81 period were made necessary by the elimination of the distinction between certificated and noncertificated carriers, and a major shift in coverage of small carriers.

382

Robert J. Gordon

value added throughout only for railroads, while switching in 1977 from gross output to value added for trucking and airlines. Although it would be desirable to use the BLS indexes throughout for consistency, the above analysis of data discrepancies suggests that a mixed set of sources is superior. Railroads. The BLS and NIPA employment series are very close, so the choice of the BLS series raises no problem. However, since 1977 the NIPA railroad output series rises almost 2 percent per annum faster than the BLS output series. About half of this difference reflects the BLS practice of weighting several hundred traffic classes by unit revenue weights, which approximates the practice of Tomqvist aggregation advocated by Caves, Christensen, and Tretheway (198 1) and is conceptually superior to the Bureau of Economic Analysis (BEA) index that is based on unweighted ton-miles. The remaining half of the difference reflects the distinction between gross output and value added; the latter increases more rapidly as a result of increased fuel efficiency. Both of these differences point to the use of the BLS gross output series for railroads and adjusting explicitly for fuel efficiency. Trucking. We concluded above that the NIPA output and employment series for trucking are much superior to the BLS series, which cover a shrinking segment of the industry. Because the NIPA output series represents value added since 1977, our MFP index for trucking since 1977 should not adjust for fuel and materials inputs, because this would amount to subtracting these inputs twice. Airlines. As noted above, the BLS employment series for airlines incorporates adjustments that make it superior to the NIPA series, and for consistency we also use the BLS output series. For 1977-87 the revised NIPA output series grows only about one percent per annum faster than the BLS output series, and much of this may reflect increased fuel efficiency that we take into account separately.

10.2 Conceptual Issues 10.2.1 MFP Growth and the Cost-Function Approach The production process in transportation is well described by the standard economic theory of production, with a few unique features. Because the formal interpretation of MFP indexes within the cost-function approach has been clearly developed elsewhere, here we limit the discussion to the implications for the MFP indexes that we develop subsequently.20

10.2.2 Issues in the Estimation of MFP Growth The cost-function approach emphasizes that standard measures of MFP growth are equivalent to the shift in the production function and cost function 20. See Denny, Fuss, and Waverman (1981, 187-95) and the appendix in Good, Nadiri, and Sickles (1989).

383

Productivity in the Transportation Sector

only in the presence of constant returns to scale. With increasing returns, the growth of MFP exaggerates the shift in the production and cost functions by including the contribution of economies of scale to economic growth. Because the proper measurement of returns to scale requires data on outputs and inputs at the level of the firm or establishment, the findings in this paper based on industry-level data must be qualified to the extent that more disaggregated studies have determined that nonconstant returns to scale are important. Other issues emerging from the cost-function literature include departures from marginal cost pricing and effective rate-of-return regulation. The first of these appears to be most important in industries that practice cross subsidization, as in the case of telephone communications studied by Denny, Fuss, and Waverman (1981), and involves the mismeasurement of output growth because of the application of incorrect weights in aggregating outputs and inputs. We are able to sidestep this issue in studying the transportation sector, because it is of secondary importance. Although airlines and railroads produce multiple outputs, their revenues are overwhelmingly dominated by a single product, scheduled passenger travel in the case of airlines and freight carriage in the case of railroads. The second issue, rate-of-return regulation, is clearly relevant for transportation. Denny, Fuss, and Waverman (1981, 199) show that, if prices of expensed factors of production and the allowed rate of return are increasing over time, then estimates of technical change that ignore rate-of-return regulation overestimate the true underlying rate of technical change. This finding is important for any investigation that includes the period of deregulation, because it could lead to an erroneous conclusion that the rate of technical change had been decreased as a result of deregulation. Although we make no adjustment for this potential bias in our study of railroads and trucking, we have sufficient data to decompose changes in airline efficiency into changes achieved by aircraft manufacturers and changes in the intensity of use of aircraft, particularly changes in load factors and in the seating density of given aircraft, that may reflect in part the influence of regulation and subsequent deregulation. Hulten (1986) and Berndt and Fuss (1986) have emphasized a problem in productivity measurement that applies to any industry, not just to the regulated sector. If output is produced by capital services, that is, by the utilized portion of the capital stock, then conventional measures of MFP growth based on data on the capital stock (implicitly assuming constant utilization) err by treating the effect on productivity of changing utilization as a shift in the production function. Below in table 10.15 we address this issue by providing estimates of MFP growth that are adjusted for changes in utilization in the national economy. 10.2.3 Causes of Changes in MFP We conclude part 10.2 by discussing causes of productivity change that are common to different subsectors of transportation, and reserve for the remain-

384

Robert J. Gordon

ing sections of the paper a detailed consideration of those causes that are specific to particular subsectors. I . Unmeasured changes in the quality of output. Because it mainly provides a consumer service rather than an intermediate input, air transportation raises more questions of unmeasured quality change than do rail and trucking. Computers, for instance, have produced unmeasured quality deterioration in the form of restrictions and penalties on airline tickets, balanced by advance seat selection and boarding passes, frequent-flyer awards, and the potential welfare gains of price discrimination to price-sensitive travelers. Other dimensions of quality change include the benefits of increased speed made possible by improved aircraft, the effects of congestion, noise, flight frequency, waiting time, and safety. Both noise and pollution are relevant for railroads and trucking, as is the increased speed of rail shipments made possible by deregulation. 2. Quality of inputs, especially capital. In the macrosources-of-growth literature there is a substantial controversy about the effects on MFP of changes in labor quality. Having summarized the issues recently, we say nothing new about this here (Baily and Gordon 1988, 370-76). Here our main emphasis is on changes in the quality of capital. The growing literature on computer prices, recently surveyed by Triplett (1989), has yielded a consensus that the proper measure of utilized capital input that appears in the production function is a vector of input characteristics of capital, defined as any attribute of a capital good that has a positive marginal product, including the horsepower and physical dimensions of a truck, or memory size and speed for a computer. Recently (Gordon 1990a) I have constructed a number of new deflators for investment goods; my approach to price measurement for capital goods emphasizes the need for accurate attribution of quality changes among producers and users of capital goods.*’ Manufacturers should be “credited’ not only with improvements in performance, but also with cost-saving innovations in energy efficiency, durability, and maintenance costs. To make sense in conjunction with my quality-adjusted measures of real capital input, calculations of MFP growth must include fuel or energy as an input. My method credits equipment manufacturers for improvements in fuel economy that are not accompanied by proportional increases in real equipment cost. Thus new technology that improves fuel efficiency enters the calculation of transportation MFP growth as an increase in the growth of capital input (which reduces MFP growth) and is balanced by a decrease in the growth of fuel input (which boosts MFP growth). If the calculation is done properly, the faster capital input growth and slower fuel input growth exactly offset each other and no change occurs in transportation MFP growth. This is the correct conclusion, because by assumption the technical achievement occurs in the manufacturing sector, not in the transportation sector. The many recent detailed studies of productivity growth in transportation have devoted 21. A brief summary of the methodology and results of this book-length study is available in Baily and Gordon (1988, 377-84).

385

Productivity in the Transportation Sector

remarkably little attention to the issue of capital quality, and hence in this example credits the transportation sector for faster MFP growth that has been achieved elsewhere.22

10.3 Air lkansportation 10.3.1 The Long-Run Behavior of Productivity and Relative Price The U.S. airline industry commenced operations in the late 1920s, and by 1935 almost all of today’s largest domestic airlines were operating under their present names. Total industry output in 1987 exceeded that in 1935 by a factor of 1650, for an annual growth rate during the intervening years of 14.2 percent. The growth performance since 1935 is summarized in the top half of table 10.6. ALP growth marched along at a rock-solid 7.1 percent throughout the period 1935-69, even though output growth in the two decades after 1948 fell by half compared to 1935-48. This casts doubt on the importance of increasing returns in the long run, because the post-1948 decline in output growth should have reduced ALP growth if scale economies were important. The bottom half of table 10.6 displays the ratio of United Airlines output and productivity to that for the air transport industry as a whole. Although United was the largest airline during 1931-38 and again from 1961 to 1988, there is no evidence that it gained any advantage from its large scale. In fact, its ALP grew slightly slower than that for the industry, 5.73 versus 6.25 annual percentage points, respectively. If an industry enjoys ALP growth that is more rapid than for the economy as a whole, its real price should decline. The final column of table 10.6 shows that this occurred for the airline industry during 1935-87, although the relationship is not exact, as the relative price of an industry’s output depends not only on relative ALP growth but also on changes in relative input costs and in the relative productivity of factors of production other than labor. Our inference that the airline industry is subject to constant returns in the long run accords with the view originally established by R. Caves (1962) and reinforced by Douglas and Miller (1974) and White (1979). Recently, D. Caves, Christensen, and Tretheway (1984) find economies of scale to “density,” adding more flights per city served, but agree with the previous literature that larger firm output accompanied by an increased size of network, holding density constant, is subject to constant returns. We return below to the effects of deregulation on route structure and density.23 22. Many papers on airline productivity cite the detailed panel data set constructed by Caves, Christensen, and Tretheway (1981) and extended in subsequent papers. These authors carry out a detailed aggregation of major aircraft types, as do we in part 10.3 below, but they weight each aircraft type by its lease cost. If lease cost is proportional to purchase price, then their procedure is equivalent to assuming that the input characteristics of different models of aircraft differ in proportion to their purchase price, which greatly understates the quality of newer models. 23. Caves, Christensen, and Tretheway (1981) also show that there are systematic differences in managerial efficiency over time that are not related to scale. In reporting these results, they stress their agreement with the results of my first professional paper (1965).

386

Robert J. Gordon

Table 10.6

Year 1935-41 1941-48 1948-59 195949 1969-78 1978-87

Long-Run Behavior of Output, Employment, and Passenger Yield, Airline Industry and United Airlines, 1985-1987 Revenue Ton Miles

Employees

Output per Employee

Real Passenger Yield

Annual Growth Rate, U.S. Domestic & International Scheduled Air Carriers 26.43 19.21 7.07 -4.02 24.58 16.61 7.08 -5.52 - 2.82 13.58 6.03 7.05 13.48 6.42 7.06 -2.72 6.26 0.60 5.66 -2.51 5.81 2.03 3.78 - 1.92

Ratio, Index of Each Variablefor United Airlines to Index for Air Transport Industry (1978 = 1.0) 1935 1.48 1.22 1.22 1.15 1941 1.11 0.84 1.32 1.03 1948 0.89 0.81 1.09 1.01 1959 0.84 0.80 1.06 1.08 1969 1.08 1.03 1.05 1.oo 1978 1.oo 1.oo 1 1 .oo 1987 0.95 1.03 0.93 1 .oo

.oo

Sources: For 1948-87, industry output and employment are obtained from the same sources as table 10.4. For 1935-48, data are obtained from the CAB Handbook ofAirline Staristics. Domestic revenue ton miles were linked to total revenue ton miles prior to 1943. Real passenger yield is passenger revenue divided by revenue passenger miles times the GNP deflator. United Airlines data come from company annual reports, selected issues.

Our treatment of airline productivity treats two main topics, unmeasured changes in output quality and new measures of inputs (especially capital). Improvements in output quality can be achieved both by aircraft manufacturers and by airline operators. The most dramatic changes in quality prior to the 1970s occurred as manufacturers made possible the shift to larger and faster piston planes, and then to jet aircraft; these are treated below in the context of input measurement. First we examine issues in the changing quality of airline output achieved within the airline industry itself, and this concentrates on the period since deregulation in the late 1970s, an interval during which interval the quality of aircraft has been relatively stable.

10.3.2 Output Quality: The Productivity Effects of Hubbing Airline deregulation is widely believed to have substantially changed the production process by shifting airline service from nonstop point-to-point service to connecting service through hubs, thereby increasing flight mileage to travel between origin and destination. In the upper left-hand of figure 10.1, the dashed line indicates the nonstop flight between origin A and destination B flown prior to deregulation, and the solid lines show the roundabout route through hub H1 flown after deregulation. If correct, this “standard model”

387

Productivity in the Transportation Sector B I

I I I

I I I

I l

a

I I

I I I

1

H1

Standard Model

New Model: Large Cities

KEY:

- - - - - - - - - Pre-deregulation Post-deregulation H3

New Model: Small Cities

Fig. 10.1 Routing effects of airline deregulation: Standard model compared to new model

would have the important implications that official measures of output in the 1980s overstate true output measured from origin to destination and that measures of yield understate the true origin-to-destination price. This standard view is frequently encountered in academic work,24 and it appears to be universally held by journalist^.^^ 24. McGowan and Seabright (1989, 326, 329) support verbally the graphical interpretation in the top left frame of figure 10.1: “a traveller from A to B takes off and lands twice instead of once, takes longer to reach the destination, travels further in total and may have to suffer the inconvenience of changing aircraft and an increased risk of baggage loss or missed connections. . . . it is important, therefore, that the true social costs of making indirect rather than direct flights should be borne by carriers.” Similarly, Good, Nadiri, and Sickles (1989, 7) state that “increased use of hub-and-spoke and loop type networks . . . allow carriers to increase load factors, but they artifically inflate the level of real production by increasing the air miles between cities and by reducing the likelihood of non-stop service.” 25. Samples include “instead of flying a ‘linear’ route system, with criss-crossing services between cities, airlines have developed more efficient hub-and-spoke systems” (The Economist,

388

Robert J. Gordon

The most widely cited advocate of the standard view is Dempsey (1990), who claims that the hub-and-spoke system has caused passengers to fly between 5 percent and 30 percent additional mileage on a given trip, implying that a portion of productivity gains measured by passenger-miles is illusory. Dempsey uses this finding (1990, 32) sharply to criticize the cost-benefit analysis of deregulation by Morrison and Winston (1986) for failing to take account of the time cost of “greater circuitry attributable to hub and spoking.” Although it might seem from Dempsey’s critique that the output and price data examined in part 10.1, above, are flawed by failing to adjust for circuitry, in fact the issue is of trivial importance. Borenstein has estimated that, if all domestic air travel were nonstops and there were no connections at all, total domestic flight mileage would be reduced only by about 4 percent, but of course there were plenty of connections before as well as after deregulation, so that the net circuitry effect must have been much less than 4 percent even if the percentage of flights involving connections has increased substantially.26 In assessing unmeasured aspects of quality change in airline output, the issue of connections and hub-and-spoke routings is central. Justifying a new assessment is that academic studies by Morrison and Winston (1986, 1989) and others use data for 1983 and earlier years produced by the U.S. Civil Aeronautics Board (CAB) prior to its 1984 “sunset.” There is virtually no evidence available for any recent year that takes account of the 1986-87 wave of mergers and the failures of numerous new entrants.27 To provide a fresh look at the routing opportunities available to travelers, we have assembled a virtually complete census of routes, and of the daily number of flights per route, flown by the air transportation industry within the 48 continuous states in August 1978 and August 1989. The results, summarized in tables 10.7 and 10.8, unambiguously contradict the standard model and reflect two simple facts. First, surprisingly few nonstop routes involving medium and large cities were discontinued. Second, critics overlook the fact that millions of people actually live in metropolitan areas where new hubs were established; the number of new nonstop hub-to-hub and hub-to-spoke routes from these new hubs greatly outnumber the small number of discontinued nonstop routes. This new model is shown in the upper-right frame of figure 10.1; deregulation allows new nonstop service from city A to new hub, H2, thus eliminating the circuitry of detouring via an old hub H 1.28 March 10, 1990, 73); “They built hub-and-spoke route systems . . . rather than a web of direct, non-stop flights” (The Economist, January 26, 1991,57); there are “far fewer direct flights” (New York Times, January 2, 1991. Al); “Many travelers now must fly farther to reach a given destination because of hub-and-spoke systems , , , yield can decline even though passengers are paying more for their tickets’’ (WallStreerJournal, April 19, 1990, B I ) . 26. The 4 percent figure is from correspondence to the author from Severin Borenstein, dated May 20, 1991, and is calculated from the Department of Transportation data base for the second quarter of 1986. 27. An exception is Borenstein (1991). to which we return below. 28. The ability of deregulation to open up new nonstop routes bypassing traditional hubs was recognized immediately by perceptive observers, whereas previously, for instance, “everyone in

389

Productivity in the Transportation Sector

Table 10.7

Effect of Deregulation on Nonstop Domestic Air Service, Top 500 Origin-Destination Markets, August 1978 and August 1989 1978 Routes

Flown both years: Hub to hub’ 71 Hub to nonhub 187 71 Nonhub to nonhub Total 329 Flown one year, not the other: 1 Hub to hub Hub to nonhub 11 Nonhub to nonhub 5 Total 17 Flown neither year: 93

1989 Flights

Routes

Flights

116

171

... 1 19 6 26

...

42 329 6 47 8 61 93

16 123 16 155 ...

Source: O@cial Airline Guide, North American Edition, August 1, 1978, and August 1, 1989 Note; The 500 top markets are ranked by revenue passenger miles, from Department of Transportation origin and destination survey, table 7, for the 12 months ending December 30, 1986. ‘The hub airports include both the original hubs and new hubs. See the listing of hubs in the notes to table 10.8.

Some accounts treat hub-to-spoke routings as a byproduct of deregulation. However, on-line connections date back to the dawn of the airline age, and the first hub operations on today’s scale began when Chicago’s O’Hare airport terminal complex was opened in 1962.29By the time deregulation occurred in 1978, United at O’Hare, as well as Delta and Eastern at Atlanta, were ulreudy operating full-scale hubs, each with roughly 250 departures per day. Prior to deregulation passengers were forced to make connections, just as they are today, but many more of those connections were interline rather than on-line, and more involved double connections. Between 1978 and 1989 interline connections fell by a factor of 10, from 41 percent of all connections to 4 percent (see table 10.10 below, sec. Id). When markets are ranked by passenger-miles, there are many long-haul markets that lacked nonstop service in both 1978 and 1989, but many more that gained service than lost service.3oThis contrast is shown in table 10.7, which provides a decomposition of nonstop routes served in the top 500 origin-destination markets (accounting for 60 percent of traffic measured by

the Carolinas or Virginias had to change planes to get beyond Atlanta or New York” (Baumgarner 1979,47). 29. This statement is supported by the American Airlines annual report for 1983, which reports that the opening of Chicago’s O’Hare terminal in 1962 represented the initiation of American’s first “true hub” (8). 30. Here it is important that markets be ranked by origin and destination passengers, i.e., the city pairs where people actually want to travel, and not by enplaned passengers on particular citypair segments, which of course respond to where the flights are actually operated.

Table 10.8

Effect of Deregulation on Nonstop Domestic Air Service, AIL Markets, August 1978 and August 1989 1978

1989

Change

Frequency*

Flights

Routes

Flights

Routes

Flights

1978

1989

97 166 468 358 1089

9 4 4 ~ 1068(23) 2394(269) 355( 1515) 4761( 1807)

3 51 77 78 209

129(0) 458(33) 376( 179) - 142(885) 82 I ( 1097)

8.7 5.4 5.4

280 880

815(0) 610( 10) 2018(90) 497(630) 3940(730)

5.3

9.7 6.6 5.7 5.2 6.0

29 146 63 238

91(18)

467(39) 53( 138) 61 1(195)

55 350

610

20s

255(69) 1263(284) 187(669) 1705(1022)

26 204 142 372

164(51) 796(245) 134(531) 1094(827)

3.8 3.5 3.0 3.4

6.0 4.4 42 4.5

175

653(58)

175

608(368)

0

- 45(3 10)

4.0

5.6

44

73(3)

61

123(44)

16

50(41)

1.7

2.8

- 44( 162)

3.2

4.4

- 82( - 30)

2.0 -

Routes 1. Original hubs:

a. To original hubst b. To new hubs c. To large nonhubs d. To small nonhubs e . Total 2. New hubs: a. To new hubs' b. To large nonhubs c. To small nonhubs d. Total 3. Large nonhubs: To large nonhubst a. Served both years b. Not other year To small nonhubs: c. Served both years d. Not other year e. Total

94 1 I5 39 I

118

162(216)

118

1 18(378)

0

126 463

108(150) 996(427)

44

26( 120) 875(910)

- 82

398

- 66

-

121(483)

4-0

2.6

4.0

4. Between small nonhub:’.’ a. Served both years b. Not other year c. Total 5 . Summary:$ a. All hubs b. Large nonhubs c. Small nonhubs

139 165 304

132(143) 60(282) 192(425)

2

139

47(353)

0

189

W446)

1233 lo00 647

5 l61(935) 3479(556) 742( 1 193)

1865 1215 752

7534(2852) 4532(1051) 606(2630)

2.0

2.9

- 115

-85(210) -43(-189) - 128(21)

2.0

2.7

632 215 105

2373(1937) 1053(495) - 136(1437)

4.9 4.0 3.0

170- 115

2-1

2.2

5.6 4.6 4.3

Source: Oficiul Airline Guide, Norrh American Edirion, August 1, 1978, and August 1 , 1989. Nores: First-listed count of flights is for jets, subsequent count in parentheses is for turboprops. The listing of routes and flights in this table includes only airports in the 48 continuous states and excludes all service from these airports to Alaska, Hawaii, or foreign countries. Every route and flight is included, except as indicated in note c, and except among cities too small to be classified as “small nonhubs.” Flight totals ignore weekend exceptions; a flight is counted as one daily frequency if it operates four or more days per week. Dejinirions: (1) Hubs: Hub airports include all those in which at least one airline operated a substantial number of on-line connecting flights in 1989. New hubs are those in which one or more airlines performed a hub operation in 1989 but not 1978 and include Baltimore, Charlotte, Chicago Midway, Cincinnati, Dayton, Detroit, Nashville, Newark, Philadelphia, Phoenix, Raleigh-Durham, Salt Lake City, and Washington Dulles. The remaining hubs are classified as original hubs and include Atlanta, Chicago O’Hare, Cleveland, Dallas-Ft. Worth, Denver, Houston, Kansas City, Los Angeles, Memphis, Miami, Minneapolis, Pittsburgh, St. Louis, and Washington National. (2) Size: Small nonhubs had nonstop service to no more than two hubs (new or original) in at least one year but had nonstop service to at least one hub in at least one year. Any airport with more than two routes to a hub in one or both years is classified as a large nonhub; airports with no routes to any hub in either year are excluded. Major airports classified as large nonhubs include Boston, Buffalo, Columbus, Oh., Indianapolis, N.Y. LaGuardia, N.Y. Kennedy, Orlando, San Diego, Seattle, and Tampa. *“Frequency” indicates total flights per route per day, including both jets and turboprops. ‘Routes and flights between airports within a single category are adjusted to eliminate double counting. *The listing for flights between small nonhubs is based on a 50 percent sample (all cities with names beginning “A” through “L”, which account for 49 percent of the pages listing flights in both the 1978 and 1989 source). §Summary totals are not adjusted to eliminate double counting; hence the total of routes and flights in section 5 is greater than the sum of routes and flights in sections 1-4, inclusive.

392

Robert J. Gordon

passenger-miles). Fully 422 of the 500 top markets show no change in the status of service, in that routes were either served nonstop or not in both years. In the remaining 78 markets, those adding nonstop service outnumbered those losing nonstop service by a margin of 61 to 17. Average frequencies (flights per day) in the discontinued markets were just 1.5, but were 2.5 in the markets adding service. Further, many 1978 nonstop markets were served sparsely, so that many passengers were forced to make stops or connect if they did not want to travel at the time of a single nonstop (e.g., nonstop service from Boston to Dallas increased from a single nonstop in 1978 to 9 per day in 1989). Critics, including Dempsey, imply that nonhub cities on the periphery of the 48 states have suffered particularly severe declines in nonstop s e r ~ i c e .Tak~’ ing as examples Boston, San Diego, and Seattle, nonstop routes from these three major cities to the other 24 of the top 25 largest metropolitan areas increased from 44 in 1978 to 56 in 1989 (out of a possible of 72). The complete census of domestic airline routes and flights appears in table 10.8. Airports are divided among four categories: original hubs, new hubs, large nonhubs, and small n o n h u b ~ The . ~ ~ number of routes served increased not only in every category involving hubs but also in routes between large nonhubs. Taking the categories in table 10.8 from line l a through 3b, which account for 90 percent of flights in 1978,33the number of routes served increases by 45 percent, the number of jet flights by 36 percent, and the number of turboprop flights by 229 percent. The bottom part of table 10.8 (lines 3c-4b) displays a sharp contrast between the 90 percent of flights on major routes and the remaining 10 percent of flights involving service between small nonhubs and other (small and large) nonhubs, where the number of routes flown decreased by 36 percent, and the number of jet flights decreased by 55 percent; the number of turboprop flights increased by 37 percent. A graphical interpretation of this shift is provided in the bottom frame of figure 10.1. Many of the abandoned flights to small cities were along linear routes dictated by regulated routes, as in the abandoned 3 1. Indeed Dempsey’s prime example of circuitry involves “the loss of pre-deregulation Boston-San Francisco nonstops” (30). This is one of Dempsey’s many factual errors: in no year since 1962 has Boston-San Francisco lacked nonstops, and in the summer of 1991 enjoyed five daily nonstop flights. His fanciful “circuitry” example involves passengers allegedly forced to fly this route via Dallas, rather than more directly through any of the many available hubs, including Chicago, Cleveland, Denver, Detroit, Minneapolis, Newark, or Salt Lake City. 32. My definition of a hub is based on the absolute volume of connecting flight and traffic activity, not the percentage of total traffic that is connecting vs. local (an alternative criterion suggested to me by Severin Borenstein). For instance, San Francisco and Memphis in 1989:Q3 were ranked 15th and 16th in the absolute volume of connecting passenger enplanements, yet San Francisco boarded only 21 percent of its total domestic traffic as connections (79 percent local traffic); Memphis boarded 63 percent as connections (37 percent local) This contrast does not make San Francisco any less of a hub than Memphis, because the volume of activity is the same, and the dominant connecting airline in San Francisco (United) gains a tremendous advantage in adding flight frequencies that allow it to dominate the local traffic as well. 33. When a turboprop flight is given a weight equal to 0.25 of a jet flight, the 1978 flights listed in lines la through 3b account for 89.3 percent of the total flights listed in lines la through 4b.

393

Productivity in the Transportation Sector

route between C and D. Because most of these routes were shorter than 200 miles in length, they were valued by relatively few passengers, most of whom used surface More than offsetting the loss of such routes was (1) the large number of new routes to hubs (e.g., from C to H1 and H2), (2) the large number of local passengers served on new routes than abandoned routes (because hubs like HI and H2 on average have much larger populations than small cities like D), (3) the greatly increased number of connection opportunities from travel beyond hubs, thus allowing many two-connection trips to be reduced to a single connection, and (4) the much greater daily frequency of service on added routes than on abandoned routes.35 Overall, it appears that the benefits to small nonhub cities of added flights to hubs outweigh the loss of direct nonstop flights, as the number of routes flown from small nonhubs increased by 16 percent, and the total number of flights increased by 67 percent (table 10.8, line 5c). The only remaining aspect of the indictment of deregulation by Dempsey and others that retains its validity is the shift from jet to turboprop aircraft. Yet even here the discomfort factor is minimal; as most of the flights involved are less than an hour, discomfort is partly offset by increased frequency. 36 Despite the widespread introduction of new nonstop routes and the virtual elimination of interline connections under deregulation, the fraction of trips involving connections actually rose slightly, from 27 percent in 1978 to 33 percent in 1989 (table 10.10 below, line Id). Thus, in view of new nonstop route opportunities, the remaining debate over hubbing remains whether passengers were forced to take the extra connections or voluntarily chose to take the extra connections that accounted for the 1978-89 increase of 6 percentage points in the fraction of trips involving connections. The forced interpretation argues that the total number of flights involving large nonhubs increased by only 33 percent between 1978 and 1989 (table 10.8, line 5b, weighting turboprops as 0.25 of a jet flight); domestic passenger enplanements increased by 67 percent. The implication is that the unavailability of seats on heavily booked nonstop flights forced demand to spill over to less desirable connections. Denying this interpretation, however, is the fact 34. Of the 123 abandoned nonstop routes between large and small nonhubs (table 10.10, line 3d), 62 percent were 200 miles or less. 35. The average daily frequency on flights from hubs to small nonhubs (table 10.10, lines Id and 2c) in 1989 was 4.8, as contrasted with 2.1 on the abandoned 1978 routes involving small nonhubs (lines 3d and 4b). 36. We can tie our study of airline routes back to the findings of CCT (1984) that there are economies of scale to increased density (traffic per number of cities served) but not from an extension of the number of cities served. For the system as a whole, increased traffic between 1978 and 1989 was not accompanied by an increase in the number of points served, and hence density increased. But the CCT results refer to individual carriers, and most carriers increased the number of points served, implying that each airport had more carriers in 1989 than 1978. The CCT results for economies of scale for individual carriers cannot be applied to the system as a whole without a carrier-by-carrier study to determine whether increased traffic offset the increase in the number of points served by each carrier.

394

Robert J. Gordon

that long-haul nonstop flights were not significantly more or less crowded than average flights before or after dereg~lation.~~ Instead, the choice interpretation suggests at least four reasons why travelers opted voluntarily for connections instead of same-plane service: The first two reasons take note of a flaw in the data on the percentage of trips involving change of plane-these neither distinguish same-plane flights making no stops, one stop, or multistops, nor do they distinguish single from double connections. Thus the first reason for voluntary choice of connections after deregulation is that a significant fraction of the same-plane 1978 traffic did not operate nonstop but involved one, two, or more stops. Much of this one or multistop traffic has been replaced by connections that are usually as fast and available at much greater frequency. Second, the proliferation of new hubs has greatly reduced not only the number of interline connections as is documented, but also the number of time-consuming double connection^.^^ Third, the greatly increased number of long-haul connection opportunities involving satellite airports (e.g., Oakland, Orange County, San Jose, White Plains, Islip) diverted traffic from the traditional nonstop flights (still routed from airports like San Francisco, Los Angeles, and New York Kennedy); passengers chose connections from nearby satellite airports voluntarily to save ground travel time, pay lower parking fees, and reduce congestion delay. Fourth, passengers may choose voluntarily to take the time penalty of a connection in order to build up frequent-flyer credits on a preferred carrier; revealed preference argues that this cost is more than offset by the benefits of frequent-flyer programs. Overall, we conclude that the forced diversion of traffic from overcrowded nonstops to connecting flights was minor compared to the diversion from one-stops to connections (involving a negligible time cost), to the benefits of reduced double connections, to the saving in ground time and congestion when travelers chose alternative smaller airports, and to the perceived benefits of frequent-flyer plans. 10.3.3 Output Quality: Other Aspects The popular literature on deregulation implies that there has been a widespread and unambiguous decline in the quality of airline service as a result of 37. Taking the nine most important transcontinental nonstop routes flown by American, TWA, and United, the weighted average load factor in October, 1977 was 58.1 percent, compared to domestic system load factors for the same three carriers of 60.5 percent. In October, 1989, the figures were 66.5 and 63.3 percent, respectively. The source is the author’s calculations from CAB and Department of Transportation market segment data. 38. Of the hundreds of examples that could be constructed from the sources used in tables 10.7 and 10.8, the first two I looked up will suffice. Travel from Portland, Maine, to Anchorage, Alaska, in July, 1978 involved a single early-morning option to take a double connection involving three airlines; in July, 1989 the same trip could be taken in mid-morning or mid-afternoon through a single connection involving a single airline, with an elapsed time shorter by 2 hours and 45 minutes. Travel from Bakersfield, California, to Savannah, Georgia, could be made twice daily in either year, by double connection involving two airlines in 1978 and by single connection involving a single airline in the other; the time saving in 1989 was only 15 minutes for an early morning trip but 2 hours for a midday trip.

395

Productivity in the Transportation Sector

airline dereg~lation.~~ This section assembles in table 10.9 a variety of indicators to provide a new evaluation. ( I ) On-time performance. Since September 1987, the U.S. Department of Transportation (DOT) has compiled a data base of on-time performance by carrier, flight, and airport, and these data are widely publicized. Shown in the second column of table 10.9, line 1, is the average percentage of flights arriving within 15 minutes for the three years ending in August 1990. It is less well known that comparable data (covering only the top 200 markets) were collected prior to 1981, and the 1977-78 average is also displayed on the same line of table 10.9. Perhaps surprisingly, the percentages are almost identical, indicating no deterioration in on-time performance. (2) Scheduledjight times. How could the airlines have maintained a constant on-time record, in view of the frequent criticism that deregulationinspired hubbing has increased congestion and led to long conga lines of aircraft waiting to take off? The answer is provided on line 2 of table 10.9, which shows that airlines have extended scheduled times in order to maintain their average on-time percentage. Our sample consists of 60 routes flown in both years, with a representative selection of routes from original hubs, new hubs, and large nonhubs, and most of the heavily congested airports are included. The sample covers roughly 5 percent of the comparable routes in each year and shows that flight times were extended by roughly 10 minutes regardless of distance, implying that ground congestion was the cause.4oHowever, the increase in flight times is uniform across airport types and shows no tendency to be greater in hubs than nonhubs. Hence the underlying culprit is more likely to be the growth in air traffic relative to air traffic control capacity rather than any effect of deregulation on route patterns. (3) Service complaints. Line 3 of table 10.9 shows a surprising decline in airline service complaints, indicating either an improvement in airline service or a reduction in the “propensity to complain.” It is unlikely that the source of this change is selection bias resulting from a change in the complaintreceiving agency from the CAB to the Department of Transportation, as the Department of Transportation telephone number has been widely publicized and in fact complaints exhibited a temporary 1987 hump as a result of airline merger^.^' ( 4 ) Safety. The fatality rate has dropped markedly, and this appears to be the result of coordinated efforts by aircraft manufacturers, airlines, and government safety regulation, rather than a by-product of deregulation. As of early 1991, more passengers had survived than died in the six fatal crashes that occurred over the three previous years. During that period 72 percent of pas39. A particularly vivid indictment is provided by Charles Kuralt (1990). 40. In August 1978, the sample includes 249 flights of the 4,727 jet flights (5.3 percent) among the airports other than small nonhubs. In August, the sample includes 296 flights of the 6,655 jet flights (4.5 percent) within the same category. 41. Complaints fell from 41,560 to 16,668 despite an increase in enplaned passengers of roughly 80 percent.

396

Robert J. Gordon

Table 10.9

Aspects of Airline Service Quality, Selected Indicators, Averages for 1977-1978and 1988-1989

1 . Percentage of flights on time (within 15 minutes) 2. Elapsed scheduled time (hours:minutes): 1. 20 short-haul routes b. 20 medium-haul routes c. 20 long-haul routes d. Average for 60 routes 3. Complaint rate per 100,000 passengers enplaned 4. Fatalities per 100,000 passengers enplaned

Average, 1977-78

Average, 1988-89

Change 1988-89 1977-78

76.8

77.9

1.1

1:07 2:05 4:08 2:27 8.03 0.17

1:15 2:19 4:18 2:37 1.84 0.06

0:08 0:14 0:10 0:10 -6.19 -0.14

Sources by line: (1) 1977-78 on-time percentage refers to top 200 markets; 1988-89 on-time percentage for all reported airports is for the 36 months from the beginning of the current data base in September 1987 through August 1990. Source for September 1987 through January 1990 is U.S. Department of Transportation, Office of Consumer Affairs, Air Travel Consumer Report, March 1990. Otherwise the source is Air Transport World, “facts and figures” page, various issues. (2) Times are for August 1978 and August 1989 and the source is the same as for table 10.8. Short-haul routes are 300-400 miles, medium-haul 700-800 miles, and long-haul routes 1500 miles and over. Of the 20 routes in each category, 8 are randomly selected among those from “original hubs” (as defined in the notes to table 10.8), 5 from “new hubs,” and 7 from “large nonhubs.” Of the most congested airports, Atlanta, O’Hare, Denver, Dallas-Ft. Worth, Los Angeles, and N.Y. Kennedy are all included. (3) Same sources as line 1, the average for the years 1977-78, and for the 24 months ending November 1990. (4) Fatalities for 1977-78, Statistical Abstract, 1982-83, table 1102, 635, and enplanements, table 1099, 633. Fatalities for 1988, Statistical Abstract, 1990, table 1066, 622 and for 1989 from New York Times, January 19, 1991, A14. 1988-89 enplanements are from Aviation Daily, various issues.

sengers in airline accidents survived, as compared to only 10 percent during the period 1980-87 (Phillips 1991). Also suggesting that deregulation had no adverse effect, Rose (1990) shows that the average accident rate was virtually the same in 1976-80 and 1981-86 and that this rate has declined by a factor of five since 1957-60. (5) Seating density. There is no more obvious source of discontent with air travel than the cramped dimensions of seats in present-day commercial aircraft. Although an increase in seating density has occurred, its timing antedates deregulation. Seats per plane for the Boeing 747 increased by 18 percent between 1972 and 1977 and by 8 percent between 1977 and 1982 (Gordon 1990a, table 4.8). The respective figures for the Boeing 727-200 were 7 percent and 9 percent. Rather than resulting from deregulation, higher seat density resulted from an overexpansion of airline capacity in the late 1960s and the timing of the airline design cycle, which led to the introduction in 197072 of overly large wide-bodied aircraft. Both seat density and load factor were temporarily depressed, and both increased as traffic recovered after 1975. (6) Frequent-Jyer beneJits. Morrison and Winston (1989, 83n.4) have estimated that frequent-flyer benefits were worth 2.3 cents per passenger-mile in 1983, fully 20 percent of the average fare in that year, and there are good

397

Productivity in the Transportation Sector

. ~ ~ represents an unmeareasons to view this figure as an u n d e r e ~ t i m a t e This sured component of airline output, in the sense that the true price of travel is overstated. Some portion of unmeasured output may be offset by free travel that is counted as part of revenue-passenger-mile output. But apparently such travel is not consistently counted in measured output, leaving a substantial residual of unmeasured Further, as long as there is an inventory of unused miles, previous travel has created a consumer asset of substantial present value. To value frequent-flyer benefits, we take the conservative Morrison-Winston estimate of a 20 percent discount and assume that one-third of award miles are claimed, one-third are held for future use, and one-third expire without use. If one-half of claimed miles are counted as revenue traffic, then the remaining unmeasured component of output is one-sixth for claimed miles and one-third for unused miles, or half the 20 percent discount figure. This implies a downward bias in output estimates of about 1 percent per year over the ten years since frequent-flyer programs began in early 1981. 10.3.4 The Value of Time By far the most important unmeasured quality attribute of airline output is the value of time saved by airline travelers, as compared to alternative means of transportation. However, the invention of aviation, and the increased speed of aircraft from the beginning of the industry through the late 1960s, should be credited to the airframe and engine manufacturers rather than to the airline industry. Unmeasured quality change in airline output refers to changes in elapsed time caused by changes in airline operations with a given fleet of aircraft. Here we focus on such changes between 1978 and 1989 and return at the end of this section to the value of time achieved by the aircraft manufacturing industry. Morrison and Winston (1989, table 2, 66) have estimated a disaggregated airline carrier choice model that yields dollar values of time saving in three categories for 1983, total travel time ($34), transit time ($74), and schedule delay time ($3). Using these estimates, we calculate in table 10.10 the time 42. The existence in the mid-1980s of a broker market for frequent-flyer awards (recently shut down by aggressive airline court actions) provides a market test for valuation. I paid in the range of $0.025 to $0.04 per mile for such awards in the period 1983-86, yet this figure understates the value to the traveler who earned the free mileage, because of innumerable bonuses (double miles, triple miles, loyalty awards, affinity credit cards, etc.). In my case, in the first ten years of frequent-flyer programs I was credited with 1.463 million frequent-flyer miles for only 0.836 million miles actually flown, for a payoff ratio of 1.75, and an estimated value of bonus miles in the range of $0.04 to $0.05 per mile actually flown. For instance, in one example by flying 100,000 miles I earned enough bonuses for a 175,000 certificate, good for two round-trip first class tickets to Australia, with a retail value of $1 1,oOO. and which I valued at $4,750 ($2,500 for the cheapest coach fare, $25 per hour per person for 35 hours in the first-class instead of economy cabin, and $500 for the included hotel and car rental certificates), or at $0.0475 per mile flown to win the award. 43. Severin Borenstein has written me that “frequent flyer plan bonus trips have not been consistently reported as revenue passenger miles by the airlines, though the Department of Transportation is now starting to enforce a consistent reporting method for these trips.”

398

Robert J. Gordon

value of shifts in routing patterns, as well as extended travel times on given flights. Because of the low estimated value of schedule delay time, we can neglect the difficult calculation of the value of increased flight frequency on given routes. All counts of flights in table 10.10 are taken from table 10.8 and are weighted, with respective weights of 1 .O for jet flights and 0.25 for turboprop flights. Line lc shows that 21 percent of 1989 flights were on new routes. Despite this, line 2 shows that total connecting traffic increased somewhat from 27 percent to 33 percent of total trips, and interline connections almost disappeared. We have argued above that this small shift to connections, despite increased nonstop routings available, mainly reflect consumer choice rather than forced diversion from overcrowded nonstop flights. To place a time value on these shifts, we use the Morrison-Winston estimates of the value of time, updated from 1983 to 1989 using aggregate compensation per hour, and make plausible estimates of the elapsed times involved in different types of flights. The resulting estimates, shown in section 4 of the table, show that the direct benefits of changes in flight routings add up to a small $1.5 billion, more than offset by the cost of lengthened flight times. The resulting time cost is about 4 percent of domestic airline passenger revenue in 1989, with the implication that measured output growth from 1978 to 1989 is overstated by roughly 0.3 percent per annum. The estimates in table 10.10 are trivial in size, however, in contrast to plausible estimates of the value of time saving achieved by the aircraft manufacturing industry. Our calculations of standardized seat miles, summarized in table 10.12 below, show that average elapsed block speed increased from 210 miles per hour in 1954 to 433 miles per hour in 1972, and then remained at this level through 1987. This implies that the average 1989 trip of 2:37 hours (table 10.9, line 2d) would have taken 5:24 hours in 1954, neglecting the greater number of enroute stops in 1954. The time saving in 1989 was worth $5 1.7 billion, or 116 percent of domestic airline passenger revenue.M The value of time saving from faster aircraft is just the tip of the iceberg, because it neglects the value of time saved when traffic shifts from surface to air transport. If we assume that intercity common carrier passenger-miles per dollar of real disposable income remained constant between 1939 and 1989, hypothetical air travel would have been 52 percent of the actual amount.45 (The remaining 48 percent represents some combination of an income elasticity for travel greater than unity and an increased demand for travel resulting from the new-product aspects of air travel). Taking an average 1989 domestic 44. If we take a more conservative approach and use the Morrison-Winston value of elapsed time for the half of air traffic that represents business travel, and use aggregate compensation per hour for the other half, the time saving falls to $35.4 billion. 45. 1939 intercity traffic from James (1982, table 1-3, xxviii); 1989/1939 real disposable income equals 5.8, from 1990 Economic Report ojrhe President, table C-27. 1989 intercity travel includes bus, rail, and air, and the share for air was 92 percent. Resulting hypothetical 1989 intercity traffic is 197.2 billion revenue passenger miles, of which 27.5 actually traveled by surface, leaving 169.7 as the amount shifting from surface to air.

399

Productivity in the Transportation Sector

Table 10.10

Changes in Value of llme in Domestic Air navel, 1978 to 1989 1978

1. Allocation of flights (weighted by aircraft size): a. Total flights b. Flights on new routes c. Flights on new routes (96) 2. Connecting flights (%): a. Interline b. On-line 3. Shifts in type of flight (%): a. Single interline to single on-line connections b. Double interline to single on-line connections c. One-stop no-plane-change to single on-line connections d. Nonstop flights to single on-line connections 4. Value of time saving (in billions of dollars): a. Interline to on-line connections b. One-stop no-plane-change to single on-line connections c. Nonstop flights to single on-line connections d. Extended flight times e. Total 1989 domestic airline passenger revenue (%)

6183

27 11 16

1989 845 1 I789 21 33 1 32 7

3.3 -0.3 - 1.5 ~

-3.1 - 1.6 -4

Sources by line: (la) Table 10.8, totals of lines l a through 4b, with jet flights weighted 1.0 and turboprops weighted 0.25 (lb). Flights on new routes are calculated by taking the number of new routes in each category of table 10.8 and estimating the frequency per route as the average of the 1978 and 1989 frequency within that category. 'hrboprop weights are applied as in line la. (lc) lblla. (2,2a, 2b) Borenstein (1991, tables 3 and 4), which refers to 1978:Q2 and 1990:Q2. Data for the first period are copied by Borenstein from Bailey-Graham-Kaplan (1985, table 4.6, 86) and for the second period are calculated by Borenstein from the Department of Transportation data base. (3a, 3b) Interline to online is divided arbitrarily by a 7-3 ratio between double interline and single-interline connections. (3c, 3d)The remaining shift to on-line connections is assumed to have been diverted equally from one-stop and nonstop flights. (4a-4c) Domestic passenger enplanements for 1988 from Statistical Abstract 1990, table 1065, 628, multiplied by 0.67 to eliminate double counting for connections. Value of time for 1983 from Momson-Winston (1989, table 2, 66), extrapolated to 1989 by business sector compensation per hour. Respective total travel times and transit travel times saved are, respectively, 2.0 and 1.5 for double interline to single on-line, 0.5 and 0.5 hours for single interline to single on-line, -0.25 and -0.25 for one-stop no change of plane to single on-line connection, and - 2.0 and - 1.O for nonstop to single on-line connection. (4d) Extra travel time 0.167 hours from table 9, line 2d. Rest of calculation uses same sources as (4a-4c).

airline trip of 791 miles and the elapsed times of 2:37 hours for air (from table 10.9) and 14 hours by surface, the implied time saving for the traffic shifting from surface to air was worth $6 1.5 billion .46 There remains the 48 percent of 1989 air travel that represents a combina46. The 14-hour surface speed is calculated as 794 miles divided by 65 miles per hour (interstate highway speed), which allows about 1.8 hours for rest and meal stops. By contrast, the fastest 1940 scheduled train between New York and Chicago took 17 hours (James 1982, xxvi).

400

Robert J. Gordon

tion of a nonunitary income elasticity and a new If, for instance, the income elasticity of travel demand with respect to real income per capita is 1.5, then this 48 percent can be divided into 16 percent for the income effect and 32 percent for the new-product effect. Usher (1964) interprets an invention as a shift from a one-dimensional to two-dimensional production possibility frontier and evaluates the social welfare created by the extra dimension as the distance between the new frontier and the community indifference curve, but his approach cannot be implemented empirically without knowledge of the slopes and intercepts of the frontier and indifference curve. A more practical approach for estimation is to interpret the demand for the new product of air travel as resulting from a decline in the total cost of travel, consisting of the money price plus the value of time. A demand curve can be drawn through two points: The first is the actual 1989 total cost of an average trip ($185) and the average quantity (416 million passengers). The second is the hypothetical 1989 total cost of the assumed surface speed ($531) and the hypothetical quantity (the actual quantity less the 32 percent new-product demand, or 283 million).48 The implied consumer surplus trapezoid is $120.9 billion. Overall, we can sum the value of time saved from shifted traffic ($61.5 billion) to the new-product value ($120.9 billion), to arrive at a total of $182.4 billion, which is 408 percent of 1989 domestic passenger revenue, or, alternatively, 3.5 percent of 1989 GNP. We cannot include the value of the increased speed of aircraft from 1954 to 1989, because this would represent double counting. Our estimate is conservative, because it applies only to the domestic, but not the international, portion of the U.S. airline industry. Balancing this is the likelihood that, in the absence of air travel, surface travel speeds would have increased by investment in an American version of the French high-speed train or Japanese bullet train. Whatever its size, this type saving should be credited to the aircraft manufacturing industry and is about 10 times as large as U.S. commercial aircraft sales in 1988, a number that would be even larger if the saving of time in international travel by U.S. and foreign airlines were included, implying a huge rate of return to research in the aircraft industry, at least through the early 1970s. 10.3.5 Input Quantity and Quality We have previously in part 10.1 discussed alternative estimates of the quantity of labor input. Our primary concern here is the measurement of capital 47. Severin Borenstein (in correspondence) cites a third source, the introduction of price discrimination under the deregulated regime, because he suspects that low discount fares have increased leisure travel by more than high undiscounted fares have reduced business travel. Thus some unknown part of our “new product” measure may be attributable to deregulation. 48. The 1989 actual cost is the average fare per passenger ($107) plus a time cost of $29.80 (the average of the Morrison-Winston estimate for elapsed travel time and compensation per hour) times 2.6 hours per trip, or a total of $184.50. The 1989 hypothetical surface cost is $184.50 plus $29.80 times the hypothetical extra time of 1 I .6 hours, or $530.20.

401

Productivity in the Transportation Sector

input, although in our MFP calculations we also make allowance for energy and materials input, and expenditures by the government on air traffic control. Our aim here is to develop alternative measures of MFP growth that correspond to different capital goods deflators, in order to determine whether improved measurement of the quality of capital goods can explain some or all of the changes in ALP growth over time in the transportation sector. Much analysis of transportation productivity treats capital as a fixed factor of production (Good, Nadiri, and Sickles 1989, 3-4). However it would be a mistake to impose too sharply the dichotomy that the manufacturing sector produces aircraft on purely technical considerations and to search for effects of deregulation only in the MFP residual that remains after the effect of capital quantity and quality is subtracted out. The quantity of service that a given aircraft can provide is determined not just by the manufacturer but also by utilization. Airlines can boost the capital services provided by a given aircraft fleet in three ways; by increasing the fraction of seats filled (load factor), by increasing the utilization of the fleet measured in hours per day or year, and by increasing seating density. In addition to affecting the ratio of capital services to aircraft characteristics, the regulatory regime feeds back to the aircraft design process itself. The mileage-based fares in the regulated era were originally based on competition with first-class rail travel, where the relation of per-mile cost to length of haul was much flatter than for airlines. As a result there was heavy cross subsidization of short-haul by long-haul travel. Gellman (1968) has argued that the highly inefficient DC-7, the last of the piston-era aircraft and the first plane designed to fly coast to coast nonstop, would not have been created without the overpricing of long-haul travel. Similarly, the wide-bodied jet aircraft (B747, DClO, and LlOll) introduced in 1970-72 might have taken a different form, or have been ordered in fewer numbers by domestic carriers, had it not been for long-haul overpricing. In turn, the effect of deregulation in sharply increasing short-haul fares relative to long-haul fares, together with the economics of hub operations, have stimulated the demand for short-haul airliners like the B737. The first concept of capital input is the real stock series developed by the BEA, using the same deflators for structures and equipment as in the NIPA accounts. The BEA capital measurement project provides a breakdown that is perfectly designed for the purpose of this study, including real and nominal investment flows and capital stocks for both structures and equipment in total transportation and in the three subsectors covered in this paper.49 For air transport two alternative capital input series are developed for comparison with the BEA. One takes the new aircraft deflator developed in my price measurement project (Gordon 1990a) and combines it with my automo49. All BEA investment and capital stock data used in this paper are taken from the latest release of the BEA “Wealth Tape.”

402

Robert J. Gordon

bile deflator as a proxy for ground equipment to form an alternative series for equipment. Because I have not developed an alternative deflator for structures, the alternative equipment series is combined with the existing BEA deflator for airline structures (which represents 5 percent or less of airline capital). By taking into account improvements in both performance and operating efficiency, my aircraft deflator declines relative to the BEA deflator by a factor of 10 and by somewhat less once ground equipment and structures are included. In order to assess the relative importance of improvements in performance as compared to improvements in efficiency, a second capital input series measures the standardized available seat mile (ASM) capacity of the industry’s aircraft fleet. Each of 35 different aircraft types is described by a standard number of seats, speed, and yearly utilization, and the total is aggregated by the actual number of each aircraft type in the fleet in each year. This measure of capacity differs from actual output in response to any divergence between actual and standard seats, speed, and utilization. In comparing new models of aircraft with the comparable older models that they replace, the standardized ASM measure always yields a smaller valuation of the quality of the new model compared to the old than is yielded by my estimate of net revenue or by a comparison of used aircraft prices, simply because it adjusts only for the increased size and speed of newer models, but not (as do the net revenue and used price ratios) for improved fuel efficiency and for the reduced number of pilots required by some types of newer aircraft. Table 10.11 shows eight examples of the 15 comparisons used to develop my aircraft price index. These eight examples cover 14 of the 35 aircraft types used to compute standardized ASMs. For each comparison, column ( 3 ) lists the ratio of the sales price of the new to the old model (in the overlap year, if any, or else in the first year of production of the new model and last year of production of the old model).50 Column (4) shows standardized ASMs for each comparison and indicates that the ASM ratio was smaller than the price ratio in six cases of eight, suggesting that airlines would not have purchased the new models if they had offered no attractive attributes other than improved size and speed. The appeal of the newer models becomes clear in column ( 3 , which shows the ratio of the net revenue that could be generated by each model at the fixed input prices of a particular year, and in column (6), which shows the ratio of the prices of the models in the used aircraft market in a particular year. The distinction between actual and standardized capacity provides an interesting decomposition of the sources of improvement in aircraft performance over time, even if it fails to take into account improvements in the efficiency of labor and fuel use. As shown in the top part of table 10.12, actual growth 50. These are true “buyers’ prices” copied from CAB records that report the price of each aircraft and engine purchased by each airline.

Productivity in the Transportation Sector

403

Table 10.11

Comparisons of Selected New and Old Model Commercial Aircraft

New Model

Sales Price Ratio

Standardized ASM Ratio

Net Revenue Ratio

Used Price Ratio

Year for columns 5 and 6

(1)

(2)

(3)

(4)

(5)

(6)

(7)

DC6-B L188 DC7 cv440 B707-3M)B DC8-61 B727-200 LlOll

L188 B727-100 DC8-50 DC9- 10 B747-100 LlOll MD80 B767-200

1.73 2.67 2.61 4.00 2.99 1.84 1.70 1.OO

1.47 1.82 2.86 2.48 3.55 1.28 1.02 0.79

3.37 1.80 20.57 10.35 4.97 1.44 3.07 0.78

2.86 4.10 19.20 9.33 6.00 3.54 3.01

1965 1965 1965 1965 1977 1977 1982 1982

Old Model

...

Columns 1-3, 5-7 from Gordon (1990a), table 4.9, 137-39, and table 4.13, 146. Column 4: see notes for table 10.12.

Sources:

in traffic largely paralleled growth in actual capacity, although there was a minor negative contribution of load factor in 1959-69, which was reversed in 1969-78. The major contribution to capacity growth in the first and last periods was the purchase of additional planes; the most important factors were larger and faster planes in 1959-69, the decade of transition from piston to jet, and larger planes in 1969-78, the decade in which the wide-bodied aircraft were introduced. The pattern for standardized capacity was similar, indicating that most changes in average size and speed were inherent in the products supplied by the manufacturing industry. Changes in the use of standardized capacity were relatively minor. Actual seats per plane fell relative to standardized seats in the first period and then rose; this reflects in part the use of relatively large low-density first-class sections on the first generation of jets, which were gradually reduced as a fraction of total seats. Once the transition to jets was complete, after 1969, the increase in seat density proceeded steadily, and there was no significant acceleration after deregulation. The only visible effects of deregulation were a minor increase in utilization (line 3d), and a slowdown in the growth of plane size (line 2b) related to the shift to smaller aircraft suitable for hub-and-spoke operations. 10.3.6 Growth in MFP The new results on changes in capital quality can now be used to compute alternative series of MFP growth for the full period 1948-87. Each of the new MFP series uses the same input data on fuel and materials inputs, and an experimental series is calculated that allows for government input in the form of spending on airports and air traffic control. Table 10.13 provides growth rates of output and input for four time intervals and begins in section 1 with the two alternative equipment deflators (BEA and

404

Robert J. Gordon Sources of Capacity Growth by Aircraft Characteristic (annual percentage growth rates)

Table 10.12

1954-59

1959-69

1969-78

1978-87

11.37 -0.95 12.32

12.38 - 2.05

14.43

6.58 2.30 4.28

6.42 0.15 6.27

5.31 3.25 1.55 2.21

2.50 6.24 5.65 0.03

0.54 3.39 0.59 -0.24

4.74 1.19 0.03 0.40

14.74 5.24 3.42 0.78

13.66 3.67 5.29 2.20

3.76 2.47 0.31 0.45

4.78 0. I6 - 0.04 -0.09

- 2.42

0.77 2.57 0.36 -2.17

0.52 0.92 0.28 -0.21

I .49

1. Actual:

a. Revenue passenger miles b. Load factor (Ic - la) c. Available seat miles ( = Id le If Ig) d. Number of planes e. Seats per plane f. Speed (MPH) g. Utilization (hours per year) 2. Standardized: a. Available seat miles b. Seats per plane c. Speed (MPH) d. Utilization (hours per year) 3. Actual - standardized: a. Available seat miles b. Seats per plane c. Speed (MPH) d. Utilization (hours per year)

+

+ +

- 1.99 - 1.87 1.43

1.02 0.07 0.49

Sources by line: (la) Revenue passenger miles are from Aerospace Facts and Figures, various issues, for 1954-83 and from Air Carrier Trajic Statistics, December of various years, for the years 1984-88. (lb) Equals l a minus Ic. (lc) Available seat miles are from Aerospace Facts and Figures, 1984185 for the years 1969-83 and from Air Carrier Trajic Statistics, December of various years for 1954-68 and 1984-88. (Id) The number of planes is a constructed series aggregating models over the time period 1954-88. The number of each model in use for each year is from the FAA Statistical Handbook of Aviation, various years, and the World Jet Airplane lnventory at Year-End 1988 (Boeing 1989), sec. 3, table 5 . (le) Data for seating density are from the measure of available seats per aircraft mile from Aerospace Facts and Figures, 1984-85, for the years 1960-83 and Air Carrier Trajic Statistics, various years, for 1954-59 and 1984-88. (If) Average speed was constructed as a weighted average of the speed of U.S. certificated air camers domestic and international operations, taken from the FAA Statistical Handbook of Aviation, various years, and the Statistical Abstract, various years. (lg) Data for total aircraft hours are revenue aircraft hours from Air Transport, various issues, for the years 1960-87 and the CAB Handbook of Airline Statistics, 1963 ed. for 1954-59. (2a-d) Standardized available seat miles were constructed by aggregating over airplane models using World Jet Airplane Inventory at YearEnd 1988 (Boeing 1989) and FAA Statistical and Handbook of Aviation. various years, for the number of planes. The number of seats for each model, annual utilization, and speed for each model come from Gordon (1990a, table 4.8), taking the figure shown for the latest year listed. For models not covered by Gordon, data for similar models were used.

my alternative) and the BEA structures deflator. These are converted in section 2 into two alternative series on total capital input, using the BEA structures deflator in each case and BEA weights for equipment and structures. Because the alternative equipment deflator (line lb) declines relative to the BEA deflator (la) throughout, but fastest during 1959-69, the corresponding alternative real capital input measure (2b) grows faster than BEA throughout, but the difference is also greatest in 1959-69. Also shown in section 2 is the capital input measure based on standardized capacity that adjusts for size and speed of aircraft but not for operating efficiency. After 1959 its growth rate

Table 10.13

Growth in Multifactor Productivity: Air lkansportation, 1948-1987 (annual percentage growth rates) 1948-59

1. Investment deflators: a. BEA equipment 3.05 b. Alternative equipment -2.89 c. BEA structures 1.59 2. Real capital input (equipment and structures): a. BEA 8.23 b. BEA with alternative equipment 10.73 deflator 9.52 c. Standardized seat miles* 3. output: 13.33 a. Unrevised BEA 13.33 b. Revised BEA 13.88 c. BLS 4. Other components of MFP growth: 5.68 a. BEA labor input 5.52 b. BLS labor input 13.73 c. Fuel input 12.88 d. Materials input e. Government input ... 5 . MFP growth: 4.24 a. With NIPA unrevised output and input 4.24 b. With NIPA revised output and labor input With BLS output and labor input: c. BEA capital input 4.85 4.48 d. BEA capital input with alternative equipment deflator 4.68 e. Standardized seat miles 3.70 f. Same as 5c with government input 5.52 g. Line 5d with output smoothing: remove effect of changes in real yield and real GNP

1959-69

1969-78

1978-87

2. I6 -8.14 2.70

6.90 1.89 8.10

4.79 2.93 5.56

10.75 21.84

3.48 9.14

0.58 4.71

13.53

3.86

4.67

10.66 10.66 13.84

4.33 5.38 5.43

2.28 6. I6 5.57

6.96 6.66 16. I6 11.32 7.61

1.49 0.46 0.53 3.15 1.21

4.26 2.35 3.95 7.26 4.33

0.65

2.22

0.65

2.73

1.73

3.97 2.49

3.69 2.81

.85 .26

3.62 3.23

3.63 2.93

.I4 .19

2.16

2.43

.27

-2.00

Sources by line: (la, lc) BEA wealth tape. (lb) The aircraft index comes from Gordon (1990a, table B.9, 620) and the ground equipment index comes from the same source (table B.8, 618), with respective weights of 0.8 and 0.2. (2a, 2b) Equipment capital (cumulated with BEA weights from lines l a and lb) is combined with structures capital using BEA weights. (2c) From table 10.12, line 2a. (3a, 3b, 3c) See sources to table 10.2. (4a, 4b) See sources to table 10.2 (4c) Total gallons of aviation gasoline and jet fuel from National Transportation Statistics. various years, and from the CAB Handbook ofAirline Statistics. The price of both types of fuel is from National Transportation Statistics, various years, and from the WPI and producer price index prior to 1970. (4d) Nominal materials input for 1969 and 1979 is from James (1982, table 1-4, 10) and for 1989 is from World Aviation Directory, winter 1990, table 101, X-17, and is interpolated for other years, and is deflated by the average of the PPI for intermediate supplies and of the revised BEA airline output deflator. (4e)Nominal expenditure on airways and airports from Transportation in America, Historical Compendium, updated with May 1989 issue, and interpolated between data available at five-year intervals before 1970. Airways deflated by the NIPA deflator for nondefense expenditure and airports by the NIPA deflator for nonresidential structures (deflators implicit before 1959, fixed weight after). (5) MFP indexes are Tornqvist indexes, with nominal shares from the sources listed for secs. 3 and 4 of this table. Methodology for output smoothing (line 5g) explained in the text. Smoothed MFP series for 1952-87 in line 5g is linked to actual MFP series for 1948-51 for the calculation of 1948-1959 growth rate. *Standardized seat miles for 1954-59 are linked to the BEA series for 1948-54.

406

Robert J. Gordon

lies between that of the alternative capital series, indicating that the effect of greater aircraft size and speed are not fully measured by the BEA deflator, but that additional improvements were made in fuel and labor efficiency that are captured by the alternative deflator and not by standardized capacity. Figure 10.2 plots the three capital input measures. Sections 3 and 4 of table 10.13 display the growth rates of alternative output measures and of the other inputs. We note a substantial reduction in the ratio of energy to output after 1969 but not before and a decline in the ratio of materials input to output before 1978 but not afterward (reflecting in part the greater importance of travel agent commissions in the 1980s). Finally, a series on real government expenditures on airports and air traffic control (line 4e) indicates a decline in the ratio to airline output throughout. Surprisingly, the ratio of government input to airline output declines least rapidly after 1978. The implied growth rates of alternative MFP indexes appear in section 5. The first (line 5a) combines the BEA unrevised output and employment series with the BEA capital stock series, while line 5b introduces the BEA revised output series and shifts to a value-added concept for calculating MFP growth since 1977.5'The remaining MFP indexes in section 5 replace the BEA output and employment series with those from the BLS. Line 5c uses the BEA capital input series and differs from the revised all-BEA series in line 5b by growing more rapidly throughout, but particularly in 1959-69. The next two series replace BEA capital with, respectively, that based on my alternative equipment deflator and on the standardized capacity measure of input. The final series (line 5f) introduces government input and appropriately reweighs all input shares. Annual values of four MFP measures are plotted in figure 10.3, corresponding to table 10.13, lines 5a through 5d. Here we see the importance of choosing reference dates at comparable stages of the business cycle. In particular, all four measures of MFP show a local peak in 1978-79 and a sharp decline through 1981, resulting from the recession and the PATCO strike. Airline MFP performance in the 1980s looks much better measured from the 1981 trough than from the 1978 or 1979 peaks. The MFP indexes for airlines are unanimous in showing a slowdown after 1978 and implicitly no efficiency gain from deregulation. Some observers, particularly Caves, Christensen, and Trethewey (CCT) (1983, 1984), date de51. In all the MFP calculations in this paper, the MFP growth rates based on a value added rather than gross concept of output (i.e., for all BEA railroad indexes, for BEA revised airlines and trucking since 1977, and for our alternative trucking index since 1977) are calculated as the value-added share in gross output (a,) times the growth rate of value-added productivity (0,). Thus if total MFP growth is given by 0 = q - (1 - a J m - avi,thenv = [q

-

(1 - a v ) m ] / a v , O= , v - i,

and the desired MFP growth rate can be calculated as 0 = a,(v - i) = aV0".Here growth rates refer to gross output (4).materials ( m ) , a weighted average of labor and capital inputs (i), and value added (v).

407

Productivity in the Transportation Sector 200

-

15010050 -

0

s II

._-I--. Standard

8

capac1ty

W

,

I

,

,

1950

I

,

I

I

I

1955

I

I

I

I

19'60

I

I

I

I

19'65

I

I

iho' 'i9'75 I

I

I

I

J

I

1980

I

I

I

,

,

1985

Fig. 10.2 Three versions of capital input: Air transportation industry, 1948-87

1950

1955

1960

1965

1970

1975

1980

Fig. 10.3 Four versions of multifactor productivity: Air transportation industry, 1948-87

1985

,

408

Robert J. Gordon

regulation prior to 1978, because fare reductions beginning in 1977 caused a jump in 1978 traffic and load factor. The debate over the date of deregulation can be easily resolved by a statistical decomposition to purge the MFP series for the effects of changing prices and aggregate demand. We first run a regression over 1950-88 of the annual change in airline output on two constants (split at 1969), the annual change in real yield, and the annual change in real GNP (both entered as the current and one-lagged change). The results are highly significant and indicate that fully 73 percent of the variance in annual output can be explained by changes in current and lagged real yield and real GNP during 1950-69 and 86 percent during 1970-88. This allows us to compute the counterfactual growth of airline output on the assumption that real yield, real GNP, or both grew at their mean 1950-69 and 1970-88 rates rather than fluctuating as actually occurred. Next, we run a regression of annual changes in MFP on changes in airline output and use these coefficients to determine the annual growth rate of MFP with the various counterfactual output series. The results are shown in line 5g of table 10.13. Comparing lines 5d and 5g, the full adjustment reduces the MFP growth slowdown between 1959-78 and 1978-87 by about one quarter, from 1.38 percentage points to 1.02 percentage points, and by a smaller relative amount if the rapid productivity period before 1959 is included. If the break point for deregulation is changed from 1978 to 1976, as CCT would recommend, the slowdown from 1959-76 to 1976-87 is raised from 0.65 points to 0.90 points. The similarity of the cyclically corrected slowdown figures, 1.02 points with a 1978 break and 0.90 with a 1976 break, shows that our cyclical and yield corrections almost totally capture the causes of rapid MFP growth in the 1976-78 interval. To conclude, we find that airline productivity growth slowed after deregulation by every measure and that this conclusion is independent of the chosen borderline date. The remaining unmeasured biases in output indexes are offsetting, with a slight upward bias of about 0.3 percent per annum owing to extended scheduled flight times (table 10.10) offset by a downward bias of perhaps 1.0 percent per annum owing to the unmeasured value of frequentflyer benefits.

10.4 Railroads The measurement of railroad ALP and MFP is more straightforward than for airlines. Railroads produce an intermediate good, and so we have less concern with the quality of output than with airlines. The most important potential measurement error for output, the changing mix of shipments of different values and labor requirements, is already taken into account in the BLS output measure that we use throughout this section for the period since 1948. There are probably unmeasured dimensions of output quality, consisting

409

Productivity in the Transportation Sector

mainly of the benefits of improved computer tracking of shipments, but these are likely to be sufficiently minor that they can be safely ignored here.52 There is a common impression that productivity in the railroad industry in the 1980s was revived by a combination of deregulation, relaxation of featherbedding work rules, mergers, and the abandonment of unprofitable Indeed, there were pathbreaking changes, particularly a reduction from 65 carriers in 1977 to 15 in 1988, and a dramatic abandonment of unprofitable track, which in turn implied a sharp decline in the capital stock (see table 10.14). However, the appearance of rapid growth in ALP, for example, 8.17 percent per annum since 1979 for the BLS data in table 10.2 may not carry over to MFP. Caves, Christensen, and Swanson (CCS) (1980), show that MFP growth, properly estimated to a modem cost-function framework, is less than half of ALP growth over the period 1951-74. Further, as we shall see, the outstanding MFP growth achieved by railroads in the 1980s is nothing new but rather represents the continuation of a longer historical process; in the late 1980s railroads carried one-third more freight traffic than in the late 1940s with only one-sixth as many workers and much less capital and fuel input. We learned in part 10.2 that MFP measures are inaccurate in the presence of nonconstant returns to scale. Indeed CCS (1980, 1981) do find significant evidence of increasing returns to scale for railroads, but the departure from constant returns is sufficiently small (about 0.09) that their estimated growth rate of MFP is an identical 1.5 percent per year with and without an allowance for increasing returns (1980, 177-78). Thus in the rest of this section we ignore the returns to scale issue. The ingredients in our calculation of MFP and value added for railroads are displayed in table 10.14. As an alternative to the BEA data on the capital stock of railroad equipment and structures, we have developed for the equipment component a Tornqvist-weighted index of the aggregate horsepower of railroad locomotives and the ton capacity of railroad freight cars. The growth rates of the BEA and alternative equipment stock indexes are compared in lines la and l b of table 10.14 and are quite consistent. Also, much more than half of railroad capital consists of structures, so MFF' estimates are robust to the choice of the two alternative measures of equipment capital .54 The implied MFP growth estimates, Tornqvist weighted with actual nominal cost shares of labor and materials, the assumed material share, and a residual share for capital, are shown in lines 4c and 4d. Over the entire 1948-87 period, the respective growth rates of the revised BEA and the two new MFP 52. The best recent general discussion of productivity and service quality improvements for railroads is Tully (1991). On the use of computers and advanced train control systems, see Machalaba (1988) and Schwartz (1989). 53. See Flint (1986) and Kupfer (1989). 54. We also experimented by varying the weights on equipment vs. structures from the BEA weights but found little sensitivity of the MF" indexes to the weighting choice.

410

Robert J. Gordon

Table 10.14

Growth in Multifactor Productivity: Railroads, 1948-1987 1948-59

I . Real capital input (equipment): a. BEA 0.50 b. Alternative -0.84 2. Real capital input (equipment & structures): a. BEA - 1.50 b. Alternative - 1.75 3. output: a. Unrevised BEA - 1.75 b. Revised BEA - 1.75 c. BLS -0.97 4. Other components: b. Labor input - 4.42 c. Fuel -2.80 e. Materials - 1.44 5. MFP growth: a. BEA unrevised output & input 1.34 b. BEA revised output & input 1.34 With BLS output & labor input: c. BEA capital input 2.04 d. Alternative capital input 2.13

1959-69 0.55 0.73

1969-78

1978-87

-0.01 1.07

- 2.41 - 2.00

-1.81

1.87

- 1.78

- 1.81

- 1.50

- 1.80

2.26 2.26 2.25

-0.88 -0.10 0.86

-4.94 1.96 0.73

1.17 0.72

1.93 0.16 0.68

-6.62 - 2.77 - 1.37

4.45 4.45

0.97 1.63

-0.68 4.90

4.58 4.55

2.38 2.28

4.56 4.50

-

- 3.54

-

Sources by line: (la) BEA wealth tape. (lb) The number of steam and diesel electric locomotives are from Railroad Facts and Yearbook of Railroad Facts, various years. Data for the horsepower and the average tractive effort of locomotives in service are from the Statistical Abstract as well as Transport Statistics in the United States and Railroad Facts. Total freight cars in service were taken from Railroad Facts and Yearbook of Railroad Facts. Data on the tons per car was from the series on average freight carload from Railroad Facts and Yearbook of Railroad Facts. (2a-2b) Both series use BEA structures capital and BEA weights to combine structures and equipment. (3a-3c) Same sources as table 10.2 and 10.4. (4a) Same source as table 10.2. (4b) Total fuel use and the price of the fuel are from Statistics of Class I Railroads and Narional Transportation Statistics, various years, as well as Railroad Facts and Yearbook of Railroad Facts. (4c) Nonfuel materials use is assumed to be a fixed 10 percent of total operating revenues and is deflated by the GNP deflator. ( 5 ) Inputs are combined with nominal expenditure weights, obtained from the above sources.

indexes are quite close-3.03, 3.35, and 3.33 percent per annum. The consistent growth rates displayed by the BEA and alternative MFP indexes are reassuring, because the first are calculated from value added without subtracting materials and fuel, whereas the second are based on gross output. However the payoff from deregulation when MFP growth in 1978-87 is contrasted with 1947-78 is, respectively, 2.44, 1.43, and 1.42, that is, less in the alternative than in the BEA indexes. For the period of overlap (195 1-74) the average growth rate of all our MFP index in line 4c is substantially higher than that constructed by CCS, 3.45 versus 1.52 percentage points. CCS provide a decomposition (1980, 177-80) showing that a similar difference between the conventional method and their results can be attributed entirely to a differing treatment of output and input weights. The essence of the difference is that CCS place greater weight on

411

Productivity in the Transportation Sector

passenger output (because they take the weight of passenger cost in total cost, not the weight of passenger revenue traffic in total traffic).55Thus the more rapid growth of MFP in this study is in part due to the cost savings of the disappearance of rail passenger traffic, which CCS largely subsume within their slow-growing output index. Overall, we have considerable confidence in our conclusion in table 10.14 that MFP growth did accelerate after 1978, but by much less than ALP growth. Alone among the three major transportation subsectors, railroads exhibited rapid MFP growth in the 1980s and helped to offset the productivity slowdown in the rest of the service sector. However, in light of the strong labor-saving effects of deregulation measured by Berndt et al., (1990), it remains surprising that the railroad industry did as well before 1978 as our alternative MFP indexes indicate. 10.5

Bucking

Trucking shares with railroads the fact that output is almost entirely an intermediate good, and so changes in the quality of output do not directly affect aggregate output and prod~ctivity.’~ However, the measurement of trucking output and employment is more prone to error than that for railroads, since (as we learned in part 10. l), alternative indexes cover differing fractions of the total trucking industry experiencing quite different productivity performance. For instance, there was so much entry and exit in the trucking industry in the 1980s that a deflator based on the shrinking part of the industry could overstate price increases for the more efficient (and nonunion) new entrants. Winston et a1 (1990, 11) report a “huge influx of entry” following the 1980 deregulation of trucking, consisting almost entirely of class I11 carriers providing truckload (TL) service. The number of class I11 carriers increased from 14,941 to 43,364; the number of class I and I1 carriers decreased from 3104 to 2477 (Salgupis 1991). The share of class I11 carriers increased from 82.8 percent to 94.7 percent over this period. The BLS data source reports only 786 class I1 carriers in 1987, indicating incomplete coverage. A major shift in the trucking industry occurred in response to deregulation from less-thantruckload (LTL) general freight carriers, the core of the BLS sample, to “advanced TL” firms using nonunion driver teams and relays for service on high5 5 . The other major difference identified by CCS, the understated capital input weights they attribute to Kendrick, does not apply to this study, where the capital share is determined as a residual and includes all of the items, e.g., rent and property taxes, that CCS advocate for inclusion. 56. This section contains no comparisons with other academic studies, because there appears to be no study analogous to CCS (1980) that presents a time-series MFP index for trucking on the basis of the cost function or production function method. There is a proliferation of studies, but they all are limited to the estimation of micro structural parameters in panels of firms without examination of time-series properties. See Chiang and Friedlaender (1984, 1985), Friedlaender and Spady (1981), Friedlaender and Chiang (1981), Friedlaender and Bruce (1985), Daughety, Nelson, and Vigdor (1985), and Ying (1990).

412

Robert J. Gordon

density traffic corridors, “thereby ensuring high vehicle use and low costs” (Winston et al. 1990, 13). New entry came also from owner operators, and this could cause a shift in output relative to employment that could be interpreted spuriously as an increase in productivity. The distinction between TL and LTL carriers is highlighted by the estimate of Winston et al. that in the absence of deregulation over the interval 1977-85 TL rates would have increased by 55 percent; LTL rates would have increased by a much larger 116 percent. The actual increases were 5 lpercent and 79 percent, respectively, indicating that deregulation had a much larger effect on LTL carriers. In this paper we develop MFP indexes based on two alternative measures of capital and two of output. The first capital stock measure is that produced by the BEA by the same procedures as for airlines and railroads, and already used in tables 10.13 and 10.14 to compute the BEA index of MFP for those two industries. The alternative capital input measure developed here is based on the alternative deflator for producers’ durable equipment investment in trucks from Gordon (1990a). This deflator combines separate deflators for automobiles (which behave quite similarly to the automobile consumer price index [CPI] after the late 1950s) and for diesel engines. However, this deflator, like the CPI and existing NIPA deflator for automobiles, assumes that the addition of antipollution equipment represents an increase in quality rather than an increase in price. Although such equipment may or may not benefit society in proportion to its cost, it does not represent an increase in quality as viewed by the firm using an automobile (or truck) as a capital input. As Triplett (1983) has emphasized, there are two correct measures of capital input: one for output deflation and one for input deflation. Here we need an input deflator that treats the cost of legislated equipment as an increase in price, not an increase in quantity. Fortunately, it is possible to adjust for this equipment, and the resulting hybrid index is likely to be a more satisfactory capital input deflator than other existing indexes. As shown in the comparison of lines l a and l b of table 10.15, and on an annual basis in figure 10.4, the new deflator implies a much more rapid increase in the capital stock in the first half of the postwar, because of a substantial reduction in the relative prices of our automobile and diesel engine deflators relative to the BEA trucking deflator. We also develop a new output measure in table 10.15, line 2c, to compare with the revised BEA output measure shown in line 2b. This takes nominal BEA output and then deflates it with the “yield” (revenue per ton-mile) index shown above in table 10.5, line 2c. Because the yield measure is only available back in 1960 and appears to agree with the BEA deflator until about 1972, the alternative output measure differs from the BEA series only in the 1970s and 1980s. An interesting aspect of these series is their implied capitaloutput ratios. The BEA capital and output series (lines l a and 2b) imply a radical shift between a falling capital-output ratio in 1948-69 to a relatively

413

Productivity in the Transportation Sector

Table 10.15

Growth in Multifactor Productivity: Ikucking, 1948-1987

1. Real capital input (equipment & structures): a. BEA b. Alternative 2. output a. BEA unrevised output b. BEA revised output c. Alternative output 3. Other components: a. BEA labor input b. Fuel c. Materials d. Highway capital 4. MFP growth: a. BEA unrevised output & input b. BEA revised output & input c. Alternative output & labor input, BEA capital input d. Alternative output & input e. Alternative output & input, with government capital

1948-59

1959-69

1969-78

1978-87

3.79 5.96

3.29 5.65

4.93 5.69

2.33 2.74

7.06 7.06 7.06

5.56 5.56 5.57

4.92 4.80 5.80

0.55 1.87 3.02

3.49 4.68 7.79

...

2.15 3.26 6.10 4.30

2.12 4.84 3.47 2.23

2.91 2.97

2.51 2.51

1.49 1.38

-0.75 0.00

2.91 2.05

2.52 1.51

2.38 2.06

0.97 0.82

...

1.47

2.36

0.86

1.64 - 2.72

-0.89 1.40

Sources by line: (la) From BEA wealth tape. (lb) Computed as in tables 10.13 and 10.14 by substituting a new equipment deflator (Gordon 1990a, table C3, 698) for the BEA deflator, while using BEA nominal equipment investment, BEA structures capital, and BEA weights for equipment and structures. (2a, 2b) Same sources as tables 10.2 and 10.4. (212)Deflate nominal, revised BEA output with alternative deflator, source given in notes to table 10.5, line 2c. (3a) Same source as table 10.2. (3b) Total fuel cost from cost of fuel per mile, total vehicle miles, and price of fuel from American Trucking Trends. (3c) Materials assumed to be 10 percent of revenue, deflated by the average of the producer price index for intermediate supplies and the revised BEA trucking output deflator. (3d) Government highway capital is gross constant-dollar capital stock of federal, state, and local highways, from Fixed Reproducible Tangible Wealth in the United States, 1925-85. 1985-87 was extrapolated from 1984-85 growth rate. (4) Inputs are combined with nominal expenditure weights, obtained from the above sources. Share of government highway input is taken to be half of the ratio of government expenditure on highways (same source as table 10.15, line Id) to intercity trucking revenue (same source as table 10.5, line lc).

stable ratio after 1969. The two new series (lines l b and 2c) imply that the capital-output ratio was roughly stable throughout. When we combine the BEA and new capital and output series with a fixed set of labor input, fuel input, and materials input series, we arrive at the MFP indexes shown in section 4 of table 10.15; annual data for the indexes on lines 4a-4d are plotted in figure 10.5.57The first in line 4a uses the unrevised BEA 57. Recall that since 1977 the revised BEA and alternative output indexes refer to value added, and thus the corresponding MFP indexes in table 10.15, lines 4b through 4e, are calculated as value-added MFP times the share of value added in gross output. See n. 51, above.

Robert J. Gordon

414

150 125

ioo

75

50

Trucking Capital wlth Alternatlve Deflator

25

I

,

~

"

'

1950

,

I

,

1955

,

,

'

I

,

1960

,

,

,

I

,

1965

,

,

,

I

,

,

1970

,

,

I

,

'

1975

,

,

I

,

,

1980

,

'

I

'

1985

Fig. 10.4 Three versions of capital input: Bucking industry, 1947-87 110

100 90 80 70 60

50

40.

1

~

,

1950

,

,

,

1

,

1955

,

,

,

1

,

1960

,

,

,

1

,

1965

,

,

,

,

,

1970

,

,

,

1

,

1975

~

,

,

1

,

1980

~

,

Fig. 10.5 Four versions of multifactor productivity: Bucking industry, 1948-87

,

1

,

1985

~

i

415

Productivity in the Transportation Sector

series for output and input and exhibits a sharp productivity growth slowdown, especially after 1978. The BEA output revisions make little difference in line 4b; MFP growth slows to zero after 1978. In line 4c we replace the BEA output series with the alternative output series based on the “yield” deflator, while retaining the BEA capital index. This makes a substantial difference but still leaves a post-1978 MFP growth slowdown. The next step in line 4d is to replace the BEA capital input index with the index based on the alternative equipment deflator. By slowing MFP growth before 1978, this reduces but does not eliminate the post-1978 slowdown, and reduces the slowdown to only 0.35 percentage points when 1969-87 is compared to the pre-1969 period. In contrast the two BEA indexes indicate post-1969 slowdowns of 2.38 and 2.06 percentage points, respectively. A final MFP index is developed in line 4e. This adds to the contribution of input growth the increase in the real gross stock of government “highway capital.’’ To obtain a share, we note that total government expenditures on highways in 1978 were 48 percent of intercity trucking revenues. Arbitrarily allocating half the highway expenditures to cars and half to trucks, we obtain a weight of 24 percent to be applied to the growth rates of highway capital (table 10.15, line 3d). For the resulting MFP index to be significantly different from the other indexes, government capital would have been required to grow at radically different rates than the average for other inputs. However, this did not occur, and the fully inclusive MFP index on line 4e of table 10.15 tells the same story as that on line 4d. Overall we should have observed some decline in the productivity of the trucking industry after the first oil shock, if only because of a decline in average highway Indeed, this is what is implied by the intermediate series using BEA capital and alternative output. However, the alternative capital series implies that MFP growth in trucking did not actually slow down appreciably in the 1970s and 1980s when the two decades are lumped together. Rather, faster growth in the conventional BEA measure in the early postwar years is attributed largely to the more rapid growth in the quality-adjusted capital stock of trucking equipment in the early postwar period, due in large part to improvements in the efficiency and durability of diesel engines.59

10.6 Conclusion The goals of this paper have been to develop new measures of MFP growth in the three main components of transportation-air, rail, and trucking-that allow for changes in the quality of both output and inputs. The new MFP 58. Average motor vehicle speed on highways dropped from 63.8 MPH in 1970 to 57.6 MPH in 1974 and then increased gradually to 59.7 MPH in 1986 and 1987 (StaristicalAbstract 1989, table 1025, and 1990, table 1047). 59. Gordon (1990a, 505-12) contains a detailed case study of diesel engine prices and quality improvements.

416

Robert J. Gordon

measures are summarized in table 10.16 and compared with the official measure implied by current NIPA (or BEA) data, both before and after the recent NIPA output revisions. Lines l a and l b of table 10.16 exhibit MFP growth for transportation, using NIPA data for output (without and with revision) and employment, together with the BEA capital stock estimates and our series on fuel and materials inputs prior to 1977 for airlines and trucking (railroads throughout are based on value added). Here as elsewhere in table 10.16 “total transportation” refers only to the three major subsectors. All MFP series for total transportation are Tornqvist indexes that use annual revised NIPA data on nominal output in the three subsectors as weights. The post-1973 slowdown on line l a is 2.61 annual percentage points but declines to 0.90 points on line l b with the recent output revisions. Line l c displays the first alternative measure, which switches to BLS measures of airline and railroad output and employment and to our new yielddeflated trucking output measure, as indicated in the notes to table 10.16. This switch boosts MFP growth both before and after 1973, but leaves the post1973 slowdown almost identical to the revised NIPA index. The second alternative on line Id substitutes our new capital input measures and reduces MFP growth more before 1973 than after, thus eliminating almost one-third of the post-1973 slowdown on line lc. However, the second alternative makes a substantial difference in the interpretation of the post- 1979 deregulation period, reducing the post-1979 slowdown almost to zero. As shown in figure 10.6, the annual plot of the four MFP indexes reveals substantial cyclical fluctuations, particularly in the late 1970s and early 1980s. As explained in the notes to table 10.16, the cyclical component of MFP fluctuations due to aggregate real GNP changes is purged, and the cyclically corrected growth rates are displayed in the bottom half of table 10.16. The cyclical correction substantially boosts MFP growth in 1973-79 and cuts it slightly in 1979-87, thus reducing the size of the post-1973 slowdown and slightly increasing the magnitude of the post- 1979 slowdown. The productivity growth story told by the revised NIPA index (line lb) and our final index (line Id) are surprisingly similar, given all the differences between them. Our adjustments boost MFP growth by switching to alternative output and employment indexes but then largely offset this by switching to faster-growing capital input indexes. However, these similarities disguise marked differences at the industry level, particularly in the first half of the postwar period. Our alternative output and employment data produce MFP indexes that rise more rapidly for airlines and railroads over 1948-69, but this is largely offset by our alternative capital input data that cut MFP growth for trucking below the rate estimated when conventional capital input indexes are used. Did deregulation boost productivity in transportation? Surprisingly, the answer is no. The great success story is the railroad industry, but all our indexes for airlines and trucking display a lamentable MFP growth record in the 1980s

Table 10.16

Four Measures of Multifactor Productivity Growth for 'hansportation, Annual Percentage Growth Rates, 1948-1987 and Selected Intervals, with and without Cyclical Correction

I . Raw data: a. BEA unrevised output & input b. BEA revised output & input c. Alternative output, BEA capital d. Alternative output & capital 2. Cyclically corrected data: a. BEA unrevised output & input b. BEA revised output & input c. Alternative output, BEA capital d. Alternative output & capital

1948-59 (1)

1959-66 (2)

1966-73 (3)

1973-79 (4)

1.90

4.20

1.25

0.39

1.90

4.20

1.25

0.99

1.82

2.36

2.37

4.55

2.32

1.64

2.33

2.08

3.96

.62

I .36

2.15

2.34

3.24

.63

0.91

2.28

3.48

.55

1.46

1.73

2.41

2.31

4.25

2.47

I .95

2.23

2.02

3.63

1.78

1.67

2.05

1979-87 (5)

-0.73

-0.80

1948-73 (6)

1973-87 (7)

Slowdown, 1973-87 1948-73 (8)

Slowdown, 1979-87 1948-79 (9)

-2.61

- 2.70

1.46

-0.90

-0.29

2.97

2.04

-0.93

-0.38

2.47

1.81

- 0.66

-0.11

2.46

-2.83

1.61

- 0.80

-0.42

2.90

2.11

-0.79

-0.52

2.40

1.89

-0.51

-0.25

2.36

2.39

-0.25

-0.07

~

Sources by Line: Tornqvist weights (nominal output shares from revised NIPA table 6. I ) are used to aggregate MFP growth for airlines, railroads, and trucking. (la) BEA unrevised concept uses NIPA unrevised output, NIPA employment, and BEA real gross capital stock of equipment and structures, together with fuel and materials inputs from tables 10.12-10.14. No allowance is made for the value of time or for government capital. (lb) BEA revised concept replaces NIPA unrevised output with NIPA revised output for 1977-87. Because revised NIPA output is a value-added concept, materials and full inputs are not subtracted out. See n . 51 in text. (lc) This measure replaced NIPA output and employment with BLS output and employment for airlines and railroads, and uses the revised NIPA output series for trucking with the new deflator, from table 10.15, line 2c. (Id) This measure starts from line lc and replaces BEA capital with the respective capital indexes (see table 10.13, line 5d; table 10.14, line 5d; and table 10.15, line 4d). (2a-2d) For the corresponding line of sec. I , the growth rate of MFP is run on five constants corresponding to the first five columns of this table, and on the current and one lagged change in the ratio of actual to natural GNP, from Gordon (l990b. appendix A, A2-A3). The cyclically corrected growth rate of MFP is the actual growth rate minus the statistical contribution of the GNP change.

418

Robert J. Gordon

130 120 110 100 0

s c)

.l

2

BBA

90 80 70 60

50

1950

1955

1960

1965

1970

1975

1980

1985

Fig. 10.6 Four versions of multifactor productivity: Ransportation, 1948-87

that more than cancels out the railroad success. These conclusions regarding the divergent performances of the three subsectors are extremely robust to alternative dating of deregulation. In conclusion, this paper has explained much but not all of the large post1973 productivity growth slowdown in the transportation industry displayed in table 10.1 above and in line l a of table 10.16, based on the NIPA and BEA data published prior to January, 1991. Much of the reinterpretation involves simple issues of data construction, reviewed in part 10.1, and pre-1991 investigators could have obtained roughly the same conclusion as in this paper by ignoring the old NIPA data and instead using BLS data on output and employment. The NIPA output revisions bring the NIPA and BLS output data much closer together for the period since 1977, and we view the prompt response of the NIPA output revisions to the earlier criticisms contained in Gordon and Baily (1988) as part of the overall contribution of our research. Our new MFP indexes rely not only on the choice of the “best” output and employment indexes, but also on the development of new capital input measures that adjust more fully for quality changes in transportation equipment than the official measures. The resulting MFP indexes grow substantially slower during the first part of the postwar period than when conventional capital input measures are used; the overall effect on transportation as a whole is limited by the relatively small weight of air transportation in the transportation aggregate during the years when “most of the action” occurred (1958-70). Several novel elements of our study are not incorporated into the final MFP

419

Productivity in the Transportation Sector

indexes in table 10.16. We have found that airline deregulation yielded a small time saving from the elimination of interline connections that was more than offset by a small time cost of extended scheduled times (which we interpret as due to inadequate government investment in airports and air traffic control). A much greater contribution was made by the value of time saved through the invention of air transport industry, which should be credited to the manufacturers of airframes and engines. This value (roughly $182 billion in 1989) amounts to a massive ten times U.S. sales of commercial aircraft, four times the domestic passenger revenue of U. S. airlines, and 3.5 percent of GNP. Our study of MFP growth in transportation has yielded additional findings: Airline deregulation greatly increased the availability of nonstop flights and forced only a negligible number of passengers off of nonstop flights onto connecting flights, contrary to the conventional wisdom. The increased use of travel agents had little effect on MFP growth, as decreases in other purchases of materials offset the increased use by airlines of purchased travel agent services. Finally, the perception that the government has shortchanged infrastructure investment in airports, airways, and highways, although plausible anecdotally in view of extended scheduled flight times, is not supported quantitatively by the government capital and investment data that we have compiled; MFP estimates are little changed when plausible adjustments are made for government inputs.

References Bailey, Elizabeth E., David R. Graham, and Daniel P. Kaplan. 1985. Deregulating the Airlines. Cambridge, Mass.: MIT Press. Baily, Martin Neil, and Robert J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 19(2): 347-420. Baumgarner, James D. 1979. Piedmont Is Finding New Markets and Shifting Fortunes Following Deregulation. Air Transport World April, 47-50. Berndt, Ernst R., and Melvyn A. Fuss. 1986. Productivity Measurement with Adjustments for Variations in Capacity Utilization and Other Forms of Temporary Equilibrium. Journal of Econometrics 33( 1): 7-29. Bemdt, Emst R., Ann F. Friedlaender, S . Judy Wang Chiang, and Christopher A. Vellturo. 1990. The Productivity and Cost Effects of Deregulation and Mergers in Class I U.S. Railroads, 1974-86. Manuscript, MIT, November 12. Borenstein, Severin. 1989. Hubs and High Fares: Airport Dominance and Market Power in the U S . Airline Industry. Rand Journal of Economics 20 (autumn): 34465. . 1991. The Evolution of U.S. Airline Competition. Univ. of California, Davis, manuscript, April 4. Card, David, 1989. Deregulation and Labor Earnings in the Airline Industry. Princeton Univ., manuscript, October. Caves, Douglas W., Laurits R. Christensen, W. Erwin Diewert. 1982. Multilateral

420

Robert J. Gordon

Comparisons of Output, Input, and Productivity Using Superlative Index Numbers. Economic Journal 92 (March): 73-86. Caves, Douglas W., Laurits R. Christensen, and Joseph A. Swanson. 1980. Productivity in U.S. Railroads, 1951-1974. Bell Journal ofEconomics 11 (spring): 166-81. . 1981. Productivity Growth, Scale Economies, and Capacity Utilization in U.S. Railroads, 1955-74. American Economic Review 71 (December): 994-1002. Caves, Dougles W., Laurits R. Christensen, Michael W. Tretheway. 1981. U.S. Trunk Air Carriers, 1972-1977: A Multilateral Comparison of Total Factor Productivity. In Cowing and Stevenson 1981,47-76. . 1983. Productivity Performance of U.S. Trunk and Local Service Airlines in the Era of Deregulation. Economic Inquiry 21 (July): 312-24. . 1984. Economies of Density versus Economies of Scale: Why Trunk and Local Service Airline Costs Differ. Rand Journal of Economics 15 (winter): 471-89. Caves, Douglas W., Laurits R. Christensen, Michael W. Tretheway, Robert J. Windle. 1985. The Effect of New Entry on Productivity Growth in the U.S. Airline Industry. The Logistics and Transportation Review 21(4): 299-336. Caves, Richard E. 1962. Air Transport and Its Regulators. Cambridge, Mass.: Harvard Univ. Press. Chiang, S. Judy Wang, and Ann F. Friedlaender. 1984. Output Aggregation, Network Effects, and the Measurement of Trucking Technology. Review of Economics and Statistics 66 (May): 267-76. . 1985. Truck Technology and Efficient Market Structure. Review of Economics and Statistics 67 (May): 250-58. Cowing, Thomas G . , and Rodney E. Stevenson, eds. 1981. Productivity Measurement in Regulated Industries. New York: Academic Press. Daughety, Andrew F., Forrest D. Nelson, and William R. Vigdor. 1985. An Econometric Analysis of the Cost and Production Structure of the Trucking Industry. In Analytical Studies in Transportation Economics, ed. A. F. Daughety, 65-95. New York: Cambridge Univ. Press. de Leeuw, Frank, Michael Mohr, and Robert P. Parker. 1991. Gross Product by Industry, 1977-88: A Progress Report on Improving the Estimates. Survey of Current Business 71 (January): 23-37. Dempsey, Paul S . 1990. Flying Blind. Washington, D.C.: Economic Policy Institute. Denny, Michael, Melvyn Fuss, and Leonard Waverman. 1981. The Measurement and Interpretation of Total Factor Productivity in Regulated Industries, with an Application to Canadian Telecommunications. In Cowing and Stevenson 1981, 179-218. Diewert, W. Erwin. 1976. Exact and Superlative Index Numbers. Journal of Econometrics 4 (May): 115-45. Douglas, George W., and James C. Miller. 1974. Economic Regulation of Domestic Air Transport: Theory and Policy. Washington, D.C.: Brookings Institution. Flint, Jerry. 1986. Here Come the Tmckbusters. Forbes, June 30, 87-90. Friedlaender, Ann F., and Sharon Schur Bruce. 1985. Augmentation Effects and Technical Change in the Regulated Trucking Industry, 1974-79. In Analytical Studies in Transportation Economics, ed. A. F. Daughety, 29-63. New York: Cambridge Univ. Press. Friedlaender, Ann F., and S . Judy Wang Chiang. 1981. Regulation and the Structure of Technology in the Trucking Industry. In Cowing and Stevenson 1981,77-106. Friedlaender, Ann F., and Richard H. Spady. Freight Transport Regulation: Equity, Eficiency, and Competition in the Rail and Trucking Industries. Cambridge, Mass. : MIT Press. Gellman, Aaron. 1968. The Effect of Regulation on Aircraft Choice. Ph.D. diss., Department of Economics, MIT. Good, David H., M. Ishaq Nadiri, and Robin C. Sickles. 1989. The Structure of Pro-

421

Productivity in the Transportation Sector

duction, Technical Change and Efficiency in Multiproduct Industry: An Application to U.S. Airlines. Research Report no. 89-14, June. New York Univ. C. V. Starr Center for Applied Economics. Gordon, Robert J. 1965. Airline Costs and Managerial Efficiency. In Transportation Economics, 61-94. Universities-National Bureau Conference Series, vol. 17. New York: Columbia Univ. Press. . 1990a. The Measurement of Durable Goods Prices. Chicago: Univ. of Chicago Press. . 1990b. Macroeconomics. 5th ed. Glenview, Ill.: Scott, Foresman. Hulten, Charles R. 1986. Productivity Change, Capacity Utilization, and the Sources of Efficiency Growth. Journal of Econometrics 33 (I): 3 1-50. James, George W., ed. 1982. Airline Economics. Lexington, Mass.: Lexington. Jorgenson, Dale W. 1990. Productivity and Economic Growth. In Fqty Years of Economic Measurements, ed. E. R. Berndt and J. E. Triplett, 19-118. NBER Studies in Income and Wealth, vol. 54. Chicago: Univ. of Chicago Press. Jorgenson, Dale W., Frank M. Gollop, and Barbara M. Fraumeni. 1987. Productivity and US.Economic Growth. Cambridge, Mass.: Harvard Univ. Press. Kendrick, John W., assisted by Maude R. Pesch. 1961. Productivity Trends in the United States. NBER General Series, vol. 71. Princeton, N.J.: Princeton Univ. Press. Kupfer, Andrew. 1989. An Outsider Fires up a Railroad. Fortune, December 18, 13346. Kuralt, Charles. Up in the Air against His Will. New York Times, October 14, 1990, travel sec., 37. Machalaba, Daniel. 1988. New Train Control Systems May Reduce Emissions, Cut Fuel Bills, Boost Capacity. Wall Street Journal, October 26, B4. Mansfield, Edwin. 1965. Innovation and technical change in the railroad industry. In Transportation Economics, 169-97. Universities-National Bureau Conference Series, vol. 17. New York: Columbia Univ. Press. Mark, Jerome A. 1988. Measuring Productivity in Services Industries. In Technology in Services: Policies for Growth, Trade, and Employment, ed. Bruce R. Guile and James Brian Quinn, 139-59. Washington, D.C.: National Academy Press. Mark, Jerome A. and William H. Waldorf. 1983. Multifactor Productivity: A New BLS Measure. Monthly Labor Review December, 3-15. McGowan, Francis, and Paul Seabright. 1989. Deregulating European Airlines. Economic Policy 9 (October: 284-344. Momson, Steven A,, and Clifford Winston. 1986. The Economic Effects of Airline Deregulation. Washington, D.C.: Brookings Institution. . 1989. Enhancing the Performance of the Deregulated Air Transportation System. Brookings Papers on Economic Activity: Microeconomics, 61-1 12. Phillips, Don. 1991. Are Major Air Crashes Getting More Survivable? (Washington Post story as transmitted by Compuserve Executive News Service, February 13, 1991). Rose, Nancy L. 1989. Financial Influences on Airline Safety. In Transportation Safety in an Age of Deregulation, ed. Leon N. Moses and Ian Savage, 93-1 14. New York: Oxford Univ. Press. Salgupis, Agis. 1991. A Whole Lot of Shaking Going On. New York Times, May 5, F5. Schwartz, James. 1989. What’s New in Freight Trains: A Drive to Compete with Trucks. New YorkTimes, August 13, F13. Sickles, Robin C. 1985. A Nonlinear Multivariate Error Components Analysis of Technology and Specific Factor Productivity Growth with an Application to the U.S. Airlines. Journal of Econometrics 27( I): 61-78.

422

Robert J. Gordon

Triplett, Jack E. 1983. Concepts of Quality in Input and Output Price Measures: A Resolution of the User-Value Resource-Cost Debate. In The US.National Income and Product Accounts: Selected Topics, ed. Murray F. Foss, 269-31 1. Studies in Income and Wealth, vol. 47. Chicago: Univ. of Chicago Press. . 1989. Price and Technological Change in a Capital Good: A Survey of Research on Computers. In Technology and Capital Formation, ed. Dale W. Jorgenson and Ralph Landau, 127-213. Cambridge, Mass.: MIT Press. Tully, Shawn. 1991. Comeback Ahead for Railroads. Fortune, June 17, 107-1 13. U.S. Bureau of Labor Statistics. 1990. Productivity Measures for Selected Industries and Government Services. BLS bulletin no. 2349. Washington, D.C.: Government Printing Office, February. Usher, Dan. 1964. The Welfare Economics of Invention. Economica August, 279-87. White, Lawrence J. 1979. Economies of Scale and the Question of “Natural Monopoly” in the Airline Industry. Journal of Air Law and Commerce 44545-73. Winston, Clifford. 1985. Conceptual Developments in the Economics of Transportation: An Interpretive Survey. Journal of Economic Literature 23 (March): 57-94. Winston, Clifford, Thomas M. Corsi, Curtis M . Grimm, and Carol A. Evans. 1990. The Economic Efects of Surface Freight Deregulation. Washington, D.C. : Brookings Institution. Ying, John S. 1990. The Inefficiency of Regulating a Competitive Industry: Productivity Gains in Trucking Following Reform. Review of Economics and Statistics 2, no. 2 (May): 191-201.

Comment

Robin C. Sickles

The paper by Robert J. Gordon follows on the heels of his excellent monograph, The Measurement of Durable Goods Prices (1990), and pursues complementary issues in the measurement of factor productivity growth for the transportation sector. The current study is at the industry level and follows the growth in factor productivity in the airline, railroad, and trucking industries for the last 40 years (1948-88). The conceptual and measurement problems that Gordon faced, and in my opinion largely overcame, were substantial. The work addresses a number of important issues on its way to making its key point. It is that the mismeasurement of output and input indexes and the use of partial instead of multifactor productivity (MFP) indexes has lead to erroneous conclusions by some researchers that there was a post- 1973 productivity slowdown in transportation mirroring the experience in the total U.S. economy. Gordon points out that this is a somewhat counterintuitive empirical finding because the transportation sector was deregulated in the mid- 1970s, and productivity should have benefited from less constrained decision makRobin C. Sickles is professor of economics and statistics at Rice University and has been affiliated with the National Bureau of Economic Research since 1982. His research on airline productivity, referred to throughout this comment, was funded by the National Science Foundation. Special thanks are given to David Good for his insightful comments and suggestions, which added significantly to the review. The usual caveat applies.

423

Productivity in the Transportation Sector

ing. The counterintuition is rendered illusory by Gordon’s new data series by which he concludes that there was not a post-1973 slowdown in MFP growth for the transportation sector. Although I am quite sympathetic to Gordon’s arguments I do have a number of points to make with regard to the research issues he addresses and with regard to complementary research that addresses these issues somewhat differently. As I have said, the major point of Gordon’s paper is that the measured productivity slowdown in the transportation industry is a measurement problem. Although the topic of the paper is on the transportation sector, Gordon gives disproportionate coverage to the airline industry; so will my comments on his paper. There are a number of convincing reasons why measurement problems plague the evaluation of MFP performance and the attendant national income and product account (NIPA) estimates for average labor productivity (ALP) in the transportation sector in general and the airline industry in particular. One reason has been discussed by a number of researchers and was pointed out by Baily and Gordon (1988). In the airline industry the Bureau of Economic Analysis (BEA) output deflators “fail to adjust properly for the introduction of discount fares.” However, Gordon points to a more fundamental reason-that ALP is not an appropriate index to use because a lot has been going on with the other factors such as energy and capital. MFP growth rates are different from ALP as one would expect because there has been a substantial change in relative input prices and relative factor intensities. But doing the right thing inevitably has a cost, and here it is in requiring the capital service flows to be estimated correctly. The measurement of capital service flows and its price and its decomposition into such sources as scale, technical change, and so on can be problematic especially when, for example, the technology (possibly endogenous) is embodied in an airplane’s design characteristics (Good, Nadiri, and Sickles 1991). It should be stressed that the transportation data series that Gordon constructs are annual aggregates. There is an acknowledged trade-off between the length of the series and the potential for mismeasurement of MFP and ALP resulting from both temporal and crosssectional aggregation. Moreover, the motivation for examining aggregated data as opposed to firm level data (which is available for all three major transportation industries) may be misplaced when one considers the substantial changes in industry structure and the menu of new technologies introduced into these industries. The 1948-88 series that Gordon constructs may in fact be nonstationary and chained indexes such as the discrete approximation to the Divisia index used herein may not properly represent shifts in the moments of the underlying data. This concern was in part what lead Sickles (1985), Sickles, Good, and Johnson (1986), and Good, Nadiri, and Sickles (1991) to break up their firm level quarterly series for the airline industry (1968-87) into epochs during which industry structure was more or less stable or to adopt modeling approaches that faced up to temporal and cross-sectional heterogeneity and changes in industry structure and incentives as a result of

424

Robert J. Gordon

deregulation. The Gordon and Good-Sickles data do indicate rather remarkable agreement with respect to industry MFP growth rates for overlapping periods. However, the claim in the paper that these aggregate annual data can be used to estimate dynamic effects, effects of the idiosyncratic confluence of high energy prices and low aggregated demand prevalent in the 1970s and 1980s, as well as be able to deliver on the aim to disentangle the contribution of macrodemand, energy prices, deregulation, and microeconomic factors in the determination of the postwar productivity performance of the transportation sector is to my mind overstated. Continuing with his critique of the input and output series constructed by the BEA and Bureau of Labor Statistics (BLS), Gordon has two specific disclaimers to the veracity of published industry data. An example in the airline industry is the capital and labor supplied to it in the form of airports and their administrative infrastructure (e.g., air traffic controllers and other Federal Aviation Administration personnel) as well as outsourcing personnel in the form of reservations clerks and sales agents at city ticket offices which are not accounted for and thus may bias labor and to some extent the capital input measures. Outsourcing is a problem that is not unique to the airline industry, for example, the U.S. Postal Service USPS outsources to firms in the form of presort discounts. As Walter Oi (chap. 4, in this vol.) has pointed out, failure to properly frame the production process as joint in household time and in formal business inputs can cause serious mismeasurement of the input mix and thus the MFP measures. Thus indirect routing, which presumably requires increased household time, can confound standard growth accounting formulas that do not explicitly recognize the joint production process and can thus distort the measurement of value-added output. Gordon convincingly addresses these points and concludes that indirect routing has indeed increased consumer surplus. Continuing on the problems with BEA and BLS approaches to both ALP and MFP growth calculations, Gordon notes that in the airline industry the two series differ largely because of the aggregation problem, because BLS employment grows less fast than the BEA figures (a fact largely attributable to the inclusion of Federal Express in the industry in 1986), and because NIPA output grows less fast than BLS figures because BEA uses deflated sales and BLS uses physical output and the deflator does not tract passenger yield well. With respect to the inclusion of Federal Express, however, is it not the case that the majority of their employees are really drivers of vans? (Also according to the Department of Transportation figures, roughly one-third of their employees are part-time.) He goes on to argue that measuring the average price of airline service is dicey and that even using yield as a deflator may overstate the growth of airline output relative to true quality-corrected output owing to the introduction of the complex regime of discount fares since 1977. Gordon makes a number of points about the mismeasurement of transportation service output. First, quality changes may not be important because the

425

Productivity in the Transportation Sector

deterioration in quality of service and the enhanced quality of service due, for example, to advanced boarding and seat reservations in airlines, are more than likely to cancel each other out. Second, frequent-flyer programs have created a significant upward bias in passenger yield. Third, changes in the efficiency of producing a “quality adjusted ton mile are of independent interest in productivity” because the production process has not been influenced by factors that have influenced quality, such as price discrimination. Here I disagree. Flight frequency and the routes themselves often cater to the business traveler and are influenced substantially by nonneutral quality changes. He concludes by stating that he has found the BLS data to be superior to the NIPA series for output and for employment. The NIPA productivity calculations are clearly suspect, but is this really a surprise in the airline industry? NIPA measures output by revenue and fares have been falling dramatically since 1977. Similarly, using revenue deflators such as the airfare component of the consumer price index does not recognize the extent of discounting of fares that has occurred. The same point could be made about the published tariffs of the LTL trucking industry. They do not adequately reflect the amount of discounting and contract rates after the trucking industry was deregulated. It is not clear that rail rate structures have changed significantly owing to the degree of competitive pressure from private-contract-exempt trucking. It is also unclear why Gordon dismisses the dramatically changing shares of passenger versus freight output for the reason that the NIPA index is so high and that the problem is with the inability of the producer price index to reflect greater pricing flexibility after the rail industry was deregulated. Winston et al. (1990) point out that rate structures, especially the discounted tariff and contract rates, appear to be quite stable before and after deregulation. The paper goes on to discuss MFP growth and its relation to the cost function. Although all of the analysis is carried in terms of a single output it could have been couched in a multiple-output setting (Denny, Fuss, and Waverman 1981). He imposes long run constant returns to scale. This is a strong assumption but one that does appear to have some empirical support. He goes on to discuss the capacity utilization issue and the mismeasurement of capital services owing to changing utilization rates (Hulten 1986; Berndt and Fuss 1986). At this point I would like to point out an alternative to the conventional view of airline service output. The production function, on which MFP estimates are based, specifies the maximum output produced by a set of inputs. The closest proxy to this is the number of available seats being moved from one place to another. Not unlike agriculture, unused seats are wastage because the distributor (marketer) of those seats has not done the job. In the case of the airlines, the farmer, the wholesaler, and the retailer are the same economic unit and a failure to correctly parcel up MFP growth among the various vertically integrated enterprises distorts measurement of output. Moreover, if revenue ton mile is used, then there are proxies for capacity utilization (other than load factor) that may be superior. In Sickles (1985) I constructed the

426

Robert J. Gordon

flying capital series by scaling down the quantity index on the basis of the discrepancy between the average time a plane was in service (ramp to ramp) during a quarter to the maximum that a plane of the same type was in service during the sample period in the entire industry. Also the work by F k e , Grosskopf, Lovell, and Pasurka (1989) and Fare, Grosskopf, Lovell, and Yaisawarng (forthcoming 1993) the producion of “goods” and “bads” could be used to evaluate the shadow prices of the “bad” output of the airlines, specifically indirect routing. Has there been a deterioration in the service provided by the carriers owing to indirect routing? Gordon counters the prevailing wisdom by convincingly pointing out that indirect routing has increased travelers’ options. However, the numbers that are cited as interlining of passengers may be systematically misleading because of code sharing. Under code sharing, a commuter carrier, for example, one of the American Eagle affiliates, uses the ticket code of a major airline. This makes it appear that the passenger is staying on the same airline, but its a rather muddy issue about whether or not it really is a different carrier because they are often at different concourses. This behavior results from airlines trying to capitalize on the benefits of feeder traffic. No assumption of increased circuitry is necessary if a cost-based study of productivity were undertaken in which various characteristics of the airline network are controlled for and thus their effects on airline costs estimated (see e.g., Good, Nadiri, and Sickles 1991). The increased options, routes, and so on that travelers face today are in place because of the tremendous economies of networking that characterize communication technologies. However, the coordination problem that exists with AT&T has essentially been resolved with the AR7 switch and more modem digital equipment. The coordination problem in the airline industry has not been resolved so costlessly. In order to assure that average arrival times coordinate in a complex network that is either in place or being pursued by most major carriers, waiting times must be higher, absent the coordination problem. Moreover, Gordon’s conclusion that ground congestion was not the cause of the increased scheduled flights times has a counter argument. Consider the following flight itinerary: A * B + C + D. Suppose that airport B is congested. The aircraft leaving at time A (a nonhub) might be held on the ground until it can get a slot for landing at B. The preclearance improves the safety at airport B by reducing the number of planes circling while waiting for a landing slot. Even if C and D are nonhub airports, their scheduled arrival time might be later than the old flight time because the arrival of the plane at C was delayed by the congestion at B. In other words, there are ripple effects of delay. Saying that they must be the results of “in route air traffic control capacity” rather than ground congestion really ignores the network aspects of airline service. With respect to the issue of complaints falling after deregulation, another indication that quality of service improved, it should be pointed out that selectivity problems with the complaints data cannot be dismissed. The filing of complaints is largely driven by expectations about the resolution of complaints. When the Civil

427

Productivity in the Transportation Sector

Aeronautics Board existed, there was an agency that could modify rewards and behavior of the carriers. T h e Department of Transportation merely keeps a tally of the letters. It has no regulatory teeth. In summary, I think the work by Gordon will stand as a focus of empirical research in the transportation sector for many years to come. I anticipate that the constructed output and input series and conclusions concerning them will remain robust to most changes that economists may argue are sensible, my comments notwithstanding.

References Baily, Martin N., and Robert J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 19(2): 347-420. Berndt, Ernst R., and Melvyn A. Fuss. 1986. Productivity Measurement with Adjustments for Variations in Capacity Utilization and Other Forms of Temporary Equilibrium. Journal of Econometrics 33: 7-29. Denny, Michael, Melvyn A. Fuss, and Leonard Waverman. 1981. The Measurement and Interpretation of Total Factor Productivity in Regulated Industries, with an Application to Canadian Telecommunications. In Productivity Measurement in Regulated Industries, ed. Thomas G. Cowing and Rodney E. Stevenson, 179-218. New York: Academic Press. Ftire, Rolf, Shawna Grosskopf, C. A. K. Lovell, and C. Pasurka. 1989. Multilateral Productivity Comparisons When Some Outputs Are Undesirable: A Nonparametric Approach. Review of Economics and Statistics 77: 90-98. Ftire, Rolf, Shawna Grosskopf, C. A. K. Lovell, and Suthathip Yaisawamg. Forthcoming. Derivation of Virtual Prices for Undesirable Outputs: A Distance Function Approach. Review of Economics and Statistics (May 1993). Good, David, M. Ishaq Nadiri, and Robin C. Sickles. 1991. The Structure of Production, Technical Change and Efficiency in a Multiproduct Industry: An Application to U.S. Airlines. NBER Working Paper no. 3939. Gordon, Robert J. 1990. The Measurement of Durable Goods Prices. Chicago: Univ. of Chicago Press. Hulten, Charles R. 1986. Productivity Change, Capacity Utilization, and the Sources of Efficiency Growth. Journal of Econometrics 33: 3 1-50. Sickles, Robin C. 1985. A Nonlinear Multivariate Error Components Analysis of Technology and Specific Factor Productivity Growth with an Application to the U.S. Airlines. Journal of Econometrics 27: 61-78. Sickles, Robin C., David Good, and Richard L. Johnson. 1986. Allocative Distortions and the Regulatory Transition of the U.S. Airline Insustry. Journal of Econometrics 33: 143-163. Winston, Clifford, Thomas M. Corsi, Curtis M. Grimm, and Carol A. Evans. 1990. The Economic Effects of Surface Freight Deregulation. Washington, D.C.: Brookings Institution.

This Page Intentionally Left Blank

11

Purchased Services, Outsourcing, Computers, and Productivity in Manufacturing Donald Siegel and Zvi Griliches

Official Bureau of Labor Statistics (BLS) multifactor productivity estimates indicate that productivity growth in manufacturing has improved substantially since the slowdown period in the 1970s.’ According to the BLS, multifactor productivity growth between 1979 and 1987 actually exceeded the preslowdown (1948-73) rate of increase. Levels of manufacturing employment have declined since the late 197Os, yet this sector’s share of total GNP has remained virtually constant (approximately 22%) during the last two decades. If accepted at face value, these findings imply that manufacturing workers have been displaced by higher growth in productivity and that the manufacturing sector is relatively healthy. Several economists have questioned the accuracy of the productivity measures that form the basis for these favorable conclusions, claiming that certain trends in the coordination of production have distorted conventional estimates of productivity, GNP, or value added by industry or sector.2 These distortions are alleged to have caused an upward bias in post-I979 estimates of productivity growth. Several trends may have resulted in an understatement of manufacturing input growth and thus (ceteris paribus) overstatement of value added or productivity change in the post-1979 period. These include the following: Outsourcing to the service sector. Examples include repair and maintenance services that might have been previously performed on site by plant workers are now contracted out to private firms. Also, there may be a greater need on Donald Siegel is assistant professor, Harriman School for Management and Policy, SUNY at Stony Brook, and a faculty research fellow of the National Bureau of Economic Research. Zvi Griliches is Paul M. Warburg Professor of Economics, Harvard University, and director of the Productivity Program at the National Bureau of Economic Research. 1. As reported in Baily and Gordon (1988). 2. See Mishel (1988) and Denison (1989). Denison’s criticism centers on recent hedonic adjustments to computer prices that, in his view, have led to an overstatement of productivity growth in manufacturing.

429

430

Donald Siege1 and Zvi Griliches

the part of manufacturing plants to purchase service sector inputs (i.e., legal, accounting, and other business services) or for their parent companies to provide them with a wide range of ~ e r v i c e s An . ~ increase in the volume of transactions between manufacturing and service establishments could affect measured productivity growth in two ways: First, the nominal value of these transactions may be unobserved. Standard measures of productivity change in manufacturing do not account for service-sector input^.^ An increase in the rate of purchase of these inputs may lead to an understatement of “true” input growth. Second, even when the nominal value of these services are properly accounted for, constant-dollar estimates of purchased services may be based on price deflators that overestimate price change, because they typically assume zero productivity growth for the respective industry providing the service.5 Outsourcing of manufacturing activities to foreign establishments. It is alleged that firms are increasingly likely to import intermediate materials and components, in order to take advantage of important differences in relative factor prices. A related issue is that due to a revision in the producer price index, price deflators for intermediate materials do not reflect import prices, which, because of a stronger dollar and other factors, have not risen as rapidly as domestic prices.6 Overestimation of input price change leads to underestimation of real input growth and overestimation of productivity growth. It could well be that accounting for service-sector inputs and foreign outsourcing is important because conventional estimates of manufacturing productivity or value added are based on the assumption that all inputs are derived from domestic factors of production within the manufacturing sector. An increase in the rate of investment in computers. This may lead to difficulties in measuring the flow of capital services. The argument for treating computers as a special type of capital is justified by the apparently large productivity gains experienced by computer manufacturers. As a result, Baily and Gordon (1988) report an average annual percentage decline of 14 percent 3. The annual survey of manufactures (ASM) and census of manufactures (COM) establishment data do not include information on central office operations and include only limited data on services purchased by manufacturing plants. 4. An exception is a paper by Gullickson and Harper (1987). which includes purchased business services as a factor of production in manufacturing. Values for purchased services, however, were not derived from data collected directly from establishments. Instead, 1977 input-output tables were used to estimate these flows. We use the I 0 data to supplement our data on purchased services at the four-digit SIC. 5. Some of the difficulties associated with productivity measurement in service industries are discussed in Griliches (1987) and Kendrick (1985). Suffice it to say that many economists are skeptical about the accuracy of productivity measures in the service sector. 6. Other important aspects of the producer price index (PPI) revision include the following: ( a ) the indexes are now constructed based on the theory of output price indexes (see Diewert 1983): ( b ) probability-based sampling techniques have been partially implemented; and (c) the PPI is now SIC based. See Triplett (1988) for a comprehensive discussion of the PPI revision.

431

Services, Outsourcing, and Computers

in the computer price index for the years 1969-87.’ Given the large increase in nominal expenditures on computers during this period, real investment in these machines and their relative weight in the capital stocks of representative industries are also substantially higher. Important technological improvements embodied in successive generations of computers may not have been properly accounted for in investment deflators associated with these capital goods. If this hypothesis is true, there might be an upward bias in measured total factor productivity (TFP) growth because of an underestimation of capital input growth in manufacturing industries that have made extensive use of computers yet do not produce them.8 The purpose of this paper is to document the incidence of these trends at the industry (four-digit SIC) level and to analyze the resulting effect on sectoral estimates of productivity. Specifically, we examine whether the post1979 improvement in measured productivity can be attributed to an increase in the rate of foreign or domestic outsourcing or to errors in the measurement of capital induced by expenditures on computers. If the incidence of outsourcing or investment in computers has risen substantially across industries since the late 1970s, we expect to find a strong correlation between an industry’s propensity to purchase computers, service sector, or foreign inputs and the difference between its post- 1979 and pre- 1979 productivity growth rates (acceleration in productivity). If our estimate of the timing of this relationship is imprecise, we would still expect to find a positive correlation between longrun measures of productivity change and an industry’s propensity to engage in these activities, although the major concern is that the measurement error is explicitly distorting recent measures. We have analyzed the following industry and establishment-level data sets to test this hypothesis: NBER productivity data base. Annual output and input measures for 450 manufacturing industries during the years 1958-86. This file is an updated version of the Penn-SRI data base created at the Census Bureau in the late 1970s and is described in full detail in Griliches and Lichtenberg (1984). NBER Immigration, trade, and labor markets data $files. Annual measures of imports, exports, and components of labor input for 450 manufacturing industries during the years 1958-86.9 Longitudinal research data base (LRDttime-series extract. Contains extremely detailed annual information on the output and inputs of approximately 7. The figures cited in Baily and Gordon (1988) are derived from the hedonic price deflators for computers developed by Cole et al., (1986), now incorporated (to some extent) in the national income accounts. 8. The BLS figures cited earlier for the entire manufacturing sector incorporate the effects of the hedonic price adjustment for computers. The BLS two-digit manufacturing data (see Gullickson and Harper 1987) apparently do not. An updated version of our four-digit SIC industry level data set includes these adjustments to output, but not to capital. 9. See Abowd (1990) for a complete description of these files.

432

Donald Siege1 and Zvi Griliches

20,000 plants for the period 1972-86. These plants were in continuous operation and were sampled annually during each of these years.'O 1977 and 1982 products and materials$le. Published tables, derived from the 1977 and 1982 censuses of manufactures (COM) on purchases of selected services, computers, and detailed data on the consumption of materials in the production process for 450 manufacturing industries. Additional data on services was obtained from input-output (1-0) tables. 1977 and 1982 censuses of auxiliary establishments. These are central and divisional offices that provide services to operating manufacturing plants (plants that produce output). R&D, clerical, managerial, administrative, sales, and other supporting activities are performed in auxiliary establishments. These establishments also report expenditures on services and computers. Another important aspect of this study is our careful auditing of the consistency of output, input, and productivity measures at the industry level. A review of the quality of these data revealed that many sectors were not consistently defined over time. Some of these anomalies may have been caused by the general decline in the magnitude of information solicited from establishments by the Census Bureau in conducting its economic surveys. Specifically, a change in the sampling framework of the annual survey of manufacturers (ASM) in 1979 reduced the number of plants sampled on an annual basis from approximately 73,000 to 56,000. In more than 15 percent of all manufacturing industries, there was a net decline of over 50 percent in the number of establishments surveyed in 1979, relative to 1978. Given that studies documenting the recent recovery in manufacturing often use 1979 (or 1981) as a base for assessing this improvement, we are concerned about the impact of attrition in the ASM sample on the variance of measured productivity change. In other words, our estimates of key variables in industries greatly affected by the change in the ASM sample design may be based on plants that are not truly representative of the industry.'.' Another consistency problem explored is the incidence of industry switching among establishments-that is, the reclassification of plants from one industry in 1977 to another in 1982 (using the LRD time-series extract). We also examine whether the acceleration in productivity is correlated with these two measures of inconsistency in data collection. The remainder of this paper is organized as follows: In section 11.1, we 10. In our version of the file, plants were sampled annually (and thus survived) through 1981. The panel data set is unbalanced after 1981. See McGuckin and Pascoe (1988) for an in-depth description of the characteristics of the full LRD. 11. In the future, we hope to analyze the full LRD file to determine whether plants dropped from the ASM panel in 1979 were low-productivity plants, possibly leading to biased estimates of productivity change in subsequent years. Olley and Pakes (1991) find that, for the telecommunications equipment industry, estimates of industry productivity growth differ substantially when one uses balanced or unbalanced establishment data.

433

Services, Outsourcing, and Computers

provide an exposition of the existence of errors in measurement in TFP growth, a problem that may have been exacerbated by recent trends in the coordination of production. In section 11.2, we present evidence on purchases of computers and service sector inputs in manufacturing. Section 11.3 examines the available data on the extent of foreign outsourcing in manufacturing. An analysis of the consistency of industry definitions and in particular, the impact of the 1979 ASM sample design change on individual industries, is contained in section 11.4. In section 11.5, we examine whether the post-1979 acceleration in productivity is correlated with the many possible sources of measurement error outlined throughout the paper. The final section consists of our preliminary conclusions and suggestions for future research.

11.1 Errors of Measurement in TFP Growth This section provides a framework for considering effects of the existence of errors in the measurement of real factor inputs on conventional estimates of TFP. We consider three possible sources of mismeasurement: l 2 ( 1 ) materials price deflators; (2) investment goods deflators; and (3) an omitted factor input-purchased services. Our estimates of TFP are calculated according to standard practice: log change in real output minus a cost share weighted average of the changes in real inputs.13 five inputs are measured-capital, production workers, nonproduction workers, energy, and nonenergy materials. The growth rates of capital and materials are assumed to be measured with error:

(2)

E;i(r)

=

E;i*(r)

+

Emr,

where and * superscripts denote observed and true growth rates, respectively. Thus, measured TFP growth is14 (3) where Q, = measured growth rate of output at time t; Sir = average share of factor i in total cost at time r; and X, = measured growth rate of factor i at time t ; and i = K , M , L,, L,, and E.I5 True TFF' growth is expressed as: 12. A fourth source of measurement error is considered in section 11.4-changes in sampling variance resulting from a change in the underlying characteristics of establishments sampled on an annual basis. 13. Where the weights are the arithmetic mean (between the current and previous year) cost shares of the respective inputs. 14. We have suppressed industry subscripts. 15. L, and L, refer to production and nonproduction workers, respectively.

434

Donald Siege1 and Zvi Griliches

2 S;,X:,, 6

DTFP; = Q:

(4)

-

I=

.

I

where the additional factor of production is X6(t) = svc*(t) = service input and all factors of production are measured without error. Note also that in our earlier specification of measured TFP we must assume that factor shares are also measured with error because of the omitted factor input (services):

+ I*.,, It can be shown that the following relationship exists between measured acceleration in productivity and true acceleration in productivity: (5)

= St;

= (DTFP;- DTFP;) (DTFP, - DTFP~)

+ (Sml-%o)

+

(%I

+ (Ski -Sm)

(E~, E ~ (p.kl )

- &mo)(Pml - Pmo) +

(S,, - S,)(svc, - svc,)

+

(&I

- IJO)

(eqI- e J ,

where the subscripts 0 and 1 refer to periods 0 and 1, respectively, and S, refers to the factor share of input i. We now consider how these errors arise. 1 1.1.1 Errors of Measurement in Capital We hypothesize that an industry’s investment deflator is measured with error when the industry has a high level, or growth rate, of investment in computers. The error in the investment deflator (PI) is transmitted to an estimate of industry j s net investment in capital (I)during year t:I6

(6)

+

IJI= vIJ;/PIJI,

where PI,^ = piJ; edkrand VI is the nominal value of new investment (capital expenditures). A recursive perpetual inventory algorithm is used to calculate the real net stock of capital in year T I 7 T

where DELTA is an estimate of the average rate of depreciation in industry j computed as the ratio of replacement investment to the net stock of capital, both in current dollars, and 7 is an estimate of the average service life of capital assets.I8 The capital stock is measured with error: 16. More specifically, the error is transmitted to estimates of the net stock of equiprnenr. 17. The procedures used to calculate the initial benchmark estimate of each industry’s capital stock are discussed in Fromm et al. (1979). 18. Measures of variables relating to capital investment are derived from the Bureau of Industrial Economics’ industry capital stocks data base. Implicit depreciation rates are calculated based on capital stock estimates and data on replacement investment that are not directly reported by firms.

435

Services, Outsourcing, and Computers log K, = log yr+

E

~

~

.

Because K is a moving average of past investments with weights related to the estimated rate of depreciation, E~~~is a moving average of investment deflator errors, with weighted depreciated (surviving) values of the respective net investments. The (cumulative) effects of overestimation of the investment deflator (PI), owing to a substantial increase in an industry's rate of investment in computers, can lead to underestimation of changes in the net stock of capital and thus, overestimation of total factor productivity growth. 11.1.2 Errors of Measurement in Materials

Constant-dollar values of materials are computed as the ratio of currentdollar values of materials to the NBER four-digit SIC industry price deflators for materials. It is likely that the materials deflator is measured with some error because of the use of foreign intermediate goods and materials in the production process. One important feature of the recent revision of the producer price index is that price deflators for intermediate materials no longer reflect import prices, which, because of a stronger dollar and other factors, have not risen as rapidly as domestic prices during the period in que~tion.'~ Overestimation of input price change leads to underestimation of real input growth and overestimation of productivity. In sections 1I .2 and 11.3, we describe the available data on the use of services, computers, and foreign materials in domestic manufacturing production. These data are used to test the various measurement error hypotheses outlined in section 1 1.1.

11.2 Service Sector and Computer Inputs in Manufacturing 11.2.1 Purchased Services Ideally, we would like to have detailed, comprehensive annual data on all types of purchased services by manufacturing establishments. With accurate measures of price change in service industries, we could then include servicesector inputs in the standard production function measures of TFP in manufacturing. Unfortunately, such data are unavailable. Beginning in 1977, data on selected purchased services have been collected from ASM establishments in census years. These data constitute the only direct information collected from manufacturing establishments on several types of service sector inputs: 19. According to the Federal Reserve Board, the multilateral trade-weighted value of the U.S. dollar (March 1973 = 100) rose from 93.1 in 1977 to 132.0 in 1985. The PPIs for industrial output and intermediate materials rose 68 percent and 58 percent respectively, during the same period.

436

Donald Siege1 and Zvi Griliches

(1) machinery repair and maintenance services; (2) building repair and maintenance services; and (3) communication services. Table 11.1 presents information on the de$uted cost of these selected purchased services in manufacturing establishments in 1977 and 1982. Although measures of price change in service industries may not accurately reflect quality change, we note that the price indexes for communication and repair services rose 25 percent and 34 percent, respectively, between 1977 and 1982.20These service expenses may play an important role in improving the quality of the flow of services derived from an establishment's capital stock. Subject to caveats concerning response rates, data on selected purchased services are published at the fourdigit SIC level. Table 11.1 demonstrates that constant-dollar expenditure on these selected services increased by 8 percent over the five-year period, with substantial increases (39 percent) in purchased communications services. The share of these services in total output, however, has not increased. For the manufacturing sector (not shown), total selected purchased services represented only 1.14 percent of nominal output in 1977 and 1.15 percent in 1982. In terms of levels of expenditure, the most striking numbers are those for SICs 22 and 23, textiles and apparel. Interestingly, these are industries that are alleged to engage heavily in foreign outsourcing. However, a more detailed analysis has revealed that the numbers for SICs 22 and 23 are based on questionable data for several four-digit industries.2' Note that these data do not constitute a complete accounting of all servicesector inputs.22This is demonstrated in columns 13 and 14 of Table 11.1, where a comparison is made between census data on selected purchased services and broader, inputed measures of purchased services by manufacturing industries (total services) from input-output The input-output data include additional service-sector inputs-finance, insurance, and real estate (FIRE), engineering and technical services, advertising, vehicle repair, medical, and educational services. The selected purchased services (communications, building and machinery repair services) accounted for 28.4 percent and 25.8 percent of total services in 1977 and 1982, respectively. Industry percentages ranged from 13.4 percent in instruments (SIC 38) to 78.3 percent in 20. Sources: the PPIs for SICS 48 11 (telecommunications) and 76 (miscellaneous repair services). respectively. 21. E.g., we found that reported purchased services declined from 318.8 million dollars in 1977 to 19.5 million in 1982 in SIC 2257 (circular knit fabric mills). The corresponding numbers for SIC 2396 (auto and apparel trimmings) were 553 million dollars in 1977 and 9.6 million in 1982. Table 11.3 of this paper contains a list of industries reporting large (possibly erroneous) changes in reported purchased services. 22. Perhaps unreported services, such as legal, accounting, and other business services, are increasingly likely to be purchased by manufacturing establishments. This would be consistent with the findings of Gullickson and Harper (1987). based on imputed data for nine types of business services. 23. Note also that different methods of collection are used for the census service data and the I 0 data.

Cost of Selected Purchased Services in Manufacturing Establishments, 1977 and 1982 (in millions of constant dollars)

Isble 11.1

Selected Purchased Services Industry name (2-digitSIC) Total manufacturing Food (20) Tobacco (21) Textiles (22) Apparel (23) Lumber (24) Furniture (25) Paper (26) Printing (27) Chemicals (28) Petroleum (29) Rubber (30) Leather (31) Stone, clay, glass (32)

Change

Machinery Repair Servicest

1977

1982

(%)

1977

1982

12,988

-2

3,025

3,296

+9

28.4

25.8

96. I

1,533 49 314 28 1 297 77 868 550 1,295

- 18

-6 18 24 + 69 - 16

1.876 31 612 695 379 93 702 67 I 1.324

-60 - 22 - 17 + 24 - 18 -2

431 18 158 305 45 31 93 I01 310

352 7 89 53 33 40 125 312

-18 -60 -43 -83 -28 +28 +35 +42 +0

33.8 78.3 62.3 69.0 42.3 13.6 42.4 19.8 18.1

20.5 79.8 26.2 20.0 34.4 11.0 43.4 18.1 15.6

119.2 62.5 153.3 145.6 125.4 77.3 104.1 72.9 102.0

+ 59

549

880

+ 60

233

349

+50

24.7

32.2

18.5

41 1 43 529

781

+90

40

-6 - 15

79 14 83

120 II 66

+52 -22 -21

27.0 14.5 33.2

55.3 13.7 38. I

90.9 44.1 76.3

Change

1982

(%)

1977

1982

(%)

1977

1982

20,691

22,361

+8

4,358

6,077

+ 39

13,308

2,742 54 1,029 1,508 508 193 920 1,211 1,941

2,228 67 466 526 408 176 1,149 1,438 1,870

- 19

-4

435 5 259 508 84 68 126 439 313

343 11 62 192 78 59 156 744 263

814

1,279

+ 57

32

50

589 72 70 1

1,282 66 79 I

+I18 -9 + 13

110 15

381 14 276

- 55 - 65

- 20 -8

+ 25

+ I9

90

Selected in Total (%)$

($6)

Change

1977

+ 24

Building Repair Servicest

Communication Services in Total, 1977(%)

Communication Services'

-21

+ 109 - 76 - 62

+ +

+ 283

+

-4 208

450

+ 59 - 49

Change

144

(continued)

'I8ble 11.1

(continued)

Selected Purchased Services Industry name (2-digitSIC) Primary metals (33) Fabricated metals (34) Nonelectric machinery (35) Electric machinery (36) Transportation equipment (37) Instruments (38) Miscellaneous manufacturing (39)

Communication Services'

Change

Machinery Repair Servicest

1982

(%)

1977

1982

140

-41

59.0

82.9

102.3

189

508

t170

39.5

60.1

100.0

-1

230

258

+I2

23.4

18.6

84.2

658

+41

189

333

+76

16.5

19.0

108.8

750

898

+ 20

205

254

+24

19.9

22.1

100.7

+ 49

107

159

+ 48

56

77

+37

13.4

13.0

80.2

- 50

Ill

80

- 28

25

26

f4

15.1

9.4

78.4

Change

Change

1977

1982

(%)

1977

1982

+ 189

1,640

1,758

+7

238

1,057

+ 169

1,204

1,322

+ 10

441

511

+ 16

706

699

+ 79

37 1

846

+ 128

467

1,532

+ 22

30 I

380

i

27

305

447

+46

142

21 1

263

170

- 35

127

64

1977

1982

(%)

2,009

2,277

+ I3

131

378

1,786

2,887

+ 62

393

1,377

1,468

+7

1,027

1,836

1.256

(%)

Sources: census of manufactures and producer price index.

'Telecommunications price deflator (SIC 481 I ) used for communication services. 'Miscellaneous repair price deflator (SIC 76) used for machinery and building repair services. 'Total services estimated from 1 0 tables (see text).

Selected in Total (a)*

Communication Services in Total, 1977(%)

Change 1977

Building Repair Services'

439

Services, Outsourcing, and Computers

tobacco (SIC 21) in 1977 and from 9.4 percent in miscellaneous manufacturing (SIC 39) to 82.9 percent in primary metals (SIC 33) in 1982.24 The decline in the percentage of census selected services in total services for the entire manufacturing sector in 1982, relative to 1977, is driven to a large extent by large percentage reductions in SICs 22 and 23. We again note that a more detailed analysis will reveal that these two-digit values may reflect anomalous data for several four-digit industries. In the last column of table 11.1, we compare census and Bureau of Economic Analysis (BEA) I0 estimates (aggregated to the two-digit SIC level) of communication services purchased by manufacturing plants. These values are roughly equivalent, although they are based on different data sources and method~logies.~~ In several sectors, most notably tobacco and petroleum, the census estimates are sharply lower than the corresponding BEA I0 estimates. The BEA I0 data also deviate sharply from the census data for SICs 22 and 23, providing additional independent evidence that some of the four-digit SIC values within these sectors may be erroneous. Descriptive statistics (not shown) on selected purchased services were calculated for 431 fourdigit SIC industries. There we observed only a relatively modest increase in purchased services by manufacturing industries. In constant dollars, the average industry spent 37 million dollars and 39 million dollars on communication, machinery, and building repair services in 1977 and 1982, respectively. In each period, over seventy percent of this expenditure was devoted to the repair of machinery and equipment. The mean cost share for selected services, or the ratio of selected purchased services to shipments, was relatively stable: 1.5 in 1977 and 1.2 percent in 1982. Industries reporting the highest cost shares of selected purchased services are examined in table 11.2. We also present levels of selected services and a measure of the importance of purchased repair and communication services, relative to the net stock of equipment. In general, the numbers seem plausible, given that many of these industries are highly capital intensive, and thus, are likely to require extensive repair and maintenance services. These ratios and measures of the importance of purchased services, relative to the industry’s capital stock, were used to identify suspected outliers, presented in table 11.3. The most striking feature of table 11.3 is the sharp changes in service cost shares observed during the sample period. Many of these values do not seem plausible. In the chewing gum industry, for example, the published figures yield a service cost share of 56 percent in 1977, which is clearly anomalous.26 24. In our empirical work in the final section of the paper, we supplement the four-digit specific service (census) measures with input-output measures at higher levels of aggregation (54 inputoutput industries within the manufacturing sector). 25. The input-output data on business services used in manufacturing industries are derived almost exclusively from indirect sources. The methodologies employed to estimate usage vary substantially across services, although they are generally based on proxy variables. E.g., the use of legal services is based on occupational distribution of lawyers by industry. Also, FDIC data on deposits by industry are utilized to estimate the use of banking services. 26. Note again that these costs shares include only selected services, not the complete array of service-sector inputs reflected in the input-output data.

440

Donald Siege1 and Zvi Griliches

Table 11.2

Industries with the Highest Shares of Selected Purchased Services in Gross Output, 1977 and 1982 (in millions of current dollars)

Ratio of Selected Purchased Services to Industry Shipments (%) Industry Name (4-digit SIC) Electron tubes (3671) Metal stampings (3469) Pressed &blown glass (3229) Aluminum foundries (3361) Newspapers (271 1) Industrial controls (3622) Carbon black (2895) Cotton finishing plants (2261) Malt (2083) Bottled & canned soft drinks (2086) Speed changers, drives, & gears (3566) Paperboard mill products (2631) Sheet metal work (3444) Gray iron castings (3321) Concrete block & brick (3271) Alkalies & chlorine (2812) Brick & structural clay tile (3251) Warp knit fabrics (2258)

cost of Selected Purchased Services (in millions of dollars)

Ratio of Machinery Repair and Communication Services to Net Stock of Equipment (beginning of year) (%I

Average 1977-82

1977

1982

1977

1982

1977

1982

9.8 6.2 4.9 4.7 4.2 3.5 3.4 2.9 2.9 2.8

10.2 3.8 1.5 5.8 4.0 0.6 3.9 4.3 4.1 1.5

9.5 8.7 8.3 3.6 4.4 6.5 3.0 1.6 1.6 4.2

19.4 177.8 31.5 142.5 527.1 15.5 18.2 32.6 20.6 147.7

26.2 559.3 225.5 108.4 931.0 280.4 18.9 11.9 10.8 703.0

25.0 19.3 3.2 25.7 16.7 3.5 8.3 7.9 31.0 6.8

41.3 41.5 24.5 15.2 26.3 53.9 8.7 2.0 8.2 28.0

2.7

3.7

I .8

44.9

28.8

19.8

8.1

2.7

2.0

3.4

142.8

321.8

3.6

5.4

2.6 2.6 2.6 2.5 2.4

3.9 2.7 2.2 2.1 2.3

1.3 2.5 3.0 2.9 2.5

191.0 200.5 24.6 34.4 18.2

86.8 153.1 38.8 45.4 16.2

30.5 8.4 5.6 2.8 5.7

11.8 5.0 7.6 3.3 4.3

2.3

3.9

0.7

54.4

10.7

22.1

3.2

In large part, the seemingly incorrect values for SICs 2337, 2396, and 2257 (along with several other four-digit SIC industries within the respective sectors that do not appear on the table) explain the sharp declines in purchased services for SICs 22 and 23 between 1977 and 1982. Initially, we hypothesized that these movements might have been caused by (a) industry redefinition, or possibly large plants switching in and out of adjacent industries (e.g., consider the changes in purchased services between 1977 and 1982 for SICs 2337, 233 1, and 2335, shown on table 11.3;*’or (b) low response rates to the 27. 1.e.. the decline in purchased services for SIC 2337 may be due to the reassignment of several plants to SICs 233 1 and 2335 (industries that experiences a sharp increase in purchased services).

441

Services, Outsourcing, and Computers

Table 11.3

Industries with the Highest Shares of Selected Purchased Services in Gross Output, 1977 and 1982 (in millions of current dollars)

Industry Name (4-digit SIC) Chewing gum (2067) Steel foundries (3325) Automotive & apparel trimmings (2396) Textile bags (2393) Hardware (3429) Plating & polishing (3471) Women’s & misses suits & coats (2337) Marking devices (3953) Circular knit fabric mills (2257) Prepared feeds (2048) Women’s & misses suits & coats (2335) Women’s & misses blouses & waists (2331)

Ratio of Selected Purchased Services to Industry Shipments (%)

cost of Selected Purchased Services (in millions of dollars)

Ratio of Machinery Repair and Communication Services to Net Stock of Equipment (beginning of year) (70)

1977

1982

1977

1982

1977

1982

56.0 1.8 25.5

0.7 33.1 0.5

317.5 42.2 553.0

6.2 693.9 9.6

274.4 1.6 253.4

3.8 101.3 5.2

18.2 1.8 I .4 13.4

0.5 13.7 13.3 0.1

58.5 95.4 26.4 389.7

2.2 788.1 363.6 6.2

147.8 8.7 5.5 209.8

5.2 53.6 68.8 3. I

11.9 10.1

0.7 0.7

29.5 318.8

2.2 19.5

81.2 29.4

4. I 2.1

4.2 0.3

0.6 3.8

368.9 11.7

62.6 176.3

39.0 3.2

8.0 56.0

0.3

I .7

6.7

66.4

4.8

48.1

questions concerning services and thus, unreliable estimates of service-sector inputs. As we will discuss in more detail in section 11.4, a special plant-level data set (a time-series extract of the LRD) was used to examine the consistency of industry definitions and reporting. These factors could not explain extreme movements in the data. Furthermore, the response rates for these industries to the questions relating to purchased services were actually quite high. One possibility is that the published figures are erroneous, or off by a few decimal points. Having analyzed the available direct evidence on purchased services by manufacturing establishments, we now examine data on intrafirm transfers of services. That is, we consider the services provided to operating manufacturing plants by central offices owned by the same parent company. 11.2.2 Central Administrative Offices It is important to note that (four-digit SIC) industry estimates of productivity and value added are based on information provided by operating manufac-

442

Donald Siegel and Zvi Griliches

turing establishments (OMEs), or plants that produce manufacturing output. In addition to purchasing services from outside vendors, OMEs are provided with services by auxiliary establishments, or central administrative offices (CAOs) operated by their parent companies. Clerical, administrative, managerial, and technical services are performed in CAOs. Many multiunit plants are serviced by these central offices and each auxiliary is assigned to a twodigit SIC.** The increasing importance of CAOs is demonstrated in table 11.4, which is based on the 1977 and 1982 censuses of auxiliary establishments. Although the number of employees in OMEs declined from 18.5 million in 1977 to 17.8 million in 1982, the number of CAO workers assigned to manufacturing establishments has increased from 1.1 million in 1977 to 1.3 million in 1982 (1 14 per establishment in 1977, 127 per establishment in 1982). All two-digit SICS, except paper (SIC 26), experienced an increase in the ratio of CAO to OME employees between 1977 and 1982. CAOs also purchase services from outside vendors. Data on selected purchased services by CAOs are presented in the last two columns of table 11.4.29On average, CAOs purchased about 10 percent as many of these selected services as OMEs, although growth in service expenditure was somewhat higher in CAOs. In the next section, we present evidence on the use of computers in manufacturing establishments. One interesting finding is that, in certain industries, substantial funds were spent on computers in CAOs (assigned to manufacturing establishments), relative to operating plants. That is, central and divisional headquarters provided important computer-driven services to OMEs as well. Estimation of the flows between the service and manufacturing sectors requires that we account for the contribution that CAOs provide to manufacturing plants. 11.2.3 Investment in Computers in the Manufacturing Sector The use of computers in manufacturing has become ubiquitous during the last two decades.3o Since 1977, manufacturing establishments have been asked to report their annual expenditure on new computers in conjunction with each COM.31As with services, these data constitute the only reliable, direct information collected from manufacturing establishments on computer expenditures. 28. See Lichtenberg and Siegel (1990) for a complete description of this file. 29. This information was not included in the published tables. However, we had access to the full microdata constituting the 1977 and 1982 censuses. Consequently, we were able to construct this table. 30. Actually, as reported in Baily and Gordon (1988). the rate of investment in computers is higher in other (nonmanufacturing) sectors of the economy, particularly communications and financial services. 31. All plants report fotal expenditures for new machinery and equipment. ASM establishments, however, are asked to provide derailed data on their total expenditures for new machinery and equipment-how much is spent on vehicles, computers, and other types of machinery and equipment.

Table 11.4

Employment and Cost of Selected Purchased Services in Central Office Establishments (CAOs) and Operating Manufacturing Establishments (OMEs), 1977 and 1982 (thousands)

Industry Name (2-digit SIC) Total manufacturing Food (20) Tobacco (21) Textiles (22) Apparel (23) Lumber (24) Furniture (25) Paper (26) Printing (27) Chemicals (28) Petroleum (29) Rubber (30) Leather (31) Stone, clay, glass (32) Primary metals (33) Fabricated metals (34) Nonelectric machinery (35) Electric machinery (36) Transportation equipment (37) Instruments (38) Miscellaneous manufacturing (39)

CAO Employment 1977

1982

1,074.1 102.1 8.0 33.2 27.5 14.5 9.0 37.6 38.9 181.4 65.3 25.2 10.8 41.1 46.8 50.1 93.1 140.0 105.5 32.8 11.2

1,275.9 108.9 14.2 32.9 34.8 21.9 10.6 31.3 48.3 206.5 76.0 31.7 9.0 41.8 47.5 51.5 137.7 191.3 108.4 58.5 13.1

OME Employment

Change(%)

+ 19 +7 + 78 -1

+ 27 +51 + 18 - 17 + 24 + 14 + 16

+ 26

- 17 +2 +I +3 + 48 + 37 +31 + 78 + 17

Sources: Census of auxiliary establishments and census of manufactures Note: CAOs that service operating manufacturing establishments.

Ratio of CAO to OME Employment

Ratio of CAO Purchased Services to OME Purchased Services

1977

1982

Change(%)

1977

1982

1977

1982

18,515.9 1,520.2 60.6 875.4 1,334.3 692.4 463.8 628.7 1,092.2 880.2 146.8 721.3 242.5 613.7 1,113.6 1,555.7 2,083.3 1,723.1 768.2 559.1 440.7

17,818.1 1,487.7 57.6 717.4 1,189.0 576.4 436.0 605.6 1,291.8 872.6 151.6 681.7 199.8 531.5 854.1 1,459.7 2,188.7 1,914.5 1,595.9 623.6 382.6

-4

,058 ,067 ,132 ,038 ,021 ,021 ,019

,072 ,073 .247 ,046 ,029 ,038 .024 ,052 ,037 ,237

,085 ,066 ,142 ,037 ,016 .059 ,062 ,061 ,052 .I28 .I13 .056 .117 .098

.i06

-2 -5 - 18 - 11 - 17 -6 -4 + 18 -1

+3 -5 - 18 - 14 - 23 -6 +5 +11 -

10

+ 12 - 13

,060

,036 ,206 ,445 .035 ,045 .067 ,042 ,032 ,045 .08 I ,060 .059

,025

,501

,047 ,045 ,079 .056 .035 ,063 .loo

,068 .094 ,034

,044

,046 ,120 .286 .I73 ,148 ,044

,092 ,364 ,097 ,081 ,146 ,070 ,053 ,045 ,171 ,105 ,035 ,163 ,094 ,047 ,033 ,157 ,258 ,163 ,229 . I19

444

Donald Siege1 and Zvi Griliches

Statistics on the rate of new investment in computers by manufacturing plants are reported in table 11.5.32It is important to note that these figures understate real investment in computers because current-dollar figures are used. Cole et al. (1986) report a 50 percent decline in computer prices between 1977 and 1982.33Table l l $ 5 also includes additional measures of the relative importance of computers and the rate at which these machines have been incorporated into the capital stocks of the purchasing industries. The largest absolute and percentage increases in new investment in computers occurred in SIC 36. The last column of table 11.5 contains a comparison of expenditures on computers by auxiliary establishments or CAOs and OMEs. Central offices spent 47 percent as much as OMEs on computers in 1977; 40 percent as much in 1982.34In 1982, the proportions of CAO to OME computer expenditure were highest in the petroleum, tobacco, chemicals, and food industries. High rates of investment in computers by central office establishments underscores the importance of accounting for CAO inputs. Table 1 1.6 presents the (four-digit SIC) industries with the largest average expenditure on computers. Four of these industries produce electric machinery, equipment, or components (SIC 36): SICS 3662 (radio and TV communication equipment), 3674 (semiconductors), 3679 (electronic components), and 3661 (telephone equipment). Not surprisingly, most are generally regarded to be high-tech industries. Several printing and publishing industries are also included on this list.35 In the next section, we consider another potential source of measurement error: the effects of foreign inputs (materials) on domestic production. Even if the nominal values of these transactions are properly accounted for, there may be errors in materials price deflators, because of differences between domestic and foreign materials prices. Current procedures involve the use of a domestic price measure in the deflation of materials input. Given that prices of domestic materials have, in general, risen more rapidly than prices of foreign materials over the sample period, estimates of real materials input may be overstated.

11.3 Foreign Outsourcing Another trend in the coordination of production in manufacturing alleged to have resulted in mismeasurement of productivity growth is foreign out32. We have excluded the electronic computing equipment industry (SIC 3573) from all calculations because it is the rate of investment in this industry’s output that we wish to examine. 33. This may be a relatively conservative estimate of the decline in the effective price of computing because it is based only on the price behavior of mainframe computers. Berndt and Griliches (1990) report more rapid price declines for microcomputers over the same period (197782); also see Cohen (1988). 34. Two-digit figures on computers expenditures by CAOs were not available for 1977. The CAO computer expenditure values are probably lower-bound estimates because only about 82 percent (87 percent in terms of employment) of these establishments respond to the inquiry concerning capital expenditures. Our interpretation of the documentation is that the Census Bureau does not weight up the sectoral data that is reported. 35. When we analyzed industries devoting the largest percentage of their capital expenditures to computers (not shown), four of the top six industries were in SIC 27 (printing).

445

Services, Outsourcing, and Computers

Table 11.5

Investment in Computers in the Manufacturing Sector, 1977 and 1982

New Capital Expenditures on Computers (in millions of current dollars) Industry Name (2-digit SIC) Total manufacturing Food (20) Tobacco (21) Textiles (22) Apparel (23) Lumber (24) Furniture (25) Paper (26) Printing (27) Chemicals (28) Petroleum (29) Rubber (30) Leather (31) Stone, clay, glass (32) Primary metals (33) Fabricated metals (34) Nonelectric machinery (35) Electric machinery (36) Transportation equipment (37) Instruments (38) Miscellaneous manufacturing (39)

1977

1982

Change (%)

640.3 35.4 0.5 19.7 13.8 8.5 9.4 18.4 138.4 49.8 2.7 8.4 2.3 40.8 34.5 30.5 69.8 70.8 42.4 37.0 7.2

1,907.6 76.4 9.6 25.1 21.6 13.5 18.3 57.4 265.0 119.2 15.5 27.6 3.7 27.6 93.3 95.3 201.9 428.3 241.5 145.1 21.2

198 +I17 t I820 + 27 57 + 59 + 95 +212 +91 + 139 474 + 229 + 61 - 32 170 +212 + 189 505 + 470 + 292 194

+

+

+ +

+ +

New Capital Expenditures on Equipment Devoted to Computers

1977

1982

Ratio of CAO to OME Expenditures on Computers, 1982

1.8

3.6 1.6 2.0 2.0 5.7 1.3 5.2 1.3 10.9 1.7 0.4 1.7 3.7 1.5 2.4 3.4 4.6 8.2 4.5 9.3 4.9

,399 ,825 ,885 ,331 .662 .496 ,104 ,157 ,068 ,763 2.026 .268 .703 ,279 ,137 ,183 .404 ,586 ,321 ,339 ,066

(%)

1.1

0.4 1.6 3.5 0.6 3.2 0.6 9.4 0.7 0.2 0.6 2.8 2.5 1.0 1.4 2.4 3.2 5.1 4.9 2.0

Sources: Census of auxiliary establishments and census of manufactures.

sourcing. This section describes the proxies we have developed to measure this activity at the industry level. Unfortunately, data on foreign outsourcing are not directly reported by manufacturing establishment^.^^ We have used two files to develop what we believe is a reasonably accurate proxy for foreign outsourcing in the production process: ( a ) the products and materials file1982 census of manufactures, which contains detailed (five- or six-digit SIC level) information on products and intermediate materials used by an industry in producing its final output; and (b)NBER trade and immigration data base, which provides data on industry imports for 450 manufacturing industries. By linking these two files, we can determine the extent to which industries are 36. The Census Bureau, recognizing the increasing affect of offshore production on value added, cost of materials, and other measures, added a special set of questions to the 1987 COM on foreign outsourcing. However, this information was requested only from plants in industries that are alleged to be actively engaged in this activity (automobiles, electrical and electronic products, and apparel).

Table 11.6

The Top Wenty Purchasers of Computers, 1977 and 1982

Industry Name (4-digit SIC) Newpapers (271 1) Radio & TV communications equipment (3662) Semiconductors (3674) Blast furnaces & steel mills (3312) Motor vehicles (371 1) Guided missles, space vehicles (3761) Electronic components (3679) Aircraft (3721) Photo equipment (3861) Instruments to measure electricity (3825) Industrial organic chemicals (2869) Commercial printing, lithographic (2752) Periodicals (2721) Telephone & telegraph (3661) Pharmaceutical preparations (2834) Aircraft engines & engine parts (3724) Book publishing (2731) Miscellaneous plastic products (3079) Motor vehicle parts (3714) Pressed & blown glass (3229)

New Capital Expenditures on Computers (in millions of current dollars) Average

1977

1982

110.4 90.6

84.1 23.9

136.7 157.2

54.5 46.6 38.6 37.2 35.8 33.6 30.2 24.3

11.7 26.5 n.a. 20.6 11.5 8.2 15.9 4.7

97.3 66.6 38.6 53.8 60.0 58.9 44.5 43.8

19.1 18.4

4.5 13.6

33.6 23.1

18.3 18.3 18.2 17.2 13.1 12.8 12.5 11.3

10.0 7.0 9.5 5.7 4.4 6.6 n.a. 18.0

26.6 29.5 26.8 28.7 21.8 18.9 12.5 4.5

Change (70)

+ 63

+ 558

+ 732 + 151 n.a. + 161 + 422 +618 + 180 + 832 + 647 + 70

+ 166 + 321 + 182 + 404 + 395 + I86 n.a.

- 75

Nore: n.a. indicates not available, except under “change” column, where it signifies not applicable.

New Capital Expenditures on Equipment Spent on Computers (%)

Ratio of Computer Expenditure to Net Stock of Equipment at Beginning of Year (%)

1977

1982

1977

1982

19.4 6.5

18.7 14.8

2.8

4.2 5.5

3.4 1.4 n.a. 19.9 6.4 5.1 6.3 6.3

8.2 3.4 2.3 25.1 10.7 9.9 6.9 19.5

0.7 0.2 n.a. 2.6 0.5 0.9 1.1

.o

3.4 0.4 0.5 6.5 3.7 3.3 2. I 7.5

0.2 4.2

1.4 2.9

0.0 0.8

0.3 1.1

14.8 3.8 3.2 3.9 6.9 0.7 n.a. 14.2

18.5 6.8 5.1 8.4 16.5 1.6 0.8 3.2

1.5 0.5 0.6 0.5 0.9 0.1 n.a. 2.3

4.4 I .8 1.2 2.2 4.0 0.3 0. I 0.5

1.1

1

447

Services, Outsourcing, and Computers

consuming (in their production processes) materials that are relatively import intensive. We calculated the shares of all products in the industry’s total cost of materials (from the products and materials file) and multiplied each share by its corresponding import share-the ratio of imports to the sum of output and imports (derived from the NBER trade file). Next, we computed the sum of these values to calculate an estimate of the percentage of foreign materials used in production. A simple example will suffice to illustrate our methodology. Assume that 80 percent of the cost of materials in the flat-glass industry is devoted to purchases of inorganic chemicals, and that the remainder is devoted exclusively to plastic materials. Using the NBER trade and immigration file, we calculate the import shares for the inorganic chemicals and plastic materials industries. Assume that these import ratios are 50 percent and 25 percent, respectively. Our estimate of the percentage of imported materials used in the flat-glass industry would be 45 percent [(.8 * .5) + (.2 * .25)]. In practice, this approach does not capture all of an industry’s outsourcing of foreign materials, mainly because, in almost all industries, a nonnegligible percentage (at least 5 percent) of the cost of materials is not specified by kind or consists of materials that fall outside the manufacturing sector (generally, commodity-based products such as rubber or precious metals). Subject to this caveat, we have calculated estimates of the percentage of foreign materials used in production in 1977 and 1982 for 414 manufacturing industries. Descriptive statistics are presented in table 11.7. The share of foreign materials in total cost of materials rose 1 percent between 1977 and 1982, averaging 4.3 percent over the period. These shares were used to calculate estimates of constant-dollar values of imported materials used in production in 1977 and 1982. The mean percentage change in the quantity of foreign materials was 48.1 percent, although the median percentage change was only 13.3 percent. Thus, the data appear to be consistent with the hypothesis that manufacturing industries are using a greater proportion of foreign goods to produce their domestic output.37 To examine the plausibility of our estimates, we have displayed the industries with the highest percentages of foreign materials on the top panel of table 11.7. Related industries appear to exhibit similar patterns of behavior in foreign outsourcing activities. Industries experiencing the largest increases in the use of foreign materials between 1977 and 1982 are presented on the bottom panel of table 11.7. Again, we find that related industries had similar increases. The largest percentage gain in foreign materials occurred in SIC 3843 37. Note that these constant-dollar values were not calculated based on separate price series for imported and domestic materials inputs. In the future, we plan to use the BLS’s PPI for imports at the detailed industry level to deflate these purchases. For our sample period, however, the BLS data were not available. When the 1987 COM becomes available, we will adjust our estimates accordingly.

448

Donald Siege1 and Zvi Griliches

Table 11.7

Imputed Measure of Foreign Outsourcing in Manufacturing: 414 Industries

Industry Name (4-digit SIC) Average manufacturing industry

PFM77 3.8

PFM82

4.8

PCRFM

CPFM

+ 1.0

+48.1

+4.9

+ 10.7

Industries using large % of foreign materials in production: Jewelry & precious metal (391 1) Nonferrous rolling & drawing, n.e.c. (3356) Wool yam mills (2283) Brass, bronze, & copper foundries (3362) Silverware & plated ware (3914) Weaving & finishing mills & wool (2231) Watches, clocks, & watchcases (3873) Nonferrous foundries, n.e.c. (3369) Cane sugar refining (2062) Cellulosic man-made fibers (2823) Hardwood veneer & plywood (2435) Steel investment foundries (3324) Paper mills (2621) Sanitary paper products (2647) Steel springs, except wire (3493)

42.1 34.3

47.0 47.9

+ 13.6

29.4 29.1

27.4 23.1

-2.1 - 6.0

21.5 26.1

23.0 18.3

+IS

-11.7

-7.8

- 30.9

17.6

25.9

+ 8.3

-22.0

18.5

22.0

+3.5

- 13.1

20.2 17.2 17.4

12.3 13.5 13.1

-7.9 -3.7 -4.4

-48.2 -37.7 -40.7

11.2 12.7 11.8 9.6

19.2 11.0 11.6 13.5

+8.0

- 1.7 0.2 +3.9

+ 178.4

- 0.2

- 35.0 -35.3

-

1.3

+ 15.6

-

-27.8

Industries experiencing large % increases in the use of foreign materials in production (1977-82): Dental equipment & supplies (3843) Nonferrous rolling & drawing, n.e.c. (3356) Watches, clocks, & watchcases (3873) Steel investment foundries (3324) Leather goods, n.e.c. (3199) Waterproof outergarments (2385) Steel wire & related (3315) Leather & sheep-lined clothing (2386) Pharmaceutical preparations (2834) Boot and shoe cut stock & findings (3 I3 1) Industrial fumances & ovens (3567) Current-carrying wiring devices (3643)

+ 1,915.9

1 .o

22.0

+21.0

34.3

47.9

+ 13.6

-0.2

17.6

25.9

+8.3

- 22.0

11.0 2.4 0.2 8.4 7.2

19.0 9.0 6.4 14.2 12.5

+8.1

i6.2 +5.7 +5.3

+2,667.8 + 19.7 + 29.3

4.6

9.9

+5.3

+ 129.9

2.7

8.0

+4.9

+ 171.4

2.0

6.9

+4.9

+ 241.8

0.6

7.5

+4.9

+ 173.4

+ 6.6

+ 178.4 + 158.5

Services, Outsourcing, and Computers

449

Table 11.7

(continued)

Industry Name (4-digit SIC) ~

_

_

PFM77

PFM82

CPFM

PCRFM

42.1 1.3 7.0

47.0 6. I 12.5

f4.9 +4.8 + 4.6

+ 10.7 +414.7 -5.3

_

Jewelry, precious metal (391 1) Printing trades machinery (3555) Leather gloves & mittens (3151)

F M ~= ~Imputed measure of percentage of foreign materials used in production (1977); P F M ~ Z= imputed measure of percentage of foreign materials used in production (1982); CPFM = change in percentage of foreign materials used in production ( P F M-~P ~F M ~ ~and ) ; PCRFM =

Notes: P

imputed measure of percent change in real foreign materials used in production (1977-82).

(dental equipment and supplies). Further analysis of the production process in this industry in both years revealed that the increase was caused by the adoption of a new semiconductor-oriented production technology during this peri~d.~~ In the next section, we report some findings based on our analysis of the consistency of the industry data.

11.4 Inconsistencies in Industry Definition and Sampling Procedures Since 1949, the Census Bureau has conducted an ASM in each year between censuses. Although the COM is designed to be a complete, comprehensive enumeration and description of the activities of all plants in the manufacturing sector, the ASM collects less detailed information (although, still quite comprehensive) for a survey sample of establishments. Approximately two years after a COM has been conducted, two types of establishments are identified. On the basis of employment, some plants are designated as “certainty” establishments and are required to report ASM data. The remaining establishments are sampled in accordance with standard statistical methods of probability sampling, where the probability of selection in the ASM panel is proportional to size, as measured by the plant’s value of shipments in its principal product class ( i n d u ~ t r y )From . ~ ~ 1949 through 1978, the sampling unit of the ASM was the firm. If a company owned at least one plant with 250 or more employees (based on the most recent COM), all its establishments were sampled with certainty.40 Small companies, or those that failed to meet the certainty cutoff level of 250 employees, were sampled with probabilities proportional to measures of firm size (value of shipments). Thus, plants owned 38. For symmetry, we analyzed industries that experienced the sharpest declines in foreign materials over the same period (not shown). Again, commodity-based products, such as paper, wool, and sugar-related products, experienced some of the most dramatic shifts. 39. The variance of annual fluctuations in shipments (and in certain cases, employment) in an establishment’s home industry is also taken into consideration. 40. Prior to 1969, all companies owning at least one plant with more than 100 employees were sampled with certainty.

450

Donald Siege1 and Zvi Griliches

by firms owning large establishments were highly likely to be included in a given ASM panel. Beginning in 1979, in an effort to reduce the cost of collecting and processing data, the Census Bureau redefined the sampling unit of the ASM to be the individual establishment, rather than the firm. The certainty cutoff level was again defined as 250 or more employees. As a result of this change in sample design, small plants owned by large, multiplant firms were no longer sampled with certainty. Instead, small establishments operated by large firms were treated in an identical fashion to small establishments operated by small firm^.^' The effect of the 1979 sample design change was to reduce the number of plants in the ASM panel from approximately 73,000 plants to 56,000 plants. Table 11.8 examines the effects of the reduction in the ASM sample across two-digit SICs. The largest absolute and relative declines occurred in SICs 20, 23, 24, 27, 28, and 32. On the other hand, SICs 35 and 38 had more plants sampled in 1979 than in 1978. This is due in part to greater representation of emerging growth industries in the 1977 COM and a better accounting of plant births. In table 11.9, we present descriptive statistics on the effect of the change in ASM sample design on four-digit SIC industries. The average industry experienced a decline of 15.6 percent in the number of plants sampled in 1979, relative to 1978. Sample size was reduced by more than 38 percent in over 25 percent of these industries (and by more than 50 percent in over 15 percent of these industries). The largest percentage declines in four-digit SIC industries are displayed on the bottom panel of table 11.9. Some of these declines (i.e., manufactured ice, SIC 2097) are very large and raise serious doubts conceming whether plants remaining in the sample accurately reflect the “true” distribution of plants in the An additional concern associated with the sample reduction is the concomitant decline in the number of potential respondents to detailed questions, such as those on purchased services and consumption of materials, that are directed only to ASM establishments (during census years). Another consistency problem explored is the incidence of sectoral switching among establishment^.^^ By definition, an industry is comprised of all establishments whose primary product is classified in a given SIC code. We 41. The revisions in the sampling methodology used in the selection of the ASM sample are described in full detail in Waite and Cole (1980) and also in U.S. Bureau of the Census (1985). 42. Several values are not reported on the table because of confidentiality concerns. When we raised the issue of whether the current industry samples of establishments might be biased with Census Bureau officials, they assured us that homogeneity of production was considered in the decision to reduce the number of plants sampled in a given industry. However, this subject was nor explicitly considered in the study conducted by Waite and Cole (1980) that describes the rationale for the change in ASM sample design. 43. Andrews and Abbot (1988) have examined this phenomenon and found it to be of importance in a number of industries.

451

Services, Outsourcing, and Computers

Table 11.8

Effect of Change in ASM Sample Design on Coverage of Manufacturing Establishments within 2-digit SIC Categories

Industry Name (2-digit SIC) Total manufacturing Food (20) Tobacco (21) Textiles (22) Apparel (23) Lumber (24) Furniture (25) Paper (26) Printing (27) Chemicals (28) Petroleum (29) Rubber (30) Leather (31) Stone, clay, glass (32) Primary metals (33) Fabricated metals (34) Nonelectric machinery (35) Electric machinery (36) Transportation equipment (37) Instruments (38) Miscellaneous manufacturing (39)

No. of ASM Plants, 1978 72,451 8,579 116 2,772 5,757 5,017 2,069 2,977 5,287 4,808 992 3,115 838 3,521 2,414 7,084 7,288 3,634 2,432 1,540 2,211

No. of ASM Plants, 1979 55,910 5,856 82 2,401 3,914 3,304 1,434 2,070 3,61 I 3,108 862 2,367 760 1,739 2,051 5,858 7,410 3,533 2,089 1,870 1,591

Change in ASM Plants - 16,541

- 2,723 - 34 -371

- 1,843 - 1,713 - 635 - 907

- 1,676 -

1,700 - 130

- 748 - 78 - 1,782 - 363 - 1,226

+ 122

- 101 - 343

+ 330 - 620

Change in ASM Plants ( S ) -22.8 -31.7 - 29.3 - 13.4 - 32.0 - 34. I - 30.7 - 30.5 -31.7 - 35.4 - 13.1 - 24.0 -9.3 -50.6 - 15.0 - 17.3 + 1.7 -2.8 - 14.1 f21.4 - 28.0

Notes: Changes are unweighted.

have demonstrated that after 1979, fewer plants were sampled on annual basis in most industries. In an industry with few plants, large plants switching out of (or into) that industry because of a change in product mix could have a dramatic effect on key sectoral variables. We examined the industrial classification of plants in the time series extract of the LRD file that could be matched across the 1972, 1977, and 1982 censuses. Our results are presented in table 11.10. We find that, on average, 13.9 percent of an industry’s plants switched four-digit SICS between 1972 and 1977; 12.2 percent between 1977 and 1982.@Rates of switching are slightly lower in terms of the output or employment assigned to that industry. The inconsistencies in the industry data outlined in this section may reflect a reduction in the quality of the data in a nonnegligible percentage of fourdigit industries. It is also possible that these anomolies may give rise to measurement error in the productivity statistics. Although it is impossible to reach a definite conclusion about the global effect of such inconsistencies without 44. Undoubtedly, some of these plants switched back in 1982 to their original classification in

1972.

452

Donald Siege1 and Zvi Griliches

Table 11.9

Effect of Change in ASM Sample Design (1978-1979)

Industry Name (4-digit SIC) Average manufacturing industry

Change in ASM Plants (%)

- 15.6

Absolute Change in ASM Plants

37

No. of ASM Plants 1978

No. of ASM Plants 1979

166

129

D 141 1066

D 20 173

D 76

D 14

- 124

62 51 227 178

12 11 55 54

-69.5

- 173

249

76

- 69.4

- 161

232

71

- 69.1

- 172

249

77

- 69.0

-591

857

266

- 66.4

- 101

152

51

-65.5

- 55

84

29

-65.1

- 84

129

45

-64.1

- 25

39

14

- 63.9

- 53

83

30

- 655

1026

37 1

- 35

55

20

Industries experiencing large % declines in ASM plants: Millinery (2351) Manufactured ice (2097) Ready-mixed concrete (3273) Buttons (3963) Engraving & plate ptinting (2753) Textile bags (2393) Marking devices (3953) Printing ink (2893) Curtains & draperies (2391) Fertilizers, mixing only (2875) Metal coating & allied services (3479) Adhesives & sealants (2891) Commercial printing, letterpress (2751) Architectural metal work (3446) Special product sawmills, n.e.c. (2429) Dog, cat, & other pet food (2047) Men’s & boy’s neckwear (2323) Nonmetallic mineral products, n . e x . (3299) Logging camps & logging contractors (241 1) Lime (3274)

D

-90.9 -85.8 - 83.8

- 893

- 83.3 -81.6

- 62

- 80.6 -78.4 -75.8 -69.7

-63.8 - 63.6

- 121

D

- 50 - 40 - 172

Note: D = not reported because of confidentiality constraints.

453

Services, Outsourcing, and Computers Industry Switching in the Time-Series Extract of the LRD File (LRDTS)

Table 11.10

LRDTS Switching 4-Digit SICS (%)

Quantiles Mean

.25

.50

.75

13.9 12.2

3.1 3.2

10.4 8.8

19.2 16.7

12.1 10.7

1 .o 1.1

6.4 5.0

16.2 14.6

12.0 11.5

1.3 1.3

6.8 6.0

16.6

Plants:

1972-71 1977-82 output:

1972-77 1977-82 Employment:

1972-77 1977-82

15.8

Notes: N = 448 manufacturing industries (approximately 18,OOO plants). We are measuring switching between censuses of manufactures. These results should be interpreted cautiously because there were fewer than five plants in certain industries in 1972, when we imposed the restriction that plants be present in 1972, 1977,and 1982.

further analysis of the characteristics of plants that were dropped from the ASM panel or those that shifted to new industries, we can examine whether these inconsistencies are systematically correlated with measures of productivity growth.

11.5 Total Factor Productivity Growth and Measures of Outsourcing and Inconsistency In sections 11.2 and 11.3, we discussed procedures for measuring the incidence of service-sector inputs, computers, and foreign outsourcing in manufacturing industries. In section 11.1, we described how increases in these activities may have exacerbated measurement error in factor inputs. In this section, we examine whether these trends are correlated with acceleration in productivity in the post-1979 period. First, we must determine whether we observe higher productivity growth in the 1980s at the detailed industry level. Current estimates of a recovery in manufacturing are based on data at higher levels of a g g r e g a t i ~ nIf. ~a~recovery is reflected in the data, we wish to determine whether the improvement in measured productivity growth is driven primarily by industries that are heavily engaged in activities that may have induced measurement error in the productivity statistics. Table 1 1 . 1 1 presents descriptive statistics on TFP growth for 392 manufacturing industries in three periods: 1959-73, 1973-79, and 1979-86. These results are essentially equivalent to TFP growth measures for all (450) manufacturing industries (not shown). TFP growth is calculated using standard growth accounting meth45. See Baily and Gordon (1988)or Mishel(1988)

454

Donald Siege1 and Zvi Griliches

Table 11.11

Descriptive Statistics on TFP Growth: ManufacturingIndustries Reporting Consistent Data on Outsourcing (%)

Variable (1) Average annual TFP growth 59-73 (2) Average annual T I T growth 73-86 (3) Average annual TFP growth 73-79 (4) Average annual TFP growth 79-86 (5) Change in TFP growth rates: (4)-(3)

Mean

Median

Standard Deviation

1.O

0.8

3.5

-4.9

6.0

0.2

0.2

4.9

-5.1

18.0

6.4

-9.3

19.3

0.0

-0.1

Minimum

Maximum

0.4

0.3

5.9

- 8.6

17.0

0.3

0.3

7.1

-7.3

15.2

Nore: N = 392. TFP measures include period-specific value-added weights.

ods-logarithmic change in real output minus a Tornqvist index of real factor inputs-capital (plant and equipment estimated separately), energy, nonenergy materials, production workers, and nonproduction The sum of cost shares is constrained to equal one, and capital’s cost share is calculated as a re~idual.~’ The productivity estimates are weighted by period-specific measures of value added. Note that these conventional measures of TFP are subject to the measurement error problems we described in section 11,l. The data reflect the slowdown in productivity during the 1970s and the subsequent recovery in recent years. The average industry experienced acceleration of L/3 percent in TFP during the period 1979-86.48 Similar patterns were observed when we calculated growth in value added. In table 11.12, we examine the relationship between TFP growth and various measures of service-sector inputs, outsourcing, and inconsistency in industry data. Variables 5-12 correspond to measures described in full detail in earlier sections of the paper. Glancing down column 4, we find that these measures are not strongly positively correlated with acceleration in productivity. This is true whether we measure these values in levels or first differences. Columns 1-3 demonstrate that these variables are generally not positively correlated with other measures of TFP growth. One exception is the correlation coefficient of .13 between acceleration in TFP and the average ratio of purchased services to output (including an adjustment for 10 services). TFP 46. For further information on the variables contained in the NBER productivity file, see Griliches and Lichtenberg (1984). 47. In the TFP calculation, the capital cost share is not divided between plant and equipment (the cost share is applied to the sum of net plant and equipment). 48. Note that our detailed industry file does not include data for 1987, which are reflected in the BLS TFF’ growth figures cited in the introduction to this paper. It is highly likely that our estimates of a recovery in manufacturing will be stronger when our file is updated to include 1987 data (a year of relatively strong economic performance).

Table 11.12

TFP Growth and Its Relationship to Purchased Services, Computers, Foreign Outsourcing and Estimates of Consistency in Industry Data

(1) Average annual TFP

growth (1973-86) (2) Average annual TFF' growth (1973-79) (3) Average annual TFF' growth (1979-86) (4) Change in TFP growth rates: (3) - (2) ( 5 ) Average ratio of purchased services to output (1977-82) (6) Change in ratio of purchased services to output (1982- 1977) (7) Average ratio of computer expenditures to capital expenditures (1977-82) (8) Change in ratio of computer expenditures to capital expenditures (1982 - 1977) (9) Change in ratio of CAO to OME employment (1982- 1977) (10) Average share of Imported materials (1977-82) (1 1) Change in share of Imported materials (1982 - 1977) (12) Decline in no. of ASM plants (1978-79)(%)

1 .OO

0.77*

1 .oo

0.83*

0.34*

1.oo

-0.04

-0.65*

0.50*

1.00

- 0.07

- 0.14*

0.01

0.14*

0.09

0.13*

0.04

0.08

0.30*

0.21*

0.23*

1.OO

-0.14*

1.OO

-0.06

-0.09

0.08

1.00

-0.03

0.02

0.34*

1.00

-0.04

0.01

-0.03

-0.04

0.01

0.09

-0.03

-0.10

0.07

0.03

0.01

0.01

0.03

0.05

-0.01

-0.05

0.01

0.12

0.06

0.10 -0.02

-0.10

-0.05

-0.04

0.06

0.10

0.12**

0.12** 1.00

-0.04

-0.14**

-0.11**

0.17*

0.10

0.14*

0.16*

-0.09 0.02

-0.03 0.11**

Note: N = 392 manufacturing industries, value-added weights. *Significant at .01 level. **Significant at .05 level.

-0.00 0.16*

1.00 1.00

0.061.00

456

Donald Siege1 and Zvi Griliches

growth (although not acceleration of TFP) is strongly positively correlated with an industry’s level of investment in computers. An additional variable measuring the incidence of industry switching among plants (not shown on the table) was also found to be uncorrelated with all measures of productivity change. Regressions of two alternative measures of industry performance: growth in value added and labor productivity growth (not shown) on the same sets of variables in table 11.12 yielded the same pattern of

11.6 Concluding Remarks These preliminary findings suggest that the recovery in measured manufacturing productivity growth cannot be attributed to increases in purchased services, foreign outsourcing, or a decline in the quality of industry data. Thus, our evidence is inconsistent with Mishel’s (1988) hypothesis that measured improvements in productivity significantly overstate true productivity growth because of the these trends. The results are consistent with the BEA’s gross product originating numbers that reflect an improvement in manufacturing performance in the 1980s. Another interesting empirical finding is the positive correlation between productivity growth (but nor acceleration in productivity) and investment in computers. We hope to investigate whether this result reflects errors of measurement of capital or is, in fact, indicative of the importance of computers as a determinant of productivity growth.50 Several important caveats must be considered. Our empirical analysis of activities that may distort conventional estimates of TFP is based only on data from the 1977 and 1982 COMs. These data may not reflect important changes that may have occurred since 1982. In this regard, we plan to extend our estimates when the 1987 census data become available in 1991. We also hope to improve our measures of the use of foreign materials by analyzing the geographic origin of materials and using exchange rates as price deflators. Also, it would be useful to test our measurement error model at higher levels of aggregation so the analysis would more closely correspond to existing studies. Although our study explores the incidence of mismeasurement of two inputs, capital and materials, we have not considered errors in the measurement of labor input that may arise from changes in the quality of hours worked by manufacturing employees (both production and nonproduction workers). Studies of aggregate economic growth (Denison 1962; Jorgenson, Gollop, and Fraumeni 1987; and Jorgenson and Fraumeni, chap. 8, this vol.) have 49. Labor productivity, which is less likely to be measured with error than TFP, was also strongly positively correlated with the level of investment in computers. 50. This result is consistent with the view that (see Bresnahan and Trajtenberg 1990) technological change can be imported into an industry through investment in computers. The authors argue that computers are a general-purpose technology that leads to substantial improvement in the technology of producing a good or service.

457

Services, Outsourcing, and Computers

included these adjustments, although controlling for quality change would be

more difficult at the detailed industry level. Our preliminary findings suggest that the recovery in measured manufacturing productivity growth cannot be attributed to increases in purchased services, foreign outsourcing, or a decline in the quality of industry data. Finally, we have highlighted certain inconsistencies in the industry data that merit additional analysis, such as changes in the sampling framework of the A S M and a high incidence of plants switching industries between economic censuses. Although w e failed to establish that measures of inconsistency are systematically correlated with levels or changes in productivity growth, further examination of the effects of such anomalies on the quality of the four-digit industry data is warranted.

References Abowd, John M. 1990. The NBER Immigration, Trade, and Labor Markets Data Files. NBER working paper no. 3351. Cambridge, Mass., May. Andrews, Stephen H., and Thomas A. Abbott 111. 1988. An Examination of the Standard Industrial Classification System of Manufacturing Activity Using the Longitudinal Research Database. Working paper. U.S. Census Bureau, Center for Economic Studies, June. Baily, Martin N., and Robert J. Gordon. 1988. The Productivity Slowdown, Measurement Issues, and the Explosion of Computer Power. Brookings Papers on Economic Activity 19(2): 347-420. Bemdt, Ernst R., and Zvi Griliches. 1990. Price Indexes For Microcomputers: An Exporatory Study. NBER working paper no. 3378, June. Bresnahan, Timothy, and Manuel Trajtenberg. 1990. General Purpose Technologies and Long-Term Growth. Paper presented at NBER Productivity Seminar, March. Cohen, Jeremy M. 1988. Rapid Change in the Personal Computer Market; A QualityAdjusted Hedonic Price Index, 1976-1987. Unpublished M. S. thesis, Alfred P. Sloan Schoolof Management, MIT, May. Cole, Rosanne, Y. C. Chen, Joan A. Barquin-Stolleman, Ellen Dulberger, Nurhan Helvacian, and James H. Hodge. 1986. Quality-Adjusted Price Indexes for Computer Processors and Selected Peripheral Equipment. Survey of Current Business 66, no. 1 (January): 41-50. Denison, Edward F. 1962. Sources of Economic Growth in the United States and the Alternatives before Us. New York: Committee for Economic Development. . 1989. Estimates of Productivity Change by Industry: An Evaluation and Alternative. Washington, D.C. : Brookings Institution. Diewert, W. E. 1983. The Theory of the Output Price Index and the Measurement of Real Output Change. In Price Level Measurement: Proceedings from a Conference Sponsored by Statistics in Canada, ed. W. E. Diewert and C. Montmarquette, 1049-1 113. Ottawa: Ministry of Supply and Services. Fromm, G., L. R. Klein, F. C. Ripley, and D. Crawford. 1979. Production Function Estimation of Capacity Utilization. Univ. of Pennsylvania. Mimeographed. Griliches, Zvi. 1987. Productivity: Measurement Problems. In The New Palgrave Dictionary of Economics, ed. Murray Eatwell, 3. London: Macmillan.

458

Donald Siegel and Zvi Griliches

Griliches, Zvi, and Frank R. Lichtenberg. 1984. R&D and Productivity Growth at the Industry Level. Is There Still a Relationship? In R&D, Patents, and Productivity. ed. Zvi Griliches, 465-96. Chicago: Univ. of Chicago Press. Gullickson, William, and Michael J. Harper. 1987. Multifactor Productivity in U.S. Manufacturing 1949-83. Monthly Labor Review, October, 18-28. Jorgenson, Dale W., F. W. Gollop, and Barbara Fraumeni. 1987. Productivity and U.S. Economic Growth, 1979-1985. Cambridge: Harvard Univ. Press. Kendrick, John W. 1985. Measurement of Output and Productivity in the Service Sector. In Managing the Service Economy, ed. R. Inman, 465-96. New York Cambridge Univ. Press. Lichtenberg, Frank R., and Donald Siegel. 1990. The Effect of Ownership Changes on the Employment and Wages of Central-Office and Other Personnel. Journal of Law and Economics 33 (October): 383-408. Mishel, Lawrence R. 1988. Manufacturing Numbers: How Inaccurate Statistics Conceal US.Industrial Decline. Washington D.C.: Economic Policy Institute. Mimeographed. McGuckin, Robert, and George Pascoe. 1988. The Longitudinal Research Data BaseResearch Possibilities. Survey of Currenr Business 68 (November): 30-37. Olley, Steve, and Ariel Pakes. 1991. The Dynamics of Productivity in the Telecommunications Equipment Industry. Working paper. U.S. Census Bureau, Center for Economic Studies, Washington, D.C.: January. Triplett, Jack E. 1990. Hedonic Methods in Statistical Agency Environments: An Intellectual Biopsy. Fifry Years of Economic Measurement, ed. Emst R. Bemdt and Jack E. Triplett. NBER Studies in Income and Wealth, vol. 54. Chicago: Univ. of Chicago Press. U.S. Bureau of the Census. 1985. Longitudinal Establishment Data (LED) Documentation. Washington, D.C. Waite, Preston J., and Stacey J. Cole. 1980. Selection of a New Sample Panel for the Annual Survey of Manufactures. Paper presented at the annual meeting of the American Statistical Association, August 11-14.

Comment

M. Ishaq Nadiri

In this interesting paper, Seigel and Griliches examine whether the observed increase in total factor productivity (TFP) growth at the total manufacturing level between 1979 and 1987 is partly due to mismeasurement of growth of inputs. They focus on three areas of potential mismeasurement: (1) outsourcing of some activities by the manufacturing sector to the service sector, (2) import of intermediate materials and components from foreign establishments and (3) increases in the rate of investment in computers by different manufacturing industries. If there are measurement errors present from these sources and they are not taken into account, the observed deflators for capital and materials are overstated, thereby underestimating the growth of real inputs and leading to an overestimation of measured TFP growth.

M. lshaq Nadiri is the Jay Gould Professor of Economics at New York University and a research associate at the National Bureau of Economic Research.

459

Services, Outsourcing, and Computers

The basic model employed in this paper is straightforward: TFP is calculated using the standard technique of logarithmic change in output minus a Tornqvist index of real factor inputs. Six inputs are considered. The first five are capital, energy, nonenergy materials, production workers, and nonproduction workers. Capital and materials are assumed to be measured with errors due to the overestimation of investment deflator (not measuring the decrease in computer prices) and material deflator (not reflecting import prices). The sixth input is purchased services such as machinery and building repair and maintenance services and communication services. The authors assemble a large body of data at the four-digit SIC industry level from a variety of sources. Careful examination of different bodies of data reveal substantial increases in the use of computers, particularly in high-tech and petroleum, chemicals, food and tobacco industries. The authors also document significant percentage changes in the use of foreign materials among different industries. They also explore the effect of sample design changes by the Census Bureau in 1979 and sectoral switching of plants among establishments. The surprising result of this paper is that with all the careful effort to document the trends that have exacerbated measurement errors in factor inputs, these errors are not correlated with productivity change or with the post 1979 acceleration of TFP growth. Siegel and Griliches report evidence of acceleration of about '/3 percent in TFP for the period 1979-86 at the disaggregated four-digit industry level. That is, the recovery in measured manufacturing productivity cannot be attributed to increases in purchased services, foreign outsourcing or decline in the quality of industry data. The analysis presented in the paper requires a great deal of effort and is a necessary requirement for solid empirical work. The authors should be commended. I would like to raise a few questions and suggest some possible extensions for future work: First, Siegel and Griliches consider mismeasurement of only two inputs: materials and capital. It could be argued, however, that errors of measurement may exist in output deflator, hours worked, energy, and employment data. Clearly the authors need to examine these potential sources of errors as well to insure the reliability and accuracy of the results presented in this paper. Second, the measurement errors of different inputs may interact depending on the underlying production structure. For example, the introduction of new computers may lead to compositional changes in labor and materials as well as changes in the quality of output. Such changes may not be captured by the relevant deflator. Third, the treatment of computer purchases excludes computer parts and, more importantly, the use of software. However, because the increased innovation in software is a primary source of enhancement of computer services, this should be reflected in the measurement of the capital stock deflator. To adjust for import prices, a disaggregation of capital and materials by country of origin would be needed because the exchange rate differs by country.

460

Donald Siege1 and Zvi Griliches

Fourth, The surprise in the paper is that after all the adjustments, some fairly sizable, the measurement errors and sampling changes are not correlated with TFP growth and particularly its acceleration after 1979. The problem could be either that the measurement errors offset each other or that the magnitudes of the changes in the period are not large enough to affect TFP growth or that the list of adjustments may not be extensive enough. Finally, if one is interested in explaining aggregate TFP growth, it may be that, as Griliches argued a quarter of a century ago, aggregation pays. However, measurement errors may affect TFP growth at the individual industry level. Because there is considerable interest in industry productivity growth, the authors might want to consider this line of research.

12

Dispersion and Heterogeneity of Firm Performances in Nine French Service Industries, 1984-1987 Elizabeth Kremp and Jacques Mairesse

The present paper has three distinct but intertwined motivations, pursuing jointly three purposes, each corresponding to one of the subsequent sections. Since the early 1980s, the French National Institute of Statistics and Economic Studies (INSEE) has been conducting an annual survey of market services, which is thought to be a very good, and in some respects rather unique, source of general information on this sector. Our first goal is to give a brief description of this survey (in section 12.1 of the paper). This survey not only is useful to ensure a knowledge of the relevant macrofacts but also provides a wealth of microeconomic information on the structure of these industries. In recent years, an increasing number of studies have taken advantage of information at the microlevel to investigate the behavior and performance of firms. Most of these studies have, however, concentrated on manufacturing industries, because the more easily accessible data bases cover primarily large publicly traded corporate companies, which are numerous in these industries. In view of the growing importance of service industries, it is clearly desirable to initiate similar studies also for them. The outlooks of economists working at the micro- and the macrolevels, and the ways they treat the data are quite different. Our interest, in section 12.2 of the paper, is to illustrate some of the basic problems involved and to provide At the time of this writing, Elizabeth Kremp was a research economist at the National Bureau of Economic Research; her work was supported by a grant of the French Ministry of Foreign Affairs. She is presently an economist with the Banque de France. Jacques Mairesse is a research associate of the National Bureau of Economic Research and a professor at Ecole nationale de la statistique et de l’administration Cconomique (ENSAE-Paris) and Ecole des hautes ttudes en sciences sociales (EHESS-Paris). The authors are grateful to Zvi Griliches for encouragement and suggestions and to the participants of the NBER productivity seminar for comments. We thank also Jean Albert, Jean Marie Chanut, Ian Cockburn, Marc Tajan, and Philippe Trogan for their help in gaining us access to the French annual survey of market services data.

461

462

Elizabeth Kremp and Jacques Mairesse

some indications of how they can be dealt with. We do this illustration in analyzing the productivity and profitability performances of firms in selected service industries, for the four recent years, 1984-87, for which the French survey was available to us. More precisely, we have concentrated on large firms with 20 or more salaried employees, because they are exhaustively surveyed and have to answer a more detailed questionnaire. We have also selected nine service industries that we thought typical in various ways. These are industries at the four-digit level of the French classification of industrial activities, Nomenclature dactivite's ef de produits (NAP), with at least 200 or 300 large firms. They all belong to the private competitive sector and fall in the category of personal services, where direct provider-customer interrelations are essential. Two of them are traditional consumer services, which have recently undergone important changes: restaurants and hotels. The seven others are producer services with different characteristics: engineering services, computer programming, computer processing, legal services, accounting services, personnel supply and building cleaning services. I We focus on four measures of performances or outcome variables. We take sales per person and (preferably) value added per person, as measures of labor productivity, and value added to sales ratio and (preferably) operating income to sales ratio (price cost margin), as measures of profitability margins.2 We consider these variables, both in levels (in the beginning and ending years, 1984 and 1987) and in rates of growth or changes (over the three-year period 1984-87).3 For the approximately 7000 large firms that were surveyed from 1984 to 1. Among the producer services, one might also distinguish between engineering services, computer programming, legal services, and accounting services, which are in the nature of counseling, and computer processing, personnel supply and building cleaning, which are more in the nature of doing. One should also note that personnel supply is not readily comparable to the other services in the sense that temporary workers could be considered as an intermediate input and not as labor (because they are actually recorded in the survey together with permanent employees). 2. The measure of these variables is straightforward enough on the basis of the information provided in the survey, and only three points need to be noted: The number of-persons includes both salaried employees and nonsalaried persons. Value-added and operating income have been corrected to include expenditures on rented capital buildings and equipment. For a number of firms, the fiscal year, for which we have their accounts, is different from the calendar year; we found, however, that this timing problem did not matter much, and we have not done any corrections for it in the present work. 3. Rates of growth are computed for sales and value added per person, as the three-year differences in logarithms; the absolute changes are considered for the value added and operating income to sales ratios. Because we had no information on the prices of services at the firm level, in order to compute our measures of the rates of growth of productivity, we have deflated sales and value added by the corresponding aggregate price indexes, which are available at the four-digit level of the industrial activity classification. These industry price indexes are themselves rather rough; the deflated figures should be, however, more akin to real productivity indicators and more comparable across industries. Although we report in this paper sales and value added per person in nominal francs per person (usually for 1987). the corresponding rates of growth are thus given in terms of volume, i.e., constant francs of 1984. There are no such problems of deflation for the profitability margins that are expressed naturally in percentages (of total sales).

463

Dispersion and Heterogeneity of Firm Performances

1987 in our nine selected industries, we have been able to construct a balanced and cleaned panel sample of 2289 firms. The first problem that we touch on is just that of constructing a sample and assessing some of the differences that arise in going from the analysis of the population to that of a sample. This problem raises in fact the difficult and more fundamental issue of the renewal of the population through the entry and exit of firms on the one hand and that of firms that should be viewed as outliers (or else that report incomplete or erroneous information) on the other hand. The second typical problem that we also illustrate is that of defining an average level and growth rate, for example, productivity, for an industry and of comparing the numbers that macro- and microeconomists usually compute. In fact, the microeconomist is concerned not only with the average characteristics of the variables of interest but also with many other aspects of their full distributions. The differences between the various averages are only the reflection, more or less transparent (and easily interpreted), of the magnitude (and changes in magnitude) of the dispersions and correlations of these distributions. One of the most striking phenomenon when analyzing microdata is precisely the extreme variability that they reveal. Part of such variability may be accounted for by heterogeneity factors, such as differences in specific activities, historical and environmental conditions, but a large part must also correspond to intrinsic or true d i ~ p e r s i o nIn . ~section 12.3 of the paper, we document the extent of the variability in the productivity and profitability variables in our sample of service firms and contrast it with the differences in the average levels of these variables across industries. We do so both cross-sectionally (in 1987) and in the time dimension (over 1984-87), in an attempt to exhibit a few of the heterogeneity categories that are usually thought to be relevant and that we could distinguish.

12.1 The French Firm Annual Survey on Services The survey on services is part of the general French system of annual firm surveys (enqu6tes annuelles d’entreprises). It is the last to have been launched in the early 1980s, and it is directly managed by INSEE. Over the yeqs, its scope has been extended, and it presently covers all market services, except health, social care, education, and research activities. Sixty-two industries at the four-digit level of the French NAP are now surveyed, involving some 600,000 service firms, and about 2,500,000 persons (2,000,000 salaried and 500,000 nonsalaried) in 1987.5 Table 12A.1 in the appendix provides some illustrative statistics at the two-digit industry level for all firms and for firms with 20 or more salaried employees in 1987. 4. Part of the variability, of course, is bound to arise also from the numerous observational and measurement errors. 5. This is a major survey with a permanent staff of over 80 employees.

464

Elizabeth Kremp and Jacques Mairesse

The survey is a survey of firms or enterprises, in the sense of juridically independent profit-making entities. Liberal professions, such as lawyers and accountants, are included, but nonprofit organizations are not. The service firms surveyed are classified according to their main activities and can have one or more different establishments.6 The survey is conducted by sending a detailed mail questionnaire to all firms with 20 or more salaried employees and a simpler one to a representative sample of smaller firms. The sample for the latter is stratified by size categories and activities (the sampling rate varying between 1 and 1/100) and is renewed by half each year. The rate and quality of the answers are deemed quite satisfactory, especially considering that a very large number of very small firms (with zero, one, or two salaried employees) are surveyed.' Basically, the survey provides detailed information on the current income accounts of the firms, as well as complementary information on their labor force and capital assets. Table 12.1 summarizes the structure and contents of the questionnaire for the larger firms (with 20 or more salaried employees). The larger firms have to report their statement of income and expense for the last accounting period (fiscal year) with a breakdown of some 30 operations (sales of merchandise, purchased goods, and produced services; purchases of goods and raw materials; changes in inventories; taxes; wages and social security costs; interest incomes and expenses; profits and losses). All firms are asked to give a detailed breakdown both of their total turnover (chi& d'ufuires) by services (400 different services or commodities for 62 activities) and of their purchases (about 30 categories, including goods purchased for resale and various intersectoral exchanges). For labor the following items are given: the total number of salaried em6. The survey is une enqu2re de secteur, covering all the activities (main and secondary ones) of the firm, and is different (in accordance to the distinction of the French national accounts between sectors and branches) to what would be une enquBre de branche, corresponding to units of production having the same activities. Branch surveys exist in manufacturing industries and other industries but not in services. The operational definition of the main activity (or primary industry) of a firm is explained in M. Tajan (1986). The problem is less difficult than in other sectors, because the majority of service firms are small, and most of them tend to be quite specialized. 7. About 70,000 questionnaires (of which 11,000 for the firms with 20 and more salaried employees) were sent for the 1987 survey in March 1988. The rate of nonresponse has been about 20 percent, nearly half of which corresponds to firms that have ceased their activities in 1987. Among the questionnaires returned, another 7 percent were also for firms interrupting their activities, and some additional 14 percent were not usable for various reasons. In terms of number of firms the rate of missing, incomplete, or erroneous data is thus about 20 percent, but is only about 6 percent in terms of number of employees or value added. Starting in 1989 for the year 1988, the sample has been expanded to 90,000 questionnaires, in order to obtain more reliable detailed results at infraregional levels. For more information, see the publications presenting the survey results for the various years. 8. The parts of the questionnaire that ask for the detailed breakdown of sales and purchases are specific to the different service sectors. Such detailed information is useful in particular to determine the main activity of firms; it is also important for the construction of branches accounts in the national accounts.

465

Dispersion and Heterogeneity of Firm Performances

Table 12.1

Detailed Questionnaire for Large Firms (with 20 or more salaried employees on December 31 of the year of the survey)

Firm characteristics: Identification number (SIREN) Address Legal form of organization Tax system Conditions of activity: End and length of fiscal year Description of the activity (creation, merger, modification of ownership, disappearance . . .) Employment and wages: Number of salaried workers: supervisory, nonsupervisory, part-timers, and family workers Quarterly distribution of salaried workers and number of hours worked Nonsalaried workers Earnings and fringe benefits Breakdown of sales (turnover) varying according to the different industries Profit and loss account: Expenditures Income Sales of produced goods Purchases of goods Sales of produced services Purchase of raw materials Changes in inventories Financial yields Taxes Wages and salaries Taxes on profits Capital and investments Total capital outlays at the beginning of the year Investment and retirement during the year Total capital outlays at the end of year Breakdown of investments between investments acquired and investments brought through a modification of ownership and according to seven categories: land, new buildings and structures, existing buildings and structures, new transportation equipment, secondhand transportation equipment, new machinery and other equipment, and secondhand machinery and other equipment. Breakdown of expenditures, varying according to the different industries Goods purchased for resale Interindustry exchanges Rented capital (equipment and properties) Subcontracting

ployees at the end of the year, with a distinction between professionals (i.e., managerial, executive, and supervisory personnel), other full-time employees, part-time employees and apprentices; the total number of nonsalaried persons with a distinction between owners and associates (or independent workers), full-time family workers and part-time ones. The total number of hours worked by salaried employees during the calendar year is also asked, together with the corresponding wage bill. For capital, larger firms report the gross book value of their fixed assets that is registered in their balance sheets at the beginning and end of their fiscal year, and they have to provide a decomposition of the change in gross book

466

Elizabeth Kremp and Jacques Mairesse

value that occurred over the fiscal year, in terms of acquisitions, cessions, discounts, revaluations, and other adjustments. For all firms, investment expenditures (measured on the basis of acquisitions) are detailed in seven categories: land; new and existing buildings and structures; new and secondhand transportation equipment; new and secondhand machinery and other equipment. 12.2 Average Productivity and Profitability Performances: From the Survey to Sample and from Macro- to Microaverages Economists working at the microlevel and those working at the macrolevel have divergent perspectives. Even when they investigate the same issues, adopt the same models, and rely on the same basic econometric techniques, because the data they use are so different, the ways they look at them in practice are also very different. This difference is already apparent with the problem of defining the scope of study: the macroeconomist considers the population as a whole (e.g., a complete industry); the microeconomist usually deals with a sample (e.g., of firms in a given industry). This difference is also clear in the supposedly simple question of measuring an average level or growth rate of an economic variable such as productivity (for a given agreed-on definition). In general, the possibilities offered by microdata (typically cross-sectional or panel data coming from surveys) are much larger than for macrodata (typically aggregate time series provided by national accounts), but the difficulties in dealing with them tend also to be greater. Although the number of observations is incomparably higher, it is also the case that interesting variables are often either more crudely measured (or less manufactured) and much more affected by errors or else are simply not available. In this section, we intend to look primarily at the average performances of our nine service industries, but at the same time we shall illustrate the different choices that arise from macroeconomic and microeconomic points of view in constructing the sample and computing averages. We first compare the two indicators of value added per person and operating income to sales margin for the survey of all firms, for the group of all large firms of 20 salaried employees and more, for the group of what we call large continuingfirms, and finally for the panel data sample, which we deem satisfactory for further econometric investigation. We then proceed on comparing the two kinds of averages usually considered in macro- and microanalyses-respectively, weighted (arithmetic) means and unweighted (eventually geometric) ones. The main numbers for comparisons across samples and between averages are given in tables 12.2 and 12.4; additional information and insight can be gained from tables 12A.2-12A.5. A number of explanations and observations could be made on these tables; we will only comment on the few points we want to stress.

467

Dispersion and Heterogeneity of Firm Performances

Table 12.2 gives the total number of persons by industry in 1987 for our various samples and helps to define more precisely what they are (table 12A.2 gives the corresponding number of firms). The figures given for all firms are the official numbers from the French survey (see references to the INSEE publications). They correspond to the complete population of firms in the nine service industries. There is in total some 165,000 firms, with a labor force of about 1,200,000 persons in 1987 (salaried and nonsalaried employees) and an average size of seven persons per firm. Most of the firms are small. Only about 5,300 of them (3 percent) have 20 salaried employees or more, for a total, however, of as much as 47 percent of the workers (570,000 persons). These firms, which we call large firms, are the ones for which we have had individual information (in anonymous form); they are surveyed exhaustively and have answered a detailed que~tionnaire.~ The proportion of large firms varies widely across our nine industries; in terms of number of persons it varies from a low 15 percent to 25 percent in restaurants, hotels, and legal services to a high 80 percent to 90 percent in personnel supply and building cleaning services. What we call continuing firms are the large firms that have kept answering the detailed questionnaire during the four years, 1984-87. The proportion of continuing firms among the large firms does not vary much across the industries; it is about 80 percent on average in terms of number of persons (and 55 percent in terms of number of firms). The firms accounting for the difference between the two samples in 1984, which we call leaving, have stopped reporting in 1985, 1986, or 1987, because they ceased their activities, went bankrupt, or were taken over, or because they shrunk in size, below the limit of 20 salaried employees. Conversely, the firms accounting for the difference between the two samples in 1987, which we call entering, began answering the detailed questionnaire in 1985, 1986, or 1987, because they went in business with already 20 or more salaried employees from the start, or because they increased their size over this limit.10Although in principle it should be pos9. The figures we give for the large firms (of 20 or more salaried employees) are those we have computed on the basis of the data to which we have had access. They differ to some extent from the corresponding figures that have been published. These are corrected in various ways to reintroduce firms that are still existing but that for some reasons have been allowed to not report or to send back incomplete questionnaires. For example, the published numbers are about 6.5 percent higher than ours in 1987 for the total number of persons and total value added (value added per person being thus equal to the first decimal). 10. Various miscellaneous reasons, such as failing to report, or being allowed not to report, can also explain why firms have been leaving or entering during the study period. However, one would think, considering the quality of survey, that these reasons affect only a few firms. In this respect, we have eliminated altogether from the large-firms sample a number of intermittent firms leaving and then reentering (these firms amount to about 3 percent of the total number of persons in 1984 or 1987). Similarly, we have not considered the firms that are present only in the intermediate years, 1985 and 1986. We have also discarded the few firms answering the detailed questionnaire, even though they had fewer than 20 salaried employees in 1984. We thought preferable, however, to keep the few firms that had 20 or more salaried employees in 1984 and that reported fewer than 20 salaried employees in the following years but that continued answering the detailed questionnaire sent to them.

468

Elizabeth Kremp and Jacques Mairesse

Table 12.2

Total Number of Persons in the Survey and Sample in 1987 No. of Persons (in thousands)

Corresponding Proportions (W)

Service Industry (4-digit NAP)

All Large Continuing Large/ Continuing/ Sample/ Firms Firms Firms Sample All Large Continuing

Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (77 13) Building cleaning (8708)

258.1 161.0 108.5 98.5 41.4 106.9 95.3 171.2 180.6

Total

40.6 38.9 59.3 44.7 25.6 16.5 35.2 159.1 149.3

28.4 26.9 45.7 25.9 21.1 12.6 26.4 142.8 114.8

19.1 23.3 32.7 19.2 13.9 8.4 19.5 123.6 97.2

15.7 24.1 54.7 45.4 61.8 15.5 36.9 92.9 82.7

70.0 69.1 77.1 57.9 82.4 75.9 75.0 89.8 72.2

67.2 86.6 71.6 13.7 65.9 66.7 73.9 86.6 84.7

1221.7 569.4

444.5

356.6

46.6

78.1

80.2

sible from the questionnaire (or from another source to which we had access), to distinguish between the two main reasons why firms have been leaving or entering, the information was missing, and we could not do it. Microdata sets are not in general immediately fit for econometric analyses; first, they have to be thoroughly cleaned from observations that can be seen as erroneous or that clearly appear as outliers. If this is not done, such observations, even if few, can influence the estimates (and statistical tests) to a very large extent (and wrongly so, significant correlations possibly arising from them only, or being masked by them). Thus in order to get a satisfactorily balanced panel sample, we had to clean the continuing-firms (balanced) data set. We did so in three steps: (1) we cleaned out firms with incoherent information or missing values for our main variables; (2) we eliminated firms with extreme outliers in the distributions of a few important ratios, either in 1984 or in 1987; and (3) we dropped out firms exhibiting huge rates of increase or decrease, over the three years, 1984-87, for some of the main variables.I1 The sample that we finally obtained (and to which we simply refer as the sample) amounts to about 80 percent of the continuing firms, both in terms of number of persons and number of firms, this percentage differing little by industry. Table 12.3 gives the average level and average growth rate (or average absolute change) of the value added per person and operating income to sales ratios, both across industries and data sets; table 12A.3 gives the average number of persons per firm and the average growth rate of number of per1 I . To be more precise, about 50 percent of the firms that have been cleaned out have been so because of missing or incoherent figures, and the remaining 50 percent have been eliminated, in roughly equal proportions, because of extreme values of important ratios in levels or to extreme rates of growth of major variables. It can be noted that about half of the firms are dropped out for two reasons or more.

469

Dispersion and Heterogeneity of Firm Performances

Table 12.3

Productivity and Profitability in the Survey and Sample Value Added per Person (in thousands)

Service Industry (4-digit NAP)

All Firms

Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708)

116.2 154.7 245.3 267.5

Total

Large Firms

Continuing Firms Sample

A. Average Levels in 1987 171.1 179.8 167.9 208.6 224.2 231.6 295.2 297.7 297.9 360.7 375.8 350.6

Operating Income to Sales Ratio (%) Large Continuing Firms Firms Sample

13.7 23.2 6.3 17.3

14.2 23.7 6.1 17.0

14.5 24.3 9.5 14.4

298.8

335.0

326.3

314.4

25.0

23.6

23.2

242.1 233.7 136.3 78.9

329.1 260.4 136.7 75.9

324.1 258.1 135.9 75.3

306.2 256.5 136.0 73.3

28.0 15.7 8.3 9.9

29.4 14.9 8.5 9.9

30.9 15.0 8.5 9.8

156.3

184.1

180.1

171.4

14.1

13.7

14.2

B . Average Rates of Growth (198447)

Restaurants (6701) Hotels (670R) Engineering (7701) 'omputer programming (7703) 2omputer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (77 13) 3uilding cleaning (8708) Total

-0.1 -1.2 -6.7 5.5

7.0 -4.6 1.4 -0.5

4.8 2.4 -1.8

1 .o

8.6 -2.9 4.1 5.6

1.9 1.0 0.1 0.5

2.2 1.6 -0.6 - 1.6

2.9 1.9 1.9 - 1.0

9.3

14.3

12.5

14.3

4.0

1.8

2.3

28.4 16.2 -2.4 2.2

37.8 10.5 -3.5 -1.6

32.2 9.0 -6.3 -0.5

29.3 11.2 -6.4 1.8

6.3 1.3 -0.1 -0.5

8.5 -0.1 -0.3

7.5 1.6 -0.1 0.3

5.1

0.7

0.3

1.4

1.4

0.9

1.4

1 .o

sons.I2Both tables show a rather clear pattern. As could be expected, because the three data sets overlap greatly, the numbers for the large firms, the continuing firms and the sample are usually close; discrepancies show up more often in growth rates than in levels and are much larger for the growth rate of employment than for the growth rate of productivity or the change in profitability. However, the numbers are much further apart in the case of all firms, with the exception of personnel supply and (to a lesser extent) of building cleaning, where large firms outweigh the smaller ones. In the seven other industries, value added per person tends to be significantly lower for firms with fewer 12. The operating income to sales ratio numbers are not available for the population of all firms, because firms with fewer than 20 salaried employees are asked only to answer a simplified questionnaire in which they do not have to report their profits and loss accounts.

470

Elizabeth Krernp and Jacques Mairesse

than 20 salaried employees. There is no such systematic difference in terms of the corresponding change in productivity and profitability or in employment. If we consider the three data sets consisting of large firms, the hierarchy of industries is quite well marked. The average size of these firms varies a great deal across industries; it is strikingly high in personnel supply, but it is also quite large in building cleaning and computer programming. Computer programming, computer processing, engineering, and legal services have the highest average levels of value added per person (300,000 francs per person in 1987 or more); personnel supply and building cleaning services have the lowest ones (respectively, about 135,000 and 75,000 francs per person). Computer programming and legal services are also at the top in terms of (gross) operating income margins (25 percent and 30 percent), together with hotels (25 percent). Personnel supply and building cleaning, joined by engineering, stand again at the bottom (with a margin of about 8 percent to 10 percent). Legal services have experienced by far the largest growth in labor productivity-about 30 percent from 1984 to 1987-as well as the biggest increase in profit shares, nearly 8 percent. They are followed by computer processing and accounting services, both having a very fast growth in productivity but only a modest increase in profit shares. These two industries have known also a relatively rapid growth of employment; legal services have been about the slowest. Personnel supply stands as the opposite case of legal services-it exhibits a huge increase in employment (about 70 percent over 1984-87) and has at the same time the worst productivity growth record. Hotels are still another case, with a very mediocre performance in both employment and productivity growth. The fact that the average productivity and profitability ratios are close enough for all the large firms and the continuing ones (these two sets largely overlapping) does not preclude that these numbers differ substantially between firms leaving and firms entering (because the weight of these firms over the three-year period remains small relatively to that of the continuing firms). It is better to compare directly these two categories of firms, as in table 12A.4. Contrary to what would appear likely, however, value added per person is not clearly higher for the entering firms than for the leaving ones; nor is it the case for the operating income to sales margin. Only computer processing and legal and accounting services seem to confirm such expectation^.'^ It is interesting to note that in all our industries the entering and leaving firms are much smaller (by about three times) than the continuing firms. However, it is again rather surprising to see that the average size of these firms is about the same, whether entering or leaving. A closer look at the individual size distributions, 13. Comparing the actual distribution of the two ratios for the firms entering and leaving (and not only their averages) shows that the differences in these three industries are real and cannot be accounted by a few outliers. In fact, one can see that the profit shares are also higher, by a small but clear margin, for the entering firms than for the leaving ones, in two more industries, engineering and computer programming.

471

Dispersion and Heterogeneity of Firm Performances

by industry, of the two groups of firms shows that they are indeed quite similar. l4 Although firms entering and leaving do not contribute much to changes in productivity or profitability, because they do not differ much, they do correspond to large flows of workers coming in and out. These flows have an important part in explaining the pattern of changes in employment in our service industries. They amount on average, over the three-year period 1984-87, to as much as 20 percent to 25 percent of the total stock of persons working in the large firms; the overall increase in the number of employees in the existing firms is about 20 percent. As can be seen from table 12A.5, such decomposition of the changes in employment varies greatly across industries. For example, although the very fast growth in personnel supply services (67 percent) is mainly due to hirings in the existing firms, that of computer programming services (61 percent) is also accounted for by the creation of new jobs in entering firms, which offsets largely (by 38 percent) the losses in jobs from the leaving firms. What we refer to as macro- and microaverages are given in table 12.4 for our ratios of interest, both in levels and in growth rates; to make them more comparable, these are computed for our (cleaned and balanced) sample. The macroaverages are the usual ones we have been looking at in the previous table 12.3. They are defined in a sense as if an industry as a whole represented only one very large firm. In terms of the underlying individual ratios at the firm level, they are the (arithmetic) weighted means of these ratios.I5 From a microeconomic point of view, there are various other possibilities. One is in fact confronted with the full distribution of the variables, and one can choose different kinds of average characteristics; one may also be very much interested in dispersion or in other aspects such as concentration. Usually, the simple unweighted means are computed, because they are most easy to interpret; medians are also often considered, being more robust in the presence of outliers. Often the original variables and ratios, when positive, are first transformed into logarithms, the main reason being to make their distri14. Considering per se the group of firms that we clean out of our sample is not a priori very interesting, because most of these firms are some sort of outlier. Although we know that they do differ in specific ways from the firms kept in the sample, there is little difference between the continuing firms sample (including them) and our proper sample, in terms of average productivity and profitability. In a sense this is reassuring. It also suggests that in a similar fashion the entering and leaving firms, which somewhat surprisingly show rather close productivity and profitability performances, may differ in fact in some other dimension, such as cash flows and debt-equity ratios. 15. In this sense, for example, the macroaverage of value added per person is the ratio of the total value added for the industry divided by the corresponding total number of persons in the industry (i.e., the ratio of the sample means of value added and total number of persons). It is also equal to the (arithmetic) mean of the individual value added per person ratios of the firms in the industry, weighted by the number of persons in these firms. This weighted mean (the ratio of the means) differs in general from the unweighted one (the mean of the ratios), the difference depending on the correlation of the individual ratios and the weights.

472

Elizabeth Kremp and Jacques Mairesse

Table 12.4

Macro- and Microaverages Computed from the Sample

Service Industry (4-digit NAP)

Sales per person (in thousands of francs) Macro

Micro

Value Added per Person (in thousands of francs) Macro

Micro

Value Added to Sales Ratio (%)

Operating Incom e to Sales Ratio ( 9

Macro

Micro

Macro

Micrc

A . Levels in 1987

Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708) Total

323.3 364.0 523.5 541.1

305.3 305.3 405.7 495.0

179.8 231.6 297.9 350.6

168.6 189.1 263.4 338.2

55.6 63.6 56.9 64.8

55.7 62.6 67.1 70.7

14.5 24.3 9.5 14.4

11.9 21.5 11.0 13.9

505.2

342.3

314.4

234.7

62.2

70. I

23.2

20. I

423.7 311.7 144.8

385.5 281.7 161.O

306. I 265.5 136.0

290.3 237.1 148.1

72.2 82.2 93.9

76.0 84.4 92.1

30.9 15.0 8.5

32.5 15.4 7.7

84.7

89.5

73.3

76.8

88.0

86.0

9.8

10.3

237.5

248.8

171.4

181.4

72.0

74.7

14.2

15.3

2.6 2.0 -0.3 -2.6

2.9 1.9 1.9 1.0

1.2 2.0 I .6 -1.3

-0.4

2.3

0.3

B. Rates of Growth (198447) Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708) Total

2.3 -5.0 - 1.6 8.6

0.2 -3.8 5.5 12.1

8.6 -2.9 4. I 5.6

4.9 -0.4 5.9 7.2

3.2 1.4 3.1 -1.9

13.6

9.3

14.3

8.9

0.3

28.2 11.4 -8.3

23.6 10.9 0.9

29.1 11.2 -6.4

24.0 10.5 1.6

0.5 -0.1 1.9

0.3 -0.3 0.7

7.5 1.6 -0.1

7.6 1.6 - 0.3

3.5

5.1

1.8

4.4

- 1.4

- 0.5

0.3

0.5

6.5

1.4

7.2

3.1

0.3

1.4

1.6

-3.1

-

bution more normal.'h What is then computed, instead of the more standard arithmetic means, are the geometric means, which can be expected to be rather close to the medians (if the distributions in logarithms fit well to the normal curve and are thus approximately symmetrical). This is what we do 16. Another advantage of taking logarithms is that dealing with ratios becomes more simple, the log of a ratio being the difference of the logs. Thus the mean of the log of a ratio is just the differenceof the means of the logs.

473

Dispersion and Heterogeneity of Firm Performances

here for the two productivity ratios, and the so-labeled microaverages in table 12.4 are precisely their geometric (unweighted) means.” Therefore, the usual departures of the microaverages from the macroaverages are twofold. The first departure (which concerns only our two productivity measures) is that between geometric and arithmetic means, and the difference between the two is related to the dispersion of the individual ratios. The second distinction (which concerns our four ratios) arises from the fact that the microaverages are unweighted contrary to the macro ones. The differences between the two reflect the magnitudes of correlations (or covariances) between the firm individual ratios and the corresponding values of the denominator variable. l y With these distinctions in mind, various observations can be made in comparing the macro- and micronumbers from table 12.4. A first look shows that what we have just said about the ranking of the industries according to their performances, on the basis of the aggregate data (i,e., the macroaverages), is still valid if we consider the microaverages. The industries performing best and those performing worst remain the same with respect both to productivity and profitability and both in terms of levels and rates of growth. However, if we go into more detail, the comparability in levels appears much more satisfactory than that in rates of growth. The rankings of industries according to the macro- and microaverage levels of value added per person and of operating income margin are (almost) the same, with very few inversions and only between adjacent industries. The rankings of the corresponding average rates of growth are not so close, with a number of inversions among more or less distant industries. Although our qualitative conclusions on the relative performances of the industries appear to be similar, particularly so in levels and much less so in rates of growth, the magnitudes of the macro- and microaverages can be widely different. Taking first the case of levels, the two kinds of averages remain rather close for the value added and operating income to sales margins and reflect the absence of a systematic (and large enough) correlation across firms between these ratios and size. They can be, on the other hand, much further apart for the sales and value-added per person productivity ratios. These differences are accounted for both by the dispersion of the individual productivity ratios and their correlation with size.2oDispersion explains why 17. We verified that these geometric means differ very little in fact from the medians, showing that the log transformations achieve symmetry well enough and also that the sample has been cleaned successfully of the most offensive outliers. Note that, because the profitability margins that we consider are proportions varying between 0 percent and 100 percent, it is not appropriate to transform them into logarithms. 18. As a first approximation the arithmetic means is larger than the geometric one by a factor equal to exp (uY2),if u is the standard deviation of the logarithm of the variable (or ratio) considered. This is the exact formula if the distribution of the variable (or ratio) is exactly log normal. 19. The formulas are straightforward for the average levels (such as value added per person as indicated in n. 15); but they are more complicated for the average growth rates. 20. The fact that the distribution of the individual ratios is not exactly log normal is a third source of difference between their (geometric unweighted) microaverages and their (arithmetic weighted) macroaverages in levels. However, this source proved to be negligible in our case.

474

Elizabeth Kremp and Jacques Mairesse

the (geometric) microaverages should be lower than the (arithmetic) macroaverages by about 5 percent to 20 percent, depending on the industry. The correlation explains the remaining gap, going in the same direction if positive and in the opposite one if negative. Thus, one can gather from the two sets of averages that the correlation between productivity levels and size (in numbers of persons) is positive (and strong) in computer processing and that it is negative in personnel supply and building cleaning services.*' In the case of rates of growth, the discrepancies between the two types of averages can be more substantial, particularly for the two productivity indicators. They are not, however, accounted for as simply as they are in levels. The differences between the productivity average growth rates can be seen as arising from the dispersion of the individual rates (as previously), from the correlation of these rates and the corresponding levels of productivity in the beginning year (1984), and from the change in the correlations of these individual levels of productivity with size (number of persons) between the last and first year of the period (1987 and 1984).**Thus, the impressive difference for the complete sample (i.e., the nine industries) between the microaverage rate of growth of value added per person and the corresponding macroaverage rate of growth-7.2 percent as against only 1.4 percent-can be decomposed in the following way: + 3.1 percent coming from the dispersion of the individual growth rates; - 2.0 percent coming from their correlation with the corresponding productivity levels; - 6.9 percent resulting from the change in correlation over the three-year period between these productivity levels and size.

12.3 Dispersion and Heterogeneity of Productivity and Profitability Levels and Changes Looking at average characteristics by industry and at the differences between them can be very misleading if one forgets about the extreme variability of these characteristics at the firm level. The economic performance of one industry may be much better than that of another one, and yet the distribution of a particular outcome measure usually overlaps in the two industries, with a large proportion of firms being lower in the first and higher in the second. In this section, we focus on such within-industry variability for the four outcome variables of productivity and profitability. We investigate to what extent it is accounted for by the more detailed four-digit NAP classification (in nine service industries), and by other attributes that are usually viewed as 21. The fact that these two industries account for about 60 percent of the total number of persons in our nine industries implies that the macroaverage levels of our two productivity indicators are smaller than the microaverages. 22. The differences in the changes of the profitability averages arise only from the last of these three sources, i.e., the change in the correlations (or more precisely the covariances) of the individual ratios with size (in terms of sales) in the first and last years (of the study period).

475

Dispersion and Heterogeneity of Firm Performances

contributing to the firm heterogeneity. These are three indicators of specialization (within five-digit subindustries), location (Paris region vs. the provinces), and form of ownership (corporate firms vs. noncorporate firms). Tables 12.5 and 12.6 summarize the results of analyses of variance relating these outcome variables to the above-mentioned attributes. Usual presentations of such results tend to stress the statistical significance of the various effects and report corresponding F-statistics. In a microdata analysis such as ours, given the large number of observations, statistical tests do not convey much information. All the main effects (and most of the interactions between them), even when they are quite small, appear to be statistically ~ignificant.~~ What matters is whether these effects actually reduce the (unexplained) dispersion of the variables of interest substantially and whether the magnitude (and sign) of the effects themselves appear to be economically meaningful. This is what is to be looked for in tables 12.5 and 12.6. Table 12.5 is set up in terms of the standard deviations of the four productivity and profitability ratios. It gives first the overall dispersion (i.e,, across industries, using up 1 degree of freedom only), then the within-industry dispersion (using up 9 degrees of freedom), and last, the dispersion within the much finer categories constructed from the cross classification of the three indicators of specialization, location, and form of ownership (using up 71 degrees of freedom).24These standard deviations are shown in the crosssectional and time dimensions of the data (1984 and 1987 levels and threeyear growth rates).25 In order to facilitate the interpretation, we have also adjusted them in terms of permanent or transitory dispersion, and we have computed the corresponding correlations between the 1984 and 1987 levels.26 The main message of table 12.5 is the extreme dispersion of firm individual productivity and profitability ratios and rates of growth, even when account is taken of systematic differences between industries and other major sources of heterogeneity. The magnitudes of the standard deviations speak for themselves. If one is ready to make the more or less crude assumption that these ratios are distributed normally, then about one-third of the firms are outside the plus or minus one standard deviation range around the mean, and these ranges can be very wide indeed.27For example, for one-third of the firms, value added per person differs by a factor of more than three across industries (2 u about l . l ) , and (by more than two, on average, within industries (2 u about 0.65). Similarly, for one-third of firms, the three-year growth rate in value added per person (or in sales per person) differs by more than 45 percent 23. At the conventional significance level of 5, or 1 percent. 24. Taking into account that the indicators are not fully interacted in order to avoid empty cells. 25. That is precisely the three-year differences of logarithms for the two productivity variables and three-year absolute changes for the two profitability ratios. 26. As an additional help to the reader, the traditional R2 coefficients of determination that parallel these standard-deviationnumbers are given in table 12A.6 in the appendix. 27. This assumption is particularly crude for the two profitability ratios but provides an acceptable approximation for the logarithms of the two productivity ratios.

476

Elizabeth Kremp and Jacques Mairesse

Table 12.5

Estimates of Dispersion: Standard Deviations Overall, within Industries and within Categories According to Specialization, Location, and Form of Ownership

Dispersion

Logarithm of Sales per Person

Logarithm of Value Added per Person

Value Added to Sales Ratio

Operating Income to Sales Ratio

0.16 0.15 0.07 0.14 0.05 0.90

0.10 0.11 0.08 0.09 0.06 0.70

0.10 0.10 0.07 0.08 0.05 0.74

0.09 0.09 0.08 0.06 0.06 0.58

0.09 0.09 0.07 0.08 0.05 0.73

0.08 0.08 0.08 0.06 0.05 0.57

Overall Dispersion 1984 1987 198711984 Permanent* Transitory** Correlation (1984, 1987)

0.63 0.65 0.23 0.62 0.17 0.93

0.54 0.56 0.24 0.53 0.17 0.91

Within Industry Dispersion* 1984 1987 198711984 Permanent* Transitory** Correlation (1984, 1987)

0.35 0.35 0.22 4.32 0.16 0.80

0.32 0.32 0.23 0.28 0. I6 0.75

Within Category Dispersionb 1984 1987 1987/1984 Permanent* Transitory** Correlation (1984, 1987)

0.32 0.32 0.22 0.28 0.16 0.77

0.29 0.30 0.23 0.25 0.16 0.70

'9 industry parameters. b71industry- and firm-type parameters. *Permanent dispersion: u,; u2= (u& + I$- u&J/2. **Transitory dispersion: ue; = (u&)/2.

across and within industry, and the operating income to sales ratio differs by more than 20 percent, either in levels for 1984 and 1987 or in the variation between these two years. To be more specific (and also more precise by considering the actual distribution of the variables by industry), it is instructive to compare legal services and personnel supply services and look at figures for these two industries. Legal services (7708) have the highest average operating income to sales margin; personnel supply services (7713) have the lowest average one. Although the operating income margin is on average four times higher in the first industry than in the second one-0.32 as against 0.08 (see fig. 12.1)-the lower tail of the distribution in the first recovers (nearly) completely the distribution in the second. Legal services and personnel supply services are also the two

477

Dispersion and Heterogeneity of Firm Performances Estimates of Main Effects in 1987

Table 12.6

Firms in First Category (%)

Service Industry (4-digit NAP)

Sales per Person

Value Added per Person

Value Added to Sales

Operating Income to Sales

A. Influence of Specialization (less specialized vs. more specialized)

Overall Within subindustries: Restaurants (6701) Engineering: Buildings (7701 I ) Infrastructures (77012) Manufacturing (77013) Other (7701R) Computer processing (7704) Accounting: Proper (77092) Other (7709R) Building cleaning: Residential (8708 I ) Commercial (87082) Industrial (87084) Other (8708R)

22.3

.lo**

.09**

14.1

.07

.13**

23.2 36. I 16. I 17.6 19.2

-.08 .I1 .25 .I3 .26*

-.I4 -.02 .15 .09 .23*

18.1 27.9

.15** .16**

.15** .17**

29.4 37.4 34.8 12.1

.02 .03 -.04 .47**

-.03 .04 -.06 .29**

-.01 .03 -.03 -.07 -.06 -.03 -.03

.01

.04** - .06 - .05 - .02 - .02

.02

.OO

.OO

.01

.02

-.03 .01 -.02 -.12**

- .02

.oo .01

- .01

B . Influence of Location (Paris vs. provinces)

Overall Within industries: Restaurants (6701) Hotels (670R) Engineering services (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708)

47.3 60.9 34.5 52.7 74.0 41.0 48.1 24.0 61.5 47.1

.17**

.19**

.01

- .01*

.16**

.20** .15* .27** .I5 .22* .26** .25** .20** .05

.02

.oo .oo

.lo**

.27** .04 .25* .44** .28** .19** .03

.04* .OO .06

-.02

.01 .00

.02

-.11**

- .06*

-.02*

- .02

.01 .01

.Ol*

- .02**

C . Influence of Form of Ownership (corporate vs. noncorporate) Overall Within industries: Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708) *Significant at the 5 percent level **Significant at the 1 percent level.

67.9

.]I**

.09**

- .01*

.oo

68.6 74.4 70.8 81.6 62.8 90.3 84.2 56.6 37. I

.09** .lo** .20** .03 .43** .03 .lo** .02

.08* .15** .06 .08 .34** - .03 .12** .02

.00

.03** .02 - .02 .03

.oo

.01

.03* - .08** .01

- .07** - .04 .01 - .01 .01

.04*

.07** .03** - .01

- .01

478

'

Elizabeth Kremp and Jacques Mairesse

Yo 6o

50

40

30

20

10

0

Fig. 12.1 Distributions of the 1987 levels of firms operating income to sales ratios for legal services and personnel supply services

industries with both the largest and (almost) the smallest changes in the operating income margin: +7.5 percent and 0 percent, respectively. In this case the lower half of the distribution in the first industry overlaps with the complete distribution in the second one (see fig. 12.2). Average value added per person in legal services is twice that in personnel supply services (260,000 francs per person as against 130,000) and the lower half of the distribution in the first industry overlaps approximately with the upper half of the distribution in the second industry (see fig. 12.3). These two industries have also both the strongest and (almost) the slowest three-year productivity increase: 24 percent and 1.6 percent, respectively, but the corresponding distributions at the firm level overlap fully, except for the lower tail in personnel supply (see fig. 12.4). Besides providing overwhelming evidence of huge dispersion, table 12.5 suggests two additional observations. The first is the predominance of industry effects in explaining the heterogeneity of productivity and profitability ra-

479

Dispersion and Heterogeneity of Firm Performances

6o

I

50

40

30

20

10

0

-.25 -.17

-09

-.01

.07

.15

.23

.31

Fig. 12.2 Distributions of the 1984-87 changes in firms operating income to sales ratios for legal services and personnel supply services

tios across firms. Comparing the overall and within-industry and withincategory standard deviations for 1984 and 1987 shows clearly that the division of the data into nine service industries, at the four-digit level of the NAP industrial classification, contributes much more to the reduction of dispersion among firms than the breakdown into finer categories by specialization, location, and form of ownership. Although such a conclusion could, in principle, depend on the order in which the various effects are considered, this is far from true here. For example, the R2s for the 1987 level of value added per person and operating income to sales ratio are about .65 and .40, respectively, if we take into account industry effects alone. They increase to about .75 and .45,when specialization, location, and the form of ownership are introduced as additional effects (see table 12A.6). But if we looked at these three effects alone, then the R2s would only amount to .15 and .05, respectively. In additional analyses of variance, not reported here, we have used also different breakdowns by size groups, in particular, interacting the form of ownership with the distinction between smaller and larger firms (with fewer and more

480

Elizabeth Kremp and Jacques Mairesse

35

n

30

25

20

15

10

5

3.6

4.0

4.4

4.8

5.2

5.6

6.0

6.4

Fig. 12.3 Distributions of the 1987 levels of firms value added per person for legal services and personnel supply services

than 40 salaried employees). Contrary to industry effects but similar to the case of the three other attributes, size characteristics account for surprisingly little of the dispersion in productivity and profitability levels.28 The second observation is related to the comparison of levels with growth rates. Although the NAP industry classification contributes importantly to re28. This statement must be, of course, qualified: it applies to firms that are already large enough, because we are only considering in our sample firms with 20 or more salaried employees. As we have noted, in the previous section, in most industries (with the two exceptions of personnel supply and building cleaning) value added per person appears lower in the firms with fewer than 20 salaried employees. In other analyses of variances, we have also experimented with the number of establishments per firm: this indicator, however, played a negligible role.

Dispersion and Heterogeneity of Firm Performances

481

25

20

15

10

W

-.90 -.72

-54

-.36

-.18

0

.18

.36

.54

.72

.90

Fig. 12.4 Distributions of the 1984-87 changes in firms value added per person for legal services and personnel supply services

ducing the variability in levels, it has only a small effect on the dispersion of the rates of growth in productivity or the changes in profitability. In other words, the contrasts between the average industry growth rates, even when they are significant (economically as well as statistically), are relatively minor compared to the wide range in the rates of growth of individual firms. If we interpret the numbers in terms of permanent and transitory components, we see that permanent dispersion has an industry component but that transitory dispersion has practically none. Comparing levels and growth rates, it is also interesting to consider the relative size of the permanent and transitory components. The productivity variables and the value added to sales margin as well appear rather stable, with a permanent dispersion much larger than the transitory dispersion, even within industry (or within category). The operating income to sales margin is more volatile, the transitory and permanent dispersions being nearly of the same size within industry (and within category). Although the three indicators of specialization, location, and form of ownership play a modest role on the whole in accounting for the heterogeneity of the levels of productivity and profitability, it is instructive to examine the magnitude of their estimated effects. These are shown in table 12.6 for 1987 lev-

482

Elizabeth Kremp and Jacques Mairesse

el^.^^ In each panel, the overall line provides what can be viewed as our average estimates, corresponding in fact to the intermediate specification in which the three effects are not interacted with the industry effects.30The first column gives the percentage of firms, which are, respectively, less specialized, located in the Paris region, and corporate owned. The indicator of specialization characterizes the firms whose activity appears highly concentrated in contrast to firms that are more diversified. Whenever it is possible, this distinction is made at the most detailed level of the NAP industrial classification used in the survey. As can be seen in panel A of table 12.6, this indicator of specialization can be defined in only five out of the nine service industries (for restaurants and computer processing, and for two subindustries in accounting services, four in building cleaning services, and four in engineering services). 3 1 The particular (and somewhat arbitrary) criterion we have adopted here is that of a share of value added above 75 percent in the main detailed activity for the more specialized firms (and below that for the less specialized ones). Surprisingly enough, a large majority of firms in the various industries or subindustries are highly specialized, over three-quarters of them being classified in the more specialized group with our a priori fairly stringent definition. No definite pattern seems to emerge in the differences between the more or less specialized firms. Although in many cases diversification goes along with an increase in sales and value added per person (of about 10 percent on average), its influence is usually insignificant, and at best a minor one, on the value-added and operating income to sales ratios. The location indicator distinguishes firms in the Paris region (Paris intra muros and he de France) and in the rest of France. That almost half of the large (more than 20 employees) service firms are located in the Paris region provides further evidence of centralization in France. The pattern of differences between the Parisian and provincial firms, although somewhat analogous, is more clear-cut than that arising from the degree of specialization. The influence on profitability ratios is rather small, except perhaps in legal services, which are significantly less profitable in the Paris region. On the other hand, the effect on the two productivity variables is quite strong and significant: for at least seven of the nine service industries, sales and value added per person are about 20 percent higher on average in the Paris region than in the provinces. It may be the case (e.g., in legal services) that competition is more intense in the Paris region and hence that firms have to be more productive and tend to be less profitable. However, more likely, the observed differences reflect largely price differentials rather than true productivity differences. 29. The estimates are only shown for 1987; they are practically the same for 1984 and most of them are negligible (and insignificant) for the 1984-87 growth rates. 30. And thus using up 9 + 3 = 12 degrees of freedom instead of 71. 31. The four others have only more specialized firms.

483

Dispersion and Heterogeneity of Firm Performances

Wages are notoriously higher in Paris and in Ile de France than in the rest of the country (because of higher costs of living and a more competitive labor market). The third indicator is based on the legal status of the firm and contrasts corporate firms to proprietary-owned ones. The proportion of firms belonging to one or the other categories varies according to the industry. In the sample as a whole, a third of the firms are noncorporate even though they have more than 20 employees. Unfortunately the distinction in the legal status of a firm does not correspond to the distinction that is a priori more relevant, of managerial and nonmanagerial ownership, because managers may also control the stock majority in corporate companies. The two should be at least positively correlated, and one might thus expect noncorporate firms to be more productive and profitable than corporate ones in a given industry or on average (controlling for industry). What we see in fact is rather the opposite picture: sales and value added per person are significantly higher in most industries for the corporate firms. This fact may correspond to the higher prices that corporate firms charge for their services on average (and to the higher wages that they pay), as much as it means a higher real productivity. The evidence is mixed for the two profitability ratios; in particular the operating income to sales ratio is higher for corporate firms in computer processing and legal services and for noncorporate firms in restaurants and in accounting services.

12.4 Concluding Remarks and Summary As stated in the introduction, this paper has tried to do three things: to present the French annual survey of market services; to illustrate some of the problems arising from the different points of view of macro- and microeconomists when assessing industry average performances; to exemplify the extreme variability of such performances at the firm level and to attempt to decompose it in terms of heterogeneity components and intrinsic dispersion. Along the way, we have touched on a number of issues that would be worth investigating further and deeper. We shall end by remarking briefly on three of these issues and by summarizing what has actually been done. Entry and exit of firms are particularly important in the services sector, as can be seen from the fact that the renewal of large firms in our nine industries is about as high as 15 percent per year (in terms of number of firms). Our somewhat puzzling (and inconclusive) findings on the differences of productivity and profitability performances between entering, leaving, and continuing firms should be reconsidered in a more focused analysis. To do such a task properly, however, one will have to be able to consider also the smaller firms (with fewer than 20 salaried employees), for which only a representative sample is surveyed. It would be particularly valuable for that purpose if firms were asked a question about their age (or date of creation) and one about their

484

Elizabeth Kremp and Jacques Mairesse

past employment record (e.g., the number of salaried employees at the end of the year, for the last three years), or if such information could be recovered satisfactorily from other sources. The discrepancies between what we have called macro- and microaverages of our indicators of firms performances are a reflection of the underlying distributions of the variables of interest and their interrelations. In fact, such discrepancies raise interesting questions about the relations between size and levels of productivity, size and growth rates of productivity, levels and growth rates of productivity, and so forth. To go about these questions through the comparison of average overall index numbers seems, however, rather awkward; it is better to study them per se either by relying on a (more straightforward) descriptive framework, or by embedding them in an explanatory model. What we have done in order to account for the variability of our productivity and profitability measures across firms is only a first step. One would like to assess the significance and magnitude of a number of explanatory factors, by specifying and estimating production functions and price cost marginstype equations. Such studies at the microlevel are still rare in service industries, and we intend to follow this route in future work. However, it is clear from the outset that not having information on individual price differentials and quality attributes of the services provided by the firms will be a major shortcoming for an in-depth productivity or profitability analysis. More generally, standard accounting data such as the ones collected by the French annual survey of market services are most valuable and even indispensable; they have, nevertheless, important limits. In order to carry out specific investigations, economists will have to rely more and more on additional sources of information and specially designed surveys for given industries. In the present study, we have taken advantage of the wealth of information provided by the French annual survey of market services, to construct a panel sample of data on about 2,300 large firms, from 1984 to 1987, in nine selected service industries (at the four-digit level of the industrial classification). We have contrasted the average performances of firms across industries, in terms of labor productivity ratios and profitability margins, both in levels and in growth rates. Going from the survey of all firms to a balanced and cleaned panel data sample of large firms, we have compared these averages indicators for more or less inclusive sample definitions and for the two kinds of averages usually considered in macro- and microanalyses. We have also indicated how major discrepancies could be related to size effects, to the different characteristics of firms entering or leaving the industry, or to the dispersion of the underlying variables and their correlations. Whatever the sample or average definitions, legal services ranks first in terms of labor productivity and profitability levels as well as rates of growth; personnel supply services ranks last (or almost). However, by contrast to legal services, which have done a little more than maintaining their level of employment, personnel supply services

485

Dispersion and Heterogeneity of Firm Performances

have known a remarkable growth (of about 70 percent in total number of persons over the three years, 1984-87). We, then, proceeded to show that the differences across industries in average productivity and profitability are usually small when compared to the range of individual differences within industries. As a striking example, the distributions of the rates of growth of firms in value added per person for legal services and personnel supply services overlap nearly completely, although these two industries have respectively the strongest and (almost) the slowest three-year productivity increase: about 24 percent and 1.6 percent. We have investigated to what extent the extreme variability in individual performances could be accounted for by other heterogeneity factors, besides the industry effects. We found that in fact the industry effects largely predominate in explaining the dispersion of the productivity ratios and profitability margins in levels and that our three other indicators of specialization (within the fourdigit-level industry), location, and form of ownership play a minor role, with location being the most significant of the three and probably reflecting price differentials. However, we found also that the dispersion in the productivity growth rates and profitability changes, contrary to levels, is only weakly related to the industry breakdown.

Appendix follows on pages 486-89.

Appendix Table 12A.1

Illustrative Statistics for the French Market Services Sector, Wo-Digit Level of the French Classification (NAP), 1987

No. of Firms Service Industry (2-digit NAP) Recycling (56) Repair services (66) Hotels, cafes, & restaurants (67) Travel agencies (74) Business services (77) Insurance (78) Real estate management (79) Automobile & equipment renting (80) Real estate agencies (81) Motion pictures & TV (86) Personal services (87)

All Firms

Large Finns

No. of Persons

Sales (in lo6 Francs)

Value Added per person (in 10’ Francs)

All Firms

Large Firms

All Firms

Large Firms

All Firms

Large Firms

Operating Income to Sales Ratio, Large Finns

m’o)

4,505 13,663 157,871 1,777 137,405 22,062 26,905 8,372

191 119 1,686 159 5,481 129 473 243

21,229 29,540 594,390 25,084 995,445 67,210 106,180 38,380

7,996 6,841 135,876 15,371 536,361 9,152 25,643 20,845

16,105 6,392 143,697 26,777 368,170 18,171 71,987 26,653

8,207 2,021 40,075 18,031 207,559 3,857 23,480 13,743

202 I17 127 222 207 I95 25 1 483

254 175 163 252 222 297 34 1 455

8.3 10.2 12.3 5.8 10.6 13.9 12.1 43.3

8,996 3,549 87,836

589 257 1,603

88,658 63,987 437,063

7 1,405 46,431 204,190

68,616 34,798 61,335

56,404 26,782 27,776

463 288 97

46 1 323

38.8 25.5 11.7

101

487

Dispersion and Heterogeneity of Firm Performances

Table 12A.2

Total Number of Firms in the Survey and Sample No. of Firms in 1987

Corresponding Proportions (%)

Service Industry (4-digit NAP)

All Firms

Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708)

61,743 28,463 15,307

797 567 658

402 297 391

312 235 277

1.3 2.0 4.3

50.4 52.4 59.4

17.6 19.1 70.8

15,351

523

171

144

3.4

32.7

84.2

3,282 20,418 12,696 742 7,232

346 413 712 451 820

231 276 416 290 497

156 216 367 205 407

10.5 2.0 5.6 60.8 11.3

66.8 66.8 58.4 64.3 60.6

67.5 78.3 88.2 70.7 81.9

165,234 5,287

2,971

2,289

3.2

56.2

77.0

Total ~

Large Continuing Large/ Continuing/ Sample/ Firms Firms Sample All Large Continuing

~~

Table 12A.3

Average Size and Growth of Employment in the Survey and Sample No. of Persons per Firm in 1987

Service Industry (4-digit NAP)

All Large Continuing All Firms Firms Firms Sample Firms

Restaurants(6701) 4.2 50.9 Hotels (670R) 5.7 68.6 Engineering(7701) 7.1 90.2 Computer programming (7703) 6.4 85.4 Computer processing (7704) 12.6 74.0 Legalservices(7708) 5.2 40.2 7.5 49.5 Accounting(7709) Personnel supply (7713) 230.8 352.8 Building cleaning (8708) 25.0 182.1 Total

Growth Rate of No. of Persons 1987/1984

7.4 107.7

Large Firms

Continuing Firms Sample

70.7 90.6 117.0

61.2 99.0 117.9

4.2 8.7 2.6

-2.5 9.7 -8.1

0.7 -5.3 -4.2

3.1 1.6 -5.8

151.2

167.7

60.6

61.3

33.0

28.3

91.3 45.6 63.4

88.8 38.6 53.2

6.3 1.8 11.9

-4.8 -7.5 21.2

5.4 2.2 9.7

8.5 2.8 10.0

492.3

602.7

64.5

66.9

76.4

74.7

230.9

238.7

20.0

16.7

1.8

2.0

149.6

155.8

16.7

22.1

18.9

20.6

488

Elizabeth Kremp and Jacques Mairesse

Table 12A.4

Service Industry (4-digit NAP) Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legalservices(7708) Accounting (7709) Personnel supply (77 13) Building cleaning (8708) Total

Table 12A.5

Compariem of Firms Leaving and Entering the Large Firms Data Set, 1984-1987 No. of Firms

No. of Persons per Firm

Value Added per Person

Operating Income to Sales Ratio

Leaving Entering Leaving Entering Leaving Entering Leaving Entering

353 215 256

395 270 267

37.9 32.8 65.8

30.7 44.4 51.0

143.6 216.7 256.6

160.4 173.5 286.6

11.3 21.4 4.6

12.6 21.8 6.8

123

352

67.3

53.5

339.9

339.9

13.4

17.8

129 203 151

115 137 296

53.2 27.6 33.3

39.2 29.2 29.9

302.0 224.9 230.3

375.4 344.5 267.0

19.3 23.8 16.8

30.8 25.0 17.9

152

161

95.0

101.8

122.5

142.9

7.4

7.2

218

323

69.4

107.0

88.2

77.7

12.2

10.0

1,800

2,316

51.5

53.9

196.0

198.4

12.1

15.1

Decomposition of the Change in Total Number of Persons for the Large Firms Data Set, 1984-1987 Decrease for Increase for Leaving Entering

Service Industry (4-digit NAP)

Resulting Variation

Variation for Continuing

Total Variation

A . Absolute Change of Total No. of Persons (in thousands) Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708) Total

13.4 7.0 16.8 8.3 6.9 5.6 5.0 14.4 15.1

12.1 12.0 13.6 18.8 4.5 4.0 8.9 16.4 34.6

1.3 4.9 - 3.2 10.6 - 2.4 - 1.6 3.8 1.9 19.4

92.6

124.9

32.2

-

0.3 2.3 61.8 2.0

1.o

- 1.1 3.4 - 5.2 17.0 - 1.4 -1.3 6.1 63.7 21.4

70.6

102.8

0.2 1.5 -2.0 6.4

~

B . Growth Rate of Total No. of Persons (%) Restaurants (6701) Hotels (670R) Engineering (7701) Computer programming (7703) Computer processing (7704) Legal services (7708) Accounting (7709) Personnel supply (7713) Building cleaning (8708) Total

32.2 19.9 25.7 29.9 25.6 31.3 17.3 15.1 11.8

29.2 33.8 20.8 68.0 16.8 22.3 30.5 17.2 27.0

-3.0 13.9 -4.9 38.1 -8.8 - 9.0 13.2 2.1 15.2

0.5 -4.2 -3.1 23.2 4.0 1.5 8.0 64.8 1.5

-2.5 9.7 -8.1 61.3 -4.8 -7.5 21.2 66.9 16.7

19.8

26.8

7.0

15.1

22. I

489

Dispersion and Heterogeneity of Firm Performances

Table 12A.6

Coefficients of Determination RZfor Industry Effects Only and for All Effects with Interaction

Coefficients of Determination

Logarithm of Sales per Person

Logarithm of Value Added per Person

Value Added to Sales Ratio

Operating Income to Sales Ratio

0.65 0.67 0.07 0.72 0.07 0.56

0.63 0.60 0.03 0.68 0.02 0.55

0.27 0.39 0.07 0.42 0.06 0.34

0.72 0.73 0.11 0.78 0.08 0.50

0.67 0.65 0.08 0.72 0.06 0.53

0.33 0.45 0.12 0.47 0.09 0.32

R2: Industty Effects

1984 1987 198711984 Permanent Transitory Squared correlation (1984, 1987)

0.68 0.70 0.10 0.74 0.09 0.64 R2:All Effects

1984 1987 198711984 Permanent* Transitory Squared correlation (1984, 1987)

0.74 0.75 0.14 0.79 0.11 0.59

~~

Note: The R2 in this table are computed from the corresponding standard deviations in table 12.2.

References Aubert, M., and P. Trogan. 1974. Les Gains de productivite dans les services traditionnels et modernes. Econornie et Statistique 55: 25-39. INSEE. 1986. EnquCte annuelle d’entreprise dans les services: Principaux resultats de 1984. Les Collections de 1’INSEE no. 521, ser. E102, July. . 1987. EnquCte annuelle d’entreprise dans les services: Principaux resultats de 1985. Les Collections de 1’INSEE no. 542, ser. E106, April. . 1988. EnquCte annuelle d’entreprise dans les services: Principaux resultats de 1986. Les Collections de 1’INSEE no. 576, ser. El 11, March. . 1989. DonnCes sectorielles sur les services en 1987. INSEE Rksultuts no. 7, systkme productif no. 2 . . 1990. Les Comptes de services en 1988. INSEE RPsultats nos. 49-5 1, March. Moyne, V., and P. Trogan. 1987. The Development of Statistics on Market Services over the Last Twenty Years: The French Experience. Contribution to the 20th General Conference of the International Association for Research on Income and Wealth (IARIW). Document interne. Paris: INSEE. Tajan, M. 1986. EnquCte annuelle d’entreprise dans les services: Guide de I’utilisateur. Document interne, Aoiit. Pans: INSEE.

This Page Intentionally Left Blank

13

Measuring Final Product Services for International Comparisons Alan Heston and Robert Summers

Services, as distinct from commodities, have been the focus of great interest in recent years. In large part this interest is the result of the apparent growing importance of services in the economic life of nations. This, in turn, could be either because the demand for services is income elastic and, as national economies have developed over time, their real incomes have risen or because of an independent secular increase in service activity relative to commodity activity in national economies, possibly for supply-side reasons. Certainly, it is commonly believed that nations’ productive activity involves services more now than in the past, as measured both by the proportion of the labor force producing services and by the proportion of total output that takes the form of services. No attempt is made here to explain different patterns of production of national outputs. Rather, this paper examines the output aspect of services in present-day economies, in both structural and secular terms. In what follows, all attention is directed at3nalproduct output, and at comparative quantities of services relative to commodities produced and consumed in a wide variety of nations. The presentation will be divided between a discussion of the problems of measurement of services in an international context and a description of the empirical results derived from a combined cross-section and time-series analysis of service-commodity output composition. The nature of the evidence in this analysis bears more on intercountry comparisons than on intertemporal ones. That is, more reliable statistical light Alan Heston is professor of economics and South Asia at the University of Pennsylvania. Robert Summers is professor of economics at the University of Pennsylvania. The authors acknowledge with thanks support for this paper provided by the National Science Foundation and the Fishman-Davidson Center for the Study of the Service Sector. Much of the original work reported on here was done in the course of carrying out benchmark studies of the United Nations International Comparison Project. Those studies were done jointly with their colleague, Irving B. Kravis. Research assistance was performed by Bettina Aten and Joon-Haeng Lee.

493

494

Alan Heston and Robert Summers

is cast here on the role of relative prices and national incomes in determining the composition of national outputs-services versus commodities-at a point in time, than on the perhaps more interesting secular changes taking place everywhere. The striking finding in this empirical work is that, contrary to the common view, the service share of total output does not go up with income. The key to this affront to the conventional wisdom is the introduction of service-commodity relative prices into the empirical investigation. It appears, but with less certainty, that the service share of output has gone up over time but not a great deal. Services here are defined in the conventional way: they are nonstorable products that are consumed simultaneously as they are produced. The extreme heterogeneity of services makes quantity comparisons relatively difficult, both between countries and across time. In principle, interspatial and intertemporal comparison problems are equivalent, but in fact time-to-time comparisons are usually thought to be easier. Although on balance this is surely true, there are counterexamples involving particular services flowing from rapidly changing technologies. The data underlying the work described here came from three benchmark studies of the United Nations International Comparison Project (ICP).' In section 13.1, immediately following, a brief review of the ICP work is given. Then section 13.2 presents a more detailed description of the methods the ICP used in quantifying services in the various countries participating in its benchmark studies. Section 13.2.1 lays out the ICP service taxonomy and methodology; 13.2.2 goes on to provide the flavor of the complexities of the comparisons; and 13.2.3 gives illustrations of some specific service comparisons. Then the cross-section and time-series analysis of the ICP data is presented in some detail in section 13.3. (In section 13.3.2 an expanded measure of services is discussed and integrated into the cross-section and time-series analysis. It is introduced in section 13.3 instead of 13.2 because it is outside the scope of the United Nation's ICP.) The paper closes in section 13.4 with a summary of results.

13.1 The United Nations International Comparison Project The ICP has conducted a number of benchmark studies since 1970 in which substantial pricing surveys were carried out in many countries. Specifically, these surveys and the analysis following them were done for the years 1970 (16 countries), 1975 (34 countries), 1980 (60 countries), and 1985 (as many as 60 countries, but the results have not yet been published for all 60). In each benchmark study a large number offinal products, both goods and services, were defined in close detail and then priced where possible in each 1. ICPbenchmark study references: 1970 (Kravis, Heston, and Summers 1978); 1975 (Kravis, Heston, and Summers 1982); 1980 (United Nations 1986).

495

Final Product Services

of the countries participating in the study. The method of definition was a written specification, often supplemented by visual illustrations. These specifications have been built up across countries over the years and often are specific to certain world areas. The criterion that countries followed in selecting an item was that it be sufficiently widely available in the country that an adequate market existed to produce a meaningful price. This often meant that items would already be collected for a national intertemporal price index like a consumer price index (CPI), and then the problem was to obtain a national average price, a nontrivial problem in countries where price relatives over time are outlet specific. Where a national specification varied from the ICP specification, special pricing was required. Special pricing was also often necessary in benchmark countries where items in the CPI were changed infrequently, perhaps every five years or more. Usually, special pricing for the ICP was piggybacked onto existing surveys using the same sample of outlets. By the nature of the problem of spatial comparisons, the sample of items priced will be different from the sample for time-to-time comparisons. In the latter case, if a particular nonseasonal specification is chosen for a country’s CPI in a particular month on a probability basis that takes account of its importance in consumption, it is generally available in the sample in the following month also. However, that specification is not necessarily included in the CPI sample of a neighboring country, even if the consumption heading of the specification is equally important in the second country. In the ICP methods to date, there is a related problem of how prices of specifications are processed at the category or basic heading level. The basic heading level is where expenditure weights are available, and these are used in further aggregation of the category parities. The individual category parities are built up from the item prices of the different specifications within the categories. Typically, each benchmark country prices only a fraction of all the individual specifications, and country expenditure weights for the specifications are usually not available. In the course of its benchmark studies, the ICP has employed two statistical procedures to get category price relatives from the country item prices: the so-called country-product-dummy method (CPD), a regression procedure involving dummy variables as indicated in the name; and the E-K-S method, a rather ad hoc procedure with somewhat unclear stochastic properties, named for the last initials of its originators. The virtue of these methods is that they can be used to find category price relatives even when item prices are not available for every specification in every country. (For details, see Kravis, Heston, and Summers 1982, 88-89.) A sobering remark should be made about the statistical quality of the category parities developed in the ICP. With only a few exceptions, the ICP parities are likely to be of lower quality than the time-to-time indexes for the same categories as estimated in the OECD countries. The ICP combined these various sets of country category price relatives with the countries’ associated national currency expenditures on the products

496

Alan Heston and Robert Summers

using an aggregation algorithm, the details of which are not repeated here, to get for each country a set of different purchasing power panties (PPPs). These PPPs cover gross domestic product (GDP) as a whole and a variety of subaggregates built up out of about 150 detailed categories of final output that exhaust GDP. Alternatively, this algorithm can be thought of as a device for repricing each of the countries’ quantities of its different products at a common set of international prices that are a weighted average of relative prices around the world.2 The present paper makes use of two kinds of country data extracted from the ICP benchmark studies: (1) estimates of quantities of final product services at various levels of aggregation (i.e., estimates of real expenditures on services, where the service quantities in all countries are valued at the same international prices); and (2) relative prices of services, expressed in the form of PPPs, at various levels of aggregation. Incidentally, an important difference between service outpur and service production must be recognized. Comparisons of service and commodity production in different countries are very sensitive to the degree of vertical integration of producing enterprises in the countries, but this is not the case with the service and commodity output comparisons of the ICP. On the production side, any service-type production is counted as commodity production if the final product of the firm engaged in it is a commodity. If, however, the same service production is performed by a free-standing firm and sold to a commodity-producing firm, the activity is classified as service-producing activity. (E.g., If General Motors did all its accounting in-house, the production of its accounting department would all be classified as commodity production because General Motors produces commodities. However, if General Motors’ accounting was done under contract by Price Waterhouse, the same accounting production would be classified as service production.) Grubel and Walker (1989) have noted that an increasing percentage of service employment is devoted to the production of intermediate product.

13.2 Services in the ICP This section gives a brief overview of the ICP’s treatment of services in its benchmark studies. A broadened definition of services (augmented services) and accompanying empirical exploration is more naturally taken up in section 13.3.1 because that work was done outside the scope of the international comparisons of the United Nations. 2. The merits and drawbacks of the algorithm, originally devised by Geary (1958). have been debated at length in the ICP literature. Suffice to say for present purposes, the algorithm makes it possible to aggregate the price parities and real expenditures of category basic headings to any selected level. In the present case, the individual categories representing services can be aggregated into total economywide services, or all of the services included in consumption, or all of the services included in broad categories of consumption like recreation or education.

497

13.2.1

Final Product Services The Place of Services in the ICP

Of the ICP’s 150-odd detailed categories of GDP, a little more than twothirds are in Consumption, a little less than a quarter in Investment, and the remaining are in Government. The ICP follows the United Nations’ System of National Accounts (SNA) convention in allocating all domestic capital formation expenditures to Investment, whether they are private or public. However, in a departure from the SNA, the ICP transfers all medical and education expenditures of government to the consumption categories covering the same activities. This is because in some countries most medical and education services are paid for by individual households as part of their consumption, and in others the equivalent payments are made out of the public purse. International comparisons of total spending on medical care and education are facilitated by ignoring the usual rule of grouping expenditures by who has paid for them. All Investment is treated as spending on commodities. On the other hand, all G-that is, the public consumption part of government spendingis treated as spending on services. (In earlier treatments of this subject [Kravis, Heston, and Summers 1982, 19831, the government-purchases-ofcommodities component of Government was treated as a commodity. Since then it has been judged that the contribution to welfare of government public consumption is more appropriately interpreted as the total stream of final services it provides. Empirically, this change does not affect the cross-section conclusions described below.) The rest of services are found in Consumption. In fact, about a third of the detailed categories of Consumption are services. Services are absent from the food summary heading (though food consumed in restaurants is a separate personal-service category) but show up in varying degrees in each of the other major consumption heading^.^ 13.2.2 Problems of Estimating PPPs for Different Kinds of Services Priced Services Service categories where prices can be found are relatively easy to deal with. Category price relatives are obtained using the CPD or E-K-S item processing methods referred to above. Some examples of the matching of specifications will now be described. Auto repairs are a category of priced services that can be treated very much like the commodity categories. The specification of an engine tune-up defines explicitly what is included: the changing of oil and spark plugs; the list of parts that, if found defective, are replaced at no extra cost; and so on. In a comparison of tune-ups across outlets within a country there undoubtedly are price differences, some of which may be associated with quality differences 3. See appendix table 2-1 of Kravis, Summers, and Heston (1982,69), for a listing of service and commodity categories used in the original analysis of the 1975 ICP service data and in the results reported in Kravis, Heston, and Summers (1983). In the present work, the classification system is the same except that government commodities here are classified as a service.

498

Alan Heston and Robert Summers

such as better testing equipment at large versus small repair garages. The problem in price collection for a national CPI is to sample appropriately the types of repair outlets so the temporal changes in repair charges at each type of outlet are properly weighted. The problem in international price comparisons is to hit on the specification of the service and the outlet type that holds constant, across countries, the quality differences in the service purchased. There is a trade-off between identity of outlet (see discussion of McDonald’s, below) and the representativeness of the service provided. For example, foods from roadside stalls have been matched in portion size but are not necessarily identical. Two spiced barbecue meat items, like sate in Malaysia and tikka in Pakistan, even if not identical, may be matched. In the auto repair case, the outlet sample may be confined to garages where repairs are the principal function, with sales dealerships excluded. In the case of restaurants, it might be thought that specifying the outlet may be more important than describing the entree. The Economist, which publishes international Big Mac price indexes, uses the McDonald’s hamburger as a basis for estimating the purchasing power parity not only for food consumed away from home but for all consumption. However, the novelty of McDonald’s in Tokyo and Moscow as compared with Peoria suggests that standardizing for chain outlets is a doubtful way to deal with the problem. Such chain pricing across countries may be too unrepresentative of the countries’ own price structures. The ICP item specifications for food consumed away from home explicitly dealt with the character of outlets. In the case of a restaurant meal, restaurants located in hotels or with entertainment are excluded; they have to be air-conditioned if the climate requires it; and they have to have table service. There are additional specifications for outdoor food stalls, for cafeterias in work places, and for outlets selling alcoholic beverages and snacks. An important difference between the treatment of priced services and commodities is that standardizing outlets is more important in the former case. For commodities, the ICP follows the principle that “a potato is a potato.” That is, the country price of a potato is taken as the average over all outlets, which in some countries might range from a village market to an urban supermarket. For priced services, then, a specification can be priced on a comparable basis in most, if not all, countries. In collecting such prices, careful account must be taken of the quality of the service, of course, but that is no different from the problem of controlling quality in pricing commodities. Services that are bought and sold, for which there are markets and prices, present special problems mainly in connection with outlet sampling. However, medical services are a major exception to the above statement. In general, for both medical and health services the ICP has moved over the years to pricing as many services as possible. One of the reasons for this is in fact the difficulty in making alternative comparisons. Before discussing priced medical services further, let us turn to nonpriced services.

499

Final Product Services

Nonpriced Services

When dealing with services that are not actually bought and sold, an alternative approach is called for. Services where prices typically are not to be found (referred to in the ICP in monumental understatement as “comparisonresistant” services) turn up principally in the consumption categories in the areas of medical care and education. (Remember, through no coincidence, these are the categories that include the transferred government spending.) In addition, the compensation of government employees is classified as a comparison-resistant service. Inevitably, in areas where output prices are missing, one must fall back from final to intermediate products, or more typically to inputs. The recommendations to the ICP from consultants and from participants at ICP conferences have been that comparisons based on indicators closer to the final product are to be preferred to those based on inputs. The number of children completing a level of instruction or the number of bed days in hospitals is to be preferred to the number of teachers or the number of hospital beds. Because there was no agreed methodology in this area, the procedures used in the various ICP benchmark studies have changed. Previously published analyses (Kravis, Heston, and Summers 1983; Summers 1985) rely on the benchmark service data of 34 countries for 1975, and these are not strictly comparable with either 1970 or 1980. In the present paper, all service quantities have been reestimated to make them as comparable as possible across all three years.4 Diferences between 1980 ICP Comparisons and Those of Earlier Years

The ICP comparisons for 1980 differed from the earlier ones in the treatment of health, education, and government expenditures. First, neither the treatment of the 1970 or the 1980 data made allowance for capital per worker in general government, as had been done for 1975 (Kravis, Heston, and Summers 1982, 142-43, 159). Here the 1970 and 1980 estimates of general government incorporate capital per worker adjustments, so they are now on the same basis as the 1975 estimates. Second, in comparing health services in 1980, price comparisons were carried out for the services of dentists and physicians outside of hospitals. For hospitals, salary comparisons were used for hospital personnel, and commodity price comparisons from other basic headings were used for other hospital 4. The treatment of hospital and other health expenditures for 1980 was an improvement over the earlier phases, but it is not possible now to get 1970 and 1975 estimates on a comparable basis. The necessary price data and expenditure classifications are not available for the earlier years. However, the quantitative importance of these conceptual differences for health are likely to be small compared with the errors that might be introduced by attempting to put all years’ estimates on the same basis. The treatment described in the text for education produced a reconciliation for 1970 and 1975 with 1980 that gave estimates that differed for education from published estimates (and therefore SNA government). The revised estimates are generally lower than the published estimates for lowincome countries.

500

Alan Heston and Robert Summers

expenditures. In the earlier benchmark studies, the classification of health expenditures was a bit different and so reduced strict comparability somewhat. Third, the education expenditure classifications in 1980 were much closer to the earlier phases than those for health. Fortunately, for 20 core countries, data were available for 1980 both on the 1980 basis (using teacher salaries for parities) and on the basis used in earlier phases, where parities were derived by dividing expenditures by quantities of teachers and pupils. Consequently, it was possible to adjust the 1970 and 1975 education estimates to put them on a basis comparable to 1980. In the areas of general government and education, these problems have their analogues in the procedures used for obtaining constant-price national accounts series. If input measures are used in time-to-time deflation, it is necessary to make some assumption about differences in productivity of employees over time. This is equivalent to the problem faced by the ICP when it uses teachers as proxies for educational output in its comparisons of education across countries. A judgment about differences in teacher productivity across countries is required. Similarly, if one draws on measures closer to output, like completed grades of school, the question of quality of education must be resolved. 13.2.3 Some Specific Service Comparisons Problems across Countries Illustrations of international comparisons of service quantities are given here for three situations: the treatment of health and education categories; the use of inputs in making government comparisons; and the use of hedonic pricing in making rent comparisons. Health and Education Standardizing the quality of labor input is not enough to assure that comparability is achieved. Rural physicians in India may see patients every two to five minutes in a typical day and produce large improvements in the health status of their clientele. On the other hand, doctors in a richer country may see many fewer patients but provide much more thorough evaluations. Despite the difficulty of controlling final output quality, the ICP has moved to pricing final comparable medical services, wherever possible, instead of comparing numbers of practitioners. Although it is possible to specify health procedures, comparisons across countries based on them is still subject to wide margins of error and possible bias. Consider an apperidectomy, which is an item that the ICP prices. Although the backup facilities of the hospital in which the operation takes place do not enter into the physician’s fee for an appendectomy, it seems probable that more of the fee in a rich country is used to pay for office facilities or enhanced training than is the case in poorer countries. One conclusion appears to emerge from examining priced services. If medical procedures are priced across countries, the derived quantity of medical services is significantly less

501

Final Product Services

than those obtained from a direct quantity comparison of physicians (Kravis, Heston, and Summers 1982, 159). Thus, moving toward the pricing of medical services is a step in the direction of better quantity comparisons. Still, a particular medical service fee may buy higher quality service as one moves to higher-income countries. This would mean that quantity estimates in low income countries are biased upward. Many price comparisons for medical services are possible, but this is not the case for general government or education. In these areas, and in medical care in 1970, the ICP worked with either input prices or input quantities. In some cases, input quantities, somehow standardized for productivity (e.g., the number of board-certified physicians or high school teachers with particular advanced degrees), provide a clue to output quantities (the number of tonsillectomies or the number of taught children). By pairing off the derived output quantities Q with corresponding category expenditures E , output prices P can be inferred from the relationship P = E / Q . What initially may appear to be an impossible task becomes merely Herculean when one realizes that relative country service prices are needed rather than absolute ones. Generally, indirect comparisons of input quantities have been preferred to direct comparisons of input quantities. For example, rather than compare numbers of teachers to obtain input quantities, salaries of teachers with different experience and training levels have been collected. These salary ratios for primary and secondary school teachers, when divided into total teacher compensation for each group, provide indirect quantity comparisons of inputs that better control for input quality. Where quantity comparisons have been relied on, an attempt has been made to allow for capital per worker differences that appear to be systematically related to level of income across countries. In the comparisons reported here, the allowance made for 1975 (Kravis, Heston, and Summers 1982, chap. 5) has also been used for 1970 and 1980. This had not been done in the original reports. In general, these adjustments have been in the direction of raising the estimates of quantities of services in richer countries compared with estimates based simply on quantities of inputs. Whether the adjustments are adequate is discussed further below. Treatment of Government Services

The ICP used 25 job specifications from the International Labor Organization to make compensation comparisons by skill level in the various countries. Wages and salaries, including all benefits paid by employers, were requested from all participating countries for as many of the specified occupations as were common in their government services. The simplest approach then was to assume that a secretary in a government office, or a truck driver transporting government publications, was equally productive in all countries. If this assumption is adopted, then appropriately weighted averages of the ratios of salaries for various jobs between countries becomes the purchasing power parities for government services. Needless to say, there has been much dis-

502

Alan Heston and Robert Summers

cussion about how to best compare government services, but for a fairly homogeneous group of countries like the European Community, the equal productivity assumption was adopted for 1975. Should it be adopted in world comparisons? The few studies that have been done on trends in the government sector for currently industrialized countries suggest some rise in productivity over time. (A full discussion appears in Kravis, Heston, and Summers 1982, chap. 5.) Furthermore, the information on capital per worker in government that the ICP was able to collect from a few participating countries, suggests the hardly surprising conclusion that a secretary in Japan has more inputs to work with than his or her counterpart in India. In the comparisons presented here, the methodology of the 1975 benchmark study was adopted, and an adjustment was made for capital per worker in general government. This allowance was based on the apparent rise in capital per worker associated with higher per capital GDP. For countries with real GDP per capita between 30 percent and 50 percent of the United States, the productivity per worker was lowered 15 percent, and for countries less than 30 percent of the United States GDP per head, the adjustment was 30 percent. This had the effect for the latter group of making the ratio of the quantity of government services 70 percent of what was obtained simply by dividing expenditure ratios by salary ratios. The methodology used in 1970 and 1980 was somewhat different from the 1975 one, so the comparisons for those years were adjusted to make them conform to the 1975 method. Anticipating a striking empirical finding in section 13.3, below, about how service shares compare across countries, it should be noted here that the direction of this adjustment operates against the flatshare finding. That is, using the 1975 method reduces the service share of the poor countries relative to the rich and so has the effect of increasing the slope of the relationship between service share and GDP per capita. Comparisons of Rental Value of Housing In the ICP countries, house rents range from 3 percent to 15 percent of total consumption expenditures, so it is a particularly important sector to measure correctly. First, what is needed is an estimate of the national average rent for each of a large number of kinds of dwelling units. The general procedure is to ask countries to provide national average rents for as many as 60 specifications or cells (the number in 1 9 7 3 , where the key variables identifying a unit are age, floor area, presence of electricity, water, central heating and airconditioning, and the existence and amount of indoor plumbing. Most countries could supply cell weights as well from recent housing censuses. The rents were obtained from special surveys, regular household surveys, or special rental questionnaires. When available, detailed rent files were collected from countries, and hedonic regressions were estimated. Monthly rent for an unfurnished unit without utilities appeared on the left-hand side, and the variables defining the cells mentioned above, plus some location variables,

503

Final Product Services

were on the right. Generally, the fit of these equations was satisfactory (see Kravis, Heston, Summers 1982, 54-59), so by allowing for location it was possible to estimate a national average rent. Rents that are controlled or subsidized present major problems. An indication of the problem could be seen in a regression covering countries with rent controls: when length of occupancy was included as an independent variable, its slope coefficient was significantly negative. However, a main purpose of the ICP is to make quantity comparisons, so, if the rent total in the national accounts include rental payments for controlled units, clearly the ICP price should be the controlled rent. One exception was made when the amount of rent subsidy was known to be very large, more than one-third of rental payments. In such a case (e.g., Hungary), an estimate was obtained of the full social cost of the rental unit, and both the rents in consumer expenditures and the rent used for the ICP comparison were increased to represent the social costs. Using 1975 data, an attempt was made to validate the ICP procedures by comparing the indirect quantity estimates they produced with independent quantity estimates derived from United Nations housing survey data on rooms per capita and housing stock amenities. The ICP indirect quantity estimate (the ratio of total housing expenditure to the ICP rental PPPs) was regressed in log-linear form against the United Nations direct rooms per capita variable after the latter was adjusted to take account of the amenities. For what it is worth, the estimated intercept and slope coefficient differed insignificantly from 0 and 1.0, respectively, and the adjusted RZ was .70. Although these results were encouraging, substantial noise remained in the regression.

13.3 The Intercountry “Demand” for Services This section presents the principal findings from a cross-section regression analysis of the 1970, 1975, and 1980 ICP service data sets. The word “demand” in the title of the section is meant to conjure up all the economist’s notions of what to think about in explaining differences in the shares of services in countries with different incomes and price structures. The quotation marks in the title are there to alert the reader that questions of demand-versussupply identification are not really resolved. (Such quotation marks will not be repeated. However, in what follows, terms like elasticity should always be interpreted with caution.) Section 13.3.1 is a digression on the main theme of this paper. It contains a discussion of the differences between services as measured in the ICP and services as recorded in a number of countries’ national accounts. Section 13.3.2 discusses the concepts of services, including a new augmented services measure, that are used in the empirical work here; and section 13.3.3 follows with new, detailed demand results based on the 1980 service data. Finally, section 13.3.4 presents an integrated analysis of the 1970, 1975, and 1980 data sets.

504

Alan Heston and Robert Summers

13.3.1 ICP Services versus National Accounts Services Before dealing in detail with services as defined within the ICP, we briefly set out a few observations on the relationship between ICP services and services as they are reported in the national accounts. For six countries, it is possible to compare the growth in real services implied by two ICP benchmark comparisons with the countries’ own national measures of service growth. (Only six of the countries that have been in more than one benchmark study between 1970 and 1980 also regularly publish price indexes for services and commodities.) Because most national measures are based on SNA concepts, as contrasted with the ICP concept where medical care and education expenditures of Government are transferred to Consumption, comparisons are confined to services covering all of GDP. The growth rate of real service expenditure for an ICP country between two benchmark years is given by the change in its volume of services at a common set of international prices. In general, the results are quite uneven across the six countrie~.~ Section 13.1, above, commented on the difference between real services as recorded in countries’ production accounts and real services as recorded in their final product accounts. Growth rates for the former concept have been compared with growth rates for the latter, where it was possible. Of course, as was remarked, real service growth measured on the production side need not be the same as the growth measured from the final expenditure side. Using the growth rates of measures of service final product implied by estimates from successive ICP benchmark studies, the growth rates of these two service measures were compared for 14 countries for 1970-75 and 1975-80, and for an additional 10 countries for one or the other of these periods. Not surprisingly, there were large differences between the growth rates, particularly for lowincome countries. No systematic patterns stood out, but this appears to be a fruitful avenue of research to pursue further. 13.3.2 The Empirical Definition of Services At the empirical level, the commodity-service distinction is easy to deal with if detailed categories are designated as either commodities or services. For ease of reference, when a category is treated uniquely as either a service 5. The two growth rates were quite close for the United States: the national growth rate for 1970-75 was 9.5 percent, and the ICP growth rate for the same years implied by the 1970 and 1975 benchmark studies was 12.2 percent; the corresponding growth rates for 1975-80 were 1 1 .O percent and 12.6 percent. The growth-rate match was not nearly as good for the five other countries. The 1970-75 comparisons that could be made for Italy, Japan, and the United Kingdom gave an average national growth rate of 19.8 percent; the corresponding ICP average was 35.5 percent. For 1975-80, when Austria and Denmark figures were also available, the same averages were 15.8 percent and 43.2 percent. In all cases, the growth rate in services implied by the ICP benchmark treatments was higher than the national growth rates, and usually substantially higher. It appears that, if any of these countries had been chosen as the ICP numeraire instead of the United States, the observed growth-rate disparities would have been smaller. This is because their national service price indexes were rising more rapidly than their GDP deflators, compared with the United States. This point will be discussed again when the regression equations are examined.

505

Final Product Services

or a commodity, we use the phrase “services narrowly defined.” One is left uneasy about this definition because the magnitude of distribution services included in the expenditure figures on apparel, food, and most other consumption items is likely to vary with the country’s income level. Some critics have noted that expenditures on, for example, yellow onions take no account of whether they were sold by a producer in a village market, by a street vendor in a city, or by a sales clerk in a specialty shop. The onion sold in the city has a great deal more transport and trade services embodied in it than does the village onion, but remember the ICP’s principle that a potato is a potato. Similarly, the service element in the sales of the street vendor is probably much more rudimentary than that of the specialty shop. It should be said, however, that there is not complete agreement on this point. Across countries, one may well get more labor time from sellers in the street than from sales persons in specialty shops, and this may even occur in the same city. Apart from the ambiguity about the direction of the effect, some difficulties remain with an empirical definition of narrowly defined services, because no account is taken of the differential time and place utilities that may be embodied in an onion sold in different outlets in different countries. Dan Usher (1968, 154), has dramatically illustrated the quantitative importance of transport and trade margins within and between countries in examining the case of rice in Thailand and in the United Kingdom. He put the cost of distributing rice from farmer to urban consumer in Thailand in 1965 at under 15 percent of the farm price; the distribution cost of the same rice from its port of entry in the United Kingdom to consumer was approximately 90 percent of the c.i.f. price. The United Kingdom transport and trade margins were 15 times the Thai ones! But here is a slippery slope. Once one begins to worry about the service components of some of the commodity detailed categories, where is the stopping point? We do not offer a conceptual answer, but at an empirical level, it appears to us that transport and trade margins are of sufficient interest to warrant at least examining whether our results are sensitive to the inclusion or exclusion of these elements from the definition of services. To clarify, under our standard service-commodity dichotomy, in the case of a strictly service category like local transport, the total expenditure is 100 percent services. No serious thought has been given to reallocating to overall commodities the possible commodity component of some of the service categories (like gasoline, in the local transport example). However, for categories of expenditures designated as commodities, an attempt has been made to remove the component of those expenditures that may be attributed to transport and trade. The term “augmented services” is used when allowance has been made for transport and trade margins. Note that modifying services to allow for transport and trade makes the definition of services from the expenditure side closer to the definition from the production side that includes all trade and transport in services.

506

Alan Heston and Robert Summers

How can this be done? One approach would be to survey outlets in each country-those selling food, apparel, appliances, housewares, and so on, and also those selling capital equipment-to learn what trade and transport resources were directly involved in the movement of their goods from producer to the consumer point of sale. Although this type of information would be fairly easy to obtain from organizations like retail trade associations in the United States or a number of western European countries, it would involve much more effort in the rest of the ICP countries. Still, this could be a fruitful line of inquiry. We report here on the use of input-output relations to estimate the trade and transport components. This is a less satisfactory approach, but one still to be regarded as a useful beginning. A significant difficulty is that the trade and transport margins in the input-output accounts refer to purchases by intermediate as well as final users. Furthermore, the overlap of expenditure and production categories is often very rough, so there is some arbitrariness in the selection of the input-output categories that match the various ICP expenditure categories. In addition, detailed input-output tables are not available for all 60 of the countries in the 1980 benchmark study. In view of the roughness of the approach, it did not seem worthwhile to use even all the individual country tables that were available. As a compromise, the technical coefficients of the input-output tables for only six countries were used to represent all of the countries in the 1970, 1975, and 1980 benchmark studies.6 Application of the input-output coefficients to expenditure data yields augmented services in national currencies, but what is needed is the value of augmented services measured in international dollars. If trade margins were 10 percent for poultry in national currencies, a plausible approach would be to increase the total services in international dollars by 10 percent of the poultry international dollar expenditures-that is, increase both the national currency and international dollar total by the same percentage. This treatment assumes that the relation of the prices of services in trade margins are the same as the relation of prices of poultry in all countries. This is certainly contrary to fact because the ratio of service prices to commodity prices rises systematically as one moves from low- to high-income countries. The method that has been adopted is to modify the trade and transport margins to reflect this. The international dollar service total of each country has been divided by the ratio of the country’s price level of priced services to its GDP price level. The price levels of priced services have been used because these estimates are less controversial than the nonpriced ones,’ and because both trade and transport are priced services. 6 . Input-output tables were used from the following countries (listed in ascending order of affluence): India, Indonesia, Korea, Israel, the United Kingdom, Japan, and the United States. Six income categories were defined on the basis of the incomes of these seven countries, and the relevant input-output coefficients of the representative countries were assigned to all countries within their income group. We believe the direction of error, if any, in this assignment of trade and transport coefficients, would be to increase the share of services of the high-income countries relative to the low-income countries. 7. See the comments above about comparison-resistant services.

507

Final Product Services

The difference between narrowly defined and augmented service shares is discussed below. Table 13.1 presents estimates of each for 60 countries in 1980. 13.3.3 The 1980 Cross Section This part of the analysis focuses on the real share of services of 60 countries in 1980, where real means that all quantities in all countries are valued at 1980 international prices. The 1980 cross section has been chosen for emphasis because it has the largest number of benchmark countries so far completed and because the commodity-service breakdown for these countries has not previously been published. For easy reference, these countries are arrayed in table 13.1 in increasing order of real GDP per capita. Two levels of aggregation are considered: the share of all service categories in all GDP and the share of consumption services in Consumption. These shares are given in real (SSGDpand SSc) and nominal (ScDpand Sc) terms, where the latter is based on valuations at national prices. The real shares for GDP and C are given in table 13.1 for both the narrow and broadened (augmented) definition of services. Column (8) gives the GDP service share in national prices measured from the production side. It is probably best compared with ScDp, given in column (9,the service share in national prices from the final product side. The production share is usually much larger than the final product share. Two of the five exceptions are centrally planned countries for which the comparisons are not really appropriate; the production-side estimates of Poland and Hungary are based on Material Product Accounts that do not match the SNA final product accounts. In the other three cases (Ethiopia, Nigeria, and Israel), the column (5) entries only slightly exceed the column (8) entries. The conclusion that should be drawn from the column (8)-column (5) comparison is that the production classification does not mesh well with final expenditures. Nor should one attempt to infer levels or trends in expenditures from levels or trends in employment or value added. (Because the production numbers in table 13.1 are based on a simple ten-sector production breakdown, it is conceivable that a more detailed analysis of services from the production side would show production and expenditure estimates to be more comparable.) Figures 13.1 and 13.2 provide visual images of the relation between service shares and output per capital in 1980. The first depicts a scatter diagram of the share of all services and GDP per capita; the second is restricted to shares of Consumption services and Consumption per capita. In both cases the per capita aggregate is real in the sense of being measured in international prices. Points corresponding to real, narrowly defined service shares (i.e., both the numerator and denominator are expressed in international prices, and the service concept does not include the transport and trade components of commodities) are clearly distinguished from points corresponding to nominal shares (where both the numerator and denominator are expressed in national prices). The regression lines best fitting the two sets of points have been drawn in. In

508

Alan Heston and Robert Summers

Table 13.1

Share of Services in Consumption and GDP, and GDP per capita: 60 ICP Countries, 1980 Share in Consumption

Country 1. Ethiopia 2. Mali 3. Tanzania 4. Malawi 5. India 6. Madagascar 7. Kenya 8. Zambia 9. Senegal 10. Pakistan 11. Zimbabwe 12. Indonesia 13. Honduras 14. Nigeria 15. Cameroon 16. Sri Lanka 17. Ivory Coast 18. Bolivia 19. El Salvador 20. Morocco 21. Philippines 22. Botswana 23. Dominican Rep. 24. Paraguay 25. Guatemala 26. Tunisia 27. Korea 28. Peru 29. Ecuador 30. Colombia 3 1. Panama 32. Costa Rica 33. Poland 34. Chile 35. Brazil 36. Portugal 37. Argentina 38. Yugoslavia 39. Hungary 40. Greece 41. Ireland 42. Uruguay

Share in GDP

~~~l ( S ~ D P ) Production, Real (SSc) Income 1980 (GDP Nominal Aug- Nominal Aug- National per capita) (Sc) Narrow mented (ScDp) Narrow mented Prices (1) (2) (3) (4) (5) (6) (7) (8)

275 381 421 485 499 672 792 791 935 938 1,119 1,140 1,159 1,189 1,215 1,291 1,430 1,524 1,576 1,578 1,617 1,641 1,861 1,991 2,083 2,330 2,394 2,594 2,604 2,635 2,794 3,237 3,403 3,515 3,568 3,581 3,717 3,743 4,038 4,233 4,686 4,805

31.3 21.7 14.5 23.9 20.3 21.0 36.0 19.1 22.4 21.1 36.6 26.7 34.5 21.0 28.0 20.8 30.5 29.3 30.8 26.4 27.7 31.5 29.4 38.4 36.9 32.8 30.4 30.0 27.1 36.5 36.2 31.0 28.6 40.8 30.9 33.3 26.5 30.8 32.5 38.5 42.7 28.2

53.7 26.3 27.4 32.0 34.6 33.4 45.7 20.7 20.5 39.8 43.3 30.1 39.6 28.0 35.0 45.8 30.3 28.6 46.3 30.1 41.4 41.6 38.0 38.3 43.6 30.7 35.0 33.5 36.7 38.8 41.1 42.9 40.1 45.7 27.2 41.4 25.8 37.7 39.4 36.2 43.6 31.6

60.5 33.7 37.9 40.7 41.2 44.2 51.8 27.7 26.3 49.4 50.3 41.8 45.1 36.7 42.0 64.8 37.9 33.5 53.3 40.4 45.5 52.8 42.9 42.3 48.8 35.3 39.5 37.3 40.3 43.4 44.0 47.4 43.9 47.6 29.8 44.7 28.7 41.3 42.4 38.7 49.3 34.5

39.0 33.2 20.4 34.3 22.7 30.1 37.9 32.8 37.6 26.3 42.6 22.7 33.8 21.3 28.2 22.9 34.3 31.1 31.6 34.6 23.8 33.7 27.1 33.8 35.5 31.6 27.3 27.2 26.9 33.6 34.4 34.1 28.6 38.7 28.2 32.9 28.0 27.3 25.5 37.7 42.3 32.2

60.4 48.0 26.5 44.3 27.9 43.0 48.2 41.9 41.7 40.2 49.0 27.9 38.6 29.3 39.9 40.7 43.3 33.6 46.8 43.0 38.3 41.5 34.5 36.9 42.7 35.4 29.8 32.0 30.8 35.8 36.8 42.2 36.8 34.9 23.8 42.0 22.4 34.1 31.6 34.7 42.5 29.4

65.7 53.2 45.4 51.8 38.3 53.1 54.4 48.2 47.1 49.9 55.4 38.3 45.9 37.2 47.1 60.0 51.5 38.8 54.6 52.7 43.3 52.8 41.2 42.0 48.7 40.9 36.2 37.1 36.0 41.2 40.7 47.8 40.8 37.1 26.9 45.7 26.0 38.1 35.3 37.6 48.6 32.8

38.1 53.2 45.0 51.9 37.5 62.8 48.5 47.9 50.6 47.4 50.0 34.3 47.6 19.6 45.5 52.5 46.0 46.8 51.5 59.8 38.9 49.7 53.2 49.2 54.0 31.6 45.1 46.7 51.6 52.4 71.1 67.4 22.3 55.6 54.9 70.5 51.6 60.6 24.4 54.7 51.9 58.9

509

Final Product Services

Table 13.1

(continued) Share in Consumption

Share in GDP

~ ~ ~ l ( ~Production, p p ) Real (SF) 1980 Income (GDP Nominal Aug- Nominal Aug- National per capita) (Sc) Narrow mented (SGDp) Narrow mented Prices Country 43. Spain U.'VenezueIa 45. Hong Kong 46. Israel 47. Japan 48. Austria 49. United Kingdom 50. Italy 5 1. Finland 52. Denmark 53. Netherlands 54. Belgium 55. France 56. Luxembourg 57. Germany 58. Norway 59. Canada 60. United States

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

5,247 5,533 6,436 6,847 7,660 7,743 7,807 7,912 8,109 8,558 8,576 8,599 8,942 9,003 9,060 9,894 11,148 11,998

35.3 32.1 41.1 54.7 53.2 48.4 50.1 39.6 42.3 46.8 43.0 45.6 43.8 43.5 45.3 40.4 54.2 52.4

31.3 40.9 31.0 53.3 52.1 48.5 49.0 42.3

37.3 43.7 34.3 57.9 54.7 51.7 51.8 46.4 47.5 50.3 39.1 44.1 43.2 44.6 41.4 49.2 46.9 46.4

33.1 26.7 29.2 60.7 38.5 40.4 47.3 36.4 36.3 46.8 39.5 42.0 39.1 39.3 41.5 34.2

29.8 33.8 25.3 59.5 33.9 41.5 48.8 38.2 33.2 46.6 33.6 37.7 35.3 36.3 35.7 36.1 36.5 40.8

35.5 37.5 29.6 63.3 37.9 45.4 51.6 42.5 37.3 50.5 37.8 42.1 39.4 40.8 39.5 40.7 40.3 44.1

51.5 60.2 68.9 60.0 53.4 47.4 63.5 53.3 51.3 66.5

44.2 46.8 35.7 40.8 39.9 41.2 38.2 45.4 44.0 43.7

44.9 48.7

64.1

57.0 56.6 62.3 48.9 58.9 64.4 65.2

figure 13.1 the positive slope of the dashed line, the regression line for SGDp, shows that the nominal service share of GDP clearly rises with per capita GDP. In sharp contrast, the slope of the solid line, the SSGDpregression line, is slightly negative. Why is the SSGDp-income relation essentially flat when the SGDp-income relation is not? Equivalently, why are the SSGDppoints in the scatter diagram nearly always above the corresponding SGDppoints for low income countries and below for high income ones? (Of the 4 4 countries with per capita GDPs less than half that of the United States, 35 real shares are above the nominals; of the 16 with per capital incomes more than half the United States, 11 real shares are below the nominals.) The explanation lies in a strong ICP finding, the systematic pattern of rising relative service prices with per capita income. (See, e.g., Kravis, Heston, and Summers 1982, 191-95.) As a consequence, the nominal share in low-income countries understates the real share, and it overstates the real share in rich countries. The regression patterns for consumption in figure 13.2 are almost but not quite the same as those in figure 13.1. The SSc slope is not negative, but it is again smaller than the slope of the Scregression. The reason, of course, is the same as in the GDP regressions: relative service prices rise with per capita income. These relations were reexamined using 1970 and 1975 data, and the same

510

0 O 67

Alan Heston and Robert Summers

1

9

05

s!

0.4

C

0.3

02

1-

0.0

0

6

10

12

Gross Domestic Product per capita (in thousonds)

-+

-Nomino1

-Real

Fig. 13.1 Service share of GDP versus GDP per capita (nominal and real shares; 1980)

+

0.291 0.012 GDP per capita (in thousands) (0.012) (0.002) Real share = 0.397 - 0.003 GDP per capita (in thousands) (0.014) (0.002) Note: Real service share is based on the narrow concept of services. Nominal share

=

conclusions were reached. The flat relationship between the real service share in GDP, narrowly defined, and GDP per capita is not just peculiar to 1980. Furthermore, the same story about real Consumption service shares rising much less with Consumption than the nominal shares also holds for 1970 and 1975. The relative prices of services play two roles here in producing these results. First, they are used in converting the nominal shares into real shares, because repricing services in international prices instead of national prices is essentially an exercise in equalizing relative service prices. Second, one would expect that the relative price of services to have a cutting edge in determining which countries had larger real service shares and which had smaller. Economists looking for explanations of variability in the real share would immediately think of the relative price of services as well as country income for their demand regressions.

Final Product Services

511 08

5 c

0.4

0.3

0.2

0

2

4

6

8

10

12

Consumption per capita (in thousands) ---CL--Nominol

-t -Real

Fig. 13.2 Service share of consumption versus Consumption per capita (nominal and real shares; 1980) Nominal share = 0.246 + 0.035 Consumption per capita (in thousands) (0.011) (0.003) Real share = 0.347 + 0.013 Consumption per capita (in thousands) (0.015) (0.004)

Note: Real service share is based on the narrow concept of services.

13.3.4 Comparing Results in the Various Benchmarks The explanation provided by price and income is examined in a set of equations where the left-hand-side variable is the log of the share of real services in Consumption or GDP and the right-hand-side variables are the log of real per capita Consumption or GDP and the log of relative prices of services in consumption or GDP. With cross sections for 1970, 1975, and 1980, the three benchmarks can be pooled if the variables are restated in a common year’s international prices. However, the pooled sample is unbalanced, and the constant-price procedure is subject to error. Therefore, the separate results of the three benchmarks, based on current international prices, will be reported first. The results appear in table 13.2. Table 13.2 shows that the major differences between 1970, 1975, and 1980 are that the price elasticities take on larger negative values over time, though

512

Alan Heston and Robert Summers

Table 13.2

Service Share, Income, and Price Relationships: Estimates Based on the 1970,1975, and 1980 ICP Benchmark Data (log-linear regression results)

Service Concept & Aggregation, by Year Narrowly Defined: Consumption: 1980 1975 1970 GDP 1980 1975 1970 Augmented: Consumption: 1980 1975 1970 GDP: 1980 1975 1970

Coefficients & Standard Errors (S.E.) Price (S.E.)

Income Variable (S.E.)

- ,623 (.124) - ,118 (.181) -.lo8 (.184)

1.179 (.028) 1.122(.050) 1.106(.037)

-.368 (.184) ,004 (.149) ,095 (. 192)

1.003 (.036) ,980 (.041) ,963 (.038)

-.704 (.119) -.373 (.170) -.253 (.143)

1.1 16 (.024) 1.068 (.039) 1.046 (.026)

- ,501 (.182) - ,109 (. 150) - .044 (. 195)

.981 (.030) ,958 (.033) ,940 (.033)

Notes: Price is the purchasing power over services in Consumption or GDP, relative to the purchasing power over all Consumption or GDP; and income is per capita real Consumption or real GDP. The variables are expressed in current international dollars of the year indicated. Augmented services include the trade and transport components of commodities as well as narrowly defined services. ICP Consumption includes public expenditures on education and health. The number of observations in each of the three cross sections is 1970: 16; 1975: 34; and 1980: 60.

the differences are probably not statistically significant. The income elasticities are not statistically different from each other for either definition of services or for GDP or consumption. There is a consistent pattern that the income elasticities rise slightly over the three benchmarks, but only the elasticities in the Consumption equations for 1980 differ from unity at the 1 percent level of significance. Note how the price and income elasticities associated with the augmented services compare with those of the narrowly defined services. Modifying the definition of services to include trade and transport margins leads to elasticity estimates that are slightly smaller for income and larger for price (accompanied by smaller standard errors). It was expected that the greater amenities of distribution in the rich countries would tilt positively the flat-share relationship found for narrowly defined services. The finding, however, is that any new tilting is down rather than up! The many problems encountered in attempting to capture the trade and transport amenities with the input-output tables makes idle any speculation on just why this happened.

513

Final Product Services

What is strikingly consistent among 1970, 1975, and 1980 is that, when the relative price of services is taken into account, there is virtually no change, or a small decline, in the share of services in GDP as GDP per capita goes up; and there is a consistent but small rise in the consumption service share as consumption per capita goes up. One is tempted to conclude from this that, contrary to Wagner’s law and much-received doctrine on the increasing role of government, the real share of government declines slightly with income but that this decline is matched by a rise in the service share of consumption. Overall this produces the relatively flat share of real services in GDP.* This question is further explored in the presentation of the pooled regression equations. Pooled Demand Regressions

The basic similarity of the equations across the three benchmarks invites a pooling of the data of the three cross sections to run single demand regressions, one for GDP and one for Consumption, which would incorporate all 110 (probably not independent) country observations. Such a regression has the efficiency gain that goes with a larger number of observations, but more importantly, it provides an opportunity to estimate any secular change in service demand over the 1970-80 period that cannot be explained by changes in prices or income. The pooled regression requires that all the share, price, and income data be expressed in the international prices of one of the years. Doing this requires the use of a national service price deflation index for one or several of the countries. In fact, most countries do not have a deflator for consumer expenditure on services, but service price deflators for Consumption and Government were obtained for Italy, Japan, the United Kingdom, and the United States. Running the pooled regression with the four different sets of national deflators gave the following results: For Consumption, the price and income elasticities (with their standard errors in parentheses) were - 0.397 (0.084) and 1.153 (0.020) regardless of the country national deflator. For GDP, the elasticities depended on the national deflator, but the price elasticities only varied between - 0.104 (0.103) and - 0.187 (0.102); the income elasticities varied narrowly as well, between 0.998 (0.023) and 0.967 (0.023). The price and income elasticity estimates were quite insensitive to which country’s national service price index was used, but the estimates of the secular change in service shares was not. Time dummies for 1970 and 1980 in the Consumption and GDP regressions show how the 1970, 1975, and 1980 intercepts compared. The coefficients of these dummy variables were estimated using each of the four country inter8. Note that two parts of government, health and education, are transferred to Consumption in the ICP definition used here. The conclusion, therefore, with respect to the real share of government is exclusive of those two categories.

514

Alan Heston and Robert Summers

temporal service price indexes. The smallest and largest coefficients among the four sets for both GDP and consumption are as follows (estimated standard errors are given in parentheses): For 1970-75, the coefficients for GDP varied between - 0.080 (0.047) and 0.056 (0.047); for consumption, between - 0.042 (0.048) and 0.233 (0.048). For 1975-80, coefficients for GDP varied between - 0.088 (0.031) and -0.025 (0.032); for consumption, between - 0.002 (0.034) and 0.002 (0.036). In every case the extreme estimate was associated with the Japanese deflator. (The estimated trend between 1970 and 1980 was negative for the consumption service share except when Japan’s service deflator was used. However, for GDP the estimated 1970-80 trend was positive using service deflators for two of the four countries, Japan and the United States.) The deflators available were all from relatively highincome countries. It is unclear whether the trend figures would display a different pattern if service deflators for some developing countries had been available. This would certainly be an interesting question to examine. Unfortunately, the potential errors of measurement in country service price deflators make estimating secular change with ICP benchmark data perilous at this time. Methods are being explored to develop time-to-time deflators between ICP benchmarks that are base country invariant, and these approaches should allow a more fruitful examination of this question.

13.4 Conclusion This paper reports on the examination of a set of data from the U.N. ICP that bears on expenditures on final-output services around the world. A variety of conceptual and measurement problems that affect comparisons of expenditures on services across countries were considered. The problems arising from specifying a common quality for a service that is to be priced in many countries of widely differing incomes were discussed. Specifying a common quality is even more difficult when the service itself is not priced in the marketplace, so output quantities must be estimated on the basis of inputs. Unpriced services, in education and general government primarily, are particularly vexing in this regard. An exploratory attempt to transfer commodity trade and transport margins to the service category was described, along with empirical results of introducing such an augmented-service concept into demand regressions. A number of reasons were also discussed why expenditures on final product services might have a different trend from services measured from the employment and production side. The study explored further an earlier finding, developed from the 1975 ICP data, that service shares did not rise with income. In cross sections examined here on data for new countries and new years, estimated income elasticities were still in the 1.1-1.2 range for Consumption, and often less than 1.O for GDP. The service flat-share conclusion, so at odds with common perception,

515

Final Product Services

was verified for other country sets and other years, and for a broader definition of services. (Elsewhere the authors have shown that very similar flat-share results are also found within countries. When national expenditure surveys for 1980 for the United States and 11 European Communities were analyzed, no significant connection between the service share and income at the household level was found (see Heston and Summers 1988). The question of whether there has been a secular rise in the share of services in national output-apart from changes induced by income or price changes-was also investigated, with the ICP cross sections providing mixed findings. This work required the use of country service price deflators, and those available did not give uniform empirical conclusions. Between 1970 and 1980 whether the secular change as registered by the ICP data was up or down depended on which country’s deflator was used. The weight of evidence certainly supports the contention that expenditure shares on services have displayed very little to moderate secular trend and there has been very little rise with respect to income. In part, this probably is because the concept of services is not an entirely tidy one. Although many expenditure categories, like entertainment, may be elastic with respect to income, the same entertainment desires can often be satisfied by either service or commodity purchases. (See Summers 1985,42-44.) A question might be raised whether our conclusions flow from the way the underlying data are used to derive service prices. To the extent the ICP underestimates the relative quality of services in high-income countries, it overstates the quantity of services in low-income countries. If such a systematic effect existed, would it modify the flat-share finding? At first glance one would think the answer is yes, but on reflection this may be less obvious for the following reason: the quality of commodities must also be held constant across countries in the ICP comparisons. To some extent this may be easier to accomplish because of branded products and fairly standard technical characteristics of some commodities. However, it would be necessary to show that systematic underestimates of quality differences across countries of different income levels were much larger for services than for commodities for such an effect to modify the flat-share finding. Were they? Maybe. Maybe not.

References Bhagwati, Jagdish. 1984. Why Are Services Cheaper in Poor Countries? Economic Journal June: 279-86. Geary, R. C . 1958. A Note on Exchange Rates and Purchasing Power between Countries. Journal of the Royal Statistical Society 121:97-99. Grubel, Herbert, and Michael A. Walker. 1989. The Canadian Service Industries. Vancouver: Fraser Institute Service Sector Project.

516

Alan Heston and Robert Summers

Heston, Alan, and Robert Summers. 1988. Consumers’ Demand for Services: Within and Across Countries. Discussion paper, Fishman-Davidson Center for the Study of Services, Univ. of Pennsylvania. Hill, T. P. 1977. On Goods and Services. Review of Income and Wealth. 23, no. 4 (December): 315-38. . 1987. The Economic Significance of the Distinction between Goods and Services. Paper presented to 20th General Conference, International Association for Research in Income and Wealth, Rocca do Papa, Italy, August 1987. Kravis, Irving B., Alan W. Heston, and Robert Summers. 1978. International Comparisons of Real Product and Purchasing Power. Baltimore: Johns Hopkins Univ. Press. . 1982. World Product and Income: International Comparisons of Real Gross Product. Baltimore: Johns Hopkins Univ. Press. . 1983. The Share of Services in Economic Growth. In Global Econometrics: Essays in Honor of Lawrence R. Klein, ed. F. Gerard Adams and Bert G. Hickman, 188-218. Cambridge, Mass.: MIT Press. Summers, Robert. 1985. Services in the International Economy. In Managing the Service Economy, ed. Robert P. Inman, 27-48. Cambridge: Cambridge Univ. Press. United Nations and Eurostat. 1986. World Comparisons of Purchasing Power and Real Product for 1980. STlESAISTATlser. Fl42 (pt. 1). Usher, Dan. 1968. The Price Mechanism and the Meaning of National Income Statistics. London: Oxford Univ. Press.

14

Measuring Public-Sector Output: The Swedish Report Richard Murray

This paper relates the large scale effort to measure productivity in the production of public-sector goods and services in Sweden. It is organized as follows: First, there is a methodological part dealing with problems of measuring output in the public sector and how they are addressed in this study. This is undertaken in the context of national accounting and deals with the problems of aggregation in choosing final output, of weighting outputs and of the treatment of quality. Next, results of the study are presented. The huge drop in productivity is analyzed, and some main conclusions as to its causes are suggested. Finally, macroeconomic implications are briefly discussed. The Swedish study was organized by a subgroup to the Expert Group on Public Finance (ESO) under the Ministry of Finance. Several of the studies were contracted out to independent researchers, agencies, and public organizations. But the conceptual framework of the study was closely controlled by the subgroup. Dr. Ingvar Ohlsson headed this group, which was set up in 1982. A summary in English was published in 1987 (Ohlsson 1987). All segments of the public sector were to be covered. Comprehensive or sample measurements were taken for roughly 70 percent of the sector, incorporating the municipal, county, and national levels (including the national social security administration). However, only the nonprofit activities were included. Public utilities like railroads, telecommunications, power and heat, and housing were omitted for they are reasonably well covered by the national accounts as of today. An effort was made to extend the measurements as far back as 1960 and to make them cover the period up to 1980. In the vein of the national accounts, measurements were undertaken with a base year for weights and price indexes. That base year was set to 1980, the end of the Richard Murray has a Ph.D. in economics and is senior economist at the Swedish Agency for Administrative Development.

517

518

Richard Murray

period. Since the first report appeared, a series of productivity measurements extending the period up to 1985 were undertaken (see Murray 1987). The framework of the study is the national accounts scheme. Its concepts, theory, principles, and practices were consequently adopted as guidelines for the study. As anyone familiar with compiling national accounts statistics knows, this includes many approximations and deviations from strict principles. Even though the United Nations has set standards for accounting, there are ambiguities stemming from a compromise of purposes for these accounts. The basic purpose has been to measure economic activity for business cycle analyses. This purpose comes very close to analyzing the laws of production and the measurement of productivity. A different purpose is to measure income, welfare, or even the quality of life. It is important to recognize that the national accounts, according to the U.N. definitions, strike a compromise between these two extremes. The stress is on production, but production should be measured in a way that is relevant to welfare. Therefore, it is output-not outcome, nor throughput-that is to be measured and with consumer evaluation as the measuring rod-even for the public nonprofit, nonmarketable goods and services.

14.1 Including the Public Sector in GNP When calculating GNP we add up value added in all sectors of the economy. The government nonprofit sector is included with the sum of its wage bill. There are two corrections we want to make to this calculation: (1) the capital stock of the government ought to have a rental value that should be added to value added from the public sector; and (2) the change in productivity-value added in relation to labor input-should be reflected in the value added by the public sector. As seen in table 14.1, GNP rose according to official estimates by 2.0 percent per year 1970-80. Adding the rental value of capital in the government sector makes the growth increase to 2.1 percent. Assuming hypothetically that the government sector had a 2 percent growth of value added per year in relation to the input of labor and capital reduces value added in 1970 by 17.071 billion Swedish kronor (SEK) in 1980 prices. Growth goes up to 2.5 percent per year. It is clear that productivity in the government sector matters. In order to include the government sector in a full account of GNP, one needs figures on value added for the government sector. Lacking this information, we may instead try to measure aggregate final output of government, calculate the rate of total factor productivity change for the sector as a whole, and from this infer the rate of change of value-added productivity.

14.2 Aggregation Poses Problems Aggregation poses several problems of measurement: (1) What outputs are to be measured and included? When aggregating one runs the risk of double

519

Measuring Public-Sector Output

Table 14.1

Correcting GNP for the Rental Value of Capital and the Change in Productivity (billions of SEK, 1980 prices)

1 . GNP at market prices, official figures 2. Of which: Government nonprofit organizations 3. Capital stock of government organizations 4. 6% interest on capital stock 5. GNP adjusted for interest on capital stock of government (1 + 4) 6. Of which: government nonprofit organizations 7. 2% productivity growth per year of government nonprofit organizations 8. GNP adjusted for capital stock and productivity increase (rows 1 + 4 + 7)

1970

1980

432.647 81.535

525.099 116.036

224.838 13.498 446.137

364.354 21.861 546.960

95.025

137.897

- 17.071

429.066

546.960

counting. Which outputs are intermediate and which are final? (2) How does one sum pears and apples? By kilos or by kronor? What weights should be used in adding outputs of different kinds? (3) How should quality be incorporated in the output measurement? These questions become especially tricky where public-sector output is concerned, but they are also relevant to other sectors of the economy. The first question is, What is to be counted as output? Outputs that directly benefit consumers and producers outside the government sector should be included. They are final in relation to the government sector. Intermediate outputs within the government sector should be excluded. There is, however, a problem of aggregating the government and private sectors. Those outputs from the government sector that we consider here are typically free of charge. Still they may benefit private producers. That means that value added in the private sector is overstated and that we would be counting this production twice if we were to include this output in the government aggregate. Some other government outputs-like environmental regulation-might not show up in the value of production in the private sector or might even reduce its reported value while benefiting citizens and consumers. A very preliminary investigation on who are the beneficiaries of the government subsidized output sheds some light on how large this problem might be. In table 14.2 government output is measured in input terms, equivalent to what is called public consumption. That part of government output that directly benefits the private sector-part of the consumption of roads, harbors, employment agencies, and so on-has diminished in share although it has grown in volume. Recalculating the growth of GNP by subtracting this part of the contribution of the public sector to GNP from that of the private sector increases the growth of GNP by 0.14 percent over the ten years. A similar calculation was undertaken by Kuznets (1971). He included a much larger part of the government output in what was to be subtracted from

520

Richard Murray

Table 14.2

The Destination of Government Output, Percentage of Public Consumption (nominal prices)

Input into production: Private industry Government Consumption: Individual goods Collective goods

1970

1975

1980

8.5 3.0

8.0 2.5

7.6 2.2

61.6 26.9

63.7 25.8

68.0 22.2

GNP as sheer costs to keep society functioning. General administration, police, military defense, and so on were looked on as a prerequisite for other kinds of economic activities. Kuznets found that over 100 years the growth of the U.S. economy was lowered by 7 percent. How does one add the output from different government branches or different outputs within the same agency? Consumer evaluation should be the guidepost, but this creates new problems because there are no market prices for these outputs. It has been suggested (Ohlsson 1987, 38) that nominal user fees could be used as weights in case “the market” is in equilibrium, that is, that there are no unsatisfied wants manifesting themselves in queues. Because over 60 percent of public consumption is individualistic in nature (see table 14.2, above), this argument carries a substantial weight. Consider the case of health care in Sweden, which is to a very small extent financed by a user fee of about 60 SEK per visit. There are queues in some areas of health care, but on the whole the demand is met by an ample supply. The logic of such weights would be a drastically reduced health care. At the price of 60 SEK per visit, Swedish hospitals cannot carry their costs. The prices would seem to indicate that health care is supplied far beyond what is optimal. Libraries, museums, schools, and such would be valued nil, also implying that a sharply reduced level of spending would come closer to the optimum. Because there is very little political and public support for cutbacks on government output, these services must be more highly valued than is reflected in the user fees. Evidently there are collective evaluations even concerning individual public goods. How to incorporate changes in quality is a well-known problem in the construction of price indexes. This becomes more problematic if there are no prices from which to infer the characteristics that command the willingness of the consumers to pay. However, techniques like conjoint analysis (Cattin and Wittink 1982; McFadden 1976) could be applied to public-sector output to reveal consumer evaluations. What counts as quality are only those characteristics of a good that appeal

521

Measuring Public-Sector Output

to the consumer. But who is the consumer? There might be a genuine conflict of interests between citizens/consumers and politicians/principals. (This conflict is also relevant in the identification of outputs and in the choosing of weights but is best treated in connection with quality.) Although politicians may rate the accuracy of the Swedish internal revenue service highly, citizens might have an opposite view. There are merit want goods that would not even qualify as outputs in the eyes of all citizens, like advertisements on the dangers of smoking. In these cases we are apt to take the view of the politicians and regard them as the final consumers.

14.3 Practical Solutions 14.3.1 Final Output It is easier to conceptualize final output than to measure it. The goods we are envisaging might well be consumed by specific individuals or organizations outside the government sector, but they are ordered and paid for by quite different bodies. There is no connection between payment and consumption, which makes identification of outputs much more complicated than of goods on a market. Payment to government organizations is for resources and activities and not for outputs. Because the financing bodies pay for office spaces, traveling, coffee breaks, staff meetings, and data machines as well as for operations, teaching, land surveys, and court proceedings, asking what the principal is paying for in order to determine what are the final outputs is not at all helpful. From that point of view it is not apparent why the latter should qualify as outputs and why the former would not. This causes a well-known lack of goal consciousness on behalf of government organizations and their staff and also makes it a cumbersome, sometimes delicate and questionable, exercise to choose the final outputs. Despite the funny way government production is financed, it was of considerable help to think about what the principal was purchasing. Would he be purchasing coffee breaks? Would he be purchasing capacity without production? Other questions that were asked: What are the ends that the services to identify should serve? What outputs serve the outside world? These questions guided the choice of output indicators in the Swedish report. Applying, for example, the question of what the principal is purchasing clarifies what are the outputs in the following instances: Is the end product of the internal revenue service the tax receipts it collects? Is the output of the social security service the benefits that are paid out? Of course not. Output is instead the handling of these payments. Agencies are doing a good job when they collect or pay the correct amount, without delay and at low administrative costs. But there are instances that are more ambiguous. From the point of view of the agency, informing citizens on their rights might look very much as an end

522

Richard Murray

product, especially because the agency is instructed to carry out this activity. However, from the point of view of the citizens, information is a prerequisite for the exercising of rights, just like information on prices and qualities in a market. So we chose not to include information in the output when seen as a supporting service. On occasion, we were forced to measure a host of outputs from one agency in order to capture it fully. However, this ran the risk of including intermediate products. Again, what are the outputs of the internal revenue service? Is it the handing out of forms to citizens, the information on how to fill in these forms correctly, the collection and the scrutinizing of the forms, and investigations into the accounting books of the firms, the processing of bank accounts, checks on employers payments of benefits to employees, and so on? In this case all these activities add up to one final product: the completion of the informational basis necessary for taxing and billing the citizens. The end product then is the complete processing of the tax form of a citizen or a company. All activities within the internal revenue service contribute to this end. But instead of measuring them all and then aggregating them in relation to their contribution, it is preferable to measure the final output directly. This is what has been done wherever possible. In the case of the internal revenue service tax forms processed are counted as final output. Agency representatives, personnel, ministerial supervisors, and politicians often see the identification of an agency’s final output as an oversimplification. It is much easier to gain the acceptance from the man in the street, who has an exceptional and under-utilized talent for disentangling the relevance. If it is possible to find a measure for the aggregate output, it simplifies matters a great deal. Government production is often hierarchical: parliament formulates the policies; ministries issue goals for production; national government agencies survey county or municipal government organizations doing the actual production-teaching, day nursing, medical care, and so on. Wherever it has been possible to identify and measure final output (like the number of children trained at school) for the whole chain of producers, that is what was done. Of course, in that way it is not possible to distinguish the contribution to the final output at each level. And the efforts of parliament, ministries, and supervising bodies are all reduced to input and resource consumption (which may not appeal to their idea of their own importance). In this way the measurement of the national government was reduced to one third of total employment. Two thirds were allocated as inputs to final outputs from the local governments. To take the most brutal example: all the administration of building control and financing and community planning was seen as input to a very crude indicator of the governmental output “physical planning”: the total volume of construction. On occasion it was not possible to find an acceptable measure of final output. For example, what is the final output of the military? There are a multitude of activities that add up to the capacity to defend the country against

523

Measuring Public-Sector Output

military intrusion. One such activity is the training of military personnel. That output may be measured. But there are also other activities, like additions to the stock of weapons and systems, maintenance and the repair thereof, planning, surveillance, and spying. In these cases there is no other way but to measure these intermediate outputs and add them up. In the case of defense, the productivity in training of recruits and pilots served as representative indicators for the whole sector. Representative parts of sector production were used in some other instances: productivity change for Statistics Sweden was calculated on a sample of 20 percent of its products, for hospitals on 30 percent of the clinics, and for libraries on public libraries of local communities only. But, aside from these examples, most of the calculations were based on aggregate output statistics and aggregate resource consumption. Of course, there is a risk of double counting when calculations are based on intermediate outputs. Adding intermediate outputs instead of adding value added or estimating overall productivity change from a partial productivity measure is quite tricky. There are instances when the change of productivity of an internal process is a good approximation to the change of productivity in the overall operation. And there are instances when this is not so. As an example one might think of a partial productivity measure like number of pupils trained per teacher. If the number of teachers and other resources remained the same and teacher productivity increased by 4 percent, it is very likely that overall productivity would also increase by 4 percent. However, if the number of pupils remained the same and the number of teachers decreased by 4 percent, while other resources remained the same, the teacher productivity increase would overstate the overall productivity increase. From the point of view of the agency, it might be very relevant indeed to include every activity that adds to the workload. However, that may end up relating input to input. The dangers in adding intermediate outputs or in inferring from productivity change in intermediate production to overall productivity change call for a quest for final outputs as far as possible. Among the 14 national government agencies in the Swedish study, the number of output indicators varied from 2 to 3 to 40. In other sectors the variation is even greater. Community planning rested on only one crude measure, as previously mentioned; the health sector built on over 300 measures: the output of the sample of clinics was treated as unique to each clinic and given a separate weight. In appendix A, I provide a sample of output indicators. 14.3.2 Weights Determining weights for the nonmarketed services of the public sector is in principle a matter of social benefit analyses. Such analyses provide values that could serve as weights in a measurement of government output that is truly welfare oriented. However, except for cases such as labor market policies,

524

Richard Murray

roads, education, and some others, estimates of social benefits have not been performed successfully. Therefore vast areas are left without weights. Discarding the need to aggregate dissimilar services and concentrating on services with similar aims, it would be quite satisfactory to have estimates of service effectiveness. All government goods and services that aim, for example, primarily at saving lives could be given weights according to their effectiveness in this respect. Such weights would indicate the relative contribution to the common aim. This works in only a few instances and on an almost microlevel, because the intended effects are so specific to the services produced. Within the healthcare sector many other goals are pursued beside saving lives, like getting people back to work, or curing for a better life or preventing possible ills. If several goals are to be included, we are back to weighing social benefits. In some instances weights were judged by service effectiveness. An example is the use of flight simulators in the training of air force pilots. For those parts of the training where simulators were used, they were deemed by professional teachers to teach the same things that would otherwise be trained in the air at a much higher cost. Those hours were added with the same weights, which contributed substantially to a productivity increase in the training of air force pilots. A second example is the customs authority that completely changed its style of work from storing and inspecting goods themselves to a system of self-declarations by importers. Output in terms of possibilities to levy taxes and duties remained the same. Regarding these different forms of outputs as equivalent makes the customs authority register a substantial increase in productivity. In addition, importers benefited from speedier deliveries. Also, in cases where we use highly aggregated measures of output there is an implied cost-effectiveness weighting: tax forms are divided up in only two types, each with a separate weight, but within each type all tax forms are treated as equal. In the public sector we have substituted politicians for the market. An interesting approach is to regard politicians’ decisions as the revealed preferences of the electorate. Marginal costs of production are then the prices at which the substitute (representative?) consumers (politicians!) would go shopping. They would buy health care up to the point where its additional value equals its marginal cost. Values would thus be reflected in marginal costs. Of course, this is based on a very optimistic view of the rationality of the political and administrative process governing the production process in the public sector. But there seems to be no other comprehensive approach that could compete with it. However, it has one severe drawback, for it leaves no room for productivity increases via a more efficient output mix. Suppose that very many elderly people are taken care of at regular hospitals at a very high cost and that one finds out that many of them might be better taken care of at retirement homes

525

Measuring Public-Sector Output

at a lower cost. Moving some of the elderly patients out of the hospitals and into retirement homes increases unit costs both at the hospitals and at the retirement homes because those patients that are transferred cost more than the average at the retirement homes and less than the average at the hospitals. Productivity decreases in both places. However, realizing that the value of a place in a retirement home is just as high or even higher than the value of a place in a hospital would cause productivity to go up in both places. The revealed preference approach has still another major drawback. It is that government policies cannot be evaluated from the outside. What politicians do is the best that can be done-as Dr. Pangloss might have said. Consequently, wherever possible we tried to insert weights that reflected the social benefit or the service effectiveness. For the remainder, that is, the main body of weights, we made do with marginal costs as weights. Using unit costs from a specific base year as weights implies another interpretation of the change in productivity aside from the change in welfare. The aggregate government output equals the change in costs assuming constant unit costs. Productivity then reflects the change in productive capacity.

14.3.3 Quality Realizing that quality lies at the heart of the service makes it easier to spot the important variables. The main qualitative aspect of weather forecasts is that they are correct: the percentage of correct prognosis in terms of temperature, wind, and rain can be measured. Social security checks should be correct in relation to legislation: random samples can be evaluated. Roads should be safe: the rate of accidents is recorded. We have recorded such quality indicators, which is half the problem. The other half is how to adjust output for quality. Dealing with quality includes the choice of output measures. One may distinguish between three kinds of output variables: throughput, output, and outcome. Throughput measures work loads and may even come close to input, like the number of cases in the in file. Output is the goods or services delivered, like the number of cases handled. Outcome is the result from the point of view of the principal or the customer, for example, the reception of valuable advice. National accounts have no standard in this respect, for they measure whatever has a price tag. Therefore, for government output one has to think about and try to find out what would have a price tag if these goods and services were sold on a market. Let us take the example of crime investigation: A crime is reported to the police, it is registered, and there is a good chance that it will be investigated by the police, and a lesser chance that it might be solved. Output could be measured by the number of reports. This is throughput. Or it could be measured by the number of cases investigated-this is output. Or it could be measured as the number of cases solved-this is output with an eye to outcome.

526

Richard Murray

The man in the street is interested in nailing the criminal-solving the crime. So are the principals, although they realize that it will not be possible ever to solve each and every crime. Suppose we were to measure the number of investigations as output and then would like to adjust that measure with a measure of the quality of that output. Various candidates could be the number of hours spent on a case, the qualification of the personnel, the number of pages of written report on a case, and the percentage of cases solved. The percentage of cases solved no doubt comes closest to the result of fulfilling the objective of the principal. So why not include the quality aspect of output in the measure, that is, counting as output the number of cases solved! That is exactly what we did. Market prices reflect the value of the output to the consumer, if markets are in equilibrium. Using prices as weights incorporates quality in the output measure. This is however, an individual evaluation, which does not encompass distributional aspects and externalities. To parallel the measurement of marketed outputs, such considerations should not enter into the measurement of government output, either. We may stop investigating whether in the long run the outputs of the police in terms of solved crimes actually produce fewer crimes, as long as there is a demand for solving crimes by both principals and citizens. This is analogous to not investigating whether cars make people happier: people buy them; that is enough for national accounts’ purposes. Far from being neglected, quality has in many ways been included in the measurements of public-sector output in the Swedish study. Here are some examples: Measurements have favored outcome indicators that have a reasonably close connection to the output. For example, number of treated patients is preferred in relation to hospital days. This statistic has its drawbacks but it captures the shortening of hospital stays. The output of highway authorities is measured by the number of miles traveled by various vehicles on those highways rather than by the number of miles of roads maintained and built. Whether highway authorities build roads for which there is demand or roads that are wasted is included in the measure in this way. Another way to treat quality change is to separate services of different qualities, assign different weights and then aggregate. For example, treated patients are divided into 312 different outputs according to the type of clinic, each with a different weight. Changes in the composition of treatmentsmore or less qualified-are thereby captured. Quality adjustment is a very similar exercise. The social insurance office increased the quality of their benefit payment by acquiring more accurate and up-to-date information on the incomes of the insured. Before this was done, output was regarded as one type of output, afterward as another kind; they were also assigned different weights. Moreover, added features may be included as new outputs, for example, separate rooms at hospitals and lunches at school. Sometimes quality may be transformed into quantity. The main output of

527

Measuring Public-Sector Output

Statistics Sweden consists of several statistical products that remain fairly constant. However, their precision changes according to enlargement of samples or new sampling techniques. Most qualitative changes may be recalculated in terms of the size of the sample and measured as such. Closely related to these ways of dealing with quality is the technique to discard all those costs associated with quality increase. The costs of health care, education, leisure activities, and some other qualitative improvements of prisons were excluded from total costs. These were considered as important improvements in prisons both from a humane perspective and from a therapeutic point of view. However, the effect on the tremendous productivity decline was negligible, and criminal recidivism remained unaltered. All these methods produce output measures either that incorporate quality or that deduct the costs associated with quality. Of crucial importance when adjusting for quality, however, is what weights are used. More often than not the weights are unit costs or marginal costs associated with the change in quality. This practice rests on the assumption that those costs would not have been incurred had the qualitative increase not been valued at least as high as that cost. The measure of output in that case is biased downward. On the other hand, the rationality assumption might be totally false and the output measure could be biased upward. Quality poses a measurement problem only insofar as it changes. One way of controlling for quality is merely to look for indicators that might tell whether quality has been changing. One may look at the effects of government programs, like health indicators, recidivism of criminals, scores in student achievement tests, and road traffic accidents. Of course, these indicators may be influenced by other factors than the quality of government output. The evidence must therefore be interpreted with care. Throughout the study checks on quality were performed. Some indicated a definite quality increase, like road safety and precision in weather forecasts, that could not be incorporated in any reasonable way in the output measure. Most of the checks indicated no change, which made it possible to leave output measures as they were.

14.4 Other Methodological Problems and Their Solutions I will now briefly mention some other methodological problems and how they were addressed in order that the reader may correctly appraise the results of the study that are to be presented subsequently. Two aggregated measures for the government sector as a whole were presented: One covered the measured parts of the government sector, where each measured entity was weighted by its cost share in 1980. Another was for the government sector as a whole, where measured entities were weighted according to the share of public consumption that the purposes they represent commanded. Aggregation for the government sector as a whole was undertaken

528

Richard Murray

only for the period 1970-80, where the coverage was best. The two measures for the sector as a whole came out almost identical. This is in large part due to the 70 percent coverage of the measurements. Therefore, only the latter, representative, productivity measure is presented in the following. The denominator in the productivity expression has been calculated along the lines of the national accounts. For some branches the cost is set equivalent to public consumption in fixed prices. It includes depreciation of capital but no interest charge. For others the cost in fixed prices is calculated from costs in nominal prices deflated by implicit price indexes for various inputs in the national accounts. Deflating costs by the consumer price index would have produced different results. Sensitivity tests have been performed in very many ways. Different ways of measuring output, different systems of weights, different deflators, and so on, have been tried. Of course, results are influenced, but on the whole the results are quite robust for variations that are reasonable and compatible with the general approach. Needless to say, the quoted figures are not exact. An interval of t 0 . 5 percentage points should be added to the stated figures on the average yearly growth rate of a sector. In the aggregate the confidence interval is likely to be smaller.

14.5 Productivity in the Government Sector of Sweden 1960-1985 This study is unique in that it covers a very long period of time, 25 years, and most of the government sector, whether national, county, or local.' The 1960-85 time period includes those years in which the large public sector, the welfare state, was created in the industrialized countries. There are reasons to believe that the patterns that emerge in the Swedish Report are indicative of what has been going on in other countries as well.2 What we are witnessing is not some special political effect of a social democratic regime. 1. References, necessarily in Swedish, are given to all separate studies in the reference list. 2. For the Netherlands there is a study that resembles the Swedish (Goudriaan, de Groot, and van Tulder 1987). It covers the period 1975-83, 56 percent of the government output-32 large, publicly provided services within the six subsectors: ten health services, seven types of education, five social and cultural services, three modes of public transport, four services of the police and justice system, and three executive branches of the tax and social security administration, and costs are deflated by consumer prices. In eight years costs per unit increased on the average by 4 percent. In Denmark partial analyses of local public services point in the same direction (Mikkelsen 1982). The conclusion from ad hoc studies for a few areas in the United States is that state and local government productivity has remained stagnant or decreased over the past several decades (Fisk 1983). Crude labor-output relations over the years 1960-80 for a few public services in some industrialized countries indicate productivity decrease (Maddison 1984). The only puzzling exception is the U.S. federal government, which, since 1967 and up to 1986, has recorded a yearly productivity increase of 1.5 percent. Measurements cover two-thirds of the employment, are made to a large extent on an intermediate level, and outputs are related only to labor inputs (Bureau of Labor Statistics 1988).

529

Measuring Public-Sector Output

Table 14.3

Productivity Growth in the Government Sector of Sweden 1960-1980 (yearly change in percent) 1960-65

General administration Justice & police Defense Education Health care Social security Social welfare Community planning Libraries Economic services Total

1965-70

1970-75

1975-80

-3.2 - 3.6 -0.4

-3.7 - 2.7 ... -6.3 -3.1 - 2.6

-4.9 1.5

+

+3.0 +2.1

-5.5 -6.1 -0.1 +0.2 - 1.4 -4.8 -2.8 +0.2 +1.1 +0.1

+4.5 +3.1 -1.0 -3.2 - 2.2 - 0.2 - 0.4 -8.9 - 1.8 0.4

+

-0.6 - 1.5 - 1.8 - 2.5 - 1.6 -4.5 -0.3 +0.2

...

...

- 1.4

- 1.6

- 1.5

... ...

...

1970-80 -0.6 - 1.6

Productivity Growth for Selected Branches of the Government 19801985 (yearly change in percent)

Table 14.4 Branch

Growth

National government administration (1980-83) Primary schools Secondary schools Colleges Health care (1980-84) Social welfare (1980-84) Libraries

3.5 -0.3 - 1.8 - 1.7 - 2.2 2.0 -2.9

+

I will now discuss results from the Swedish study, as presented in tables 14.3, 14.4, and 14.5. The general picture of Swedish public-sector productivity is one of decline. With few exceptions, all the studied government branches and individual agencies show a negative productivity change. In business it may happen that the productivity of a branch decreases, but only for a short period of time. After such a period, forces are set in motion-that is, competition-to correct the course, and the branch gets back on the track of productivity increase. In government production the reverse seems to be true: productivity may increase, but only for short periods of time. Then it is typically followed by continual productivity decreases. There are exceptions, but they are few: the National Agency for Roads (the main part of economic services), the Board of Customs, the National Housing Board, the Meteorological Institute, Statistics Sweden, the Salaries and Pensions Board, and county-council-operated psychiatric care. These activities have experienced a positive productivity change on the average throughout the period. Over time there are some sparks of light. The productivity of the national

Table 14.5

Productivity Growth in the National Government Administration, 1960-85 (yearly change in percent)

Agency National Labor Market Board National Housing Board courts Prisons Enforcement service National Board of Agriculture National Land Survey Police Social security Tax administration Board of Customs Meteorological Institute Patent & Registration Office Statistics Sweden Salaries & Pensions Board (not included in the aggregate) Total

1960-65 - 1.9

5.0 -5.4 -5.6 -5.0 -4.0 -

1.0

- 2.9

5.0 -3.1

- 2.0

1965-70

1970-75

- 7.4 - 0.6 -0.9 - 6.0 -4.1 -1.6 0.3 - 1.8 - 2.6 -7.1 5.2 4.2

- 3.5

-3.3

6.6 1.3 -11.0 -4.9 0.6 -2.9 - 6.2 -4.8 - 6.4 -4.3 -3.7 -4.3 2.4

-5.2

1975-80

1981

1982

1.9 2.0 2.8 0.3 3.1 1.1 2.5 3.6 -0.2 5.1 4.1 4.7 -3.2 1.o

22.8 5.2 - 1.2 -0.2 0.8

- 2.3 -0.5 -0.7 10.9 - 0.5

5.8 3.7 0.2 2.9 4.3

1.2

3.6

2.4

2.5

13.1 - 8.2

11.2 3.2 3.4

1983

1984

1985

2.4 2.9 -0.9 -9.7 -5.8

-2.9 - 1.3 -0.2 - 4.5 - 3.0

- 1.3

4.0 0.9 3.8 - 12.0 2.0

-6.7 -4.4

-0.5

-0.5

- 3.6

3.5 -1.3 8.9

8.3

- 3.0

1.3 7.3 -5.5

531

Measuring Public-Sector Output

government administration plunges very deep in the years 1960-75, but from 1975 and on, there is a marked increase in productivity (tables 14.4 and 14.5). Social welfare turns from productivity decline in the early 1970s to productivity increases in the early 1980s. Studying the trends more closely, one finds a very definite relationship between the growth of output and the rate of productivity change. A faster increase in output is connected with a lesser decline of productivity or even with an increase in productivity. We can also see that there are instances when agencies have absorbed large increases in work loads despite an unchanged capacity, indicating that they have had an excess capacity. For example, the employment agency managed to handle a 23 percent increase in the number of job seekers in 1981 with constant resources. Decreases in output invariably lead to declining productivity. Resources are not cut back in proportion to diminishing work loads, if at all. In the 1980s there are some examples of national government agencies that manage to decrease inputs in relation to a decrease in outputs, for example, the housing board, the enforcement service, and the social insurance offices. The sluggish response of government production units has been observed by many others. It causes substantial cost increases because “demand” for, and output from, the public sector varies a great deal. This we now know, thanks to the studies performed on public-sector output. The pattern of demand is very clearly countercyclical. Increases in demand, for reasons to be investigated by future researchers, take place in periods of recession; demand stagnates in periods of boom. This increases the cost of sluggish response in adjusting the resource requirements. In addition to different conditions of survival, are there different demands on private and public organizations that explain the sluggish response? When facing a rising demand, a private firm may either raise the price or refuse to serve some customers if production capacity is not adequate. A public organization cannot do that. It does not control price. It has an obligation to serve and treat all alike. There is a case for running the show permanently with excess capacity, but not with a continuously growing excess capacity, which is what the productivity decline seems to indicate. What characterizes those activities that have had a long-run increase in productivity? Roads have had the most spectacular growth in demand. Output, measured by vehicle miles, has risen steadily with an average of 4.5 percent per year. This, plus the technical advance in road construction and maintenance, should be enough to boost productivity to the highest levels. It is a real surprise that productivity under these circumstances increases by only 0.4 percent per year. It makes a great difference whether capital costs are included or not. If costs are calculated as the sum of consumption and investments, productivity increases at roughly 4 percent per year. But this is due to a sharp decline in road investments. If, instead, depreciation on the accumulated capital stock is included (as it should be), the rise in productivity almost disap-

532

Richard Murray

pears. An increase in quality in terms of safer roads and roads that save fuels, time, and wear and tear is not included in the measurement. The board of customs has also faced a steadily rising demand of 3-4 percent per year in the number of shipments to be declared, and the number of vehicles and passengers crossing national borders. But there are also examples of radical changes in work styles. In addition to the previously mentioned introduction of a self-declaration system, Sweden and its neighbors share the responsibility for border control. These changes in work style show up in a productivity increase. Periodically there has been a large increase in the demand for meteorological forecasts. A large part of the production is sold on an almost commercial basis. Production has been heavily computerized. This shows up in a productivity increase, but only of 0.5 percent per year. The importance of computerization lies in dramatically increased capacity to process large amounts of information, which has resulted in more correct prognoses. In conjunction with massive housing programs, the output of the housing board in terms of mortgages handled has increased steadily at a rate of more than 5 percent per year. For some reason this agency has not received much political attention and consequently not much resources. Very little technical change has taken place. The central part of the agency has diminished in relation to the local parts. Increased productivity of around 3 percent per year is best interpreted as the exploitation of economies of scale. Even in-patient psychiatric care at hospitals has expanded strongly in the period, at roughly 5 percent per year, measured as the number of patients admitted. The productivity increase of 2.2 percent per year originates out of a shortening of the average length of stay from 300 days to 100 days, mainly because of the use of psychiatric drugs, that is, an example of technological advance. It is very difficult indeed to see in what way these areas differ fundamentally from other areas of government production. They have all experienced a strong expansion of demand, but so have some other areas. It is evident, though, that a strong increase in output helps productivity growth. The question remains, What makes productivity decrease? From economic theory it is very difficult to deduce the causes of technological retrogression. Of course, there is the possibility that there are diseconomies of scale and an increasing marginal cost of output. Because production in the government sector more often than not is organized by a single producer, it might be especially susceptible to diseconomies of scale. In addition to too-large production units there is the diseconomy of stretching government programs to cover ever-larger proportions of the population. Travel assistance to the handicapped and the elderly is less costly to organize in densely populated areas. The costs increase as this service is offered in more sparsely populated areas. The cost of secondary education increases more than in proportion to the number of students as enrollment approaches 100 percent. Reaching the very last

533

Measuring Public-Sector Output

households with TV programs is very costly in relation to average cost, clearly. This hypothesis needs to be tested empirically. Another hypothesis is that regulation has increased within government oPerations and that this has caused a falling productivity. The productivity increase of the National Housing Board is evidence that this need not be so. Its operations have been bounded by an ever more complicated legislation. General regulation in areas such as environment, employment, and taxes should have harmed the private sector as well, but we see no sign of that in terms of productivity decline. Shortened work hours do not in themselves cause productivity to decrease, because we have deflated costs with an index of the effective wage rate. However, indirectly this might increase costs, for example, by raising the number of square meters of office space needed per employee. This should happen in private and government organizations alike. But we know that government organizations utilize a lot more office space than private organizations, in for example, dentistry, schools, hospitals, and consultancy. An hypothesis connected with the former is that agencies do not adapt their input mix to changes in relative prices of inputs. This is either because of general lack of cost consciousness or because of detailed regulation regarding what inputs to use. Such a hypothesis is not supported by evidence from the study of state agencies. In the aggregate of these agencies there is a substantial change in the input mix, and it is in line with the changes in relative prices (see table 14.6). The labor share diminishes from 81.4 percent in 1960 to 73.7 percent in 1980, while, for example, office space increases from 4.1 percent to 10.0 percent. At the same time, wages increased by 456 percent and rents by 300 percent. For all but one agency the growth in labor productivity is larger than that in total factor productivity. This demonstrates, contrary to this hypothesis and some popular beliefs, that state agencies do plan their resource mix in accordance with the relative prices of inputs and that there are possibilities for input substitution. Most importantly, though, is that the measurements of output in the government sector might have missed an important qualitative improvement of the services. All the studies made serious efforts to detect qualitative change. But, Table 14.6

Input Shares and Price Change on Inputs (%) Input Shares

Labor Intermediary inputs Office space Capital

1960

1980

81.4 14.3

13.1 15.0

4.1

10.0

0.2

1.4

Price Change 1960-80

+ 456

+ 400 + 300 -t257

534

Richard Murray

of course, they are hampered by a lack of data. As far as one may judge from the evidence at hand, though, the qualitative changes left out ought not affect the conclusions dramatically. The most difficult area to judge in this respect is health care. In order that the 3 percent decrease in productivity be compensated by a quality increase the value of the health-care services must on the average be twice as high in 1985 as they were in 1960. Although the measures employed capture some elements of quality like the shortening of hospital stays and the shift of work loads from more costly clinics to less costly, there are shortcomings in the measures of output. We know of spectacular advances in medicine in very narrow disciplines that have not been accommodated. We know that the measurements of output would be better if made in diagnostically related groups; then we would, for example, capture the productivity increase in the treatment of ulcer with drugs instead of by surgery. But over the period there are no clear signs of improved health, fewer sick days, or longer life expectancy, and so on. This is astounding, because it is not only a matter of the quality of health care but should also be influenced by the massive quantitative increase in output-roughly 65 percent from 1960 to 1985. Have conditions that influence health really undergone such a dramatic deterioration? In education one source of decreasing productivity is claimed to be the diminishing size of classes. Also, it is claimed that decreasing the studentteacher ratio should be a qualitative improvement (although educational research does not support that contention). There are some studies of student achievement that roughly cover the period of investigation, and they show no sign of improvement. Qualitative changes in social welfare services have been investigated thoroughly in connection with the Swedish study particularly in two areas: child care and elderly care. A host of qualitative indicators has been analyzed, but with no definite answer. And so the story goes. In area on area, with few exceptions, there is no evidence of a qualitative change that would upset our measurements. Enforcement services collect slightly less SEK out of what they should collect; criminals go back into crimes to the same extent after treatment in prisons; crimes committed continue to increase despite an increased police output; and so on. Except for in the areas mentioned there is no evidence to support the belief that quality has increased. This is a bold statement, and it needs reexamination from another point of view. Agencies often point to added features of the output. At the hospitals, patients are nowadays lodged in rooms of their own or in smaller groups than before. Citizens today receive advance notice that their passports expire. These are examples of valuable improvements in output. There are many more. An hypothesis is that the increase in costs and lowered productivity may have been caused by a proliferation of added features of government output. Because there is no real market-but the willingness of the politicians to raise

535

Measuring Public-Sector Output

the taxes-in which to test the value of the products, new features and new products may be launched even if their value is far below the added cost. Substantial effort has been put into investigating this issue. One case is that of prisons. It has been said that costs have increased because of a series of improvements, like the introduction of vocational training, leisure activities, extra costs incurred with more lenient rules for prisoners to leave the prisons temporarily, and so forth. Interestingly enough, the costs that were identified by the prison authorities were of a rather smallish nature. Out of a total increase of unit costs by 252 percent these costs were calculated to add 28 percentage points. The investigators identified other sources of cost increase that were much more important (see table 14.7). Fewer clients per supervisor added 38 percentage point and reduced crowdedness 35 percentage points. However, the main part of the cost increase remained unexplained. Libraries are another example. It has been contended that libraries nowadays are much more than book-lending machines. They serve the general public with information, they arrange cultural events, they serve as public sitting rooms where people go to read newspapers. The extra cost for these byproducts is, however, estimated at only 3 percent of total operating costs of libraries. It is negligible in relation to the 25 percent decrease in productivity. If classes in primary schools had not become smaller productivity would still have declined - 1.7 percent per year. Not accounting for this “quality increase,” productivity dropped - 2.8 percent. Excluding the cost of increased room standard at hospitals reduces the productivity decline by 0.1 percent per year. Hence, although the costs incurred from added features on occasion may be quite impressive, they are far from explaining the long-run productivity decline. What other causes of productivity decline are there? Excess capacity is one piece of evidence that slack exists. Another is the productivity increase brought about by less lenient budgetary appropriations to state agencies that occurred in the period 1975-80. It is very difficult to pinpoint slack in an organization, because it does not show up with personnel just sitting around doing nothing. Slack may just as well consist in a hectic life at the workplace, because of disorganization and inappropriate priorities. It is not until people in the organization receive a clear understanding of what their goals are that it Table 14.7

Prisons, Unit Cost Increase in Fixed Prices, 1960-1980 (%) Total cost increase Of which: Vocational training, health care, leisure activities, permissions to leave Reduced size of prisons Reduction in the number of clients per supervisor Reduction of crowdiness Unexplained

+ 252 + 28

+ 10 + 38 + 35 + 141

536

Richard Murray

becomes possible to organize the work in a suitable way and to make the right priorities. Even in such a labor intensive business as child care there may be slack. This is exposed by large differences in unit costs among different day-care centers. The range may be up to 50 percent and increases when it is calculated on hours of child care instead of the number of places. Also, comparing the actual number of people at work, including all kinds of personnel, with what has been recommended nationally disguises a slack of 45 percent.

14.6 Some Tentative Conclusions We may conclude that there has been a long period of productivity decline in the public sector of Sweden. This period coincides with the buildup of the welfare state. Being able to measure outputs instead of just inputs, we may observe that there has been a real expansion of government services, but not in proportion with the increase in resources spent. Productivity decline means that services become more costly. We have measured only the average cost. Measurement of the marginal cost of output would have shown an even steeper upward slope. It might be said that our measures understate the increase in output by not capturing qualitative improvements, added features, and new products and that they do not include the advantages of an improved allocation of outputs in relation to effectiveness or values of services. But if that is so, we must conclude that when our measures of output are compared with the ultimate state of affairs-that is, with the general health of the population, the attainment of students, the crime propensity, tax receipts per SEK spent on tax administration, and so on-output seems to be less and less effective. The reasons for the general productivity decline seem to be a blend of sluggish response in resource use to variations in demand, in diseconomies of scale, rising marginal costs of government programs as they are made more comprehensive, additions of new outputs and features with little effectiveness, and increasing slack. All these reasons are susceptible to manipulation. Beginning in 1975, the state government embarked on a fiscal policy of selective restraint. Although the budget deficit exploded at that time, state government agencies were forced by, on the one hand, less permissive appropriations in real terms-a substantial wage inflation occurred at the time-and on the other hand, by a rapid increase in demand (work loads) to increase productivity. In the 1980s these agencies experienced zero growth of real resources. Public consumption of local governments grew at a faster rate than of the state government in the late 1970s and in the beginning of the 1980s. Productivity continued to decline, though slightly less in schools, and there was in fact a productivity improvement in social welfare services. In health care productivity declined at an unchanged speed.

537

Measuring Public-Sector Output

The achievement is also discouraging in comparison to the private service industry (table 14.8). Private service industries as measured in the national accounts may have a 1-2 percent increase in productivity per year.3 Some public sector services belong in fact to the private sector in that they are financed on a commercial basis. Those branches with a large public share are shown separately in table 14.8. Two of the branches have experienced a long-run positive productivity development-transportation and telecommunications. In transportation there is a fairly large share of private firms, enough to make competition real. In telecommunications there have been strong technological developments, but there are reasons to believe that a large part of the productivity increase stems from economies of scale or scope. The other branches are housing, with a majority share of public housing and heavily regulated, postal services (100 percent public) and sanitation, water and sewage (mainly public). The Swedish study found productivity decreases for all three branches. This reinforces the impression of a specific productivity problem in the public sector.

14.7 Some Macroeconomic Implications The aggregated productivity decline for the public sector in Sweden in the 1970s is - 1.5 percent per year. Is this sufficient to cause any alarm? In table 14.9 the productivity decline in the 1970s is used to recalculate the GNP. The figures sum up to the conclusion that GNP rose 25 percent less than officially recorded if account of the productivity decline is taken. The officially recorded growth is 2.0 percent per year. Decreasing the growth of public consumption by 1.5 percent yearly reduces the growth of GNP to 1.5 percent per year. Knowing that the change in value-added productivity and total factor productivity come out almost alike when the share of intermediary inputs is small, one might recalculate GNP in value added terms taking the change in total factor productivity to be equal to the change in value-added productivity in the public sector. Another way of doing this is to calculate the change in value-added productivity in the public sector by setting the value of its output equal to the unit cost in 1980. It produces exactly the expected result. In particular, the decline of value-added productivity in the public sector as a whole is somewhat larger, - 1.8 percent per year over the period 1970-80. Accumulating the increased cost of the public-sector production, due to decreased productivity, to borrowing requirements and adding to that also an interest charge of 10 percent-all in nominal terms-adds up to 106.7 billion 3. There are deficiencies in the measures of output of the private service industry that may make many of those of the public sector in the Swedish report compare quite favorably. Note that the measured productivity in table 14.8 is not value-added productivity, but, in line with all other calculations, total factor productivity. In comparison this measure of productivity produces smaller variations than value-added productivity.

538

Richard Murray

Table 14.8

Productivity Growth in Private Services, 1965-1980 (yearly change in percent) 1965-80

Total Of which: Transportation Postal services Telecommunications Housing Sanitation, water, & sewage

Table 14.9

1.1 2.2 1.5 1.3 -1.1 -0.7 -

A Recalculation of GNP, 1980 Prices (billions of SEK) 1970

Private consumption Investments Export-import Public consumption, assuming: Constant productivity Productivity decline 1.5% per Year GNP, assuming: Constant productivity in public consumption Productivity decline of 1.5% per year in public consumption

1975

1980

230.892 113.590 - 23.429

259.903 121.556 - 19.827

270.049 111.89 1 - 9.997

111.594 17.774

129.9 14 10.646

153.156

432.647

49 1.546

525.099

450.421

502.192

525.099

...

SEK. During the 1970s the state government debt increased by 192 billion SEK. We may conclude that the productivity decline caused more than half of that. The increased cost of production is 21 billion SEK. One can add to that 8.6 billion SEK of added interest payments on the accumulated state government debt. Had these costs not been incurred the tax ratio could have been lowered from 49.5 percent to 43.8 percent of GNP. In addition to this impressive amount, there is the deadweight loss of taxes. This exercise in calculation leads to a final conclusion. Suppose output measures have completely neglected quality improvements and that productivity has in fact been constant or even increased. Of course, this would affect the rate of growth. However, the effect on the tax rate would be the same: quality improvement must be paid for just as well as productivity decline. Therefore, we may conclude, that this development, no matter whether it is caused by productivity decline or by quality improvement, cannot go on very

539

Measuring Public-Sector Output

much longer. The development of the welfare state undermines its own foundations. Either taxes must be raised continuously or the volume of output must be reduced, implying that some people will not be served. With present day resources the - 2.2 percent productivity decrease in health care implies that in 20 years only two-thirds of the patients may be treated. Despite its shortcomings the Swedish study has provided many valuable insights into the public sector. It seems to be the case that the public sector generally, but with notable exceptions, has a mounting problem of production efficiency. This observation has caught political attention and administrative solutions are to a large extent sought in the application of output measurements and productivity monitoring along the same methodological lines that the study employed. The study has made quite clear the implications of declining productivity in the public sector for the Swedish economy as a whole. It has shown that it is both conceptually and practically possible to incorporate the public sector within the national accounts on a more realistic basis than on the zeroproductivity-change assumption. This requires, though, that one sticks to the original purpose of national accounts-that of describing production-and abstains from ambitions to measure the well-being of nations. Of course, there are many improvements to be made in the measurements of publicsector outputs, just as there are improvements to be made in the measurement of private-sector outputs.

Appendix A Sample of Output Indicators Branch

Indicator

Health care

Patients admitted Outpatient visits Bed days for inpatients Hours of attendance at school by pupils Children admitted to day-care centers Hours of care of elderly people Number of recipients of benefits Bed days Hours of attendance at flight training Days of training of conscripts Vehicle-miles Book loans Volume of building construction

Education Social welfare

Defense Roads Public libraries Community planning

540

Richard Murray

Labor market board Housing board Courts Prisons Land survey Tax administration Enforcement service Board of agriculture Customs authorities Social insurance offices Police Meteorological Institute

Job applicants Hours of attendance at training Housing loans processed Housing loans administered Sentencing of offenders Internment places used Number of maps Revenue in fixed prices (deflated by user charges) Income tax returns processed Proceedings Consultations with farmers Inspections carried out Customs declarations People insured Number of disbursements Crimes solved Patrol hours Prognoses made Revenue in fixed prices (deflated by user charges)

References Bjurek, H . , L. Hjalmarsson, and F. R. Forsund. 1988. Parametric and Non-Parametric Estimation of Efficiency in Service Production: A Comparison. Journal of Econometrics. Bureau of Labor Statistics. 1988. Productivity Measures for Selected Industries and Government Services. BLS bulletin no. 2296. U.S. Department of Labor. Cattin, P. and D. R . Wittink. 1982. Commercial Use of Conjoint Analysis: A Survey. Journal of Marketing 46 (summer): 44-53. Fire, R. 1988. Fundamental ofProduction Theory. Berlin: Springer-Verlag. Fisk, Donald M. 1983. Measuring Productivity in State and Local Government. BLS bulletin no. 2166. U.S. Department of Labor, Bureau of Labor Statistics. Goudriaan, R., H. de Groot, and F. van Tulder. 1987. Public Sector Productivity: Recent Empirical Findings and Policy Applications. Proceedings of the 41st Congress of the International Institute of Public Finance, 193-209, Madrid 1985. Detroit: Wayne State Univ. Press. Kuznets, S. 1971. Economic Growth of Nations: Total Output and Production Structure. Cambridge, Mass.: Harvard Univ. Press. Kvalitetsutvecklingen inom den komniunala aldreomsorgen 1970-1980 (Quality in elderly care 1970-1980). Ds Fi 1987:6. Kvalitetsutvecklingen inom den kommunala barnornsorgen (Quality in child care). Ds Fi 1988:l. Maddison, A. 1984. Comparative Analysis of the Productivity Situation in the Ad-

541

Measuring Public-Sector Output

vanced Capitalist Countries. In International Comparisons of Productivity and Causes of the slowdown, ed. J . W. Kendrick. Cambridge, Mass.: Ballinger. McFadden, D. 1976. Quanta1 Choice Analysis: A survey. Annals of Economic and Social Measurement 514: 363-390. Mikkeken, P. 1982. Offentling produktion, produktivitet og effektivitet (Public Production, Productivity, and Effectiveness [in Danish Local Government]). Amtskommunernas och kommunernas forskningsinstitut.Copenhagen. Murray, R. 1987. Den offentliga sektorn-produktivitet och effektivitet. (The Public Sector-Productivity and Effectiveness). Bilaga 2 1 till LU 87, SOU 1987:3. Nordman, P. and T. Pettersson. 1983. Minskadproduktivitet i offentlig sektor-en studie av patent-och registreringsverket (Productivity Decline in the Patent and Registration Office). Ds Fi 1983:18. Ohlsson, I. Swedish Ministry of Finance. 1987. Public Services: A Searchlight on Productivity and Users. Report to the Expert Group on Public Finance. Produktions-, kostnads-, och produktivitetsutveckling inom offentligt bedriven halsooch sjukvird 1960-1980 (Production, Cost, and Productivity Growth in Health Care 1960-1980). Ds Fi 1985:3. Produktions-, kostnads- och produktivitetsutveckling inom den sociala sektorn 19701980 (Production, Cost, and Productivity Growth in Social Welfare). Ds Fi 1985:4. Produktions-, kostnads- och produktivitetsutveckling inom viigsektorn (Production, Cost, and Productivity Growth in Roads). Ds Fi 1985:9. Produktions-, kostnads- och produktivitetsutveckling inom arm& och Jlygvapnet (Production, Cost, and Productivity Growth in the Army and Air force). Ds Fi 1986:l. Produktions-, kostnads- och produktivitetsutveckling inom den offentligt jinansierade utbildningssektorn 1960-1980 (Production, Cost, and Productivity Growth in Education). Ds Fi 1986:17. Produktivitetsmatning av folkbibliotekens utliningsverk-samhet (Productivity in Library Services). Ds Fi 1989:42. Produktivitetsutvecklingeni kommunal barnomsorg 1981-1985 (Productivity in Child Care). Ds Fi 19885. Produktkostnader for offentliga tjanster med tillampningar p i kulturomridet (Cost of Production of Cultural Services). Ds Fi 1987:lO. Produktkostnader for offentliga tjanster-Detaljstudie rorande Historiska museet. (Cost o f Production: Museum of History). Bilaga 3 till Ds Fi 1987:10 (Expert Group on Public Finance). Produktkostnader for offentliga tjanster-Detaljstudie rorande Sveriges Television. (Cost of Production: Television). Bilaga 4 till Ds Fi 1987:10 (Expert Group on Public Finance). Statskontoret. 1985. Statling tjansteproduktion. Produktivitetsutvecklingen 1960I980 (Productivity in National Government Services 1960-1980: Main Report). Huvudrapport 1985: 15. Statskontoret. 1985. Produktivitetsutvecklingen Arbetsmarknadsverket (Productivity in National Labor Market Board). Rapport 1985:17. . ProduktivitetsutvecklingenBostadsverket (Productivity in the National Housing Board). Rapport 1985:18. . Produktivitetsutvecklingen Domstolsvasendet (Productivity of Courts). Rapport 1985:19. . ProduktivitetsutvecklingenKriminalvirden (Productivity of Prisons). Rapport 1985:20. . ProduktivitetsutvecklingenLantmateriet (Productivity in the National Land Survey). Rapport 198521. . Produktivitetsutvecklingen Skatteforvaltningen (Productivity in the Tax Administration). Rapport 1985:22.

542

Richard Murray

. Produktivitetsutvecklingen Kronofogdemyndigheten (Productivity in Enforcement Service). Rapport 1985:23. . Produktivitetsutvecklingen Lantbruksverket (Productivity in the National Board of Agriculture). Rapport 1985:24. . Produktivitetsutvecklingen Tullverket (Productivity in the Board of Customs). Rapport 1985:25. . ProduktivitetsutvecklingenSocialjiorsiikringar (Productivity in Social Security). Rapport 1985:26. . Produktivitetsutvecklingen Polisen (Productivity of the Police). Rapport 1985:27. . ProduktivitetsutvecklingenSMHI (Productivity in the Meteorological Institute). Rapport 198528. . ProduktivitetsutvecklingenSCB (Productivity of Statistics Sweden). Rapport, 1985:29. . 1986. PRISA 11, Uppfoljning av PRISA-projektet 1980-1985 (Productivity in National Government Services 1980-1985). Rapport 1986:31 .

Contributors

Paul A. Armknecht U.S. Bureau of Labor Statistics Washington, DC 20212-0001 Martin Neil Baily Department of Economics Tydings Hall University of Maryland College Park, MD 20742 Allen N. Berger Monetary and Financial Studies Section Board of Governors of the Federal Reserve System 20th and C Streets, NW Washington, DC 2055 1 Emst R. Berndt Sloan School of Management, E52-452 Massachusetts Institute of Technology 50 Memorial Drive Cambridge, MA 02139 Timothy F. Bresnahan Department of Economics Encina Hall Stanford University Stanford, CA 94305-6072 Edwin R. Dean Office of Productivity and Technology U.S. Bureau of Labor Statistics 200 Constitution Avenue, NW, room S-4325 Washington, DC 20212

543

W. Erwin Diewert Department of Economics University of British Columbia Vancouver, BC V6T 1W5 Canada Dennis J. Fixler Division of Price and Index Number Research U.S. Bureau of Labor Statistics 600 E Street, NW, room 4013 Washington, DC 20212 Barbara M. Fraumeni Department of Economics Northeastern University Boston. MA 021 15 Daniel H. Ginsburg U.S. Bureau of Labor Statistics Washington, DC 20212-0001 Robert J. Gordon Department of Economics, G- 174 Northwestern University 2003 Sheridan Road Evanston. IL 60208-2400 Zvi Griliches National Bureau of Economic Research 1050 Massachusetts Avenue Cambridge, MA 02138-5398

544

Contributors

Diana Hancock Mail Stop 7 1 Division of Monetary Affairs Board of Governors of the Federal Reserve System Washington, DC 20551 Alan Heston Department of Economics University of Pennsylvania Philadelphia, PA 19104-6297 David B . Humphrey Department of Finance College of Business R-53G Florida State University Tallahassee. FL 32306-1042 Dale W. Jorgenson Department of Economics 122 Littauer Center Harvard University Cambridge, MA 02138 Elizabeth Kremp Banque de France Direction Gentrale du Credit Centrale des Bilans 39, rue croix des petits champs 75001 Paris France Kent Kunze Office of Productivity and Technology U.S. Bureau of Labor Statistics 200 Constitution Avenue, NW Washington, DC 20212 Robert E. Lipsey National Bureau of Economic Research 269 Mercer Street, 8th floor New York. NY 10003 Jacques Mairesse Institut National de la Statistique et des Etudes Economiques 18, Boulevard Adolphe-Pinard 75675 Paris, Cedex 14 France

Marilyn E. Manser Office of Research and Evaluation U.S. Bureau of Labor Statistics 441 G Street, NW, room 2126 Washington, DC 20212 Paul Milgrom Department of Economics Stanford University Stanford, CA 94305-6072 Michael F. Mohr Bureau of Economic Analysis U.S. Department of Commerce Tower Building, Room 925 14th Street and Constitution Avenue, NW Washington, DC 20230 Swati Mukerjee Department of Economics Bentley College Waltham, MA 02154-4705 Richard Murray Statskontoret Box 34107 Stockholm S-100 26 Sweden M. Ishaq Nadiri National Bureau of Economic Research 269 Mercer Street, 8th Floor New York. NY 10003 Walter Y. Oi Department of Economics University of Rochester Harkness Hall Rochester. NY 14627 Jonathan Paul School of Business University of Michigan 701 Tappan Street Ann Arbor, MI 48 109-1234 Sherwin Rosen Department of Economics University of Chicago 1126 East 59th Street Chicago, IL 60637

545

Contributors

Michael Rothschild Office of the Divisional Dean University of California, San Diego 9500 Gilman Drive La Jolla, CA 92093-0064 Robin C. Sickles Department of Economics Rice University P.O. Box 1892 6100 S. Main Street 'Houston, TX 77251 Donald Siege1 Harriman School for Management and Policy 314D Harriman Hall State University of New York at Stony Brook Stony Brook, NY 11794-3775 Robert Summers Department of Economics University of Pennsylvania 3718 Locust Walk Philadelphia, PA 19104-6297

Jack E. Triplett Bureau of Economic Analysis U.S. Department of Commerce Room 710A, Tower Building 1401 K Street, NW Washington, DC 20230 Ann Dryden Witte

Department of Economics Wellesley College Wellesley, MA 021 8 1

Frank C. Wykoff Economic Inquiry 109 Seaver North Pomona College 645 North College Avenue Claremont, CA 9171 1-6363 Kimberly D. Zieschang Division of Price and Index Number Research U.S. Bureau of Labor Statistics 600 E. Street, NW, room 4103 Washington, DC 20212

This Page Intentionally Left Blank

Author Index

Abbott, Thomas A,, 111, 4501143 Abowd, John, 431119 Adar, Z. T., 300119 Admati, A., 203,205n12, 206n13, 210 Agmon, T., 300x19 Aly, Hassan Y.,257 American Petroleum Institute, 186n26 American Trucking Association, 379t. 380n15 American Trucking Trends, 381nn16,17 Andrews, Stephen H., 450n43 Appel, D., 177 Armknecht, P. A,, 112, 354 Arndt, H. W., 284,285 Arrow, Kenneth J., 169n9 Auromotive News, 143 Averch, H., 281111 Bagehot, W., 219x11 Bailey, M. J., 174 Bailey, Elizabeth, 399t Baily, Martin Neil, 14, 32, 376, 384,418, 423,429n1,431n7,442n30,453n45 Baltensperger, Ernst, 289 Barger, Harold, 162, 163t Barnett, William A,, 220, 223, 225n7, 229, 241, 248111, 284-85, 291,292n3, 297n2 Barron, J., 189n29 Baurngarner, James D., 338n28 Baurnol, W. J., 3, 197,211 Bear, Donald V. T., 339111 Becker, G. S., 12, 187, 303, 307n6, 312

547

Belsley, D. A,, 237, 2381123 Benston, G., 219n2, 250,290 Berger, Allen N., 227111 1, 241,248112, 250, 253t, 2581119, 259n20, 2611, 262t, 263n23, 264n25,266n28, 268, 272n32, 284 Berndt, Ernst R., 143, 272, 383.41 1, 425, 444n33 Blackman, S. A. B., 3 Blanciforti, Laura A,, 142 Bliss, C., 174 Blozan, William, Jr., 177n18 Bluestone, Barry, 182-83 Booz-Allen and Hamilton, 263n24 Borenstein, Severin, 388, 4001147 Boskin, Michael, 67 Bowen, B., 1771118, 182n21 Bradford, D. F., 344 Branscome, James M., 123 Bresnahan, Timothy, 456n50 Brown, C., 183n22 Brown, James R.,165 Bucklin, Louis P., 161, 175 Bureau of Economic Analysis (BEA), 26, 32n3, 51nn18,19, 314nn16.18 Bureau of the Census, 305n4 Burkhead, J., 344113 Calore, Daniel, 363 Campbell, B., 314n20 Card, David, 381 Cattin, P., 520

548

Author Index

Caves, Douglas W., 10, 228, 240, 372n4, 3751110, 381-82, 385, 393n36,406, 409-1 1 Caves, Richard E., 385 Chiang, S. Judy Wang, 41 In56 Christensen, L. R., 100, 228, 240, 372n4, 3751110, 381-82, 385, 3931136,406, 409-1 1 Christy, Paul T., 32 Coase, R. H., 174, 286 Coelen, Craig, 363 Cohen, Jeremy M., 444n33 Cole, Roseanne, 444 Cole, Stacey J., 450nn41,42 Council of Economic Advisers, 671128 Courant, P. N., 307n7 Daughtey, Andrew F., 41 In56 Dean, Edwin R.,290, 314n15, 374117 Debreu, G., 227 de Groot, H., 528112 de Leeuw, Frank, 373, 379 Dempsey, Paul S., 388, 392 Denison, Edward F., 16, 429n2, 456 Denny, Michael, 651126, 3821120, 383,425 DeVaney, A. S., 169 Diewert, W. Erwin, 65n25, 102n1, 228, 237n22, 240, 263, 281111, 282, 298117, 375n10,430n6 Domar, E. D., 102111, 105 Donovan, D. J., 220, 241, 248111 Douglas, Edna, 175 Douglas, George W., 170n10, 385 Dow, G. K., 307117 Early, J. F., 133 Ehrlich, Isaac, 172 Eisner, Robert, 307, 314n20 Engerman, S., 314n20 Fama, Eugene, 289 Fire, R., 2281112, 230,426 Farrell, M. J., 227 Feldstein, Paul J., 140 Fenier, G. D.,227nl1, 256 Fisher, I., 103n2 Fisher, Lawrence, 172 Fisk, Donald M.,528n2 Fixler, Dennis J., 220-22, 224n6, 227, 229, 239, 241, 248-49, 250n6, 252, 2531115, 283-84, 292, 293n4 Flint, Jerry, 4091153

Forbes, 143 Ford, I. K., 130 Francois, Joseph F., 123 Fraumeni, B., 15-16, 310n13, 311, 314nn15, 17, 316, 372114,456 Friedlaender, Ann F., 41 In56 Frornm, G., 4341117 Froomkin, J. T., 315n21 Fuchs, Victor, 3, 19 Fukuyama, H., 230 Fuss, Melvyn A,, 272, 382n20, 383,425 Galvin, John M., 142 Gates, J. A., 307nn6.7, 3141119 Geary, R. C., 496112 Gellman, Aaron, 401 Gilbert, M., 12, 354n27 Gillingham, Robert F., 110, 123, 143 Ginsberg, Daniel H., 354 Glantz, Frederic, 363 Gollop, Frank M., 105, 306, 314nn15,17, 316, 372n4,456 Good, David H., 372n4, 3821120, 387n24, 401,423,426 Goodhart, C. A. E., 222, 286n5 Gordon, Robert J., 14, 16, 32, 99n5, 100, 266n26, 376,384, 396,401,403t, 404t, 405t, 412,413t, 415n59,417t, 418, 422-23, 429n1,431n7,442n30,453n45 Gorman, John A,, 288 Goudriaan, R., 528112 Grabowski, Richard, 257 Graham, David R., 399t Graham, J. W., 308118, 315n22 Greene, William H., 256 Griliches, Zvi, 12, 16, 133, 188,294, 303111, 430115,431,4441133,454n46 Gross, Donald, 169n9 Grosskopf, S., 230n14,426 Grossman, Sanford, 200, 203-4, 206n13, 210 Grubel, Herbert, 496 Gullickson, William, 430n4,43 ln8,436n22 Hall, A., 346nn7,8 Hall, H., 165115, 175 Hancock, Diana, 220, 225117, 226, 228-29, 241, 248, 249n4, 284-85, 290-92, 297n3 Hannan, Timothy, 248112, 264n25, 268n30 Hanushek, E. A , , 304112 Hanweck, G., 219n2, 250, 290

549

Author Index

Harper, Michael J., 143,43On4,431n8, 436n22 Harris, Carl M., 169n9 Haveman, R. H., 304113 Hayashi, F., 202, 211, 2121114,214 Hayes, C. D., 351nn14,17,18, 352n20 Hellwig, M. F., 203 Heston, Alan W., 494n1,495,497,499, 501-3, 509,515 Hewson, John, 289 Hill, T. P., 5 Hinich, M., 223 Hirshleifer, J., 14 Hofferth, S., 349-50 Hogarty, T.F., 189n29 Holdren, Bob R., 161, 174-75 Hollowell, S., 346, 355 Holmstrom, Bengt, 202118, 204n10, 209 Hornstein, A,, 14 Hulten, Charles R., 102111, 105, 383, 425 Humphrey, David B., 219n2, 2271111, 241, 250, 2511111, 253t, 258n19, 259n20, 261t, 262t, 263n23, 266n28, 268t, 284, 290 Hunter, William C., 246, 254 Hutt, W. H., 169 Ingene, C. A,, 161 Inman, R., 347 Jablonski, Mary, 78 James, George W., 398n45, 399n46 Jamison, D. T., 315n21 Jensen, Michael C., 211 Johnson, L., 281nl Johnson, Richard L., 423 Jorgenson, D. W., 9n4, 15-16, 188, 306-7, 308nll,309n12, 310n13, 311, 314nn15,17,316, 373n4,456 Juster, Thomas, 307nn6.7 Kaplan, Daniel P., 399t Kelly, Henry, 32, 61n22 Kendrick, John W., 19, 307, 314n20, 315, 344n2, 371111, 373t, 430115 Klein, Michael A., 241112 Klerman, J., 345116 Knapp, J., 165115, 175 Kopp, Raymond J., 263 Kosary, Carol L., 123 Kravis, Irving B., 494111,495,497,499, 501-3,509

Kremp, E., 18, 355129 Kroch, E., 312n14, 315n22 Kuh, E., 237, 2381123 Kunze, Kent, 78, 290, 374n7 Kupfer, Andrew, 4091153 Kuralt, Charles, 3951139 Kuznets, Simon, 1, 12n6, 519-20 Kwast, Myron L., 266n27 Lal, Kishori, 401114 Land, K. C., 305114 Landes, W. M., 177n18 Lane, WalterF., 110, 121, 123, 143 Lange, M., 345n5 Leaver, Sylvia G., 114 Leibowitz, A,, 345116 Lester, R. A. 183n22 Levhari, D., 169n9 Lewis, W. Arthur, 161 Lichtenberg, Frank R., 431,442n28, 454n46 Liegey, Paul R., 133 Lindley, James, 247, 284 Lingren, B., 2301114 Lovell, C. A. K., 2271111, 230n14, 256,426 Lumsden, Kenneth, 339111 Lusch, R. F.. 161 McDonald, R., I ln5 MacDonald, R. J., 133 McFadden, D., 228, 520 McGowan, Francis, 387n24 McGuckin, Robert, 4321110 Machalaba, Daniel, 4091152 Machlup, F., 314n20 McMillen, M. M., 305114 Maddison, A., 528112 Mairesse, J., 18, 3551129 Malmquist, S., 227 Malt, R. A,, 344 Mamalakis, Markos J., 247, 251n12, 285n3 Manser, M., 11115, 3501112, 353t Mansfield, Edwin, 372n2 Marcoot, John L., I16 Marimont, Martin L., 421115 Mark, Jerome A,, 374n5, 375118 Marshall, N. L., 366n37 May, Doug, 651126 Medoff, J., I83n22 Mellow, W., 183n22 Mester, Loretta J., 290 Michael, R. T., 304113 Mikkelsen, P., 528112

550

Author Index

Milgrom, Paul, 2041110, 209 Miller, H. P., 308118 Miller, James C., 111, 170n10, 385 Mincer, Jacob, 303, 312, 314n15, 317n23 Mishel, Lawrence R., 32, 61n22,429n2, 453n45,456 Mohr, Michael F.,2t, 32.65n25, 93, 95, loo, 373,379 Moorsteen, R. H., 227 Momson, Steven A,, 388, 396-97, 399t Mukerji, Swati, 346, 355 Mulligan, J. G., 169 Murname, R., 314n15 Murphy, K., 211, 303nl Murphy, M., 307nn6,7 Murray, R., 518 Nadiri, M. Ishaq, 372n4, 382n20, 387n24, 401, 423,426 NAEYC (National Academy of Early Childhood Education), 35 1 National Center for Education Statistics, 305115 National Center for Health Statistics, 308 Nelson, Forrest D., 411n56 Newhouse, J. P., 345, 352n22, 354n27 Nicholson, J. L., 354n27 Niehans, Jurg, 289 NIPA (National Income and Product Accounts), 374116 Nooteboom, B., 176 Nordhaus, W. D., 12n6, 307, 315 North, Douglass C., 161 Norwood, Janet L., 122 NYSE (New York Stock Exchange) Fact Book, 196n5 Oates, W. E., 34 Offenbacher, Edward K., 292n3 Office of Management and Budget (OMB), 35118, 67n28 Ohlsson, I., 517, 520 Oi, Walter Y., 182n20, 183n22, 286,424 Olley, Steve, 432n11 O’Neill, D. M., 305n5 Orgler, Y. T., 300119 Otto, Phyllis F., 78 Pakes, Ariel, 4321111 Palmer J., 351nn14,17,18, 352n20 Parker, Robert P., 175, 373, 379 Pascoe, George, 4321110

Pashigian, B. Peter, 1771118, 182n21 Pasurka, Carl, 257,426 Paul, Jonathan, 209-10 Pesek, Boris P., 289 Peskin , J., 3 14nn19,20 Pfleiderer, P., 203.2051112, 206n13, 210 Phillips, D., 350, 396 Pollak, Robert A,, 110 Posner, R. A., 177n18 Prescott, E. C., 14 Primont, D., 230 Progressive Grocer, 178t, 182n20, 183 Radner, R., 315n21 Rangan, Nanda, 257 Ratchford, Brian T.,165 Reinsdorf, M., 11115 Revankar, N., 355 Ricardo, David, 219 Robinson, E. A. G., 169 Roos, P., 230n14 Rose, John T.,266n27 Rosen, S., 20, 303n1, 314n20 Ross, J. P., 344113 Ross, Stephen, 202118 Ross, Tom, 188 Ruggles, Richard, 289 Ruopp, R., 346n7, 347, 351n16.352n19 Salgupis, Agis, 41 1 Samuelson, P. A,, 14 Santomero, Anthony, 286n4 Saving, Thomas R., 289 Schofield, J. A,, 345115 Schultz, T. W., 303, 314n20 Schwartz, James, 4091152 Schwartzman, D., 1771117 Seabright, Paul, 387n24 Sealey, S. Calvin, 247, 284 Searle, A. D., 344n3, 345n4 Securities and Exchange Commission Annual Report, 195n1, 196n3 Securities Industry Association, Securities Industry Yearbook, 196n5 Sepielli, P., 305115 Shephard, R. W., 227-28 Sheshinski, E., 169n9 Shleifer, Andrei, 214n18 Sickles, Robin C., 372n4, 382n20, 387n24, 401,423,425-26 Siegel, Donald, 14, 442n28 Siegel, Irving H., 75

551

Author Index

Sinclair, J. H., 133 Singleton, Kenneth, 297n2 Sjoblom, K., 3121114, 314n22 Smith, Dale, A,, 135t Smith, Henry, 161 Sommers, John P., 123 Spady, Richard H., 41 In56 Spence, A. Michael, 202118 Stafford, F. P., 307n6 Stallings, J., 346, 3511113 Steiner, P. O., 186n24 Steiner, Robert L., 187, 189 Stevenson, Rodney E., 256 Stiglitz, Joseph, 200, 203-4, 2061113, 210 Sturm, P., 130 Summers, Robert, 494111,495,497,499, 501-3, 509,515 Sunga, Preetom S., 289 Survey of Current Business, 8-9, 26, 34n7, 611123, 67n27 Swanson, Joseph A,, 409-1 1 Syrquin M., 169n9 Tajan, M., 464116 Timme, Stephen G., 246, 254 Tobin, James, 12n6, 202, 211, 307, 315 Towey, Richard E., 289 Trajtenberg, Manuel, 456n50 Tretheway, Michael W., 100, 372n4, 381-82, 385, 393n36,406 Triplett, Jack, 65n25, 133, 284-85, 288111, 294, 384,412,430n6 lblly, Shawn, 4091152 Umbeck, I., 189n29 United Nations, 34n6,40n14, 289,494nl United Nations Statistical Office (UNSO), 220n3 U.S. Bureau of Labor Statistics (BLS), 251, 374n5,528n2 Handbook of Methods, 1 13, 1 16, 124 U.S. Bureau of the Census, 45On41 Statistical Abstract of the United States, 195111 U.S. Department of Commerce, 346 U.S. Department of Labor, 3491110, 350111 1, 352, 353t U.S. General Accounting Office (GAO), 3511115, 3551131 Usher, Dan, 102nl,400,505

Van Home, J. C., 225118 van Tulder, F., 528112 Vigdor, William R., 41 In56 Vishny, Robert W., 214n18 Waite, C. A., 344113, 345n4 Waite, Preston J., 450nn41,42 Waldorf, William H., 374n5 Walker, Michael A , , 496 Wallis, John J., 161 Warburton, Clark, 289 Waverman, Leonard, 382n20, 383,425 Webb, R. H., 308n8, 3151122 Weber, W., 223 Weisbrod, B. A,, 308118 Welch, Finis, 303nl Welsch, R. E., 231, 2381123 Weyback, D., 112 White, Lawrence J., 385 Whitin, T. M., 170 Wilcox, M., 346, 351n13 Williamson, 0. E., 168 Willis, R. J., 16 Wilson, R., 202118 Windle, Robert J., 372n4 Winsten, C., 165n5, 175 Winston, Clifford, 372n4, 380n12, 388, 39697, 399t, 411-12,425 Witte, Ann Dryden, 346, 355 Wittink, D. R., 520 Wolf, E. N., 3 Wolfe, B. L., 304n3 Wood, David O., 143 Wyckoff, Andrew, 32,61n22 Wykoff, Frank C., 284-85 Yaisawamg, S., 2301114,426 Ying, John S., 380n14,411n56 Yntema, Dwight B., 289 Young, Allan H., 65 Yun, K.-Y., 307, 3081111, 3091112 Zaslow, M., 351nn14,17,18, 352n20 Zeckhauser, R., 202n8 Zellner, A., 355 Zieschang, Kimberly D., 221-22, 227, 2281112, 230n14, 239, 248-49, 250n6, 252, 253n15, 263, 283-84, 292, 293n4

This Page Intentionally Left Blank

~~~

Subject Index

Air transportation industry: ALP growth in, 385; effect of deregulation on, 386-95, 419; hedonic regression analysis for fares, 132-35; input quality and quantity, 400-408; new measures for, 381-82; quality of, 394-97; value of time in, 397-400 ALP. See Average labor productivity Annual survey of manufactures (ASM), 432, 435,449 Asset approach (bank output), 247-48, 29091 Automobile insurance, CPI pricing of, 13539, 156-57 Automobile leases, 143-46 Automobile rental market, 142-46 Average labor productivity (ALP), 376; comparison of NIPA and BLS measures, 374-81; growth in airline industry of, 385; growth levels in transportation sector, 37 1-72; measurement for railroads, 408-1 1; transportation sector data, 37576 Banking: characteristics of services in, 28687; cost dispersion in, 254-56, 275; implicit and explicit revenues, 252-54; maximization of economic profit, 220; measurement in commercial, 245, 274, 287-88; pricing of services, 252 Bank output: disagreement in choice of, 252; in finance literature, 289; measures of,

553

273-74, 288-89, 293; methods to define, 246-54; production function approaches to, 289-92 Bank production model, 226-41. See also Distance function; Opportunity cost rate BEA. See Bureau of Economic Analysis (BEN BLS. See Bureau of Labor Statistics Blue Cross/Blue Shield data, 129-30 Bureau of Economic Analysis (BEA): capital measurement project of, 401-8; comparison with BLS output measures, 95-100; criticism of services measures of, 32-33; GPO estimates of, 25-26, 32, 34-39, 61-67, 69-70; improvements in GNP estimates, 42-61; improvements in GPO data, 61-65; measure of real day-care expenditures, 346-47; measures show industry contribution to GNP, 93; NIPA of, 4,25, 314, 374-81,418; transportation measures of, 412-15 Bureau of Labor Statistics (BLS): ALP measures of, 374-76; comparison with BEA output measures, 95-100; criticism of productivity measures and response by, 101-6; health insurance data of, 129-31; output and employment measures of, 374-82, 41 8; physician service indexes, 117-18; service sector measures of, 7889; trucking industry data of, 41 1-12. See also Consumer price index Bureau of the Census: data used by BEA, 65-

554

Subject Index

Bureau of the Census (continued) 66; data used by BLS, 75-76. See also Census of manufactures (COM): Annual survey of manufactures (ASM) Call report data, 279 CAOs. See Central administrative offices (CAOs) Capital gains rate (in user cost price), 220 Capital input: air transportation industry, 400-408; measures of, 412 Capital market imperfections, 222 Category price relatives, 495. See also Country-product-dummy method (CPD); E-K-S method CE. See Consumer expenditure survey (CE) Census of manufactures (COM), 432,44245,449. See also Annual survey of manufactures (ASM) Central administrative offices (CAOs), 44244 Child-care industry. See Day-care industry Child-rearing investment, 304, 317 COL. See Cost of living index (COL) COM. See Census of manufactures (COM) Commodity production, 3, 496 Communication services: BLS productivity measures of, 81; productivity growth in, 91-92 Competition: among banks, 235-36, 284; effect on store size and market concentration, 175-80; full marginal cost in monopolistic, 173-74; in securities markets, 198-99 Computer investment: data for manufacturing sector, 442-45; effect on measurement of manufacturing input, 430-31.459 Consumer: inventory model of demand, 17073; may decide quality of service, 5-6; price of time expenditure by, 157, 317; role in distributive trade changes, 16880, 185-88, 192. See also User cost Consumer demand full-price model, 170-73 Consumer expenditure survey (CE), 113-16, 123, 140, 143, 145-46, 349 Consumer price index (CPI): home ownership measurement in, 121-22; improvements in measuring consumer services, 13246; improvements in service sector items, 124-32; measures price changes, 1 1 1; quality in measures of, 11 1-13; quality of CPI series, 13; relation to

cost-of-living index, 110-1 1; rentalequivalence technique in, 122-24, 139, 142-46; service items in, 116-21, 12432; shift in services component of, 109; in United Nations International Comparison Project, 495; value of consumer service in, 142-46; weighting and item structure of, 113-15. See also Consumer expenditure survey (CE); Expenditure classes (ECs); Entry-level items (ELIs); Point-of-purchase survey (POPS) Consumer services: developing hedonic models for, 134; future improvements in measurement of, 140-46; measuring user cost of, 142 Consumption, public, 519-21 Cost-function approach (MFP growth), 38283 Cost function (bank), 257 Cost of living index (COL), 110-1 1 Country-product-dummy method (CPD), 495, 497 CPI. See Consumer price index (CPI) Current point-of-purchase survey (CPOPS), 146 Data envelopment analysis (DEA), 257 Data sources: for analyzing real output of banks, 221; annual survey of manufactures (ASM), 432,435,449-50; for bank production model, 227, 231, 279; BLS examination for deficiencies, 7678; for capital input in airline industry, 401; census of manufactures (COM), 432, 449-50,456; for CPI health insurance pricing, 129-32; criticism of BLS collection for, 103; for day-care output analysis, 347-50, 363-66; FCA (functional cost analysis) program, Federal Reserve system, 250-5 1,253-54; for foreign outsourcing, 445, 447; for industry output measurement, 75-76; longitudinal research data base (LRD), 431-32, 451; to measure lifetime labor income, 305; measurement errors in, 26, 69; need for expanded data collection by, 66-67; NIPA and BLS discrepancies, 374-8 1 ; for pricing of insurance, 137-38; problem of labor input, 77-78; for productivity in transportation sector, 374-8 1 ; quality of CPI series, 13; segments of service sector requiring improvement,

555

Subject Index

70; for services in ICP, 495-96; for transportation sector analysis, 374-82 Day-care industry: adjustment of output for quality, 358-60; output measurement for, 344-47; pricing of services in, 354; quality analysis of, 350-61, 366-67 DEA. See Data envelopment analysis (DEA) Decontrol Act (1981), 284 Deregulation: of bank deposits, 252, 263; effect on airline industry, 408, 419; effect on banking, 235-36, 245-46, 27lt, 273t, 275-77, 284; effect on stock exchanges, 198; effect on transportation industry, 380,411-12,416,418 Diagnosis-related groups (DRGs), 140-41 Disaggregation method (in CPI construction), 118-21, 149-50 Dispersion: of cost in banking, 254-56, 275; of productivity and profitability in French firms, 474-83 Distance function: in bank output analysis, 227-29; estimating, 230-3 1; use for estimates of, 221-22, 242 Distributive trades: development of, 165, 168; differences in production function of, 188-89; growth of, 161-65; measuring services of, 193; productivity change in, 188-89; quality and composition of workforce in, 162-65. See also Grocery stores; Middlemen; Retail firms; Supermarkets Double deflation procedure, 42-50,62, 93, 95 DRGs. See diagnosis-relatedgroups (DRGs) Durable good theory: in automobile purchase, 142; for home ownership, 121-22. See also Flow-of services approach Economic growth measurement, 3 14 Economic Statistics Initiative (1992), 67 Economies of massed reserves, 62; for retail firm production function, 169, 192-93 ECs. See Expenditure classes (ECs) Education sector: differences in cost- and income-based estimates of, 315-16; estimates of investment in, 310-15; human capital as product of, 303; international comparisons of spending for, 499-501, 514-15; measuring output of, 317-20; productivity measurement for, 3 16; value in current and constant prices for, 328t, 329t. See also School enrollment

E-K-S method, 495,497 Electric, gas, and sanitary services: BLS productivity measures for, 75, 81-83; productivity growth in, 91-93 ELIs. See Entry-level items (ELIs) Employee hours index, 74-76 Employment data: of BLS and NIPA for transportation sector, 379-80; of Bureau of the Census, 76; requirement for improvement in, 71 Entry-level items (ELIs), 113-16, 118-20 Expenditure classes (ECs), 116-18 Externalities (banking), 245-46 FAA. See Functional cost analysis (FCA) program, Federal Reserve System Fertility behavior, 316-17 Finance, insurance, and real estate service measure, 85, 87-89 Financial firm model, 223-26 Financial services, 220, 222-23; measurement and pricing of input and output of, 296-300. See also Banking Financial services bundle, 220, 224-26 Flow-of-servicesapproach, 122-24, 139, 142-46. See also User cost; Rental equivalence Frontier cost function, 256-57. See also Thick-frontier cost function Full marginal cost (FMC), 173-74 Full price: model of consumer demand, 17073; supermarket, 186 Functional cost analysis (FCA) program, Federal Reserve System, 250-51, 253-54, 279 Government services: international comparison of spending for, 499-502, 514-15; measurement of output of, 519-27 GPO estimates. See Gross product originating (GPO) estimates Grocery stores, 180, 182; changes in output mix, 185-88; composition of work force, 182-85 Gross national product (GNP): current- and constant-dollar industry estimates of, 39-62; evaluation of source data for, 26; growth of service industry share in, 2730; recalculation of Swedish, 537-39. See also Gross product originating (GPO) estimates

556

Subject Index

Gross product originating (GPO) estimates: current-dollar, 34-39; improvement program, 61-67, 69-70; issues in, 25-26; services industries in, 26-32 Growth accounting method, 453-54 Health Care Financing Administration (HCFA), 129-30 Heath insurance: CPI pricing method for, 124, 129-32, 156-57 Hedonic regression analysis: for hospital and physician services, 139-40; to measure quality changes in CPI, 132-34, 146 Heterogeneity: of bank loans, 294; in productivity and profitability in French firms, 474-83; in services sector, 3, 7, 19 Home ownership: BLS changes to component for, 122-23, 146; durable good theory in CPI, 121-22. See also Rental equivalence Homogeneity, in services sector, 7 Housing, rental value, 502-3 Human capital: differences in cost- and income-based approaches to, 3 15-16; education as investment in, 303-4; investment in other forms, 304, 316-17; measurement of investment in, 305, 314; as output of education, 39 ICP. See United Nations International Comparison Project (ICP) Imputation procedure. See Linking procedure Indexes: for air transportation, 81f; BLS service sector, 75-76, 117-18; for gas and electric utilities, 82, 83f; of labor productivity, 74-76; measurement basis for output, 75-76; quality, 355, 362; quantity, 242, 298; superlative 64-65, 69. See also Output indexes Industry GNP estimates. See Gross product originating (GPO) estimates Industry measures of productivity, 73-74 Inefficiency: approaches to measuring, 25666; effects of changes in, 274-75; possible explanations of bank, 281-83. See also Data envelopment analysis (DEA); Thick-frontier cost function Information: bank’s role in providing, 222; outputs of stock exchanges related to, 197-98; value in managerial decision making, 21 1-14; value in stock prices, 200-1 1

Infrastructure investment, 419 Input: BEA improvements in estimates for, 51-52;.BLS commercial banking, 88; data problems for labor, 77-78; to educational system, 304; factors influencing measurement of manufacturing, 429-33; for financial, insurance, and real estate, 85, 87-89; growth at stock exchanges member firms, 196-97; for trade sector, 82-85. See also Capital input Input, intermediate: deflators and estimates of, 51-52; service sector share of, 27, 30-32 Input measures: BLS automotive repair shops, 88-89; BLS communications, 81; BLS electric, gas, and sanitary services, 81-82; BLS retail trade, 85; BLS transportation industry, 80-81 Input-output: analysis for financial firm model, 223-26; classification of firm’s, 101-2; of banks, 284-85, 287-88.296; requirement for adjustment in tables for, 70-71 INSEE. See National Institute of Statistics and Economic Studies (INSEE), France Interest rate (in user cost price), 220 Intermediation services, bank, 247 International Comparison Project (ICP). See United Nations International Comparison Project (ICP) Inventory costs, 170-71, 179 Inventory model of consumer demand, 17073 Labor activities, 312-14, 325-27t Labor activities, market: current and constant dollar value for, 306, 324-25t; in estimates of educational investment, 31 1; in estimates of human wealth value, 31314, 338t. See nlso Labor (or work) time Labor activities, nonmarket: allocation of time among, 307; comparison of two estimates of, 315, 338t; in estimates of educational investment, 31 1-12; value in current and constant prices for, 304-5, 307-8, 316, 326-27t Labor income, 307-9. See also Lifetime labor income Labor (or work) time, 306-8 Labor productivity: effect on size and focus of retail firms, 185-88; growth in distributive trades of, 165-67; positive relation

557

Subject Index

to store size, 175-77; in service sector, 3. See also Average labor productivity (ALP); Multifactor productivity (MFP) Labor productivity measures: BLS bank, 273-74; computation of, 73-76; criticism of BLS, 101-6; as information source, 104 Lifetime labor income: measurement of, 3059; and educational attainment, 310, 315 Linking procedure, 112-13, 132 Longitudinal research data base (LRD), 43132 Malmquist quantity index, 298 Managerial compensation model, 201-4 Marginal holding cost function (MHC), 171 Measurement problems: in commercial banking, 245, 274-75; in financial services, 220, 223 Medical services: alternatives to pricing individual, 140-42; international comparisons for spending on, 499-501; measuring price and quality of, 139-40, 156; pricing across countries, 500-501; pricing alternatives for, 140-42; pricing in CPI of, 156-57 Middlemen: function of, 161-62; increased specialization of, 164-65 Multifactor productivity (MFP): computation of capital quality changes, 403-8; conceptual issues for, 382-85; development of BLS measures of, 104; growth of, 372, 384-85; indexes for trucking industry, 412-15; measures for railroads, 4081 1 ; new BLS and NIPA measures for, 381-82; produced by Bureau of Labor Statistics, 74; and transportation growth, 403-8,415-19 National Association for the Education of Young Children (NAEYC), 344, 351 National income and product accounts (NIPA), 314; measures of, 374-82; output revisions, 418; transportation sector data of, 375-76 National Institute of Statistics and Economic Studies (INSEE), France, 461 Not-for-profit organizations (NPOs), 346, 354-55 NPOs. See Not-for-profit organizations (NPOs)

Operating manufacturing establishments (OMEs), 44-44 Opportunity cost of capital (in user cost price), 220, 297-98 Opportunity cost rate: in differing accounting systems, 221; modeling of, 229-30; in user cost formula, 241-42 Output: adjustment for quality in day-care industry, 358-61; air transport quality of, 394-408; asset approach to measuring and defining bank, 246-48, 290-91; of bank financial services, 222-26; definition of service, 496; deflation for service sector in BEA data, 42-5 1; effect of hubbing on air transport service quality, 386-94; improvement of gross estimates of BEA, 43; intermediate (for railroads and trucking), 4 1 1; measurement of educational system, 304-5; measurement of final, 189; quality changes in airline industry, 386-97; of securities markets, 197-202; user cost approach to measuring bank, 248-50, 291; value-added approach to measuring bank, 250-54, 274, 290,298-99 Output indexes, 74-76, 81-84 Output measures: of BLS and BEA, compared, 93, 95-100; BLS communications, 81; BLS electric, gas and sanitary, 8 1-82; BLS fire, insurance, and real estate, 85-88; BLS retail trade, 82-85; BLS service sector, 88; BLS transportation series, 80; for trucking industry, 411-15 Outsourcing: effect on measurement of manufacturing input, 429-30; measurement of, 453-56; proxies to measure manufacturing sector, 444-49 Physician services index, 117-18 PMOs. See Profit-making organizations (PMOs) Point-of-purchase survey (POPS), 115-16. See also Current point-of-purchase survey (CPOPS) Poisson distribution, services sector, 169-70 PPPs. See Purchasing power parities (PPPs) PPS methodology. See Probability proportionate to size (PPS) Price: coverage in services sector for, 6-7; effect in retail firms of competition in, 175-80; measurement of medical ser-

558

Subject Index

Price (continued) vice, 156-57; measuring consumers’ time, 157; of services in international comparison, 497-98. See also Full price; Linking procedure Price changes: comparing new service with old, 132-34; in CPI, 111-13, 123; improvements in CPI measures for, 146; measurement in services sector of, 109; selection of items in CPI for, 116-21 Price indexes, 70 Pricing: of automobile leases, 145-46; of bank financial services, 222-26; BLS and CPI methods for health insurance, 129-32, 156-57; in CPI of automobile and tenants’ insurance, 135-39, 156-57; in monopolistically competitive store, 173-75; of service items in CPI, 116-21 Probability proportionate to size (PPS), 114115, 117 Producer price index (PPI), 6 Productivity: BLS measures of, 73-74, 7889; causes for changes in transportation, 383-85; changing levels of, 68; in France, 474-83; growth levels of, 20, 25; industry measures of, 73-74; in services, 1, 3 , 6 , 25, 68-69; in Sweden, 528-39. See also Average labor productivity (ALP); Multifactor productivity (MFP); Total factor productivity (TFP) Productivity measures: of Bureau of Labor Statistics, 73-74, 85, 87-89, 93; characteristics of, 76-78; criticism of BLS, 101-2; labor productivity measures in BLS program, 104; service industries covered by BLS, 79-80t. See also Labor productivity measures Profitability, 474-83 Profit-making organizations (PMOS), 346, 354-55 Purchasing power parities (PPPs), 496-503 Quality: in air transport output, 394-408; dependence of service activity on, 5-6; indicators in airline industry of, 394-400; measurement for Swedish public sector output analysis, 525-27; productivity effects of air transport hubbing, 386-94; of replacement services, 111; of retail work force, 164; specification in international comparison, 514-515

Quality change: and CPI, 11 1-13, 139-40; in health insurance data, 131-32; hedonic regression analysis to measure, 132-40; methods for, 112-13; in multifactor productivity analysis, 383-85. See also Linking procedure Quality index, day-care industry, 355, 362 Quantity indexes, Tornqvist, 242 Queueing theory, 192-93 Railroads: measurement of ALP and MFP in, 408-1 1; new measures for multifactor productivity in, 381-82 Ramsey rule, 174 Regulation: of day-care centers, 351; effect on bank activities, 222-23, 245; protection mechanism of, 281-82. See also Deregulation Rental equivalence: automobile leases, 14346; cross-country comparison of, 502-3; measurement to separate service from asset, 142-46; measure of home ownership, 122-24, 146 Replacement services, 11 1 Reserve requirement (banks), 223 Retail firms: BLS output index for, 75; economies of massed reserves in, 169-70, 192-93; output of, 165-68; production function of, 168-69. See also Grocery stores; Supermarkets Revenues, bank (implicit and explicit), 25254 School enrollment, 312, 323t Securities markets, 196-200 Service bundles: changing structure of, 162; in financial services, 220, 224-26; of retail firms, 168, 188 Services sector: BLS productivity measures for, 88-90; consumers’ time in measuring price of, 157; contribution to productivity, 68-69; criticisms of BEA measures of, 32-33; defined, 4-6,494, 496, 504-7; differences in data among countries for, 503-14; exit and entry in France of, 483; factors influencing manufacturing sector input, 429-33; in GPO estimates, 26-32; homogenous and heterogenous, 7 , 19; ICP compared to national accounts, 504; measurement of productivity and price in, 6-7, 157;

559

Subject Index

measurement of industries in, 78-79, 344-45; methods to separate asset from, 142; nonpriced, 499-501,514; output and production, 496; problems in crosscountry comparisons of, 500-503; slow growth in, 1, 3-4; in United Nations International Comparison Project, 496503. See also Banking; Retail firms; Stock exchanges; Transportation sector Shadow price, and opportunity cost rafe, 22122, 228-29, 242. See also Distance function Shephard-Hotellinglemma, 228 SIC codes. See Standard industrial classification (SIC) code changes SIPP data. See Survey of income and program and participation (SIPP) data Social benefit analyses, 523-24 Standard industrial classification (SIC) code changes, 76 Stock exchanges, 195-96, 198. See also Information; securities markets; Stock prices Stock market: with fully rational price, 2014; model of information gathering and trading, 200-201, 204-7 Stock prices: managerial compensation based on, 201-4; value of information in, 20021 1 Superlative indexes, 64-65, 69 Supermarkets: changes in output, 188; factors contributing to increase in, 175-88. See also Grocery stores Survey of income and program and participation (SIPP) data, 347 Surveys: CPI consumer expenditure, 113-15; CPI current point-of-purchase survey, 146; CPI point-of-purchase, 115-16; French system for services, 463-66 Teachers (cross-country comparison), 501 Technology: effect on stock exchanges of new, 198; impact on retail firm functions and size, 162, 168-70, 179-80; innovation in goods-producing industries, 188 Tenants’ insurance, 135-39 TFP. See Total factor productivity (TFP) Thick-frontier cost function: estimates and benefits of, 257-58; shift over time in, 266-75; specification of, 258-59 Time: allocation of, 307; price of consumer

expenditure of, 157; as quality attribute in airline industry, 397-400; of shopper in distributive trades, 192; value of market and nonmarket, 304-5, 307-8, 31214, 316. See also Labor (or work) time Tornqvist indexes, 74, 242, 298-99 Total factor productivity (TFP), 165, 188; calculation of growth, 453-54, 458-59; in distributive trades, 165; measurement errors in growth, 433-35; measuring, 1013; relationship between growth and measures of service-sector inputs, 454-56; usefulness of measures, 101; value of measures of, 101-02 Trade sector: BLS productivity measures in, 82-87; productivity growth in, 91-93 Training, on-the-job, 317 Transaction costs: approach in analysis using, 286; bank role to reduce, 222 Transportation sector: BLS productivity measures of, 78-81; flow-of-servicesin CPI, 142-46; growth of output per hour, 373t; new multifactor productivity measures for, 381-82; productivity growth in, 70, 91-92; value of consumed, 142-46. See also Air transportation industry; Automobile rental market; Railroads; Trucking industry Trucking industry: measurement of output and employment in, 41 1-15; new measures for, 381-82; nominal and real output data, 379t Unit costs (bank), 279-81 United Nations International Comparison Project (ICP), 494-500 United Nations system of national accounts (SNA) convention, 497 User cost: components in financial services of, 220; as price of financial services, 224; methodology, 142-46, 248-50, 291-92. 297 Value: of air travel time, 397-400; of consumed transportation services, 142-46; for education, 315, 328-29t; estimates of wealth, 313-14; 336-337t; of information, 200-11; of market and nonmarket labor activities, 306-8, 313-14, 316, 324-27t, 338t; as measure of quality,

560

Subject Index

Value (conrinued) 397-400; of on-the-job training, 317; of rental housing, 502-3 Value-added approach, 69-70, 93, 250-54, 274, 290, 298-99 Wealth, human: definition and estimates of, 313, 334-3%; estimates based on lifetime labor incomes, 313-14, 338t; values per person in current and constant prices, 313, 336-37t

Weights: for analysis of Swedish public sector output, 523-25; in BLS trade sector measures, 82-84; in CPI construction, 11, 113-15, 123, 129, 135-36;forexpenditure in CPI, 135; in flow-of-services approach, 145-46; in shift from direct to indirect pricing, 138-39; use in BLS measures of. 107

This Page Intentionally Left Blank