The Economics of Climate Change: Adaptations Past and Present 9780226479903

Citation preview

The Economics of Climate Change

A National Bureau of Economic Research Conference Report

The Economics of Climate Change Adaptations Past and Present

Edited by

Gary D. Libecap and Richard H. Steckel

The University of Chicago Press Chicago and London

GARY D. LIBECAP is the Donald Bren Distinguished Professor of Corporate Environmental Management and professor of economics at the University of California, Santa Barbara, and a research associate of the National Bureau of Economic Research. RICHARD H. STECKEL is the Social and Behavioral Sciences Distinguished Professor of Economics, Anthropology, and History and a Distinguished University Professor at Ohio State University, and a research associate of the National Bureau of Economic Research.

The University of Chicago Press, Chicago 60637 The University of Chicago Press, Ltd., London © 2011 by the National Bureau of Economic Research All rights reserved. Published 2011. Printed in the United States of America 20 19 18 17 16 15 14 13 12 11 1 2 3 4 5 ISBN- 13: 978- 0- 226- 47988- 0 (cloth) ISBN- 10: 0- 226- 47988- 9 (cloth)

Library of Congress Cataloging-in-Publication Data The economics of climate change : adaptations past and present / edited by Gary D. Libecap and Richard H. Steckel. p. cm. — (National Bureau of Economic Research conference report) ISBN-13: 978-0-226-47988-0 (cloth: alk. paper) ISBN-10: 0-226-47988-9 (cloth: alk. paper) 1. Climatic changes—Economic aspects—Congresses. 2. Climatic changes— Economic aspects—United States—Congresses. 3. Ecology— Economic aspects—Congresses. 4. United States—Environmental conditions—Economic aspects—Congresses. I. Libecap, Gary D. II. Steckel, Richard H. (Richard Hall), 1944– III. Series: National Bureau of Economic Research conference report. HC79.E5E2778 2011 333.71⬘4—dc22 2010037305 o 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 Z39.48- 1992.

National Bureau of Economic Research Officers John S. Clarkeson, chairman Kathleen B. Cooper, vice-chairman James M. Poterba, president and chief executive officer Robert Mednick, treasurer

Kelly Horak, controller and assistant corporate secretary Alterra Milone, corporate secretary Gerardine Johnson, assistant corporate secretary

Directors at Large Peter C. Aldrich Elizabeth E. Bailey Richard B. Berner John H. Biggs John S. Clarkeson Don R. Conlan Kathleen B. Cooper Charles H. Dallara George C. Eads

Jessica P. Einhorn Mohamed El-Erian Jacob A. Frenkel Judith M. Gueron Robert S. Hamada Peter Blair Henry Karen N. Horn John Lipsky Laurence H. Meyer

Michael H. Moskow Alicia H. Munnell Robert T. Parry James M. Poterba John S. Reed Marina v. N. Whitman Martin B. Zimmerman

Directors by University Appointment George Akerlof, California, Berkeley Jagdish Bhagwati, Columbia Glen G. Cain, Wisconsin Alan V. Deardorff, Michigan Ray C. Fair, Yale Franklin Fisher, Massachusetts Institute of Technology John P. Gould, Chicago Mark Grinblatt, California, Los Angeles

Marjorie B. McElroy, Duke Joel Mokyr, Northwestern Andrew Postlewaite, Pennsylvania Uwe E. Reinhardt, Princeton Nathan Rosenberg (Director Emeritus), Stanford Craig Swan, Minnesota David B. Yoffie, Harvard

Directors by Appointment of Other Organizations Bart van Ark, The Conference Board Jean-Paul Chavas, Agricultural and Applied Economics Association Martin Gruber, American Finance Association Arthur B. Kennickell, American Statistical Association Thea Lee, American Federation of Labor and Congress of Industrial Organizations William W. Lewis, Committee for Economic Development

Robert Mednick, American Institute of Certified Public Accountants Alan L. Olmstead, Economic History Association Harvey Rosenblum, National Association for Business Economics John J. Siegfried, American Economic Association Gregor W. Smith, Canadian Economics Association

Directors Emeriti Andrew Brimmer Carl F. Christ George Hatsopoulos Saul H. Hymans

Lawrence R. Klein Franklin A. Lindsay Paul W. McCracken

Rudolph A. Oswald Peter G. Peterson Nathan Rosenberg

Relation of the Directors to the Work and Publications of the National Bureau of Economic Research 1. The object of the NBER is to ascertain and present to the economics profession, and to the public more generally, important economic facts and their interpretation in a scientific manner without policy recommendations. The Board of Directors is charged with the responsibility of ensuring that the work of the NBER is carried on in strict conformity with this object. 2. The President shall establish an internal review process to ensure that book manuscripts proposed for publication DO NOT contain policy recommendations. This shall apply both to the proceedings of conferences and to manuscripts by a single author or by one or more coauthors but shall not apply to authors of comments at NBER conferences who are not NBER affiliates. 3. No book manuscript reporting research shall be published by the NBER until the President has sent to each member of the Board a notice that a manuscript is recommended for publication and that in the President’s opinion it is suitable for publication in accordance with the above principles of the NBER. Such notification will include a table of contents and an abstract or summary of the manuscript’s content, a list of contributors if applicable, and a response form for use by Directors who desire a copy of the manuscript for review. Each manuscript shall contain a summary drawing attention to the nature and treatment of the problem studied and the main conclusions reached. 4. No volume shall be published until forty- five days have elapsed from the above notification of intention to publish it. During this period a copy shall be sent to any Director requesting it, and if any Director objects to publication on the grounds that the manuscript contains policy recommendations, the objection will be presented to the author(s) or editor(s). In case of dispute, all members of the Board shall be notified, and the President shall appoint an ad hoc committee of the Board to decide the matter; thirty days additional shall be granted for this purpose. 5. The President shall present annually to the Board a report describing the internal manuscript review process, any objections made by Directors before publication or by anyone after publication, any disputes about such matters, and how they were handled. 6. Publications of the NBER issued for informational purposes concerning the work of the Bureau, or issued to inform the public of the activities at the Bureau, including but not limited to the NBER Digest and Reporter, shall be consistent with the object stated in paragraph 1. They shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The Executive Committee of the Board is charged with the review of all such publications from time to time. 7. NBER working papers and manuscripts distributed on the Bureau’s web site are not deemed to be publications for the purpose of this resolution, but they shall be consistent with the object stated in paragraph 1. Working papers shall contain a specific disclaimer noting that they have not passed through the review procedures required in this resolution. The NBER’s web site shall contain a similar disclaimer. The President shall establish an internal review process to ensure that the working papers and the web site do not contain policy recommendations, and shall report annually to the Board on this process and any concerns raised in connection with it. 8. Unless otherwise determined by the Board or exempted by the terms of paragraphs 6 and 7, a copy of this resolution shall be printed in each NBER publication as described in paragraph 2 above.

Contents

1.

2.

3.

4.

5.

Foreword

ix

Climate Change: Adaptations in Historical Perspective Gary D. Libecap and Richard H. Steckel

1

Additive Damages, Fat-Tailed Climate Dynamics, and Uncertain Discounting Martin L. Weitzman

23

Modeling the Impact of Warming in Climate Change Economics Robert S. Pindyck

47

Droughts, Floods, and Financial Distress in the United States John Landon-Lane, Hugh Rockoff, and Richard H. Steckel The Effects of Weather Shocks on Crop Prices in Unfettered Markets: The United States Prior to the Farm Programs, 1895–1932 Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

73

99

Information and the Impact of Climate and Weather on Mortality Rates during the Great Depression 131 Price V. Fishback, Werner Troesken, Trevor Kollmann, Michael Haines, Paul W. Rhode, and Melissa Thomasson

vii

viii

6.

7.

8.

9.

10.

11.

Contents

Responding to Climatic Challenges: Lessons from U.S. Agricultural Development Alan L. Olmstead and Paul W. Rhode

169

The Impact of the 1936 Corn Belt Drought on American Farmers’ Adoption of Hybrid Corn Richard Sutch

195

The Evolution of Heat Tolerance of Corn: Implications for Climate Change Michael J. Roberts and Wolfram Schlenker

225

Climate Variability and Water Infrastructure: Historical Experience in the Western United States Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe Did Frederick Brodie Discover the World’s First Environmental Kuznets Curve? Coal Smoke and the Rise and Fall of the London Fog Karen Clay and Werner Troesken

253

281

Impacts of Climate Change on Residential Electricity Consumption: Evidence from Billing Data 311 Anin Aroonruengsawat and Maximilian Auffhammer Contributors Author Index Subject Index

343 345 349

Foreword

Many economic historians and other social scientists have observed the climate change debate with dismay for its lack of attention to past experience and the diversity in the historical record. Climate change can be sudden, intense, and geographically focused, as in the Dust Bowl days of the 1930s, or gradual, irregular, and widespread as in the cooling of the Little Ice Age. Knowledge of the mechanisms of global climate change is far from complete, and forecasts of average temperature for the coming decades, much less centuries, are always hedged into alternative scenarios and wide confidence intervals. The place of regional climate change within the global system is heavily laden with complexity and seldom addressed by climate models. If climate change over the decades is very difficult, if not impossible, to predict at the geographic level where national or regional policy is made, it is prudent to investigate how the economy and the political system have responded to climate change in the past when even less was known about the physical system that determines temperature, precipitation, and the like. We, therefore, welcomed the interest of the National Bureau of Economic Research (NBER) in a project that views the past as a laboratory for understanding future scenarios when the economy must adapt to climate change. In this regard, we are grateful for the leadership and advice of James Poterba, president of the NBER, and Claudia Goldin, director of the program on Development of the American Economy, in support of this project. We also acknowledge the valuable organizational support of the Conference Department at the NBER. Study of climate change and its implications is inevitably an interdisciplinary effort, and the chapters in the volume benefited from the contributions, advice, and expertise of several researchers in the NBER program

ix

x

Foreword

on Environment and Energy, including Maximilian Auffhammer, Maureen Cropper, Olivier Deschênes, Michael Greenstone, Robert Pindyck, Wolfram Schlenker, Martin Weitzman, and Catherine Wolfram. Several researchers from the program on Development of the American Economy presented earlier papers, discussed others, or provided advice to authors, including Hoyt Bleakley, Karen Clay, Price Fishback, Michael Haines, Zeynep Hansen, Richard Hornbeck, Paul Rhode, Hugh Rockoff, Richard Sutch, Melissa Thomasson, and Werner Troesken. We also thank Melissa Dell, Haggay Etkes, Erica Field, Raghav Gaiha, Benjamin Jones, John Landon-Lane, Valerie Muellera, Cormac Ó Gráda, Benjamin Olken, Alan Olmstead, Daniel Osgood, Michael Roberts, Stephen Salant, James Stock, Daniel Sumner, and Hedrick Wolff, who also presented papers and discussed others at our conferences. Gary Libecap Richard H. Steckel August 3, 2010

Climate Change Adaptations in Historical Perspective Gary D. Libecap and Richard H. Steckel

News of global warming or climate change has inundated the public in recent years. Every major newspaper has published multiple editorials or op- ed pieces on the topic, the broadcast media regularly discuss the issue, and thousands of Web pages and blogs provide definitions and information and suggest causes and consequences of action or inaction. So why are we adding to congestion on the subject? As readers will see, analysis of the past has much to contribute, especially in understanding prospects of adapting to climate change, which has received relatively little study and comment relative to the standing it should take in the debate. At the outset, we note that the chapters in this book do not evaluate the science of climate change, a subject of media attention in recent months, because we do not claim the expertise to contribute on this matter. Instead, the chapters respond to proceedings of the past couple of decades, especially the efficacy of collective action in light of reports of the Intergovernmental Panel on Climate Change (IPCC). We perceive an imbalance in discussions of methods to cope with climate change relative to the viable options before the world community, given technologies now in use. The types of adaptation to climate that are described in this volume have received less attention in the economics literature than have policies for miti-

Gary D. Libecap is the Donald Bren Distinguished Professor of Corporate Environmental Management and professor of economics at the University of California, Santa Barbara, and a research associate of the National Bureau of Economic Research. Richard H. Steckel is the Social and Behavioral Sciences Distinguished Professor of Economics, Anthropology, and History and a Distinguished University Professor at Ohio State University, and a research associate of the National Bureau of Economic Research The authors thank David Stahle, Henri Grissino-Mayer, and two anonymous reviewers for comments and suggestions.

1

2

Gary D. Libecap and Richard H. Steckel

gation of emissions. Researchers have emphasized the design of optimal carbon taxes and implementation of cap and trade systems, including carbon credits. Because of the global nature of the greenhouse gas (GHG) externality, however, any effective mitigation effort necessarily requires international collaboration. Otherwise, emissions reductions in some regions will likely be offset by increases elsewhere, and the overall stock of GHG may continue to rise, generating further climate change. This point highlights the importance of international collective action to address climate change. The record, however, is not encouraging. The United Nations (UN) has been active on this front, beginning with the 1992 UN Conference on Environment and Development in Rio de Janeiro and the 1997 UN Framework Convention on Climate Change in Kyoto, followed by various UN Conferences of Parties in Buenos Aires, 2004; Bali, 2007; and Copenhagen, 2009. Yet there has been little concrete, coordinated global action to reduce emissions. Disputes over the science, over the magnitudes, nature, speed, and distribution of climate change across the planet, and, more critically, over the distribution of the costs of addressing it have blocked meaningful action. These problems of mobilizing collective action on climate change are similar to those encountered with other environmental or resource externalities. In fact, experiences in effectively confronting overharvest, overextraction, and excessive pollution in common pool fisheries, water resources, oil and gas reservoirs, and local air quality reveal that distributional conflicts are the norm. Disagreements over the size of the problems to be addressed and the assignment of the costs and benefits of mitigating it inhibit cooperation. Delay in response until a crisis occurs is the usual pattern. An emergency— the collapse of the resource stock or severe health consequences from air pollution, provide new information about the immediacy and size of the problem. This information causes the cost- benefit calculations to change, and distributional concerns become displaced by agreement on the need to address the problem. At that time, more effective collective action takes place (Libecap 2008). If this empirical pattern holds true for international climate change negotiations, as seems likely, then more attention should be placed on adaptation because meaningful global mitigation of GHG emissions is unlikely to occur before there is a major crisis to galvanize international collaboration. By that time, however, the stock of GHG may be of sufficient size that major climate changes will follow in any event. Accordingly, adaptation is key, and an understanding of the potential responsiveness of the economy is essential. Our volume addresses this pressing need. The empirical papers emphasize weather shocks rather than broader climate trends. Nevertheless, if the weather events examined here become more representative of patterns under climate change, then the analyses can provide useful insights into likely responses of the economy, governments, and society.

Climate Change: Adaptations in Historical Perspective

3

Although climate change is a global matter, by now readers will have surmised that the volume emphasizes analytical studies of the United States (eight out of the nine empirical papers). It would be desirable to incorporate the experiences of many other countries in the broader agenda, but we feel that the present focus is justified. The United States is a large continental country, which includes assorted ecological environments that represent or typify large areas of the world. Because the economy, methods of production, government policies, and so forth changed considerably over the past couple of centuries, far more diversity is represented in the chapters than suggested by studies of a single country in the contemporary world. In addition, researchers on other countries have undertaken very little work of the type represented in this volume. Therefore, we hope the approach will be imitated elsewhere; certainly the methods included here can be replicated for many other countries or regions of the world. The Record and Projections The Intergovernmental Panel on Climate Change (IPCC) is the most widely regarded source of information on climate change (http://www.ipcc .ch/index.htm). Created in 1988 by the World Meteorological Organization and the United National Environment Programme, the IPCC does not carry out its own original research or collect climate data, but its panels of experts and interested parties assess the risks and evaluate the implications of climate change based on their reading of the scientific literature. The IPCC is organized into three working groups, which study the Physical Science Basis of Climate Change; Climate Change Impacts, Adaptation, and Vulnerability; and Mitigation of Climate Change. Every few years, the IPCC publishes assessment reports, the fourth of which arrived in 2007 in the form of a synthesis and reports by each working group (Pachauri and Reisinger 2007). How diverse are weather patterns observed in the past one to two centuries around the globe, and what are the projections for the future? As we will see in the following, the diversity in the past at the regional level is considerable, relative to global patterns that we argue make it difficult to plan for future events at the geographic level of countries, where most decisions are made. Even if average global forecasts are reliable, those at the regional level are much less so, making coordinated international responses more difficult when national experiences may vary so much. National political leaders and their constituencies may find it hard to relate the conditions faced in their region to broader global weather patterns. The results suggest that adaptation should be part of any strategy to cope with climate change. Therefore, it is important to know the extent to which economic adaptation to weather changes has been successful in the past at the regional level. The most recent synthesis reports observed temperatures for the twentieth

4

Gary D. Libecap and Richard H. Steckel

Fig. I.1 Multimodel global averages of surface warming (relative to 1980–1999) under alternative scenarios Source: Pachauri and Reisinger (2007, 46).

century and gives projections up to 2100 (IPCC 2007, chapter 3). As one can see from figure I.1, the scenarios differ vastly depending upon the projected rates of economic growth, population increase, and technological change, all of which affect the concentration of greenhouse gasses.1 The IPCC makes no predictions about the likelihood of these various outcomes but the range of plausible possibilities is enormous, relative to average global temperature change in the preceding century. Various global warming skeptics have cast doubt on the historical temper1. A1 assumes very rapid economic growth, a global population that peaks in midcentury, and rapid introduction of new and more efficient technologies. A1 is divided into three groups that describe alternative directions of technological change: fossil intensive (A1Fl), nonfossil energy resources (A1T), and a balance across all sources (A1B). B1 describes a convergent world, with the same global population as A1, but with more rapid changes in economic structures toward a service and information economy. B2 describes a world with intermediate population and economic growth, emphasizing local solutions to economic, social, and environmental sustainability. A2 describes a very heterogeneous world with high population growth, slow economic development, and slow technological change. No likelihood has been attached to any of the Special Report on Emissions Scenarios (SRES). See IPCC (2007, chapter 3).

Climate Change: Adaptations in Historical Perspective

5

ature record and the models used to project the future. Steve McIntyre, for example, notes that an increasing share of temperature stations were located in urban areas, which retain heat and induce an upward bias to recorded average temperatures (McIntyre 2007). He has recalculated the temperature record using level 1 (excellent) or level 2 (good) reporting stations, finding that the warmest years were 1934 and 1921 followed by 1998 and 2006. If the commonly accepted temperature record is distorted, obviously projections based on this record will be deficient. We are not climate scientists, and here we do not engage the debate over the scientific plausibility of the reported temperature record and projections based upon it. We do know, however, that the climate system is enormously complex and difficult to model. Even if one believes the model, or a synthesis of models, widely accepted by climate scientists, the future is highly uncertain because many of the moving parts in these models are difficult to project. How reliably can we forecast economic activity, population growth, and technological change, for example? And especially the process of global political decision making that will influence these outcomes? In sum, it is prudent to act as if change is on the way, but we should prepare for many alternatives. Our capacity to adapt to climate change is even more important in light of past regional variability. The temperature record and the forecasts of figure I.1 appear placid because they average out or smooth local variation. Figures I.2 to I.5 disaggregate the record of temperature change from hemispheric to regional levels, and by comparing the charts by decreasing level of aggregation, one can see that fluctuations increase. Temperature change over the past century or so has been highly variable at the regional level around the globe. Mean annual temperatures in Australia, for example, have generally increased since 1910, but the amount of warming and the trends in maximum and minimum temperatures has been far from uniform across the continent. The largest increase in minimum temperature occurred in the Northeast, while the largest increase in maximum temperature took place in the Northwest. Experience also suggests that the effects of climate variability, and the corresponding need for adaptation, will be regional or local due to patterns of specialization induced by resource endowments, transportation links, and supplies of labor and capital. The forces that drove local climate variation in the past, whatever they may be, are likely to continue and will not be negated by other processes that provide the foundation for global warming. Therefore, the models required to forecast regional climates are far more complex and the projections more uncertain relative to those operating at the global level. Hundreds of regional projections exist, of course, but their standard errors are enormous. We argue that flexibility and adaptability are the best insurance for the planet and its various regions in our uncertain future. It, therefore, becomes important to know how successful we have been in this dimension in the past.

6

Gary D. Libecap and Richard H. Steckel

Fig. I.2

Annual mean temperature change for hemispheres

Source: http://data.giss.nasa.gov/gistemp/graphs/.

Methods of Measuring Weather Sources of weather data are fundamental for any study of the economics of climate change. Today, researchers have a vast array of instruments to measure weather conditions around the globe from satellites and weather stations based on land and sea and in the atmosphere. Unfortunately, the time- depth of these readings is inadequate for historical studies that reach back to the era when farming was a dominant source of national income, but sources are available to cover the earlier period. Understanding the practical limitations of empirical research requires a brief discussion of technology, concepts, private efforts, and government action that unfolded since the early 1700s. Instrument Readings Ideally, one would have a geographically dense array of comparable instrument readings that cover various aspects of weather over the past several centuries. Devices for measuring temperature and rainfall, however, were

Fig. I.3

Mean annual temperature change for three latitude bands

Source: http://data.giss.nasa.gov/gistemp/graphs/.

8

Gary D. Libecap and Richard H. Steckel

Fig. I.4

Mean annual temperature change in the United States

Source: http://data.giss.nasa.gov/gistemp/graphs/.

crude until the early eighteenth century, when Daniel Fahrenheit invented the mercury thermometer (1714) and Reverend Horsely developed the rain gauge (pluviometer) consisting of a funnel placed at the top of a cylinder (1722). In the 1720s, Anders Celsius assisted Erik Burman in recording temperatures in Uppsala, and soon thereafter observational sites appeared elsewhere in Sweden. Although the new technologies solved the problem of consistency, enabling people in distant places to take comparable readings, these early efforts were based on various thermometer scales that had to be converted into a common metric. Recordkeeping efforts were largely private or handled by scientific societies until well into the nineteenth century. Systematic study of weather patterns and their causes, much less climate change and its implications, could not begin effectively until a substantial body of evidence had accumulated. In the United States, the Smithsonian Institution took the lead in developing a weather network in 1849 based on 150 voluntary observers. In 1874, the task passed to the U.S. Army Signal Service, whose functions were transferred to the newly created U.S. Weather Bureau in 1891. By 1900, numerous countries had created national meteorological services.

Climate Change: Adaptations in Historical Perspective

Fig. I.5

9

Temperature trend in northeast Australia

Source: http://www.bom.gov.au/climate/change/retrends.shtml.

Concepts After data collection became routine, researchers suggested ideas for measuring drought (Heim 2002). In the early 1900s, drought was defined by a rule of thumb: twenty- one or more days with rainfall below one- third of normal. Thornthwaite (1931) proposed the concept of evapotranspiration or the sum of evaporation and plant transpiration, which led to ideas of water balance and alternative time scales. Palmer (1965) developed four measures: (a) a hydrological drought index based on measures of groundwater and stream flow with a horizon of 1 ⫹ years; (b) a drought severity index with a nine- month horizon; (c) a Z index that measures moisture anomaly in a particular month; and (d) a crop moisture index that monitors weekly conditions. Soon other measures appeared, such as the standardized precipitation index, a surface water supply index, and the vegetation condition index measured from satellite images. Several chapters in the volume analyze the Palmer Drought Severity Index (PDSI), which has a time scale useful for assessing the economic consequences of yearly fluctuations or clusters of annual patterns in effective moisture (precipitation minus losses from evapotranspiration). Its water balance equation includes rainfall, runoff, evaporation, transpiration, and soil recharge, which are converted to a scale of –6 (extreme drought) to

10

Gary D. Libecap and Richard H. Steckel

⫹6 (extremely wet), relative to normal or average conditions for a locality (Palmer 1965). It ignores stream flow, reservoir levels, and snowfall and is less useful in mountain areas or regions of microclimates (Alley 1984). It is fortunate for our purposes that the PDSI can be reconstructed from tree rings, which provide a chronology over hundreds, or even thousands, of years if especially long- lived trees are available (such as bristlecone pines) or preserved logs can be extracted from peat bogs or old buildings. Dendochronology Leonardo da Vinci recognized that rings in the branches of trees show annual growth and the thickness of the rings indicate the years that were more or less dry. Andrew Douglas was a pioneer who formulated the scientific basis of the field in the 1920s and 1930s and, importantly, the principle of cross dating, whereby overlapping chronologies from different trees could be merged to form long series. He founded the Laboratory of Tree Ring Research at the University of Arizona, where he and coworkers collected a vast number of cores using a tubular boring device.2 To understand the meaning of tree rings, he decomposed growth into five parts: (1)

Rt ⫽ At ⫹ Ct ⫹ ϕD1t ⫹ ϕD2t ⫹ Et

where R denotes ring width; t the year; ϕ is a presence or absence indicator (taking values of 0 or 1); A is the age trend of tree growth; C indicates climate; D1 represents external disturbance processes such as a fire; D2 represents internal growth disturbance such as a disease; and E is an error term. The goal is to decipher or solve for C given R and information on A, D1, and D2. To reduce the effects of disturbances it is useful to collect large samples in a particular locality. The Drought Database Researchers have estimated the PDSI from instrument records, which cover the period 1900 to the present, and from tree rings, which go back as much as two thousand years depending upon the locality. Thus, there is a break in the data source in 1900, and for this reason, it is prudent to divide analyses into two corresponding parts. Contrasts in the results across the two periods might be attributable to different data sources or to a structural shift in the relationship between drought and economic activity. Note that the PDSI estimated from tree rings measures net moisture available for plant growth and is, therefore, a function of precipitation, temperature, and wind patterns, as well as the timing of these effects during the growing 2. We thank Henri Grissino-Mayer for concepts important for this discussion. For additional information, see his Web page: http://web.utk.edu/~grissino/.

Climate Change: Adaptations in Historical Perspective

11

season. Nevertheless, the PDSI estimated from tree rings tracks the instrument record reasonably well. Assembling tree ring chronologies and estimating PDSI is an ongoing process, which is described in The North American Drought Atlas (Cook and Krusic 2004). Various chapters in the volume use PDSI values estimated from 835 tree ring chronologies scattered across North America, which researchers used to estimate PDSI values at 286 grid points (the raw data are available at http://www.ncdc.noaa.gov/paleo/pdsi.html). The points are evenly spaced over a 2.5 degree grid (roughly 175 miles apart), which provides a useful approximation to annual net moisture conditions at the local (state) level.3 Researchers use two sets of drought indexes obtained from the North American Drought Atlas. The first is a set of instrumental PDSI readings for the sample period from 1900 to 1978. The second is a set of reconstructed PDSI readings using tree ring data that can be used to extrapolate the PDSI data far back into the past. Overview The inherent variability of regional climates in the past and projections of the future suggest that climate change poses serious and potentially dramatic challenges to the American economy. In part, the magnitude of these challenges depends upon the nature of the overall weather response to the buildup of green house gases (GHG). As indicated in the following two chapters of this volume, there is considerable uncertainty as to the feedback mechanism between GHG accumulations and the climatic reaction. Accordingly, we may not know the nature of the problem before us for some time. In this case, an understanding of the likely adaptability of the American economy will be important for forging any private or government action. Historical experiences can give us longer term assessment of just how well the American economy is positioned to meet climate change. The economic impact also will depend on the time frame under which climate changes will occur. As with temperature projections, there is no consensus on a specific time period for major economic damages to materialize. One possibility is that they will be small and isolated for twenty to fifty years, after which they will be cumulatively larger. If this is correct, then it 3. Cook and Krusic used point- by- point regression, which is a sequential, automated fitting of single- point principal component regression models to a grid of climate variables taken from instrument readings. The method assumes that only those tree ring chronologies close to a given PDSI grid point are good predictors of drought at that location. They used instrument data from 1928 to 1978 (the calibration period) to develop each regression equation. The remaining data from 1900 to 1927 (the verification period) of the instrument record were used to test the validity of the PDSI estimates. Additional details, including a map of the grid points and a discussion of statistical methods, are available at http://iridl.ldeo.columbia.edu/SOURCES/ .LDEO/.TRL/.NADA2004/.pdsi- atlas.html.

12

Gary D. Libecap and Richard H. Steckel

may make sense for modest emissions abatement programs initially while the economy begins to adjust, more technology and learning are developed, and more information is generated.4 A major question, then, is how adaptable is the American economy? Agriculture will be particularly vulnerable if temperature and precipitation become more erratic with larger swings. Six of the chapters in this volume address agricultural responses to new weather conditions in American economic history. Indeed, the expansion of agriculture across North America in the nineteenth and twentieth centuries encountered greater climatic variation than is predicted by at least some climate change models (Olmstead and Rhode 2008). And as we see, agriculture was amazingly adaptive through new crop types, mixes, and methods of cultivation. Health will be affected, and two chapters address the issues. Here we also see responsiveness through demands for reduced air emissions as incomes rose and associated regulations to limit pollution were adopted. We also see that once individuals had access to information regarding health threats related to climate, they generally were able to respond effectively. Electricity demand and pressure on utilities also likely will increase. The final chapter of the volume provides information on the size and heterogeneity of demand changes as temperature increases. This information is valuable for preparing for new energy sources if temperatures become hotter, especially in areas already warm and dry. Overall, research in economic history reveals both how closely twined are climate, weather, and the economy and how remarkably resilient and adaptive is the economy. This is a valuable insight both because it suggests adjustments are likely to occur as new information, new learning, and new technologies emerge and because it augments contemporary climate change studies that typically rely upon either simulations or very limited data sets. Adaptation takes time, and history is the best provider of information about how it has unfolded over time. It also is important to note that the record of adjustment and flexibility described here is apt to underestimate the general responsiveness of the current and future economy. In the nineteenth and early to mid- twentieth centuries, the time period for most of the research presented here, information sources and transmission mechanisms were much more limited and costly than today. There were far fewer government, business, and agricultural organizations for coordination and mobilizing responses. Technology and knowledge were necessarily more primitive. It seems reasonable to conclude that if the economy could adapt in the manner described in the following in the past, it ought to be able to do so even more effectively in the future. 4. For a sense of the debate over the magnitude of the impact and the time frame involved, see National Academy of Sciences (2008), Nordhaus (2008), and Kousky, Rostapshova, et al. (2009).

Climate Change: Adaptations in Historical Perspective

13

This is not to say that we should be too sanguine about the ability of the economy to adapt to climate change. But the historical record, at least, is one that brings considerable optimism about the future responsiveness of the society to the important challenges of a climate that may change in unknown ways, raising uncertain costs—and, possibly, benefits for both current and future generations. This record has not been presented in one place before, and how the economy has responded to past climatic shocks has not been sufficiently integrated into contemporary debates over climate policy. The evidence reported here reveals that much can be learned not only about specific responses in the past, but also about the overall flexibility and ingenuity found in sectors of the American economy. Future climatic challenges will require similar innovation and elasticity, perhaps in ways that have not been encountered before. But the research in this volume provides reasons for confidence and expectations for creativity that should not be overlooked as the economy moves to address a changing climate environment. The chapters in this volume describe research findings regarding historical climate- related events as they have been faced in the American economy; the responses of individuals, organizations, and government institutions to those climate challenges; and assessments of their successes in addressing potential disruptions and in promoting the continued economic growth and welfare. The chapters also provide new data sources for measuring and evaluating how economic agents have adjusted to and progressed even in light of formidable environmental concerns. This record of adaptation and progress has involved both individuals and organizations in the private sector—farmers, bankers, consumers, and firms, as well as government agencies at the state and federal level—agricultural experiment stations and extension services, the Bureau of Reclamation and Army Corp of Engineers, the National Weather Service, health agencies, and regulatory bodies. The Uncertain Economic Implications of Climate Change The first two chapters describe the environmental problems before us and the uncertainties involved in assessing the magnitude of climate change in determining its long- term impact on the economy and society and in formulating government policies and private strategies to address it. Chapter 1, “Additive Damages, Fat-Tailed Climate Dynamics, and Uncertain Discounting” by Martin L. Weitzman, provides a theoretical framework for thinking about climate change. As Weitzman notes, climate change is so complicated, involving so many different disciplines and viewpoints that a model can only address one or two facets of the problem. His chapter focuses on the economic implications of unusually large structural uncertainties surrounding climate change extremes and why aggressive emissions

14

Gary D. Libecap and Richard H. Steckel

mitigation policies might be justified. He argues that additive damages are appropriate for analyzing the economic impact of global warming in light of highly uncertain feedback mechanisms between GHG accumulations and temperatures with possibly catastrophic results (losses of habitats and species, dramatic rises in sea levels, shrinkage of inhabitable regions and associated collapses in production and consumption). With uncertain rates of time preference and discount rates approaching zero, these climate change damages and uncertain response mechanisms lead to very large expected present disutility. The social willingness to pay to avoid potential catastrophically high temperatures in the future could be infinite, whereby society sacrifices all current consumption to prevent future warming. Because current CO2 and CH4 atmospheric accumulations appear to be at unparalleled levels and changing more rapidly than ever before, and the speed and magnitude of the global temperature response is so unclear with possible irreversible, self- enhancing feedback effects, Weitzman is doubtful that there are opportunities for learning, midcourse corrections, or adaptive adjustments of the types described elsewhere in this volume to respond to climate change. Chapter 2, “Modeling the Impact of Warming in Climate Change Economics” by Robert S. Pindyck, also examines the issues raised by Weitzman, but with a different approach and a different conclusion. He incorporates distributions for temperature change and its possible economic impacts derived from studies assembled by the IPCC and from integrated assessment models (IAMs) into an analysis of climate change policy. He estimates the fraction of consumption society would be willing to sacrifice to ensure that future temperature changes are limited to some target level, say 2° or 3°C. He models the relationship between temperature change and the growth of gross domestic product (GDP) using a thin- tailed distribution for temperature change inferred from studies surveyed by the IPCC and a zero discount rate. Pindyck finds that the willingness to pay is less than 2 percent to avoid temperature increase greater than 3°C over 100 years, a result that supports only moderate abatement policies. Pindyck’s approach involves fitting parameters based on the existing state of knowledge regarding temperature change and its economic impact. In contrast to Weitzman, the analysis supports slower, more adaptive policies to address climate change that rely on new learning, new technologies, and market adjustments to lower costs and avoid the losses of more preemptive policies that turn out to be inappropriate or ineffective. This approach certainly would take advantage of the adjustment mechanisms described in the empirical work presented in this volume. The Costs of Climate Change Because Weitzman and Pindyck point to the large uncertainties in assessing the potential costs of climate change, it is worthwhile to continue the

Climate Change: Adaptations in Historical Perspective

15

volume with three chapters that examine historical climate- related costs. Chapter 3 by John Landon-Lane, Hugh Rockoff, and Richard H. Steckel, “Droughts, Floods, and Financial Distress in the United States,” examines the relationships among weather, agriculture, and financial markets. There has been a long- standing sense, going back at least to Stanley Jevons (1884) and the notion that financial crises were linked to sunspots, that weather shocks as they curtained agricultural production could bring about bank failures and thereby disrupt the macroeconomy. Even though this linkage was noted by Keynes (1936) and Friedman and Schwartz (1963), there has been little empirical work on the causal mechanisms and the overall magnitude of the impact. The authors examine how drought has affected both bank failures and rates of return to bank equity in the United States. Two major droughts are emphasized, those between 1874 and 1896 in Kansas and those during the 1930s in Oklahoma. Both of these climatic events crippled agriculture, a major sector of the state economies. The PDSI has been constructed from meteorological data systematically collected by the U.S. Weather Bureau since 1900. Examining drought in nineteenth century Kansas requires that the index be reconstructed using tree ring data. These data are combined with measures of financial distress. The authors show that the early droughts, especially, resulted in drops in the rate of return to national bank equity and in bank failures as banks faced defaults on agricultural loans and mortgages. Because of available data, it is possible to more systemically examine the linkage between drought and bank stress in Oklahoma during the 1930s. Surprisingly, despite the extent of the drought and the corresponding Dust Bowl in western and central Oklahoma, national bank rates of return reveal little effect. Much of the distress instead appears to have been with local, small state banks that did not have the institutional framework available to national banks for accessing funds and smoothing risk. Data for state bank rates of return, however, are not available for econometric analysis. The authors proceed, however, with other time series analysis of the impact of the PDSI on national bank capital rates of return, farm foreclosure rates, and farm income between 1850 to 1976 and various subperiods for the United States as a whole and by region and subregion. Generally, they find a positive relationship between the PDSI and rates of return—more rain, higher rates of return, with extreme drought and wet periods reducing rates of return. The effects are most pronounced in the period through 1940 and in the farming region of the Midwest. Similar results are reported for farm foreclosures and farm income. After 1940, however, the relationship between the financial sector and weather is much weaker. This suggests considerable adjustment through institutional changes via branch banking that allowed banks to better sustain localized droughtinduced economic stress and through agricultural adjustments through a

16

Gary D. Libecap and Richard H. Steckel

shift to new crops and new production methods. In sum, the authors find that weather events have had less economic cost since the mid- twentieth century than one might have imagined. They also suggest that organizational innovations that lowered the cost of accessing capital across banks during times of climate shocks was an important factor in mitigating financial and overall macroeconomic distress. Chapter 4, “The Effects of Weather Shocks on Crop Prices in Unfettered Markets: The United States Prior to the Farm Programs, 1895–1932” by Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode, also addresses the cost of climate- related events by examining the impact on agricultural prices. Because of current government intervention into commodity markets in many leading agricultural countries, it is more difficult to gauge the linkages between weather on contemporary agricultural prices. This was not the case in the early twentieth century. The authors assemble a thirty- seven- year panel of state data from the U.S. Department of Agriculture (USDA) for cotton, corn, wheat, and hay and from the National Climatic Data Center for temperature and precipitation, including the PDSI. They econometrically test for linear and nonlinear effects of temperature and precipitation on prices through time series analysis controlling for state and year fixed effects. They find that prices for cotton and wheat, two crops most traded internationally, were not sensitive to local state- level weather shifts. Prices for more locally consumed commodities, corn and hay, however, were much more affected. Indeed, the effect of severe drought and high temperatures especially on corn crops helped to mobilize political action for federal government intervention to stabilize prices and farm incomes. Accordingly, in these cases, the costs of weatherrelated events depended in part on the extent of the market. Where markets were narrower, regional weather events had more pronounced costs, and these, in turn, likely were a source of subsequent government intervention. Chapter 5, “Information and the Impact of Climate and Weather on Mortality Rates during the Great Depression” by Price V. Fishback, Werner Troesken, Trevor Kollmann, Michael Haines, Paul W. Rhode, and Melissa Thomasson, continues the examination of the cost of climatic events. Contemporary discussion of climate change often points to its possible health implications. In this chapter, Fishback et al. present new data on mortality rates for 3,054 U.S. counties between 1930 and 1940, a decade that saw unusually severe droughts and high temperatures as well as economic collapse associated with the Great Depression. These conditions could be similar to conditions in many developing countries today should climate change disrupt the macroeconomy. Combining data on mortality, temperatures, precipitation, and various socioeconomic correlates, the authors examine both cross- section and time series variation in temperature and precipitation and death rates across the over 3,000 counties during the 1930s. Respiratory and diarrheal diseases,

Climate Change: Adaptations in Historical Perspective

17

as well as the prevalence of disease- carrying insects could be influenced by weather fluctuations, and infant children are particularly susceptible. Even so, the weather variables show little significant impact on infant mortality. The most important explanatory factors are those associated with information—literacy and access to radios and magazines. These results are repeated in statistical tests where noninfant mortality is the dependent variable. The findings underscore the importance of improved information flows to promote public health as a way of reducing the effect of weather shocks on welfare. The authors acknowledge that their analysis focuses on weather fluctuations around long- term climate norms and that shifts in climate may have larger effects on mortality. Evidence of Adaptation to Climatic Challenges As noted by Robert Pindyck in chapter 2, moderate climate policy can benefit from adaptation in the economy that could reduce both the cost of climate change and efforts to mitigate it. Three chapters address adaptation by private markets and governments in American economic history. Chapter 6, “Responding to Climatic Challenges: Lessons from U.S. Agricultural Development” by Alan L. Olmstead and Paul W. Rhode, provides a longterm perspective for understanding how American farmers have responded to new and challenging growing environments that in some cases involved greater temperature variations than are forecast in many contemporary climate change scenarios. Moreover, the adjustments Olmstead and Rhode chronicle took place prior to major understanding of plant genetics. Although the authors point out that they cannot say how future farmers, aided by new breakthroughs in plant science, might respond to climate change, the historical record they present reveals the malleability of the agricultural enterprise that countered past expert predictions of failure. These predictions failed to see the impact of biological innovations that would transform commodity production. Olmstead and Rhode examine the regional climatic barriers that were encountered in the expansion of the agricultural frontier across North America and the collaborative innovations of farmers, agricultural experiment and extension agents, and seed companies in responding to them for wheat, corn, and cotton (Olmstead and Rhode 2008). The authors point out that between 1839 and 1929, U.S. wheat production increased by ten times, and production shifted to areas that were very different in climate and soil. New varieties, particularly hard red winter and spring wheat, improved yields and were more tolerant to extremes in heat, cold, and drought while also being resistant to rust and to other plant diseases. New dry farming techniques introduced by farm organizations and the USDA further allowed for production to be extended into areas that previously had been considered too inhospitable. Similar progress is described for corn and cotton.

18

Gary D. Libecap and Richard H. Steckel

The response of farmers to new seed types and the roles of private firms and government agencies in developing them is described in chapter 7, “The Impact of the 1936 Corn-Belt Drought on American Farmers’ Adoption of Hybrid Corn” by Richard Sutch. Sutch reexamines Zvi Griliches famous study of the adoption of hybrid corn (Griliches 1957). Griliches argued that diffusion could be described by a logistic function, whereby individuals learned from first adopters. As more parties considered the innovation, overall penetration would rise initially at an increasing rate to an inflection point; after that, adoption would rise more slowly. Sutch’s study shows that the process was far more complex and that there was critical involvement of both government and private companies. Indeed, until the 1936 drought, Sutch argues that there was little or no economic advantage to hybrid corn over standard, open- pollinated varieties due to the high cost of seed and the need for more fertilizer and greater use of mechanization at a time of low corn prices. Hybrid corn seeds had been marketed commercially in the United States since 1925, with slow adoption. But the drought of 1936 revealed advantages of hybrid corn that previously had been unrecognized—its drought tolerance. This advantage led to more rapid adoption, especially in the more drought- prone western Corn Belt in 1937 and 1938. Further, there was path dependence. Greater sales revenues financed research and development (R&D) by private seed companies, particularly Pioneer Hi-Bred Seed that improved seed productivity in normal years, encouraging more adoption. A sudden climate event, the extraordinarily severe drought of 1936, served then as a tipping point by generating new information about the advantages of hybrid corn seeds. Additionally, Pioneer was founded by Agricultural Secretary Henry A. Wallace, who used his position to encourage the diffusion of new seed types. Sutch’s analysis uses data from the Iowa Corn Yield tests of 1926 to 1940 as well as Griliches’s data and relies upon yield per harvested acre, a measure of the severity of the drought in 1936. He tests the hypothesis that adoption of hybrid corn after 1936 was influenced by drought conditions. His analysis provides a more complete and interesting story about the development and dissemination of innovation and the important role of collaboration among farmers, firms, and government agencies as emphasized by Olmstead and Rhode. Chapter 8, “The Evolution of Heat Tolerance of Corn: Implications for Climate Change” by Michael J. Roberts and Wolfram Schlenker, continues the analysis of technological change in seed types as responses to climatic shocks. They provide new evidence on the relationship between weather and corn yields by assembling data from 1901 to 2005 for Indiana, a major cornproducing state, on yields, the daily temperature range between minimum and maximum temperature, and precipitation. Overall, they find that crop sensitivity to extremes in temperature and precipitation evolved over time in a manner described in the previous chapter as new seed varieties, supple-

Climate Change: Adaptations in Historical Perspective

19

mental irrigation, and new production and management techniques were introduced. Using a panel of county- level yields in Indiana in the econometric analysis, Roberts and Schlenker, however, find that the evolution of tolerance for excessive heat has been less successful than that for swings in precipitation. They find that improvements in heat tolerance have been nonlinear, growing with the introduction of new double- cross hybrids in the 1940s and 1950s, peaking in 1960, and then declining after the introduction of single- cross hybrid corn. The key question they raise is whether the next seed innovation cycle can bring both increases in average yields and heat tolerance as was the case from 1940 to 1960, or if increases in average yields can only be obtained while lowering heat tolerance as was the case from 1960 onward. In the latter case, there may be little scope for response to higher temperatures while maintaining yields in areas where corn is currently grown, as all climate models predict significant increases in extreme heat in these areas. Government Policy and Adaptation to Variable Climatic Conditions The final three chapters address government policies and how they have affected the response to new climate conditions. Chapter 9, “Climate Variability and Water Infrastructure: Historical Experience in the Western United States” by Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe, examines how a major policy initiative, massive investment in dams and related canals, largely for irrigation and flood control in the twentieth century, affected crop yields and mixes during times of extreme drought and wetness. Because the water infrastructure constructed by the Bureau of Reclamation and the Army Corps of Engineers was importantly influenced by political constituent pressures (Pisani 2002), it is possible that the infrastructure may have had little impact on smoothing output, but rather served to expand production into areas where it otherwise might not have been possible. To explore these issues, the authors assemble a county- level data set of 3,620 observations for five western states, Idaho, Montana, North Dakota, South Dakota, and Wyoming using census data for the twentieth century. These states have similar temperature and precipitation patterns, crops, and soil types, but the availability of irrigation varies widely. The crop data are for hay, wheat, barley, corn, and potatoes. The first two crops, especially, can also be dry farmed without irrigation. The data set includes total planted acreage, total failed acreage, total fallow or idle acreage, and total harvested acreage by crop, along with information on topography, soil quality, water storage and distribution, temperature, and precipitation. The authors examine variation in agricultural production and crop mix before and after the water infrastructure was installed, and across counties with and without such infrastructure during times of excessive drought and precipitation. Various econometric techniques are employed to address

20

Gary D. Libecap and Richard H. Steckel

endogeneity. They find that agriculture within counties with water storage and distribution facilities generally was better able to deal with climate variability, smoothing output, and crop mixes in times of drought and heavy precipitation, relative to agriculture in counties without such facilities. The results underscore how important the water infrastructure has been for long- term adaptation strategies to respond to highly variable climatic conditions. In Chapter 10, “Did Frederick Brodie Discover the World’s First Environmental Kuznets Curve? Coal Smoke and the Rise and Fall of the London Fog,” Karen Clay and Werner Troesken examine the incidence of coal smoke in fogs in and around London and the pattern of such fogs over time. Brodie attributed these fogs to intense coal smoke emissions between 1871 and 1903. After that, however, he argued that the fogs largely ended. These were the famous London fogs that led Arthur Cecil Pigou to describe taxes (Pigouvian taxes) as a means of internalizing the costs of externalities (Pigou 1920). Clay and Troesken analyze to see if this claim is correct, why the pattern of fogs existed, and address related health effects of the so- called killer fogs. In doing so, they reconsider Brodie’s limited data and assemble additional information on coal consumption per capita, gas and electricity use, abatement legislation, and mortality from respiratory diseases. The authors construct a reverse event study, using spikes in mortality to predict severe fogs and then compare those predictions against other evidence regarding their occurrence. As indicated by Brodie, the authors find that between 1855 and 1910, there were recurring fogs, but none after 1900. With this information, they then conjecture why the fogs declined after 1900. The smoke density in London fell for a variety of reasons: the city’s population became more dispersed; the inhabitants became richer; and associated regulations, such as the 1891 Public Health Act instituted fines for dense smoke emissions, promoting a shift to the use of gas and hard coal that burned more cleanly. Clay and Troesken conclude that Brodie was correct in his assessment of the source of London’s killer fogs and that the city had to reach a threshold level of income and technological advancement before it could address the problem of coal smoke. The final chapter, “Impacts of Climate Change on Residential Electricity Consumption: Evidence from Billing Data” by Anin Aroonruengsawat and Maximillian Auffhammer, does not directly address government policy, but it provides information critical for formulating policy in the energy sector. The chapter describes analysis of an unusually complete panel data set for California from 2003 to 2006. Supplementing these data with weather data and census information by zip code and controlling for household, month, and year fixed effects, the authors examine the electricity consumption response to changes in temperature across sixteen climate zones. Flexible temperature response functions are estimated by zone, and they find heterogeneity in the response.

Climate Change: Adaptations in Historical Perspective

21

The greatest impact occurs in areas with the largest number of high and extremely high temperatures—in the central and southeastern parts of the state where electricity demand could rise as much as 55 percent by the end of the century. The authors also explore the impact of other variables on electricity consumption, including the percent of households using electricity for heating, the percent using natural gas, urban location, and age of the structure. Overall, the study finds larger and more nonlinear impacts on electricity consumption from temperature extremes than had been previously found. The authors caution that the study does not allow for changes in consumption due to price shifts, movements to different locations, or other structural modifications that might reduce potential increases in demand. The chapters in this volume provide important new empirical information on and analyses of the economics of climate change. They examine responses to past climatic events and in so doing indicate the range of possible future adaptations. As climatic events unfold, this knowledge is critical for understanding how society has reacted to similar occurrences in the past and for developing effective, new private and governmental policies to address them.

References Alley, William M. 1984. The Palmer Drought Severity Index: Limitations and assumptions. Journal of Climate and Applied Meteorology 23 (7): 1100–09. Cook, E. R., and P. J. Krusic. 2004. North American Drought Atlas. Palisades, NY: Lamont-Doherty Earth Observatory and the National Science Foundation. Friedman, Milton, and Anna Jacobson Schwartz. 1965. A monetary history of the United States, 1867–1960. Princeton, NJ: Princeton University Press. Griliches, Zvi. 1957. Hybrid corn: An exploration in the economics of technological change. Econometrica 24 (4): 501–22. Heim, Richard R., Jr. 2002. A review of twentieth- century drought indices used in the United States. Bulletin of the American Meteorological Society 83 (8): 1149–65. Intergovernmental Panel on Climate Change (IPCC). 2007. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, ed. S. Solomon, D. Qin, M. Manning, Z. Chen, M. Marquis, K. B. Averyt, M. Tignor, and H. L. Miller. Cambridge, UK: Cambridge University Press. Jevons, William Stanley. 1884. Investigations in currency and finance. London: Macmillan. Keynes, J. M. 1936. William Stanley Jevons, 1835–1882: A centenary allocation on his life and work as economist and statistician. Journal of the Royal Statistical Society 99 (3): 516–55. Kousky, Carolyn, Olga Rostapshova, Michael Toman, and Richard Zeckhauser. 2009. Responding to threats of climate change mega- catastrophes. World Bank Policy Research Working Paper no. 5127. Washington, DC: World Bank. Libecap, Gary D. 2008. Open access losses and delay in the assignment of property rights. Arizona Law Review 50:379–408.

22

Gary D. Libecap and Richard H. Steckel

McIntyre, Steve. 2007. Gridding from CRN1- 2. http://climateaudit.org/2007/10/04/ gridding- from- crn1- 2/. National Academy of Sciences. 2008. Understanding and responding to climate change: Highlights of National Academies reports. Washington, DC: National Academies. Nordhaus, William. 2008. A question of balance: Weighing the options on global warming. New Haven, CT: Yale University Press. Olmstead, Alan L., and Paul W. Rhode. 2008. Creating abundance: Biological innovation and American agricultural development. New York: Cambridge University Press. Pachauri, R. K., and A. Reisinger, eds. 2007. Climate change 2007: Synthesis report: Contribution of Working Groups I, II, and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Geneva, Switzerland: IPCC. Palmer, Wayne C. 1965. Meteorological drought. U.S. Weather Bureau Research Paper no. 45. Washington, DC: U.S. Weather Bureau. Pigou, Arthur Cecil. 1920. The economics of welfare. New York: Macmillan Press. Pisani, Donald J. 2002. Water and American government: The Reclamation Bureau, national water policy, and the West, 1902–1935, Berkeley, CA: University of California Press. Thornthwaite, C. Warren. 1931. The climate of North America according to a new classification. Geographical Review 21 (4): 633–55.

1 Additive Damages, Fat-Tailed Climate Dynamics, and Uncertain Discounting Martin L. Weitzman

1.1

Introduction

Climate change is so complicated, and it involves so many sides of so many different disciplines and viewpoints that no analytically- tractable model or paper can aspire to illuminate more than one or two facets of the problem. This is a chapter in applied theory, where the application is to climate change. The chapter ranges widely but is primarily about some economic implications of some of the unusually large structural uncertainties surrounding climate change extremes. In particular, I focus on implications of fat tails. The 2008 financial crisis has brought home the idea that people should think not just in terms thin- tailed distributions, the primary example of which is the normal. In the analysis of catastrophes, whether in finance or climate change, it is important to focus on the implications of fat- tailed distributions like the Pareto. One major structural uncertainty in climate change economics concerns the appropriate way to represent damages from global warming. The functional form used most frequently in the literature is a nested utility specification, within which consumption is reduced multiplicatively by a

Martin L. Weitzman is a professor of economics at Harvard University and a research associate of the National Bureau of Economic Research. Without tying them to the contents of this chapter, I am grateful for critical comments on an earlier version by Frank Ackerman, Marcia Baker, Stephen DeCanio, Simon Dietz, David Frame, Olle Häggström, John Harte, Peter Huybers, Michael MacCracken, Torsten Persson, Alan Robock, Michael Schlesinger, Robert Stavins, Jörgen Weibull, and Thomas Wigley. This chapter was originally published in the e-Journal Economics: The Open-Access, Open-Assessment E-Journal (Martin L. Weitzman, “Additive Damages, Fat-Tailed Climate Dynamics, and Uncertain Discounting,” 3 (2009- 39), http://www.economics- ejournal.org/ economics/journalarticles/2009- 39).

23

24

Martin L. Weitzman

quadratic- polynomial function of temperature change. In this chapter, I argue that an additive form, in which welfare is the difference between the utility of consumption and a quadratic disutility- loss function of temperature change, may make just as much sense for evaluating extreme climate damages. The distinction between multiplicative and additive welfare specifications may seem arcane, but I explain why it can make a surprisingly significant difference in the evaluation of future scenarios involving both high temperatures and high consumption. There are a great many structural uncertainties about climate change extremes other than the specification of damages, which represents just one part of the economic- welfare side. To represent structural uncertainty on the science side, I also use one specific example although several others might have served in this capacity. So- called climate sensitivity is the equilibrium mean surface temperature response to a doubling of atmospheric CO2. An oft- asked question is: why has it been so difficult, after some three decades of intensive scientific research, to narrow down the upper- tail probability density function (PDF) of climate sensitivity enough to exclude very high values (say substantially higher than 4.5°C)? A standard answer is that seemingly tiny uncertainties concerning the possibility of a large feedback factor f near 1 are naturally amplified into broad uncertainties about very large values of climate sensitivity  by a highly nonlinear transformation of the form   0/ (1 – f ). A detailed examination of the generic analytical mechanism behind such an explanation reveals that the implied upper- tail distribution of climate sensitivity is so “fat” (or “heavy” or “thick”—all synonyms) with probability that its variance is infinite. In other words, essentially the same argument used by most scientists to explain why high values of climate sensitivity cannot be excluded contains within itself the seeds of a generic argument not just for a fat upper tail of the PDF of climate sensitivity, but for a very fat tail, which is so spread out that it has infinite variance. With an additive quadratic loss function, this infinite variance translates into infinite expected disutility. Climate sensitivity is a long- run equilibrium concept that abstracts away from the transient dynamics by which it is approached as an asymptote. The next two logical questions concern the transient dynamic phase: (a) what happens to the PDF of temperatures along the uncertain dynamic trajectory that leads to climate sensitivity as an asymptotic limit (and whose variance approaches infinity, but only at an infinitely distant future time)? (b) What is the welfare evaluation of the uncertain transient trajectory of temperatures? The simplest diagnostic energy- balance model is used to specify quantitatively how the PDF of temperatures varies over time in approaching its infinite- variance limiting PDF. In the case of an additive damages function (here quadratic in temperature changes), welfare evaluation then mainly becomes centered on the issue of how future disutilities should be discounted. I show that when the “rate of pure time preference” or “utility discount rate” is uncertain, but it has a PDF with infinitesimal probability

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

25

in a neighborhood of zero, then the limiting expected present discounted disutility of additive quadratic temperature damages approaches infinity. The chapter closes by mentioning some possible welfare and policy implications of this disturbing theoretical finding. 1.2

Temperature Damages: Multiplicative or Additive?

This section of the chapter argues that it can make a big difference for climate change policy whether high- temperature damages are specified as entering the overall net utility function multiplicatively or additively with consumption. Most modelers use a multiplicative formulation, perhaps not realizing the degree to which their model’s outcomes depend sensitively on this particular assumption. Here I argue that an additive form might possibly make as much sense as a multiplicative form and indicate why this seemingly obscure distinction might matter a lot, especially at high temperatures conjoined with high consumption. Although generalizations are possible, suppose for the sake of specificity here that the utility of consumption is isoelastic with coefficient of relative risk aversion two. Let C be consumption, while T stands for temperature change above the prewarming level. A utility function commonly used in the economics of climate change (for coefficient of relative risk aversion two) is, up to an affine transformation, of the multiplicative form (1)

冤冢 冣

冥

1 UM (C,T )     (1  MT 2) , C

where M is a positive coefficient calibrated to some postulated loss for T ≈ 2 to 3°C.1 Equation (1) is essentially a single- attribute utility function, or, equivalently, a multiattribute utility function with strong substitutability between the two attributes. This would be an appropriate formulation if the main impact of climate change is, say, to drive up the price of food and increase the demand for air conditioning. Instead of the multiplicative functional form (1), suppose we now consider, up to an affine transformation, the analogous additive functional form (with a quadratic loss function) (2)

冤冢 冣

冥

1 UA(C,T )     (1  AT 2) , C

where A is a positive coefficient calibrated to some postulated loss for T ≈ 2 to 3°C. Equation (2) is a genuine multiattribute utility function. It describes a situation where the main impact of climate change is on things that are not readily substitutable with material wealth, such as biodiversity and health. 1. Such type of calibration is done in Nordhaus (2008) and Sterner and Persson (2008) among others.

26

Martin L. Weitzman

The 1 in the right side of equation (2) is an inessential constant, intended only to facilitate comparison of equation (1) with equation (2). When C is normalized to unity, then UM(1,T )  UA(1,T ), and the same calibration can be used to fix the same value of A  B   in both cases. Note that equations (1) and (2) are symmetric, with the only difference being the “” sign in equation (1) and the “” sign in equation (2). I think it is fair to say that it is hard to argue strongly for one form over the other from any basic principles so that, at first glance, there might seem to be little basis for choosing between the multiplicative form (1) and the additive form (2). I do not want to take a decisive stand on which of equations (1) or (2) are “better” formulations of temperature damages. The main purpose of this chapter is to point out that a seemingly arcane theoretical distinction between additive and multiplicative disutility damages may have surprisingly strong implications for economic policy. There is not much difference between equations (1) and (2) for small values of C and T, but when they are large, which is the domain of the utility function about which we are most unsure, I show that the distinction becomes significant. In an effort to compare and contrast in familiar language the basic properties of the multiplicative form (1) with the additive form (2), I ask the following question. What is the willingness to pay as a fraction of consumption that the representative agent would accept to reduce temperature change to zero? This welfare- equivalent fraction of consumption w must satisfy the equation (3)

U [(1  w)C, 0]  U(C,T ).

Plugging equation (3) into equations (1) and (2), one obtains, respectively, (4)

T 2 wM  2 1  T

and (5)

CT 2 wA  2 . 1  CT

If C is normalized to unity, then for all temperature changes, the two specifications are identical and wM  wA. Notice, though, what happens as C increases from its initial value of 1. Under the multiplicative specification (4), the fraction of consumption willing to be paid to eliminate temperature change, wM, is independent of C. This might appear to be odd because one might think that in a rich world the fraction of consumption people would be willing to sacrifice to eliminate a given temperature change would be higher than in a poor world. Note that wA in equation (5) has just this property. Let time be denoted t. Thus, consumption at time t is C(t), while temperature change at time t is T(t). For notational simplicity, normalize so that

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

27

C(0)  1, T(0)  0. To compare and contrast in familiar language the basic dynamic properties of the multiplicative form (1) with the additive form (2), I now ask the following question. What is the willingness to pay as a fraction of current consumption C(0)  1 that the representative agent would accept to reduce the temperature T(t) 0 at time t 0 down to T(t)  0? Call this value W. Suppose that the rate of pure time preference or “utility discount rate” is . Then W must satisfy the equation (6) U(1, 0)  U [(1  W ), 0]  exp( t){U[C(t), 0]  U[C(t), T(t)]}. Suppose consumption grows at rate g so that C(t)  exp(gt). Then plugging equations (1) and (2) into equation (6), after some algebraic rearranging one obtains (7)

exp[(g  )t]T(t)2 WM   1  exp[(g  )t]T(t)2

and (8)

exp( t)T(t)2 WA  2 . 1  exp( t)T(t)

The difference between the multiplicative formulation (7) and the additive formulation (8) is that the latter is free of the powerful dampening term exp(–gt). To give a numerical example emphasizing the significance of this kind of distinction, suppose that g  2 percent,  0, t  150. By the Ramsey formula with coefficient of relative risk aversion   2, the corresponding real interest rate in this example is r   g  4 percent. Calibrate  so that 2 percent of welfare- equivalent consumption is lost (at C [0]  1) when T  2°C. Then straightforward calculations show that the willingness to pay at time t  0 to avoid T(150)  4°C under the multiplicative specification is WM  0.4 percent, while under the additive specification it is WA  7.5 percent—a difference of almost twenty times. Another way to see this dramatic difference is ask how much of a welfare- equivalent temperature reduction in 150 years would 7.5 percent of current consumption buy. With additive utility (8), the answer (from the preceding) is 4°C. With multiplicative utility (7), the answer is 18°C! In this spirit it might be argued that, relative to the multiplicative form (1), the additive formulation (2) does not trivialize the welfare impacts of large future temperature changes. One lesson to be drawn from this simple numerical example is that a seemingly arcane distinction between an additive and a multiplicative interaction of temperature change with consumption might have big consequences. If so, then it becomes another example of structural uncertainty exerting a decisive influence on climate change policy (here the structural uncertainty concerns the functional form of utility damages). In an important article, Sterner and Persson (2008) tested on a leading

28

Martin L. Weitzman

integrated assessment model (IAM) a utility function of the constant elasticity of substitution (CES) form (9)

冤

1 U(C,E )   (1  b)C ( 1)  bE ( 1) 1

冥

[(1) ]/( 1)

,

where the “environmental” good is (10)

E0 E  2 . 1  aT

With elasticity of substitution  1/2 and coefficient of relative risk aversion   2, the above formulation (9), (10) is equal (up to an affine transformation) to (11)

冢

冣

1 US (C,T )     1  T 2 , C

where   ab/[(1 – b)E0]. Comparing equation (11) with equation (2), US (C,T ) is identical with UA(C,T ). Thus, for the parameter values  1/2 and   2 chosen by Sterner and Persson, the additive version (2) and the CES version (11) are the same and will, therefore, give the same results when plugged into any IAM. Consequently, one is free to view this utility function either through the lens of an additive form or through the lens of a CES form, using whichever lens gives more insight for a particular application. Importantly, Sterner and Persson (2008) found empirically that plugging their CES utility function (11) into William Nordhaus’s (2008) pathbreaking Dynamic Integrated model of Climate and the Economy (DICE) model yields a far more stringent emissions policy than Nordhaus found with his multiplicative utility form (1).2 As an empirical matter, therefore, the seemingly obscure distinction between multiplicative and additive interactions of consumption with temperature change makes a significant difference for optimal climate change policy. This demonstrates how seemingly minor changes in the specification of high- temperature damages (here from multiplicative to additive) can dramatically change the climate change policies recommended by an IAM. I think the underlying reason is more or less transparent from the previous discussion of the comparison of equations (7) with (8). Fragility of policy to forms of disutility functions is a disturbing empirical finding because the outcomes of IAMs are then held hostage to basic structural uncertainty about the way in which high temperatures and high consumption interact. Furthermore, this big difference comes from a deterministic IAM (DICE with no numerical simulations of probability distributions) having a relatively high rate of pure time preference  1.5 per2. See Nordhaus (2008).

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

29

cent per year. What I show theoretically in the rest of this chapter is that if one introduces fat- tailed climate change uncertainty, along with even infinitesimal probabilities of low rates of pure time preference, the difference in optimal policies between additive and multiplicative utilities can become overwhelmingly dominant. 1.3

Deep Structural Uncertainty about Climate Extremes

In this section, I try to make a brief intuitive case for the plausibility of there being big structural uncertainties in the science of extreme climate change. I would interpret this as heuristic evidence that an IAM might be missing something important if its results do not much depend on the treatment of these big structural uncertainties. Ice core drilling in Antarctica began in the late 1970s and is still ongoing. The record of carbon dioxide (CO2) and methane (CH4) trapped in tiny ice core bubbles currently spans 800,000 years.3 The numbers in this unparalleled 800,000- year record of greenhouse gas (GHG) levels are among the very best data that exist in the science of paleoclimate. Almost all other data (including past temperatures) are inferred indirectly from proxy variables, whereas these ice core GHG data are directly observed. The preindustrial- revolution level of atmospheric CO2 (about two centuries ago) was 280 parts per million (ppm). The ice core data show that CO2 varied gradually during the previous 800,000 years within a relatively narrow range roughly between 180 and 280 ppm. Currently, CO2 is at 385 ppm and climbing steeply. Methane was never higher than 750 parts per billion (ppb) in 800,000 years, but now this extremely potent GHG, which is twenty- two times more powerful than CO2 (per century), is at 1,780 ppb. The sum total of all carbon- dioxide- equivalent (CO2 - e) GHGs is currently at 435 ppm. An even more startling contrast with the 800,000- year record is the rate of change of GHGs: increases in CO2 were below (and typically well below) 25 ppm within any past subperiod of 1,000 years, while now CO2 has risen by 25 ppm in just the last ten years. Thus, anthropogenic activity has elevated atmospheric CO2 and CH4 to levels far outside their natural range at an extremely rapid rate. The unprecedented scale and speed of GHG increases brings us into uncharted territory and makes predictions of future climate change very uncertain. Looking ahead a century or two, the levels of atmospheric GHGs that may ultimately be attained (unless decisive measures are undertaken) have likely not existed for tens of millions of years, and the speed of this change might be unique even on a time scale of hundreds of millions of years. 3. See Dieter et al. (2008), from which my numbers are taken (supplemented by data from the Keeling curve for more recent times, available online at ftp://ftp.cmdl.noaa.gov/ccg/co2/ trends/co2_mm_mlo.txt).

30

Martin L. Weitzman

Another disturbing issue concerns the ultimate temperature response to such kind of unprecedented increases in GHGs. “Climate sensitivity” is a key macro indicator of the eventual temperature response to GHG changes. It is defined as the average global surface warming in equilibrium following a sustained doubling of carbon dioxide concentrations. Other things being equal, higher values of climate sensitivity raise temperatures in every period by shifting up their dynamic trajectory, but it also takes longer for temperatures to reach any given fraction of their asymptotic limit. Left unanswered by my simplistic treatment here are many questions, including whether enough can be learned sufficiently rapidly about high climate sensitivity—relative to tremendous systemic inertias and lags—to be able to undertake realistic midcourse corrections (more on this later). A total of twenty- two peer- reviewed studies of climate sensitivity published recently in reputable scientific journals and encompassing a variety of methodologies, along with twenty- two imputed PDFs of climate sensitivity, are cited by IPCC-AR4 (Intergovernmental Panel on Climate Change [IPCC] 2007). How to aggregate climate sensitivity PDFs from various studies is currently a serious unresolved issue. The aggregated PDF should have a thinner tail than the individual studies to the extent that the PDFs from the different studies are conceptualized as independent draws from the same “correct” model specification. Against this, the aggregate PDF tail should be fattened to the extent that individual models overlap and are correlated in their mutual omission of important geophysical processes (like ice sheets) or carbon cycle processes (like methane releases). The upper- tail distribution of climate sensitivity remains poorly constrained even after thirty years of research. For what it is worth, the median upper 5 percent probability level over all twenty- two climate- sensitivity PDFs cited in IPCC-AR4 (IPCC 2007) is 6.4°C, which is the number that I use here. Only so- called fast feedback processes are included in the concept of climate sensitivity, narrowly defined. Additionally there are slow feedback components that are currently omitted from most general circulation models (mainly on the grounds that they are too uncertain to be included).4 A prime omitted components concern the potentially powerful self- amplification potential of greenhouse warming due to heat- induced releases of sequestered carbon. One vivid example is the huge volume of GHGs currently trapped in tundra permafrost and other boggy soils (mostly as methane, a particularly potent GHG). A more remote (but even more vivid) possibility, which in principle should also be included, is heat- induced releases of the even- vaster offshore deposits of CH4 trapped in the form of hydrates (aka clathrates)—which has a decidedly nonzero probability over the long run of having destabilized methane seep into the atmosphere if water temperatures 4. The distinction between “fast feedbacks” and “slow feedbacks” is explained in Hansen et al. (2008).

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

31

over the continental shelves warm just slightly. The amount of CH4 involved is huge although it is not precisely known. Most estimates place the carbonequivalent content of methane hydrate deposits at about the same order of magnitude as all fossil fuels combined. Over the long run, a CH4 outgassingamplifier process could potentially precipitate a disastrous strong- positivefeedback warming. Thus, the possibility of a climate meltdown is not just the outcome of a mathematical theory, but has a real physical basis. Other examples of an actual physical basis for catastrophic outcomes could be cited, but this one will do here. The preceding methane- release scenarios are examples of slow carbon cycle feedback effects that I think should be included in the interpretation of a climate- sensitivity- like concept that is relevant for the economics of uncertain extremes. The main point here is that the PDF of fast plus slow feedback processes has a tail much heavier with probability than the PDF of slow feedback processes alone. Extraordinarily crude calculations suggest that, when slow and fast feedback processes are combined, the probability of eventually exceeding 10°C from anthropogenic doubling of CO2 is very roughly 5 percent, which presumably corresponds to a scenario where CH4 and CO2 are outgassed on a large scale from degraded permafrost soils, wetlands, and clathrates.5 To summarize the major implication for this paper, the economics of climate change consists of a very long chain of tenuous inferences fraught with big uncertainties in every link, of which anthropogenic climate sensitivity (incorporating fast and slow feedbacks) is but one component. The uncertainties begin with unknown base- case GHG emissions; then they are compounded by big uncertainties about how available policies and policy levers will transfer into actual GHG emissions; compounded further by big uncertainties about how GHG flow emissions accumulate via the carbon cycle into GHG stock concentrations; compounded by big uncertainties about how and when GHG stock concentrations translate into global mean temperature changes; compounded by big uncertainties about how global mean temperature changes decompose into regional climate changes; compounded by big generic uncertainties about the appropriate structure of damage functions and how to discount their disutilities; compounded by big uncertainties about how adaptations to, and mitigations of, climatechange damages are translated into welfare changes at a regional level; compounded by big uncertainties about how future regional utility changes are aggregated—and then how they are discounted—to convert everything into expected- present- value global welfare changes. The result of this lengthy cascading of big uncertainties is a reduced form of truly extraordinary uncertainty about the aggregate welfare impacts of catastrophic climate change, which mathematically is represented by a PDF that is spread out and 5. These calculations are explained in Weitzman (2009a).

32

Martin L. Weitzman

heavy with probability in its tails. The fat tail of the PDF of overall welfare is the reduced form that concerns economic analysis, not the PDF of climate sensitivity per se, which is but one limited illustrative example representing the overall science component of structural uncertainty. In other words, uncertain climate sensitivity serves in this chapter as a particular example of uncertain climate science as a whole, which itself is a subset of overall uncertainty. 1.4

Why is Climate Sensitivity So Unpredictable?

Taking climate sensitivity as a metaphor for climate science, it appears to bother scientists a lot that, even after some three decades of intensive research, essentially no progress has been made on excluding very high values of climate sensitivity (substantially higher than 4.5°C, say). A longstanding informal explanation for this state of affairs—which focuses on the mechanism by which small individual forcing feedbacks are amplified into a large climate- sensitivity multiplier—was formalized recently in a culminating Science article by Roe and Baker (2007; henceforth R&B) entitled “Why is Climate Sensitivity So Unpredictable?,” from which the title of this section was taken.6 A troubling economic implication of R&B, which I explain later in more detail, is that the asymptotic tail probability of large climate sensitivity appears to be declining toward zero so slowly, relative to its impacts, that this fat- tail aspect seems like it should play a significant role in welfare analysis. Here I reformulate at a high level of abstraction the analytical essence of a R&B- style explanation to emphasize that it implicitly contains a theoretical argument pointing toward very fat tails of climate sensitivity. In other words, I am making the simple point that the same (or at least a very similar) argument used by climate scientists to explain why high values of climate sensitivity cannot be excluded contains within itself the seeds of a more general argument not just for fat tails, but for very fat tails, which spread the PDF so far apart that its variance is effectively infinite. By focusing on the feedback- multiplier nub of this argument, I naturally pass over a great many important details of the underlying science. In estimating climate sensitivity (denoted ), feedbacks are everything. The ultimate temperature response of a climate system to imposed GHG shocks is unknown mostly because the exogenous initial forcing is amplified by uncertain endogenous feedback factors like albedo, water vapor, clouds, and so forth. These feedback factors have complex, nonlinear, and even chaotic features. Overarching this messiness, R&B (2007) argue, feedbacks still combine additively, and linear systems analysis is still a useful way of seeing the forest for the trees. Climate sensitivity in this R&B- style view is 6. See Roe and Baker (2007). There is a long list of predecessors, ranging from Wigley and Schlesinger (1985), or maybe even earlier, to Allen et al. (2006).

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

33

a derived concept able to be portrayed abstractly as an amplifier (or multiplier) for a forcing impulse to a linear feedback process. More basic in this process than the multiplier- amplifier  is the aggregate feedback factor or coefficient f. Not only does f act on the original CO2 forcing, but it also acts on the results of its own forcing action, and so forth, ultimately causing an infinite series of feedback loops as described by the differential equation in the next section. The aggregate feedback coefficient f has a critical additivity property in its components: n

(12)

f  ∑ f˜j, j =1

where each of the n primitive f˜j represents a feedback subfactor, such as the albedo, water vapor, or clouds previously alluded to. In the R&B- style worldview, the total feedback- forcing factor f is considered more fundamental than the climate- sensitivity multiplier  because f scales additively in its primitive subcomponents f˜j (whereas  does not scale additively in j  0/[1 – f˜j ]) and because each f˜j is (at least in principle) directly measurable, in the laboratory or in the field. By contrast,  is observable only indirectly, as the equilibrium limit of an iterative multiplier process that requires (at least in principle) the passage of an infinite number of multiplier rounds over real time by the formula (13)

∞ 0   0 ∑ ( f )i   . 1 f i =1

If each primitive f˜j is an independently distributed random variable (RV) and n is large, then from equation (12) and the central limit theorem a case could be made that the RV f might be approximately normally distributed. A normal PDF for f is the prototype case considered in the R&B- style modeling tradition. To draw out the generic implications of a R&B- style explanation for the derived fatness of the upper tail of  requires showing formally how the argument generalizes from the normal to essentially any reasonable PDF of f. The base- case normal PDF in the R&B- style tradition is presumably truncated from above at f  1, or else it could be argued that the implied unstable dynamics would have produced a runaway feedback amplification at some time in the past. However, it is far from clear how exactly this truncation at f  1 (of the normal or any other PDF) is to be carried out in practice or even how it is to be conceptualized—and the formal R&B- style argument for   0 /(1 – f ) having a PDF with a fat upper tail is left somewhat dangling on this point. Let the PDF of the RV f be ( f ) with upper support at f  1. The spirit of a R&B- style explanation is that high values of f ≈ 1 – having “small but nonnegligible” probability get nonlinearly skewed upward into a fat- upper- tail PDF of   0 /(1 – f ). Without further ado, I assume that a fair translation

34

Martin L. Weitzman

of the idea that high values of f ≈ 1 – have “small but nonnegligible” probability is that the PDF ( f ) has the properties (1)  0, (1)  0.

(14)

It is important here to understand that I am not assuming a positive point probability of occurrence for a feedback value f  1, in which case the results to follow would be trivial. I am not even assuming that the probability density at f  1 is positive. I am only assuming in equation (14) that in the limit the probability density of f changes linearly (from an initial value of zero) within an arbitrarily small neighborhood of f ≈ 1 –. With the usual Jacobian change- of- variable transformation ( f )df → ψ()d  [ f ()] f ()d, applied to   0/(1 – f ), the derived PDF of  is  1 ψ()  0  1  0 2  

冢

(15)

冣

for all  0. The mean of  is given by the expression E() ⬅ lim

(16)

M →∞

冕

M 0

ψ()d  ,

while its variance is M

(17)

V() ⬅ lim % 冕 [  E()]2 ψ()d  , M →∞

0

where the integral in equation (17) is blowing up essentially because, from equation (15), the integrand inside of equation (17) approaches 1/ as  → , making the integral (17) approach –(l) ln M as M → . The significance of equation (17) for economic policy is not subtle. As climate sensitivity goes, so goes the eventual mean planetary temperature response to increased GHGs. While global warming is just one example of a fat- tailed f →  feedback- multiplier process, it is special because of the enormous potential damages to worldwide welfare associated with very large values of . If additive economic damages increase at least as fast as quadratically in temperatures, as in equation (2), then equation (17) indicates that the probability- weighted expected value of climate damages is infinite. A standard criticism of my (or any) oversimplified reliance on the timeindependent long- run equilibrium concept of climate sensitivity is that the catastrophically high temperature values will materialize (if they materialize at all) only in the remote future. If one brought back the time element by focusing more on the transient dynamics and less on the stationary limit, this line of argument goes, short- to medium- term concerns would dominate— and the climate sensitivity issue might recede. Some have even interpreted the R&B- style explanation of fat- tailed climate sensitivity as signifying that

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

35

the concept itself has run into diminishing returns, and the scientific community should essentially “call off the quest” of trying to make more precise estimates of  in favor of concentrating greater effort on analyzing short- to medium- term temperature dynamics and implications. In the rest of this chapter, I will show that when one formalizes the uncertain trajectory of temperature dynamics, along with analyzing carefully the issue of discounting utility (or disutility) under uncertain rates of pure time preference, then the long- run behavior of the system can, in principle, continue to play a significant role in economic analysis and policy discussion. 1.5

A Dynamic Aggregative Model of Global Warming

This section compresses into a single differential equation what is arguably the simplest meaningful deterministic- dynamic model of the physical process of global warming.7 Of course, this particular one- differential- equation model cannot possibly capture the full complexity of climate change. However, I think that the highly aggregated approach taken here is realistic enough to serve as a springboard for meaningful discussions of some basic climate- change issues, which, for the purposes of this chapter, may actually be clarified when tightly framed in such stark simplicity. Factors that affect climate change are standardly segregated into “forcings” and “feedbacks.” A climate forcing is a direct, primary, or exogenous energy imbalance imposed on the climate system, either naturally or by human activities. Examples include changes in solar irradiation (the prototype, in whose units all other forcings may be expressed), volcanic emissions, deliberate land modification, or anthropogenic changes in atmospheric stocks of greenhouse gases, aerosols, and their precursors. (The radiative forcing from CO2 happens to be proportional to the logarithm of its atmospheric concentration, but this is not true in general for all GHGs.) A climate feedback is an indirect, secondary, or endogenous radiative imbalance that amplifies or dampens the climate response to an initial forcing. An example is the increase in atmospheric water vapor that is induced by an initial warming due to rising CO2 concentrations, which then acts to amplify the warming through the greenhouse properties of water vapor, further accelerating the process. Suppose, for simplicity, that in pre-Industrial Revolution times (t  0), the planetary climate system had been in a state of (relative) equilibrium at a constant temperature with constant radiative forcing and no radiative imbalance. Let F(t) stand for radiative forcing at time t. Normalize F(t)  7. This supersimple diagnostic energy- balance model is sprinkled throughout the scientific literature and appears formally in, for example, Andrews and Allen (2008) or Roe (2009), both of which contain further references to it, including who created it and more realistic extensions of it. My only possible originality here is in expositing this basic one- differential- equation model to a broader audience, primarily economists.

36

Martin L. Weitzman

T(t)  0 for t  0. Imagine, in a thought experiment, that a sustained radiative imbalance of constant magnitude RΣ has been additionally imposed. (Whether this constant additional radiative imbalance RΣ is itself exogenous or endogenous is not relevant in this context because only the reducedform total imbalance matters for its expository role here.) Let T(t) be the temperature response at time t 0. If the earth were a blackbody planet, with no atmosphere and no further feedbacks, the long- run “reference” nonfeedback temperature response as t →  would be T(t) → 0RΣ, where 0 is the feedback- free constant defined by the fundamental physics of a blackbody reference system as described by the Stefan-Boltzman law. Even in richer, more- realistic situations with atmospheric feedbacks and complicated dynamics, other things being equal, it is not a terrible approximation that at any time the temperature moves with an instantaneous velocity roughly proportional to the reference imbalance. This means that the linearized differential equation of temperature motion is (18)

冤

冥

1 T(t) T˙ (t)   R Σ(t)   , k 0

where the positive coefficient k in equation (18) represents the aggregate thermal inertia or effective capacity of the system as a whole to absorb heat. In this application, k essentially stands for the overall planetary ability of the oceans to take up heat. The full temperature dynamics of an idealized nonblackbody planetary system can now most simply be described as follows. Count time in the conventional modeling format where the present corresponds to t  0. At any time t 0, suppose that the system is subjected to an exogenously imposed additional radiative forcing of F(t) (relative to its pre-Industrial Revolution equilibrium rest state of zero). In the application here, the exogenously imposed additional radiative forcing is essentially the logarithm of the relative increase of atmospheric CO2 over pre-Industrial Revolution levels. Without loss of generality, it is convenient throughout this chapter to normalize the unit of forcing to correspond to a doubling of CO2. If G(t) is the concentration of atmospheric carbon- dioxide- equivalent (CO2 - e) GHGs at time t (in parts per million [ppm]), and G (≈280ppm) is the pre-Industrial Revolution CO2 - e concentration of atmospheric GHGs, then (19)

冢 冣

G(t) 1 F(t)   ln . ln 2 G

The trajectory of exogenous (or primary) radiative forcings {F(s)} for 0  s  t (here essentially standing for past anthropogenic increases in atmospheric GHG stocks) causes temperatures to rise over time, which induces feedbacklike changes in secondary radiative imbalances (such as cloud formation, water vapor, ice albedo, lapse rates, and so forth). Lumped together, these “secondary” radiative imbalances are typically more powerful

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

37

in ultimate magnitude than their “primary” inducers. Let the endogenously induced overall radiative imbalance at time t be denoted RI (t). Let the total change in radiative imbalances at time t be denoted RΣ (t). Then RΣ(t)  F(t)  RI (t).

(20)

In the problem at hand, the temperature change T(t) induces a (comparatively- fast- acting, relative to equation (18) endogenous radiative imbalance RI (t) according to the formula RI (t) 

(21)

f T(t), 0

where the (linear) feedback factor f is a basic parameter of the system. Not only does f act on the original CO2 forcing, but it also acts on the results of its own forcing action, and so forth, ultimately causing an infinite series of feedback loops. As mentioned, the relevant feedback factors in climate change involve cloud formation, water vapor, albedo, and many other effects. A key property of linear feedback factors is that (as with radiative forcings or radiative imbalances) the various components and subcomponents can be aggregated simply by adding them all up because they combine additively. Plugging equations (21) and (20) into equation (18) then yields, after simplification, the basic differential equation (22)

T˙ (t) 

1 1 f F(t)  T(t) k 0

冤

冥

with the initial conditions F(0)  T(0)  0. The closed- form solution of equation (22) is (23)

1 T(t)  k

t

∫ F(s)exp冤(s  t)冢 0

1 f 0

冣冥ds.

The oversimplifications of physical reality that have gone into the onedifferential- equation temperature change trajectory (22) are numerous. As just one example, the parameters that appear in equations (22) or (23) are not true constants because they might covary over time in complicated ways that this simplistic formulation is incapable of expressing. The only defence of this ultramacro approach is a desperate need for analytical simplicity in order to see the forest for the trees. It seems fair to say that equation (22) captures the dynamic interplay of forces along a global warming path decently enough for the purposes at hand—and almost surely better than any alternative formula based on one simple linear differential equation. Even accepting the enormous oversimplifications of reality that go into equations like (22) or (23), there remain massive uncertainties concerning the appropriate values of the structural parameters. For simplicity, the critical feedback parameter f is chosen to be the only uncertainty, but it should be appreciated that the relevant values of k and of forcings {F(t)} are also very

38

Martin L. Weitzman

uncertain and covary in ways that are not fully represented here, even leaving aside the model itself being an uncertain simplification of a much more complicated reality. Generally speaking, additional uncertainty strengthens the message of this chapter. Just glancing at equation (23) is suggestive of why it is so difficult in practice to infer f directly from data. The record of past forcing histories is extremely noisy, and such components as aerosol concentrations are notoriously difficult to identify. Furthermore, it is readily shown that the firstorder response of a system like equation (23) to a change in forcings does not involve long- run parameters like f at all but more centrally concerns the overall ability of the oceans to take up heat as embodied in the thermal inertia coefficient k. The parameter k itself is not very well known in this aggregative context and can be interpreted as having time varying values for different stages of different heat- absorbing processes. It is statistically very difficult to distinguish between a high-f low-k world and a low-f high-k world. To be able to infer f at all precisely would require a long and fairly accurate time series of past natural forcings along with a decent knowledge of the relevant thermal inertias—none of which are readily available. From this, from the difficult politics of the situation, and from the very long pipeline commitment of atmospheric CO2 stocks, it follows that prospects for a meaningful “wait and see” reactive policy for GHG flow emissions may be quite limited. For notational convenience and analytical sharpness, I restrict the situation here to the most basic case of the dynamic temperature reaction to a step function forcing represented by an instantaneous doubling of CO2 - e GHGs: F(t) ⬅ 1 for t  0 and F(t) ⬅ 0 for t  0. Then equation (23) simplifies down to (24)

0 1 f T(t | f )   1  exp   t . 1 f k0

冦

冤 冢

冣 冥冧

Note that the right- hand side of equation (24) approaches t/k when going to the limit as f → 1. This implies from equation (24) that T(t | f )  T(t | 1)  t/k for f  1, implying that the bounded random variable (RV) T(t | f ) must have finite (but increasing) variance no matter what is the PDF of f satisfying equation (14). Equilibrium climate sensitivity is defined as (25)

 ⬅ lim T(t) t→ ∞

and it is apparent from applying equation (25) to equation (24) that (26)

0  , 1 f

which is one of the most basic relationships of climate change. Conventional as if deterministic point estimates might be 0 ≈ 1.2, f ≈ .65,  ≈ 3,

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

39

with the standard deviation of f approximately f ≈ .13. Relevant values of k might vary widely in this aggregate context, depending on the time scale of the heat absorption process. Other things being equal, higher values of f shift up temperatures T all along the trajectory (24), but with higher values of f it also takes a longer response time to reach any given fraction of the asymptotic value  represented by equation (25), (26). The third link in my chain of reasoning concerns the welfare and policy implications of the infinite limiting variance of  from the second link. Under any foreseeable technology, elevated stocks of CO2 are committed to persist for a very long time in the atmospheric pipeline. Ballpark estimates imply that, for every unit of CO2 anthropogenically added to the atmosphere, ≈70 percent remains after 10 years, ≈35 percent remains after 100 years, ≈20 percent remains after 1,000 years, ≈10 percent remains after 10,000 years, and ≈5 percent remains after 100,000 years.8 It can also take a long time to learn about looming realizations of uncertain, but irreversible, climate changes. Thus, the CO2 stock inertia, along with slow learning, makes it unreliable to react to unfolding disasters by throttling back CO2 flow emissions in time to avert an impending catastrophe. Here I just simplistically assume that the planet will never be able to react to bad future scenarios by stabilizing atmospheric concentrations below a doubling of CO2 - e GHG concentrations relative to pre-Industrial Revolution levels. Such a stark approach may be an acceptable proxy for reality in the context of the message I am trying to convey because it seems to me, alas, that CO2 - e GHGs ≈560ppm are essentially unavoidable within the next half century or so and will plausibly remain well above this level for one or two centuries thereafter, no matter what new information is received in the meantime. This represents an extreme and perhaps unrealistic interpretation, but the modeling strategy of this chapter is to lay out the essential structure of my argument as simply as possible, leaving more realistic refinements for later work. 1.6

How Should Climate Change Disutilities be Discounted?

The analysis of last section showed that pushing a R&B- style explanation of fat- tailed climate sensitivity all the way to its logical conclusion implies a PDF having infinite variance. With an additive quadratic disutility of temperature change, this implies an infinite loss of expected welfare, but this infinite loss is occurring at an infinitely remote future time. The obvious next question is: what happens to expected present discounted welfare when the disutility damages of high temperatures are discounted at the appropriate rate of pure time preference? Suppose that the damages of temperature changes are quadratic in T and of the additively separable form UA(C,T )  –(1/C  1  T 2) from 8. See Archer (2007, 122–24) and the further references he cites.

40

Martin L. Weitzman

equation (2), which has the analytical convenience of allowing one to calculate separately the disutility impact of temperature changes, irrespective of the time trajectory of consumption. The superstrong result that follows depends critically upon additively separable utility and would not hold (in such strength) for the multiplicative form (1): UM (C,T )  –(1  T 2)/C. The temperature change at time t conditional on the realization f is T(t | f ) given by formula (24). The expected disutility damages at time t are then given (up to an inessential multiplicative constant) by the expression 1

(27)

DE (t) 

∫ [T(t | f )]2 ( f )df,

−∞

and it is straightforward to show that DE(t) increases monotonically over time, approaching a limit of , which is consistent with climate sensitivity having an infinite variance in equation (17). If the rate of pure time preference used for discounting future utiles or disutilities is 0, and if the future is artificially truncated at time horizon H, then expected present discounted disutility is H

(28)

D∗( ; H)  ∫ DE(t)e t dt. 0

It is essential to realize that the number being discussed here for discounting future disutilities is the so- called rate of pure time preference or utility discount rate, an elusive concept that is subjective and not directly observable. The “utility discount rate” is not the much- more- familiar number that is used to discount ordinary goods (r   g by the Ramsey formula) and which is identified with the everyday concept of an interest rate. It is much harder to argue that this utility discount rate should be significantly above zero than it is to make such an argument for the “goods interest rate” r, which is far more directly tied to observed market rates of return on capital that are significantly positive.9 The next obvious question is: what are appropriate values of and H to use in evaluating equation (28)? For H the answer is relatively easy: by longstanding economic logic and practice, in principle the horizon ought to be infinite, and equation (28) should be evaluated by taking the limit at H → . The more difficult and more controversial issue concerns the appropriate rate of pure time preference to be used for discounting intergenerational disutility damages from future climate change. The question here is: what is ? I think an honest direct answer is somewhere between zero and very roughly about 1 percent per year—some people might have opinions, but nobody really knows. I think it is also fair to point out that a notable minority of some very distinguished economists believe that the appropriate rate of pure time preference for discounting intergenerational utilities 9. Dasgupta (2007) has an insightful discussion of some of the main issues here.

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

41

generally, and disutility damages from future climate change particularly, should be arbitrarily close to zero. Without taking sides directly on this issue, I approach the problem indirectly by postulating some given distribution of subjective probabilities representing overall “degrees of belief ” in the appropriate value of to be plugged into equation (28).10 Let the subjective PDF of the RV be h( ), with a lower support at  0. The formal treatment of the RV that follows in this section parallels the formal treatment of the RV f from the last section. In a spirit of giving at least some limited voice to the opinion that the rate of pure time preference for intergenerational discounting might be arbitrarily close to zero, I assume that low values of ≈ 0 have “small but nonnegligible” probability in the sense that the PDF h( ) obeys (29)

h(0)  0, h(0) 0,

which is the analogue here of condition (14). Essentially, uncertainty concerning the possibility of a small rate of pure time preference near zero is naturally amplified into uncertainty about very large values of the present discounted value  of a unit flow by a highly nonlinear transformation of the form   1/ . It is critical to understand here that I am not assuming a positive point probability of occurrence for a zero rate of pure time preference, in which case the main result of this chapter would be trivial. I am not even assuming that the probability density of a zero rate of time preference is positive. I am only assuming in equation (29) that the PDF of increases linearly (from an initial value of zero) within an arbitrarily small neighborhood of ≈ 0. There are now two RVs: f and . What might be called the “expected expected” present discounted disutility of temperature change is H

(30)

D∗∗(H )  ∫ D∗( ; H)h( )d . 0

The final question to be addressed here is: what happens to D∗∗(H) in the limit as H → ? The answer is not obvious. Other things being equal, as the time horizon recedes, the expected disutility damage from more- variable future temperatures is increasingly dominated by the limiting infinite variance of climate sensitivity. However, other things also being equal, discounting at a positive rate counteracts this infinite- variance asymptote. Yet a third wild card here is that pure time preference itself is a legitimately unknown RV in this context, with some “small but nonnegligible” probability of being close to zero, which tends to favor lower effective discount rates at longer horizons—again, other things being equal.11 The value of the “expected expected” present discounted disutility of temperature change in equation 10. The logic of this position is spelled out further in Weitzman (2001). 11. This effect is described in Weitzman (1998).

42

Martin L. Weitzman

(30), which emerges from this pulling of different forces in different directions, is the main result of the chapter that is exposited in the next section. 1.7

A Dismal Proposition

The following “dismal proposition” hints that fat- tailed infinite- variance climate sensitivity may have economic ramifications conceivably impacting current policy analysis. THEOREM 1. In the above model with “small but nonnegligible” probabilities, lim D∗∗(H )  .

(31)

H→∞

PROOF. Consider the expression ∞

(32)

∫[T(t; f )]2 exp( t)dt 0

冢

0   1 f

∞

1 f t exp( t)dt. 冣 ∫冦1  exp冤冢 k 冣 冥冧 2

2

0

0

By brute- force integration, the right- hand side of equation (32) is shown to be proportional to (33)

1/{ [  (1  f )/k0][  2(1  f )/k0]}.

If it were true that (34)

lim

→ 0 , f → 1

 冧  , 冦 [

 (1  f )/k ][  2(1  f )/k ] [( f )][h( )] 0

0

then it would follow that ∞

冦∫ [T(t; f )] exp( t)dt冧  ,

E , f

(35)

2

0

which in turn would imply equation (31). But expression (34) must hold because (36)

lim

→ 0 , f → 1

[(1  f )(1)][ h(0)]

 冧  , 冦 [

 (1  f )/k ][  2(1  f )/k ] 0

0

which concludes this streamlined proof. ■ Were quadratic disutility to be discounted at a zero rate of time preference, then it would be straightforward that an infinite- variance tail of eventual temperature change would have a big impact on present discounted expected welfare. What is perhaps surprising is that this high- impact result can continue to hold even when the probability density at a zero rate of time

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

43

preference is zero and is merely increasing linearly (in the small) with time preference. The infinite limit of the theorem is coming in main part from a perhaps counterintuitive implication of equation (29) that, with uncertain rates of pure time preference, utilities in the far- distant future are discounted at the lowest possible rate of pure time preference, here zero.12 This holds even when the probability of ≈ 0 is infinitesimal. In a sense, the model has been reverse engineered via equation (29) to put weight on the limiting PDF of temperature changes, whose variance approaches the infinite variance of the PDF of climate sensitivity. With economic damages quadratic in temperatures, expected present discounted disutility then approaches infinity as the horizon recedes. Of course any interpretation must be based on an assessment of the model’s overall assumptions. The model of this chapter is really more of a suggestive example than a fully general formulation, and an example that has been reverse engineered at that. The theorem depends on a conjunction of several basic assumptions, none of which is beyond criticism. Analyzing what happens for H →  stretches the logic even further. Still, taken as a whole (and even admitting that it has been somewhat rigged), I think this “dismal proposition” makes it somewhat less easy to dismiss the significance of unpredictable climate sensitivity on the grounds that high values will have impact only in the distant future. Theorem 1 indicates that the willingness to pay to avoid climate change is unbounded. There are several possible ways to escape this disturbing paradox of infinity. The troubling infinite limit is technically eliminated by imposing ad hoc inequality constraints like H  200 years, or T  7°C, or f  .99, or  .001, or so forth. However, removing the  symbol in this way does not truly eliminate the underlying problem because it then comes back to haunt in the form of an arbitrarily large expected- present- discounted disutility, whose exact value depends sensitively upon obscure bounds, truncations, severely dampened or cutoff prior PDFs, or whatever other formal mechanisms have been used to banish the  symbol. The take- away message here is that reasonable attempts to constrict bad- tail fatness can leave us with uncomfortably big numbers whose exact value may depend nonrobustly upon artificial constraints or parameters, the significance of which we do not honestly comprehend. Theorem 1 should, therefore, be taken only figuratively as holding for some “uncomfortably big number”—but not for infinity. A reader interested in understanding more about how the infinite limit of the “dismal proposition” is to be interpreted and applied in a finite world should consult the much fuller discussion of this set of issues in Weitzman (2009a) and Weitzman (2009b). Here I restrict myself just to commenting on the widespread notion that the extreme realizations being described in the tails are so improbable that they can effectively be ignored. 12. For an exposition of this logic, see Weitzman (1998).

44

Martin L. Weitzman

One may not legitimately discard the bad tail of the PDF of a disaster on the grounds alone that the probabilities are “too small to matter.” Such de minimis truncation requires some asymptotic argument along the lines that the limiting product of the decreasing probability times the increasing disutility is “too small to matter.” The most practical way to avoid this vexing tail- evaluation issue is when there exists strong a priori knowledge that meaningfully restrains the extent of total damages. If a particular type of idiosyncratic uncertainty affects only one small part of an individual’s or a society’s overall portfolio of welfare, which is the usual case, exposure is naturally limited to that specific component and bad- tail fatness is not such a paramount concern. However, some very few but very important real- world situations have potentially unlimited exposure due to structural uncertainty about their potentially open- ended catastrophic reach. In these unusual situations, there is no choice but to evaluate somehow or other the limiting product of probability times disutility. Climate change potentially affects the whole worldwide portfolio of utility by threatening to drive all of planetary welfare to disastrously low levels in the most extreme scenarios. This tail- evaluation feature, which is essentially ignored by most conventional IAMs, can understandably dominate the economics of climate change. Such a feature makes an economic analysis of climate change look and feel uncomfortably subjective, but, at least in the formulation of this chapter, it is the way things can be with fat tails and unlimited liability. Because climate change catastrophes develop slower than some other potential catastrophes, there is perhaps more chance for learning and midcourse corrections with global warming, relative to some other catastrophic scenarios. The possibility of “learning by doing” may well be a more distinctive feature of global warming disasters than of some other disasters and, in that sense, deserves to be part of an optimal climate- change policy. The other horn of this dilemma, however, is that the ultimate temperature responses to CO2 stocks have tremendous inertial pipeline- commitment lags that are very difficult to reverse once they are in place. This nasty fact can be brutal on illusions about the easy corrective potential of “wait and see” reactive policies. As I already noted, it seems implausible to me that ultimate stabilized values of GHGs will end up being much less than twice pre-Industrial Revolution values, no matter what realistic future responses to global warming are undertaken. Reacting to an impending climate disaster by changing a CO2 emissions- flow instrument (to control the CO2 stock accumulation inducing the disaster) seems offhand like using an outboard motor to maneuver an ocean liner away from an impending collision with an iceberg. The role of learning and midcourse corrections is a subject worthy of further detailed study, the outcome of which could potentially soften all of my conclusions. However, a conservative position, at least for the time being, might be to consider that by the time we learn that a climate change disaster is impending, it may be too late to do much about it.

Additive Damages, Fat-Tailed Climate Dynamics, and Discounting

1.8

45

Concluding Comments

Issues of uncertainty and discounting are fundamental to any economic analysis of climate change. This chapter combines together three forms of structural uncertainty: how to formulate damages, how to discount these damages, and how to express future temperature dynamics. The chapter shows that the single most widespread scientific explanation of why climate sensitivity is so uncertain at the upper end contains within itself a generic argument in favor of a very fat upper tail of temperature changes. When this is merged with an additively separable damages function and a rate of pure time preference that is unknown but might conceivably be close to zero, the combination can in principle dominate an economic analysis of climate change. Such a message is not intended to cause despair for the economics of climate change nor to negate the need for further study and numerical simulations to guide policy. The message is just a cautionary note that this particular application of cost- benefit analysis to climate change seems more inherently prone to being dependent on subjective judgements about structural uncertainties than most other, more ordinary, applications.

References Allen, M., N. Andronova, B. Booth, S. Dessai, D. Frame, C. Forest, J. Gregory, G. Hegerl, R. Knutti, C. Piani, D. Sexton, and D. Stainforth. 2006. Observational constraints on climate sensitivity. In Avoiding dangerous climate change, ed. H. J. Schellnhuber, 281–89. Cambridge, UK: Cambridge University Press. Andrews, David G., and Myles R. Allen. 2008. Diagnosis of climate models in terms of transient climate response and feedback response time. Atmospheric Science 9:7–12. Archer, David. 2007. Global warming: Understanding the forecast. Oxford, UK: Blackwell Publishing. Dasgupta, Partha. 2007. Commentary: The Stern Review’s Economics of Climate Change. National Economic Review 199:4–7. Dieter, Lüthi, Martine Le Floch, Bernhard Bereiter, Thomas Blunier, Jean-Marc Barnola, Urs Siegenthaler, Dominique Raynaud, Jean Jouzel, Hubertus Fischer, Kenji Kawamura, and Thomas F. Stocker. 2008. High- resolution carbon dioxide concentration record 650,000–800,000 years before present. Nature 453:379–82. Hansen, James, Makiko Sato, Puskher Karecha, David Beerling, Valerie MassonDelmotte, Mark Pagani, Maureen Raymo, Dana L. Royer, James C. Zachos. 2008. Target atmospheric CO2: Where should humanity aim? Open Atmospheric Science Journal 2:217–31. Intergovernmental Panel on Climate Change (IPCC). 2007. Climate change 2007: The physical science basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC-AR4). Cambridge, UK: Cambridge University Press. Nordhaus, William. 2008. A question of balance. New Haven, CT: Yale University Press.

46

Martin L. Weitzman

Roe, Gerard. 2009. Feedbacks, timescales, and seeing red. Annual Review of Earth and Planetary Sciences 37:93–115. Roe, Gerard H., and Marcia B. Baker. 2007. Why is climate sensitivity so unpredictable? Science 318:629–32. Sterner, Thomas, and U. Martin Persson. 2008. An even sterner review: Introducing relative prices into the discounting debate. Review of Environmental Economics and Policy 2 (1): 61–76. Weitzman, Martin L. 1998. Why the far- distant future should be discounted at its lowest possible rate. Journal of Environmental Economics and Management 36 (3): 201–8. ———. 2001. Gamma Discounting. American Economic Review 91(1): 260–71. ———. 2009a. On modeling and interpreting the economics of catastrophic climate change. Review of Economics and Statistics 91:1–19. ———. 2009b. Reactions to Nordhaus. http://www.economics.harvard.edu/faculty/ weitzman/files/ReactionsCritique.pdf. Wigley, Thomas M. L., and Michael E. Schlesinger. 1985. Analytical solution for the effect of increasing CO2 on global mean temperature. Nature 318:649–52.

2 Modeling the Impact of Warming in Climate Change Economics Robert S. Pindyck

2.1

Introduction

Any economic analyses of climate change policy must include a model of damages, that is, a relationship that translates changes in temperature (and possibly changes in precipitation and other climate- related variables) to economic losses. Economic losses will, of course, include losses of gross domestic product (GDP) and consumption that might result from reduced agricultural productivity or from dislocations resulting from higher sea levels but also the dollar- equivalent costs of possible climate- related increases in morbidity, mortality, and social disruption. Because of the lack of data and the considerable uncertainties involved, modeling damages is probably the most difficult aspect of analyzing climate change policy. There are uncertainties in other aspects of climate change policy—for example, how rapidly greenhouse gases (GHGs) will accumulate in the atmosphere absent an abatement policy, to what extent and how rapidly temperature will increase, and the current and future costs of abatement—but damages from climate change is the area we understand the least. It must be modeled, but it is important to understand the uncertainties involved and their policy implications. Most quantitative economic studies of climate change policy utilize a “damage function” that relates temperature change directly to the levels Robert S. Pindyck is the Bank of Tokyo-Mitsubishi Professor of Economics and Finance at the Sloan School of Management, Massachusetts Institute of Technology, and a research associate of the National Bureau of Economic Research. My thanks to Andrew Yoon for his excellent research assistance and to Larry Goulder, Charles Kolstad, Steve Salant, Martin Weitzman, and seminar participants at the National Bureau of Economic Research (NBER) Conference on Climate Change and at Massachusetts Institute of Technology (MIT) for helpful comments and suggestions.

47

48

Robert S. Pindyck

of real GDP and consumption. Future consumption, for example, is taken to be the product of a loss function and “but- for consumption,” that is, consumption in the absence of any warming. With no warming, the loss function is equal to 1, but as the temperature increases, the value of the loss function decreases. This approach is reasonably simple in that any projected path for temperature can be directly translated into an equivalent path for consumption. In other words, consumption at time t depends on temperature at time t. Given a social utility function that “values” consumption, one can then evaluate a particular policy (assuming we know its costs and can project its effects on GHG concentrations and temperature). In a recent paper (Pindyck 2009), I have argued that on both theoretical and empirical grounds, the economic impact of warming should be modeled as a relationship between temperature change and the growth rate of GDP as opposed to the level of GDP. This means that warming can have a permanent impact on future GDP and consumption. It makes the analysis somewhat more complicated, however, because consumption at some future date depends not simply on the temperature at that date, but instead on the entire path of temperature and, thus, the path of the growth rate of consumption, up to that date. The issue I address in this chapter is the extent to which these two different approaches to modeling damages—temperature affecting consumption directly versus temperature affecting the growth rate of consumption—differ in terms of their policy implications. This issue must be addressed in the context of uncertainty, which is at the heart of climate change policy. It is difficult to justify the immediate adoption of a stringent abatement policy based on an economic analysis that focuses on “most likely” scenarios for increases in temperature and economic impacts and uses consensus estimates of discount rates and other relevant parameters.1 But one could ask whether a stringent policy might be justified by a cost- benefit analysis that accounts for a full distribution of possible outcomes. In Pindyck (2009), I showed how probability distributions for temperature change and economic impact could be inferred from climate science and economic impact studies and incorporated in the analysis of climate change policy. The framework I used, which I use again here, is based on a simple measure of “willingness to pay” (WTP): the fraction of consumption w∗() that society would be willing to sacrifice, now and throughout the future, to ensure that any increase in temperature at a specific horizon H, TH , is limited to . It is important to understand the limitations of this approach. Whether the reduction in consumption corresponding to some w∗() is sufficient to limit warming to  is a separate question that is not addressed; in effect, WTP applies to the “demand” side of policy analysis. 1. The Stern Review (Stern 2007) argues for a stringent abatement policy, but as Nordhaus (2007), Weitzman (2007), Mendelsohn (2008), and others point out, it makes assumptions about temperature change, economic impact, abatement costs, and discount rates that are outside the consensus range.

Modeling the Impact of Warming in Climate Change Economics

49

The advantage of this approach, however, is that there is no need to project GHG emissions and atmospheric concentrations or estimate abatement costs. Instead the focus is on uncertainties over temperature change and its economic impact. My earlier paper was based on what I called the current “state of knowledge” regarding global warming and its impact. I used information on the distributions for temperature change from scientific studies assembled by the Intergovernmental Panel on Climate Change (IPCC; 2007a,b,c) and information about economic impacts from recent integrated assessment models (IAMs) to fit displaced gamma distributions for these variables. But unlike existing IAMs, I modeled economic impact as a relationship between temperature change and the growth rate of consumption as opposed to its level. I examined whether “reasonable” values for the remaining parameters (e.g., the starting growth rate and the index of risk aversion) can yield values of w∗() well above 3 percent for small values of , which might support stringent abatement. I also used a counterfactual—and pessimistic—scenario for temperature change: under business as usual (BAU), the atmospheric GHG concentration immediately increases to twice its pre-Industrial level, which leads to an (uncertain) increase in temperature at the horizon H, and then (from feedback effects or further emissions) a gradual further doubling of that temperature increase. In this chapter, I use the same displaced gamma distributions for temperature change and economic impact, but I compare two alternative damage models—a direct impact of temperature change on consumption versus a growth rate impact. I calibrate and “match” the two models by matching estimates of GDP/temperature change pairs from the group of IAMs at a specific horizon. I then calculate and compare WTPs for both models based on expected discounted utility, using a constant relative risk aversion (CRRA) utility function. I find that for either damage model, the resulting estimates of w∗() are generally below 2 percent or 3 percent, even for  around 2 or 3°C. This is because there is limited weight in the tails of the calibrated distributions for T and its impact. Larger estimates of WTP result for particular combinations of parameter values (e.g., an index of risk aversion close to 1 and a low initial GDP growth rate), but overall, the results are consistent with moderate abatement. A direct impact generally yields a larger WTP than a growth rate impact, and the sign and extent of the difference varies with changes in parameter values. Overall, there are no substantial differences between the two models in terms of policy implications.2 2. As with my earlier study, I ignore the implications of the opposing irreversibilities inherent in climate change policy and the value of waiting for more information. Immediate action reduces the largely irreversible build up of GHGs in the atmosphere, but waiting avoids an irreversible investment in abatement capital that might turn out to be at least partly unnecessary, and the net effect of these irreversibilities is unclear. For a discussion of the interaction of uncertainty and irreversibility, see Pindyck (2007).

50

Robert S. Pindyck

The next section discusses the probability distribution and dynamic trajectory for temperature change. Section 2.3 discusses the two alternative ways of modeling the economic impact of higher temperatures and the corresponding probability distributions that capture uncertainty over that impact. Section 2.4 explains the calculation of willingness to pay. Numerical results are presented and discussed in sections 2.5 and 2.6, and section 2.7 concludes. 2.2

Temperature

According to the IPCC (2007c), under BAU, that is, no abatement policy, growing GHG emissions would likely lead to a doubling of the atmospheric CO2e concentration relative to the pre-Industrial level by the end of this century. That, in turn, would cause an increase in global mean temperature that would “most likely” range between 1.0°C to 4.5°C, with an expected value of 2.5° to 3.0°C. The IPCC report indicates that this range, derived from a “summary” of the results of twenty- two scientific studies the IPCC surveyed, represents a roughly 66 to 90 percent confidence interval, that is, there is a 5 to 17 percent probability of a temperature increase above 4.5°C. The twenty- two studies also provide rough estimates of increases in temperature at the outer tail of the distribution. In summarizing them, the IPCC translated the implied outcome distributions into a standardized form that makes the studies comparable and created graphs showing multiple outcome distributions implied by groups of studies. Those distributions suggest that there is a 5 percent probability that a doubling of the CO2e concentration relative to the pre-Industrial level would lead to a global mean temperature increase of 7°C or more and a 1 percent probability that it would lead to a temperature increase of 10°C or more. I fit a three- parameter displaced gamma distribution for T to these 5 percent and 1 percent points and to a mean temperature change of 3.0°C. This distribution conforms with the distributions summarized by the IPCC. Finally, I assume (consistent with the IPCC’s focus on temperature change at the end of this century) that the fitted distribution for T applies to a 100- year horizon H. 2.2.1

Displaced Gamma Distribution

The displaced gamma distribution is given by (1)

r f(x; r, , )   (x )r 1e (x ), (r)

x ,

where (r)  冕0sr–1e–sds is the gamma function. The moment generating function for equation (1) is

冢

冣

 r Mx(t)  E(etx)   et.  t

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.1

51

Distribution for temperature change

Thus the mean, variance and skewness (around the mean) are E(x)  r/ , V(x)  r/2, and S(x)  2r/3, respectively. Fitting f(x; r, , ) to a mean of 3°C, and the 5 percent and 1 percent points at 7°C and 10°C, respectively, yields r  3.8,   0.92, and   –1.13. The distribution is shown in figure 2.1. It has a variance and skewness around the mean of 4.49 and 9.76, respectively. Note that this distribution implies that there is a small (2.9 percent) probability that a doubling of the CO2e concentration will lead to a reduction in mean temperature, consistent with several of the scientific studies. The distribution also implies that the probability of a temperature increase of 4.5°C or greater is 21 percent. 2.2.2

Trajectory for Tt

Recall that this fitted distribution for T pertains to the 100- year horizon H. To allow for possible feedback effects or further emissions, I assume that Tt → 2TH as t gets large. As summarized in Weitzman (2009), the simplest dynamic model relating Tt to the GHG concentration Gt is the differential equation

52

(2)

Robert S. Pindyck

冤

冥

ln(Gt /G0 ) dT m2 Tt .   m1 dt ln 2

Making the conservative (in the sense that it would lead to a higher WTP) assumption that Gt immediately (i.e., at t  0) doubles to 2G0, Tt  TH at t  H , and Tt → 2TH as t → , implies that Tt follows the trajectory

冤 冢 冣 冥.

1 Tt  2TH 1  2

(3)

t/H

Thus if TH  5°C, Tt reaches 2.93°C after 50 years, 5°C after 100 years, 7.5°C after 200 years, and then gradually approaches 10°C. This is illustrated in figure 2.2, which shows a trajectory for T when it is unconstrained (and TH happens to equal 5°C) and when it is constrained so that TH   3°C. Note that even when constrained, TH is a random variable and (unless   0) will be less than  with probability 1; in figure 2.2, it happens to be 2.5°C. If   0, then T  0 for all t. 2.3

Impact of Warming

Most economic studies of climate change assume that T has a direct impact on GDP (or consumption), modeled via a “loss function” L(T ), with L(0)  1 and L  0. Thus GDP at some horizon H is L(TH )GDPH , where GDPH is but- for GDP in the absence of warming. This “direct impact” approach has been used in all of the integrated assessment models that I am aware of. However, there are reasons to expect warming to affect the growth rate of GDP as opposed to the level. At issue is how these two alternative approaches to modeling the impact of warming—direct versus growth rate—differ in their implications for estimates of willingness to pay to limit warming. 2.3.1

Direct Impact

The most widely used loss function has been the inverse quadratic. For example, the recent version of the Nordhaus (2008) Dynamic Integrated model of Climate and the Economy (DICE) uses the following loss function: 1 L   . [1 1T 2(T )2] Weitzman (2008) introduced the exponential loss function, which is very similar to the inverse quadratic for small values of T but allows for greater losses when T is large: (4)

L(T )  exp[ (T )2].

I will use this loss function of equation (4) when calculating WTP under the direct impact assumption.

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.2

53

Temperature change: Unconstrained and constrained so ⌬TH < −␶

To introduce uncertainty over the impact of warming, I will treat the parameter  as a random variable that, like temperature change, can be described by a 3- parameter displaced gamma distribution. Although the IPCC does not provide standardized distributions for lost GDP corresponding to any particular T as it does for climate sensitivity, it does survey the results of several IAMS. As discussed in the following, I use the information from the IPCC along with other studies to infer means and confidence intervals for . 2.3.2

Growth Rate Impact

There are three reasons to expect warming to affect the growth rate of GDP as opposed to the level. First, some effects of warming are likely to be permanent: for example, destruction of ecosystems from erosion and flooding, extinction of species, and deaths from health effects and weather extremes. If warming affected the level of GDP directly, for example, as per equation (4), it would imply that if temperatures rise but later fall, for example, because of stringent abatement or geoengineering, GDP could return to its but- for path with no permanent loss. This is not the case, however, if T affects the growth rate of GDP.

54

Robert S. Pindyck

Fig. 2.3

Example of economic impact of temperature change

Note: Temperature increases by 5°C over fifty years and then falls to original level over next fifty years. C A is consumption when T reduces level, C B is consumption when T reduces growth rate, and C 0 is consumption with no temperature change.

Suppose, for example, that temperature increases by 0.1°C per year for fifty years and then decreases by 0.1°C per year for the next fifty years. Figure 2.3 compares two consumption trajectories: C At , which corresponds to the loss function of equation (4), and C Bt , which corresponds to the following growth rate loss function: (5)

gt  g0 Tt

The example assumes that without warming, consumption would grow at 0.5 percent per year—trajectory C 0t —and both loss functions are calibrated so that at the maximum T of 5°C, CA  CB  .95C 0. Note that as T falls to zero, C At reverts to C t0, but C tB remains permanently below C t0. Second, resources needed to counter the impact of higher temperatures would reduce those available for research and development (R&D) and capital investment, reducing growth. Adaptation to rising temperatures is equivalent to the cost of increasingly strict emission standards, which, as Stokey (1998) has shown with an endogenous growth model, reduces the rate of return on capital and lowers the growth rate. As a simple example,

55

Modeling the Impact of Warming in Climate Change Economics

suppose total capital K  Kp Ka(T ), with Ka(T )  0, where Kp is directly productive capital and Ka(T ) is capital needed for adaptation to the temperature T (e.g., stronger retaining walls and pumps to counter flooding, new infrastructure and housing to support migration, more air conditioning and insulation, etc.). If all capital depreciates at rate K, K˙p  K˙ – K˙a  I – K K – Ka(T )T˙ so that the rate of growth of Kp is reduced, and, thus, the rate of growth of output is reduced. Third, there is empirical support for a growth rate effect. Using historical data on temperatures and precipitation over the past fifty years for a panel of 136 countries, Dell, Jones, and Olken (2008) have shown that higher temperatures reduce GDP growth rates but not levels. The impact they estimate is large—a decrease of 1.1 percentage points of growth for each 1°C rise in temperature—but significant only for poorer countries.3 To calculate WTP when T affects the growth rate of GDP, I assume that in the absence of warming, real GDP and consumption would grow at a constant rate g0, but warming will reduce this rate according to equation (5). This simple linear relation was estimated by Dell, Jones, and Olken (2008), and can be viewed as at least a first approximation to a more complex loss function. I introduce uncertainty by making the parameter , like , a random variable drawn from a displaced gamma distribution. 2.3.3

Distributions for  and 

To compare the effects of a direct versus growth rate impact on estimates of WTP, we need to fit and “match” the distributions for  and . This is done as follows. Using information from a number of IAMs, I fit the three parameters in a displaced gamma distribution for  in the exponential- quadratic loss function of equation (4). I then translate this into an equivalent distribution for  using the trajectory for GDP and consumption implied by equation (5) for a temperature change- impact combination projected to occur at horizon H. From equations (3) and (5), the growth rate is gt  g0 – 2TH[1 – (1/2)t/H]. Normalizing initial consumption at 1, this implies

冤

冥

2HTH 2HTH 冕t (6) Ct  e 0 g(s)ds  exp  (g0 2TH )t  (1/2)t/H . ln(1/2) ln(1/2) Thus,  is obtained from  by equating the expressions for CH implied by equations (4) and (6): 3. “Poor” means below- median purchasing power parity (PPP)- adjusted per capita GDP. Using World Bank data for 209 countries, “poor” by this definition accounts for 26.9 percent of 2006 world GDP, which implies a roughly 0.3 percentage point reduction in world GDP growth for each 1°C rise in temperature. In a follow- on paper, Dell, Jones, and Olken (2009) estimate a model that allows for adaptation effects so that the long- run impact of warming is smaller than the short- run impact. They find a long- run decrease of 0.51 percentage points of growth for each 1°C rise in temperature, but again only for poorer countries.

56

Robert S. Pindyck

2HTH HTH (g0 2TH )H  (7) exp   exp[g0H (TH )2] ln(1/2) ln(1/2)

冤

冥

so that  and  have the simple linear relationship (8)

1.79TH    . H

Distribution for  To fit a displaced gamma distribution for , I utilize the IPCC’s survey of several IAMS. This information from the IPCC, along with other studies, allow me to infer means and confidence intervals for . These IAMs yield a rough consensus regarding possible economic impacts: for temperature increases up to 4°C, the “most likely” impact is from 1 percent to at most 5 percent of GDP. (Of course, this consensus might arise from the use of similar ad hoc damage functions in various IAMs.) Of interest is the outer tail of the distribution for this impact. There is some chance that a temperature increase of 3°C or 4°C would have a much larger impact, and we want to know how that affects WTP. Based on its survey of impact estimates from four IAMs, the IPCC (2007a, 17) concludes that “global mean losses could be 1–5% of GDP for 4°C of warming.”4 In addition, Dietz and Stern (2008) provide a graphical summary of damage estimates from several IAMs, which yield a range of 0.5 percent to 2 percent of lost GDP for T  3°C and 1 percent to 8 percent of lost GDP for T  5°C. I treat these ranges as “most likely” outcomes and use the IPCC’s definition of “most likely” to mean a 66 to 90 percent confidence interval. Using the IPCC range and, to be conservative, assuming it applies to a 66 percent confidence interval, I take the mean loss for T  4°C to be 3 percent of GDP and the 17 percent and 83 percent confidence points to be 1 percent of GDP and 5 percent of GDP, respectively. We can then use equation (4) to get the mean, 17 percent, and 83 percent values for , which 2 I denote by 苶 , 1 and 2, respectively. For example, .97  e–苶(4) so that 苶  .00190. Likewise, 1  .000628 and 2  .00321. Fitting a displaced gamma distribution to these numbers yields r  4.5;   1,528; and   苶  – r/  –.00105. Figure 2.4 shows the fitted distribution for . Also shown is the fitted distribution when “most likely” is taken to mean a 90 percent confidence interval so that 1 and 2 instead apply to the 5 and 95 percent confidence points.

4. The IAMs surveyed by the IPCC include Hope (2006), Mendelsohn et al. (1998), Nordhaus and Boyer (2000), and Tol (2002). For a recent overview of economic impact studies, see Tol (2009).

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.4

57

Distributions for loss function parameter ␤

Distribution for  The mean, 17 percent, and 83 percent values for  applied to a TH  4°C at a horizon H  100 years, so from equation (8),   .0716. Thus, the mean, 17 percent, and 83 percent values for  are, respectively, 苶   .0001360, 1  .0000450, and 2  .0002298. Now suppose f(x; r, , ) is the displaced gamma distribution for x, and we want the distribution f( y; r1, 1, 1) for y  ax. We can make use of the fact that the expectation, variance, and skewness of x and of y are related as follows: E(y)  aE(x)  ar/ a, V(y)  a2V(x)  a2r/2, and S(y)  a3S(x)  2a3r/3. This implies that 1  a, r1  r, and 1  /a. Thus, the matched distribution for  will be the same as that for , except that 1  1528/.0716  21,340 and 1  .0716(–.00105)  –.0000752. The distribution for , shown graphically in Pindyck (2009), will have the same shape as the distribution for  but a different scaling. 2.4

Willingness to Pay

Given the distributions for T and  or , I posit a CRRA social utility function:

58

Robert S. Pindyck

C 1  t . U(Ct)   (1 )

(9)

where  is the index of relative risk aversion (and 1/ is the elasticity of intertemporal substitution). Social welfare is measured as the expected sum over time of discounted utility: ∞

W  E ∫ U(Ct )e t dt.

(10)

0

where  is the rate of time preference, that is, the rate at which utility is discounted. Note that this rate is different from the consumption discount rate, which in the Ramsey growth context would be Rt   gt. If T affects consumption directly, then Rt   g0 and does not change over time. If T affects the growth rate of consumption, then Rt   g0 – 2TH[1 – (1/2)t/H], so Rt falls over time as T increases.5 For both the direct and growth rate impact models, I calculate the fraction of consumption—now and throughout the future—society would sacrifice to ensure that any increase in temperature at a specific horizon H is limited to an amount . That fraction, w∗(), is the measure of willingness to pay.6 2.4.1

WTP: Direct Impact

Using equation (3), if TH and  were known, social welfare would be given by ∞

(11)

∞

1 t/H 2t/H W  ∫ U(Ct)e t dt   ∫ e0 1t 20(1/2) 0(1/2) dt, 1  0 0

where (12)

0  4( 1)(TH )2,

(13)

1  ( 1)g0 .

Suppose society sacrifices a fraction w() of present and future consumption to keep TH . With uncertainty over TH and , social welfare at t  0 is ∞

(14)

[1 w()]1  t/H 2t/H W1   E0, ∫ e ˜0 ˜1t 2˜0(1/2) ˜0(1/2) dt, 1  0

where E0, denotes the expectation at t  0 over the distributions of TH and  conditional on TH . (Tildes are used to denote that 0 and 1 are 5. If 2TH   g0, Rt becomes negative as T grows. This is entirely consistent with the Ramsey growth model, as pointed out by Dasgupta, Mäler, and Barrett (1999). 6. The use of WTP as a welfare measure goes back at least to Debreu (1954), was used by Lucas (1987) to estimate the welfare cost of business cycles, and was used in the context of climate change (with   0) by Heal and Kriström (2002) and Weitzman (2008).

Modeling the Impact of Warming in Climate Change Economics

59

functions of two random variables.) If no action is taken to limit warming, social welfare would be ∞

(15)

1 t/H 2t/H W2   E0 ∫ e ˜0 ˜1t 2˜0(1/2) ˜0(1/2) dt, 1  0

where E0 again denotes the expectation over TH and  but now with TH unconstrained. Willingness to pay to ensure that TH  is the value w∗() that equates W1() and W2.7 Given the distributions f(T ) and g() for TH and , respectively, denote by M(t) and M(t) the time-t expectations (16)

1 M(t)   F()

 ∞

∫ ∫ e ˜ ˜ t 2˜ (1/2) 0

1

t/H  ˜ 0(1/2)2t/H

0

f(T )g()dTd

T  

and ∞ ∞

(17)

M(t) 

∫ ∫ e ˜ ˜ t 2˜ (1/2) 0

1

0

t/H  ˜ (1/2)2t/H 0

f(T )g ()dTd,

T  

where ˜ 0 and ˜ 1 are given by equations (12) and (13), T and  are the lower  limits on the distributions for T and , and F()  冕 T f(T )dT. Thus, W1() and W2 are ∞

(18)

[1 w()]1  [1 w()]1  W1()   ∫ M(t)dt ⬅  G 1  1  0

and ∞

(19)

1 1 W2   ∫ M(t)dt ⬅  G. 1  0 1 

Setting W1() equal to W2, WTP is given by (20)

冢 冣

G w∗()  1  G

1(1 )

.

The solution for w∗() depends on the distributions for T and , the horizon H  100 years, and the parameters , g0, and  (values for which are discussed in the following). We will examine how w∗ varies with ; the cost of abatement should be a decreasing function of , so given estimates of that cost, one could use these results to determine abatement targets. 2.4.2

WTP: Growth Rate Impact

If TH instead affects the growth rate of consumption as in equation (5), and if TH and  were known, social welfare would be 7. I calculate WTP using a finite horizon of 500 years. After some 200 years, the world will likely exhaust the economically recoverable stocks of fossil fuels so that GHG concentrations will fall. In addition, so many other economic and social changes are likely that the relevance of applying CRRA expected utility over more than a few hundred years is questionable.

60

Robert S. Pindyck ∞

∞

1 t/H W  ∫ U(Ct)e t dt   ∫ e0 1t 0(1/2) dt. 1  0 0

(21) where (22)

HTH 0  2( 1)  , ln(1/2)

(23)

1  ( 1)(g0 2TH ) .

If society sacrifices a fraction w() of present and future consumption to keep TH  and there is uncertainty over TH and , social welfare at t  0 is ∞

[1 w()]1  t/H W1()   E0, ∫ e ˜ 0 ˜1t ˜ 0(1/2) dt. 1  0

(24)

If no action is taken to limit warming, social welfare would be ∞

t/H 1 W2   E0 ∫ e˜ 0 ˜ 1t ˜ 0(1/2) dt. 1  0

(25)

Once again, WTP is the value w∗() that equates W1() and W2. Defining M(t) and M(t) as before, but with g () instead of g (), equations (18), (19), and (20) again apply. 2.5

Results

Willingness to pay is essentially a measure of the demand side of policy— the maximum amount society would be willing to sacrifice to obtain the benefits of limited warming. The case for an actual GHG abatement policy will depend on the cost of that policy as well as the benefits. The framework I use does not involve estimates of abatement costs—I only estimate WTP as a function of , the abatement- induced limit on any increase in temperature at the horizon H. Clearly the amount and cost of abatement needed will decrease as  is made larger, so I consider a stringent abatement policy to be one for which  is “low,” which I take to be at or below the expected value of T under a BAU scenario, that is, about 3°C, and w∗() is “high,” that is, at least 3 percent. At issue in this chapter is the extent to which estimates of WTP depend on whether T is assumed to affect the level of consumption directly versus the growth rate of consumption. In addition to the distributions for T and the impact parameters  or , WTP depends on the values for the index of relative risk aversion , the rate of time discount , and the base level real growth rate g0. To explore the case for a stringent abatement policy, I make conservative assumptions about , , and g0 in the sense of choosing numbers that would lead to a higher WTP.

Modeling the Impact of Warming in Climate Change Economics

61

The finance and macroeconomics literature has estimates of  ranging from 1.5 to 6 and estimates of  ranging from .01 to .04. The historical real growth rate g ranges from .02 to .025. It has been argued, however, that for intergenerational comparisons  should be close to zero on the grounds that society not should value the well- being of our great- grandchildren less than our own. Likewise, while values of  well above 2 may be consistent with the (relatively short horizon) behavior of investors, we might use lower values for intergenerational welfare comparisons. Because I want to determine whether current assessments of uncertainty over temperature change and its impact generate a high enough WTP to justify stringent abatement, I will stack the deck in favor of our great- grandchildren and use relatively low values of  and : around 2 for  and 0 for . Also, WTP is a decreasing function of the base growth rate g0, so I will set g0  .02, the low end of the historical range. 2.5.1

No Uncertainty

It is useful to begin by considering a deterministic world in which the trajectory for T and the impact of that trajectory are known with certainty. Then (18) and (19) for the direct impact case would simplify to ∞

(26)

[1 w()]1  t/H 2t/H W1   ∫ e0 1t 20(1/2) 0(1/2) dt, 1  0

(27)

t/H 2t/H 1 W2   ∫ e0 1t 20(1/2) 0(1/2) dt, 1  0

∞

where now 苶 , the mean of , replaces  in equation (12) for 0. (I will use the means of  and  as their certainty- equivalent values.) Likewise, equations (24) and (25) for the case of a growth rate impact would simplify to ∞

(28)

[1 w()]1  t/H W1()   ∫ e0 1t 0(1/2) dt, 1  0

(29)

t/H 1 W2   ∫ e0 1t 0(1/2) dt, 1  0

∞

where now the mean 苶  replaces  in equations (22) and (23) for 0 and 1. For both impact models, I calculate the WTP to keep T zero for all time, that is, w∗(0), over a range of values for T at the horizon H  100. For this exercise, I set   2,   0, and g0  .020. The results are shown in figure 2.5, where w ∗c (0) applies to the case where T affects C directly, and w∗g (0) applies to the case where T affects the growth rate of C. The graph says that if, for example, TH  6°C, w∗c (0) is about .03, and ∗ w g (0) is about .022. Thus, if T affects consumption directly, society should be willing to give about 3 percent of current and future consumption to keep T at zero instead of 6°C. But if T affects the growth rate of con-

62

Robert S. Pindyck

Fig. 2.5

WTP, known temperature change: ␩ ⴝ 2, g0 ⴝ .020, and ␦ ⴝ 0

sumption, the willingness to pay is only about 2.2 percent. (Remember that the “known T” applies to time t  H. Tt follows the trajectory given by equation [3].) Note that both w∗c (0) and w∗g (0) become much larger as the known TH becomes larger than 8°C; such temperature outcomes, however, have low probability. In addition, these curves have different shapes: w∗c (0) is a convex function of TH , while w∗g (0) is a (nearly) linear function of TH.8 This means that for small changes in temperature, a growth rate impact model will yield a slightly higher WTP, but for very large changes in temperature, the direct impact model yields much larger WTPs. Whether this difference matters for estimates of WTP under uncertainty depends on the probability distributions for TH and  and . With sufficient probability mass in the right- hand tails of the distributions, the two impact models should yield different numbers for WTP. We explore this in the following. 8. To see that wc∗(0) is a convex function of TH , note that if   0, equations (26) and (27) imply that

冤

冥

0

w∗c (0)  1 –  ∫ e–t–4(1–)苶T

2 H

∞

1/(1–)

ψ(t)

dt

where ψ(t)  [1 – (1/2)t/H ]2. Just differentiate to see that dwc∗/dTH  0 for all values of TH and , and d 2wc∗/dTH2  0 for sufficiently small values of TH and 苶  (in our case, as long as TH  苶 15.8°C). Similarly, we can show that dw∗g /dTH  0 and d 2wc∗/dTH2 is a small negative number (in our case –.000063), a curvature small enough so that in figure 2.5, w∗g (0) appears linear.

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.6

2.5.2

63

WTP, both ⌬T and ␥ uncertain: ␩ ⴝ 2 and 1.5, g0 ⴝ .020, and ␦ ⴝ 0

Uncertainty over Temperature and Economic Impact

I now allow for uncertainty over both T and the relevant impact parameter (either  or ), using the calibrated distributions for each. Willingness to pay is given by equations (16) to (20) for the direct impact model and equations (24) and (25) for the growth rate impact. The calculated values of WTP as are shown as functions of  in figure 2.6 for   0, g0  .020, and   2 and 1.5. Note that if   2, WTP is always less than 1.5 percent, even for   0. To obtain a WTP above 2 percent requires a lower value of . As figure 2.6 shows, if   1.5, w∗() reaches about 3 percent for  around 0 or 1°C, but only when the impact of warming occurs through the growth rate of consumption. When the impact is direct, w∗ is always below 2.5 percent. Because relatively large values of WTP can only be obtained for small values of , the top two lines in figure 2.6 have the greatest policy relevance. But note that when   1.5, the difference between w∗g () and w∗c () is only significant for  below 2°C. It seems unlikely that a politically and economically feasible policy would be adopted that would prevent any warming, or limit it to 1 or even 2°C. If we believe that a “feasible” policy is one that limits T to its expected value of around 3°C, then as figure 2.6 shows, the direct and growth rate impact models give similar values for WTP. On the other hand, what if we take the view that the “correct” value of  is less than 1.5? Figure 2.7 shows the dependence of WTP on the index of risk

64

Robert S. Pindyck

Fig. 2.7

WTP versus ␩ for ␶ ⴝ 3: g0 ⴝ .020 and ␦ ⴝ 0

aversion, . It plots w∗(3), that is, the WTP to ensure TH 3°C at H  100 years, for g0  .02, as a function of . Although w∗(3) is below 2 percent for values of  above 1.5, it approaches 5 percent as  is reduced to 1 (the value used in Stern 2007). The reason is that while future utility is not discounted (because   0), future consumption is implicitly discounted at the initial rate g0. If  is made smaller, potential losses of future consumption have a larger impact on WTP. Also, as discussed further in the following, w∗c (3)  ()w∗g (3) when   ()1.3. These estimates of WTP are based on zero discounting of future utility. While there may be an ethical argument for zero discounting,   0 is outside the range of estimates of the rate of time preference obtained from consumer and investor behavior. However, estimates of WTP above 3 percent depend critically on this assumption of   0. Figure 2.8 again plots w∗c (3) and w∗g (3), but this time with   .01. Note that for either impact model, discounting future utility, even at a very low rate, will considerably reduce WTP. With   .01, w∗(3) is again below 2 percent for all values of , and for either impact model. The results so far indicate that for either impact model, large values of WTP require fairly extreme combinations of parameter values. However, these results are based on distributions for TH , , and  that were calibrated to studies in the IPCC’s 2007 (2007a,c) report and concurrent economic

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.8

65

WTP versus ␩ for ␶ ⴝ 3: g0 ⴝ .020 and ␦ ⴝ .01

studies, and those studies were done several years prior to 2007. More recent studies suggest that “most likely” values for T in 2100 might be higher than the 1.0°C to 4.5°C range given by the IPCC. For example, a recent report by Sokolov et al. (2009) suggests an expected value for T in 2100 of around 4 to 5°C, as opposed to the 3.0°C expected value that I used. Thus, I recalculate WTP for both impact models, for both   0 and .01, but this time shifting the distribution for TH to the right so that it has a mean of 5°C, corresponding to the upper end of the 4 to 5°C range in Sokolov et al. (2009). (The other moments of the distribution remain unchanged, and H is again 100 years). The results are shown in figures 2.9 and 2.10. Now if   0 and  is below 1.5, w∗(3) is above 3 percent when the impact of T occurs through the growth rate, and above 4 percent when the impact is direct, and reaches around 10 percent if   1. Even if   .01, w ∗c (3) exceeds 4 percent when   1 (although w∗g [3] only reaches 2.5 percent). Thus, there are parameter values and plausible distributions for T that yield a large WTP. Those parameter values and distributions are outside the current consensus range, but that may change as new studies of warming and its impact become available. As figures 2.7 to 2.10 show, for either value of , w ∗c (3) is usually higher than w∗g (3), and when   .01 it is considerably higher. In the Ramsey growth context, the consumption discount rate is  gt, so even if   0, future

66

Robert S. Pindyck

Fig. 2.9

WTP versus ␩ for ␶ ⴝ 3: E(⌬TH) ⴝ 5°C, g0 ⴝ .020, ␦ ⴝ 0

consumption (although not utility) is discounted (less so for small values of ). When T affects consumption directly, the loss of consumption is greater at shorter horizons (but smaller at long horizons), making w ∗c (3)  w ∗g (3). (In figure 2.7, w ∗c [3]  w ∗g [3] when   1.3 because with a low consumption discount rate, the larger long- run reduction in consumption from a growth rate impact overwhelms the smaller short- run impact, even in expected value terms.) 2.6

Modeling and Policy Implications

The integrated assessment models that I am aware of all relate temperature change to the level of real GDP and consumption. As we have seen, this will often yield a higher WTP—and thus yield higher estimates of optimal GHG abatement—than will a model that relates temperature change to the growth rate of GDP and consumption. How important is the difference, and what do these results tell us about modeling? In Pindyck (2009), using a model that related temperature change to the growth rate of consumption, I found that for temperature and impact distributions based on the IPCC and “conservative” parameter values (e.g.,   0,   2, and g0  .02), WTP to prevent even a small increase in temperature is around 2 percent or less, which is inconsistent with the immediate adop-

Modeling the Impact of Warming in Climate Change Economics

Fig. 2.10

67

WTP versus ␩ for ␶ ⴝ 3: E(⌬TH) ⴝ 5°C, g0 ⴝ .020, ␦ ⴝ .01

tion of a stringent GHG abatement policy. To what extent do those results change when temperature change directly affects the level of consumption? And more broadly, what are the policy implications of the results in this chapter? 2.6.1

Implications for Modeling

The difference in WTPs for a direct versus a growth rate impact is largest for large temperature changes and for higher consumption discount rates. As we saw in figure 2.5 for the case of no uncertainty, w ∗c (0) is a convex function of T and thus becomes increasingly greater than w ∗g (0) as T gets larger. Likewise, when there is uncertainty but the expected temperature change is increased from 3°C to 5°C, the difference between w ∗c (3) and w ∗g (3) becomes larger. And note from figures 2.9 and 2.10 that the difference between w ∗c (3) and w ∗g (3) is proportionally larger when the consumption discount rate ( g0) is larger, that is, when  is larger or when  is .01 rather than 0. If the consumption discount rate is large (i.e., if  is large or   0), almost any model will yield estimates of WTP and optimal abatement levels that are small. This is simply the result of discounting over long horizons (greater than fifty years). That is why model- based analyses that call for stringent abatement policies assume   0 and relatively low values for . (Stern [2007, 2008], for example, uses   0 and   1.) Thus, if we limit our analyses to

68

Robert S. Pindyck

the low end of the consensus range for  (around 1.5), even with   0, the choice of impact model will matter if evolving climate science studies yield increasingly large estimates of expected temperature change. Which impact model—direct versus growth rate—should one use for modeling? A direct impact model is simpler, easier to understand, and perhaps easier to estimate or calibrate. But as I have argued at the beginning of this chapter, there are strong theoretical and empirical arguments that favor the growth rate impact. Until new studies demonstrate otherwise, it seems to me that it is difficult to make the case for a direct impact. 2.6.2

Implications for Policy

The results in this chapter supplement those in Pindyck (2009) in terms of implications for policy. We can summarize those implications as follows. First, although the direct impact model often yields higher estimates of WTP, it is still the case that using temperature and impact distributions based on the IPCC (2007a,c) and concurrent economic studies, for most parameter values our WTP estimates are still too low to support a stringent GHG abatement policy. Of course, these estimates do not suggest that no abatement is optimal. For example, a WTP of 2 percent of GDP is in the range of cost estimates for compliance with the Kyoto Protocol.9 In addition to the effects of discounting discussed in the preceding, our low estimates of WTP are due to the limited weight in the tails of the distributions for T and the impact parameter  or . The probability of a realization in which T 4.5°C in 100 years and the impact parameter is 1 standard deviation above its mean is less than 5 percent. An even more extreme outcome in which T  7°C (and the impact parameter is 1 standard deviation above its mean) would imply about a 9 percent loss of GDP in 100 years for a growth rate impact, but the probability of an outcome this bad or worse is less than 1 percent. And this low- probability loss of GDP in 100 years would involve much smaller losses in earlier years. Second, although these estimates of WTP are consistent with the current consensus regarding future warming and its impact as summarized in IPCC (2007a,b,c) that consensus may be wrong, especially with respect to the tails of the distributions. Indeed, based on recent studies, that consensus may already be shifting toward more dire estimates of warming and its impact. As we saw from figures 2.9 and 2.10, shifting the temperature distribution to the right so that E(TH ) is 5°C instead of 3°C results in substantially higher estimates of WTP. Thus, if the consensus (or “state of knowledge”) shifts toward a higher expected value for the amount of warming, or more mass in the tails of the distribution, WTP might increase enough to justify more aggressive abatement policies. 9. See the survey of cost studies by the Energy Information Administration (1998) and the more recent country cost studies surveyed in IPCC (2007b).

Modeling the Impact of Warming in Climate Change Economics

2.7

69

Conclusions

If we are to use economic models to evaluate GHG abatement policies, how should we treat the impact of possible future increases in temperature? One could argue that we simply do not (and cannot) know much about that impact because we have had no experience with substantial amounts of warming, and there are no models or data that can tell us much about the impact of warming on production, migration, disease prevalence, and a variety of other relevant factors. Instead, I have taken existing IAMs and related models of economic impact at face value and treated them analogously to the climate science models that are used to predict temperature change or its probability distribution. In this way, I obtained a (displaced gamma) distribution for an impact parameter that relates temperature change to consumption or to the growth rate of consumption. We have seen that in most cases, a direct impact yields a higher WTP than a growth rate impact. The reason is that when T affects consumption directly, the loss of consumption is greater at short horizons (but smaller at long horizons). Consumption discounting can give these short- horizon effects more weight. Even if future utility is not discounted (  0), the consumption discount rate ( g0) is still positive and can be large if  is large. Overall, I would argue that the choice of a direct versus growth rate impact should be based on the underlying economics, and the growth rate specification has both theoretical and empirical support. But even with a direct impact model, using temperature and impact distributions based on the IPCC (2007a,b,c) and concurrent economic studies, for most parameter values our WTP estimates are still too low to support a stringent GHG abatement policy. Of course, there are parameter values and plausible distributions for T that yield a large WTP—and those that can yield a much smaller WTP. In particular, if the rate of time preference, , is 1 or 2 percent, WTP will generally be very low. On the other hand, if   0, a shift in the temperature change distribution such that E(TH ) is 5°C or a shift in the accepted value of  to put it close to 1 can lead to a WTP of 6 or 8 percent. Such distributions and parameter values are outside the current consensus range, but that range may change as new studies of warming and its impact are completed and disseminated.

References Dasgupta, Partha, Karl-Göran Mäler, and Scott Barrett. 1999. Intergenerational equity, social discount rates, and global warming. In Discounting and intergenerational equity, ed. P. Portney and J. Weyant. Washington, DC: Resources for the Future.

70

Robert S. Pindyck

Debreu, Gerard. 1954. A classical tax- subsidy problem. Econometrica 22:14–22. Dell, Melissa, Benjamin F. Jones, and Benjamin A. Olken. 2008. Climate change and economic growth: Evidence from the last half century. NBER Working Paper no. 14132. Cambridge, MA: National Bureau of Economic Research, June. ———. 2009. Temperature and income: Reconciling new cross- sectional and panel estimates. American Economic Review 99:198–204. Dietz, Simon, and Nicholas Stern. 2008. Why economic analysis supports strong action on climate change: A response to the Stern Review’s critics. Review of Environmental Economics and Policy 2:94–113. Energy Information Administration. 1998. Comparing cost estimates for the Kyoto Protocol. Report no. SR/OIAF/98- 03. Washington, DC: Energy Information Administration, October. Heal, Geoffrey, and Bengt Kriström. 2002. Uncertainty and climate change. Environmental and Resource Economics 22:3–39. Hope, Chris W. 2006. The marginal impact of CO2 from PAGE2002: An integrated assessment model incorporating the IPCC’s five reasons for concern. Integrated Assessment 6:1–16. Intergovernmental Panel on Climate Change (IPCC). 2007a. Climate change 2007: Impacts, adaptation, and vulnerability. Cambridge, UK: Cambridge University Press. ———. 2007b. Climate change 2007: Mitigation of climate change. Cambridge, UK: Cambridge University Press. ———. 2007c. Climate change 2007: The physical science basis. Cambridge, UK: Cambridge University Press. Lucas, Robert E., Jr. 1987. Models of business cycles. Oxford, UK: Basil Blackwell. Mendelsohn, Robert. 2008. Is the Stern Review an economic analysis? Review of Environmental Economics and Policy 2:45–60. Mendelsohn, Robert, W. N. Morrison, M. E. Schlesinger, and N. G. Andronova. 1998. Country- specific market impacts of climate change. Climatic Change 45:553–69. Nordhaus, William D. 2007. A review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature 45:686–702. ———. 2008. A question of balance: Weighing the options on global warming policies. New Haven, CT: Yale University Press. Nordhaus, William D., and J. G. Boyer. 2007. Warming the world: Economic models of global warming. Cambridge, MA: MIT Press. Pindyck, Robert S. 2007. Uncertainty in environmental economics. Review of Environmental Economics and Policy 1:45–65. ———. 2009. Uncertain outcomes and climate change policy. NBER Working Paper no. 15259. Cambridge, MA: National Bureau of Economic Research, August. Sokolov, A. P., P. H. Stone, C. E. Forest, R. Prinn, M. C. Sarofim, M. Webster, S. Paltsev, C. A. Schlosser, D. Kicklighter, S. Dutkiewicz, J. Reilly, C. Wang, B. Felzer, and H. D. Jacoby. 2009. Probabilistic forecast for 21st century climate based on uncertainties in emissions (without policy) and climate parameters. MIT Joint Program on the Science and Policy of Global Change Report no. 169. Cambridge, MA: MIT. Stern, Nicholas. 2007. The economics of climate change: The Stern Review. Cambridge, UK: Cambridge University Press. ———. 2008. The economics of climate change. American Economic Review 98: 1–37. Stokey, Nancy. 1998. Are there limits to growth? International Economic Review 39:1–31.

Modeling the Impact of Warming in Climate Change Economics

71

Tol, Richard S. J. 2002. Estimates of the damage costs of climate change. Environmental and Resource Economics 21 (Part 1: Benchmark estimates): 47–73; 21 (Part 2: Dynamic estimates): 135–60. ———. 2009. The economic effects of climate change. Journal of Economic Perspectives 23:29–51. Weitzman, Martin L. 2007. A review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature 45:703–24. ———. 2008. Some dynamic economic consequences of the climate- sensitivity inference dilemma. Harvard University. Unpublished Manuscript. ———. 2009. Additive damages, fat- tailed climate dynamics, and uncertain discounting. Harvard University. Unpublished Manuscript.

3 Droughts, Floods, and Financial Distress in the United States John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

The relationships among the weather, agricultural markets, and financial markets have long been of interest to economic historians, but relatively little empirical work has been done, especially at the regional or state level. We push this literature forward by using modern drought indexes, which are available in detail over a wide area and for long periods of time to perform a battery of tests on the relationship between these indexes and sensitive indicators of financial stress. The financial literature in the area can be traced to William Stanley Jevons, who connected his sunspot theory to rainfall patterns. The Dust Bowl of the 1930s brought the weather- finance link to the attention of the general public. Here we assemble new evidence to test various hypotheses involving the impact of extreme swings in moisture on financial stress in the United States. 3.1

Prior Work on Weather, Financial Markets, and Business Cycles

The idea that weather affects agriculture and through agriculture, financial markets and the economy as a whole has a long, if not always persuasive, history among economists. The British economist William Stanley Jevons (1884, 221–43) famously argued that financial crises were produced, John Landon-Lane is an associate professor of economics at Rutgers University. Hugh Rockoff is a professor of economics at Rutgers University, and a research associate of the National Bureau of Economic Research. Richard H. Steckel is the Social and Behavioral Sciences Distinguished Professor of Economics, Anthropology, and History and a Distinguished University Professor at Ohio State University, and a research associate of the National Bureau of Economic Research. The authors thank conference participants, David Stahle and Henri Grissino-Mayer, for comments and suggestions. We also thank Scott A. Redenius who was kind enough to share his estimates of the rates of return to bank equity.

73

74

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

ultimately, by sunspots. Financial crises had occurred with an average frequency of ten to twenty years in Jevons’s time (1825, 1836, 1847, and 1866). Could it be an accident, Jevons asked, that spots appeared on the surface of the sun at (approximately) the same intervals? The connection, Jevons concluded, was through India. Sunspot activity disrupted rainfall and harvests in India. Low incomes in India depressed imports from Britain. The disruption of British trade with India in turn produced the financial crises. Jevons’s son, H. Stanley Jevons (1933), attempted to defend and extend his father’s theory. He recognized that the business cycle was the result of several factors. However, he argued that a harvest cycle of three or more years was part of the business cycle and that the harvest cycle was related to meteorological conditions (shown in part in tree ring data), and the regular fluctuations in meteorological conditions were partly the result of fluctuations in solar radiation. Although Jevons’s sunspot theory was often ridiculed, John Maynard Keynes’s (1936, 531) cautious conclusion is to be preferred: “The theory was prejudiced by being stated in too precise and categorical a form. Nevertheless, Jevons notion, that meteorological phenomena play a part in harvest fluctuations and that harvest fluctuations play a part (though more important formerly than to- day) in the trade cycle, is not to be lightly dismissed.” A. C. Pigou was an influential contemporary of Keynes who gave his imprimatur to the idea that fluctuations in the weather contributed significantly to the trade cycle, especially in countries such as the United States where agriculture was an important part of overall economic activity (Pigou 1927). The American economist Henry Ludwell Moore (1921) argued that the business cycle was produced by the “transit of Venus.” Every eight years, Venus stands between the Earth and the Sun, disrupting the Sun’s radiation on its path to the earth. The result, according to Moore, was a regular eightyear rainfall cycle (identifiable in part by evidence from tree rings), a regular eight- year crop cycle, and a regular eight- year business cycle. Weather driven fluctuations in harvests also play a role in accounts of particular episodes. Indeed, the business cycle at the end of the nineteenth century has often been described by economic historians as a product of fluctuations in weather. Milton Friedman and Anna J. Schwartz (1963, 98) argued that the cyclical expansion from 1879 to 1882 was reinforced by “two successive years of bumper crops in the United States and unusually short crops elsewhere.” Katherine Coman (1911, 315) thought that the bumper crop of 1884 had produced the opposite effect because it sold for low prices: “The wheat crop of 1884 was the largest that had ever been harvested, and the price fell to sixty four cents a bushel, half that obtained three years before.” As a result, there was a rash of bankruptcies in the wheat growing areas and the “inability of the agriculturists to meet their obligations to Eastern capitalists and to purchase the products of Eastern mills and workshops, extended and prolonged the industrial depression.” Wesley

Droughts, Floods, and Financial Distress in the United States

75

Clair Mitchell (1941, 2) argued that the recovery from the 1890 financial crisis was partly a harvest driven event: “Unusually large American crops of grain, sold at exceptionally high prices, cut short what was promising to be an extended period of liquidation after the crisis of 1890 and suddenly set the tide of business rising.” O. M. W. Sprague (1910, 154) attributed the severity of the depression that followed the crisis of 1893 to low farm prices and high farm mortgages. Ernest Ludlow Bogart (1930, 690) agreed that the farm sector was heavily involved in the depression of the 1890s because of “the ruinous failure of the corn crop in 1894, and the falling off of the European demand for wheat, the price of which fell to less than fifty cents a bushel.” The poor corn harvest was the result of drought (New York Times, August 4, 1894, 1; August 5, 1894, 8; and subsequent stories). Friedman and Schwartz (1963, 140) concluded that the economic revival after 1896 was reinforced by “another one of those fortuitous combinations of good harvest at home and poor harvests abroad that were so critical from time to time in nineteenth- century American economic history.” A. Piatt Andrew (1906) surveyed many of these individual episodes. He concluded that although corn, cotton, and wheat were the most important U.S. crops and that all influenced the business cycle, the latter two especially through exports, it was fluctuations in the value of the wheat crop that had the most impact on the business cycle. The reason was that wheat was an international crop and, hence, the influence of the American harvest could be offset or reinforced by the success or failure of wheat crops abroad. Recent work by Davis, Hanes, and Rhode (2009) has reinforced the view that weather- driven harvest events influenced the macroeconomy in the period between the U.S. return to the gold standard after the Civil War and World War I. The channel ran through the balance of payments: strong cotton exports produced increased imports of gold, expansion of the money supply, and lower interest rates. However, they challenge the claim of earlier writers that wheat and corn harvests mattered, finding little statistical evidence for a relationship running from the wheat or corn to industrial production. A related literature that focuses more on regions and individual states emphasizes that the restrictions on branch banking in the United States weakened the U.S. banking system, especially when compared with foreign systems that permitted branch banking, such as the Canadian system (Bordo, Rockoff, and Redish 1994; Calomiris 2000, chapter 1; Ramirez 2003). Why? There are several possibilities. A recent paper by Carlson and Mitchener (2009), for example, argues that branch banking increased stability in the 1930s by increasing competition and, thus, forcing more prudent behavior on competing banks and branches. Clearly, however, an obvious potential explanation for the apparent stability of branch banking systems is that branch banking permitted banks to diversify local weather- related agricultural shocks. The main purpose of this study is to determine the frequency and severity of weather generated banking stress in American financial history.

76

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

We first describe (in sections 3.2 and 3.3) two classic examples of weatherdriven economic distress: Kansas during the Populist era and Oklahoma during the Dust Bowl. These case studies allow us to bring in some qualitative evidence on the chain running from weather to agriculture to financial markets and allow us to explore alternative measures of bank stress that can be used in formal econometric tests. These examples, it seems to us, establish a strong prima facie case that that weather driven economic distress can have an important impact on the local financial system. We then turn to formal panel regressions. First, we examine the effects of extreme weather on farm incomes and mortgage foreclosure rates. We find significant and substantial effects. We then turn to panel regressions that directly test the effects of extreme weather on banking markets. 3.2

“In God we trusted, in Kansas we busted”

One of the best known examples of weather- driven financial distress in U.S. history comes from Kansas between the Civil War and 1900. This was the period in which the Populist movement took hold, and Kansas became famous for a motto emblazoned on the covered wagons of farm families leaving Kansas: “In God we trusted, in Kansas we busted.” There were three severe droughts in Kansas in the postbellum era. These show up clearly in figure 3.1, which plots the Palmer Drought Severity Index (PDSI) reconstructed from data on thickness of tree rings for Kansas from 1870 to 1900. The first year of severe drought after the Civil War, 1874, was the year the famous locust swarms that devastated plains farmers. A second postbellum drought followed in 1879 to 1881. Four years of good rain from 1882 through 1885 contributed to a land boom in western Kansas, but drought struck again in 1886 through 1888. Again, several years of good rain followed. But another drought, one of the most prolonged of the century, hit from 1893 through 1896. We have found little discussion in the financial history literature about Kansas during the first drought. More is available, however, about the second and subsequent droughts. In particular, we have Allan G. Bogue’s classic Money at Interest (1955, 103–9), which describes the experience of J. B. Watkins and Company, a major supplier of mortgage money in western Kansas, and other mortgage bankers. When crops failed during the 1879 to 1881 drought farmers besieged Watkins’s agents, hoping for loans to tide them over or to provide the basis after they defaulted for a new start elsewhere. In those circumstances, it was hard for Watkins to make safe loans because desperate farmers and their friends were willing to attest to any value for a property in order to get some cash. In the end, Watkins was stuck with a large number of defaults, and for a time he stopped lending in some of the western counties. This experience, however, failed to prevent a rapid surge of development

Droughts, Floods, and Financial Distress in the United States

Fig. 3.1

77

The Reconstructed Drought Severity Index for Kansas, 1870–1900

Source: See the discussion of the reconstructed Palmer Drought Severity Index.

in the early 1880s when rain again became abundant. Drought, however, as shown in figure 3.1, returned in 1886. Again, Watkins responded by cutting off lending in the affected areas (Bogue [1955] 1969, 144–45). Even so, Watkins ended up holding large amounts of land as a result of foreclosures (Bogue [1955] 1969, 167). It is not surprising that in these years of drought, poor crops, and foreclosures, the local farmers turned to Populism. Indeed, James H. Stock (1984) has shown that support for Populism, nationally, was closely tied to mortgage foreclosures. The final drought in Kansas in the nineteenth century lasted, as shown in figure 3.1, four years from 1893 through 1896. This was an unusually prolonged drought. One would have to go back to the Civil War Years or forward to the 1950s to find periods in which a four- year average of the PDSI was as low as it was in the mid- 1890s.1 It was also a period of international financial distress following the Panic of 1893, and as was often the case when there was a depression of international scope, a period of low prices for basic agricultural products. Kansas, in other words, was hit by a perfect storm (perfect lack of storms?): insufficient rain to grow familiar crops, an international financial crisis and depression, and low prices for agricultural products. The drought and depression of the 1890s had a severe impact on the finan1. It is interesting to note that there was also a sustained drought during 1855 through 1857, the years of “bleeding Kansas.”

78

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

cial system of Kansas. Most of the western mortgage companies, including J. B. Watkins, failed (Bogue [1955] 1969, 187–92). These companies had been raising capital in the eastern United States and in Europe, some of it with mortgage- backed securities (Snowden 1995)—securities similar to those that underlay today’s financial crisis. Therefore, Kansas’s financial difficulties spread quickly. Once again, farmers left Kansas with the motto “In God We Trusted, in Kansas we Busted” emblazoned on their “prairie schooners.” The farmers who used the motto, despite the hardships they had endured, were not always done with pioneering. The Emporia Daily Gazette (Emporia, KS, Monday, August 21, 1893) reported a line of prairie schooners bearing the motto “In God we trusted, in Kansas we busted. So now let ’er rip for the Cherokee Strip.” How did these developments that were clearly important to farmers and mortgage brokers like Watkins affect the national (federally chartered) banks in Kansas? This is an important issue for us because the data that are available consistently and for long periods of time are for the national banking system. We, therefore, need to know whether the national banking system was affected by the Kansas droughts and, if so, which measures of the health of the system were sensitive to the distress produced by extreme weather and poor harvests. Although it may seem plausible that the national system was affected along with other sectors, it is by no means a sure thing. The national system was subject to a different set of regulations than other sectors of the financial system, particularly with respect to real estate loans, and may even have profited to some degree during periods of financial distress from the transfer of funds to what was regarded as the safer part of the financial system. The Comptroller of the Currency, the regulatory authority for the national banks, in those days provided an explanation for every failure of a national bank. Table 3.1 shows the thirty- four national banks that failed in Kansas between 1875 and 1910 and the explanations given by the comptroller. The 1890s were the hard years. In 1890, the worst year, there were seven failures. Most of the banks that failed in 1890 and 1891 had been in operation for only a few years (the average was five): they were creatures of the boom. The exception was the First National Bank of Abilene, which had been in existence for eleven years before it failed. In the early 1890s, as in other periods, the comptroller tended to attribute failures vaguely to injudicious banking, or more informatively to excessive loans to particular stakeholders, or fraud. In 1890 and 1891, however, the comptroller mentions real estate four times. After that, however, real estate is cited only once more, in 1896. “Stringency” in the money market, on the other hand, is not mentioned before 1893, but is given as a reason in three of the failures that occur in that year. Thus, this evidence is consistent with the notion that the national banking sector in Kansas was hard hit by the real estate boom and bust and that the distress resulting from the boom and bust was aggravated by the international financial crisis.

1876 1878 1890 1890 1890 1890

1890 1890 1890 1891 1891 1891

1891 1891

1892 1892

1892 1893

1893 1893

FNB Belleville FNB Meade Center American NB, Arkansas City FNB Ellsworth Kansas SNB McPherson Kansas Pratt County NB

FNB Kansas City FNB, Coldwater Kansas

FNB, Downs Kansas Cherryvale NB

FNB, Erie Newton NB

FNB, Arkansas Citya FNB, Marion

Failure

1885 1883

1889 1885

1886 1890

1887 1887

1885 1887 1889 1884 1887 1887

1872 1872 1879 1886 1886 1887

Organized

50,000 75,000

50,000 65,000

50,000 50,000

100,000 52,000

50,000 50,000 100,000 50,000 50,000 50,000

50,000 50,000 50,000 50,000 75,000 50,000

Capital

National bank failures in Kansas, 1875–1910

FNB Wichita Merchants NB, Fort Scott FNB Abilene State NB, Wellington Kingman NB FNB Alma

Bank

Table 3.1

Defalcation of officers and fraudulent management Investments in real estate and mortgages, and depreciation of securities Excessive loans to others, injudicious banking, and depreciation of securities Injudicious banking and failure of large debtors Investments in real estate and mortgages, and depreciation of securities Excessive loans to officers and directors, and investments in real estate and mortgages Excessive loans to officers and directors, and depreciation of securities Injudicious banking and depreciation of securities Excessive loans to officers and directors, and depreciation of securities Excessive loans to others, injudicious banking, and depreciation of securities Fraudulent management and injudicious banking Excessive loans to officers and directors, and investments in real estates and mortgages Excessive loans to officers and directors, and depreciation of securities Excessive loans to officers and directors, and investments in real estates and mortgages Injudicious banking and depreciation of securities Fraudulent management, excessive loans to officers and directors, and depreciation of securities Injudicious banking and depreciation of securities General stringency of the money market, shrinkage in values, and imprudent methods of banking Excessive loans to officers and directors, and depreciation of securities General stringency of the money market, shrinkage in values, and imprudent methods of banking (continued)

Reason for failure

1898 1898 1899 1899 1905 1908

NB of Paola FNB Emporia Atchison NB FNB McPhersonb FNB Topeka FNB Fort Scott

1887 1872 1873 1886 1882 1871

1886 1882 1883 1887 1888 1882 1883

1884

Organized

100,000 50,000 70,000 50,000 50,000 50,000

52,000 50,000 50,000 60,000 75,000 50,000 50,000

50,000

Capital

General stringency of the money market, shrinkage in values, and imprudent methods of banking Excessive loans to others, injudicious banking, and depreciation of securities Depreciation of securities Injudicious banking and depreciation of securities Injudicious banking and failure of large debtors Investments in real estate and mortgages, and depreciation of securities Injudicious banking General stringency of the money market, shrinkage in values, and imprudent methods of banking Injudicious banking and failure of large debtors Fraudulent management Excessive loans to others, injudicious banking, and depreciation of securities Failure of large debtors Failure of large debtors Fraudulent management and injudicious banking

Reason for failure

Source: Annual Report Comptroller of the Currency 1910, table 44. Notes: FNB stands for First National Bank. The location of the bank is shown when it is not part of the name of the bank. a Temporarily restored to solvency before finally failing in 1899. b In voluntary liquidation, prior to failure.

1894 1894 1895 1896 1896 1896 1896

State NB, Wichita Wichita NB FNB Wellington Humbolt FNB Sumner NB, Wellington FNB, Larned FNB Garnett

Failure

1893

(continued)

Hutchison NB

Bank

Table 3.1

Droughts, Floods, and Financial Distress in the United States

81

The most important economic reason for failure, taking the period as a whole, was “depreciation of securities,” which was mentioned in sixteen cases. The nature of these “securities” is not clear from the information in the Comptroller’s Reports. It might be possible to learn more in the archives of the comptroller, where detailed records of the liquidation of closed national banks are available. One possibility is that they were mortgage- backed securities issued by the land companies. Holding these securities was probably not consistent with the provisions of National Banking Act then in effect that prohibited lending on real estate, but banks might have held them anyway. After all, they were “securities.” Our guess, however, is that many of the securities held by the banks that failed were railroad and municipal bonds. There had been a railroad construction boom in Kansas in the years leading up to the debacle of the 1890s fueled by expectations of rapid expansion of agriculture (Miller 1925, 470–71). Many of these railroads went bankrupt. And it seems probable that many of the securities issued by these railroads had been taken initially by local banks. Overall, capital in national banks in Kansas expanded rapidly during the boom of the 1880s, reached a peak in 1890, and then declined for a decade.2 Total capital began to rise at the turn of the century and finally surpassed the 1890 level in 1908, but even then the par value of outstanding shares was still below the 1890 level. This suggests that most of the growth after 1899 was due to reinvestment of bank profits rather than outside investment. Another indicator of the health of the national banks, national bank lending rates, however, does not provide a clear indicator of Kansas’s struggles. When we compared bank lending rates in the Western Plains with the national average for 1888 (the first year that is available) to 1910, we found that the droughts and agricultural distress of the 1880s and 1890s did not leave a clear imprint on rates charged, despite the clear story told by the chronology of failures. The rate of return to national bank equity in Kansas, however, shown in figure 3.2, tells a story closer to what we learn from Bogue, from the comptroller’s analyses of national bank failures, and from the aggregate capital figures. Here we can clearly see the boom of the mid- 1880s and then the collapse as the bubble burst, a downturn that precedes the national business downturn. The drought of 1887 leaves a strong impact on rates of return to equity. From 1894 on, however, the returns to national bank capital in Kansas follow the national average, suggesting that the adaptation to the postdistress world had begun. 3.3

The Dust Bowl

The Dust Bowl of the 1930s, another classic case of weather- driven economic distress, was most severe in Oklahoma, the Texas Panhandle, and 2. The story is much the same whether one looks at bank capital in nominal or real terms.

82

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Fig. 3.2 The rate of return to equity in national banks, Kansas and the national average, 1870–1910 Source: State level rates of return to bank equity compiled by Scott A. Redenius; see text.

parts of New Mexico during 1930 to 1936. The dust storms are legendary, and the economic distress became an enduring part of the nation’s cultural landscape with the publication of John Steinbeck’s The Grapes of Wrath (1939). Somewhat surprisingly, given the role of the Dust Bowl in American cultural history, the Instrumental Palmer Drought Severity Index for Oklahoma and Texas shows that the drought of the 1930s, although severe, was far from extraordinary. By this measure the drought was much less severe than the drought that hit in the 1950s. Hansen and Libecap (2004) explain that the severity of the agricultural crisis in the 1930s was due in part to the prevalence of small farms that did not engage sufficiently in practices to limit wind erosion. The economic suffering, of course, was greatly aggravated by the low farm prices that prevailed during the Depression. Nevertheless, a closer look at this episode will shed some light on how weather- driven agricultural distress challenges local financial markets. Historians of western banking are clear that the distress that resulted from the Dust Bowl was felt much more intensely by the small state chartered banks and private banks that served rural parts of the affected regions, rather than by the national banks that served urban areas. In particular, the rural banks lost money on livestock loans (Doti and Schweikart 1991, 144). Indeed, the national banks may have benefited to some degree from an

Droughts, Floods, and Financial Distress in the United States

83

attempt by depositors to switch funds from risky rural banks to the larger and safer national banks. Loan losses, of course, were the main problem faced by the rural banks, but these banks also faced another problem that also became part of the cultural landscape: a rash of daring bank robberies. The most notorious of all the Oklahoma bank robbers was Charles “Pretty Boy” Floyd, who by 1934 had become the FBI’s “public enemy number one” (Smallwood and Oklahoma Heritage Association 1979, 120–21). Bonnie and Clyde ranged over a wider area of the Midwest. Financial distress in turn may have aggravated agricultural distress in the Dust Bowl. It would have been more difficult, for example, for a farmer who wanted a loan to tide him over the hard times to get one if the bank that he had developed a relationship with had failed or was deeply distressed. It would also have been harder for a farmer who wanted to borrow to expand his holdings by purchasing smaller failed farms to get the credit to do so. Conceivably, the Canadian banking system in which banks in rural areas, including drought stricken areas, were branches of large nationwide systems was better able to provide services in areas affected by extreme drought.3 Although the effects of the Dust Bowl were felt most keenly by the state and private banks, we need to look at the indicators for the national banks because these are the indicators that are available on a consistent long- term basis for most states. It appears that national bank lending rates were somewhat higher in the Dust Bowl region in the period 1933 to 1936 than might have been expected on the basis of long- term trends: the gap between rates in this region and in other regions rose relative to trend during the Dust Bowl years. The bulge in the premium began to decline in 1937 although it was the end of the decade before the West Lower South premium had returned to trend. Figure 3.3 shows rates of return on national bank equity for Oklahoma and the United States as a whole, for the years 1925 to 1965. The Oklahoma returns were somewhat more volatile in the 1920s and 1930s than the national average. But in general, the Oklahoma nationals do not seem to have fared worse than national banks in other regions. Possibly the increase (compared with trend) in the regional risk premium in the lending rate served to protect national bank earnings in Oklahoma. The great drought of the 1950s, moreover, appears to have left virtually no impact on the rate of return to equity, which was close to the national average. By the 1950s, economic conditions had changed in Oklahoma. The banks had new fields in which to invest: beef production on large scale ranches and, most important, oil (Smallwood and Oklahoma Heritage Association 1979, 149–55). Although space constraints have forced us to leave out many details, the case studies of Kansas prior to 1900 and Oklahoma in the Dust Bowl, we 3. This conjecture is suggested by Bernanke (1983).

84

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Fig. 3.3 The rate of return to equity in national banks, Oklahoma and the national average, 1925–1965 Source: State level rates of return to bank equity compiled by Scott A. Redenius; see text.

believe, establish a strong prima facie argument that extreme weather can cause distress in agriculture and local financial markets. We now turn to formal econometric analysis to determine systematically the impact of extreme weather conditions on agriculture and through agriculture on the banking system. 3.4

Econometric Analysis

We first look at the effect of extreme weather on farm income and farm mortgage foreclosures. Once we show that extreme weather has a negative effect on farm income and farm mortgage foreclosures, we then turn to the effect of extreme weather on local banking systems. In all cases, we regress our various measures of financial and economic stress on our drought indexes. Thus, we are isolating the combined effect of weather on our measures of bank stress (by a linear projection of the determinants of stress onto the space spanned by weather). As weather is clearly exogenous in the short run, we are able to get consistent estimates of the aggregate effect of weather on our measures of financial distress. The effect that we estimate is a reduced form amalgam of the effects of weather through various channels, some of which we do not observe. Therefore, the omission of relevant variables for

Droughts, Floods, and Financial Distress in the United States

85

the determination of bank stress does not affect our estimates of the effect of weather on financial markets. We would expect there to be two possible effects of drought (or excessive rainfall) on the banking sector: a demand effect and a supply effect. (a) A demand effect would arise because drought affects the income of farmers and businesses related to farming. Low farm incomes would mean that farmers were likely to fall behind on loan repayments, which would directly reduce rates of return to equity. Drought, moreover, would lower aggregate demand in the region affected by the drought, which would lower the demand for new loans, which in turn would reduce interest rates and rates of return to equity. (b) A supply effect could arise if the adverse weather conditions and poor harvests led to bank runs and bank failures that in turn reduced the stock of bank capital, reduced the supply of loanable funds, and increased interest rates and rates of return. If the demand effect dominates, then we would expect the rate of return to be positively related to the drought index—that is, in periods of drought, we would see the rate of return on bank capital declining, and in periods of abundant rainfall, we would see the rate of return on bank capital increasing. If the supply effect dominates, we would expect to see the rate of return on bank capital negatively correlated with drought severity. Conceivably, the demand effect could dominate during normal periods with the supply effect only apparent during periods of “high stress,” such as severe drought or flood. In this case, we would expect to see the drought index having a nonlinear effect on the rate of return. We will test for this by allowing for farm income, the foreclosure rates, and the rate of return to equity to be nonlinearly related to the drought index. 3.4.1

Our Bank “Stress Test”

The use of farm income or farm mortgage foreclosures as measures of the effect of extreme weather on agriculture is straightforward, but a word is in order about why we settled on the rate of return to equity in national banks as our “stress test” for banks. There were a number of considerations. First, the rate of return to bank equity was a key decision variable for banks, banking authorities, and the public. It reflected losses due to late payments and reductions in surplus due to the writing down of the value of nonperforming loans, but ultimately it was the return to equity that would determine whether more capital was invested in a bank or whether, because it earned no income, a bank had to be closed. Second, The data for national banks about their assets and liabilities and their income and expenditures was regularly reported in standardized form to the Comptroller of the Currency. Data on state- chartered banks would be valuable because state banks were important in many of the agricultural states where extreme weather played an important role, as we saw in our case studies of pioneer Kansas and Dust Bowl Oklahoma. Unfortunately, the form in which state bank balance sheets were reported varied from state

86

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

to state, and the crucial income and expenditure data was almost always missing. Third, typically, the national bank data has been used to compute regional bank lending rates. Major contributions include Lance Davis (1965), Richard Sylla (1969), Gene Smiley (1975), and John James (1976). The most recent, and in our view, best data now available are the estimates prepared by Scott A. Redenius (2007a). But as our case studies indicated, the bank lending rate often provided an ambiguous signal concerning the stress that had been placed on the banking system by poor harvests and adverse weather. The rate of return to equity, moreover, is computed from data that were regularly reported to the comptroller with relatively few judgment calls by the historian and is available from 1869. The estimates of bank lending rates, on the other hand, are somewhat synthetic measures and are available from a later date. Scott A. Redenius compiled the data we use and was kind enough to share it with us.4 Fourth, the national banks were subject to various regulations that isolated them somewhat from weather- related problems in agriculture: rules, for example, limiting their investments in mortgages and forcing investment in government bonds. Therefore, our choice national bank profit rates as a measure of stress biases our estimates of the effects of extreme weather on banking downward. The true effect of weather on banking markets is likely to be greater than those we find. The aim of this analysis is to determine whether there was any relationship between drought and our variables of financial stress: farm income, foreclosures on farm mortgages, and the rate of return to bank capital. We also want to determine if there were systematic relationships across the country as a whole or whether any effect we find is confined to only certain regions of the country. We do this by estimating a panel (fixed effects) regression with farm income, the farm foreclosure rate, and the rate of return to bank capital as dependent variables and our drought indexes as the explanatory variables. 3.4.2

Unit Root Analysis Tests

Table 3.2 contains various panel unit root tests for real farm income, the foreclosure rate, the rate of return on bank capital, and for both of our drought indexes. We report two types of panel unit root tests: the first type are tests that assume a common unit root among the variables—in this case, we assume a common unit root across states—and the second type of tests assume that each state has individual unit roots. In all tests, the null hypothesis is that the time series contains a unit root and the alternative is that the time series is stationary. The number of lags used in the panel unit root tests 4. Redenius (20007b) uses this data to explore several hypotheses about the integration of the American banking market. Scholars who wish to use this data should contact Professor Redenius at Brandeis University.

Droughts, Floods, and Financial Distress in the United States Table 3.2

87

Panel unit root p-values for drought, rate of return and foreclosure data Palmer Drought Severity Index

Test Assuming common unit root Levin, Lin, and Chu Breitung Assuming individual unit root Im, Pesaran, and Shin ADF-Fisher

Atmospheric

Reconstructed

Rates of return

Foreclosures

Farm income

0.0000 0.0000

0.0000 0.0000

0.0000 0.0000

0.0024 0.0000

0.9023 0.1522

0.0000 0.0000

0.0000 0.0000

0.0000 0.0000

0.0251 0.0497

0.3965 0.3849

was chosen using the Schwarz-Bayesian Information Criterion (SBIC). The results for all the tests in the following show that the unit root hypothesis can be rejected for all time series except for real income. In almost all cases, the null hypothesis can be rejected at the 1 percent level with only a few tests resulting in a rejection of the null hypothesis at the 5 percent level. Given the results of the unit root tests, we will treat each time series except farm income as a stationary time series so that each series on drought severity, the rate of return on bank capital, and the farm foreclosure rates will enter into our regression equations in levels. For farm income, we use the percentage change in farm income as the dependent variable. 3.5

The of Effect of Drought on Farm Income

In this section, we look at the relationship between drought and farm income for the period 1926 to 1948. The results of these regressions are shown in table 3.3 (nominal farm incomes) and table 3.4 (real farm incomes).5 Although it makes more sense in most situations to expect real shocks such as drought to affect real variables, here the effects on nominal income are of interest because farm loans were fixed in nominal terms. We see significant effects for the United States as a whole both in the regressions explaining nominal income and the regressions explaining real income. If we look at finer census divisions and focus on nominal farm incomes (table 3.3), we see significant linear effects in all of the central farming regions. If we focus on real farm income (table 3.4), we find statistically significant linear effects of drought in the East North Central and West South Central regions. The estimated effect is actually largest for the West North Central region although it is not statistically significant. 5. To get real farm incomes, we simply deflated nominal incomes by the gross domestic product (GDP) deflator. This procedure adjusts for broad movements affecting the whole economy but not for interregional variations.

88

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Table 3.3

Farm income panel regressions (nominal values) Linear specification

Region United States Northeast Midwest South West East North Central West North Central East South Central West South Central

Quadratic specification

PDSI

PDSI

PDSI2

0.0423∗ (0.0061) 0.0294∗∗∗ (0.0141) 0.0544∗∗∗ (0.0094) 0.0346∗∗∗ (0.0073) 0.0406∗∗ (0.0144)

0.0408∗ (0.0058) 0.0284 (0.0146) 0.0438∗∗∗ (0.0084) 0.0344∗∗∗ (0.0070) 0.0396∗∗ (0.0139)

–0.0083∗∗ (0.0034) 0.0091∗∗∗ (0.0039) –0.0180∗∗ (0.0066) 0.0009 (0.0031) –0.0069 (0.0048)

0.0479∗∗ (0.0170) 0.0573∗ (0.0118) 0.0193∗∗∗ (0.0066) 0.0420∗∗∗ (0.0137)

0.0536∗∗ (0.0193) 0.0423∗ (0.0096) 0.0192∗∗∗ (0.0070) 0.0422∗∗ (0.0124)

0.0094 (0.0059) –0.0261∗∗ (0.0071) 0.0061 (0.0108) –0.0005 (0.0038)

Notes: PDSI ⫽ Palmer Drought Severity Index. All standard errors are computed using clustered robust standard errors (in parentheses). ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

3.6

The Effect of Drought on Farm Foreclosures

Drought reduced farm incomes, and as shown by Alston (1983) for the interwar period, thereby increased the rate of farm mortgage foreclosures. Here we provide some additional evidence. Using data from 1926 to 1948 on the number of farm foreclosures per 1,000 farms, we estimated a panel regression with foreclosures as the dependent variable and the atmospheric PDSI as the independent variable. The results from these regressions can be found in table 3.5. There appears to be a significant and nonlinear effect of drought on foreclosures for the whole panel. The region with the biggest effect is the Midwest. When we go to the finer census divisions, we see that the largest effects were in the East North Central and West North Central regions. Both regions also show substantial quadratic effects although the quadratic coefficient is statistically significant only for the wheat growing West North Central states. In this region, a one unit decrease in PDSI caused an increase in farm foreclosures of about 4 per 1,000 farms.

Droughts, Floods, and Financial Distress in the United States Table 3.4

89

Farm income panel regressions (real values) Linear specification

Region United States Northeast Midwest South West East North Central West North Central East South Central West South Central

Quadratic specification

PDSI

PDSI

PDSI2

0.0679∗∗ (0.0276) 0.0329 (0.0160) 0.0884 (0.0542) 0.0344∗∗ (0.0150) 0.0843 (0.0666)

0.0651∗∗ (0.0267) 0.0316 (0.0163) 0.0761 (0.0599) 0.0338∗∗ (0.0148) 0.0811 (0.0643)

–0.0152∗ (0.0091) 0.0114∗∗∗ (0.0046) –0.0210 (0.0126) 0.0034 (0.0042) –0.0224 (0.0231)

0.0673∗∗∗ (0.0268) 0.0978 (0.0805) 0.0431 (0.0239) 0.0458∗∗∗ (0.0173)

0.0727∗∗∗ (0.0312) 0.0807 (0.0877) 0.0430 (0.0236) 0.0466∗∗∗ (0.0159)

0.0089 (0.0096) –0.0298 (0.0157) 0.0089 (0.0162) –0.0017 (0.0043)

Notes: See table 3.3 notes. ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

The previous two sets of results show that weather did affect farm income and the farm mortgage foreclosure rate. We now turn to our central concern: the effect of extreme weather on the banking system as measured by rates of return to capital. 3.7

The Effect of Drought on Rates of Return to Bank Equity

We estimated a linear panel regression with fixed effects for each state in our sample.6 Cluster robust standard errors are used with the clusters 6. The states included in our sample are North East (Massachusetts, Maine, Vermont); Mid-Atlantic (New York, Pennsylvania); East North Central (Illinois, Indiana, Michigan, Ohio, Wisconsin); West North Central (Iowa, Kansas, Minnesota, Missouri, Nebraska, North Dakota, South Dakota); South Atlantic (Florida, Georgia, Maryland, North Carolina, South Carolina, Virginia); East South Central (Alabama, Kentucky, Mississippi, Tennessee); West South Central (Arkansas, Louisiana, Oklahoma, Texas); Mountain (Arizona, Colorado, Idaho, Montana, New Mexico, Nevada, Utah, Wyoming); and Pacific (California, Oregon,

90

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Table 3.5

Farm foreclosures panel regressions Linear specification

Region United States Northeast Midwest South West East North Central West North Central East South Central West South Central

Quadratic specification

PDSI

PDSI

PDSI2

–2.1083∗ (0.2969) –0.8164∗∗ (0.2729) –3.6036∗ (0.4378) –1.4526∗ (0.1775) –1.2055∗∗ (0.4018) –2.1912∗ (0.2879) –4.2253∗∗∗ (0.4866) –1.2081∗ (0.4787) –1.3863∗∗∗ (0.1133)

–2.0014∗∗∗ (0.2744) –0.7223∗ (0.3011) –3.4046∗∗∗ (0.4402) –1.4377∗∗∗ (0.1777) –1.1933∗∗∗ (0.3693) –1.9125∗∗ (0.4530) –4.0059∗∗∗ (0.4768) –1.3690 (0.6477) –1.3953∗∗∗ (0.0696)

0.2033∗∗∗ (0.0493) –0.2998 (0.1859) 0.1589∗∗∗ (0.0506) 0.0621 (0.0765) 0.1199∗ (0.0646) 0.1820 (0.1859) 0.1944∗∗∗ (0.0477) –0.2303 (0.3348) 0.0786 (0.0627)

Notes: See table 3.3 notes. ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

defined by the nine census divisions.7 The robust standard errors are robust to unknown autocorrelation within the time series and unknown heteroscedasticity across the cluster units. We estimate first a model that is linear in the drought severity index (PDSI) and then add a quadratic term (PDSI2) to check whether extreme weather conditions have a nonlinear effect on the rate of return to bank capital. The results from the panel regressions with the rate of return of bank capital are found in tables 3.6 through 3.8. In table 3.6, the results of the linear and quadratic specifications are reported for a panel consisting of the forty- two states in our sample. For both drought severity indexes, we find that there is a significant positive effect of the drought index on the rate of return on bank capital: more rain means a higher return. The quadratic term is significant for the regression over the whole period (from 1900 to 1976 for atmospheric PDSI Washington). Our sample is determined by the availability of the drought index and rate of return to bank equity. 7. We do not include time effects as there is a high degree of correlation between the drought severity indexes across states. By including time effects, we run the risk of capturing weather effects as time effects that we do not want to happen.

Droughts, Floods, and Financial Distress in the United States Table 3.6

91

Panel regression results for rates of return on bank capital (U.S. sample) Full sample period

PDSI-Actual PDSI-Reconstructed

PDSI-Actual PDSI2-Actual PDSI-Reconstructed PDSI2-Reconstructed

1850–1900

1900–1940

1940–1976

Linear specification 0.5065∗ (0.0572) 0.3758∗ 0.1027 (0.0482) (0.0775)

0.7557∗ (0.0754) 0.6971∗ (0.0887)

0.1038∗∗ (0.0466) 0.1886∗ (0.0523)

Quadratic specification 0.4879∗ (0.0528) –0.0870∗ (0.0233) 0.3528∗ 0.0791 (0.0489) (0.0843) –0.0700∗ –0.0664∗∗∗ (0.0206) (0.0406)

0.7185∗ (0.0691) –0.0863∗∗ (0.0344) 0.6509∗ (0.0871) –0.1002∗ (0.0330)

0.1038∗∗ (0.0468) –0.0011 (0.0135) 0.1970∗ (0.0517) 0.0400∗ (0.0146)

Notes: PDSI ⫽ Palmer Drought Severity Index. Using actual atmospheric readings, the full sample period is from 1900 to 1976, and using the drought index reconstructed from tree rings, the full sample period is from 1850 to 1976. The standard errors reported are clustered robust standard errors (in parentheses). ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

and from 1850 to 1976 for the tree ring reconstructed PDSI).8 The negative sign on the quadratic term indicates that periods of extreme drought have a worsening effect on the rate of return to bank capital and that periods of extreme wet can also adversely affect the rate of return to bank capital. When we reestimate the model for different subperiods, also reported in table 3.6, we see that the period from 1900 to 1940 is where the biggest effect is found. The coefficient on PDSI is over 50 basis points higher for this period than for the later period from 1940 to 1976. This shows that the effect of weather on the banking system has not been uniform over time. When using the tree ring reconstructed data, we also see lower estimated coefficients on the PDSI term suggesting that the period from 1900 to 1940 was different in terms of how weather affected the banking system. We had expected to see results for the pre- 1900 period that were similar to those found in the 1900 to 1940 period. However, this was not the case. A possible explanation is that a combination of limited farming activity in 8. The data on the reconstructed PDSI runs from 1850 until 1976, but the sample period for the rate of return on bank capital varies by state. Therefore, we have an unbalanced panel when using pre- 1900 data. Most state rate of return data starts in the late 1860s, but some states only have data from 1890.

92

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

some states in the early part of the sample and a limited number or total absence of national banks in some western states early in the sample have biased the results. Given our results for the United States as a whole (or at least for the fortytwo states in our sample), we now look at the different regions to see if there are regional differences in the effect of drought severity on the rates of return to banks for our sample periods. These results are reported in tables 3.7 and 3.8. In table 3.7, results are reported for the four major U.S. Census subregions of the Northeast, the Midwest, the South, and the West. The results show that the biggest effect of drought on banking stress can be found in the Midwest with a one unit increase in the drought index causing upward of a 100 basis points increase in the rate of return to bank capital. Again, we see that this result is biggest during the period from 1900 to 1940. Table 3.7 reports the results from the quadratic specification for those regions where the quadratic term was significant at the 10 percent level. Unlike the earlier results for the whole sample, we do not find a consistent nonlinear relationship. However, for the Midwest region, we see the same pattern with a large positive coefficient on PDSI and a smaller, but significant, negative coefficient on the PDSI2. One interesting point is that there is a significant effect of drought on rates of return for the pre- 1900 period for the South but not for other regions. The result is not a complete surprise. The South was more dependent on agriculture than the Northeast or Midwest and, hence, more likely to be affected by droughts that reduced farm incomes. Indeed, Davis, Hanes, and Rhode (2009) show that the cotton crop was an important determinant of the business cycle in the late nineteenth and early twentieth centuries. In some of the western areas of new settlement that were highly dependent on agriculture, moreover, we have fewer observations because the national banks came in only after economic development could support larger banks. The expectation that drought- related banking problems should have been greater before 1900 than after may be based on the idea that the internal capital market of the United States was completely integrated after 1900. As shown by Landon-Lane and Rockoff (2007), however, tight integration seems better identified with the post–World War II era. Finally, we break the regions up into smaller Census divisions and report these results in table 3.8.9 We report only those divisions making up the center of the country (coefficients for other regions were uniformly insignificant) and see substantial effects although even in these mainly farming states, the effect varies from region to region. The West North Central region had the biggest effect of drought on rates of return to bank capital. During the period 1900 to 1940, a one unit increase in the drought index increased the rate of return to bank equity by 100 basis points. There was a nonlinear effect for this division: at severe levels of drought, the effect of additional 9. For these regressions, the clusters are the individual states.

Droughts, Floods, and Financial Distress in the United States Table 3.7

Region

Northeast

Rate of return panel regressions by census region

Variable

PDSI PDSI2

Midwest

PDSI PDSI2

South

PDSI PDSI2

West

PDSI PDSI2

Northeast

PDSI

Midwest

PDSI2 PDSI PDSI2

South

PDSI PDSI2

West

93

PDSI PDSI2

Full sample

1850–1900

PDSI-Actual 0.6489∗∗ (0.1442) 0.1996∗ (0.0403) 0.6362∗ (0.0813) –0.2135∗ (0.0391) 0.5064∗ (0.1024) 0.0565∗∗∗ (0.0276) 0.2866∗ (0.0822) –0.0674∗∗ (0.0250)

1900–1940

1940–1976

0.9149∗∗ (0.2823) —

0.4559∗ (0.1287) 0.1287∗∗ (0.0399) 0.0782 (0.0522) —

0.08772∗ (0.1379) –0.2361∗ (0.0421) 0.6327∗ (0.1365) 0.1918∗ (0.0624) 0.6569∗ (0.1072) —

PDSI-Reconstructed 0.3492∗∗ –0.1692 (0.0927) (0.1063) — — 0.4854∗ 0.0766 (0.0692) (0.1458) –0.1075∗ — (0.0210) 0.3267∗ 0.3169∗∗ (0.0267) (0.1125) — —

0.4951∗∗ (0.1592) — 0.8161∗ (0.0871) –0.2105∗ (0.0504) 0.1790∗∗∗ (0.0916) —

0.2490∗∗ (0.1109) –0.0988∗∗ (0.0391)

0.8746∗ (0.1706) –0.1030∗∗∗ (0.0550)

0.0402 (0.1617) —

0.3752∗ (0.0535) –0.0425∗ (0.0139) –0.1667∗∗ (0.0536) —

0.5658∗ (0.1178) — 0.2433∗ (0.0283) — 0.4396∗ (0.0526) 0.0747 (0.0135) –0.1621∗∗∗ (0.0751) —

Notes: PDSI ⫽ Palmer Drought Severity Index. All standard errors (in parentheses) are computed using clustered robust standard errors. Results are only reported for the quadratic term if it was significant. When the quadratic term is not significant, the results for the linear specification are reported. Dashes indicate variable was not used in the estimating equation. ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

drought was larger. However, we do not see a significant effect for this region in the period from 1850 to 1900. As our qualitative discussion of Kansas in the 1890s showed, it is likely that there were effects, but the absence of long runs of data in this region, which was a region of new settlement, means that it is hard to detect the effects econometrically.

94

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Table 3.8

Rate of return panel regressions by selected census subregions

Region

East North Central

Variable

PDSI PDSI2

West North Central

PDSI PDSI2

East South Central

PDSI PDSI2

West South Central

PDSI PDSI2

East North Central

PDSI PDSI2

West North Central

PDSI PDSI2

East South Central

PDSI

West South Central

PDSI2 PDSI PDSI2

Full sample

1850–1900

PDSI-Actual 0.5043∗ (0.0958) –0.3171∗∗∗ (0.0863) 0.6644∗ (0.1118) –0.1933∗ (0.0423) 0.6002∗ (0.0991) — 0.1889∗∗∗ (0.0708) — PDSI-Reconstructed 0.3810∗∗ –0.1369 (0.0850) (0.1375) –0.1161∗∗∗ — (0.0523) 0.5417∗ 0.2121 (0.0979) (0.2146) –0.1004∗ — (0.0236) 0.4499∗ 0.0831 (0.0285) (0.1306) — — 0.2207 0.5878∗∗ (0.1103) (0.1698) — –0.2271∗∗∗ (0.0745)

1900–1940

1940–1976

0.4617∗∗ (0.1377) –0.4032∗∗∗ (0.1637) 0.9590∗ (0.1815) –0.2164∗ (0.0408) 0.7474∗ (0.0681) 0.2491∗∗∗ (0.1052) 0.1398 (0.0767) —

0.1488 (0.1085) — 0.0483 (0.0560) — 0.4872∗∗ (0.1372) — 0.2398∗∗ (0.0497) –0.0502∗∗ (0.0147)

0.8694∗ (0.1111) —

0.1908∗∗ (0.0456) —

0.8389∗ (0.1328) –0.1874∗ (0.0421) 0.4506∗∗ (0.0843) — 0.0169 (0.1216) —

0.2652∗ (0.0358) — 0.6045∗∗ (0.1351) — 0.3021∗ (0.0396) 0.0560∗ (0.0077)

Notes: See table 3.7 notes. ∗∗∗p-value ⬍ 0.1. ∗∗p-value ⬍ 0.05. ∗p-value ⬍ 0.01.

We also find strong effects of drought for the East North Central and East South Central regions for the period 1900 to 1940 and for the West South Central for the period 1940 to 1976. We had expected to see a strong effect for the West South Central before World War II because this was the region hit by the Dust Bowl. This region, however, also includes Texas and Louisiana. It may be that the effects of the Dust Bowl are being obscured by

Droughts, Floods, and Financial Distress in the United States

95

the inclusion of neighboring areas that did relatively well during the Depression because of the growth of petrochemicals or for other reasons. It may also be, as suggested by our case study of Oklahoma, that the damage was concentrated in the state banks and that the shift of funds from the state banks to the national banks offset some of the pressures on the national banks, the source of our data. Our results, to sum up, suggest that drought affected the rates of return to bank capital, and in some cases, this effect was economically significant. We find that the effect was largest in the early twentieth century and in the Midwest. 3.8

Conclusions and Conjectures

Did drought or excessive rainfall produce distress in agriculture and the banking systems in rural areas? In many cases it did. We explored two famous historical cases, Kansas after the Civil War and Oklahoma during the Dust Bowl of the 1930s, in some detail. In both cases, there were several factors at work producing distress in the banking system, but drought made things worse. To explore these relationships more systematically, we turned to panel data regressions. First, we tested for relationships between drought and farm income and drought and farm mortgage foreclosures in the interwar years. We found statistically and economically significant relationships for the central farming regions. Then, we tested for relationships between the rates of return to bank equity (a sensitive measure of the challenges facing a banking system) to the Palmer Drought Severity Index and where appropriate to estimates of the Palmer index derived from data on the thickness of tree rings. These regressions also revealed many statistically and economically significant relationships. Thus, for some regions and periods, we can document a chain running from drought to farm income to farm foreclosures to bank distress. While the evidence for weather related banking distress is clear, we also found evidence of adaptation. Our case studies showed that a combination of extreme weather and macroeconomic disturbances could stagger a state banking system for a time but also that people and institutions adapted. Farmers began to grow new crops, turned to grazing, or simply moved on to other activities or other places; bankers learned to finance less vulnerable sectors of the local economy. Our econometric evidence shows that weatherrelated bank stress was more important prior to 1940. The declining role of agriculture and the increased integration of financial markets in the postwar era seem to have cushioned local banks from the full effects of local weather shocks after 1940. Our results may provide some useful lessons as global warming begins to take a larger toll. One implication may be that large branch banking systems are better able to sustain localized drought induced economic stress than smaller systems. This consideration argues against recent calls for breaking

96

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

up large banks on the grounds that it would be easier to avoid the adverse incentive effects of “too big to fail,” the argument that when banks know they are too big too fail they take excessive risks. However, big banks that branch across regions, as the larger American banks now do, or as the Canadian banks did throughout their history, may be better able to offset temporary regional losses resulting from droughts or excessive rainfall with surpluses earned in other regions. The American experience, ironically, may have special relevance for small nations facing the problem of climate stress. The American states, in the periods we examined, in many ways resembled small open economies linked by fixed exchange rates and free trade. Each state had its own banking system. An adverse weather event, if piled on top of a general economic depression, had the potential to create severe distress within the local banking system. The creation of the Federal Reserve, which produced high- powered money acceptable in all states, ameliorated the problem. Branch banking that linked the banks in vulnerable states to larger national systems also contributed to breaking the relationship between local droughts and local banking market distress. The analogs in the international sphere would be multinational banks that branched into small nations and international financial institutions, such as the International Monetary Fund and the World Bank, which helped integrate financial markets. Perhaps there are some lessons here for policymakers wrestling with the question of how best to prepare small open economies for the risks of weather- driven banking problems.

References Alston, Lee J. 1983. Farm foreclosures in the United States during the Interwar Period. Journal of Economic History 43 (4): 885–903. Andrew, A. Piatt. 1906. The influence of the crops upon business in America. Quarterly Journal of Economics 20 (3): 323–52. Bernanke, Ben S. 1983. Nonmonetary effects of the financial crisis in the propagation of the Great Depression. American Economic Review 73 (3): 257–76. Bogart, Ernest Ludlow. 1930. Economic history of the American people. New York: Longmans. Bogue, Allan G. [1955] 1969. Money at interest: The farm mortgage on the middle border. Lincoln, NE: University of Nebraska Press. Bordo, Michael D., Hugh Rockoff, and Angela Redish. 1994. The U.S. banking system from a northern exposure: Stability versus efficiency. Journal of Economic History 54 (2, Papers presented at the fifty- third annual meeting of the Economic History Association): 325–41. Calomiris, Charles W. 2000. U.S. bank deregulation in historical perspective. New York: Cambridge University Press. Carlson, Mark, and Kris James Mitchener. 2009. Branch banking as a device for discipline: Competition and bank survivorship during the Great Depression. Journal of Political Economy 117 (2): 165–210.

Droughts, Floods, and Financial Distress in the United States

97

Coman, Katharine. 1911. The industrial history of the United States. New and revised ed. New York: Macmillan. Davis, Joseph H., Christopher Hanes, and Paul W. Rhode. 2009. Harvests and business cycles in nineteenth- century America. NBER Working Paper no. 14686. Cambridge, MA: National Bureau of Economic Research. Davis, Lance E. 1965. The investment market, 1870–1914: The evolution of a national market. Journal of Economic History 25 (3): 355–99. Doti, Lynne Pierson, and Larry Schweikart. 1991. Banking in the American west: From the gold rush to deregulation. Norman, OK: University of Oklahoma Press. Friedman, Milton, and Anna Jacobson Schwartz. 1963. A monetary history of the United States, 1867–1960. Princeton, NJ: Princeton University Press. Hansen, Zeynep K., and Gary D. Libecap. 2004. Small farms, externalities, and the Dust Bowl of the 1930s. Journal of Political Economy 112 (3): 665–94. James, John A. 1976. The evolution of the national money market, 1888–1911. Journal of Economic History 36 (1): 271–75. Jevons, H. Stanley. 1933. The causes of fluctuations of industrial activity and the price- level. Journal of the Royal Statistical Society 96 (4): 545–605. Jevons, William Stanley. 1884. Investigations in currency and finance. London: Macmillan. Keynes, J. M. 1936. William Stanley Jevons 1835–1882: A centenary allocation on his life and work as economist and statistician. Journal of the Royal Statistical Society 99 (3): 516–55. Landon-Lane, John, and Hugh Rockoff. 2007. The origin and diffusion of shocks to regional interest rates in the United States, 1880–2002. Explorations in Economic History 44 (3): 487–500. Miller, Raymond Curtis. 1925. The background of populism in Kansas. Mississippi Valley Historical Review 11 (4): 469–89. Mitchell, Wesley Clair. 1941. Business cycles and their causes. Berkeley, CA: University of California Press. Moore, Henry Ludwell. 1921. The origin of the eight- year generating cycle. Quarterly Journal of Economics 36 (1): 1–29. Pigou, A. C. 1927. Industrial fluctuations. London: Macmillan. Ramirez, Carlos D. 2003. Did branch banking restrictions increase bank failures? Evidence from Virginia and West Virginia in the late 1920s. Journal of Economics and Business 55 (July–August): 331–52. Redenius, Scott A. 2007a. New national bank loan rate estimates, 1887–1975. In Research in economic history. Vol. 24, ed. Alexander J. Field, Gregory Clark, and William A. Sundstrom, 55–104. San Diego: JAI Press. ———. 2007b. Regional economic development and variation in postbellum national bank profit rates. Business and Economic History On-Line, vol. 5. http://h- net .org/~business/bhcweb/publications/BEHonline/2007/redenius.pdf. Smallwood, James, and Oklahoma Heritage Association. 1979. An Oklahoma adventure: Of banks and bankers. Oklahoma Horizons Series. 1st ed. Norman, OK: University of Oklahoma Press. Smiley, Gene. 1975. Interest rate movement in the United States, 1888–1913. Journal of Economic History 35 (3): 591–620. Snowden, Kenneth. 1995. Mortgage securitization in the United States: Twentieth century developments in historical perspective. In Anglo-American financial systems: Institutions and markets in the twentieth century, ed. M. Bordo and R. Sylla, 261–98. New York: New York University Press. Sprague, O. M. W. 1910. History of crises under the national banking system. 61st Cong., 2d sess., S. Doc. 538. Washington, DC: GPO.

98

John Landon-Lane, Hugh Rockoff, and Richard H. Steckel

Stock, James H. 1984. Real estate mortgages, foreclosures, and midwestern agrarian unrest, 1865–1920. Journal of Economic History 44 (1): 89–105. Sylla, Richard. 1969. Federal policy, banking market structure, and capital mobilization in the United States, 1863–1913. Journal of Economic History 29 (4): 657–86.

4 The Effects of Weather Shocks on Crop Prices in Unfettered Markets The United States Prior to the Farm Programs, 1895–1932 Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

4.1

Introduction

Recently, much attention has been given to studying the effects of human contributions toward an increasingly warmer, wetter, and more variable climate. If we assume that the climate change scenarios are correct, it is important to determine the economic implications of not only global increases in temperature and precipitation, but also how local variation in weather may affect productive activities. Changes in temperature and precipitation, and weather disasters like droughts, floods, heat waves, and blizzards, have direct effects on crop yields and the vitality of farm animals. These weather events may affect the prices that farmers receive for their products, thus, also farm incomes and land values. The price effects of localized supply shocks at the farm- gate level will differ across types of farm commodities. Local weather shocks may have minimal influence on the local prices of crops sold in international markets. However, for crops primarily used and sold at the local level, such disasters may lead to significant local price responses.1 During the World Trade Organization negotiations, the rest of the world has been pressuring the United States and Europe to cease interfering with agricultural markets while trying to support their domestic farmers. Most modern studies of agricultural price responses to weather shocks have been Jonathan F. Fox is a postdoc at the Max Planck Institute for Demographic Research. Price V. Fishback is the Thomas R. Brown Professor of Economics at the University of Arizona, and a research associate of the National Bureau of Economic Research. Paul W. Rhode is a professor of economics at the University of Michigan, and a research associate of the National Bureau of Economic Research. This research was supported, in part, by NSF SES- 0921732. 1. This could be due to tariffs or other trade barriers or, alternatively, is simply a result of the inherent properties of the good (i.e., the good is heavy, bulky, or does not store well).

99

100

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

focused on these heavily regulated markets, but such studies provide little information on how unfettered markets, which might arise out of the trade negotiations, will operate. Our goal is to examine the sensitivity of agricultural prices and output to local and nonlocal weather fluctuations over a large span of time in the United States prior to 1932, when markets were relatively unfettered by farm programs. In this chapter, we examine the United States’ three great staple crops (cotton, corn, and wheat) as well as hay. Cotton and wheat are crops with high value- to- weight ratios and ones that are not heavily used in other agricultural productive activities. During the period of consideration, cotton prices, adjusted to 1982 to 1984 dollars, averaged about 113 cents per pound. Wheat prices averaged about 15 cents per pound in 1982 to 1984 dollars. Corn and hay, on the other hand, are both used as feed for livestock and have value- to- weight ratios less than that of cotton or wheat. For comparison, between 1895 and 1932, corn averaged about 9 cents per pound in 1982 to 1984 dollars and hay about 5 cents per pound in the same denomination. Both corn and hay are primarily used and sold at the local level. For example, at the beginning of our period of consideration, over 77 percent of corn was retained and consumed in the county of production, whereas only about 40 percent of wheat and only a negligible fraction of cotton was consumed in the locality where it was grown. We expect that when agricultural commodities have high transportation costs and are heavily used in productive activities at the local level, prices will be sensitive to changes in local weather. Conversely, for commodities sold in nonlocal markets, prices will be affected much less by changes in local weather but will be sensitive to geographically broad changes in weather conditions. While we do not explicitly estimate the relationship between transportation costs and price volatility, by looking at the differences between two crops with relatively high value- to- weight ratios and two crops with relatively low value- to- weight ratios, we can explore how the reduction of transactions costs through globalization might potentially mitigate the effects of localized weather shocks. 4.2

Local Weather and Drought

Between 1895 and 1932, there was a great deal of variation in weather conditions, yields on different harvests, and commodity prices. In his 1926 Business Annals, published for the National Bureau of Economic Research, Willard Thorp assembled narrative information from commercial sources on the success or failures of the cotton, corn, and wheat harvests over the previous century as well as data on crop prices. Within our period, he reported multiple instances of record crop harvests as well as several years of failures. For example, 1921 witnessed poor harvests of both corn and wheat and a failure of the cotton harvest, leading to price increases even in the midst of

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

101

a recession (Thorp 1926, 144). As predicted from a supply- driven equilibrium model, good harvests generally led to lower prices and bad harvests to higher prices. Some crops hewed more closely to this relationship than others. Inspection of Thorp’s annals reveals that when the cotton harvest was poor, short, or failed, cotton prices were always listed as rising or being high. This relationship was only generally true for corn and wheat. Many different reasons contributed to the good and bad harvests reported by Thorp in his business annals. Defective seed, pests, and disease played a small role in reducing yields. As has been well documented, the boll weevil, which entered the United States around 1892, was particularly devastating to cotton harvests in the South (Lange, Olmstead, and Rhode 2009). Table 4.1 reports U.S. Department of Agriculture (USDA) estimates of reduction of cotton yield by cause between 1909 and 1932. As was often the case with plant diseases and other insect and animal pests, the effects of the weevil was tied to weather conditions. Specifically, weevil damage was worse in wet, warm years. Even though most people have focused on the destruction wrought by the boll weevil, drought and other weather fluctuations caused more crop losses than did that nasty pest (USDA 1923; Kramer 1983). Droughts, floods, hot winds, and other climatic shocks destroyed numerous harvests across the United States. Heavy rain led to multiple devastating floods on the Ohio, Missouri, and Mississippi basins, ruining harvests and destroying farmland that sometimes took several years to recover. In the Mississippi Basin, there were twelve separate instances of major flooding, culminating in 1927 with the famous flood that led to the Flood Control Act of 1928 (Trotter et al. 1998).2 Drought was also a major problem. According to the USDA estimates, drought conditions in 1911 destroyed a quarter of the corn, wheat, and hay harvests. The drought eliminated thirty- five pounds per acre of the cotton harvest, a loss that is 50 percent higher than the amount of cotton destroyed by the boll weevil in any of the surrounding four years (USDA 1923). Deficient moisture conditions severely curtailed the corn harvests again in 1916 and 1918, and in 1917, frost reduced the corn and wheat yields by nearly 14 and 12 percent, respectively (USDA 1923). The droughts in the late 1910s also contributed to an outbreak of stem rust, which devastated the spring wheat crop in the Northern Plains. The Great Drought of 1930, which USDA Secretary Arthur M. Hyde called “the worst drought ever recorded in this country,” served as a prelude to the infamous Dust Bowl conditions of the 1930s (Hamilton 1982, 850). Given the importance of weather in determining the strength of the cotton, corn, and wheat harvests, it is not surprising that so many studies have been devoted to determining the relationship between temperature, precipi2. These were 1903, 1907, 1908, 1912, 1913, 1916, 1920, 1922, 1923, 1927, 1929, and 1932.

102

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Table 4.1

Year 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 Average 1909–1919 1920–1932

Percent reduction in crop yields by cause, 1909–1932 Deficient moisture 14.9 12.2 9.8 8.1 15.2 7.9 6.8 9.2 15.1 23.8 2.7 2.2 8.6 10.2 7.2 14 24.6 5.3 6.4 4.4 10.8 27.7 8.3 8 11.43 10.59

Excessive moisture 6 5.1 2.6 7.6 2 2.9 5.7 9.1 1.7 0.9 15.3 8.8 4.3 4.9 8 4.9 1.4 3.2 4.9 7.3 7.2 2.8 2.6 3.9 5.35 4.94

Other climatic

Plant disease

Boll weevil

Other insects

7.7 5.3 3.0 5.0 5.9 3.0 6.9 7.0 8.7 4.5 3.2 2.1 3.1 2.4 2.8 2.4 3.0 2.9 2.8 4.9 6.0 6.3 3.5 6.1

4.2 4.3 0.5 4.3 0.5 0.2 1.9 0.9 1.3 2 1.3 1.1 1 0.8 0.7 0.8 2.1 1.5 1.5 1.9 2.3 1.7 2 3.2

6.1 5.1 1.3 3.5 7.5 6.1 10.2 14.2 8.6 5.4 13 19.7 31.2 23.3 19.2 8.1 4.1 7.1 18.5 14.1 13.3 5 8.3 15.2

1.8 2.4 6.6 3 1.4 3.7 2 1.6 3.7 2.6 5.8 4.3 4.2 3.4 7.4 3.9 2.2 8.9 4.4 3.4 2.5 1.9 1.8 3.1

5.48 3.72

1.95 1.58

7.36 14.39

3.15 3.95

Sources: 1925 USDA Yearbook of Agriculture, and the May 1928, July 1930, June 1931, and June 1934 editions of the USDA Crops and Markets publication.

tation, and crop yields. Just within our sample years, Annie Hannay counted over 2,000 studies that examined the influence of weather on crops (Hannay 1931). Standard practice in the agronomic literature is to measure weather through a combination of both linear and nonlinear effects on crop yields. Because both very low and very high levels of precipitation and temperature adversely affect crop yield, assuming a simple linear relationship between weather and prices and subsequently profit would be a misspecification. Generally, the nonlinearity introduced is a quadratic in precipitation and temperature. We choose a similar path but also include measures for drought and wetness conditions using the Palmer Drought Severity Index (PDSI).3 3. We define the term “drought” as a prolonged and abnormal moisture deficiency.

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

4.3

103

Data and Summary Statistics

To study the effects of state- level weather fluctuations on farm- gate prices, we combine two existing data sets; one containing historical crop information, the other historical weather information. State- level information on yield, harvests, and prices for commodities produced and sold across the United States was taken from the Agricultural Time Series-Cross Section Dataset (ATICS), compiled from USDA records by Thomas Cooley, Stephen DeCanio and M. Scott Matthews (1977).4 This data set covers the contiguous United States from 1866 to 1969 although the limited availability of weather data constrained the analysis to the period after 1895. After 1932, the federal government intervened heavily in many agricultural markets, with payments to limit acreage under production, price supports for some agricultural commodities, and the formation of federal crop insurance in 1938. This new legislation, changes in technology, increases in the use of fertilizer, and other factors affecting the relationship between weather and crop yields caused a variety of effects that have yet to be sorted out. To isolate the activities of relatively unregulated markets, we limit the analysis to the thirty- seven years between 1895 and 1932. To adjust for inflation, commodity prices are adjusted to reflect 1982 to 1984 dollars using the Consumer Price Index (CPI) series developed by Lawrence Officer (2008). Table 4.2 gives summary statistics on output and prices per pound in 1982 to 1984 dollars for the selected farm commodities, and table 4.3 gives the summary statistics for the different weather variables used in estimation. There was great temporal and spatial variation in both output and prices for all four of the crops; second, there is also large variation between the commodities. Within any one year, cotton and wheat prices displayed much less regional variation than prices for corn and hay. This is likely a function of two key differences. Corn and hay were used as animal feed on the farm and in local markets, while cotton was used primarily as an input for manufacturing in U.S. and English cities, and wheat was marketed internationally. Further, cotton was much less costly to transport than the other crops. The differences in markets lead us to believe that state- level corn and hay prices will fluctuate more with state- level weather shocks than will wheat and cotton prices. Figure 4.1 shows the distribution of production across the contiguous United States for the four crops in 1929, a year relatively free of inclement weather. The cotton, corn, and wheat belts are clearly visible. Also evident is that corn and hay production is more widely distributed across the country, consistent with the existence of local markets for these commodities. Every 4. Crop prices represent the farm gate price on December 1. The data are freely available from the National Agricultural Statistics Service.

Table 4.2

Statistics for farm commodities, 1895–1932

Crop prices (cents per pound, 1982–1984 dollars) Cotton Corn Hay Wheat Crop output (for producing states) Cotton (bushels) Corn (bushels) Hay (tons) Wheat (bushels)

Table 4.3

Mean

Standard deviation

Min.

Max.

113.50 9.08 5.31 15.16

44.90 3.49 2.25 5.15

31.71 2.08 0.68 3.61

356.71 26.30 13.25 39.00

396,235.10 55,160.22 1,763.120 17,750.530

453,382.700 86,691.08 1,631.243 25,269.70

144 21 40 13

2,697,848 509,507 7,303.5 251,885

Summary statistics for climate variables, 1895–1932

Climate variables Cotton states Average yearly temperature Average yearly precipitation Months of extreme or severe drought Months of extreme or severe wet Moisture index standard deviation Corn states Average yearly temperature Average yearly precipitation Months of extreme or severe drought Months of extreme or severe wet Moisture index standard deviation Hay states Average yearly temperature Average yearly precipitation Months of extreme or severe drought Months of extreme or severe wet Moisture index standard deviation Wheat states Average yearly temperature Average yearly precipitation Months of extreme or severe drought Months of extreme or severe wet Moisture index standard deviation

Mean

Standard deviation

Min.

Max.

61.198 3.644 0.750 0.926 1.680

5.004 1.054 1.640 1.914 0.407

50.564 0.648 0 0 0.754

73.324 6.193 9.857143 11.750 4.021

51.813 2.927 1.196 0.996 1.672

8.029 1.131 2.335 2.022 0.426

35.495 0.366 0 0 0.643

73.324 6.193 12 12 4.021

51.907 2.886 1.154 1.059 1.659

8.073 1.149 2.300 2.109 0.409

35.495 0.366 0 0 0.643

73.324 5.930 12 12 4.021

51.474 2.778 1.299 1.001 1.671

7.374 1.145 2.430 2.036 0.422

35.495 0.366 0 0 0.643

69.046 5.889 12 12 4.021

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

Fig. 4.1

105

Crop shares

state engaged in at least some production of corn and hay, whereas cotton was limited to the more southern latitudes. Wheat was concentrated in the Midwest, with no production occurring in Florida, Mississippi, Alabama, New Hampshire, Massachusetts, Rhode Island, or Connecticut. Monthly data on temperature, precipitation, and the Palmer drought measures were compiled by the National Climatic Data Center (2002). To correct for biases in raw weather data that arose from different measurement times across the stations, the average temperature and precipitation data are adjusted for time of day using the model suggested by Karl et al. (1986). Because the agricultural data are measured annually and the planting and harvesting dates differ among states, for simplicity we convert all weather variables to yearly averages. Summary statistics for average temperature, average precipitation, months of extreme or severe drought, months of

106

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

extreme or severe wetness, and the Palmer Z standard deviation are given in table 4.3. For the period under consideration, these series are stationary in the time series sense.5 The weather in most states producing cotton tended to be warmer and wetter than in states producing corn, hay, and wheat. Temperatures also displayed less variability in the cotton states.6 In all of the states, average yearly precipitation, which is an average of average monthly precipitation from January to December, ranges from about a third of an inch per month to just over six inches per month. Temperature is averaged similarly to precipitation and also represents a twelve- month average of the January to December monthly averages. In our sample, this ranges from about 35°F to about 74°F in the corn, hay, and wheat producing states and from about 50°F to about 74°F in the cotton producing states. Months of extreme or severe drought and extreme or severe wetness are calculated using a form of the PDSI, the Palmer Hydrological Drought Index (PHDI).7 Between 1895 and 1932, some states endured years with serious drought conditions for all twelve months, while other states enjoyed years with twelve months straight of normal moisture levels. The number of months of extreme or severe wetness ranged similarly. 4.4

Price Effects and Transportation Costs

State- level price responses from state- level supply shocks will be mitigated for tradable goods sold in an international market because individual states will play too small a role to affect prices significantly. At the extreme, if a farm commodity is perfectly tradable internationally, the observed state average farm- gate (local) price is a function only of nonlocal, or international, supply and demand, and farm- gate prices vary only with changes affecting the aggregate market.8 Conditional on weather that affects producers as a whole, state farm- gate commodity prices would not be influenced much by state- level weather shocks. Conversely, for perfectly nontradable commodities with prohibitively high transportation or storage costs, the observed state farm- gate price is a function only of supply and demand in the local market. If a severe drought hits, local traders do not import goods from other states or countries to mitigate the price shock. Additionally, weather affecting producers outside of the local area will not affect local prices. 5. This is tested using the Fisher test suggested by Maddala and Wu (1999). 6. During the period under consideration, arid western states began producing cotton using irrigation. 7. The PHDI represents about a year’s worth of abnormal moisture conditions, while the PDSI represents about nine months worth. 8. This statement assumes that transport costs between the farm- gate and markets stay constant over time.

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

Fig. 4.2

107

High transport costs and weak local market, adverse supply shock

Commodities generally fall in between these two extreme cases. Assuming tradability is only a function of the specific properties for a certain crop and it does not vary temporally or spatially, we follow the setup of Mundlak and Larson (1992) and write the observed logged price of commodity c in state s during year t as a function of both the international and local supply and demand: (1)

ln(P os,bc,ts )  c ln(P c,intt)  (1  c)ln(Pslo,cc,t ),

where c is an index of the strength of the local market for crop c. The strength of the market itself is a function of transaction costs and local uses for the crop. Crops such as cotton and wheat, which are easy to store, easy to transport, have a relatively high value- to- weight ratio, and few local uses have a c closer to one. On the other hand, crops such as corn and hay, which are used in other areas of agricultural production and have higher transport costs and stronger local markets, have a c much closer to zero. Figures 4.2 and 4.3 show the differential effects of negative local supply shocks for goods with relatively high and low transport costs. In each of these figures, local supply is upward sloping, relatively inelastic and given by S1 and S2. Also present in each is the internationally determined prices P binuty, the price paid by international buyers and P sinetll, the effective price received

108

Fig. 4.3

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Low transport costs and strong local market, adverse supply shock

by commodity producers after subtracting their marginal cost of bringing the commodity to the international market.9 The distance between P binuty and P sinetll is Ctint, the size of the marginal cost to bring commodity i to the international market. Within each of the figures, the bolded line represents the effective market demand curve faced by suppliers in the local area, and the bolded dashed line represents the relevant areas of the market supply curves for local suppliers. Figure 4.2 represents the local market for commodities such as corn and hay that have higher transport costs and strong local markets. Because the cost of bringing these types of goods to international market is high compared to a good such as cotton, fluctuations in the local market play a larger role. Additionally, fluctuations in the international market will play a relatively smaller role. In figure 4.2, producers begin in an exporting situation on supply curve S1L, where the market quantity Qb is the amount sold at the local level, and the quantity Qa – Qb is exported to the outside market. Because the marginal transport cost Ciint is large, very little of the good is exported, and a relatively small supply shock would cause local producers to sell exclusively to the local market. For example, an adverse weather shock 9. If we assume constant marginal transaction costs, this is also the average cost.

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

109

that causes supply to fall from S1L to S2L drives the quantity produced from Qa to Qc and causes the price at which the farmers sell to rise from Pa to Pb. If the supply reduction were particularly severe, it could potentially drive the local market price above P binuty and cause local markets to import from the outside market. The effect of the international market on the local market is also mitigated. Changes in price determined by the international market, or P binuty, will affect local market prices if it pulls the “band” between P binuty and P sinetll in figure 4.2 above or below the local equilibrium price. Local prices tend to be less susceptible to changes in the international market for goods with higher transport costs and wider bands. Figure 4.3 represents a local market for commodities such as cotton that have low transportation costs and are not generally used and sold at the local level. In this case, the costs associated with bringing the good to the international market, C iint, are small. We again begin in an exporting situation with a local market price at Pa, total production of Qa, and local purchases of Qc. In this situation, however, the production exported to outside markets (Qa – Qb) is much larger. It takes a much larger adverse supply shock to force the local market to rely exclusively on local production. In the event of a supply shock that moves the local supply curve to S2L, the price shock is much lower than in figure 4.2, as the price rises much less from Pa to Pb. Because of the lower transport costs, a local supply shock would be more likely to lead to a situation where local consumers pay the international price paid by buyers. Considering a simplified version for farm commodity prices where weather is the only input, then (2)

ln(Ps,c,t)  c ln{P[Q c,intt (OPWs,t)]}  (1  c) ln{P[Qs,c,t(ws,t)]},

where Q int is the international level of the commodity sold, ws,t is a measure of weather conditions in state s and year t, and OPW is a measure of weather conditions across all other states producing the commodity, defined in section 4.5.1. 4.5

Empirical Model

To test the significance and magnitude of the relationship between farm commodity prices and adverse weather, we use the year- to- year variations in both weather and commodity prices to specify a regression model including state and year fixed effects. Including these fixed effects will net out much of the unobserved variable bias that seems to plague the cross- sectional models present in much of the agronomic literature (Deschênes and Greenstone 2007). In this way, we can look at the entire United States, instead of, for instance, limiting our scope to nonirrigated counties or to states that were

110

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

net exporters of the different crops. Additionally, we can conduct the analysis without worry that states in different climate zones may have different levels of transportation structure. Local weather is measured using time- bias corrected temperature and precipitation, their squared terms, the number of months of extreme or severe drought, the number of months of extreme or severe wetness, and moisture variation. The last three variables are derived from the Palmer Z Index. In addition to the effect of local weather on local prices, we are also interested in the effect of weather fluctuations by other producers who are competing in the national and international market. For this reason, we construct a set of Other Producer’s Weather (OPW) variables. 4.5.1

Other Producer’s Weather

For goods with weak local markets and sold primarily as exports, geographically broad changes in weather conditions affecting the aggregate market play the dominant role in determining local prices. To capture the effect of weather- driven supply shocks in the outside market, we create a set of variables that measure changes in weather affecting all other producing states in the United States. If the United States’ economy was completely closed and trade occurred only between the different states, including these variables would completely capture the effect of an international market. However, while the United States was not a closed economy and cotton and wheat were being bought and sold in a true international market, we argue that including aggregate measures of weather affecting domestic producers will proxy well for shocks to the entire market outside the local state market. This proxy works well because the United States is a sufficiently large portion of the overall international supply for each of the different crops. We construct these OPW variables for each of the four commodities. These weather variables are weighted by each state’s relative share of national production in 1929, which is chosen as the weighting reference year due to its near absence of inclement weather. These variables control for the effect on local farm- gate prices from weather shocks affecting other producers of the commodity. To create the weights, we first calculate the national share of production within state s for commodity c in 1929. This is represented by s,c and defined as (3)

s ,c 

Qs ,c

,

S

∑Q

j ,c

j =1

where Qs,c is the total output of commodity c produced by state s in 1929. We then use this to construct the weighted averages of the different weather variables (OPW). For weather variable W, OPW is defined as

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

111

S

(4)

1 OPWWs,c,t   ∑ j,c ∗ Wj,t. 1  s,c j ≠ s

In the preceding equation, W could be average annual precipitation in state j and year t, average annual temperature in state j and year t, or one of the other weather variables included in the analysis. 4.5.2

Reduced Form Models

For each time period, we estimate the following reduced form models for price and quantity: (5)

ln(Ps,c,t)  s  t  Ws,t c  OPWs,c,tc  ε1,s,c,t,

(6)

ln(Qs,c,t)  s  t  Ws,t c  OPWs,c,tc  ε2,s,c,t,

where ln(Ps,c,t) is the logged real price for commodity c in year t and state s, ln(Qs,c,t) is the logged quantity, s is a set of state fixed effects that control for unmeasured time- invariant determinants of the farm- gate price, t is a set of year indicators that control for unmeasured annual shocks common to all states, Ws,t is a vector of weather variables in year t and state s that could potentially affect local prices, and OPWs,c,t is the vector of weather variables in year t for other producers of commodity c outside state s.10 The disturbance terms ε1,s,c,t and ε2,s,c,t are assumed to have conditional mean zero and defined as the other factors influencing farm- gate prices and output besides weather. Although there are certainly other factors that could potentially affect farm commodity prices, after controlling for fixed effects, it is not likely that these unobserved effects will cause the local weather variables to be correlated with the error term. While the variables that proxy for weather fluctuations in other producing states are weighted by that state’s share of national production, it is also unlikely that the OPW variables will be correlated with the error term. The share of production used to weight the weather in the other states is fixed in 1929 and, thus, cannot vary over time in response to weather shocks. Any influence of the production share in 1929 on the error term will be controlled with the state fixed effects. The dependent variables in the analysis are the logged values of the real prices and quantities for the different commodities. Corn and hay had higher transport costs and were more commonly used in agricultural production than cotton and wheat. Therefore, we expect that the state farm- gate prices of corn and hay were more responsive to adverse local weather shocks than were the state farm- gate prices of cotton and wheat. We might expect all of these commodities to experience changes in prices due to fluctuations in the weather in the rest of the nation. Such fluctuations will influence the placement of the upper and lower prices in the price band created by trans10. After examination of the different output distributions, we concluded that they were all closer to a log- normal distribution than a normal distribution.

112

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

portation and transactions costs. How much the weather outside the state will matter is an empirical question. 4.6

Results

Tables 4.4 to 4.7 show regression results with the logged prices of cotton and corn as the dependent variables (4.4: cotton, 4.5: corn, 4.6: hay, and 4.7: wheat), and tables 4.8 to 4.11 give regression results with logged quantity as the dependent variable. All of the models present in tables 4.4 to 4.11 include state and year fixed effects.11 Column (1) in each table presents results from measuring weather using just temperature, precipitation, and their squared terms. Including this basic model allows comparison to the prior work that used only measures of temperature and precipitation and sets a baseline for comparison when the additional weather variables are included. Column (2) includes the variables controlling for the number of months of extreme or severe wetness and extreme or severe drought. For all of the different crops, including the extreme or severe wetness and drought measures did not affect the coefficients on average yearly temperature or its squared term. Their inclusion tended to slightly attenuate the coefficients for precipitation and its squared term because these variables are measures of the extreme parts of the distribution of drought and wetness that arises from changes in current and prior precipitation. Columns (3–5) represent the different models that include the OPW variables. Column (3) includes just temperature, precipitation, and their squared terms. Column (4) adds in the number of months of extreme or severe drought and wetness, and column (5) includes the standard deviation of the Palmer Z Index to control for effects of changes in weather variability. Tables 4.8 to 4.11 show the results when the logged quantity within the state is estimated as a function of the weather variables. The different crops exhibited different sensitivities to local and nonlocal weather events although there were some commonalities across the tables. In comparisons of specifications for a crop, the coefficient estimates tended to be similar across the different model specifications. The inclusion of the OPW variables and the Palmer drought and wetness measures had little effect on the coefficient estimates for local average temperature and average precipitation. 4.6.1

The Dominant Shifts in Supply or Demand Associated with Weather Changes

Our analysis of local supply and demand adjustments in figures 4.2 and 4.3 focuses on supply shifts because in most cases they are the dominant shifts 11. The inclusion of the fixed effects cause the R2 to be so close to one. Without the fixed effects, the R- squared ranges from 0.15 to 0.35 across the specifications.

The Effects of Weather Shocks on Crop Prices in Unfettered Markets Table 4.4

113

Dependent variable: ln(price of cotton in $1982) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared

(2) Local weather 0.0183 (0.0508) –0.0002 (0.0004) –0.0246 (0.0240) 0.0020 (0.0028) 0.0035 (0.0032) –0.0015 (0.0024)

0.0185 (0.0510) –0.0002 (0.0004) –0.0142 (0.0229) 0.0016 (0.0029)

Months of extreme wetness Months of extreme drought

(3)

0.0491 (0.0640) –0.0004 (0.0005) –0.0319 (0.0235) 0.0037 (0.0029)

(4)

0.0493 (0.0645) –0.0004 (0.0005) –0.0399 (0.0253) 0.0039 (0.0030) 0.0038 (0.0033) –0.0006 (0.0023)

0.0427 (0.0633) –0.0004 (0.0005) –0.0197 (0.0277) 0.0021 (0.0032) 0.0047 (0.0033) –0.0001 (0.0022) –0.0182 (0.0109)∗

–0.0321 (0.0211) 0.0003 (0.0002) –0.0665 (0.0270)∗∗

–0.0331 (0.0213) 0.0003 (0.0002) –0.0614 (0.0296)∗∗

0.0116 (0.0039)∗∗∗ –0.0011 (0.0021) 0.0040 (0.0019)∗∗

0.0108 (0.0042)∗∗ –0.0007 (0.0022) 0.0043 (0.0019)∗∗

Palmer Z Index standard deviation Other producer’s weather –0.0186 (0.0214) 0.0002 (0.0002) –0.0854 (0.0277)∗∗∗ 0.0130 (0.0040)∗∗∗

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation Constant No. of observations Adjusted R2

(5)

–0.0038 (0.0085) 4.1852 (1.5478)∗∗∗ 559 0.97

4.1704 (1.5423)∗∗∗ 559 0.97

5.3998 (2.6576)∗∗ 559 0.97

6.9183 (2.6464)∗∗∗ 559 0.97

7.1733 (2.6305)∗∗∗ 559 0.97

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

associated with local weather changes. For corn and hay, where narrative evidence suggests a great deal of local consumption, the weather coefficients in the tables 4.4 to 4.7 price regressions and the tables 4.8 to 4.11 quantity regressions are consistent with a supply shift dominating any demand effects associated with changes in the weather. The coefficient of local temperature in the crop price equations had the opposite sign of the coefficient of local

114

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Table 4.5

Dependent variable: ln(price of corn in $1982) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness

–0.0411 (0.0178)∗∗ 0.0005 (0.0002)∗∗∗ –0.1306 (0.0311)∗∗∗ 0.0180 (0.0041)∗∗∗

Months of extreme drought

(2) Local weather –0.0421 (0.0182)∗∗ 0.0005 (0.0002)∗∗∗ –0.1166 (0.0336)∗∗∗ 0.0168 (0.0043)∗∗∗ –0.0019 (0.0021) 0.0015 (0.0018)

(3)

(4)

(5)

–0.0435 (0.0180)∗∗ 0.0005 (0.0002)∗∗∗ –0.1277 (0.0312)∗∗∗ 0.0179 (0.0041)∗∗∗

–0.0437 (0.0182)∗∗ 0.0005 (0.0002)∗∗∗ –0.1166 (0.0334)∗∗∗ 0.0170 (0.0042)∗∗∗ –0.0017 (0.0020) 0.0018 (0.0018)

–0.0441 (0.0182)∗∗ 0.0005 (0.0002)∗∗∗ –0.1349 (0.0351)∗∗∗ 0.0186 (0.0043)∗∗∗ –0.0022 (0.0020) 0.0012 (0.0018) 0.0164 (1.82)∗

–0.1545 (0.0346)∗∗∗ 0.0015 (0.0003)∗∗∗

–0.1589 (0.0341)∗∗∗

Palmer Z Index standard deviation Other producer’s weather –0.1357 (0.0376)∗∗∗

Average temperature Average temperature squared Average precipitation

0.0013 (0.0004)∗∗∗ –0.1154 (0.0609)∗

Average precipitation squared Months of extreme wetness

0.0186 (0.0091)∗∗

–0.2021 (0.0697)∗∗∗ 0.0303 (0.0101)∗∗∗ 0.0177 (0.0050)∗∗∗ 0.0101 (0.0036)∗∗∗

Months of extreme drought Palmer Z Index standard deviation Constant No. of observations Adjusted R2

7.0815 (0.4743)∗∗∗ 1,796 0.9

7.1046 (0.4804)∗∗∗ 1,796 0.9

9.6347 (0.8512)∗∗∗ 1,796 0.91

0.0015 (0.0003)∗∗∗ –0.2240 (0.0662)∗∗∗ 0.0332 (0.0095)∗∗∗ 0.0159 (0.0054)∗∗∗ 0.0088 (0.0037)∗∗ 0.0164 (0.0090)∗

10.0489 (0.8175)∗∗∗

10.1610 (0.8062)∗∗∗

1,796 0.91

1,796 0.91

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

temperature in the crop output equations for nearly every crop. This was also true for local precipitation. For example, the local temperature coefficients for corn output in table 4.9 showed that a rise in local temperature raised corn output at a diminishing rate. The coefficients of local temperature in table 4.5 in the corn price regressions had the opposite sign so that increases in local temperature lowered

The Effects of Weather Shocks on Crop Prices in Unfettered Markets Table 4.6

Dependent variable: ln(price of hay in $1982) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness

0.0585 (0.0254)∗∗ –0.0005 (0.0002)∗∗ –0.3334 (0.0471)∗∗∗ 0.0363 (0.0064)∗∗∗

Months of extreme drought

(2) Local weather 0.0584 (0.0255)∗∗ –0.0005 (0.0002)∗∗ –0.2749 (0.0464)∗∗∗ 0.0320 (0.0062)∗∗∗ –0.0055 (0.0033)∗ 0.0129 (0.0027)∗∗∗

(3)

(4)

(5)

0.0850 (0.0298)∗∗∗ –0.0007 (0.0003)∗∗ –0.3340 (0.0473)∗∗∗ 0.0355 (0.0065)∗∗∗

0.0870 (0.0297)∗∗∗ –0.0008 (0.0003)∗∗∗ –0.2734 (0.0468)∗∗∗ 0.0311 (0.0064)∗∗∗ –0.0054 (0.0034) 0.0131 (0.0027)∗∗∗

0.0849 (0.0302)∗∗∗ –0.0008 (0.0002)∗∗∗ –0.3209 (0.0506)∗∗∗ 0.0354 (0.0068)∗∗∗ –0.0063 (0.0034)∗ 0.0118 (0.0027)∗∗∗ 0.0424 (0.0136)∗∗∗

–0.0487 (0.0817) 0.0002 (0.0008) 0.3022 (0.1341)∗∗ –0.0444 (0.0196)∗∗

–0.0598 (0.0820) 0.0003 (0.0008) 0.3101 (0.1503)∗∗

Palmer Z Index standard deviation Other producer’s weather –0.0360 (0.0852) 0.0002 (0.0008) 0.2584 (0.1257)∗∗ –0.0420 (0.0189)∗∗

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought

–0.0088 (0.0092) 0.0090 (0.0082)

–0.0457 (0.0217)∗∗ –0.0080 (0.0098) 0.0091 (0.0085) 0.0036 (0.0415)

7.4538 (0.8671)∗∗∗ 1,152 0.88

7.6300 (0.8729)∗∗∗ 1,152 0.88

Palmer Z Index standard deviation Constant No. of observations Adjusted R2

115

8.0209 (0.6695)∗∗∗ 1,152 0.87

7.8679 (0.6795)∗∗∗ 1,152 0.87

8.1451 (0.8887)∗∗∗ 1,152 0.87

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

farm- gate corn prices at a diminishing rate. The coefficients of local precipitation have similar opposing signs in the table 4.5 price regressions and the table 4.9 quantity regressions in table 4.9. Comparisons of the coefficients in the hay regressions in tables 4.9 and 4.10 show the same opposing signs for the local weather coefficients in the price and quantity regressions. Increases in local temperature decreased hay output at a diminishing rate while raising

116

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Table 4.7

Dependent variable: ln(price of wheat in $1982) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness

0.0158 (0.0144) –0.0001 (0.0001) –0.1012 (0.0249)∗∗∗ 0.0124 (0.0034)∗∗∗

Months of extreme drought

(2) Local weather 0.0185 (0.0147) –0.0002 (0.0001) –0.0852 (0.0253)∗∗∗ 0.0109 (0.0034)∗∗∗ 0.0001 (0.0015) 0.0041 (0.0016)∗∗

(3)

(4)

(5)

0.0216 (0.0138) –0.0002 (0.0001) –0.0870 (0.0248)∗∗∗ 0.0103 (0.0034)∗∗∗

0.0232 (0.0143) –0.0002 (0.0001) –0.0704 (0.0249)∗∗∗ 0.0088 (0.0033)∗∗∗ –0.0009 (0.0015) 0.0030 (0.0015)∗∗

0.0248 (0.0142)∗ –0.0002 (0.0001) –0.0718 (0.0253)∗∗∗ 0.0088 (0.0033)∗∗∗ –0.0012 (0.0015) 0.0028 (0.0015)∗ 0.0028 (0.0066)

–0.1646 (0.0252)∗∗∗ 0.0016 (0.0002)∗∗∗

–0.1721 (0.0255)∗∗∗

Palmer Z Index standard deviation Other producer’s weather –0.1745 (0.0245)∗∗∗

Average temperature Average temperature squared Average precipitation

0.0016 (0.0002)∗∗∗ 0.1655 (0.0403)∗∗∗

Average precipitation squared Months of extreme wetness

–0.0242 (0.0058)∗∗∗

Months of extreme drought

0.1761 (0.0470)∗∗∗ –0.0262 (0.0063)∗∗∗

–0.0211 (0.0068)∗∗∗

–0.0039 (0.0034) –0.0041 (0.0025)

–0.0066 (0.0037)∗ –0.0062 (0.0028)∗∗

Palmer Z Index standard deviation Constant No. of observations Adjusted R2

0.0016 (0.0002)∗∗∗ 0.1364 (0.0497)∗∗∗

0.0448 (4.12)∗∗∗ 6.6023 (0.3803)∗∗∗ 1,597 0.94

6.5575 (0.3848)∗∗∗ 1,597 0.94

6.3888 (0.3637)∗∗∗ 1,597 0.95

6.3510 (0.3729)∗∗∗ 1,597 0.95

6.2901 (0.3733)∗∗∗ 1,597 0.95

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

the farm- gate hay price at a diminishing rate. Meanwhile, increases in local precipitation lowered hay output at a diminishing rate and raised hay prices at a diminishing rate. In the cotton and wheat markets, where transport costs were low and there was limited local consumption, the relationships of local weather with prices and quantities had the opposing signs associated with dominant supply

The Effects of Weather Shocks on Crop Prices in Unfettered Markets Table 4.8

Dependent variable: ln(cotton output) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

1.2643 (0.2333)∗∗∗ –0.0106 (0.0019)∗∗∗ –0.1255 (0.2301) –0.0077 (0.0267)

Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

No. of observations Adjusted R2

(2)

(3)

Local weather 1.2567 1.4058 (0.2302)∗∗∗ (0.2657)∗∗∗ –0.0106 –0.0118 (0.0019)∗∗∗ (0.0022)∗∗∗ –0.0891 –0.1119 (0.2205) (0.2287) –0.0067 –0.0100 (0.0256) (0.0265) –0.0450 (0.017)∗∗∗ –0.0218 (0.0113)∗

Other producer’s weather 0.0247 (0.0850) –0.0002 (0.0008) 0.2369 (0.1357)∗ –0.0307 (0.0192)

Average temperature

Constant

117

(4)

(5)

1.3664 (0.2591)∗∗∗ –0.0115 (0.0021)∗∗∗ –0.0898 (0.2157) –0.0081 (0.0250) –0.0499 (0.017)∗∗∗ –0.0255 (0.0122)∗∗

1.3143 (0.2568)∗∗∗ –0.0110 (0.002)∗∗∗ 0.1281 (0.2220) –0.0271 (0.0253) –0.0417 (0.0165)∗∗ –0.0216 (0.0117)∗ –0.1780 (0.062)∗∗∗

0.0982 (0.0907) –0.0009 (0.0009) 0.3501 (0.1557)∗∗ –0.0486 (0.0209)∗∗

0.0658 (0.0883) –0.0005 (0.0009) 0.3005 (0.1563)∗ –0.0428 (0.0207)∗∗

–0.0255 (0.0109)∗∗ –0.0241 (0.0069)∗∗∗

–0.0281 (0.0109)∗∗ –0.0254 (0.0072)∗∗∗ 0.0744 (0.0468)

–24.4308 (7.389)∗∗∗ 559 0.95

–24.2303 (7.2743)∗∗∗ 559 0.95

–34.0117 (11.3647)∗∗∗ 559 0.95

–41.4904 (11.3199)∗∗∗

–36.9132 (10.902)∗∗∗

559 0.95

559 0.95

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

shifts for precipitation, but not for temperature. In both the cotton and wheat markets, increases in local precipitation raised output at a diminishing rate and lowered price at a diminishing rate (see tables 4.4 and 4.8 for cotton and 4.7 and 4.11 for wheat). On the other hand, increases in local temperature increased both prices and quantities in both the cotton and wheat markets. As we will see in the following, the effects of local weather

118

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Table 4.9

Dependent variable ln(corn output) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

0.2142 (0.0701)∗∗∗ –0.0027 (0.0006)∗∗∗ 0.2934 (0.106)∗∗∗ –0.0396 (0.0134)∗∗∗

(2) Local weather 0.2137 (0.0689)∗∗∗ –0.0027 (0.0006)∗∗∗ 0.2881 (0.1106)∗∗∗ –0.0392 (0.0135)∗∗∗ 0.0001 (0.0089) –0.0012 (0.0059)

Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

No. of observations Adjusted R2

(4)

(5)

0.2203 (0.0714)∗∗∗ –0.0028 (0.0006)∗∗∗ 0.2877 (0.1073)∗∗∗ –0.0390 (0.0136)∗∗∗

0.2193 (0.0703)∗∗∗ –0.0027 (0.0006)∗∗∗ 0.2818 (0.1121)∗∗ –0.0385 (0.0138)∗∗∗ –0.0004 (0.0089) –0.0017 (0.0059)

0.2233 (0.0699)∗∗∗ –0.0028 (0.0006)∗∗∗ 0.3501 (0.1173)∗∗∗ –0.0442 (0.0141)∗∗∗ 0.0014 (0.0091) –0.0001 (0.0059) –0.0641 (0.0349)∗

0.2013 (0.0808)∗∗ –0.0020 (0.0008)∗∗

0.1895 (0.081)∗∗ –0.0019 (0.0008)∗∗

0.2369 (0.1357)∗ –0.0301 (0.0205)

0.3501 (0.1557)∗∗ –0.0278 (0.0228) 0.0041 (0.0105) 0.0020 (0.0073)

0.3005 (0.1563)∗ –0.0186 (0.0229) 0.0003 (0.0108) –0.0006 (0.0075) 0.0631 (0.0486)

2.9457 (2.5138) 1,806 0.96

3.0185 (2.5523) 1,806 0.96

2.9937 (2.5529) 1,806 0.96

Other producer’s weather 0.2039 (0.0786)∗∗∗ –0.0020 (0.0008)∗∗∗

Average temperature

Constant

(3)

6.8718 (1.9907)∗∗∗ 1,806 0.96

6.8787 (1.9706)∗∗∗ 1,806 0.96

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

on prices in the corn and wheat markets were weak relative to the effects in the corn and hay markets, which is consistent with a setting where local conditions had little effect on prices. Many of the same patterns arise when examining the impact of weather outside the state on the state’s prices and quantities. The corn market results for the impact of weather elsewhere on local price in table 4.5 and the general effect of weather on output in table 4.9 display the exact same pattern as for

The Effects of Weather Shocks on Crop Prices in Unfettered Markets Table 4.10

Dependent variable ln(total hay output) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

–0.0712 (0.0325)∗∗ 0.0005 (0.0003) 0.2478 (0.0557)∗∗∗ –0.0205 (0.0076)∗∗∗

Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

No. of observations Adjusted R2

(2) Local weather –0.0827 (0.0321)∗∗ 0.0006 (0.0003)∗ 0.2295 (0.05538)∗∗∗ –0.0187 (0.0074)∗∗ –0.0064 (0.0035)∗ –0.0111 (0.0038)∗∗∗

(3)

(4)

(5)

–0.0638 (0.0355)∗ 0.0004 (0.0004) 0.2476 (0.0554)∗∗∗ –0.0199 (0.0076)∗∗∗

–0.0772 (0.0352)∗∗ 0.0006 (0.0004) 0.2278 (0.0552)∗∗∗ –0.0180 (0.0074)∗∗ –0.0063 (0.0035)∗ –0.0109 (0.0038)∗∗∗

–0.0734 (0.0353)∗∗ 0.0005 (0.0004) 0.3126 (0.058)∗∗∗ –0.0257 (0.0074)∗∗∗ –0.0045 (0.0034) –0.0085 (0.0037)∗∗ –0.0751 (0.0181)∗∗∗

–0.1397 (0.1018) 0.0014 (0.0010) –0.1984 (0.1733) 0.0291 (0.0250) 0.0138 (0.0108) 0.0015 (0.0088)

–0.1161 (0.1000) 0.0011 (0.0010) –0.1585 (0.1953) 0.0243 (0.0279) 0.0158 (0.0115) 0.0040 (0.0095) –0.0458 (0.0497)

8.7924 (1.0656)∗∗∗ 1,152 0.98

8.4668 (1.0702)∗∗∗ 1,152 0.98

Other producer’s weather –0.1230 (0.0965) 0.0012 (0.0009) –0.1582 (0.1600) 0.0248 (0.0236)

Average temperature

Constant

119

7.9382 (0.8018)∗∗∗ 1,152 0.98

8.2315 (0.7996)∗∗∗ 1,152 0.98

8.3900 (1.0474)∗∗∗ 1,152 0.98

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

the impact of local weather. Increases in temperature raised corn output at a diminishing rate in general, and the temperature rise elsewhere was associated with declines in farm- gate corn prices at a diminishing rate. Increases in precipitation show the same patterns. Similarly, the hay market results for the impact of weather elsewhere on farm- gate prices in table 4.6 and the general effect of weather on output in table 4.10 display the exact same pattern as for the impact of local weather. More precipitation raised hay

120

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Table 4.11

Dependent variable: ln(total wheat output) (1)

Average temperature Average temperature squared Average precipitation Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation

0.0233 (0.0818) –0.0011 (0.0008) 0.3528 (0.144)∗∗ –0.0545 (0.0216)∗∗

Average temperature Average temperature squared Average precipitation

(2)

(3)

(4)

(5)

–0.0049 (0.0819) –0.0008 (0.0008) 0.3667 (0.1451)∗∗ –0.0561 (0.0216)∗∗∗

0.0158 (0.0820) –0.0010 (0.0008) 0.2958 (0.1546)∗ –0.0504 (0.0219)∗∗ 0.0189 (0.0087)∗∗ 0.0044 (0.0082)

0.0058 (0.0819) –0.0009 (0.0008) 0.3186 (0.1565)∗∗ –0.0520 (0.0218)∗∗ 0.0211 (0.00901)∗∗ 0.0063 (0.0082) –0.0308 (0.0427)

Other producer’s weather –0.2739 (0.1389)∗∗ 0.0029 (0.0013)∗∗

–0.3457 (0.13595)∗∗ 0.0038 (0.0013)∗∗∗

–0.2953 (0.1401)∗∗

–0.1582 (0.1600) 0.0732 (0.0339)∗∗

–0.1984 (0.1733) 0.1521 (0.0381)∗∗∗

0.0033 (0.0013)∗∗ –0.1585 (0.1953) 0.1175 (0.0417)∗∗∗

0.1023 (0.0188)∗∗∗ 0.0412 (0.0128)∗∗∗

0.1206 (0.0195)∗∗∗ 0.0552 (0.0144)∗∗∗

Local weather 0.0400 (0.0816) –0.0012 (0.0008) 0.2771 (0.1575)∗ –0.0485 (0.02248)∗∗ 0.0172 (0.0087)∗∗ –0.0001 (0.0084)

Average precipitation squared Months of extreme wetness Months of extreme drought Palmer Z Index standard deviation Constant No. of observations Adjusted R2

–0.3052 (0.0749)∗∗∗ 7.1252 (2.19021)∗∗∗ 1,595 0.93

6.8021 (2.1839)∗∗∗ 1,595 0.93

7.7913 (2.1939)∗∗∗ 1,595 0.93

7.5424 (2.1957)∗∗∗ 1,595 0.94

7.9058 (2.2029)∗∗∗ 1,595 0.94

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

output at a diminishing rate in general, and more precipitation elsewhere lowered hay prices at a diminishing rate; higher temperatures lowered hay output at a diminishing rate, and higher temperatures elsewhere raised hay prices at a diminishing rate. The markets for cotton, described in tables 4.4 and 4.8, and for wheat, described in tables 4.7 and 4.11, again have mixed effects of weather elsewhere on output and farm- gate prices. In both the cotton and wheat mar-

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

121

kets, a rise in temperature lowered output at a diminishing rate while raising prices at diminishing rate. On the other hand, demand shifts seemed to have been more dominant with respect to changes in precipitation elsewhere. In the cotton market, increases in precipitation raised output at a diminishing rate in general, but increases in precipitation elsewhere also raised the state farm- gate price at a diminishing rate. In the wheat market, increases in precipitation lowered output at a diminishing rate, while increases in precipitation elsewhere also raised the state price. 4.6.2

Cotton Prices and Weather: Tables 4.4 and 4.8 and Figure 4.4

Cotton output was sensitive to fluctuations in temperature, extreme or severe wetness, and changes in weather variability, as shown by the coefficients of the temperature and extreme or severe wetness variables in table 4.8. Despite this sensitivity of output, state cotton prices barely responded to local weather changes. Increases in local weather variability tended to slightly decrease the state farm- gate price of cotton, but state prices were most sensitive to changes in precipitation and drought conditions in other producing states. Figure 4.4 plots price response functions that show the percentage change in the state price associated with an increase of 1°F in local state temperature (panel A) and in temperature elsewhere (panel B). Panel C of figure 4.4 shows the percentage change in the state farm- gate price in response to an increase of one inch of rainfall in that state’s precipitation. Panel D of figure 4.4 shows the percentage change in the state farm- gate price in response to an additional inch of average precipitation experienced by producers in the rest of the United States. These estimates are derived from the coefficients in column (5) of table 4.4. We plot the relationships because the inclusion of both linear and squared terms for temperature in the log price equations causes the relationship between temperature and the price to change as the temperature rises. The same linear and squared terms are used in the precipitation measures. In general, the results suggest that state cotton prices were not very sensitive to fluctuations in local weather. None of the local temperature or precipitation coefficients in table 4.4 are statistically significant, and the price response functions are much flatter and closer to 0 percent change than for any other crop. As seen in panel A of figure 4.4, the percentage change in cotton prices associated with an increase in local state temperature decreases slightly as temperature increases. Until about 60°F, a 1° rise in annual average temperature is associated with very little change in price. As the temperature approaches the mid- 70s°F, a 1° rise in local temperature is associated with roughly a 1 percent drop in state cotton prices. The other price response functions for cotton in panels B to D of figure 4.4 are flatter and even closer to zero than for local temperatures. The only statistically significant relationship between weather and prices is for precipitation in areas outside the state, but the plotted relationship in panel D of figure 4.3

A

B

C

D Fig. 4.4 Percent change in cotton price: A, From percent change in local average temperature; B, From percent change in other producers’ (OP) average temperature; C, From percent change in local average precipitation; D, From percent change in OP average precipitation

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

123

shows very weak responsiveness even at the upper ranges of precipitation. The cotton price responses to precipitation and temperature changes are much weaker than those for corn and hay described in the following. 4.6.3

Corn Prices and Weather: Tables 4.5 and 4.9 and Figure 4.5

Corn output at the state level responded to increases in temperature and precipitation in roughly the same way. The coefficients in table 4.9 show that increases in average annual temperature led to increases in corn production at a diminishing rate, as did increases in precipitation. The local increases in corn output are associated with increases in temperature and precipitation, which themselves are also associated with reductions in local corn prices as shown in table 4.7. Similarly, increases in temperature and precipitation that likely would have increased output in other states also contributed to lower corn prices in the state of interest. Our findings for the era before the powerful influences of the federal farm programs are similar in that regard to Mundlak and Larson’s (1992) findings that international markets played an important role in determining local prices. Comparisons of panels A of figures 4.4 and 4.5 show that state farm- gate corn prices were far more responsive to local temperatures than were cotton prices. Once the average annual temperature exceeded 40°F, corn prices started rising in response to increases in temperature, and the responsiveness rose significantly from there. Similarly, the state corn price response function for temperature changes occurring in the rest of the country (figure 4.5, panel B) had a much stronger positive slope than state cotton price response in figure 4.4, panel B. However, comparisons of panels C of figures 4.4 and 4.5 and panels D of figures 4.4 and 4.5 show that state precipitation and precipitation elsewhere had much stronger impacts at higher levels of precipitation on corn prices than on cotton prices. 4.6.4

Hay Prices and Weather: Tables 4.6 and 4.10 and Figure 4.6

Hay, like corn, was sensitive to many of the different weather variables, both local and nonlocal. Local weather variables that had statistically significant coefficients included the temperature and precipitation variables, as well as the variables that proxy for extreme or severe drought and wetness conditions. The price response function to local temperature was negatively sloped for hay in panel A of figure 4.6. After about 54°F, as state- level temperatures rose, prices fell. At levels below that, a 1 percent rise in average yearly temperature was associated with up to a 1 percent rise in the farm- gate price. In contrast, the relationship between hay prices and local precipitation in panel C of figure 4.6 had a mild U shape. Local monthly precipitation had little impact on state hay prices until it reached the upper end of the range. When precipitation approached seven inches per month, an additional inch rise in precipitation led to more than a 1 percent rise in that state’s hay price. This is nearly double the effect seen for corn in panel C of figure 4.5.

A

B

C

D Fig. 4.5 Percent change in corn price: A, From percent change in local average temperature; B, From percent change in other producers’ (OP) average temperature; C, From percent change in local average precipitation; D, From percent change in OP average precipitation

A

B

C

D Fig. 4.6 Percent change in hay price: A, From percent change in local average temperature; B, From percent change in other producers’ (OP) average temperature; C, From percent change in local average precipitation; D, From percent change in OP average precipitation

126

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

The price response function in panel B of figure 4.6 for temperatures in the rest of the country shows that a one unit increase in temperature elsewhere contributed to about a 1 percent decrease in the state farm- gate price at every temperature level. However, none of the temperature coefficients were statistically significant, so there may have been no effect. The precipitation coefficients were statistically significant, and the path of the price response to precipitation outside the state in panel D of figure 4.6 is nearly a direct contrast to the response path to local precipitation in panel C of figure 4.6. The response function for out- of- state precipitation is hump- shaped, and as the precipitation approaches an average of seven inches per month elsewhere, a one unit increase leads to 2 percent reduction in hay prices in the state. This response is nearly twice as large in a negative direction in comparison to the positive response to increased local precipitation. In general, the state hay price responses to higher levels of precipitation either locally or elsewhere are much larger in magnitude than for any other crop. 4.6.5

Wheat Prices and Weather: Tables 4.7 and 4.11 and Figure 4.7

Wheat is sold in an international market and, as might be expected, the responses of state’s prices to local weather shocks were muted for both local temperature and precipitation. The response functions for both types of weather in panels A and C of figure 4.7 are much flatter and closer to the origin throughout the range than for hay in panels A and C of figure 4.6 and corn in panels A and C of figure 4.5. They more closely resemble the responses seen for cotton, the other strongly international crop, in panels A and C of figure 4.4. The slight sensitivity to local weather fluctuations is likely due to wheat being grown in many of the different states, even though it is primarily concentrated in the Dakotas and Kansas. If a local weather shock did not affect the local price too much, it would still make sense to purchase locally grown wheat at a slightly higher price. However, if local supply was hit hard, then it would not be too difficult to import from the outside market. Wheat prices were very sensitive to temperatures in other parts of the country. The wheat price response function to temperatures elsewhere in panel B of figure 4.7 looks very similar to the corn price response function in panel B of figure 4.5. A 1° rise in average temperature elsewhere as temperatures elsewhere were near 33°F led to a reduction in state corn prices of nearly 2 percent. At higher temperatures, there was a much stronger response in the other direction. A rise in temperature elsewhere by 1° as temperatures elsewhere neared the high end around 70°F led to an increase of state corn prices of nearly 5 percent. The effects of precipitation elsewhere on wheat prices were also statistically significant, as seen in panel D of figure 4.6. The response function to precipitation elsewhere looks similar to the one for hay, but the negative effect of more precipitation elsewhere at high levels of precipitation is about half the size of the one for hay.

A

B

C

D Fig. 4.7 Percent change in wheat price: A, From percent change in local average temperature; B, From percent change in other producers’ (OP) average temperature; C, From percent change in local average precipitation; D, From percent change in OP average precipitation

128

4.7

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Concluding Remarks

The study of the impact of weather on crop prices in unfettered markets has become increasingly important for two reasons. First, one of the major worries associated with climate change relates to increased fluctuations in weather, which, in turn, will influence food supplies and food prices. Weather shocks that lead to reductions in output and rising food prices can have major negative effects on health as people shift their consumption to lower- priced foods, often with less nutritional quality.12 Second, in the World Trade Organization negotiations, less- developed countries have been pressuring the developed nations to change their farm policies to stop interfering with markets and to stop propping up farms in the developed countries. In evaluating these changes, therefore, it is important to examine the way that unfettered crop markets work. Studies of the United States since the 1930s cannot illuminate much about the operation of unfettered markets because of the extensive farm programs in place; therefore, we need to look at the preceding period. We estimated the responsiveness of crop prices to both localized weather shocks, as well as weighted measures of the weather shocks experienced by other producers of each crop. Four crops, cotton, corn, hay and wheat, were chosen for study not only because they were fundamental to the U.S. economy, but also because their inherent characteristics differed in an important way. Both cotton and wheat have relatively high value- to- weight ratios and between 1895 and 1932 were sold in a true international market. For these crops, localized weather shocks might have affected the size of the harvest within a state, but the effects on state commodity prices should have been limited. Over the thirty- seven- year period studied, that is what we find. Corn and hay represent the flip side of that coin. Corn and hay have lower value- to- weight ratios than wheat and cotton and, thus, higher transport costs. While cotton and wheat were not generally consumed locally, most of the value from corn and hay came from local uses and local agronomic activity. For corn and hay, localized weather shocks would have been expected to influence the prices in a state. Indeed, the results of the analysis show that hay and corn prices were substantially more responsive to local weather shocks than were cotton and wheat prices. We identified the differential weather effects using the year- to- year variation in weather and commodity prices after controlling for time- invariant features of each state and controlling for national shocks such as warfare that would have influenced the markets. As a result, the analysis avoids much 12. See Komlos (1987); Haines, Craig, and Weiss (2003); Fishback et al. (2010); and Steckel (1992). As another example, Galloway (1985) used annual data for London from 1670 to 1830 to show how bad weather and poor harvests raised both agricultural prices and mortality.

The Effects of Weather Shocks on Crop Prices in Unfettered Markets

129

of the unobserved variable bias that seems to plague cross- sectional models. Furthermore, we focused on the period between 1895 and 1932 because there were not only substantial fluctuations in weather that influenced the yields of different commodities, but also because there was much less government interference in markets to protect farmers from falling prices. Despite the absence of price supports, state- level cotton and wheat prices were not affected much by the weather shocks within the state. The nationwide weather shocks and decline in prices that followed World War I helped usher in support for the federal farm programs of the New Deal. Several of these programs, such as the 1938 Federal Crop Insurance Act, were intended to address the income and production variability that the weather shocks induced. Others such as the Soil Conservation Act sought to remedy the environmental damage associated with the Dust Bowl. These federal farm programs persisted and expanded over the next eighty years and strongly influenced the ways that farm prices at the farm- gate responded to weather shocks within states and across the nation. Our next move is to investigate how the relationships between weather and prices changed as a result of these farm policies. Many of the policies were designed to diminish the downward volatility in farm gate prices and control the sale of the crop within the state and in outside markets. Such controls potentially reduced income volatility for farmers, although given the fluctuations in prices in response to local shocks for some crops in the unfettered markets, perhaps not as much as might have been thought. On the other hand, the programs may well have led to higher crop prices in the long run, with the consequent impact on health, within the United States. Understanding these trade- offs will contribute to improvements in the quality of the policies chosen in the future.

References Cooley, Thomas F., Stephen J. DeCanio, and M. Scott Matthews. 1977. An agricultural time series- cross section data set. NBER Working Paper no. W0197. Cambridge, MA: National Bureau of Economic Research. Deschênes, Olivier, and Michael Greenstone. 2007. The economic impacts of climate change: Evidence from agricultural profits and random fluctuations in weather. American Economic Review 97:354–85. Fishback, Price V., Werner Troesken, Trevor Kollman, Michael R. Haines, Paul W. Rhode, and Melissa Thomasson. 2010. Information and the impact of climate and weather on mortality rates during the Great Depression. University of Arizona, Working Paper. Galloway, P. R. 1985. Annual variations in deaths by age, deaths by cause, prices, and weather in London, 1670–1839. Population Studies 39:487–505. Haines, Michael R., Lee A. Craig, and Thomas Weiss. 2003. The short and the dead:

130

Jonathan F. Fox, Price V. Fishback, and Paul W. Rhode

Nutrition, mortality, and the “antebellum puzzle” in the United States. Journal of Economic History 63:382–413. Hamilton, David E. 1982. Herbert Hoover and the Great Drought of 1930. Journal of American History 68 (4): 850–75. Hannay, Annie M. 1931. The influence of weather on crops, 1900–1930: A selected and annotated bibliography. USDA Miscellaneous Publication no. 118. Washington, DC: U.S. Department of Agriculture. Karl, Thomas R., Claude Williams, Pamela Young, and Wayne Wendland. 1986. A model to estimate the time of observation bias associated with monthly mean maximum, minimum and mean temperatures for the United States. Journal of Climate and Applied Meteorology 25:145–60. Komlos, John. 1987. The height and weight of West Point Cadets: Dietary change in antebellum America. Journal of Economic History 47:897–927. Kramer, Randall A. 1983. Federal crop insurance, 1938–1982. Agricultural History 57 (2): 181–200. Lange, Fabian, Alan L. Olmstead, and Paul W. Rhode. 2009. The impact of the boll weevil, 1892–1932. Journal of Economic History 69 (3): 685–718. Maddala, G. S., and Shaowen Wu. 1999. A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and Statistics 61 (S1): 631–52. Mundlak, Yair, and Donald F. Larson. 1992. On the transmission of world agricultural prices. World Bank Economic Review 6 (3): 399–422. National Climatic Data Center. 2002. Data set 9640: Time Bias Corrected Divisional Temperature-Precipitation-Drought Index. Asheville, NC: National Climatic Data Center. Officer, Lawrence H. 2008. The annual Consumer Price Index for the United States, 1774–2007. http://measuringworth.com. Steckel, Richard. 1992. Stature and living standards in the United States. In American economic growth and standards of living before the Civil War, ed. Robert Gallman and John Wallis, 265–308. Chicago: University of Chicago Press. Thorp, Willard Long. 1926. Business annals. New York: National Bureau of Economic Research. Trotter, Paul S., G. Alan Johnson, Robert Ricks, and David R. Smith. 1998. Floods on the lower Mississippi: An historical economic overview. National Weather Service Forecast Center, National Oceanic and Atmospheric Administration. http:// www.srh.noaa.gov/topics/attach/html/ssd98- 9.htm. U.S. Department of Agriculture (USDA). 1923. Yearbook 1922. Washington, DC: USDA.

5 Information and the Impact of Climate and Weather on Mortality Rates during the Great Depression Price V. Fishback, Werner Troesken, Trevor Kollmann, Michael Haines, Paul W. Rhode, and Melissa Thomasson

Global warming has become a watchword for environmental policy over the past three decades. Daily temperature highs were thought to have reached the highest levels in recorded history within the past decade. Each month, there are reports of new studies of melting glaciers, thinning of ice caps on mountains, and warming in various areas throughout the world. Al Gore shared an Academy Award for his association with the movie An Inconvenient Truth, a film warning of global warming and its potential dire consequences. He then shared a Nobel Peace Prize with a group of scientists warning of the dangers of global warming. Much of the force of Gore’s warnings about global warming comes from his predictions about the impact of warming on human populations and the economy. Yet the large volume of studies of climate change has not been matched by nearly as many studies of the impact of climate and weather on populations and economies or how populations and economies will respond. If the claims Price V. Fishback is the Thomas R. Brown Professor of Economics at the University of Arizona, and a research associate of the National Bureau of Economic Research. Werner Troesken is a professor of economics at the University of Pittsburgh, and a research associate of the National Bureau of Economic Research. Trevor Kollmann is a PhD candidate in economics at the University of Arizona. Michael Haines is the Banfi Vintners Professor of Economics at Colgate University, and a research associate of the National Bureau of Economic Research. Paul W. Rhode is a professor of economics at the University of Michigan, and a research associate of the National Bureau of Economic Research. Melissa Thomasson is an associate professor of economics at Miami University in Ohio, and a research associate of the National Bureau of Economic Research. We would like to thank Hoyt Bleakley, Olivier Deschênes, Michael Greenstone, Sok Chul Hong, Shawn Kantor, Gary Libecap, Robert Margo, Rick Steckel, James Stock, participants at the National Bureau of Economic Research (NBER) Universities Conference on Climate Change in May 2008, and participants at the NBER Conference on Climate and History in May 2009 for their helpful comments.

131

132

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

that global temperatures will warm over the next few decades no matter what policy steps we take today are true, such studies are invaluable. Here is a situation where history can serve as a guide to the impact of climate and weather on populations. We measure the impact of climate and weather fluctuations on infant mortality and noninfant mortality in United States counties throughout the Great Depression of the 1930s. The Great Depression was a period of great climate stress. It is arguably one of the two hottest decades in the 130 years in which the time- of- day adjusted temperature records have been readily available throughout the United States.1 The heat created problems with droughts and Dust Bowls that contributed to the economic problems of the era as well as long- run responses to adapt to climate extremes.2 Second, the Great Depression was a period of great economic vulnerability when climate might have had more impact on death rates. Unemployment rates were higher than 9 percent in every year between 1930 and 1940, over 14 percent in nine of those years, and exceeded 20 percent in the four years from 1932 through 1935. Annual real gross domestic product (GDP) in America was roughly 30 percent below its 1929 peak in both 1932 and 1933 and did not reach the 1929 level again until 1937.3 We have developed a database that combines information on infant and noninfant mortality rates, daily high temperatures and inches of precipitation, and a rich set of socioeconomic correlates for over 3,000 counties in the United States for each year between 1930 and 1940. We focus on infant mortality because infant mortality has long been seen as a key nonincome measure of standards of living, the death of an infant is an extraordinarily painful event, and infants are likely the most sensitive of populations to variations in conditions. We also examine the noninfant death rate to see if the patterns seen for infant deaths carry over to death rates for people in all age groups. The results of the Great Depression analysis show the importance of controlling for access to information when measuring the relationship between mortality and climate. In analyses that do not control for measures of access to information, there is a strong positive relationship between mortality and temperature. When measures of illiteracy, access to radios, and access to magazines are incorporated in the analysis, the strong positive relationship 1. Steve McIntyre of www.climateaudit.org discovered an anomaly in the temperature data circa 1999 to 2000 that caused the National Aeronautics and Space Administration (NASA) to readjust its temperature rankings. In the United States, 1934 ranks slightly above 1998 as the hottest year on record. The years 1931, 1938, and 1939 also rank in the top ten. See http://www .climateaudit.org/?p1880 and http://data.giss.nasa.gov/gistemp/graphs/Fig.D.txt. 2. See Hansen and Libecap (2003) and Cunfer (2005). 3. The 1930s also offer better data than earlier decades. It is the first decade in which infant mortality data were collected on a consistent basis for all states, and it is the first decade in which a large number of weather stations consistently reported daily information on high and low temperatures.

The Impact of Climate on Mortality Rates during the Great Depression

133

between mortality and temperature is no longer present. Researchers on the impact of climate, therefore, need to be mindful of the potential for such omitted variable bias when drawing conclusions about the impact of climate on various socioeconomic measures. 5.1

The Pathways between Climate, Weather, and Mortality

There is a long history of research linking climate to disease and mortality. Carl Spinzig (1880) developed an elaborate meteorological model designed to forecast yellow fever epidemics in American cities. Similarly, in his monumental History of Epidemics in Britain, Charles Creighton ([1894] 1965) argued that a wide range of diseases, including typhus, plague, pneumonia, influenza, and infantile diarrhea had seasonal or climatic components.4 Leonard Rogers (1923, 1925, 1926) sought to forecast the likelihood of epidemics in India using climate variables. “Based on his conclusions, it was recommended that climatic variables be used for forecasting epidemics of TB, smallpox, and pneumonia and for mapping worldwide incidence of leprosy. However, such systems were never implemented on a wide scale” (World Health Organization [WHO] 2004, 12). More recently, Olivier Deschênes and Michael Greenstone (2007) have begun exploring the impact of fluctuations of weather on mortality. In this chapter, we will focus on the influence on mortality of climate and weather fluctuations that can be effectively evaluated using county data during the Great Depression in America.5 In discussing the role of climate and weather, the distinctions between the two are somewhat fluid. Climate is often defined by long- term weather patterns, while some people define weather as short- term deviations from the long- run patterns. A shift in climate occurs when what had been deviant weather patterns last for an extended period of time. When we translate these definitions for the empirical work in the chapter, the impact of climate will be analyzed when 4. Sadly, history often remembers Creighton for his ludicrous opposition to smallpox vaccination, but this in no way undermines the significance of his exhaustive and scholarly two volume history. 5. Weather extremes that damage crops can generate increases in food prices that can lead to famine in autarkic and subsistence economies. We do not focus on that mechanism much in this chapter because we are using county- level data in the United States in a period where the markets extended beyond county boundaries and often beyond state and national boundaries. Thus, the effect of local weather on food prices was not as strong. For further evidence on this issue, see Fox, Fishback, and Rhode (chapter 4 in this volume) on the impact of weather fluctuations on state- level prices of corn and hay. In more- developed economies, increased food prices can induce consumers to switch to cheaper, low- quality foods. The “antebellum puzzle” prior to 1860 offers a prime example. Despite rising per capita incomes, mortality rates rose and access to nutrition declined as increases in food prices, especially for meat, encouraged American consumers to switch away from high- protein meat products to lower quality foods. See Komlos (1987); Haines, Craig, and Weiss (2003); and Steckel (1992). As another example, Galloway (1985) used annual data for London from 1670 to 1830 to show how bad weather and poor harvests conspired to raise both agricultural prices and mortality.

134

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

both cross- sectional and time series variation are used as the sources of identification of the relationship between temperature and precipitation and mortality. Weather fluctuations will be addressed when the analysis shifts to the use of differencing to control for time- invariant features of the counties and, thus, the source of identification is variation across time in the same county. The two key components of climate and weather examined in the analysis are temperature and precipitation. 5.1.1

Temperature

There are a variety of ways in which temperature and precipitation might influence mortality. The most obvious relate to exposures to extreme heat or cold. For example, New York City was struck by an intense heat wave in August 1896. The New York Times reported that the severe heat led to 500 early deaths and many more instances of heat prostration. Out- of- town newspapers put the numbers afflicted in New York City even higher. Due to the oppressive temperatures, many of the city’s working horses dropped dead in the street; in the age before automobiles, the carcasses could not be moved without putting other horses at risk.6 Local charities and governments responded to such extreme temperature events by providing relief in various forms (free ice, fuel, or access to protected space). And indeed, during the Great Depression period under study, record cold temperatures in the winter of 1933 to 1934 induced New Deal authorities to extend work relief programs that were set to terminate. Fluctuations in temperature were identified by public health officials as contributors to mortality in many other ways during the nineteenth and early twentieth century. These officials argued that infant mortality spiked upward during July, August, and September because the warm weather was conducive to the proliferation and spread of bacteria in milk and water. Milk samples in Washington, DC in the summers of 1906 and 1907, for example, contained average counts of 11 to 22 million bacteria per cubic centimeter, two to four times the level found in sewage from major American and European cities at the time (Rosenau 1909). Such food or waterborne pathogens were considered to be likely suspects because most infant deaths during the summer were from diarrheal diseases, and there was no summer spike in mortality for infants who were breastfed and, therefore, not exposed to bacteria in water and cow’s milk.7 Water- related diseases compounded the problem of milk- related diar6. See New York Times (6 August 1896, 1; 7 August 1896, 5; 8 August 1896, 5; 9 August 1896, 1; 10 August 1896, 1; 11 August 1896, 1–2; 12 August 1896, 1; 13 August 1896, 4; 16 August 1896, 8); Chicago Tribune (13 August 1896, 5; 14 August 1896, 4; 15 August 1896, 1); Washington Post (9 August 1896, 1; 12 August 1896, 1); and Los Angeles Times (12 August 1896, 3). 7. The literature on the summertime spike in infant mortality is voluminous. For a few representative examples, see Phelps (1910); Eghian (1905); Lancet (November 15, 1884, 882); Sedgwick and MacNutt (1910); and Routh (1879, 35–42). On the viability of bacteria in warm milk, see Science (August 16, 1889, 116–18).

The Impact of Climate on Mortality Rates during the Great Depression

135

rheal infections because parents and vendors often used water to dilute the milk. Typhoid, the most serious waterborne disease in the United States at the turn of the century, also peaked during the late summer and early fall, although the mechanisms that drove this spike are far from clear (Whipple 1908, 123–27). Surprisingly, experiments from this era repeatedly showed that typhoid bacteria in water were more common and more vital during the winter months than the summer. Direct sunlight, more common in the summer, also inhibited the growth of waterborne bacteria (Journal of the American Medical Association, March 16, 1895, 415). Given the relative vitality of typhoid bacteria during colder months, we can only speculate that typhoid peaked during the summer because people drank more water in the summer heat or because there was some unidentified interaction between tainted water and the broader environment. The warm, more tropical weather of summer led to rapid multiplication of the number of insects, which represent another possible vector through which climate change could affect disease rates and overall mortality. The summer proliferation of flies creates a serious public health risk whenever populations used privies and cesspools to dispose of human waste. The flies interacted with excreta and waste and then contributed to the spread of pathogens associated with typhoid fever and other diarrheal diseases. The pervasiveness of flies led public health officials to emphasize the importance of public sewer systems and well- screened privies in forestalling the transmission of typhoid and diarrhea (Whipple 1908, 123–27; Bergey 1907; Hewitt 1912). While not so relevant for the United States, tsetse flies are also carriers of sleeping sickness in central Africa (Hewitt 1912). Mosquitoes, too, might have been important carriers of disease in early twentieth century America. Although yellow fever, malaria, dengue, and other mosquito- related illnesses were not as common in the United States as they were in Africa and parts of Asia, the available data suggest malaria was not uncommon in the American South and represented a serious public health threat during the nineteenth and early twentieth centuries (Herrick 1903; Humphreys 2003). In Mississippi, malaria was the seventh- leading cause of death in the state in 1900. In a handful of cities such as Paducah (Kentucky), Jacksonville (Florida), Savannah (Georgia), and Wilmington (North Carolina), the death rate from malaria was between 100 and 200 deaths per 100,000 persons, rivaling the death rates from pneumonia, influenza, and typhoid fever (U.S. Bureau of the Census 1908, 34–35). Studying the early twentieth- century United States, Brazil, Colombia, and Mexico, Bleakley (2007) shows that mosquito eradication raised labor productivity significantly.8 8. There is evidence to suggest that the extent of malaria in the United States during this period was overestimated. Malaria cases were frequently misdiagnosed cases of typhoid fever, particularly among African Americans. Typhoid and malaria shared common symptoms and

136

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

Even as typhoid, diarrheal diseases, and insect- borne diseases spiked during the summer months, respiratory diseases spiked during the winter months. Pneumonia, influenza, tuberculosis, bronchitis, and, to a lesser extent, diphtheria, all rose sharply when the temperature fell (Clemow 1903, 14–21). The connection between cold temperatures and respiratory diseases was well- documented and understood before the development of the germ theory of disease. In 1864, the Massachusetts Board of Health and Birth and Death Registry (1866, 59) gave examples of the well- known pattern of winter peaks in deaths from pneumonia: “The greatest number of deaths (281) was in March, and the least (42) in August. More than half of the deaths (53.8) occurred during the first four months [of the year], and only 15.33 per cent from June to October, inclusive; showing the well development of this disease in the cold season.” Similarly, monthly data from the City of Chicago between 1871 and 1906 in figure 5.1 show a strong negative relationship between the monthly temperature and the pneumonia death rate.9 There are at least three reasons to expect respiratory diseases to be more common during the winter months. Historical observers emphasized that cold weather caused people to spend more time indoors, where respiratory diseases were more easily spread in crowded and poorly ventilated homes. Some bacteria and viruses grow and reproduce more rapidly in cooler temperatures than in warm ones or simply find cooler temperatures more amenable.10 Viruses, for example, become more stable at lower temperatures (Zinsser et al. 1980, 157). Respiratory viruses are also inhibited by summer heat and solar radiation, and recent work suggests that in temperate climates viral activity is greatest during the winter months (Yusuf et al. 2007; Sagripanti and Lytle 2007). Moreover, environmental forces such as cold weather and humidity are more important than factors such as population density and migration in the propagation of the influenza virus (Alsono et al. 2007; were routinely conflated by physicians under the misleading name “typho- malaria” fever. In compiling mortality statistics for the country during the early 1900s, the United States Census Bureau (1908, 34–35) wrote: “Death rates from malarial fever are usually of little importance, and may be subject to possible correction for inclusion of deaths actually due to typhoid fever, a disease which is frequently confused in the returns with malarial fever.” Most telling, when American cities began filtering water supplies—which should have affected typhoid rates but not malaria because filtering water did not kill mosquitoes—malaria rates fell sharply (Troesken 2004, 170–78). The upshot of this discussion is that high temperatures and excessive rainfall might affect malaria rates in the United States, but given the questionable prevalence, malaria might not prove to be an important source of variation in overall death rates. 9. The plot reveals a strong statistical correlation with an R2 of 0.32 for the regression line in the chart with a coefficient on temperature of –0.23, which is statistically significant at the .0001 level. 10. For microbial activity in general, the available evidence suggests most microbes become less active or dormant during the winter months. See Jones and Cookson (1983). This, however, does not rule out the possibility that some subset of microbes become more active during the winter. Recent research, for example, indicates that listeria can reproduce and multiply even at low temperature levels (Chan and Weidmann 2009).

The Impact of Climate on Mortality Rates during the Great Depression

137

Fig. 5.1 Pneumonia deaths per 100,000 people plotted against monthly temperature, city of Chicago, 1871–1906 Source: City of Chicago (various years between 1871 and 1906).

Reichert et al. 2004). Finally, during the winter months, people were exposed to more pollution. Before the widespread adoption of gas heating, emissions from coal, oil, fires, and stoves rose in winter as people heated their homes. In analyzing the long- run time series relationships between temperature and mortality within the year, it is important to identify whether the relationship is determined by the long- term month- to- month variation in temperature across the year or fluctuations in temperature around the longterm norms. To illustrate the differences in effects, we use ordinary least squares (OLS) to estimate the relationship between the pneumonia death rate and temperature in the data for Chicago from 1871 to 1906 in figure 5.1. In the analysis, we control for the long- term trend using a year counter and perform the estimation without and with month fixed effects. Without month fixed effects, the results in table 5.1 show a very strong and positive relationship between temperature and pneumonia death rates, as an additional degree Fahrenheit of temperature was associated with an additional pneumonia 0.687 deaths per 100,000 people. This relationship might have been driven by long- run relationships between temperatures across months of the year—July is much hotter than January, for example—or by fluctuations in temperature around the typical temperatures seen at particular times of year. In the second regression in table 5.1, we include month fixed effects to

138 Table 5.1

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson Ordinary least squares regression results with and without month fixed effects for monthly data on pneumonia deaths per 100,000 people as a function of temperature in Chicago, 1871–1906 Coefficient (t-statistics) Constant Temperature Year trend

2,435.9 (12.16) 0.687 (11.14) –1.283 (–12.10)

Month fixed effects February

–1.740 (–0.42) 2.650 (0.59) 0.582 (0.10) –3.898 (–0.56) 0.228 (0.03) 55.670 (6.11) 33.401 (3.72) 6.395 (0.80) –7.991 (–1.24) –11.860 (–2.44) –5.388 (–1.27)

March April May June July August September October November December No. of observations R2

2,461.8 (16.12) 0.079 (0.47) –1.284 (–15.89)

432 0.387

432 0.653

Source: Data collected from the city of Chicago (various years between 1871 and 1906). Note: For tables 5.1 through 5.6, coefficients with t-statistics listed in parentheses.

control for long- run differences across months that do not vary from year to year. The results show that time- invariant features of the month of July were associated with spikes of 55.67 in the number of pneumonia deaths per 100 thousand people, relative to January. The spike for August was 33.4 and death rates were 11.86 lower in November than in January. After controlling for these time- invariant features of each month, the relationship between pneumonia deaths and temperature was cut sharply from 0.687 to 0.079 and is no longer statistically significant. This second set of results suggest that the long- run unchanging differences in conditions between July, August, and

The Impact of Climate on Mortality Rates during the Great Depression

139

November are the key factors influencing differences in pneumonia death rates over the course of the year. The long- run differences might well be related to the long- run core differences in temperature between each month. If this is the case, the much lower impact of temperature in the regression with month fixed effects shows that fluctuations in temperature around the long- run core temperatures in a month have only a weak influence on the pneumonia death rate. This finding foreshadows one of the findings when we examine the county panel data for the entire United States. It appears that long- run differences in temperature conditions across the country influence mortality rates. After we control for those long- run conditions, however, short- term fluctuations in weather around those long- run differences have much smaller impact. 5.1.2

Rainfall

Rainfall and the resulting pools of water that stimulate the breeding of mosquitoes have also been found to be contributors to disease and mortality although the impact varies by type of disease. Rogers’s original studies of rainfall data in the Northwest Provinces of India indicated that smallpox epidemics were unheard of during periods of heavy rainfall, erupting instead when rain was limited. Nishiura and Kashiwagi (2009) have reproduced Rogers’s findings using modern econometric and epidemiological techniques, while MacCallum and McDonald (1957) found evidence that humidity and warm temperatures undermine the viability and lifespan of the smallpox virus. Although vaccination programs launched by American states during the nineteenth century had mostly (though not entirely) eliminated smallpox by the 1930s, it is not difficult to postulate other mechanisms linking rainfall to disease and mortality. For example, sewage- tainted water was a common transmission vector of both diarrheal disease and typhoid fever. To the extent excessive rainfall diluted the sewage found in public water sources, it would have also reduced the amount of waterborne illness. Consistent with this line of thought, serious flooding in the rivers around Pittsburgh (from which the city drew its water) during the mid- 1890s was associated with unusually large drops in the city’s diarrhea and typhoid rates (Troesken 2004, 29, 56). This connection, though speculative, might help explain some findings reported later in the chapter that suggest an inverse correlation between rainfall and infant mortality. 5.2

Mortality, Weather, Information, and Economic Development

The relationships between climate, weather, and mortality are highly specific to context, and they are mediated through institutions and technologies developed by humans. As people understood more about the mechanisms that connected climate to disease, they developed means of prevention

140

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

that served to reduce the measured impact of climate. To illustrate, during the late nineteenth century, scientists began to understand that the rise in diarrheal deaths during the hot summer months was related to pathogens that thrived in unpasteurized milk and unfiltered water supplies. Methods for pasteurization and water purification were developed to destroy nearly all of the pathogens that cause typhoid and diarrhea. As pasteurized milk became more common and cities filtered public water supplies, the rates of typhoid and diarrhea no longer varied much by season or in response to temperature. Similarly, flies were much less likely to spread disease once cities replaced outdoor cesspools and privies with public sewer systems and indoor toilets. Similar improvements in mortality were seen in smaller cities and rural areas even though they were slower to adopt sewer systems and filtered water. The lower population densities allowed such areas to have lower mortality in the late 1800s by alleviating problems, like the spread of infectious disease, associated with tightly packed populations. Even though smaller cities and rural areas were slower to adopt sewers and filtered water, people were able to limit the impact of flies by using more screens and introducing concrete vault privies with chemical treatments that limited the impact of the privies on local water supplies and the fly problem (Fishback and Lauszus 1989). To the extent that vaccinations minimized the propagation and spread of influenza, the winter spike in mortality tended to moderate. The influence of public health education and prevention on the relationship between temperature and infant mortality is illustrated in figure 5.2, which plots the infant mortality rate in Chicago against temperature using monthly data for two periods, 1871 to 1890 and 1895 to 1907. Smoothed lines that capture the typical relationships between temperature and infant mortality rates for the two periods are included to make the typical differences easier to see. The black triangles representing months between 1871 and 1890 show a flat relationship between average monthly temperature and infant mortality for temperatures between 15° and 60°F. After reaching 65°, however, the infant mortality rate leaps in response to higher temperatures, rising from less than 50 deaths per 100,000 (per month) to 100 to 250 deaths. This leap illustrates why nineteenth- century observers were so concerned about disease during the summer. Between 1890 and 1895, Chicago introduced water purification, mandated tougher milk inspection, and the diphtheria antitoxin was introduced. What happened? The 1895 to 1907 observations marked by empty circles are typically 50 percent lower than the triangles for 1871 to 1890. It is even more remarkable that the strong correlation between high temperatures and infant mortality above 65° is essentially eliminated (Ferrie and Troesken 2008). To the extent that the cities and urban areas contained in our Depression-Era sample had made similar investments, we do not expect to observe strong correlations between infant mortality (or mortality in general) and temperature and rainfall.

The Impact of Climate on Mortality Rates during the Great Depression

141

Fig. 5.2 Relationships between infant deaths per 100,000 people and the monthly average of time-of-day adjusted temperature in Chicago, 1855–1890 (triangles) and 1895–1906 (dots) Source: City of Chicago (various years between 1871 and 1906).

The introduction of these new public health technologies reduced the measured relationship between climate, weather, and mortality, but not everybody gained access to the information or the technologies. In the 1910s and 1930s, public health officials at all levels developed education programs to teach people simple ways to reduce the spread of disease with emphasis on washing hands and food and making sure that pools of water did not form in mosquito season (Fox 2009). The illiterate and people with limited access to information were less likely than the rest of the population to receive these messages. If there were enough of the ill- informed who drank unpasteurized milk or unfiltered water or did not adequately deal with privies, the long- run climate and mortality relationships still would have continued. The success of public health programs at eliminating such interactions is, therefore, an empirical question that we begin to address in the next section. 5.3

Data and Estimation

To examine the impact of climate/weather on death rates during the 1930s, a series of OLS regressions are estimated with White- corrected robust standard errors clustered at the state level. The regressions take the basic form DRit  0  1Wit  2Xit  εit,

142

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

where DR is the death rate in county i in year t. We estimate separate regressions for infant mortality rates; the number of infant deaths per 1,000 live births; and, for the noninfant death rate, the number of deaths of people over the age of one per 1,000 people. Wit is a vector of climate/weather variables in county i in year t. We use several different measures of weather that either focus on annual averages of rainfall and temperature or on distributions of the number of days at different temperatures over the course of the year. The Xit vector refers to a wide range of correlates that include demographic, economic, New Deal spending, and geographic variables describing county i in year t. Appendix table 5A.1 contains a list of the correlates with information on means and standard deviations for the panel. The data set for estimation is annual data for 3,054 counties (or groupings of counties designed to match up with New Deal spending information) each year for the years 1930 through 1940. The data on daily high temperatures and precipitation are aggregated from information originally collected by the United States Historical Climatology Network from 362 weather stations that were operational by 1930 and had complete daily weather data between 1930 and 1960. To measure the daily weather at each county seat, we used the Haversine formula to convert information on latitude and longitude from two locations to measure the distances between weather stations and county seats. The daily weather at the nearest weather station was used as a proxy for the weather in the county. The information on infant deaths, noninfant deaths, and births used to construct death rates and are from annual volumes of Birth, Stillbirth and Infant Mortality Statistics for the Continental United States and Mortality Statistics (U.S. Bureau of the Census, various years). The sources for the correlates are in the appendix. We start with an analysis of the role of climate/weather on infant mortality that takes into account both cross- sectional and time series variation. In the following, we discuss the impact of weather changes when we incorporate geographic fixed effects to control for long- term climate. The initial analysis starts with an OLS regression of infant mortality as a simple linear function of annual average high temperature and annual precipitation. Table 5.2 shows a series of OLS regressions with and without correlates. In the sparest specification (1), the number of infant deaths rises by a statistically significant 0.72 per 1000 live births with an increase of 1°F in annual average temperature. Meanwhile, greater precipitation has a small and imprecisely estimated negative effect on infant mortality of –0.12 deaths per life birth for a one- inch increase in annual precipitation. The most interesting feature of table 5.2 is what happens to the impact of temperature as correlates are added to the analysis. The sizeable effect of temperature on infant mortality largely goes away when we add one correlate to the analysis, the percentage illiterate. Just the addition of that one variable cuts the effect of high temperature from 0.72 in specification (1) to a statistically insignificant 0.095 in specification (2). Meanwhile, the percent

143

The Impact of Climate on Mortality Rates during the Great Depression Table 5.2

Coefficients and t-statistics from regressions of infant deaths per thousand live births on annual average high temperature, annual precipitation, and other correlates Coefficient (t-statistics)

Variable

Spec. 1

Spec. 2

Spec. 3

Spec. 4

Spec. 5

Average daily high temperature in year Inches of precipitation during year % illiterate

0.722 (6.18) –0.121 (–1.36)

0.095 (1.04) –0.288 (–3.48) 2.049 (5.91)

–0.062 (–0.62) –0.276 (–3.32) 1.867 (5.17) –0.261 (–6.13) 0.348 (4.52)

0.427 (1.92) –0.108 (–1.91)

Included

–0.183 (–1.58) –0.160 (–3.7) 2.069 (3.99) –0.413 (–10.56) –0.220 (–2.96) Included

32,423

32,421

% owning radio Per capita circulation of 15 magazines, 1929 Remaining correlates included N

32,598

32,598

32,584

Notes: The regressions have White-corrected robust standard errors, which are clustered at the state level. Reported R2 range from 0.039 to 0.22. The remaining correlates are retail sales per capita; auto registrations per capita; tax returns filed per capita; crop value, percent home ownership; Public Works Administration grants per capita; Agricultural Adjustment Administration grants per capita; relief grants per capita; Public Roads Administration grants per capita; Disaster Loan Corporation, loans per capita; farm loans per capita; Reconstruction Finance Corporation loans per capita; U.S. Housing Authority loans per capita; Civilian Conservation Corps camps established in year t; Civilian Conservation Corps camps established in year t – 1; Civilian Conservation Corps camps established in year t – 2; hospital beds per female aged fifteen to forty-four potentially available for infants, employment in polluting industries, 1930; coal tonnage, results of bovine tuberculosis testing; births per woman aged fifteen to forty-four; percent women aged twenty to twenty-four of women aged fifteen to forty-four; percent women aged twenty-five to twenty-nine of women aged fifteen to fortyfour; percent women aged thirty to thirty-four of women aged fifteen to forty-four; percent women aged thirty-five to forty-four of women aged fifteen to forty-four; percent urban; percent foreign-born, percent African American; population per square mile; percent families with electricity; manufacturing employment per capita; retail employment per capita; number of lakes; number of swamps; maximum elevation; elevation range; percent church membership; number of rivers that pass through eleven to twenty counties in county; number of rivers that pass through twenty-one to fifty counties in county; number of rivers that pass through over fifty counties in county; number of bays; number of beaches, on Atlantic Coast, on Pacific Coast, on Gulf Coast, on Great Lakes; land area in square miles, and a constant term.

illiterate in the population has a strong and statistically significant impact of raising the infant mortality rate by two deaths per thousand for a 1 percent increase. The importance of knowledge is reinforced by the addition of two more measures of access to information to the analysis, the share of households with radios, and the per capita circulation of fifteen news magazines in 1929. When both are added to the analysis, the coefficient of temperature falls from 0.09 in specification (2) to –0.06 in specification (3). While the presence of the radio is associated with reductions in infant mortality, the impact of

144

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

the magazine circulation variable is unexpectedly positive. However, there appears to be a positive omitted variable bias to this coefficient because when a full set of income, demographic, and geographic correlates is added to the analysis, the coefficient has the expected negative effect. It is dicey to argue for the importance of a small number of variables by adding them to the analysis without the other correlates because of crosscorrelations between correlates. In this case, however, the importance of the information variables stands out when all of the other correlates are included. Specification (4) of table 5.2 shows the climate coefficients when all of the correlates except for the information variables are included in the analysis. The inclusion of the other correlates as a group cuts the impact of the average high temperature in half from 0.72 to 0.427. When the information variables are added to the rest of the correlates in specification (5), the temperature coefficient is cut dramatically from 0.427 to –0.183. In this specification, the coefficients of the information variables are all statistically significant with the expected signs: infant mortality is positively related with illiteracy, less access to radios, and less readership of magazines. This sequence of results shows the importance of incorporating access to knowledge in studies of the relationship between climate and mortality. Had the measures not been included, we would have concluded that high temperatures were strongly related with higher infant mortality. In fact, once measures of access to knowledge were included, the results show that the real culprit that contributed to higher infant mortality was less access to knowledge, and people with less access to knowledge were much more likely to live in areas with higher temperatures, on average. There are a huge number of potential specifications for the temperature and precipitation variables that could be tried. We explored a number of higher- order polynomial specifications with squared and cubed terms. However, there is relatively little gain to this with the annual average data primarily because average annual temperatures only ranged from 47° to 91°F in the sample. Given the small range and the relative inflexibility of the polynomials, other approaches are preferred. We estimated a model with a relatively flexible formulation for temperature by using the share of days of the year that the daily high temperature was in different temperature bands. Table 5.3 shows the relationships between infant mortality and climate with and without the information variables and the remaining correlates. Because the shares of the temperature bands sum to one, we excluded a reference temperature band for days with daily highs at or above 50° and below 60°. The simplest specification is somewhat surprising. We anticipated that more days above 100° would lead to higher infant mortality. The coefficient was a positive 2.4, but the effect was not statistically significant. Relative to the 50 to 60° range, higher infant mortality was associated with a higher share of days with temperatures in the 70s and less than zero. Greater precipitation was also associated with lower infant mortality.

Table 5.3

Coefficients and t-statistics from regressions of infant mortality rate on share of days during year in temperature bands, annual precipitation, and other correlates Coefficient (t-statistics)

Share of days in year that high temperature High  100 100  high  90 90  high  80 80  high  70 70  high  60 60  high  50 50  high  40 40  high  30 30  high  20 20  high  10 10  high  0 0  high  –10 –10  high Inches of precipitation during year % illiterate

Spec. 1

Spec. 2

Spec. 3

Spec. 4

Spec. 5

2.380 (0.06) 2.365 (0.11) 24.540 (1.18) 61.178 (2.62) 12.142 (0.5) — 1.206 (0.05) –60.914 (–2.23) –4.604 (–0.15) –89.439 (–2.26) –57.800 (–1.1) 131.196 (1.72) 184.886 (1.39) –0.209 (–1.95)

13.166 (0.44) –45.003 (–2.53) –11.216 (–0.72) 1.87115 (0.12) –30.066 (–1.81) — 8.824 (0.5) –48.932 (–2.33) –23.524 (–0.98) –97.978 (–3.31) –56.908 (–1.42) 42.228 (0.61) 41.908 (0.42) –0.307 (–3.39) 2.147 (6.24)

23.970 (0.81) –39.928 (–2.11) –0.878 (–0.07) –4.687 (–0.32) –22.320 (–1.61) — 18.793 (1.26) –31.442 (–1.68) –3.196 (–0.14) –64.273 (–2.63) –33.339 (–0.87) 32.505 (0.48) 49.850 (0.62) –0.281 (–3.03) 1.965 (5.51) –0.252 (–6.55) 0.310 (3.72)

38.723 (1.17) –19.275 (–0.78) 9.674 (0.56) 18.363 (1.02) 7.312 (0.47) — 5.300 (0.34) –40.911 (–2.19) –20.393 (–1.06) –47.681 (–1.8) –116.910 (–2.13) 12.780 (0.21) 130.359 (1.29) –0.146 (–2.36)

Included

23.875 (0.99) –26.444 (–1.78) 0.692 (0.06) –1.651 (–0.15) –2.311 (–0.18) — 27.597 (2.3) –16.862 (–1.18) –4.084 (–0.25) 5.786 (0.23) –8.869 (–0.23) 45.641 (0.63) 135.520 (1.3) –0.155 (–3.3) 2.065 (3.92) –0.408 (–10.45) –0.220 (–2.97) Included

32,423

32,421

% owning radio Per capita circulation of 15 magazines, 1929 Remaining correlates included N

32,598

Note: See table 5.2 notes. Dash indicates reference temperature band.

32,598

32,584

146

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

The effects of climate/weather are transformed once again when we include additional correlates, but the story is not as simple as the one told in the preceding. The inclusion of all but the information variables in specification (4) in table 5.3 leads to a sharp rise in the effect of shares of days over 100° from 2.4 to 38.7, such that a 1 percent increase in the share raises the infant mortality rate by 0.387, but the effect is statistically insignificant. Many of the effects in the spare specification (1) are weakened sharply. Adding the information variables in specification (5) cuts the impact of days over 100° roughly in half to 23.9, while leading to a statistically significant effect of the share of days with temperatures in the 40s. In general, most of the temperature bands do not have much effect on infant mortality rates. 5.4

Infant Mortality and Annual Fluctuations in Temperature and Precipitation

The prior section focused on the impact on infant mortality of climate because so much of the variation in the analysis was cross- sectional across counties. In this section, we perform a difference analysis that controls for time- invariant features of each county and for common shocks to infant mortality throughout the country that occurred in specific years. The equation estimated takes the following form: DRit  DRit1  0  1 (Wit  Wit1)  2 (Xit  Xit1)  t  εit  εit1. where (DRit – DRit–1) is the change in the mortality rate (infant or noninfant) infant mortality from the previous year, (Wit – Wit–1) is a vector of changes in weather from the previous year, (Xit – Xit–1) is a vector of changes in other correlates from year to year, t is a vector of year dummies, and (εit – εit–1) is the change in unobservable factors that vary across time. By estimating the relationship between the change in infant mortality and the change in weather, the analysis controls for factors that vary across counties but did not change over time. To the extent that the climate in the area is considered time- invariant, the analysis controls for the climate, and the vector of 1 coefficients captures the relationship between changes from year to year in the weather and changes in infant mortality. An alternative description is that the analysis captures the effects of weather deviations from the long- run climate on infant mortality. The differencing also controls for time- invariant features of the geography. The inclusion of a vector of year dummies controls for factors like the introduction of sulfa drugs in 1936 and 1937 that would have affected all of the counties simultaneously (Thomasson and Treber 2008). A number of the variables that we included as controls in the prior section were based on census information reported only in 1930 and 1940. As

The Impact of Climate on Mortality Rates during the Great Depression

147

seen in the data descriptions in appendix table 5A.1, we used straight- line interpolations between the census years to fill in values for these variables in the intervening years. So the values in the prior section were basically trend values for those variables. Because the change in these variables would be the same in each year, we do not include them in the differencing specification. The variables that we do include in the (Xit – Xit–1) correlates vector either vary from year to year (share of tax returns, hospital beds per capita, measures of bovine tuberculosis, the general fertility rate [births/interpolated value for share of women aged fifteen to forty- four]), or we could use changes in state measures to interpolate between various years throughout the period (retail sales per capita, auto registrations per capita, crop values per farm population, the New Deal program measures.) Specifications (I- 1) and (I- 2) in table 5.4 show the results of the difference analysis using the simplest form of the changes in average annual high temperatures and changes in annual inches of precipitation, while specifications (I- 3) and (I- 4) show the results using changes in the share of days in different temperature bands. The simple specifications of annual averages in specifications (I- 1) and (I- 2) suggest that weather fluctuations had little or no effect on infant mortality rates. The coefficients on both precipitation and average daily highs are small and statistically insignificant in specifications both with and without the extra correlates. The coefficients of the changes in the shares of days within each temperature band in specification (I- 4) suggest a similar story. There are only two coefficients of the change in the percentage days in each temperature band that are statistically significant at the 10 percent level or better, the ones for temperatures in the thirties and temperatures in the teens. The effects are small, however, with elasticities for all coefficients below –0.06, such that a 1 percent rise in the share of days in the temperature band would have led to at most a –0.06 percent reduction in infant mortality. The analysis in table 5.4 also includes information on the coefficients of the other time- varying correlates in the analysis. A number of the relationships with infant mortality have been seen in other studies of death rates, in some cases including infant mortality rates. As seen here, a number of studies show a positive relationship between death rates and the number of hospital beds in an area. There are several potential reasons for this effect. One is that the data on deaths report the location of the death not the residence of the deceased. Areas with more hospital beds tend to report more deaths because people with potentially fatal illnesses from areas without hospitals often came to areas with hospitals to receive treatment. A second possibility is that there was endogeneity bias because areas with higher death rates were more likely to add more hospital beds per capita. Because increased numbers of hospital beds involved capital expenditures, this effect may have been weakened to the extent that the addition of hospital beds lagged a rise in the death rate by a year or two.

10  high  0

20  high  10

30  high  20

40  high  30

60  high  50 50  high  40

70  high  60

80  high  70

90  high  80

100  high  90

Change in share of days with high temperatures between: High  100

Change in average daily high temperature

–2.869 (–3.36) 0.007 (0.41) 0.001 (0)

Spec. 1 –1.823 (–1.73) –0.002 (–0.13) 0.050 (0.27)

Spec. 2

–15.499 (–1.15) –19.015 (–1.17) –12.930 (–1.02) –19.488 (–1.23) –12.959 (–0.92) — –25.799 (–1.90) –18.553 (–1.63) –6.591 (–0.52) –52.942 (–3.90) 27.853 (0.80)

–3.034 (–3.34) 0.005 (0.25)

Spec. 3

Change in infant mortality rate

–4.053 (–0.28) –16.892 (–1.14) –7.543 (–0.62) –19.481 (–1.24) –14.077 (–1.03) — –19.539 (–1.58) –17.970 (–1.65) –7.066 (–0.58) –42.102 (–3.30) 28.578 (0.87)

–1.532 (–1.43) 0.005 (0.23)

Spec. 4 –0.366 (–6.26) –0.002 (–1.4) 0.005 (0.29)

Spec. 1

Dependent variable

–0.309 (–3.06) –0.003 (–1.58) –0.001 (–0.04)

Spec. 2

–0.320 (–0.24) –2.335 (–2.05) –1.702 (–1.82) –3.076 (–2.96) –0.996 (–1.24) — –0.793 (–0.90) –1.561 (–2.33) –2.674 (–3.17) 0.117 (0.08) 1.096 (0.55)

–0.350 (–5.19) –0.003 (–1.27)

Spec. 3

Change in noninfant death rate

–0.447 (–0.33) –2.438 (–2.12) –1.721 (–1.84) –3.217 (–2.96) –1.190 (–1.46) — –1.127 (–1.35) –1.348 (–1.89) –2.142 (–2.65) 0.292 (0.20) 1.433 (0.76)

–0.257 (–2.55) –0.002 (–1.17)

Spec. 4

Coefficients and t-statistics from regressions of change in death rates as functions of changes in climate variables and change in other correlates, U.S. counties, 1931–1940: coefficients (t-statistics)

Change in inches of precipitation during year

Constant

Table 5.4

Disaster Loan Corporation loans per capita

Public Roads Administration grants per capita

Agricultural Adjustment Administration grants per capita Relief grants per capita

Public Works Admin. Grants Per Capita

Tax Returns Filed Per Capita

Crop value

Auto registrations per capita

% owning radio

Retail sales per capita

Hospital beds per female aged 15–44 potentially available for infants Births per woman aged 15–44

Change in: Results of bovine tuberculosis testing

–10  high

0  high  –10

0.029 (1.14) 0.020 (0.6) 0.163 (0.72) –0.086 (–0.18)

0.525 (1.19) 0.077 (1.98) –0.261 (–7.51) 0.027 (2.84) –0.171 (–2.28) 13.967 (0.87) 0.000 (–0.32) –3.417 (–0.24) –0.024 (–1.01)

–53.798 (–1.21) –149.597 (–2.50)

0.029 (1.13) 0.018 (0.53) 0.162 (0.72) –0.118 (–0.24)

0.470 (1.05) 0.077 (2.00) –0.262 (–7.42) 0.029 (2.92) –0.192 (–2.67) 12.396 (0.80) 0.000 (–0.33) 3.865 (0.25) –0.024 (–1.01)

–32.765 (–0.88) –74.660 (–1.23)

0.003 (1.87) 0.008 (3.14) –0.006 (–0.57) –0.090 (–2.4)

–0.032 (–1.14) 0.002 (1.56) 0.018 (8.69) 0.001 (0.75) 0.002 (0.24) 0.575 (0.62) 0.000 (–0.2) 2.563 (2.11) 0.001 (1.31)

–12.052 (–4.44) 0.454 (0.07)

0.002 (1.46) 0.007 (2.85) –0.005 (–0.53) –0.088 (–2.36) (continued )

–0.030 (–1.08) 0.002 (1.61) 0.018 (8.67) 0.001 (1.16) 0.000 (–0.06) 0.439 (0.52) 0.000 (–0.03) 2.806 (2.03) 0.002 (1.52)

–11.301 (–4.02) 3.049 (0.62)

(continued)

–0.812 (–0.58) 3.700 (2.56) 4.665 (3.28) –0.720 (–0.49) 3.510 (3.4) 0.188 (0.14) 0.474 (0.31) 0.428 (0.31) 1.680 (1.07)

Spec. 1

0.941 (0.62) 2.518 (1.39) 3.819 (2.44) –2.869 (–1.81) 2.642 (2.05) –0.974 (–0.64) 1.339 (0.75) –1.177 (–0.76) 2.674 (1.69)

0.048 (0.93) –0.094 (–1.67) –0.007 (–0.12) 0.016 (0.08)

Spec. 2

–0.409 (–0.26) 3.981 (2.86) 4.714 (3.24) –0.697 (–0.45) 4.118 (3.61) 0.015 (0.01) 0.570 (0.34) 0.519 (0.35) 2.002 (1.14)

Spec. 3

Change in infant mortality rate

0.827 (0.51) 2.487 (1.39) 3.326 (2.13) –3.190 (–1.81) 2.387 (1.94) –1.152 (–0.71) 1.207 (0.67) –1.613 (–0.95) 2.391 (1.43)

0.053 (1.03) –0.094 (–1.65) 0.001 (0.01) 0.000 (0.00)

Spec. 4

0.381 (4.31) 0.227 (2.47) 0.728 (7.35) 0.294 (2.93) 0.863 (9.19) –0.072 (–0.76) –0.192 (–1.87) 0.294 (5.13) 0.456 (6.07)

Spec. 1

Dependent variable

0.363 (3.93) 0.261 (2.28) 0.453 (3.57) 0.194 (1.79) 0.871 (6.68) –0.165 (–1.56) –0.260 (–1.96) 0.193 (2.66) 0.406 (3.17)

0.006 (2.18) –0.001 (–0.48) –0.010 (–2.66) –0.002 (–0.12)

Spec. 2

0.411 (3.67) 0.200 (2.12) 0.685 (7.11) 0.257 (2.29) 0.846 (9.02) –0.075 (–0.87) –0.134 (–1.19) 0.221 (2.70) 0.450 (5.21)

Spec. 3

Change in noninfant death rate

Note: The regressions have White-corrected robust standard errors (in parentheses), which are clustered at the state level. Dash indicates reference temperature band.

Year 1940

Year 1939

Year 1938

Year 1937

Year 1936

Year 1935

Year 1934

Year 1933

Year dummies Year 1932

Civilian Conservation Corps camps established in year t

Reconstruction Finance Corporation loans per capita U.S. Housing Authority loans per capita

Farm loans per capita

Table 5.4

0.371 (3.09) 0.221 (1.91) 0.379 (3.10) 0.133 (1.04) 0.799 (5.99) –0.194 (–1.88) –0.225 (–1.62) 0.093 (0.96) 0.343 (2.59)

0.005 (2.03) –0.002 (–0.49) –0.009 (–2.25) –0.004 (–0.28)

Spec. 4

The Impact of Climate on Mortality Rates during the Great Depression

151

This still would not resolve the problem if there were serial correlation in the death rates.11 The measure of economic activity, retail sales per capita, displays a positive relationship with infant mortality rates during the 1930s. It has long been thought that improved incomes would reduce infant mortality rates (Antonovsky and Bernstein 1977; Clifford and Brannon 1978; Dehejia and Lleras-Muney 2004; Kaplan et al. 1996; Kennedy, Kawachi, and ProthrowStith 1996; and Waldmann 1992). Recently, however, much evidence has emerged to challenge this commonplace assumption. For example, Fishback, Haines, and Kantor (2007) found a positive relationship between economic activity and several types of death rates in their fixed effects estimates using a panel of annual data for 114 cities between 1929 and 1940. Christopher Ruhm (2000) also found similar procyclical effects for various death rates in fixed effects analyses in the 1970s, 1980s, and 1990s. Further in the past, the antebellum puzzle is perhaps the quintessential example of rising death rates having been associated with increased economic activity. Haines, Craig, and Weiss (2003) show that the positive correlation between economic activity and poor health is driven, in part, by the greater transmission of germs during associated with movement of people and goods. Nor, it should be noted, was this dynamic limited to the United States. In early stages of development, England and Wales also exhibited a negative correlation between health and growth (Fogel 1994; Steckel 1992). In a study of yellow fever and smallpox, Beeson and Troesken (2006) find evidence that severe epidemics were positively correlated with economic activity. Fast growing port cities were ripe targets for the inflow of new infections and new populations of vulnerable (i.e., previously unexposed) migrants; sleepy backwaters did not have such a dubious honor. Due to problems with pollution from leaded gasoline, we had expected a substantial effect of the change in automobile registrations on infant mortality. More automobiles led to more lead emissions from the leaded gasoline that was widely used at this time. It has long been suspected that lead emissions harm fetal and infant development. In fact, the phase out of leaded gasoline during the 1970s was associated with small but statistically significant reduction is in infant mortality (Reyes 2002). Similarly, Greenstone and Chay (2003) find that reductions in pollutants are associated with lower infant mortality in cities in the modern era. However, the coefficients of automobile 11. In a separate unreported analysis, the positive relationship between hospital beds and infant mortality was essentially eliminated in the years after 1936 when the use of sulfa drugs had spread throughout the nation. Thomasson and Treber (2008) found that the number of deaths of mothers during child birth had been slightly negatively related to the number of hospital beds for most of the period from 1920 through 1936. They found evidence that there was a greater likelihood of sepsis infections in hospitals than outside hospitals. Doctors could do little about the infections into the introduction of sulfa drugs throughout the country around 1937. However, once the drugs were available, there was a more negative correlation between access to hospitals and maternal mortality.

152

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

registrations per capita in table 5.4 imply that a 1- percent rise in automobile registrations would have been associated with less than a 0.02 percent rise in infant mortality and the effect is statistically insignificant. There is weak evidence that greater problems with bovine tuberculosis (BTB) were associated with greater infant mortality. Alan Olmstead and Paul Rhode (2008) have reported that large numbers of children and infants were killed by the transmission of the disease into the milk supply from diseased dairy cattle in the late 1800s and early 1900s. An extensive BTB eradication program between 1900 and 1930 had greatly diminished the problem but not fully eliminated it. Bovine tuberculosis was much less widespread in the 1930s and may have been less virulent. The positive coefficient suggests still some effects, but the coefficient is not statistically significant. Areas with higher general fertility rates, births per woman aged fifteen to forty- four, were associated with lower infant mortality rates. Measures of spending and loan activity from a series of major New Deal programs are also included. None of the programs appear to have strong reductive effects on infant mortality. There is the potential of endogeneity bias that might have weakened the effectiveness of the programs. When Fishback, Haines, and Kantor (2007) controlled for endogeneity bias in their study of major cities in the 1930s, they found evidence that greater relief spending helped reduce infant mortality rates. 5.5

Climate, Weather, and Death Rates for the Noninfant Population

The relationships between climate/weather and mortality rates for the rest of the population above the age of one are similar to what we see for infant mortality rates. As in the case of infant mortality, greater literacy and more access to radios and magazines are associated with lower death rates. A feature that is different from the infant mortality pattern, however, is that adding the information variables to a specification that includes only climate/weather variables does not change the relationship between noninfant mortality and climate. The real test, however, is what happens when we add the measures of access to knowledge to specifications that include all correlates. As was the case with infant mortality, the addition of the information variables reduced the effects of climate/weather are reduced. This finding highlights once again the importance of controlling for access to knowledge when measuring the impact of climate on death rates. Without such controls, other studies might well overstate the impact of temperature on mortality. Tables 5.5 and 5.6 document these patterns. In table 5.5, the climate/ weather patterns are measured with the average high temperature for the year and the total inches of precipitation during the year. The simplest relationships in specification (1) show that higher noninfant mortality is associated with lower average temperatures and more precipitation. The addition of the information variables to the simplest specification, moving from specification (1) to those in columns (2) and (3), has little impact on

The Impact of Climate on Mortality Rates during the Great Depression Table 5.5

153

Coefficients and t-statistics from regressions of noninfant deaths per thousand people on annual average high temperature, annual precipitation, and other correlates Coefficient (t-statistics)

Average daily high temperature in year Inches of precipitation during year % illiterate % owning radio Per capita circulation of 15 magazines, 1929

Spec. 1

Spec. 2

Spec. 3

Spec. 4

Spec. 5

–0.051 (–2.12) 0.031 (2.63)

–0.063 (–2.12) 0.028 (2.4) 0.038 (0.96)

–0.032 (–1.61) 0.043 (4.94) 0.127 (3.35) –0.009 (–1.73) 0.151 (9.58)

0.022 (1.62) 0.006 (1.39)

–0.033 (–3.34) –0.002 (–0.42) 0.134 (3.61) –0.053 (–10.1) –0.021 (–2.48)

Notes: The regressions have White-corrected robust standard errors, which are clustered at the state level. Reported R2 range from 0.039 to 0.22. The remaining correlates are retail sales per capita; auto registrations per capita; tax returns filed per capita; crop value; percent home ownership; Public Works Administration grants per capita, Agricultural Adjustment Administration grants per capita; relief grants per capita; Public Roads Administration grants per capita; Disaster Loan Corporation loans per capita; farm loans per capita, Reconstruction Finance Corporation loans per capita; U.S. Housing Authority loans per capita; Civilian Conservation Corps camps established in year t; Civilian Conservation Corps camps established in year t – 1; Civilian Conservation Corps camps established in year t – 2; hospital beds per female aged fifteen to forty-four potentially available for infants; employment in polluting industries, 1930; coal tonnage; results of bovine tuberculosis testing; births per woman aged fifteen to forty-four; percent of population aged five to nine, ten to fourteen, fifteen to nineteen, twenty to twenty-four, twentyfive to twenty-nine, thirty to thirty-four, thirty-five to forty-four, forty-five to fifty-four, fifty-five to sixtyfour, sixty-five to seventy-four, and seventy-five and over; percent urban; percent foreign-born; percent African American; population per square mile; percent families with electricity; manufacturing employment per capita; retail employment per capita; number of lakes; number of swamps; maximum elevation; elevation range; percent church membership; number of rivers that pass through eleven to twenty counties in county; number of rivers that pass through twenty-one to fifty counties in county; number of rivers that pass through over fifty counties in county; number of bays; number of beaches, on Atlantic Coast, on Pacific Coast, on Gulf Coast, on Great Lakes; land area in square miles; and a constant term.

the relationship between climate/weather and noninfant mortality. When we add all of the correlates except the information variables in specification (4), the relationship between noninfant mortality and the average daily high temperature switches signs from negative to positive. Further, the positive relationship with precipitation is cut dramatically from a statistically significant 0.03 in column (1) to 0.006 in column (4). The addition of the information variables to specification (4) to create specification (5) causes the temperature coefficient to switch signs again to a negative and statistically significant –0.033. Meanwhile, the precipitation coefficient turns negative but with an even smaller magnitude than in specification (4). The importance of access to knowledge is highlighted by the statistically significant positive relationship of noninfant mortality with the percent illiterate and the negative coefficients on radio ownership and magazine circulation. When climate/weather is measured with the share of days in each

Note: See table 5.5 notes. Dash indicates reference temperature band.

Inches of precipitation during year

–10  high

0  high  –10

10  high  0

20  high  10

30  high  20

40  high  30

60  high  50 50  high  40

70  high  60

80  high  70

90  high  80

100  high  90

5.9253199 (1.74) –4.427665 (–1.44) 3.0792415 (1.2) 6.4659766 (2.3) 3.6034486 (1.48) — 5.2266821 (1.29) 8.114672 (2.5) –3.238951 (–0.86) 6.430615 (0.99) –31.83301 (–2.74) –26.27229 (–2.16) 28.442934 (1.6) 0.025201 (2.08)

Spec. 1 6.245898 (1.89) –5.8174 (–1.9) 2.030206 (0.71) 4.723384 (1.62) 2.368396 (0.93) — 5.455708 (1.3) 8.473622 (2.64) –3.79263 (–1.01) 6.184703 (0.97) –31.803 (–2.76) –28.8894 (–2.4) 24.24312 (1.37) 0.022298 (1.93)

Spec. 2 5.6537376 (1.7) –5.11953 (–2.07) –0.01870 (–0.01) –1.21041 (–0.51) –0.76792 (–0.37) — 0.7692672 (0.26) 3.3481582 (1.06) –1.16735 (–0.34) 3.7969074 (0.74) –21.71878 (–2.17) –26.97869 (–2.43) 18.496452 (1.23) 0.0436157 (4.45)

Spec. 3

Coefficient (t-statistics)

4.0611318 (2.04) –2.09142 (–1.33) 0.1518418 (0.12) 1.0813113 (0.78) 1.5105014 (0.97) — –0.34326 (–0.25) –2.72522 (–1.66) –0.37342 (–0.18) –0.253966 (–0.09) –12.8939 (–2.31) –3.27581 (–0.7) 39.282855 (5.39) 0.0048751 (1.04)

Spec. 4

2.3489633 (1.48) –3.10996 (–2.97) –1.1193712 (–1.12) –0.8415495 (–0.73) 0.3858891 (0.27) — 1.242737 (1) –0.6973798 (–0.49) 0.76637834 (0.36) 3.4170776 (1.29) –3.1538209 (–0.67) –0.9729119 (–0.29) 33.403666 (4.88) 0.00116884 (0.26)

Spec. 5

Coefficients and t-statistics from regressions of noninfant mortality rate on share of days during year in temperature bands; annual precipitation, and other correlates

Share of days in year that high temperature High  100

Table 5.6

The Impact of Climate on Mortality Rates during the Great Depression

155

temperature band in the table 5.6 regressions, the same pattern still arises. When no correlates aside from climate are included in the analysis in specification (1) in table 5.6, noninfant mortality is higher with greater precipitation. The temperature comparisons are relative to the share of days when the high temperature is in the 50s. Noninfant mortality is statistically significantly higher when there were relatively more days with high temperatures exceeding 100°, in the 70s, in the 30s, and below minus 10°. It was lower when there were more days in between 0° and 10° and between –10° and 0°. When we add all correlates except the information variables to create specification (4) the precipitation coefficient is cut by three- fourths, while the only statistically significant temperature band coefficients are at the extremes. Temperatures over 100° and under minus 10° are associated with higher death rates, while temperatures in the 0° to 10° ranges were associated with lower death rates. When the information correlates are added in specification (5) in table 5.6, the coefficient of the over 100° temperature band is cut nearly in half, and the coefficient at the other extreme is cut by about 15 percent. The impact of weather fluctuations on noninfant mortality are examined by using differencing to control for long- run climate and other timeinvariant factors in specifications (N- 1) through (N- 4) in table 5.4. We also incorporate time fixed effects to control for nationwide shocks. Like the situation with infant mortality, fluctuations in the annual average high temperature and precipitation had small and statistically insignificant effects on changes in noninfant mortality with and without other time- varying covariates. When we examine the differences in the number of days in different temperature bands, there are statistically significant effects. The coefficients of the changes in the share of days in the 90s, 80s, 70s, 30s, 20s, and between 0 and minus 10 all were statistically significant and negative. However, the economic magnitude of the effects are even smaller than they were for infant mortality. None of the elasticities are more negative than –0.034. The coefficients of the remaining correlates in table 5.4 show that annual noninfant death rates also rose with increases in the number of hospital beds, the general fertility rate, a higher share of the population with enough income to pay federal income taxes, and in areas where the New Deal spent more per capita on relief programs and the farm programs, both loans and Agricultural Adjustment Act (AAA) grants. Death rates were lower in areas where more was spent on loans from the Disaster Loan Corporation. 5.6

Conclusions

Climate and health interact in a variety of complex ways that are strongly influenced by human decisions, locations, insect and animal populations, and a variety of different factors. We explore the raw correlations between climate and mortality during the Great Depression to see if we can discern any patterns and then incorporate a wide range of demographic, economic, and geographic correlates to examine whether the raw correlations are still

156

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

present. The results show that variations across the country in climate were associated with differences in infant mortality and noninfant death rates. However, much of the influence of climate is muted once the other correlates are included. One key finding in the study is the importance of controlling for access to information when measuring the relationship between mortality and climate. In specifications where measures of access to information are not included, the results often show a strong positive relationship between mortality and temperature. However, that relationship appears to be due to a positive bias arising from the omission of measures of access to knowledge. When measures of illiteracy, access to radios, and access to magazines are incorporated in the analysis, the strong positive relationship between mortality and temperature is no longer present. Public health scholars have long touted the health benefits of improved information flows during the campaigns to promote public health during the 1910s, 1920s, and 1930s. Certainly, we saw sharp declines in infant mortality during this period that cannot be fully explained by changes in income and sanitation. The results here provide support for this view. Both infant mortality and noninfant mortality rates were higher in areas where there was more illiteracy and lower in areas where people had more access to radios and the circulation of news magazines was greater. These effects are not just indirectly related to higher incomes because we control for urbanization and economic activity in the analysis. Finally, the results suggest that differences in climate rather than fluctuations in weather around the long- term climate norms have bigger effects on mortality. In Chicago in the late 1800s, the differences in mortality due to pneumonia were much higher in July and August than in the rest of the year, while fluctuations in temperature around the normal differences across months had relatively weak effects. In the county sample, the results show strong effects of weather when we do not control for time- invariant features of the climate. Once we control for the time- invariant features of the climate, the impact of weather fluctuations around the core climate are not very large. There is still much to explore about the relationship between climate, weather, and mortality. This is just a start that focuses on overall mortality rates. We plan further work to examine the specific weather patterns that scholars have identified for specific diseases. The specific mechanisms identified for these diseases can be complicated. As one example, St. Louis encephalitis (SLE) was the name given a disease that led to 1095 hospital cases and 201 deaths in St. Louis in the summer of 1933.12 St. Louis encepha12. Scholars suggest that Paris, Illinois reported thirty- eight cases and fourteen deaths from the same disease in 1932 but somehow escaped having the disease named after the town (Chamberlain 1980, 7).

The Impact of Climate on Mortality Rates during the Great Depression

157

litis is a mosquito- borne disease as well, but Thomas Monath (1980) found that later epidemics were typically associated with above- average temperatures and abnormally high precipitation in January and February, below normal temperature in April, above- average temperatures in May through August, and an abnormally dry July. In general, the warm conditions help the virus multiply within the mosquito population and the other requirements (e.g., for April) are associated with specific life- cycle events in the host populations. The conditions in St. Louis during the year of 1933 epidemic fit Monath’s ideal conditions. The winter of 1932 to 1933 was the second warmest on record, April was cool, and June through August were the driest months on record (Reiter 1988, 245–55). Other studies suggest that fluctuations in temperature throughout the day and throughout the month may influence the extent of the disease. More work, therefore, is needed to take the specific bioscience conditions into account when designing the weather variables used for further study.

Appendix Data Source The sources of information for the weather, death rates, and birth information are in the text. Population in 1910 and 1930, percent illiterate in 1930, percent of families with radios in 1930 and 1940, retail sales in 1929 and 1939, crop values in 1929 and 1939, percent homeowners in 1930 and 1940, percent urban in 1930 and 1940, percent foreign- born in 1930 and 1940, percent African American in 1930 and 1940, population density in 1930 and 1940, manufacturing employment in 1929 and 1939, and retail employment in 1929 and 1939 can be found in the data sets incorporated into Michael Haines (2004) ICPSR 2896 data set. Percent illiterate in 1940 was calculated using procedures developed in U.S. Bureau of the Census 1948 from data on education in Haines (2004). Retail sales in 1933 and 1935 are from U.S. Department of Commerce, Bureau of Foreign and Domestic Commerce (1936, 1939). Information on the number of federal individual income tax returns filed in county for 1929 is from U.S. Department of Commerce, Bureau of Foreign and Domestic Commerce (1932); 1930, 1933, 1937, and 1938 from U.S. Bureau of Internal Revenue (1932, 1935, 1939, and 1940, respectively); 1931, 1932, 1935, 1936, and 1939 from Rand McNally (1934, 1935, 1938, 1939, and 1943, respectively); 1934 from U.S. Department of Commerce, Bureau of Foreign and Domestic Commerce (1939). Information on the number of hospital beds in each county was compiled by Melissa Thomasson from reports by the American Medical Association (various years). See Thomasson and Treber (2008) for more details. Data

158

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

for the New Deal programs by county and state come from U.S. Office of Government Reports (1940a, b, respectively). Auto registrations by county in 1930 are from U.S. Bureau of Foreign and Domestic Commerce (1932). Auto registrations by county in 1931 are from the October 31, 1931 issue of Sales Management. County auto registrations in 1936 are from U.S. Bureau of Foreign and Domestic Commerce (1939). Annual state automobile registrations are from U.S. Public Roads Administration (1947). Circulation of fifteen national magazines as of January 1, 1929 is from U.S. Bureau of Foreign and Domestic Commerce (1936). The number of Civilian Conservation Corps camps started in fiscal year t in each county were determined by starting with camp lists from the Civilian Conservation Corps Legacy Web site. We then added some additional camps listed in the U.S. National Archives finding aid for Record Group 35, Civilian Conservation Corp. The camp lists listed the nearest railroad station and the nearest post office. We matched the camps to post offices by downloading post office locations by county from the http://www.usps.com/ postmasterfinder/welcome.htm Web site in 2007. The number of people employed in polluting industries of chemicals, cigars and cigarettes, glass, bread, meat packing, autos, iron and steel, nonmetals, planing mills, lumber mills, boots and shoes, printing, paper, cotton textiles, and rubber comes from U.S. Bureau of the Census, Fifteenth Census (1932). Population by age group in 1930 and 1940 is found in Gardner and Cohen (1992). Results of Bovine Tuberculosis Status tests are discussed in Olmstead and Rhode (2007), and the data are from U.S. Bureau of Animal Industry (various years). County coal tonnage is estimated by using the number gainfully employed in coal mining in 1930 in each county from U.S. Bureau of the Census, Fifteenth Census (1932), and the coal produced in 1929 from U.S. Bureau of the Census (1933a) to determine a figure for tonnage per miner in each county. A ratio of coal miners in 1930 in the county to coal miners in 1930 in the county was then determined. The number of miners in each county for the other years of the 1930s was then determined by multiplying the county/state employment ratio in 1930 by the state coal employment from the U.S. Bureau of Mines (various years) for each year. Then the ratio of coal tonnage per miner was multiplied by the estimated county employment to obtain coal tonnage in each year. The sources and information on coastal location, access to large rivers, and topographical information are described in Fishback, Horrace, and Kantor (2006) and are available online at http://economics.eller.arizona.edu/faculty/fishback.asp under data sets from published research studies. The number of church members is from church membership data from the Census of Religious Bodies, 1926, as reported in U.S. Bureau of the Census (1980).

Disaster Loan Corporation loans per capita

Percent homeowners Public Works Administration federal and nonfederal grants per capita Agricultural Adjustment Administration grants per capita Relief spending per capita by WPA, FERA, CWA, SSA, and FSA grants Public Roads Administration grants per capita

Tax returns per capita Crop values

Auto registrations per capita

8.854 5.157

5.935 2.395

(continued)

0.166

17.256

5.550

0.007

12.575 31.416

0.024 2,060,620

0.158

108.174

51.310 1.758

0.020 1,749,595

0.163

193.469

5.337 22.834 10.656

5.541 48.174 15.081

1929, 1933, 1935, 1939, interpolated using state personal income in between 1930, 1931, 1936, interpolated using state information in between All years 1929, 1939, interpolated using state information on crop value in between 1930, 1940, straight-line interpolation in between County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information

3.158

10.063

Deaths all years, population in 1930, 1940, and straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1929, same value throughout

28.094

SD

56.633

Mean

All years

All years or interpolation procedure

Summary statistics and discussion of nature of data used in panel

Infant mortality rate, number of infant deaths per thousand live births Noninfant mortality rate, number of deaths of people over age one per thousand population Percent illiterate Percent families with radios Circulation of 15 national magazines as of January 1, 1929 per person in 1930 Retail sales per capita

Description

Table 5A.1

(continued)

General fertility rate, births per 1,000 women aged 15–44 Percent of population Aged 5–9

Results of bovine tuberculosis status tests

No. of Civilian Conservation Corps camps started in fiscal year t No. of Civilian Conservation Corps camps started in fiscal year t – 1 No. of Civilian Conservation Corps camps started in fiscal year t – 2 Hospital beds per 1,000 women aged 15–44, hospitals that might help infants No. of people employed in polluting industries of chemicals, cigars and cigarettes, glass, bread, meat packing, autos, iron and steel, nonmetals, planing mills, lumber mills, boots and shoes, printing, paper, cotton textiles, and rubber County coal tonnage in year t

Reconstruction Finance Corporation loans per capita U.S. Housing Authority loans per capita

Farm loans per capita

Description

Table 5A.1

0.153 9.100

All years All years

1930, 1940, straight-line interpolation in between

Coal tonnage based on tonnage/employment ratio in 1930 and then interpolated using state estimates of coal tonnage Annual based on tests in May through July for 1930 through 1937, October in 1938 and 1939, and January 1941 for 1940 Annual birth information divided by trend number of women aged 15–44 interpolated between 1930 and 1940 Census

1930, same value throughout

0.161

All years

10.380

86.013

1.350

170.526

2,325.430

1.869

24.757

0.748

1,479.607

13,224.580

16.968

0.634

0.639

0.650

0.902

0.043 0.173

5.087

7.672

SD

1.466

3.639

Mean

County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information County total for June 1933 through June 1939 distributed using state information All years

All years or interpolation procedure

Retail employment as a percentage of the population Average number of lakes in county Average number of swamps in county Average max elevation in county Average elevation range in counties Percent church members 1926/population 1930 Average number of rivers that pass through 11–20 counties in county, population weight

Aged 10–14 Aged 15–19 Aged 20–24 Aged 25–29 Aged 30–34 Aged 35–44 Aged 45–54 Aged 55–64 Aged 65–74 Aged 75 Percent urban Percent foreign-born Percent African American Population density Percent of families with electricity Manufacturing employment as a percentage of the population

Same value throughout

1930, 1940, straight-line interpolation in between Same value throughout Same value throughout Same value throughout Same value throughout Same value throughout

1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between 1930, 1940, straight-line interpolation in between Manufacturing employment in 1930, 1940, interpolated between years using census of manufacturing county evidence for 1929, 1931, 1933, 1935, 1937, and 1939, and state information on manufacturing employment in between, population is 1930 and 1940 with straight-line interpolation in between

0.241

1.990 21.442 2.417 2,415.793 1,539.714 48.228

10.359 10.074 8.578 7.506 6.754 12.295 10.337 7.319 4.535 1.961 21.856 4.287 10.883 103.603 49.156 4.956

0.452 (continued)

1.282 56.027 8.110 2,989.049 2,382.894 23.451

1.514 1.208 0.993 0.990 0.917 1.554 1.640 1.770 1.472 0.810 24.529 5.271 18.160 780.869 27.816 45.592

(continued)

All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if All years, nearest weather station if

no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county no station in county

Same value throughout Same value throughout Same value throughout Same value throughout Same value throughout Same value throughout Same value throughout 1930, same value throughout All years, nearest weather station if no station in county All years, nearest weather station if no station in county

Same value throughout

All years or interpolation procedure

0.023 0.134 0.198 0.172 0.144 0.119 0.096 0.070 0.028 0.011 0.004 0.001 0.000

0.093 3.107 0.510 0.044 0.014 0.017 0.028 969.230 67.7 35.2

0.136

Mean

0.037 0.081 0.067 0.038 0.040 0.039 0.050 0.058 0.033 0.017 0.009 0.005 0.002

0.296 14.143 3.193 0.205 0.117 0.128 0.165 1,329.276 8.3 15.0

0.372

SD

Note: SD  standard deviation; WPA  Works Progress Administration; FERA  Federal Emergency Relief Administration; CWA  Civil Works Administration; SSA  Social Security Administration; FSA  Farm Security Administration.

Average number of rivers that pass through 21–50 counties in county, population weight Average number of rivers that pass through 51 and over counties in county, population weight Average number of bays in county Average number of beaches in county County on Atlantic Ocean County on Pacific Ocean County on Gulf of Mexico County on Great Lakes 1930 area in square miles Average daily high temperature, Fahrenheit No. of inches of precipitation Percentage of days with high temperature within temperature band High  100 100  high  90 90  high  80 80  high  70 70  high  60 60  high  50 50  high  40 40  high  30 30  high  20 20  high  10 10  high  0 0  high  –10 –10  high

Description

Table 5A.1

The Impact of Climate on Mortality Rates during the Great Depression

163

References Alonso, Wladimir J., Cecile Viboud, Lone Simonsen, Eduardo W. Hirano, Lucaine Z. Daufenbach, and Mark Miller. 2007. Seasonality of influenza in Brazil: A traveling from the Amazon to the subtropics. American Journal of Epidemiology 165: 1434–42. American Medical Association. Various years. Hospitals registered by the American Medical Association. Chicago: American Medical Association. Antonovsky, Aaron, and Judith Bernstein. 1977. Social class and infant mortality. Social Science and Medicine 11:453–70. Beeson, Patricia E., and Werner Troesken. 2006. When bioterrorism was no big deal. NBER Working Paper no. 12636. Cambridge, MA: National Bureau of Economic Research. Bergey, D. H. 1907. The relation of the insect to the transmission of disease. New York Medical Journal 37:1120–25. Bleakley, Hoyt. 2007. Malaria eradication in the Americas: A retrospective analysis of childhood exposure. University of Chicago. Unpublished Working Paper. Chamberlain, Roy W. 1980. History of St. Louis encephalitis. In St. Louis encephalitis, ed. Thomas P. Monath, 3–64. Washington, DC: American Public Health Association. Chan, Yvonne C., and Martin Wiedmann. 2009. Physiology and genetics of listeria monocytogenes and growth at cold temperatures. Critical Reviews in Food Science and Nutrition 49:237–53. Chicago, City of. Various years. Annual report of the Department of Health of the City of Chicago. Chicago: Cameron and Amberg. Civilian Conservation Corps Legacy. n.d. Camp lists by state. http://www.ccclegacy .org/camp_lists.htm. Clemow, Frank G. 1903. The geography of disease. Cambridge, UK: Cambridge University Press. Clifford, William B., and Yevonne S. Brannon. 1978. Socioeconomic differentials in infant mortality: An analysis over time. Review of Public Use Data 6:29–37. Creighton, Charles. [1894] 1965. A history of epidemics in Britain. 2 vols. London: Frank Cass. Cunfer, Geoff. 2005. On the Great Plains: Agriculture and environment. College Station, TX: Texas A&M University Press. Dehejia, Rajeev, and Adriana Lleras-Muney. 2004. Booms, busts, and babies’ health. Quarterly Journal of Economics 119:1091–1130. Deschênes, Olivier, and Michael Greenstone. 2007. Climate change, mortality, and adaptation: Evidence from annual fluctuations in weather in the U.S. NBER Working Paper no. 13178. Cambridge, MA: National Bureau of Economic Research. Eghian, Setrack G. 1905. Relation of feeding to infant mortality. Pediatrics 12: 154–64. Ferrie, Joseph P., and Werner Troesken. 2008. Water and Chicago’s mortality transition, 1850–1925. Explorations in Economic History 45:1–18. Fishback, Price V., Michael R. Haines, and Shawn Kantor. 2007. Births, deaths, and New Deal relief during the Great Depression. Review of Economics and Statistics 89:1–14. Fishback, Price V., William C. Horrace, and Shawn Kantor. 2005. The impact of New Deal expenditures on local economic activity: An examination of retail sales, 1929–1939. Journal of Economic History 65:36–71.

164

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

———. 2006. The impact of New Deal expenditures on mobility during the Great Depression. Explorations in Economic History 43 (April): 179–222. Fishback, Price V., and Dieter Lauszus. 1989. The quality of services in company towns: Sanitation in coal towns during the 1920s. Journal of Economic History 49:125–44. Fogel, Robert W. 1994. Economic growth, population theory, and physiology: The bearing of long- term processes on the making of economic policy. American Economic Review 84:369–95. Fox, Jonathan. 2009. Public health movements, local poor relief and child mortality in American cities: 1923–1932. University of Arizona, Economics Department, Working Paper. Galloway, P. R. 1985. Annual variations in deaths by age, deaths by cause, prices, and weather in London, 1670–1839. Population Studies 39:487–505. Gardner, John, and William Cohen. 1992. Demographic characteristics of the population of the United States, 1930–1950: County level. Interuniversity Consortium for Political and Social Research File no. 0020. Greenstone, Michael, and Kenneth Chay. 2003. The impact of air pollution on infant mortality: Evidence from geographic variation in pollution shocks induced by a recession. Quarterly Journal of Economics 118:1121–67. Haines, Michael R., Lee A. Craig, and Thomas Weiss. 2003. The short and the dead: Nutrition, mortality, and the “antebellum puzzle” in the United States. Journal of Economic History 63:382–413. Haines, Michael R., and the Interuniversity Consortium for Political and Social Research (ICPSR). 2004. Historical, demographic, economic, and social data: The United States, 1790–2000. ICPSR File no. 2896. Hansen, Zeynep, and Gary Libecap. 2003. Small farms, externalities, and the Dust Bowl of the 1930s. Journal of Political Economy 112:665–94. Herrick, Glenn W. 1903. The relation of malaria to agriculture and other industries of the South. Popular Science Monthly 62:521–25. Hewitt, Charles Gordon. 1912. House- flies and how they spread disease. Ann Arbor, MI: University of Michigan Press. Humphreys, Margaret. 2003. Malaria: Poverty, race, and public health in the United States. Baltimore, MD: Johns Hopkins University Press. Jones, B. L., and J. T. Cookson. 1983. Natural atmospheric microbial conditions in a typical suburban area. Applied and Environmental Microbiology 45:919–34. Kaplan, George A., Elsie R. Pamuk, John W. Lynch, Richard D. Cohen, and Jennifer L. Balfour. 1996. Inequality in income and mortality in the United States: Analysis of mortality and potential pathways. British Medical Journal 312:999–1003. Kennedy, Bruce P., Ichiro Kawachi, and Deborah Prothrow-Stith. 1996. Income distribution and mortality: Cross sectional ecological study of the Robin Hood index in the United States. British Medical Journal 312:1004–7. Komlos, John. 1987. The height and weight of West Point cadets: Dietary change in antebellum America. Journal of Economic History 47:897–927. MacCallum, F. O., and J. R. McDonald. 1957. Survival of variola virus in raw cotton. Bulletin of the World Health Organization 16:247–54. Massachusetts, Commonwealth of. Various years. Annual report on births, marriages, and deaths. Public Document no. 1. Boston: Wright & Potter. Monath, Thomas P. 1980. Epidemiology. In St. Louis encephalitis, ed. Thomas P. Monath, 239–312. Washington, DC: American Public Health Association. Nishiura, Hiroshi, and Tomoko Kashiwagi. 2009. Smallpox and season: Reanalysis of historical data. Interdisciplinary Perspectives on Infectious Diseases 2009: 1–10.

The Impact of Climate on Mortality Rates during the Great Depression

165

Olmstead, Alan L. and Paul W. Rhode. 2007. Not on my farm! Resistance to bovine tuberculosis eradication in the United States. Journal of Economic History 67 (3): 768–809. ———. 2008. Creating abundance: Biological innovation and American agricultural development. New York: Cambridge University Press. Phelps, Edward Bunnell. 1910. A statistical survey of infant mortality’s urgent call for action. An address delivered at the first annual meeting of the American Association for the Study and Prevention of Infant Mortality. 9–11 November, Baltimore. Rand McNally. 1934. Commercial Atlas and Marketing Guide. New York: Rand McNally. ———. 1935. Commercial Atlas and Marketing Guide. New York: Rand McNally. ———. 1938. Commercial Atlas and Marketing Guide. New York: Rand McNally. ———. 1939. Commercial Atlas and Marketing Guide. New York: Rand McNally. ———. 1943. Commercial Atlas and Marketing Guide. New York: Rand McNally. Reichert, Thomas A., Lone Simonsen, Ashutosh Sharma, Scott Pardo, David Fedson, and Mark Miller. 2004. Influenza and the winter increase in mortality in the United States, 1959–1999. American Journal of Epidemiology 160:492–502. Reiter, Paul. 1988. Weather, vector biology, and arboviral recrudescence. In The arboviruses: Epidemiology and ecology. Vol. 1, ed. Thomas P. Monath, 245–55. Boca Raton, FL: CRC Press. Reyes, Jessica Wolpaw. 2002. The impact of lead exposure on crime and health, and an analysis of the market for physicians: Three essays. PhD diss., Harvard University. Rogers, Leonard. 1923. The world incidence of leprosy in relation to meteorological conditions and its bearing on the probable mode of transmission. Transactions of the Royal Society of Tropical Medicine and Hygiene 16:440–64. ———. 1925. Climate and disease incidence in India with special reference to leprosy, phthisis, pneumonia, and smallpox. Journal of State Medicine 33:501–10. ———. 1926. Small- pox and climate in India: Forecasting of epidemics. Medical Research Council Reports 101:2–22. Rosenau, Milton J. 1909. The number of bacteria in milk and the value of bacterial counts. In Milk and its relation to the public health. Treasury Department. Hygienic Laboratory Bulletin no. 56. Washington, DC: GPO. Routh, Charles H. F. 1879. Infant feeding and its influence on life; or the causes and prevention of infant mortality. New York: W. Wood. Ruhm, Christopher J. 2000. Are recessions good for your health? Quarterly Journal of Economics 115:617–50. Sagripanti, Jose-Luis, and C. David Lytle. 2007. Inactivation of influenza virus by solar radiation. Photochemistry and Photobiology 83:1278–82. Sedgwick, William T., and J. S. MacNutt. 1910. On the Mills-Reincke phenomenon and Hazen’s theorem concerning the decrease in mortality from diseases other than typhoid fever following the purification of public water supplies. Journal of Infectious Diseases 7:589–664. Spinzig, Carl T. 1880. Yellow fever; nature and epidemic character caused by meteorological influences: Verified by the epidemics of Shreveport and Memphis in 1873, by that of Savannah in 1876, by the great epidemic of the Mississippi Valley in 1878, and (in the appendix) by the one of Memphis in 1879. St. Louis, MO: G. O. Rumbold. Steckel, Richard. 1992. Stature and living standards in the United States. In American economic growth and standards of living before the Civil War, ed. Robert Gallman and John Wallis, 265–308. Chicago: University of Chicago Press.

166

Fishback, Troesken, Kollmann, Haines, Rhode, and Thomasson

Thomasson, Melissa, and Jaret Treber. 2008. From home to hospital: The evolution of childbirth in the United States, 1928–1940. Explorations in Economic History 45:76–99. Troesken, Werner. 2004. Water, race, and disease. Cambridge, MA: MIT Press. U.S. Bureau of Animal Industry. Various years. Status of bovine tuberculosis eradication on area basis. Washington, DC: GPO. U.S. Bureau of the Census. Various years. Birth, stillbirth and infant mortality statistics for the continental United States. Washington, DC: GPO. ———. Various years. Mortality statistics. Washington, DC: GPO. ———. 1908. Mortality statistics. Washington, DC: GPO. ———. 1932. Fifteenth census of the United States: 1930. Population. Vol. III. Reports by states, parts I and II. Washington, DC: GPO. ———. 1933a. Fifteenth census of the United States: Mines and quarries: 1929. General report and reports for states and for industries. Washington, DC: GPO. ———. 1933b. Manufacturing census: Reports by counties, cities, and industry. Mimeograph. University of Washington’s government documents section of the library. ———. 1934. Mortality statistics, 1930. Thirty- first annual report. Washington, DC: GPO. ———. 1935a. Mortality statistics, 1931. Thirty- second annual report. Washington, DC: GPO. ———. 1935b. Mortality statistics, 1932. Thirty- third annual report. Washington, DC: GPO. ———. 1936a. Mortality statistics, 1933. Thirty- fourth annual report. Washington, DC: GPO. ———. 1936b. Mortality statistics, 1934. Thirty- fifth annual report. Washington, DC: GPO. ———. 1936c. United States census of agriculture, 1935, statistics by counties with state and U.S. summaries. 2 vols. Washington, DC: GPO. ———. 1937a. Manufacturing census: Reports by counties, cities, and industry. Mimeograph. University of Washington’s government documents section of the library. ———. 1937b. Mortality statistics, 1935. Thirty- sixth annual report. Washington, DC: GPO. ———. 1938. Mortality statistics, 1936. Thirty- seventh annual report. Washington, DC: GPO. ———. 1939. Vital statistics of the United States, 1937: Part I. Washington, DC: GPO. ———. 1940. Vital statistics of the United States, 1938: Part I. Washington, DC: GPO. ———. 1948. Illiteracy in the United States, October 1947. Current Population Reports: Population Characteristics Series P- 20 no. 20 (September 22). ———. 1975. Historical statistics of the United States: Colonial times to 1970. Washington, DC: GPO. ———. 1980. Censuses Of religious bodies, 1906–1936 [Computer file]. ed. ICPSR. Ann Arbor, MI: Interuniversity Consortium for Political and Social Research [producer and distributor]. U.S. Bureau of Economic Analysis. 1989. State personal income: 1929–1987. Washington, DC: GPO. U.S. Bureau of Foreign and Domestic Commerce. 1932. General consumer market statistics. Market data handbook of the United States. Supplement no. 1. Washington, DC: GPO.

The Impact of Climate on Mortality Rates during the Great Depression

167

———. 1936. Consumer market data handbook, 1936. Washington, DC: GPO. ———. 1939. Consumer market data handbook, 1939. Washington, DC: GPO. U.S. Bureau of Internal Revenue. 1932. Statistics of income for 1930. Washington, DC: GPO. ———. 1935. Number of individual income tax returns for 1935 by states, counties, cities, and towns. Mimeograph. ———. 1939. Number of individual income tax returns for 1937 by states, counties, cities, and towns. Mimeograph. ———. 1940. Number of individual income tax returns for 1938 by states, counties, cities, and towns. Mimeograph. U.S. Bureau of Mines. Various years. Mineral industries in the United States. Washington, DC: GPO. U.S. National Archives. n.d. Finding aid: Civilian Conservation Corps, Record Group 35. College Park, MD: National Archives. U.S. Office of Government Reports. 1940a. Direct and cooperative loans and expenditures of the federal government for fiscal years 1933 through 1939. Statistical Section Report no. 9. Mimeograph. ———. 1940b. County reports of estimated federal expenditures March 4, 1933– June 30, 1939. Statistical Section Report no. 10. Mimeograph. U.S. Public Roads Administration. 1947. Highway statistics: Summary to 1945. Washington, DC: GPO. Waldmann, Robert J. 1992. Income distribution and infant mortality. Quarterly Journal of Economics 107:1284–1302. Whipple, George C. 1908. Typhoid fever: Its causation, transmission, and prevention. New York: Wiley. World Health Organization (WHO). 2004. Using climate to predict infectious disease outbreaks: A review. Geneva, Switzerland: World Health Organization. Yusuf, S., G. Piedmonte, A. Auais, S. Krishnan, P. Van Caeseele, R. Singleton, S. Broor, et al. 2007. The relationship of meteorological conditions to the epidemic activity of respiratory syncytial virus. Epidemiology and Infection 135:1077–95. Zinsser, Hans, Wolfgang K. Joklik, Hilda P. Willet, and Dennis Bernard Amos. 1980. Microbiology. New York: Appleton-Century-Crofts.

6 Responding to Climatic Challenges Lessons from U.S. Agricultural Development Alan L. Olmstead and Paul W. Rhode

Prominent climate researchers project that by the end of the twenty- first century, temperatures on the North American continent will be 4 to 6°F higher at its coasts and 9°F higher at the more northern latitudes.1 Sea levels may rise between 0.5 and 2 feet. Such changes will have profound impacts on economic activity, including agricultural production. Researchers at the International Maize and Wheat Improvement Center anticipate North America wheat farmers will have to cease production at the southern end of the grain belt but may be able extend cultivation 600 to 700 miles northAlan L. Olmstead is the Distinguished Research Professor of Economics and a member of the Giannini Foundation of Agricultural Economics at the University of California, Davis, and serves on the board of directors of the National Bureau of Economic Research. Paul W. Rhode is a professor of economics at the University of Michigan, and a research associate of the National Bureau of Economic Research. We would like to acknowledge the participants at the National Bureau of Economic Research (NBER) Climate Change Conference; the World Economic History Congress in Utrecht, the Netherlands; the economic history colloquia at the University of Oxford and the University of Copenhagen; the Department of Economics at the University of Guelph; the Department of Agricultural and Resource Economics at the University of California, Berkeley; the Department of Economics at Duke University; the 2010 Allied Social Science Association Meetings; the Western Economics Association International Conference; and the Conference on Environmental Change, Agricultural Sustainability, and Economic Development in the Mekong Delta of Vietnam at Can Tho University. Lee Craig, Michael Haines, Thomas Weiss, Gary Libecap, Richard Steckel, Price Fishback, and Wolfram Schlenker supplied data and/or comments. This chapter builds on our analysis in Creating Abundance: Biological Innovation and American Agricultural Development (New York: Cambridge University Press, 2008). This research was supported, in part, by NSF SES- 0550913, SES- 0551130, and SES- 0921732. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the National Science Foundation. 1. See Field et al. (2007, 627). More recent research suggests that the climate changes may be much greater (Sokolov et al. 2009).

169

170

Alan L. Olmstead and Paul W. Rhode

ward from the current northern limit of production. Alaska is projected to become a wheat growing region.2 Such projections necessarily must account for future technological changes, and this is an iffy proposition. This chapter seeks to provide long- run perspective for understanding future adjustments to variation in climatic conditions. Drawing on the record from the past two centuries, we analyze how American farmers learned to produce in unfamiliar and challenging environments. We do not explicitly examine the responses to fluctuations over time in the climate at a set of fixed locations. Instead, we seek insight by investigating the behavior of settlers moving climate- sensitive production activities to new locations, locations with significantly harsher, drier, and more variable environments. These changes for the most part occurred before a modern understanding of plant genetics informed breeding activities. Our evidence says nothing directly about the ability of future farmers aided by rapid advances in plant sciences to respond to climatic changes, but the historical adjustment process does indicate that the malleability of the agricultural enterprise rendered obsolete the predictions of many past experts. In the mid- nineteenth century, John Klippart of the Ohio State Board of Agriculture was arguably the most informed individual in the United States on wheat culture. In 1858, he published a 700- page tome detailing much of what was then known about the wheat plant and wheat farming around the world. For the age, this was a remarkable piece of scholarship. In his view, agro- climatic conditions limited the permanent commercial wheat belt to the region between the 33rd and 43rd latitudes encompassing Ohio, the southern parts of Michigan and New York, Pennsylvania, Maryland, Delaware, and Virginia. The soils in the latter three states had been largely exhausted and, without considerable investment in fertilizer, production would soon decline. Klippart was aware of the large increase in output to the west of Ohio, but he maintained that the soils and climates of Illinois, Iowa, and Wisconsin would doom those states to the haphazard production of low-quality and low- yielding spring wheat. The region beyond the 98th parallel stretching from Lake Winnipeg through eastern Nebraska to Gulf of Mexico was mostly “an unproductive desert.” Rust infestations would forever limit production in the South. Unless the country husbanded its resources, it would soon be an importer of wheat.3 Figure 6.1 maps of Klippart’s vision of the potential long- term wheat- producing area of the United States. Klippart was so far off the mark because he failed to 2. See Ortiz et al. (2008). 3. See Klippart (1860). Because of a perceived deterioration in productivity, Klippart (1860) argued that “Canada may be left out of the wheat region” (323). Lest one think that Klippart was simply an isolated alarmist, note that Genesee Farmer debated at length “Shall We Have to Abandon Wheat Growing in Western New York?” after the arrival of the highly- destructive wheat midge (63 [2]: 41–43).

Responding to Climatic Challenges

Fig. 6.1

171

The “potential wheat-producing area” in the United States in 1858

Source: Compiled from Klippart (1860).

anticipate the biological innovations that would transform North American wheat production.4 Agricultural production is location specific, at the mercy of conditions that differed across regions and even neighboring farms. Settlement was intrinsically a biological process that required farmers to harmonize production practices with specific local soil and climatic conditions. Learning did not end when the first settlers gained an agricultural foothold because, as areas matured, farmers generally switched to more intensive production patterns requiring new rounds of experimentation. The movement of production into more arid regions with more variable climates was one of the hallmarks of American agricultural development. Biological innovation was a necessary condition for this expansion. Some of 4. Klippart was one of many prominent observers who predicted impending crises in grain production. Among the most prominent was Sir William Crookes, whose prophecies of starvation in his presidential address to the British Association for the Advancement of Science in 1898 received wide currency in the popular and scientific press. For two early twentieth century views of the land suitable for wheat, see Unstead (1912) and Baker (1928, especially 402).

172

Alan L. Olmstead and Paul W. Rhode

America’s most distinguished historians, including Fredrick Jackson Turner, Walter Prescott Webb, and their many disciples, explored the broader causes and consequences of the westward movement of agriculture. Our quantitative analysis provides a better perspective on the magnitude of the challenges that farmers confronted and offers a hint as to the flexibility of farmers to respond to future challenges. In this chapter, we analyze the changing location and climatic conditions faced by the producers of America’s three great nineteenth century staples—wheat, corn, and cotton. 6.1

Wheat

From 1839 to 2002, U.S. wheat production increased nearly nineteen times, rising from roughly 85 million to 1.6 billion bushels. By 1929, the geographical center of U.S. wheat output shifted nearly 1,000 miles from near Wheeling, (West) Virginia to the Iowa/Nebraska borderlands.5 But even more impressive than these changes in geographic center of wheat production were the shifts in the ranges of growing conditions. According to Mark Alfred Carleton, a prominent U.S. Department of Agriculture (USDA) agronomist, the regions of North America producing wheat in the early twentieth century (see figure 6.2) were as “different from each other as though they lay in different continents.”6 Table 6.1 displays the main features of the changing geographic distribution of the U.S. wheat crop across latitudes, longitudes, elevation, annual mean temperature and precipitation, and January and July mean temperature for six selected years—1839, 1869, 1899, 1929, 1969, and 2002. The series combine county- level production data from the Census of Agriculture with fixed characteristics for each county.7 For example, the climatic variables reflect average conditions in each county recorded over the 1941 to 1970 period by the National Oceanic and Atmospheric Administration.8 5. We calculated the center from Census county- level production data and the location of the county’s population centroid. The data include only U.S. production. As a result, the changes do not capture the spread of grain cultivation onto the Canadian prairies. 6. See Carleton (1900, 9). 7. The Interuniversity Consortium for Political and Social Research’s (ICPSR). Historical Demographic, Economic, and Social Data, 1790–2000, ICPSR 2896, linked to county characteristics from the U.S. Department of Health and Human Services, the Health Resources and Services Administration, and the Bureau of Health Professions Resource File, ICPSR 9075. We owe a large debt to Lee Craig, Michael Haines, and Thomas Weiss for making available machine- readable crop data for 1839 to 1909. See Craig, Haines, and Weiss (2000). The information for 1969 to 2002 comes from machine- readable files from the Census of Agriculture compiled and made readily accessible by Michael Haines. We have entered the 1929 data from the U.S. Bureau of the Census (1932). 8. See ICSPR- no. 9075 Codebook, 96. The available series include mean temperature (January, July, annual) and mean precipitation (January, July, annual), among other information. This study notes “Counties with more than one weather station include data for the station closest to the county’s population center(s). For those counties not having a weather station,

Responding to Climatic Challenges

Fig. 6.2

173

Wheat regions

Source: Olmstead and Rhode (2008).

These variables do not capture year- to- year changes in the weather and predate the more recent secular climate changes associated with the global warming. The top panel of table 6.1 shows the distribution of wheat production by latitude. It indicates the median is relatively constant hovering between roughly 40 and 41 degrees, but the most northern one- quarter of production (see the 75 percent line) moved nearly 6 degrees, or over 400 miles between 1839 and 2002. Over 80 percent of this movement took place by 1929, well before the Green Revolution. The next panel on longitude indicates that the median location of production shifted by more than 18 degrees (970 miles) between 1839 and 2002. The rapid movement in the most westward fringe of wheat production (track the 30 degree longitude shift in the 90 percent row) before 1899 captures the rapid expansion in the Pacific region. The changes in the median annual and January temperatures were small. But the range of temperature conditions greatly widened, with production moving into both hotter and colder areas. The movement into more frigid zones was most pronounced. Between 1839 and 2002, the average annual temperature of the coldest 10 percent of production (the 10 percent line) dropped 7.6°F; and the January temperature, the coldest 10 percent fell the U.S. Weather Bureau’s climate regions were used to extrapolate data from other similar climatic areas.”

Wettest

10 25 50 75 90

33.2 36.0 39.0 42.4 47.0

291 482 705 920 1,145

76.02 77.53 80.65 84.07 87.02

10 25 50 75 90

10 25 50 75 90

35.19 38.46 39.95 41.16 42.94

1839

10 25 50 75 90

Percent

Distribution of U.S. wheat production

Highest Annual precipitation (inches) Driest

West Elevation (feet) Lowest

North Longitude (degrees) East

Latitude (degrees) South

Table 6.1

28.2 31.4 35.6 39.7 43.1

291 554 770 940 1,150

77.46 83.20 87.89 91.88 94.59

37.73 39.12 40.56 42.57 43.77

1869

18.0 22.0 29.2 37.6 42.6

390 700 992 1,330 1,778

81.49 86.84 95.69 97.93 117.25

36.84 38.79 41.17 44.82 46.88

1899

13.9 17.4 22.0 31.3 38.7

603 983 1,630 2,620 3,650

84.23 96.06 99.33 102.54 116.91

36.39 38.17 40.50 45.76 47.65

1929

13.3 17.2 20.8 30.3 38.4

620 1,014 1,610 2,460 3,537

87.39 97.14 99.36 102.97 116.89

35.94 37.84 40.31 46.74 48.18

1969

11.8 16.7 21.0 31.3 42.0

392 883 1,482 2,284 3,537

86.04 96.53 99.00 104.78 117.25

35.19 37.60 40.84 46.86 48.21

2002

Sources: See text.

Hottest

Hottest July temperature (°F) Coldest

Hottest January temperature (°F) Coldest

Annual temperature (°F) Coldest 47.8 49.7 52.6 55.3 58.7 23.6 26.3 30.1 33.9 38.5 70.5 71.8 74.1 76.2 78.1

10 25 50 75 90

10 25 50 75 90

10 25 50 75 90

70.5 71.8 73.8 76.2 78.1

16.3 21.2 27.1 31.6 38.5

45.5 47.9 51.0 54.5 57.8

69.2 71.2 73.4 77.0 79.8

7.5 13.1 26.8 32.6 38.7

41.0 44.7 50.4 55.5 59.8

68.3 70.7 75.1 78.7 80.6

8.3 21.5 27.8 32.0 36.0

40.9 46.2 51.6 55.8 57.9

68.1 70.1 75.6 79.3 81.7

6.6 18.3 27.8 32.8 37.2

40.1 44.3 51.8 56.1 59.9

67.5 70.0 74.8 79.5 82.1

5.9 16.7 27.8 33.8 40.5

40.2 43.7 51.3 56.9 61.2

176

Alan L. Olmstead and Paul W. Rhode

17.7°F. At the other extreme, the dividing line for the annual average temperature demarking the warmest 10 percent of production shifted 2.5°F. Between 1839 and 2002, the median elevation of production increased by 777 feet. The most pronounced changes occurred in the distribution of production by annual precipitation. By 2002, median precipitation had fallen by 46 percent since 1839, and most production took place in a drier environment than virtually anything recorded at the earlier date. Most of the changes in almost all of the geographic and climatic variables occurred before 1929. As wheat culture moved westward, settlers encountered climatic conditions far different from those prevailing in the eastern states or in Western Europe. This was especially true as farmers moved onto the Great Plains, which had long been considered the “Great American Desert.”9 Though the region was arid, it was not technically a desert. Still it “was long considered to be incapable of agricultural development. Gradually, however, farmers began to displace cattlemen, and by experimentation, attempted to establish a crop system.”10 The first waves of settlers moved into the High Plains during the relatively wet years of the 1880s. The efforts of these farmers, who emigrated mostly from the humid East, to cultivate the soils of the Plains without irrigation constituted: . . . an experiment in agriculture on a vast scale, conducted systematically and with great energy, though in ignorance or disregard of the fairly abundant data, indicating desert conditions, which up to that time the Weather Bureau had collected. Though persisted in for several years with great determination, it nevertheless ended in total failure.11 The successful spread of wheat cultivation across the vast tracts extending from the Texas Panhandle to the Canadian prairies was dependent on the introduction of hard red winter and hard red spring wheats that were entirely new to North America. Over the late nineteenth century, the premier hard spring wheat cultivated in North America was Red Fife (which appears identical to a variety known as Galician in Europe). According to the most widely accepted account, David and Jane Fife of Otonabee, Ontario selected and increased the grain stock from a single wheat plant grown on their farm in 1842. The original seed was included in a sample of winter wheat shipped from Danzig via Glasgow. It was not introduced into the United States until the mid- 1850s. Red Fife was the first hard spring wheat grown in North America and became the basis for the spread of the wheat frontier into Wisconsin, Minnesota, the Dakotas, and Canada. It also provided much of the parent stock for later wheat innovations, including Marquis. At the time of the first reliable USDA survey of wheat varieties in 1919, farmers 9. See Frazier (1989, 8–9) and Webb (1959). 10. See Goodrich et al. (1936, 207) and Olmstead and Rhode (1989, 24–25). 11. See Johnson (1901, 681).

Responding to Climatic Challenges

177

in North Dakota, South Dakota, and Minnesota grew hard red spring and durum wheats to the virtual exclusion of all others.12 Another notable breakthrough was the introduction of “Turkey” wheat, a hard red winter variety suited to Kansas, Nebraska, Oklahoma, and the surrounding region. The standard account credits German Mennonites, who migrated to the Great Plains from southern Russia, with the introduction of this strain in 1873.13 James Malin’s careful treatment describes the long process of adaptation and experimentation, with the new varieties gaining widespread acceptance only in the 1890s. In 1919, Turkey- type wheat made up about “83 percent of the wheat acreage in Nebraska, 82 percent in Kansas, 67 percent in Colorado, 69 percent in Oklahoma, and 34 percent in Texas. It . . . made up 30 percent of total wheat acreage and 99 percent of the hard winter wheat acreage in the U.S.”14 A similar story holds for the Pacific coast. The main varieties that would gain acceptance in California and the Pacific Northwest differed in nature and origin (Chile, Spain, and Australia) from those cultivated in the humid East in 1839. As a rule, breeders and farmers were looking for varieties that improved yields, were more resistant to lodging and plant enemies, and as the wheat belt pushed westward and northward, varieties that were more tolerant of heat and drought and less subject to winterkill. Canadian experiment station data and other sources show that changes in cultural methods and varieties shortened the ripening period by about twelve days between 1885 and 1910. Given the region’s harsh and variable climate, this was often the difference between success and failure.15 The general progression in varieties allowed the North American wheat belt to push hundreds of miles northward and westward and significantly reduced the risks of crop damage everywhere. One of the most important of the early twentieth century innovations was Marquis, a cross of Red Fife with Red Calcutta, bred in Canada by Charles Saunders. The USDA introduced and tested Marquis seed in 1912 to 1913. By 1916, Marquis was the leading variety in the northern grain belt, and by 1919, its range stretched from Washington to northern Illinois.16 The spread of Marquis was not an isolated case. Following extensive expeditions on the Russian plains, Carleton introduced Kubanka and several other durum varieties in 1900.17 These hardy spring wheats proved relatively 12. See Olmstead and Rhode (2008, 26–27). 13. Although the Mennonites were the most notable group of immigrants to bring new seed varieties to the United States, the practice must have been fairly common, especially in the early years of settlement. We have not seen evidence that would indicate that migrants were more receptive to new varieties released by experiment stations. See Ball (1930, 63). 14. See Quisenberry and Reitz (1974) and Malin (1944). 15. See Norrie (1975) and Ward (1994). Buller (1919, 175–76) credits Marquis with giving adopters about one extra week between harvest and freeze- up (which put an end to fall plowing). 16. See Clark, Martin, and Ball (1922, 901). 17. See Ball and Clark (1918, 3–7) and Clark and Martin (1925, 8–9).

178

Alan L. Olmstead and Paul W. Rhode

rust resistant. By 1903, durum production, which was concentrated in Minnesota and the Dakotas, approached 7 million bushels. In 1904, the region’s Fife and Bluestem crops succumbed to a rust epidemic with an estimated loss of 25 to 40 million bushels, but the durum crop was unaffected. By 1906, durum production soared to 50 million bushels.18 Varietal change also redefined the hard winter wheat belt. Early settlers in Kansas experimented with scores of soft winter varieties common to the eastern states.19 According to the Kansas State Board of Agriculture, “as long as farming was confined to eastern Kansas these [soft] varieties did fairly well, but when settlement moved westward it was found they would not survive the cold winters and hot, dry summers of the plains.”20 The evidence on winterkill lends credence to this view. Data for four east- central counties for 1885 to 1890 show that over 42 percent of the planted acres were abandoned. For the decade 1911 to 1920, after the adoption of hard winter wheat, the winterkill rate in these counties averaged about 20 percent.21 Mark Carleton also left his imprint on Kansas. In 1900, he introduced Kharkof from Russia. This hard winter wheat adapted well to the cold, dry climate in western and northern Kansas, and by 1914, it accounted for about one- half of the entire Kansas crop.22 Drawing on decades of research, S. C. Salmon, O. R. Mathews, and R. W. Luekel noted that for Kansas “the soft winter varieties then grown yielded no more than two- thirds as much, and the spring wheat no more than onethird or one- half as much, as the TURKEY wheat grown somewhat later.”23 In 1920, Salmon concluded that without these new varieties, “the wheat crop of Kansas today would be no more than half what it is, and the farmers of Nebraska, Montana and Iowa would have no choice but to grow spring wheat” which offered much lower yields.24 In addition to introducing new varieties, western farmers experimented with a range of dry-farming techniques.25 The moisture- conserving techniques involved creating a layer of dust to retain precipitation in the soil. Between 1900 and 1930, dry farming was “responsible for a considerable advance into the semiarid region.” Yet the new methods created problems, too. They quickly destroyed the humus layer and left the soil unprotected

18. See Carleton (1915, 404–8). 19. See Malin (1944, 96–101). 20. See Salmon (1920, 210). 21. See Malin (1944, 156–59). Winterkill rates for 1911 to 1920 are calculated using data from Salmon, Mathews, and Luekel (1953, 6, 78–79). The search for varieties suitable for Kansas echoed the earlier experiences of settlers in other states. In the 1840s pioneer farmers attempted to grow winter wheat on the Wisconsin prairie. Repeated failures due to winterkill eventually forced the adoption of spring varieties. See Hibbard (1904, 125–26). 22. See Carleton (1915, 404–8). 23. See Salmon, Mathews, and Luekel (1953, 14). 24. See Salmon (1920, 211–12). 25. See Hansen and Libecap (2004a, b) and Libecap and Hansen (2002).

Responding to Climatic Challenges

Fig. 6.3

179

Spring wheat as a share of U.S. output and acreage

Sources: USDA Crop Reporter (February 1908, 13); USDA Yearbook 1916 (573); USDA Yearbook 1920 (table 21); USDA Commissioner of Agriculture 1886 (410).

against the wind, leading to disastrous effects during the Dust Bowl droughts of the 1930s. “Even after 40 years of trial, a permanently successful system had not been evolved.”26 Adjustment took time.27 Wherever it is feasible, farmers prefer to grow winter wheat instead of spring wheat. Winter wheat generally offers higher yields and is much less subject to damage from insects and diseases. The problem is that in colder climates, winter wheat suffers high losses to winterkill. The agronomy literature commonly recognizes that the development of more hearty winter varieties that could be grown in harsher climates was a great achievement. Just how much land was affected by this fundamental change in farming practices? County- level data on spring and winter wheat production found in the agricultural censuses of 1869 and 1929 show that over this sixty- year period, winter wheat displaced spring wheat in most of Kansas, Iowa, Nebraska, and significant portions of several other states. This area accounted for almost 30 percent of U.S. wheat output in 1929. Figure 6.3 charts the ratio of spring wheat to total wheat acreage and pro26. See Goodrich (1936, 207, 215). 27. See Hargreaves (1957) and Hargreaves (1993).

180

Alan L. Olmstead and Paul W. Rhode

duction in the United States. It uses the best available data from the USDA. Official revised data segregating the two types of wheat begin in 1909. Earlier unadjusted data from USDA allow us to extend the series back to the 1880s. Note the acreage share of spring wheat is typically greater than its production share, consistent with lower yields per acre for spring wheat relative to winter wheat. The spring wheat shares of acreage and output rose over the late nineteenth century as grain production moved into the northern Great Plains. This exerted a drag on overall wheat yields. The share declined subsequently, due in part to the northward shift in the spring- winter wheat line.28 The low spring wheat shares in the 1930s are related to the Dust Bowl era droughts. As with wheat, farmers pushed the frontiers of corn and cotton production into areas previously thought unsuitable for the crops. Success in overcoming the climatic challenges required new varieties and new farming methods. 6.2

Maize

The location of U.S. corn production shifted dramatically over the late nineteenth and early twentieth centuries. Richard Steckel highlighted the significance of the photoperiodic properties of maize to explain the eastwest pattern of U.S. migration during the nineteenth century.29 His analysis demonstrated the importance of latitude in shaping the spread of corn cultivation. Corn is classified as a short- day plant. Such plants flower after the number of hours of daylight falls below a certain maximum threshold. For corn, the shortening days in the latter part of summer trigger flowering. Long- day plants such as wheat and small grains, by way of contrast, time their flowering to occur after the number of hours of daylight rises above a certain minimum. Steckel further observes that “Long- day or short- day plants that are grown outside their latitude of adaptation mature too early or too late for optimal performance.”30 Steckel quantifies the effect of growing the “right” corn by using historical data from experiment station trials. Between 1888 and 1894, the Illinois Agricultural Experiment Station at Champaign tested a variety of corn seeds adapted “to about 80 different locations” in the Midwest and Northeast. Drawing on these trials, Steckel’s econometric analysis found that the “yields of seeds adapted 250 miles south and 250 miles north were only 62 and 72%, respectively, of the yield of seed adapted to Champaign. Yields of seeds adapted up to 250 miles east were slightly higher than those adapted to Champaign, whereas the yield of seeds adapted 250 miles west was 93% 28. These long- run movements inform the debate between Fisher-Temin, Higgs, and Page over the share of the spring crop in U.S. wheat production. See Fisher and Temin (1970); Higgs (1971, 101–2); Fisher and Temin (1971, 102–3); Page (1974, 110–14); and Fisher and Temin (1974, 114–15). 29. See Steckel (1983). 30. See Steckel (1983, 20).

Responding to Climatic Challenges

181

of the yield of seeds adapted to Champaign.”31 Here is solid evidence of the importance of matching corn varieties to geoclimatic conditions. For corn, north- south variations mattered significantly, but east- west variations (within the range Steckel considers) were relatively minor. Steckel argued that pioneering farmers learned that their seed corn was adapted to the seasonal daylight conditions of their own latitude. When moving to new areas of settlement, they “probably took their own supplies of seed grain.” Thus, they would be disinclined to change latitudes significantly for fear that their seeds would generate substantially lower yields. “Farmers who went too far north or south had poor yields and sent relatively unfavorable reports back to the community from which they left.”32 Although westward settlement occurred across a broad front, for many it involved movement along an east- west line. Table 6.2 replicates the previous exercise by showing the changing distribution of U.S. corn production by location and climatic conditions. Again, it is important to recall the corn crop expanded tremendously after 1839. The crop in 2002 was about twenty- three times larger than in 1839. The panel on longitude captures the movement in corn production—the median location shifted by about 8 degrees between 1839 and 2002. But there was also a shift in median latitude of 3.6 degrees, or roughly 250 miles to the north. In addition, the range of latitudes and climatic conditions where corn was grown widened considerably. The median annual temperature under which corn was grown fell by over 6°F from 56.3° in 1839 to 49.9° in 2002. (This is of the same magnitude but in the opposite direction of the change that the Intergovernmental Panel on Climate Change [IPCC] predicts will occur in the grain growing parts of North America over the next century.) Median annual precipitation fell by 11.5 inches, from 43.9 inches in 1839 to 32.4 inches in 2002. Median elevation rose by 370 feet. As with wheat, the bulk of the changes in location and climate occurred by 1929. This was before the widespread diffusion of hybrid corn. The movement of corn production to drier and colder environments required biological innovation. M. L. Bowman and B. W. Crossley observed in 1908 that “the cultivation of corn has been gradually extended northward in the United States. Today this cereal is grown successfully, where twentyfive years ago its cultivation was impossible.”33 It is possible to identify specific breakthroughs that facilitated the shift of the Corn Belt several hundred miles to the north. Of special significance was the work of Andrew Boss, C. P. Bull, and Willet Hays at the University of Minnesota who developed Yellow Dent Minnesota No. 13 and Yellow Dent Minnesota No. 23: “These varieties had remarkable early ripening properties that reduced the ripening 31. See Steckel (1983, 22). 32. See Steckel (1983, 23). 33. See Bowman and Crossley (1908, 90).

Wettest

10 25 50 75 90

36.9 39.6 43.9 49.4 53.3

103 385 590 811 1,010

76.70 80.37 84.54 87.29 90.12

10 25 50 75 90

10 25 50 75 90

33.31 35.34 37.78 39.54 40.58

1839

10 25 50 75 90

Percent

Distribution of U.S. maize production

Highest Annual precipitation (inches) Driest

West Elevation (feet) Lowest

North Longitude (degrees) East

Latitude (degrees) South

Table 6.2

33.3 35.6 39.0 44.5 50.3

255 482 695 880 1,055

77.62 83.45 87.80 91.06 94.02

34.21 37.20 39.51 40.79 42.02

1869

27.9 32.0 36.0 40.3 47.3

390 603 819 1,110 1,401

82.64 86.50 90.75 95.36 97.25

34.71 38.06 40.10 41.36 42.63

1899

25.1 29.8 34.8 40.4 48.4

375 620 870 1,200 1,615

83.00 87.31 92.05 95.86 97.86

34.40 38.00 40.48 41.95 43.29

1929

25.6 29.8 34.5 37.8 42.3

464 640 833 1,160 1,494

83.47 87.36 90.47 94.92 97.25

38.13 39.80 41.02 42.41 43.66

1969

22.1 27.0 32.4 36.5 40.5

520 689 960 1,250 1,841

85.70 88.92 92.68 95.97 98.73

37.79 40.04 41.38 43.07 44.42

2002

Sources: See text.

Hottest

Hottest July temperature (°F) Coldest

Hottest January temperature (°F) Coldest

Annual temperature (°F) Coldest 50.9 53.3 56.3 60.0 63.3 26.7 30.1 35.1 40.6 45.4 73.0 75.1 77.1 79.0 80.4

10 25 50 75 90

10 25 50 75 90

10 25 50 75 90

72.3 74.2 76.0 78.2 80.1

21.5 25.6 29.6 35.7 43.0

48.7 50.9 53.6 57.3 61.9

72.4 74.1 75.9 78.2 80.7

18.8 22.6 26.9 32.9 41.7

47.7 50.0 52.5 56.1 61.4

72.4 73.9 75.5 77.8 80.6

16.3 20.4 26.2 33.4 42.4

46.4 48.9 51.5 56.1 61.8

71.9 73.0 74.8 76.3 78.0

14.7 19.2 23.9 28.0 33.6

45.4 48.1 50.7 52.8 56.2

71.4 72.6 74.4 76.3 78.2

12.9 17.0 22.7 27.4 33.2

44.5 46.7 49.9 52.5 56.1

184

Alan L. Olmstead and Paul W. Rhode

time from 120 to 125 days to about 90 days (for No. 23). These and other early ripening varieties also allowed farmers in the Canadian plains to grow corn for ensilage.”34 According to Andrew Boss and George Pond, “the development of earlymaturing varieties of corn combined with adapted hybrid varieties, and improved cultural practices are steadily drawing the Corn Belt northward and westward into the Spring Wheat area. Accompanying this movement has been a steady increase in cattle and hog production in the area which furnished the chief outlet for the corn crop.”35 Minnesota No. 13 was a potent factor in pushing corn grown for grain fifty miles northward in a single decade.36 Between 1869 and 1929, the corn- wheat frontier moved about 400 miles at some longitudes. An enormous area including most of Minnesota, South Dakota, and Nebraska shifted from the wheat belt to the Corn Belt; these shifts would not have been possible without researchers developing earlier maturing varieties. 6.3

Cotton

Cotton, the country’s third major nineteenth century staple crop, also required extensive adaptation as its culture spread across the American South and Southwest. According to J. O. Ware, a leading USDA cotton expert, the varieties that became the basis for the South’s development were a distinctly “Dixie product”: “Although the stocks of the species were brought from elsewhere, new types, through [a] series of adaptational changes, formed this distinctive group the final characteristics of which are a product of the cotton belt of the United States.”37 This process of molding cotton was repeated over and over again as new varieties were introduced and as production moved into new areas. According to Ware, “The vast differences in climate and soil that obtain over the Cotton Belt undoubtedly brought about a kind of natural selection which eliminated many of the kinds that were tried, while others became adapted to the several conditions under which they were grown and selected over a period of years.”38 Cottons cultivated in the United States belong to one of two species. Sea Island (G. barbadense) was grown primarily along the coasts and on the offshore islands of Georgia, South Carolina, and Florida. Sea Island produced high quality, long staple fibers (over 1 ¼ inches), but it was low yielding and difficult to pick. Cottons of the second and more important 34. See Buller (1919, 187–90). 35. See Boss and Pond (1951, 65) and Troyer and Hendrickson (2007, 905–14). 36. See Troyer (2004, 176) and Hays (1904, 19, 82). Minnesota No. 13 was selected over several years from local seed purchased in 1893. It was first released in 1897. A number of even earlier Dents were subsequently developed at experiment stations in Minnesota, the Dakotas, and Montana. See Will (1930, 65, 85–88, 147). 37. See Ware (1951, 1). 38. See Ware (1936, 659) and Handy (1896).

Responding to Climatic Challenges

185

species (G. hirsutum) were commonly referred to as upland cottons because they were grown in the more variable climates away from the coast. As of the turn of the twentieth century, cotton experts grouped the upland varieties into eight general types. Most of these types could be developed to fit specific environmental and economic situations and would be ill suited for other conditions. None of these cottons were native to British North America. Adaptation was essential for the successful cultivation of upland cotton. In its native environment in Central America, G. hirsutum was a frostintolerant, perennial shrub with short- day photoperiod response. As a short- day plant, its flowering was triggered when the nights began to grow longer and cooler in the late summer or autumn. This strategy was adapted to a semitropic, semiarid environment where the rains came in the autumn. The greater variation in day length over the seasons at the higher latitudes of the American South meant that the date with the right conditions to trigger flowering occurred later in the year. This meant that many of the introduced cotton varieties either did not mature before the first frost set in or did not flower at all. Initial attempts to grow upland cotton in the areas that now constitute the United States faced severe challenges. Success depended on finding a mutation/cross or a variety with the appropriate photosensitivity characteristics. “Following generations of repeated selection, these initial stocks were molded into early maturing, photoperiod- insensitive cultivars adapted for production in the southern United States Cotton Belt.”39 Adaptation was made easier because, as John Poehlman and David Sleper note, the cotton stocks first introduced to the region “were largely mixed populations with varying amounts of cross- pollination and heterozygosity that gave them plasticity and potential for genetic change.”40 Cotton breeders confront a number of trade- offs because improving one plant characteristic often requires sacrificing another desirable quality. Breeders strive for high yields, long staple lengths, soft and strong fibers, good spinning characteristics, ease of picking, high lint- to- seed ratios, whiteness, and more. In addition, breeders work to develop cotton varieties to match local soil and climatic conditions (especially the length of the growing season), to resist specific diseases and pests, survive high winds, and to appeal to special market niches.41 The importance of wind resistance became 39. See Poehlman and Sleper (1995, 376) and Stephens (1975). 40. See Poehlman and Sleper (1995, 376). They further note that “the adjustments were hastened by the contributions of large numbers of early cotton breeders who worked without the genetic guidelines available to cotton- breeders today.” 41. Cotton’s intolerance to cold limited the geographic extent of its cultivation. A freeze (below 32°F) will kill the tissues of this subtropical plant, and even temperatures below 60°F will inhibit growth. Seven months (or, more precisely, 200 days) of frost- free weather or a latitude of 37 degrees is generally considered to set the northern limit to production in the United States. See Hake and Kerby (1996, 325). Nonetheless, there were pockets of production in selected areas of Kansas, Missouri, Illinois, Kentucky, and Virginia above the 37th parallel. See Hart (1977, 308) and U.S. Department of Agriculture (1924, 153).

186

Alan L. Olmstead and Paul W. Rhode

more significant as cotton cultivation moved onto the Texas plains, and the incentive to develop cotton that could be picked more rapidly increased as wages rose. In the antebellum period, the South developed and grew three main “types” of upland cotton: the Petit Gulf or long- limbed cottons, which were late maturing, spreading plants producing long staple fibers, and best suited for fertile lands; the cluster cottons, based on Sugar Loaf (1843) and Boyd’s Prolific (1847), which were earlier, more compact plants producing shorter staple lint; and semicluster cottons, another variant of Boyd’s Prolific with a more moderate tendency for the bolls to cluster. The 1870s saw the development of two additional types—Peterkin and Eastern Big Boll. Three more types gained prominence over the late nineteenth century—Early or King, Long Staple or Allen, and Western Big Boll. The types known as Western Big Boll, Stormproof, and Texas Big Boll cotton were noted for two characteristics. They were resistant to shedding or breaking in high winds, and they were relatively easy to pick because of their large bolls. Whereas the Eastern Big Boll cottons likely evolved from a Mexican variety imported in the 1850s by a Georgia planter named Wyche via Algeria, Texas Big Boll cotton likely evolved out of varieties imported directly from the dry plains of northern Mexico. The process of selection was similar in both regions. “Under the conditions of the great climatic change, pronounced environmental shock was effective in breaking up or isolating favorable responding genotypes. These better balanced and, therefore, more fruitful forms were readily recognized by growers who would save the seed from them. In this way desirable plant habit having the necessary production characteristics for the new adaptation or ecological area in question was established.”42 Mexican stocks imported into different parts of the South thus took on different characteristics—presumably due to different origins, but also due to breeding to fit local environmental conditions. The first Texas Stormproof variety of note was called Supak or Bohemian in honor of the German immigrant who developed the variety around 1860. Probable derivatives of this variety were Meyer and Texas Stormproof. These three varieties gained wide acceptance in Texas, and Texas Stormproof was distributed extensively across the South. In addition, these varieties provided the germplasm for breeders such as W. L. Boykin and A. D. Mebane who developed improved Western Big Boll lines. In 1869, Boykin commenced a decade- long program of carefully selecting Meyer seed from the best plants on his farm near Terrill, Texas. Around 1880, he began planting his improved Meyer amongst Moon, a long staple variety, in a quest for a favorable hybrid. To breed storm resistant cotton, Boykin attached a string with a one pound weight to the 42. See Ware (1951, 83).

Responding to Climatic Challenges

187

tip of the locks and then held up the boll by the slender stock holding the fruit. He only selected seeds from bolls with stocks that didn’t break under the pressure. Boykin’s cotton was similar in appearance to Meyer, easy to pick, and exceptionally storm resistant. It had a high seed- to- lint ratio with a lint length of greater than one inch. Mebane began studying cotton near Lockhart, Texas in the mid- 1870s. Over the next quarter century, he bred cotton in pursuit of a number of characteristics, including storm and drought resistance, higher lint ratios and yields, and larger easy- to- pick bolls. He succeeded in most of these areas and in the process changed his cotton’s appearance, creating a stocky and compact plant that would not whip around in the wind. The high cotton so prized in the Mississippi Delta was a detriment in the windswept plains. When the boll weevil entered Texas, Mebane’s variety became especially important because it was early to mature. Its success in weevil- infested areas led Seaman A. Knapp to name it “Triumph.” Breeders created many other Western Big Boll varieties in the pre-World War II era. Much of this effort focused on satisfying the critical need for early varieties.43 The early twentieth century was a challenging period for cotton producers. The boll weevil, which entered the country around 1892, spread across the traditional Cotton South as a “wave of evil.”44 The pest invasion caused a wholesale transition in the traditional cotton belt to earlier maturing cottons. Among the additional consequences were the push of cotton culture onto the High Plains of the Texas and Oklahoma and the introduction of the crop into New Mexico, Arizona, and California. These environments were far drier and hotter than those found in the traditional Cotton South. Adaption involved finding new cotton varieties and developing new growing practices. In the early twentieth century, USDA scientists scoured Mexico and Guatemala for new varieties. In June 1906, O. F. Cook stumbled upon a single plant growing by the roadside in eastern Chiapas that had a longer and denser fiber than any Big Boll cottons then grown in the United States. Cook deduced from observing local cottons and interviewing farmers that the prized cotton was not cultivated in the immediate area, but because of heavy rains, failed to discover its exact origins. Adding to the problem, the seeds that Cook had extracted from his one plant rotted due to the humid weather. In December 1906, G. N. Collins and C. B. Doyle carried on the quest to find the home of Cook’s chance discovery. They tracked the probable home to the village of Acala, where they acquired seeds for trials in the United States. Beginning in 1907, the laborious process of planting and repeatedly selecting the best plants ensued. Throughout this process, breeders worked to adapt strains for specific areas. Although growers across the West adopted 43. See Olmstead and Rhode (2008, chapters 4–5). 44. See Lange, Olmstead, and Rhode (2009).

188

Alan L. Olmstead and Paul W. Rhode

Acala varieties, the type had its greatest impact in California where it became the only variety planted on any scale for over forty years.45 In addition to adapting plants to meet new environments, farmers changed cultural practices as they pushed into new areas—the time of planting and harvesting, plowing and cultivating methods, and fallowing schemes were all subjects of experimentation. Irrigation was especially important for the movement of cotton into the Southwest and California. In the traditional Cotton Belt, irrigation was a rarity; in the new regions, it was nearly universal. At first, irrigation typically involved individual farmers or small collectives diverting rivers with ditches or employing pumps to tap into underground aquifers. These local efforts were supplemented by massive social overhead investments to dam large rivers and move water long distances via canals. These were all methods used to deal with climate differences as agriculturists adapted to new environments. The responses to these new challenges are evident in table 6.3, which displays the changing distribution of cotton production by location and climatic conditions from 1839 to 2002. In the early decades of twentieth century, a significant fraction of production moved to more western lands, with hotter annual and July temperatures, lower precipitation, and higher elevations. Between 1839 and 2002, median precipitation fell by 6.4 inches; median production in 2002 occurred in areas that would have ranked in the driest 10 percent of production in 1839. The changes in the medians in the other characteristics are less dramatic. But those of the fringe, in the tails of the distribution, show what was possible. The swings were quite dramatic. In the 1960s, 1970s, 1980s, traditional cotton areas in the Southeast temporarily dropped out of production.46 And in the mid- 1970s and the 1980s, output in California doubled. A major feature in table 6.3 is the relative stability of the northern frontier of cotton production; very little production has occurred above the 37th parallel, which has long been the crop’s traditional boundary. With global warming, this barrier may be breached. 6.4

Conclusion

During the nineteenth and twentieth centuries, scientists and agriculturalists created new biological technologies that allowed U.S. farmers to repeatedly push cultivation of the three major staple crops into environments previously thought too arid, too variable, and too harsh to farm. The climatic challenges that these farmers overcame in adopting crops to new areas rivaled the magnitude of the climatic changes predicted for the next hundred years in the United States. 45. See Collings (1926, 206–13), Turner (1981, 40–41), and Ware (1951, 116–27). The dominance of Acala in California was more a function of government policy than of the variety’s superiority. See Constantine, Alston, and Smith (1994). 46. See Hart (1977, 307–22) and Fite (1984).

Wettest

64 130 250 555 700 45.4 48.5 51.8 54.4 57.0

10 25 50 75 90

81.11 83.55 87.63 91.06 91.54

10 25 50 75 90

10 25 50 75 90

30.88 31.69 32.76 33.96 34.98

1839

10 25 50 75 90

Percent

Distribution of U.S. cotton production

Highest Annual precipitation (inches) Driest

West Elevation (feet) Lowest

North Longitude (degrees) East

Latitude (degrees) South

Table 6.3

44.4 47.2 50.2 52.8 55.1

82 175 310 476 655

81.39 84.49 89.08 91.52 94.81

31.04 32.04 32.85 34.29 35.2

1869

34.0 43.6 49.0 52.2 54.4

116 201 360 570 770

81.34 84.78 90.24 95.45 97.12

30.61 31.78 32.95 34.1 35.07

1899

23.4 36.1 47.7 51.4 54.0

117 215 413 745 1,470

81.98 86.52 90.82 96.88 99.54

31.22 32.39 33.6 34.83 35.62

1929

8.1 17.2 40.7 50.1 52.4

107 194 355 1,403 3,254

87.29 90.19 96.35 102.33 118.88

31.42 32.79 33.69 35.2 36.24

1969

10.2 17.7 45.4 50.4 52.3 (continued)

82 160 324 1,405 3,370

83.58 89.68 91.85 102.26 118.88

31.27 32.73 33.96 35.45 36.32

2002

(continued)

Sources: See text.

Hottest

Hottest July temperature (°F) Coldest

Hottest January temperature (°F) Coldest

Annual temperature (°F) Coldest

Table 6.3

60.9 62.0 64.1 65.9 66.8 41.2 44.1 46.2 49.2 50.9 78.9 79.5 80.6 81.6 81.9

10 25 50 75 90

10 25 50 75 90

1839

10 25 50 75 90

Percent

79.0 79.8 81.0 81.9 83.0

40.5 43.1 45.9 47.8 49.9

60.8 62.0 64.1 65.6 66.9

1869

79.0 80.0 81.2 83.0 84.5

40.9 43.1 45.4 47.8 50.0

61.0 62.2 64.1 65.8 67.5

1899

78.9 80.0 81.4 83.0 84.6

38.9 40.9 43.8 46.8 49.9

60.0 61.2 63.4 65.2 66.9

1929

79.2 80.1 81.3 83.0 84.8

38.3 40.2 43.7 46.8 51.2

59.7 61.1 62.7 64.9 70.1

1969

78.9 79.7 80.7 82.1 84.3

38.3 39.1 43.7 46.9 51.2

58.9 60.5 62.3 64.9 67.6

2002

Responding to Climatic Challenges

191

The effects of climate change are likely to appear both in gradual terms and in episodic crises such as outbreaks of new pests and in the onset of severe droughts. This chapter bears on the historic responses to the equivalent of gradual changes. The chapter does not address related shocks that are the predicted dire consequences to agriculture of global warming, including the depletion of already stressed aquifers, a worsening of insect and disease problems, an increase in wildfires, and possible atmospheric changes that will adversely affect crops. But the historical record does show that farmers were able to develop technologies to push crop production into areas previously thought unsuitable for agriculture because of the harsh climatic conditions. There is little reason to think that future technological advances and crop substitutions will not partially offset some of the problems created by global warming. Plant scientists offer a mixed view on the prospects of breeding advances to stave off the consequences of global warming. Breeders of annual crops have expressed confidence that they can employ both traditional and transgenic methods to develop varieties of maize and other crops to keep up with the gradual effects of climate change. However, their ability to deal with episodic events such as the introduction of new pests is far more problematical. The adaptation of perennial crops will be more difficult than adapting annuals in part because the breeding cycle is longer and many perennials prosper commercially only in small geoclimatic niches.47 There will be enormous challenges to the agricultural sector associated with impending climate changes. As in the past, public and private research will be crucial in meeting the new environmental realities.

References Baker, Oliver E. 1928. Agricultural regions of North America. Part VI—The Spring Wheat Region. Economic Geography 4 (4): 399–433. Ball, Carleton R. 1930. The history of American wheat improvement. Agricultural History 4 (2): 48–71. Ball, Carleton, R., and J. Allen Clark. 1918. Experiments with durum wheat. USDA Bulletin no. 618. Washington, DC: U.S. Department of Agriculture. Boss, Andrew D., and George A. Pond. 1951. Modern farm management: Principles and practice. Saint Paul, MN: Itasca Press. Bowman, M. L., and B. W. Crossley. 1908. Corn: Growing, judging, breeding, feeding, marketing. Ames, IA: Privately Printed. Buller, A. H. Reginald. 1919. Essays on wheat. New York: Macmillan. Carleton, Mark Alfred. 1900. The basis for the improvement of American wheats. USDA Division of Vegetable Physiology and Pathology Bulletin no. 24. Washington, DC: U.S. Department of Agriculture. 47. See Hest (2008) and Koski (1996, 235–39).

192

Alan L. Olmstead and Paul W. Rhode

———. 1915. Hard wheats winning their way. In USDA Yearbook, 1914, 391–420. Washington, DC: GPO. Clark, J. Allen, and John H. Martin. 1925. Varietal experiments with hard red winter wheats in the dry areas of the western United States. USDA Bulletin no. 1276. Washington, DC: U.S. Department of Agriculture. Clark, J. Allen, John H. Martin, and Carleton R. Ball. 1922. Classification of American wheat varieties. USDA Bulletin no. 1074. Washington, DC: U.S. Department of Agriculture. Collings, Gilbeart H. 1926. The production of cotton. New York: Wiley. Constantine, John H., Julian M. Alston, and Vincent H. Smith. 1994. Economic impacts of the California One-Variety Cotton Law. Journal of Political Economy 102 (October): 951–74. Craig, Lee A., Michael R. Haines, and Thomas Weiss. 2000. Development, health, nutrition, and mortality: The case of the “antebellum puzzle” in the United States. NBER Working Paper Series on Historical Factors in Long-Run Growth, Historical Paper no. 130. Cambridge, MA: National Bureau of Economic Research. Field, C. B., L. D. Mortsch, M. Brklacich, D. L. Forbes, P. Kovacs, J. A. Patz, S. W. Running, and M. J. Scott. 2007. North America. Climate Change 2007: Impacts, adaptation and vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, ed. M. L. Parry, O. F. Canziani, J. P. Palutikof, P. J. van der Linden, and C. E. Hanson, 617–52. Cambridge, UK: Cambridge University Press. Fisher, Franklin M., and Peter Temin. 1970. Regional specialization and the supply of wheat in the United States, 1867–1914. Review of Economics and Statistics 52 (May): 134–49. ———. 1971. Regional specialization and the supply of wheat in the United States, 1867–1914: A reply. Review of Economics and Statistics 53 (February): 102–3. ———. 1974. Regional specialization and the supply of wheat in the United States, 1867–1914: A reply. Review of Economics and Statistics 54 (February): 114–15. Fite, Gilbert C. 1984. Cotton fields no more: Southern agriculture, 1865–1980. Lexington, KY: University Press of Kentucky. Frazier, Ian. 1989. Great Plains. New York: Penguin. Genesee Farmer. 1857. 63 (2): 41–43. Goodrich, Carter, Bushrod W. Allin, C. Warren Thornthwaite, Hermann K. Brunck, Frederick G. Tryon, Daniel B. Creamer, Rupert B. Vance et al. 1936. Migration and economic opportunity: The Report of the Study of Population Redistribution. Philadelphia: University of Pennsylvania Press. Haines, Michael R., and Interuniversity Consortium for Political and Social Research (ICPSR). 2004. Historical, demographic, economic, and social data: The United States, 1790–2002 [Computer file]. ICPSR02896- v3. Ann Arbor, MI: Interuniversity Consortium for Political and Social Research [distributor], 2010- 05- 21. doi:10.3886/ICPSR02896. Hake, K. D., and T. A. Kerby. 1996. Cotton and the environment. In Cotton production manual, ed. S. J. Hake, T. A. Kerby, and K. D. Hake, 324–34. University of California, Division of Agriculture and Natural Resources Publication no. 3352. Oakland, CA: Regents of University of California. Handy, R. B. 1896. History and general statistics of cotton. In The cotton plant: Its history, botany, chemistry, culture, enemies, and uses, ed. Charles W. Dabney, 17–66. USDA Office of Experiment Stations Bulletin no. 33. Washington, DC: GPO. Hansen, Zeynep K. and Gary D. Libecap. 2004a. The allocation of property rights

Responding to Climatic Challenges

193

to land: U.S. land policy and farm failures in the northern Great Plains. Explorations in Economic History 41 (2): 103–29. ———. 2004b. Small farms, externalities, and the Dust Bowl of the 1930s. Journal of Political Economy 112 (3): 665–94. Hargreaves, Mary W. M. 1957. Dry farming in the northern Great Plains, 1900–1925. Cambridge, MA: Harvard University Press. ———. 1993. Dry farming in the Northern Great Plains: Years of readjustment, 1920– 1990. Lawrence, KS: University Press of Kansas. Hart, John Fraser. 1977. The demise of king cotton. Annals of the Association of American Geographers 67 (3): 307–22. Hays, W. M. 1904. Breeding animals and plants. St. Anthony Park, MN: Farm Students’ Review. Hest, David. 2008. Seeds for global warming. Farm Industry News 1 (January): 1. Hibbard, Benjamin Horace. 1904. The history of agriculture in Dane County Wisconsin: A thesis submitted for the degree of doctor of philosophy, University of Wisconsin, 1902. Bulletin of the University of Wisconsin no. 101, Economics and Political Science Series 1 (2): 67–214. Higgs, Robert. 1971. Regional specialization and the supply of wheat in the United States, 1867–1914: A comment. Review of Economics and Statistics 53 (February): 101–2. Johnson, Willard D. 1901. The High Plains and their utilization. In Twenty-First Annual Report of the United States Geological Survey to the Secretary of the Interior, 1899–1900: Part IV, Hydrography. Washington, DC: GPO. Klippart, John H. 1860. In The wheat plant: Its origin, culture, growth, development, composition, varieties, diseases, etc., etc., 296–327. New York: A. O. Moore. Koski, Veikko. 1996. Breeding plans in case of global warming. Euphytica 92 (1–2): 235–39. Lange, Fabian, Alan L. Olmstead, and Paul W. Rhode. 2009. The impact of the boll weevil, 1892–1932. Journal of Economic History 69 (3): 685–718. Libecap, Gary D., and Zeynep K. Hansen. 2002. “Rain follows the plow” and dryfarming doctrine: The climate information problem and homestead failure in the Upper Great Plains, 1890–1925. Journal of Economic History 62 (1): 86–120. Malin, James C. 1944. Winter wheat in the golden belt of Kansas. Lawrence, KS: University of Kansas Press. Norrie, K. H. 1975. The rate of settlement of the Canadian prairies, 1870–1911. Journal of Economic History 35 (2): 410–27. Olmstead Alan L., and Paul W. Rhode. 1989. Regional perspectives on U.S. agricultural development since 1880. University of California, Davis, Agricultural History Center Working Paper no. 43. ———. 2008. Creating abundance: Biological innovation and American agricultural development. New York: Cambridge University Press. Ortiz, Rodomirio, Kenneth D. Sayre, Bram Vovaerts, Raj Gupta, G. V. Subbarao, Tomohiro Ban, David Hodson, John M. Dixon, J. Ivan Ortiz-Monasterio, and Matthew Reynolds. 2008. Climate change: Can wheat beat the heat? Agriculture, Ecosystems and Environment 126 (1–2): 46–58. Page, Walter P. 1974. Wheat culture and productivity trends in wheat production in the United States, 1867–1914; A comment. Review of Economics and Statistics 56 (February): 110–14. Poehlman, John Milton, and David Allen Sleper. 1995. Breeding field crops. 4th ed. Ames, IA: Iowa State University Press. Quisenberry, Karl S., and L. P. Reitz. 1974. Turkey wheat: The cornerstone of an empire. Agricultural History 48 (10): 98–114.

194

Alan L. Olmstead and Paul W. Rhode

Salmon, S. C. 1920. Developing better varieties of wheat for Kansas. In Wheat in Kansas, 210–17. Topeka, KS: Kansas State Board of Agriculture. Salmon, S. C., O. R. Mathews, and R. W. Luekel. 1953. A half century of wheat improvement in the United States. In Advances in agronomy. Vol. 5, ed. A. G. Norman, 1–151. New York: Academic Press. Sokolov, A. P., P. H. Stone, C. E. Forest, R. Prinn, M. C. Sarofim, M. Webster, S. Paltsev, C. A. Schlosser, D. Kicklighter, S. Dutkiewicz, J. Reilly, C. Wang, B. Felzer, J. M. Melillo, and H. D. Jacoby. 2009. Probabilistic forecast for twenty- firstcentury climate based on uncertainties in emissions (without policy) and climate parameters. Journal of Climate 22 (19): 5175–5204. Steckel, Richard. 1983. The economic foundations of East-West migration during the nineteenth century. Explorations in Economic History 20:14–36. Stephens, S. G. 1976. Some observations of photoperiodism and the development of annual forms of domesticated cottons. Economic Botany 30 (4): 409–18. Troyer, A. Forrest. 2004. Persistent and popular germplasm in seventy centuries of corn evolution. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, 133–231. New York: Wiley. Troyer, A. Forrest, and Lois G. Hendrickson. 2007. Background and importance of “Minnesota 13” corn. Crop Science 47 (May): 905–14. Turner, John. 1981. White gold comes to California. Bakersfield, CA: California Planting Cotton Seed Distributors. Unstead, J. F. 1912. The climatic limits of wheat cultivation, with special reference to North America. Geographical Journal 39:347–66 and 39:421–41. U.S. Bureau of the Census. 1932. Fifteenth Census of the United States: 1930. Agriculture. Vol. II, Reports by states, with statistics for counties and a summary for the United States. Pt. 1, The Northern States, Pt. 2, The Southern States, Pt. 3, The Western States. Washington, DC: GPO. U.S. Department of Agriculture. 1924. Farthest north of the cotton crop. Crops and Markets 1 (May): 153. U.S. Department of Health and Human Services, Health Resources and Services Administration, Bureau of Health Professions. Bureau of Health Professions Area Resource File, 1940–1990: [United States], [Computer file]. 2nd ICPSR release. Rockville, MD: U.S. Department of Health and Human Services, Office of Data Analysis and Management [producer], 1991. Ann Arbor, MI: Interuniversity Consortium for Political and Social Research [distributor], 1994. doi:10.3886/ ICPSR09075. Ward, Tony. 1994. The origins of the Canadian wheat boom, 1880–1910. Canadian Journal of Economics 27 (4): 864–83. Ware, Jacob Osborn. 1936. Plant breeding and the cotton industry. In Yearbook of Agriculture, 1936, 657–744. Washington, DC: GPO. ———. 1951. Origin, rise and development of American upland cotton varieties and their status at present. University of Arkansas College of Agriculture, Agricultural Experiment Station. Mimeograph. Webb, Walter Prescott. 1959. The Great Plains. Boston: Ginn. Will, George W. 1930. Corn for the Northwest. St. Paul, MN: Webb.

7 The Impact of the 1936 Corn Belt Drought on American Farmers’ Adoption of Hybrid Corn Richard Sutch

A severe and sustained drought struck central North America during the 1930s. Centered on eastern Kansas, it extended north into the Canadian prairies, east to the Illinois-Indiana border, south to the Gulf of Mexico, and west into Montana and Idaho. See figure 7.1. The seven- year period of low rainfall and high temperatures, 1932 to 1938, was unprecedented in the memory of the Euro-Americans who inhabited the region in its extent, severity, and duration. It has been described by climate scientists as “one of the most severe environmental catastrophes in U.S. history” (Schubert et al. 2004, 1855). The period is best remembered for the Dust Bowl conditions created on the panhandles of Texas and Oklahoma and adjacent parts of New Mexico, Colorado, and Kansas.1 Richard Sutch is the Distinguished Professor of Economics at the University of California, Riverside, and a research associate of the National Bureau of Economic Research. Thanks to Vilma Helena Sielawa Ferreira, Connie Chow, and Hiroko Inoue for research assistance; to Susan B. Carter for critical advice; and to Norman Ellstrand for assistance with the plant biology. Comments by Paul David, Bronwyn Hall, Gary Libecap, Michael Roberts, Paul Rhode, Hugh Rockoff, Oscar Smith, Richard Steckel, and Gavin Wright on earlier drafts are gratefully acknowledged. Parts of this chapter were previously released as National Bureau of Economic Research Working Paper number 14141 (Sutch 2008). Financial support was provided by a National Science Foundation grant: “Biocomplexity in the Environment, Dynamics of Coupled Natural and Human Systems.” Administrative support was provided by the Biotechnology Impacts Center and the Center for Economic and Social Policy at the University of California, Riverside. 1. The Dust Bowl is not synonymous with the drought area. The Dust Bowl had a naturally semiarid climate and was settled during an untypical period of favorable climatic conditions. The first farmers imposed a “system of agriculture to which the Plains are not adapted to bring into a semiarid region methods which, on the whole, are suitable only for a humid region” (http://newdeal.feri.org/hopkins/hop27.htm). Arid conditions returned with the North American precipitation anomaly of the 1930s. For more detail, see the Report of the President’s Great Plains Drought Area Committee (1936), Hansen and Libecap (2004), and Hornbeck (2009).

195

196

Richard Sutch

A

B

Fig. 7.1 A, 1932–1938 composite precipitation anomaly. (Based on Schubert et al. 2004, figure 1, 1855.) B, the American Corn Belt. (Based on the 1950 Census of Agriculture. Counties with 25 percent or more of acreage planted to corn.)

My interest in this chapter is not with the Dust Bowl but with the Corn Belt that lies to the northeast of the Dust Bowl. Figure 7.1 also displays the outline of the Corn Belt. As can be seen, the eastern portion of the Corn Belt (western Ohio and Indiana) was largely outside the region struck by the severe drought. By contrast, the western Corn Belt (southwest Minnesota, western Iowa, southeast South Dakota, and eastern Nebraska) was hard hit. This geographical contrast will allow me to explore the adaptations made by corn farmers to sudden climate change. The lens through which I will look is the adoption of hybrid corn in the 1930s.

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

197

Fig. 7.2 Proxies for the severity of the drought: Iowa, 1926–1950

The suggestion that I make in this chapter is that the severe drought of 1936 revealed an advantage of hybrid corn not previously recognized— its drought tolerance. This ecological resilience motivated some farmers to adopt hybrids despite their commercial unattractiveness in normal years. But that response to climate change had a tipping effect. The increase in sales of hybrid seed in 1937 and 1938 financed research at private seed companies that led to new varieties with significantly improved yields in normal years. This development provided the economic incentive for late adopters to follow suit. Because post- 1936 hybrid varieties conferred advantages beyond improved drought resistance, the negative ecological impact of the devastating 1936 drought had the surprising, but beneficial, consequence of moving more farmers to superior corn seed selection sooner than they might otherwise. There is no doubt that the drought decimated corn crops in 1934 and 1936. One index of the impact is the fraction harvested of each year’s acreage planted to corn. When the damage to the crop is extensive, it is not worthwhile to attempt a harvest. If the damage is total, there is no crop to harvest. Figure 7.2 presents the percentage of the corn acreage planted that was harvested in the state of Iowa for the years 1926 to 1950. The two

198

Richard Sutch

Fig. 7.3 Proxies for the severity of the drought: Illinois and Kansas, 1926–1950

years 1934 and 1936 stand out as quite depressed. Figure 7.3 displays the data for Illinois (top panel), a state that was less affected by the participation shortfall, and Kansas (bottom panel), a hard hit state at the epicenter of the drought. Another index of drought is the yield (in bushels of corn) per harvested

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

199

acre. Figures 7.2 and 7.3 also display the yield statistics.2 As figure 7.3 suggests, the yield data is somewhat less satisfactory as an index. Unfortunately, all we have for most counties and crop districts is the yield per harvested acres. In a state like Kansas, only a very small fraction of the acreage was harvested, presumably located in areas that escaped the worst of the drought. On those privileged farms, relative yields were depressed but not to the extent in percentage terms as in Iowa. In a state like Illinois, the drought effect is more evident in the harvest- to- planting ratio than in the yield per harvested acre. 7.1

The Adoption of Hybrid Corn

Hybrid corn (technically “double- cross inbred- hybrid corn”) was “invented” by Donald F. Jones in 1917 to 1918 and was developed and introduced on a trial basis in 1924 by Henry Agard Wallace. In the 1920s, the Iowa Experiment Station began scientific field trials. Wallace’s hybrid was first entered in the Iowa tests in 1921. It won first place in 1924. It was first sold commercially in 1925. Competitors began sales in 1928. Widespread commercial adoption began in 1932 (Olmstead 2006, IV–9; Zuber and Robinson 1941, 589). The U.S. Department of Agriculture (USDA) began tracking the adoption of the new varieties in the following year, 1933. At that time, only about 0.1 percent of the nation’s corn acreage was planted to the new seed. In 1936, the USDA proclaimed significant increases in yield per acre could be achieved by adopting hybrid corn (Jenkins 1936, 481). Yet it took another decade before 70 percent of the corn acreage had been planted with hybrid seed. By 1960, 96.3 percent of acreage was planted to hybrid varieties (USDA, Agricultural Statistics, 1962, table 46, 41; USDA, Historical Track Records, 2004, 19). Figure 7.4 reproduces the annual data for the country as a whole and reveals a lazy S- shaped cumulative diffusion pattern.3 Zvi Griliches made the adoption of hybrid varieties in the United States the exemplar for a statistical model of technological diffusion and fitted the cumulative adoption patterns reported by the USDA to a logistic curve (Griliches 1957b). Despite the nearly three decades between initial adoption and full acceptance, Griliches considered the adoption pattern displayed in figure 7.4 to have been remarkably rapid (502). By examining the pattern of adoption, first, state by state, and then, crop district by district, he argued that at the local level adoption proceeded rapidly. But, he continued, the 2. Unpublished state- and county- level data on acreage planted and harvested were made available by Michael Haines. I am most grateful. 3. National- and state- level data on the percentage of corn acres planted to hybrids are available in various annual issues of the USDA’s Agricultural Statistics. I have relied on the volumes for 1945 (table 46, 42), 1948 (table 50, 48), 1950 (table 49, 47), 1952 (table 43, 40), 1954 (table 38, 30), 1957 (table 40, 39), 1959 (table 43, 33), and 1961 (table 43, 33). The year 1960 is the last date this data is available.

200

Richard Sutch

Fig. 7.4 Percent of corn acreage planted to hybrid varieties and acres planted to corn Source: USDA, Agricultural Statistics, (1962, table 46, 41); USDA, National Agricultural Statistics Service, Historical Track Records (April 2004, 19).

initial introduction into a local farming community proceeded slowly as commercial seed producers developed and marketed acceptable hybrids in different locales. Thus, the rate of diffusion across the United States proceeded slowly (507). Griliches’s explanation for the rapid and complete abandonment of openpollinated corn in favor of the new hybrid varieties was based on a simple set of “stylized facts.” He considered hybrid corn superior to the traditional open- pollinated varieties from the beginning and suggested that that superiority was established in 1935 and persisted thereafter.4 The advantage of 4. The date 1935 is the year that acreage planted to hybrid corn exceeded 10 percent of the total in the district at the heart of the hybrid revolution. Griliches chose 10 percent “as an indi-

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

201

hybrids, according to Griliches, could be objectively measured by the relative increase in yield over the open- pollinated corn (Griliches 1957b, 516–17). He assumed that the new varieties required no significant increase in capital investment or annual inputs. According to this analysis, the adoption process in a given crop district was one of disequilibrium transition (503). Griliches attributed the lags in the process to “imperfect knowledge.” It “takes time to realize that things have in fact changed” (516). The spread of hybrid corn geographically was slowed by the supply lags in developing and introducing hybrid varieties tailored to the specific soil type, weather conditions, and latitude of the peripheral regions.5 But even this process was rapid. Using the rule of thumb suggested by Griliches to mark the start of an adoption process as the date that 10 percent of acreage was planted to hybrid corn, Iowa in 1936 was followed by Illinois, Indiana, and Wisconsin in 1937; by Minnesota in 1938; and by Ohio, Nebraska, and Missouri in 1939. The development of hybrid corn and its rapid adoption were, nearly from the beginning, hailed as a triumph of twentieth- century biotechnology and one that carried with it enormous welfare benefits (Sprague 1946, 101).6 In a chart that is perhaps even more famous, at least among plant scientists, than Griliches’s logistic, the rise in corn yields per acre is employed to suggest that hybrid corn was responsible for a biotechnological revolution that abruptly ended a sustained period of “biological stasis.” Figure 7.5 displays the “hockey stick” graph reproduced dozens of times in the scientific literature.7 The chart plots USDA statistics on corn yields per acre dating back to 1866. There was a remarkable stability in yields with no discernable trend before 1936.8 Thereafter, yields began to increase, and they have continued cator that the development had passed the experimental stage and that superior hybrids were available to farmers in commercial quantities.” The region where this breakthrough occurred was Iowa Crop Reporting District 6 (Griliches 1957b, 507 and table II, 508). District 6 comprised Bremer, Black Hawk, Benton, Buchanan, Linn, Delaware, Jones, Dubuque, Jackson, and Clinton Counties, all in Iowa. 5. Paul David, in an insightful review, has criticized the Griliches approach “for lacking any real micro- level technology choice model” (David 2006, 4). Edwin Mansfield (1961) can be credited with supplying such a model to explain the logistic shape (although, as David points out, Mansfield’s model is simply one of many formulations consistent with the data). Mansfield suggested that the probability that a nonuser would switch to a new technology would be a function of the number of those in the immediate neighborhood who had already accepted the technology. This “contagion” model, borrowed from epidemiology, leads to the logistic diffusion curve. Bronwyn Hall (2004) provides a review of the theoretical literature on diffusion from both sociology and economics. 6. Griliches estimated the rate of return on hybrid corn research to have been at least 700 percent annually as of 1955 (Griliches 1958, 419). 7. As an indication of how ubiquitously figure 7.5 appears, I note that a standard textbook on corn for plant scientists (Smith, Betrán, and Runge 2004) reproduces a version of this chart four times in four separate chapters (Troyer 2004, chapter 1.4, figure 32, 218; Betrán, Bänziger, and Menz 2004, chapter 2.3, figure 6, 351; Wisner and Baldwin 2004, chapter 3.8, figure 2, 759; and Halauer 2004, chapter 4.4, figure 1, 901). 8. Alan Olmstead and Paul Rhode have challenged the notion of a biological stasis before 1936. They view the stability of yields before 1936 as due to a balance of conflicting forces, some of which would depress yields and counterbalancing ones that worked to raise yields (Olmstead and Rhode 2008, 64–97).

202

Richard Sutch

Fig. 7.5 U.S. corn yields, 1866–2002 Source: Carter et al. (2006, series Da693- 694); USDA, National Agricultural Statistics Service, (2004, 7, 9). Note: The percentages shown indicate the percentage of corn acreage that was planted to hybrid corn.

upward ever since. Yields per acre rose from an average of 25 bushels per acre before 1936 to 135 bushels per acre in the years 2000 to 2002 (Carter et al. 2006, series Da693- 694), more than a fivefold increase. Perhaps too casually, this increase has been attributed (a) to the continuing adoption of the hybrid varieties between 1936 and 1960 and (b) to the continuing improvement of hybrid traits as new varieties were introduced between 1936 and 1989 (Duvick 1992). I say “perhaps too casually” because the introduction of hybrids was also accompanied by the increased use of synthetic nitrogen fertilizers, increased planting densities, and the adoption and improvements in planting and harvesting machinery. However, these developments were intimately interrelated. One of the hybrid traits introduced improved the plant’s ability to absorb nitrogen fertilizers, and, indeed, the use of fertilizer was required to reach the potential of the hybrids. Similarly, the increased planting densities were possible only because of traits that reduced the plant’s requirements for full sunlight and that increased its resistance to lodging—the tendency of the plant to lie down or fall over when beaten down by the wind. Even then high density was possible only with the heavy application of fertilizer. Increased planting densities also required the abandonment of the horse and the horse- wide path between the rows of corn. Thus, the adoption of machinery was a necessary component for achieving the full potential of hybrid corn.9 Because hybrid seed, synthetic fertilizer, and gasoline tractors 9. On these points, see Castleberry, Crum, and Krull (1984, 33); Shaw and Durost (1965, table 21, 39); Johnson (1960); and Sutch (2008).

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

203

Fig. 7.6 Hybrid grain yield by year of introduction of the variety Source: Duvick (1992, 69–79, table 3, 73).

were a necessary triad, it is not really possible to partition responsibility for the yield increases among them.10 The continuing improvement in the performance of hybrids after their initial introduction is an important part of the story. Figure 7.6 reproduces the results of field experiments conducted in 1989 and 1990 in central Iowa. Forty- one varieties introduced between 1934 through 1989 by Pioneer Hi-Bred (a leading seed producer and a key player in the story to follow), “all popular in their time,” together with the most famous open- pollinated variety, Reid’s Yellow Dent, were planted in adjacent fields in a demonstration designed to illustrate the advance of yields due to genetic improvement (Duvick 1992, 70). As figure 7.6 illustrates, yields advanced at an average rate of 1.16 bushels per acre per year throughout this fifty- five- year period.11 Despite the undeniable improvement in plant traits and the obvious appeal of the Griliches’s adoption story, I make the following claims: (a) There was not an unambiguous economic advantage of hybrid corn over the openpollinated varieties at the time of planting in 1936. (b) The early adoption 10. The biological revolution in corn, commonly associated with the introduction of hybrid varieties, was not a unique phenomenon. Indeed, I find remarkably similar “hockey stick graphs” for the yields per acre in cotton, wheat, tobacco, oats, potatoes, and barley (Sutch 2008, 19–26). The simultaneous increase in the yields of so many different crops during this period is more properly attributed to the discovery of an economical process for synthesizing ammonia and the consequent increase in the use of synthetic fertilizers (Smil 2001). 11. Also see Cardwell (1982); Russell (1984); and Castleberry, Crum, and Krull (1984).

204

Richard Sutch

of hybrid corn before 1937 can be better explained by a sustained propaganda campaign conducted by the USDA at the direction of the Secretary of Agriculture, Henry Agard Wallace. The USDA’s message echoed that of the commercial seed companies. Wallace was the founder of the Pioneer Hi-Bred Seed Company, the first and largest producer of hybrid seed. (c) Later adopters of hybrid seed were motivated by the drought resistance of one experimental hybrid variety vividly demonstrated during the drought of 1936. The eventual improvement of yields as newer varieties were introduced and the imitative force of “collective logic” explain the continuation and acceleration of the process. Given the required capital investments in fertilizer tanks and tractors and the inability to save and plant one’s own seed, adoption tended to be irreversible. 7.2

The Iowa Corn Yield Tests and Hybrid Superiority

Griliches assumed that hybrid corn had an economically significant and unambiguous superiority over open- pollinated corn from the time it was first introduced. He reported that this superiority could be gauged by a 15 to 20 percent higher yield achieved with hybrid corn over the traditional openpollinated varieties. Griliches also suggested that this relative advantage applied to both high- and low- yielding soils, good years and bad (Griliches 1957b, 516–17; 1958, 421). His citations to support this estimate of the yield advantage were from an unpublished Federal Commodity Insurance Corporation source dated 1942 and published sources dated 1940 (USDA 1940), 1946 (Sprague 1946), and 1952 (Rogers and Collier 1952). None of these sources referred to the 1932 to 1936 period of early adoption. The 1940 USDA report cited the claims of “plant breeders” (Griliches 1957b, 517).12 G. F. Sprague, an agronomist at Iowa State College, based his 20 percent estimate on the increase in per acre yields observed in Iowa between 1933, when only 0.7 percent of the corn acreage was hybrid, and 1943, when 99.5 percent hybrid planting was reported, and he attributed the entire advance to the use of hybrid seed (Sprague 1946, figure 1, 101).13 John Rogers, a professor of agronomy at College Station, Texas, and Jesse Collier, at the Texas Blackman Experiment Station, simply reported without citation “experience in other corn- growing regions” (1952, 7). None of these reports seems a very reliable source, and none explicitly examine the relative superiority of hybrid corn in the first half of the 1930s. 12. There was a survey of “scientists engaged in crop breeding” taken (probably) in 1938 that reported estimates of the hybrid yield advantage that ranged from 5 to 25 percent; the authors concluded that the probable range was 10 to 15 percent (Dowell and Jesness 1939, table 1 and 480–81). This may have been the source for the USDA’s 1940 report of the opinions of “plant breeders.” 13. The actual increase in yields between those two years was 38 percent (USDA 2004), but 1945 was a very poor year for Iowa corn, so perhaps Sprague, writing in 1945, tempered his estimate.

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

205

Fig. 7.7 Iowa crop reporting districts Source: Zuber and Robinson (1941).

The best and most appropriate data on the relative yields of different corn varieties are the reports of field trials conducted by the agricultural experiment stations. These remain unexploited by quantitative historians.14 Beginning with the Iowa Agricultural Experiment Station in the early- 1920s, many of the stations in Corn Belt states conducted controlled plantings of open- pollinated, experimental hybrid, and commercially available hybrid seeds and published the results in the stations’ Bulletins. This chapter relies on the data available from the Iowa corn yield tests. These are the most complete. They begin at the earliest date. And they are the most relevant. Iowa was both the heart of the Corn Belt and the first state to widely and most quickly adopt hybrid corn. For the Iowa tests, the state was partitioned into twelve districts, shown on the map in figure 7.7. A volunteer farmer from each district, who was 14. Although not cited in his published Econometrica article (1957b), Griliches’s unpublished PhD thesis for the University of Chicago contains a comment in an appendix that rejects the Agricultural Experiment Station data: The data raise several difficult problems. They represent results on one or several fields in the whole state, conducted under varying and better than average conditions. The relation between the experiment station results and what the farmer may expect on his own farm is not clear. In particular, this relation may not remain constant between different states. For example, while the average yield in Iowa tests was around 80 bushels per acre at a time when the average yield for the state was around 40 bushels, the North Carolina tests averaged more than 100 bushels, but at the same time the average state yield was only around 30 bushels. (Griliches 1957a, 56–57) These considerations may make the Iowa test results an exaggerated estimate of the absolute advantage, but all that is needed for Griliches’s disequilibrium model is an estimate of the relative advantage. Elsewhere, Griliches argued that the relative advantage was independent of the yield per acre. Moreover, to the extent that the test results exaggerated the absolute gain, they bias the farmer’s decision calculus toward adoption, and, thus, they would bias the argument against the claim I make that the hybrid advantage was not large enough to encourage early adoptions.

206

Richard Sutch

also a member of the Iowa Crop Improvement Association, planted several varieties in adjacent fields and employed a uniform cultivation practice to raise them to maturity. At harvest, the yields were measured separately for each variety and reported back to the experiment station. Table 7.1 summarizes the results for the years 1926 through 1940. For each district and each year, the average yield for all hybrid varieties tested is expressed relative to the average for all open- pollinated varieties. It is immediately clear why the introduction of hybrid corn caused such excitement. Of the 166 observations in the table, only two recorded a relative below 101 (District 3 in 1926 and District 8 in 1927). These data provide strong support for the concept of hybrid vigor, or “heterosis” to use the scientific term. As we will see, however, hybrid vigor is not the same as economic superiority. 7.3

Hybrid Vigor

American corn or, more properly, “maize” (Zea mays, L.) is native to North America.15 It originated in Mexico where farmers cultivated it for millennia, gradually improving the plant by the selection for genetically based traits. Before the development of hybrid corn, the seed used for farmplanted corn was the result of natural cross- pollination. Under these conditions, corn is said to be “open- pollinated.” Pollen, produced by the corn plant’s tassels, is released and carried on the air. Some of the pollen typically reaches the cornsilks (the “ear shoots,” which are the stigmas of the female flower) of one or more nearby plants. The geminating pollen tube grows down the silk and fertilizes the egg cell, thereby starting the growth of a seed. In principle, each seed on an ear of corn could have a different male parent. The fraction of corn seeds set by self- pollination is known to be very low. If this fertilization process is left to the wind, selective breeding consists of choosing individual ears of corn on the basis of desirable plant or grain properties and saving those seeds for the following year’s crop. A great deal of natural hybridization between separate corn populations and varieties took place in this way.16 As a consequence of repeated selection under open- pollination, corn lines evolved that were adapted to new climates and soil conditions such that corn cultivation spread across the North American continent in the nineteenth century. The next step, deliberate control of parentage, produced “varietal hybrids” 15. The sources for this and the next several paragraphs are many, but the science is wellknown, so a detailed list of sources will be omitted. References to the names and historical dates can be found in Duvick (2001). For a history of corn varieties (germplasm), see Troyer (2004). Much of the science is elaborated in Smith, Betrán, and Runge (2004). A useful discussion of the relevant agricultural history is provided by Bogue (1983). 16. For a brief review of the history of maize cross pollination, see Olmstead and Rhode (2008, 71–76). The most popular open- pollinated variety at the time that the first hybrids were introduced was Reid’s Yellow Dent. This was an accidental hybrid between a reddish semigourd and a yellow flint. The story is told by Russell Lord (1947, 147) and Troyer (2004).

117 105 97 116 107 105 105 111 104 103 105 110

1 2 3 4 5 6 7 10 8 11 9 12

109 117 103 105 111 110 103 102 98 114 102 107

1927

110 120 109 110 108 103 114 111 115 108 114 104

1928

109 124 114 110 108 103 109 108 109 112 114 106

1929 114 113 111 116 114 105 113 102 124 111 106 103

1930

106 112 113 109 107 105 108 106 107 102

a

116

1931 115 102 102 107 108 106 112 102 110 111 106 100

1932

122

a

149 115

109 114 105

105

113

111 109 106 121 107 103

1935

112 101 119 111 108 106

1934

114 110 107 129 128 116

1933

140 127 154 149 141

a

129 117

a

107 118 126

1936 107 109 112 108 114 108 150 133 109 114 114 118

1937 115 109 118 114 112 117 131 120 112 134 106 115

1938

Iowa corn yield tests, 1926–1940 (relative average yield for all hybrid varieties, all open-pollinated varieties ⴝ 100)

Source: Zuber and Robinson (1941). Note: For 1933 to 1935, Districts 10, 11, and 12 were combined with Districts 7, 8, and 9, respectively. a Crop lost due to drought. b Poor crop. Data not calculated. c Crop abandoned because of wire worms.

1926

District

Table 7.1

110 115 114 113 107 116 120 141 112 115 110 108

1939

122 118

c

121 127 115 121 132 116

b

108 122

1940

208

Richard Sutch

in experiments conducted by farmers and agronomists in the late nineteenth and early twentieth centuries. Ever since Charles Darwin’s experiments with inbred and cross- pollinated corn, reported in 1876, it was known that the progeny of inbred plants were inferior to those of the cross- bred hybrids.17 In Darwin’s terms, hybrid plants had “innate constitutional vigour.” The lack of this vigor in the inbreds is known as “inbreeding depression.” Not surprisingly, Darwin’s results stimulated experimentation with deliberate cross- variety hybrids. Neither natural hybrids nor the deliberate varietal hybrids are the hybrid corn of the hybrid revolution under discussion. Hybrid corn as it is known today is more accurately described as a hybrid of inbred lines. Due to their inferior quality, the inbreds were generally avoided by plant breeders. So it took a leap of imagination when George Shull and Edward East, working independently, crossed two pure inbred lines of corn (homozygous strains) and produced plants superior to the run- of- field open- pollinated varieties. The results were published in 1908. The Shull-East “single- cross” hybrid of inbred lines, in principle, could revolutionize corn farming. Seeds could be produced on a field- wide basis by removing the tassels from one inbred line and allowing it to be fertilized by the pollen from a second inbred line planted in the neighboring row. It seemed evident, however, that this approach was impractical. Producing the inbred lines that were to be crossed involved laborious hand pollination, and these parent lines were so depressed by inbreeding that their seed yields were extremely low, making the input costs to large- scale seed production prohibitive. The problem of producing hybrid seed that the farmer could afford was further compounded by the need to plant freshly made hybrid seed each year. If the seeds of an inbred hybrid were planted, the yields achieved the following season would drop significantly because seed from a hybrid field would suffer from inbreeding depression (Jugenheimer 1939, 18–19). The practical problem was solved in 1918 by Donald F. Jones. He found that a “double- cross” hybrid could be made by crossing two single- cross varieties. The progeny, while generally not as productive as their single- cross parents, nevertheless out performed the open- pollinated varieties. Because single- crosses were prolific parents (unlike the pure inbred lines), production costs for the double- crosses were reduced to an economical level. All of the hybrids in the Iowa corn tests recorded in table 7.1 were doublecross varieties. The trial results for the period 1926 to 1933 (before the drought of 1934) are plotted both as a histogram and as a density estimate in figure 7.8. The average yield advantage of the hybrids was 9.3 percent (averaging across the twelve districts and ninety- five observations). The median yield 17. Darwin married his first cousin, and their first child was mentally retarded. He had a life- long interest in inbreeding.

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

209

Fig. 7.8 Density estimate of relative hybrid yield, 1926–1933: Iowa corn yield tests

advantage was 9 percent. An advantage of 15 to 20 percent would be an exaggeration for this period. The average advantage in District 6, where the adoption of hybrid corn first took place, was only 7 percent. Although the 15 to 20 percent advantage cited by Griliches is an exaggeration for the period of first adoption, perhaps the story of a disequilibrium transition would be just as valid with the more modest 7 to 9 percentage advantage reported by the Iowa Experiment Station. The typical yield with open- pollinated corn reported by the farmers conducting the Iowa corn yield tests, 1926 through 1933, was 61.3 bushels per acre (Shaw and Durost 1965, table 12, 28). Presumably, this represents the yield achieved with best- practice farming by experienced farmers. The statewide average for that period, however, was only 38.2 bushels per acre. Thus, the typical farmer could anticipate a yield gain of less than four bushels per acre, and an experienced best- practice farmer could anticipate perhaps as much as a six- bushel gain. Before we can conclude that an advantage in physical yield translated into an economic advantage, we must factor in the depressed value of the corn crop and the high price of hybrid corn seed. With the advent of the Great Depression in the 1930s, the market price of corn came down from a high of eighty to eighty- five cents a bushel in the late 1920s to thirty- two cents in 1931 and 1932 (Carter et al. 2006, series Da697). During the Depression, hybrid seed was selling in Iowa for $6.00 a bushel. The average price for

210

Richard Sutch

hybrid seed sold by Pioneer Hi-Bred for the 1932 crop was $6.51 and in the following year it was $5.41 (Urban 1979).18 Because a bushel of seed would plant three acres (Duvick 1992, 71), a farmer would have to expect a financial gain exceeding $2.00 an acre to be tempted to pay full price.19 Expecting no more than thirty- two cents per bushel for the crop when sold, the advantage of the hybrid seed would have had to exceed six bushels per acre, not the four to six bushels that could be anticipated on the basis of the Iowa corn tests. 7.4

Henry Agard Wallace and Hybrid Hype

The puzzle then is why some adventurous farmers were willing to adopt hybrid corn before its economic superiority was demonstrated by controlled tests. My suggestion has two parts: (a) there was an aggressive marketing campaign launched by the commercial seed companies directed at potential adopters, and (b) the Secretary of Agriculture, Henry Agard Wallace, a commercial promoter of hybrid seed and former President of the major seed company, Pioneer Hi-Bred International, put the full weight of the federal government behind an advocacy of hybrid corn. Henry Agard Wallace, Franklin Roosevelt’s first Secretary of Agriculture, was a multifaceted, complex, prolific, and eccentric man.20 He was an early champion of scientific farming, a path- breaking plant scientist, a talented statistician and geneticist, America’s first econometrician, author of a dozen books, a journalist, and the influential editor of Wallaces’ Farmer, from 1921 to 1933, the most prominent agricultural magazine of its time.21 Later he became the editor of The New Republic, 1946 to 1948. Wallace was a successful entrepreneur who made a personal fortune as the leading founder of the Hi-Bred Seed Company (later Pioneer Hi-Bred International, Inc.). Today, Pioneer is a wholly owned subsidiary of E. I. du Pont de Nemours 18. Culver and Hyde report a figure of $6.00 per bushel, but elsewhere they report “at the depths of the Depression, corn sold in Iowa for ten cents a bushel” and that Pioneer’s price was $5.50 a bushel (Culver and Hyde 2000, 91, 147). Pioneer’s average price for hybrid seeds for the 1930 plantings was $10.43 per bushel, for the 1931 crop it was $7.83 (that year the posted price was $10.60 but the company offered a free bushel for every two purchased), for 1932 prices fell to $6.51, for 1937 to $5.41, and for the 1934 crop the average price rose to $6.58 (Urban 1979). 19. Corn seeding rates varied widely depending upon the local practice and climate. States with abundant rainfall, such as Ohio, supported heavier seeding rates (Fernandez-Cornejo 2004, 8). I thank Oscar “Howie” Smith, a former employee of Pioneer Hi-Bred International, for clarifying the seeding rate practices of the Corn Belt in the 1930s. 20. As one index of his eccentricity, I note that Wallace was a mystic and an ambidextrous, vegetarian, teetotaler before any of these affectations was considered legitimate. Republican teetotalers holding high office in Roosevelt’s New Deal administration were rare indeed. The best biography of Wallace is by John C. Culver and John Hyde (2000), from which I draw the details in this paragraph. 21. It was the USDA’s statistician Louis Bean that named Henry A. Wallace the first American econometrician based on Wallace’s book, Agricultural Prices (1920). See Culver and Hyde (2000, 51).

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

211

and Company and is the largest seed company in the world with a market share in 1997 of 42 percent (Fernandez-Cornejo 2004, table 12, 26; Beck 2004, 568). Henry Agard Wallace was, according to historian Arthur Schlesinger Jr. (2000), America’s best Secretary of Agriculture.22 He was Vice President of the United States during World War II—the most influential and powerful vice president before Dick Cheney. Wallace served as Secretary of Commerce during the economic transition to peace time (1945 to 1946). He ran for president on the Progressive Party ticket in 1948. The Des Moines Register identified Henry A. Wallace the “Most Influential Iowan of the 20th Century” on December 31, 1999. His biographers identified him as the “state’s greatest son” (Culver and Hyde 2000, ix). When he died of Lou Gehrig’s disease in 1965, the then- reigning Secretary of Agriculture, Orville Freeman, could declare without hyperbole that “No individual has contributed more to the abundance we enjoy today than Henry Wallace” (531). A unifying theme—an obsession, really—for Wallace throughout this prolific and many- sided career was hybrid corn. In 1910, two years after Shull and East reported on their single- cross inbred hybrid experiments, Wallace was debating the findings with Iowa State College agronomists in Ames. In 1912, he conducted his own experiments to produce single- cross hybrids. At the time, he concluded that the difficulty of hand pollination “was too laborious” (Wallace quoted by Culver and Hyde 2000, 67). Over the next several years, Wallace experimented with varietal hybrids without achieving consistent success. But when Edward East visited Wallace in 1919 and introduced him to Donald Jones’s results with double- cross hybrids, Wallace immediately saw the commercial potential and began his own experiments with the new technique. He also used the pages of Wallaces’ Farmer to proclaim the coming revolution (Culver and Hyde 2000, 68). In 1920, the circulation of Wallaces’ Farmer was 65,200 (Galambos 1968, 344). The journal was read by a high proportion of corn and hog farmers. In 1920, Wallace convinced the Iowa State Agronomist, H. D. Hughes, to establish the Iowa corn yield tests. The idea was to challenge the current practice of judging corn by the physical appearance of the ear and instead focus on yields per acre. Wallace did not have enough seed to offer an entry of his own that first year, and his entry for 1921 failed to outperform the best of the open- pollinated varieties. He entered a new hybrid, named Copper Cross, in the 1922 tests and again in 1923, but it too failed to outyield the best open- pollinated entries. However, Copper Cross was successful enough that Wallace was able to draw up a contract for its commercial release with the Iowa Seed Company. When Copper Cross won the gold medal at the 1924 22. Precision requires that I use Wallace’s middle name because his father, Henry Cantwell Wallace, was also Secretary of Agriculture (1921 to 1924), appointed by Warren Harding. Another Henry Wallace in the family was Henry A. Wallace’s grandfather and the founder of Wallaces’ Farmer. This Wallace had no middle name (Lord 1947).

212

Richard Sutch

test, the commercialization of hybrid corn was launched.23 Wallace himself wrote the first advertising copy. “An Astonishing Product—Produces Astonishing Results . . . If you try it this year you will be among the first to experiment with this new departure, which will eventually increase corn production of the U.S. by millions of bushels” (quoted by Culver and Hyde 2000, 71). In 1926, Wallace founded the Hi-Bred Seed Company (Culver and Hyde 2000, 82–83). He continued to use the pages of Wallaces’ Farmer to proclaim the virtues of hybrid corn and, of course, to advertise his company’s seed. In that same year and the following year, hybrid maize did reasonably well in the Iowa corn yield tests, recording about a 7 percent greater yield than their open- pollinated rivals. But that was an insufficient advantage to create much demand. There were several obstacles, not the least of which was the astonishing price that Wallace was asking for his “astonishing” seed—it was $52 bushel in 1924 (May 1949, 514; Culver and Hyde 2000, 71). Two founding principles of the Hi-Bred Seed Company were first, total honesty in advertising and, second, high prices. “High prices, Wallace believed, were necessary to convince farmers they were buying something special” and, of course, high profits helped cover the cost of ongoing research intended to improve the varieties (Culver and Hyde 2000, 91, 148). Another obstacle that had to be overcome was the reluctance of many farmers to abandon their reliance on their own homegrown seed and instead entertain a visit by, and the commercial pitch of, the traveling seed salesman. The role of the salesman was not so much to educate the farmer—the genetics of inbred- hybrid crosses and the “magic” of heterosis exceeded the common- sense knowledge of most farmers and indeed of most seed salesman. The claims of superiority had to be accepted, if they were, on faith. It was a particularly sore point with many farmers that seeds saved from a hybrid crop could not themselves be planted the next season with any hope of success. So old habits were challenged. A commitment to hybrid seed was tantamount to an agreement to deal with the seed salesman every subsequent year as well as the current year. And that commitment meant that the farmer’s skill in selecting seed corn from his own crop, a skill that many took great pride in, would be no longer needed or esteemed (Fitzgerald 1993). Pioneer Hi-Bred designed a sophisticated marketing plan to address these problems. Seed salesmen working for Wallace offered to provide the reluctant farmer enough seed free of charge to plant half of his acreage. The farmer would plant the remaining land with the open- pollinated seed he preferred. In exchange, the Hi-Bred company would reclaim one- half of the increased crop produced by its seed judged against the farmer’s regular crop 23. The Funk Brothers Seed Company introduced its first double- cross hybrid in 1928 (Fitzgerald 1990, 218).

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

213

(Lee 1984, 43; Culver and Hyde 2000, 91). Typically, only one farmer on each lane was offered the deal with the hope that a demonstration would spread interest to the neighborhood. Other farmers were given yield guarantees (May 1949, 514). According to Culver and Hyde, it often took several years to persuade a farmer that the higher yields achieved with the Hi-Bred seed were not a fluke (91). With the advent of the Great Depression in the 1930s, marketing the new seed became even more difficult. As I have noted, the Depression had sent the market price of corn tumbling from a high of eighty- five cents a bushel in the late 1920s to thirty- two cents in 1932, and Iowa farmers, many who faced ruin, were hardly in the mood for experimentation and risk taking. Safety first was the general rule. Moreover, the average yield gains from switching to hybrids were, as I have already pointed out, generally insufficient to justify the cost of seed. Henry Agard Wallace became President Roosevelt’s Secretary of Agriculture in 1933. The public position did not damp his enthusiasm for hybrid corn. For many years, the USDA had published an annual volume, the Yearbook of Agriculture, devoted to reporting on the activities of the department, of advances in many fields, and offering both general and specific advice to farmers. The Yearbooks had large press runs and were widely distributed by members of Congress to their farming constituency. For the 1936 edition, Wallace made an unusual decision. As he explained: The 1936 Yearbook of Agriculture differs . . . from those published in recent years. . . . This year it is devoted to a single subject—the creative development of new forms of life through plant and animal breeding. (Wallace 1936, foreword) The article on “Corn Improvement” for this Yearbook was written by Merle T. Jenkins, the USDA’s Principal Agronomist. A headline exaggeratedly claimed “Yield Advances up to 35 Percent over Open-Pollinated Varieties” (Jenkins 1936, 481). The report was based on the Iowa corn yield test despite the exaggeration of the headline. In retrospect, and perhaps even at the time, the focus of the 1936 Yearbook was in jarring contrast to other efforts of the Roosevelt Administration to deal with the Great Depression. For the first several years of his administration, Wallace presided over the acreage reduction and crop destruction policies of the Agricultural Adjustment Administration. He was the one who ordered the plowing up of ten million acres of cotton in 1933 and the slaughter of six million baby pigs and sows in September (Culver and Hyde 2000, 123–25). Yet Wallace looked into the future beyond the current crisis to foresee a time when the yield increases to be made possible by the spread of hybrid corn would be welcome. By today’s standards, the glaring conflict of interest between Wallace’s financial interest in the Pioneer Hi-Bred Company and the use of the govern-

214

Richard Sutch

ment agency he controlled to advertise and advocate his product would be outrageous. But even this propaganda barrage combined with the innovative marketing strategy of his company might not have been successful in tipping the balance in favor of hybrid corn. It took two other factors to put the company on the road to success. 7.5

Drought and Research

The eventual success of hybrid corn was due, first, to a tipping event and then to the self- reinforcing momentum of biotechnology. The factor that acted to tip the balance in favor of hybrid adoption was paradoxically another disaster to bedevil corn farmers in the 1930s. As if the Depression, with its devastating impact on agricultural prices, was not enough, there were catastrophic droughts in 1934 and 1936. An index of the severity of these droughts, as we have suggested, is the fraction of the crop planted that was harvested. In Iowa and in the country overall, 30 to 40 percent of the acreage planted was so devastated by drought that it was not worth harvesting. In Nebraska and Kansas, the losses were nearly total. Consult figures 7.2 and 7.3. What the droughts starkly demonstrated was that the relative yield of hybrid corn was greatest when the absolute yields were generally depressed. Figure 7.9 reveals the relationship using, once again, the Iowa corn yield test results to illustrate the correlation. In the extreme drought conditions of the mid- 1930s, the yield differences between the new and traditional varieties were stark. Edward May, President of the May Seed Company, recalled: Yield differences became plainly evident in 1936, which was also a severe drouth year in Iowa. At this time nearly all farmers who were testing hybrid seed corn planted only a limited acreage. Yields of hybrids under these conditions in many areas of the state were approximately double the yields of other corn grown on the farm. The results were so convincing that it marked the end of the vast efforts of initial adoption. (May 1949, 514) “Almost overnight, demand for hybrid seed exploded” (Culver and Hyde 2000, 149). Big percentage point gains in adoption came in 1937: 22.3 percentage points accounted for by new adoptions in Illinois, 21.2 percentage points Iowa, 18.3 points in Ohio, 17.4 in Indiana, 12.9 in Wisconsin (see note 3 for sources). Pioneer Hi-Bred’s sales and profits also exploded. In 1936, as the impact of that year’s drought became evident, Pioneer’s sales of hybrid seed for the 1937 crop jumped to 40,586 bushels, up from 16,525 bushels sold in 1933 even as the price rose from $6.58 to $9.96 per bushel (Urban 1979). Once the move to hybrid corn was launched—and only because the switch was made—the technological diffusion process became self- sustaining and

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

215

Fig. 7.9 Relative advantage of hybrid varieties were best when yields were depressed by unfavorable weather: Iowa, 1935–1938

irreversible. The steady improvement of the yield advantage of hybrid corn began in 1937. See figure 7.6, but also figure 7.10, which illustrates the shift in the relative hybrid advantage after 1936. Farmers might have switched to hybrid corn out of fear of continued drought, but soon the genetic advance in hybrid corn made open- pollinated corn obsolete even though the price of hybrid seed remained high and a farmer using it would need to purchase fresh seed each season. This genetic improvement was achieved thanks to continuing research funded by the seed companies using retained earnings generated by soaring sales and high prices. Wallace believed that his hybrid revolution would have collapsed without a continuing, well- financed research effort (Culver and Hyde 2000, 148). Research by the federal government also played a supporting role.24 The research in both sectors was closely coordinated. According to Sprague (1946, 101), there was unrestricted interchange of ideas and seed stock between government researchers and the private companies. Most observers agree that the for- profit research was the driving partner of the private24. In 1922, when Henry Agard Wallace’s father, Henry C. Wallace, was Secretary of Agriculture a well- funded hybrid corn research program was established by the department in cooperation with the experiment stations in several Corn Belt states. This federal program was vital during the 1920s. Donald Duvick suggests that “the commercial maize breeders probably could not have succeeded in the early years [without the contributions from the public sector], for individually they simply did not have enough inbred lines . . .” (Duvick 2001, 71).

216

Richard Sutch

Fig. 7.10 Density estimates of relative hybrid yield before and after the droughts of 1934–1936: Iowa

federal joint effort after 1937 (Griliches 1958, 420–21 and table 1, 424; Duvick 2001, 71; Fuglie, et al. 1996, 45; Fernandez-Cornejo 2004, 41–50). Wallace claimed that his company spent more money on corn research than the USDA and the state experiment stations combined (Culver and Hyde 2000, 148). Ironically, the drought of 1934 was, in part, responsible for the remarkable improvement in hybrid development seen thereafter. One of the farmers that Hi-Bred recruited as part of its experimental research on new hybrid strains suffered greatly in the drought of 1934. Most of his experimental plants were lost. But he continued to work with the few plants that had managed to survive. The result was the unexpected discovery of a hardy new hybrid, number 307, with a remarkable ability to withstand drought. Consult figure 7.6 again, where number 307 is labeled for easy identification. The experimenter remarked that this plant “proved very valuable when we found ourselves in another serious drought condition in the summer of 1936” (Culver and Hyde 2000, 149). What we have, then, is a story of the diffusion of hybrid corn that is more complex and more interesting than the one usually told by Grilichesinspired plant scientists (Griliches 1960). Rather than disequilibrium transition slowed by information imperfections that were gradually overcome by commercial advertising and agricultural extension education, the history reveals that neither the innovation of 1918, nor the commercial product of 1924, nor the highly touted seeds of 1936 were economically and culturally

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

217

attractive. The advertising and marketing campaigns of the seed companies were effective in the late 1920s or early 1930s not because they educated farmers, but because they offered inducements designed to lower the costs and risks of adoption, shifting those costs and risks to the seed companies. The tipping point came in 1936 when farmers chose what seeds to plant in 1937. How much credit should be given to the Yearbook of Agriculture that year and how much to the drought would be difficult to say given their simultaneity. But what is clear is that the genetic advance in hybrid corn varieties beginning with hybrid 307 (which was introduced commercially in 1936) is what locked in the transitional adopters and made the hybrid revolution seem inevitable in retrospect. Had Wallace not used the bully pulpit of the USDA to promote his own commercial and financial interests, had the USDA not supported the research effort in the late 1920s and early 1930s, had the droughts of 1934 and 1936 not occurred, had Hi-Bred not continued a major research effort following 1936, the Wallace crusade might have succumbed as just another fatality of the Great Depression. 7.6

The Impact of the Drought on Adoption: An Illustration

The relationship between the drought and the adoption of hybrid corn can be illustrated using crop district data for the years 1926 to 1960. For the purposes of reporting the data on the percentage of acreage planted to hybrid varieties and the statistics of acreage and output, the USDA partitioned each of the corn states into nine (or fewer) districts that aggregated contiguous counties. Unfortunately, the county- level data on acreage planted are not complete, and the data on yields are available only for Ohio, Indiana, Illinois, Wisconsin, Iowa, Missouri, eastern South Dakota, and eastern Nebraska. Michigan and Kansas are not yet available. Because the data on acreage planted are not available at the county or district level for most of these states, I use the yield per harvested acre as my measure of the severity of the drought in 1936. Specifically, severity is measured as the percentage yield shortfall in 1936 when compared to the average yield for 1926 to 1935, excluding 1934. The more severe the 1936 drought (the greater its depressing impact within a crop district), the faster we expect the adoption of the new droughtresistant hybrids would have been. Zvi Griliches collected annual data on the percentage of acreage planted in each crop district that was devoted to hybrid varieties. District- by- district, he fit a simple three- parameter logistic curve to estimate the timing and speed of adoption. We used the parameters that he published (Griliches 1957b, table 2) to calculate three measures:25 25. A purist might prefer to use the original data rather than the fitted curves. I have been unable to locate the original data. However, the fit of each curve as reported by Griliches is very high. Moreover, there is certain logic in using the fitted values because they create a continuous variable, smooth the data, and correct for noise in the original crop reporters’ estimates.

218

Richard Sutch

Fig. 7.11 Year that 40 percent penetration of hybrid corn was achieved (44 crop districts in the Corn Belt)

(a) the year that the district achieved a 40 percent adoption rate, (b) the number of years that elapsed between the time a 10 percent adoption rate was achieved and a 70 percent adoption rate was recorded, and (c) the percentage of new acreage adopting hybrid corn in 1937 compared with 1936. I name these three variables “penetration,” “cautiousness,” and “jump.” We also conjecture that the eagerness to adopt the new varieties would be correlated with the productivity of corn farming within the district. The higher the yield per acre in normal years, the larger will be the pecuniary advantage of the higher yields the new varieties seemed to offer. 7.6.1

Penetration

A 40 percent rate of adoption would indicate that hybrid corn had made significant inroads in the region. Penetration is the date (in years) this threshold was achieved. The average over the forty- four crop districts was 1939. The smaller this number, the earlier substantial penetration was achieved. I performed a simple statistical calculation to predict the year that 40 percent penetration was achieved for a cross section of the forty- four crop- reporting districts in the cotton belt. See figure 7.11. The average yield reported for the 1926 to 1933 and 1935 period and our measure of the yield shortfall in 1936 were used as linear regressors. The larger the shortfall in yield, the greater severity of the drought, and, thus, the earlier penetration should have been achieved. The regression confirms this relationship. The higher the normal yield per acre, the earlier 40 percent penetration was achieved. Both

The 1936 Corn Belt Drought and Adoption of Hybrid Corn Table 7.2

219

Regression estimates of the reaction to the severity of the 1936 drought (44 crop districts in the Corn Belt) Dependent variable

Independent variables Yield Shortfall Constant R2 Mean of dependent variable Standard error of dependent variable

Penetration

Cautiousness

Jump

–0.257 –0.015 1,948.2

–0.068 –0.011 6.02

1.067 0.074 –30.17

0.668

0.125

0.464

3.272 0.138

9.145 1.143

1,938.89 0.23

coefficients are statistically significant at the 96 percent level. The regression statistics are presented in table 7.2. 7.6.2

Cautiousness

I expect that a severe drought would reduce caution and accelerate the speed of adoption. The results, also in figure 7.11, confirm this conjecture. The larger the shortfall in the 1936 crop, the faster the diffusion process proceeded. 7.6.3

Jump

The size of the jump in adoption rates, measured between 1936 and 1937, would be expected to be positively related to yield and to the shortfall. These results are also presented in table 7.2. Both coefficients have the predicted sign. Large yields and large losses in 1936 are associated with a bigger jump between 1936 and 1937. 7.7

Conclusions

We conclude that the drought of 1936 sped the process of adoption after it revealed the drought resistance of hybrid corn. The commercial seed producers were quick to spread the word. An advertisement for DeKalb seed published in the Prairie Farmer of October 24, 1936 prominently featured hybrid corn’s drought resistance and claimed it had been proven in eight Corn Belt states from Nebraska to Ohio (reproduced in Fitzgerald 1990, 176). After 1937, a new dynamic was set in motion. The explosion of demand for hybrid corn generated large profits for the major hybrid seed companies: Pioneer, Funk, and DeKalb. As a result, the companies invested heavily in research with new hybrid strains. They not only perfected the drought resistance of the plant but also found ways to permit increased planting

220

Richard Sutch

density, increase the resistance to lodging, and increase responsiveness to artificial fertilizer. The result was a steady improvement in the yields per acre that hybrid corn could achieve. Once these post- 1937 improvements were recognized, adoption of hybrid corn became economically advantageous; before 1937, it had not been so. Figure 7.11 traces the diffusion of hybrid corn from district to district. The map illustrates when the various districts achieved 40 percent penetration. The crop district in the center of the Corn Belt (Iowa District 6) was the epicenter for the spread of the innovation. It had reached 40 percent in 1936. Ten districts, all located in a concentric ring around the origin, achieved that threshold in 1937 in the rush to achieve some protection in case another dry year were to follow. In 1938, the concentric ring expanded to the north and west. In 1938, there were also four districts in northern Indiana and northwest Ohio that reached the penetration threshold. These districts were less affected by the drought of 1936. It is possible that a geographical contagion coupled with new and superior varieties introduced for the 1938 crop year explain this move into northern and central Indiana and Ohio. Central Indiana and Ohio filled in the map in 1939. The southern tier of those states did not reach 40 percent until 1940 or after. The sociologists Bryce Ryan and Neal Gross, writing in 1950, studied the diffusion of hybrid corn in two communities located in Greene County, Iowa (Ryan and Gross 1950). In their view, late adopters were farmers bound by tradition. They were irrational, backward, and “rural.” The early adopters by contrast were flexible, calculating, receptive, and “urbanized.” “Certainly,” they summarized, “farmers refusing to accept hybrid corn even for trial until after 1937 or 1938 were conservative beyond all demands of reasonable business methods” (672). They drew a policy implication: “The interest of a technically progressive agriculture may not be well served by social policies designed to preserve or revivify the traditional rural- folk community” (708). In part, this view was based on Ryan and Gross’s (incorrect) belief that hybrid corn was profitable in the early 1930s (668). I have suggested that this was not the case. Figure 7.11 should also give pause to the view that rural laggards delayed the adoption of hybrid corn. It would be hard to argue that the farmers in Iowa Crop Reporting District 6 were predominantly forward- thinking leaders, attentive, and flexible, while those in Indiana and Ohio were predominately backward rustics trapped by inflexible folk tradition. I think an implication of this study is that farmers (even those of rural America in the 1930s) are remarkably resilient and adaptive. Sudden and dramatic climate change induced a prompt and prudent response. An unexpected consequence was that an otherwise more gradual process of technological development and adoption was given a kick start by the drought and the farmers’ response. That pushed the technology beyond a tipping point and propelled the major Corn Belt states to the universal adoption

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

221

of hybrid corn by 1943. The country as a whole reached universal adoption by 1960. While this process was driven by individual farmers and privately owned seed companies, there was also a role played by the government. The USDA not only campaigned vigorously for hybrid corn from 1936 onward, but engaged in the years before 1936 in its own research, and subsidized the dissemination of knowledge and seed samples. That this engagement was to some extent promoted by the Secretary of Agriculture, Henry Agard Wallace, a hybrid researcher and the founder of the major commercial producer of hybrid seed, should not blind us from recognizing the importance of the government subsidizes in preparing the new technology for the leap forward.

References Beck, David L. 2004. Hybrid corn seed production. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, 565–630. New York: Wiley. Betrán, Javier, Marianne Bänziger, and Mónica Menz. 2004. Corn breeding. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, chapter 2.3. New York: Wiley. Bogue, Allan G. 1983. Changes in mechanical and plant technology: The Corn Belt, 1910–1940. Journal of Economic History 43 (1): 1–25. Cardwell, V. B. 1982. Fifty years of Minnesota corn production: Sources of yield increase. Agronomy Journal 74 (6): 984–90. Carter, Susan B., Scott Sigmund Gartner, Michael Haines, Alan Olmstead, Richard Sutch, and Gavin Wright, eds. 2006. Historical statistics of the United States. Millennial ed. New York: Cambridge University Press. Castleberry, R. M., C. W. Crum, and C. F. Krull. 1984. Genetic yield improvements of U.S. maize cultivars under varying fertility and climatic environments. Crop Science 24 (1): 33–36. Culver, John C., and John Hyde. 2000. American dreamer: The life and times of Henry A. Wallace. New York: Norton. David, Paul A. 2006. Zvi Griliches on diffusion, lags and productivity growth . . . Connecting the dots. Paper presented at conference on R&D, Education and Productivity held in Memory of Zvi Griliches (1930–1999), 25–27 August, Paris. Dowell, A. A., and O. B. Jesness. 1939. Economic aspects of hybrid corn. Journal of Farm Economics 21 (2): 479–88. Duvick, Donald N. 1992. Genetic contributions to advances in yield of U.S. maize. Maydica 37:69–79. ———. 2001. Biotechnology in the 1930s: The development of hybrid maize. Nature Reviews: Genetics 2:69–74. Fernandez-Cornejo, Jorge. 2004. The seed industry in U.S. agriculture: An exploration of data and information on crop seed markets, regulation, industry structure, and research and development. Agriculture Information Bulletin no. 786. Washington, DC: GPO.

222

Richard Sutch

Fitzgerald, Deborah. 1990. The business of breeding: Hybrid corn in Illinois, 1890– 1940. Ithaca, NY: Cornell University Press. ———. 1993. Farmers deskilled: Hybrid corn and farmers’ work. Technology and Culture 34 (2): 324–43. Fuglie, Keith, Nicole Ballenger, Kelly Day, Cassandra Klotz, Michael Ollinger, John Reilly, Utpal Vasavada, and Jet Yee. 1996. Agricultural research and development: Public and private investments under alternative markets and institutions. U.S. Department of Agriculture Agricultural Research Service Agricultural Economics Report no. 735. Washington, DC: GPO. Galambos, Louis. 1968. The agrarian image of the large corporation, 1879–1920: A study in social accommodation. Journal of Economic History 28 (3): 341–62. Griliches, Zvi. 1957a. Hybrid corn: An exploration in economics of technological change. Thesis, University of Chicago. ———. 1957b. Hybrid corn: An exploration in the economics of technological change. Econometrica 25 (4): 501–22. ———. 1958. Research costs and social returns: Hybrid corn and related innovations. Journal of Political Economy 66 (5): 419–31. ———. 1960. Hybrid corn and the economics of innovation. Science (132): 275–80. Hall, Bronwyn. 2004. Innovation and diffusion. In Oxford handbook of innovation, ed. Jan Fagerberg, David C. Mowery, and Richard R. Nelson, 459–85. New York: Oxford University Press. Hallauer, Arnel R. 2004. Specialty corns. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, chapter 4.4. New York: Wiley. Hansen, Zeynep K., and Gary D. Libecap. 2004. Small farms, externalities, and the Dust Bowl of the 1930s. Journal of Political Economy 112 (3): 665–94. Hornbeck, Richard. 2009. The enduring impact of the American Dust Bowl: Shortand long- run adjustments to environmental catastrophe. NBER Working Paper no. 15605. Cambridge, MA: National Bureau of Economic Research, December. Jenkins, Merle T. 1936. Corn improvement. In Yearbook of agriculture, 1936, ed. Henry A. Wallace, 455–522. Washington, DC: GPO. Johnson, Paul R. 1960. Land substitutes and changes in corn yields. Journal of Farm Economics 42 (2): 294–306. Jugenheimer, Robert W. 1939. Hybrid corn in Kansas. Kansas Agricultural Experiment Station Circular 196:1–19. Lee, Harold. 1984. Roswell Garst: A biography. Ames, IA: Iowa State University Press. Lord, Russell. 1947. The Wallaces of Iowa. New York: Houghton Mifflin. Mansfield, Edwin. 1961. Technical change and the rate of imitation. Econometrica 29 (4): 741–66. May, Edward. 1949. The development of hybrid corn in Iowa. Iowa Agricultural Experiment Station Research Bulletin 371 (1): 512–16. Olmstead, Alan L. 2006. Agriculture: Introduction. In Historical Statistics of the United States. Vol. 4, ed. Susan B. Carter, Scott Sigmund Gartner, Michael Haines, Alan Olmstead, Richard Sutch, and Gavin Wright, 7–10. New York: Cambridge University Press. Olmstead, Alan L., and Paul W. Rhode. 2008. Creating abundance: Biological innovation and American agricultural development. New York: Cambridge University Press. President’s Great Plains Drought Area Committee. 1936. Report. http://newdeal.feri .org/Hopkins/hop27.htm.

The 1936 Corn Belt Drought and Adoption of Hybrid Corn

223

Rogers, John S., and Jesse W. Collier. 1952. Corn production in Texas. Texas Agricultural Experiment Station Bulletin 746:1–76. Russell, W. A. 1984. Agronomic performance of maize cultivators representing different eras of breeding. Maydica 29:375–90. Ryan, Bryce, and Neal Gross. 1950. Acceptance and diffusion of hybrid corn seed in two Iowa communities. Iowa Agricultural Experiment Station Research Bulletin 372:662–708. Schlesinger, Arthur, Jr. 2000. Who was Henry A. Wallace? The story of a perplexing and indomitably naive public servant. Los Angeles Times, 12 March 2000. Schubert, Siegfried D., Max J. Suarez, Philip J. Pegion, Randal D. Koster, and Julio T. Bacmeister. 2004. On the cause of the 1930s Dust Bowl. Science 303:1855–59. Shaw, Lawrence H., and Donald D. Durost. 1965. The effect of weather and technology on corn yields in the Corn Belt, 1929–62. USDA, Economic Research Service, Agricultural Economic Report no. 80. Washington, DC: GPO. Smil, Vaclav. 2001. Enriching the Earth: Fritz Haber, Carl Bosch, and the transformation of world food production. Cambridge, MA: MIT Press. Smith, C. Wayne, Javier Betrán, and E. C. A. Runge, eds. 2004. Corn: Origin, history, technology, and production. New York: Wiley. Sprague, G. F. 1946. The experimental basis for hybrid maize. Biological Reviews of the Cambridge Philosophical Society 21 (3): 101–20. Sutch, Richard. 2008. Henry Agard Wallace, the Iowa corn yield tests, and the adoption of hybrid corn. NBER Working Paper no. 14141. Cambridge, MA: National Bureau of Economic Research, June. Troyer, Forest. 2004. Persistent and popular germplasm in seventy centuries of corn evolution. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, chapter 1.4. New York: Wiley. Urban, Nelson. 1979. A history of Pioneer’s first ten years. Pioneer Hi-Bred International. Manuscript, Iowa State University Library. U.S. Department of Agriculture. 1940. Technology of the farm. Washington, DC: GPO. ———. 1945–1962. Agricultural statistics. Washington, DC: GPO. U.S. Department of Agriculture, Bureau of Agricultural Economics. 1948. Tobaccos of the United States. Washington, DC: GPO. U.S. Department of Agriculture, National Agricultural Statistics Service. 2004. Historical track records [Internet publication]. Wallace, Henry Agard. 1920. Agricultural prices. Des Moines, IA: Wallace Publishing Company. ———. 1936. Yearbook of agriculture, 1936. Washington, DC: GPO. Wisner, Robert N., and E. Dean Baldwin. 2004. Corn marketing. In Corn: Origin, history, technology, and production, ed. C. Wayne Smith, Javier Betrán, and E. C. A. Runge, chapter 3.8. New York: Wiley. Zuber, Marcus S., and Joe L. Robinson. 1941. The 1940 Iowa corn yield test. Iowa Agricultural Experiment Station Bulletin P19 (NS): 519–93.

8 The Evolution of Heat Tolerance of Corn Implications for Climate Change Michael J. Roberts and Wolfram Schlenker

8.1

Introduction

With evidence accumulating that greenhouse gas concentrations are warming the world’s climate, there is growing interest in the potential impacts that may occur under different warming scenarios and on how economies might adapt to changing climatic conditions. Agriculture is of particular interest due to the fact that climate is a direct natural input in the production process. Agriculture in developed nations, and particularly in the United States, has received considerable attention. This attention may derive from the fact that wealthier nations produce a disproportionate share of the world’s agricultural commodities, at least partly due to their relatively more temperate climates. Accordingly, climate change impacts on agriculture in developed nations, and particularly the United States, the world’s largest producer, have broad implications for food supply and prices worldwide. In recent research, we conducted detailed statistical analyses of the relationship between weather and crop yields of corn, soybeans, and cotton in 1950 to 2005. These crops are among the four largest U.S. crops, all of which are important for world commodity prices (Schlenker and Roberts 2009). Corn and soybeans are two of the world’s four key staple commodities that comprise about three- quarters of calories produced worldwide (rice and wheat are the other two). The U.S. produces about 40 percent of world production in these two crops, making it, by far, the world’s largest producer and exporter. While less important for global food supply, cotton is grown in Michael J. Roberts is an assistant professor in the Department of Agricultural and Resource Economics at North Carolina State University. Wolfram Schlenker is an assistant professor in the Department of Economics and the School of International and Public Affairs at Columbia University, and a faculty research fellow of the National Bureau of Economic Research.

225

226

Michael J. Roberts and Wolfram Schlenker

the warmer Southern areas of the United States and might be better suited to warmer temperatures. We found that yields of all three crops grow roughly linearly in temperature up to a threshold, above which yield growth declines sharply. The threshold varies by crop: 29°C (84°F) for corn, 30°C (86°F) for soybeans, and 32°C (90°F) for cotton. For all three crops, the slope of the decline above the optimum temperature for yield growth is significantly steeper than the incline below the optimum temperature. Cumulative exposure to heat above the threshold is the strongest single predictor of yield outcomes. One implication is that a modest amount of warming could change, markedly, the best locations for growing these key crops. In this chapter, we extend the analysis and construct a fine- scaled weather data set for the entire twentieth century in Indiana. This prolonged period covers weather extremes of the 1930s that led to the Dust Bowl and includes observations both before and after the Green Revolution, allowing us to examine how the relationship between weather and corn yields evolved over time as new seed varieties (double- and single- crossed hybrids) were introduced. Historic adaptation to weather extremes, or the failure to do so, can give valuable insights on how difficult it is to adapt to conditions that are predicted to become more frequent under climate change. We find that the relationship between various weather measures and yield evolves over time. Most notably, the detrimental effects of too much or too little precipitation vanishes continuously over time, while tolerance to extremely warm temperatures peaks around 1960. Extrapolating the relationship we previously discovered for the entire United States while holding growing areas fixed results in severe impacts: average yields decrease by 30 to 46 percent before the end of the century under the slowest (B1) warming scenario and decrease by 63 to 82 percent under the most rapid warming scenario (A1FI) under the Hadley III climate model. These projected declines are driven by sharp yield reductions when temperatures exceed 29°C to 32°C combined with the sizable increase in the projected frequency of these extreme temperatures. There are several reasons why these projected damages might overstate actual potential damages. As the climate warms, agricultural production will work to adapt to this warming. The most difficult economic questions pertain to how large these adaptation possibilities may be. As climates change, so will geographical comparative advantages. We should not expect crops to be grown in the same locations as they are grown today. Ascertaining the potential impact of climate changes, therefore, calls for an analysis of the yield potential of major crops across the globe, even in places where agricultural production does not exist today. Such analysis can be quite complex and requires strong assumptions about the potential suitability of many crops in various climates and soils. For example, there is uncertainty about soil dynamics in the tundra, a region that is currently too cold to farm

The Evolution of Heat Tolerance of Corn

227

but might become farmable under warming. Chapin et al. (1995) conduct experiments of soil changes in Alaska and find that the three- year response in experimental plots are a bad predictor of nine- year changes in experimental plots. The authors emphasize the difficulty of predicting long- term changes using short- term heat waves. A recent study by the International Food Policy Research Institute (Nelson et al. 2009) conducts a comprehensive, worldwide analysis that incorporates shifts in growing locations. Given its inherent complexity, many assumptions enter their model. The amount of uncertainty surrounding their projections is probably unquantifiable. But this is the most recent, careful, and comprehensive study to date. The study predicts significant declines in commodity production and increases in commodity prices stemming from global warming. Calorie availability will not only be less than the no- climate- change scenario, but less than availability in 2000. South Asia will be hit particularly hard as yields for rice and wheat decrease significantly. In neither our earlier work nor in this chapter do we attempt such a comprehensive analysis. Rather, by focusing on major crops in the United States, a climatically diverse country that generates the world’s largest agricultural output and exports the most, we examine forms of potential adaptation observable in historical data. These historical adaptations (or lack thereof) may provide some insight into the scope and nature of potential adaptations that may be available as the climate changes. 8.2

Implications of Earlier Findings for Adaptation

Our earlier research found the same nonlinear relationship between yield growth and temperatures, described in the introduction, when the analysis is narrowed to consider only cooler northern U.S. states or only warmer southern U.S. states. This evidence supports the idea that the nonlinear temperature relationship is a generalizable phenomenon. Adding to this evidence, we found the same relationship if we examined only the early half of the sample (1950 to 1977) or only the latter half of the sample (1978 to 2006). This was particularly surprising given the significant increase in average yields. These comparisons suggest that innovations since 1950, while increasing average yields approximately threefold, did not increase relative heat tolerance. And because most regions of the United States currently have temperature distributions that are warmer than optimal, there has been some incentive to breed or engineer more heat tolerance into plants. Our earlier examination of heat tolerance over time was relatively crude: we merely split the sample into an earlier and later period. A key focus of new analysis presented in the following is to examine the evolution of heat tolerance more thoroughly and over a longer time period. The stability of the nonlinear temperature- yield relationship over different

228

Michael J. Roberts and Wolfram Schlenker

subsets of the data helps to provide powerful evidence of a causal link. This is particularly true as each specification includes county fixed effects to control for time- invariant heterogeneity of soils and farming practices. While cross- sectional variation in temperatures may be associated with other factors correlated with geography, county fixed effects purge this variation from the regression. Remaining variation in weather outcomes over time are arguably random from the vantage point of farmers and thereby constitute a viable natural experiment. The stability of results combined with strong exogeneity of weather variations in a fixed location are what make the empirical results persuasive.1 While correlations between time series weather variations and economic outcomes are persuasively causal, a problem with focusing on time series variations is that they cannot account for adaptation. When farmers operate in a different climate, the set of adaptation strategies will be very different compared to unanticipated changes in weather. One might be tempted to interpret short- run response to weather as a useful lower bound of the impact stemming from climate change. The idea is that adaptation would mitigate damages and exploit new opportunities that are not available in the short run. Thus, the argument goes, adaptation necessarily improves the outcome relative to the short- run response to weather. In our view, such an inference is incorrect. It is true that some decisions are available in the long run that are unavailable in the short run. But the converse is also true. For example, an aquifer with limited replenishment may provide irrigation water to help a farmer cope with a temporary drought but may be insufficient for maintaining crop production if precipitation were permanently reduced.2 Adaptations to changing climate conditions are better captured by crosssectional comparisons. The potential downside is that cross- sectional comparisons are more easily confounded by unobserved factors that happen to correlate with location. Because many economic and social factors correlate with geography, and climate itself is correlated with geography, there is a distinct possibility that any observed association between climate and an economic outcome is not causal, but rather reflects the influence of some unmeasured factor associated with location and climate. Considering the strengths and weaknesses of both cross- sectional climate variations and time series weather variations, it is important to consider both. And in this respect perhaps the most compelling finding of our earlier 1. Deschênes and Greenstone (2007) use year- to- year variation in weather to estimate the relationship between profits or yields and weather. They find that agricultural profits and yields are independent of weather. However, their weather data set contains many irregularities, and their profit measure, which is the difference between sales in a given year minus expenditures, does not account for storage behavior that smoothes profits between periods. Once the data errors are corrected, projected climate change effects on yields are again unambiguously negative (Fisher et al., forthcoming). 2. Other examples are provided in Fisher et al. (forthcoming).

The Evolution of Heat Tolerance of Corn

229

research is that both methods of comparison give very similar results. We isolated the pure cross- sectional relationship between climate and yields by pairing the average distribution of temperature and precipitation outcomes with each county’s average deviation from the nationwide U.S. yield.3 To isolate the pure time series, we paired the nationwide average yield with the crop- area- weighted average weather distribution in each year. Both of these methods of identification show the same distinctly nonlinear relationship between temperature and yield growth described in the introduction. While the cross- sectional relationship may be potentially confounded by omitted variables, it is robust to inclusion or exclusion of controls for soils and other factors. Moreover, we also find it unlikely that unobserved confounding factors would happen to align in such a manner that would give rise to the same nonlinear relationship as observed in the time series relationship that is identified with presumably random weather fluctuations. The fact that these relationships are similar suggests that, at least historically from 1950 to 2005, there has been little scope for adaptation conditional on the locations where these key crops were grown. This finding is consistent with some earlier work using the hedonic approach, which considers cross- sectional variations in climate to land values (Schlenker, Hanemann, and Fisher 2006).4 The hedonic approach also accounts for crop switching in response to climate change. 8.3

The Evolution of Weather-Yield Relationships over the Twentieth Century

Our earlier work found little evidence of adaptation to warmer temperatures between 1950 and 2006. In this chapter, we extend the analysis to include the earlier and potentially more interesting period between 1901 and 1950. Our focus on this period is motivated in large part by Sutch’s (2011) 3. Subtracting each year’s nationwide yield from each county’s yield removes the aggregate upward trend, which is substantial. 4. Mendelsohn, Nordhaus, and Shaw (1994) first introduced the Ricardian method to measure the effects of climate change on agriculture by estimating a cross- sectional relationship between county- level farmland values and climatic variables in the United States. The predicted impact of changing climatic variables depends largely on the set of weights. Under the cropland weights (fraction of a county that is cropland), the predicted impacts are severely negative, and under the crop- revenue weights (the value of agricultural production sold), the effects are beneficial. The reason why the results diverged under various weights is access to highly subsidized irrigation water rights in the western United States. These subsidized water rights capitalize into farmland values (Schlenker, Hanemann, and Fisher 2007). Because access to subsidized water rights is correlated with temperature, an increase in temperature implicitly assumes an increase in subsidies, which should not be counted as a societal benefit. The crop- revenue weights aggravate the problem because highly irrigated counties in the western United States account for a large share of overall revenues, yet the fraction of the county that is cropland (cropland weights) is small. Schlenker, Hanemann, and Fisher (2005) show that if the analysis is limited to rainfed agriculture, the results converge and become unambiguously negative under both sets of weights.

230

Michael J. Roberts and Wolfram Schlenker

research. Sutch argues that the adoption of hybrid corn, one of history’s most remarkable and well- documented technological revolutions, was precipitated in part by the extreme weather events of the 1930s. In particular, he argues that hybrid corn demonstrated high yields relative to open- pollinated (nonhybrid) corn during 1934 and 1936, which (by our own key crop- related temperature measures) remain the most extreme on record. Thus, it could be that our earlier analysis did not look back far enough to the timing of the key innovation leading to the Green Revolution. Specifically, in this chapter, we examine a panel of corn yields from 1901 to 2005, a time period that includes a full thirty- five years before the beginning of the Green Revolution as well as some seventy years after the first adoption of hybrid corn. Our analysis focuses on the state of Indiana, which sits in the middle of the so- called Corn Belt and is the nation’s third largest corn growing state. Our focus on Indiana is mainly due to data availability: it turns out that Indiana has the most comprehensive record of detailed daily weather records in the station data maintained by the National Climatic Data Center. Detailed daily weather data are necessary to estimate the effect of the entire temperature distribution on yields. The data accounts for variations in temperatures, both within and across all days of each growing season. This detail facilitates correct identification of nonlinear temperature effects, which can be diluted from measurement error, or if temperatures are averaged over time or space. The key focus of our analysis is to examine how heat tolerance and drought tolerance has changed over time, with some particular focus on the time period following the great heat waves of 1934 and 1936 and subsequent widespread adoption of hybrid corn. 8.3.1

Data: A Century of Yields and Weather in Indiana

Figure 8.1 shows corn yields in Indiana over the twentieth century. These yield data are publicly available from the U.S. Department of Agriculture’s National Agricultural Statistical Service (USDA-NASS). All of our data sources are described in further detail in the data appendix. The graph shows the average yield in the state for all years between 1901 and 2005 as black diamonds. For years after 1928, when county- level data becomes available, a box plot shows the range and interquartile range of yields across counties in Indiana. Before 1940, there was no discernible trend in yields. This is true even if one were to extend the time series back many decades before 1901, the earliest year shown in the figure. Around 1940, yields started a sharp upward trend that appears ongoing even today. Typical yields in Indiana were between 30 and 40 bushels per acre before 1940, yet today, a typical Indiana farmer can expect 150 to 160 bushels per acre. Yield variance increased along with typical yields, so we model the natural log of yield per acre. As discussed in great detail by Sutch (2011), the beginning of the upward

The Evolution of Heat Tolerance of Corn

231

Fig. 8.1 Average yields in Indiana, 1901–2008 Notes: The graph shows history of corn yields in Indiana. State- level averages are shown as diamonds. The range of yields among Indiana’s counties is shown as boxplots: The box gives the 25 percent to 75 percent quartile range, the median is shown as a solid line, and whiskers extend to the minimum and maximum. A locally weighted regression of the state averages (bandwidth of ten years) is shown as a black line.

trend in yields began around the time when many key events occurred simultaneously. The Great Depression in the 1930s was followed by the onset of World War II in 1938, which caused large fluctuations in commodity prices. At least equally important was the early adoption of hybrid corn, starting in Iowa and quickly expanding to Illinois, Indiana, and beyond. The superior yields of hybrid corn was discovered in 1918, but it was not until later, perhaps after 1936, that seed production became commercially viable and high- yielding enough for farmers to adopt. Also, in the decade before 1940, the Midwest, including Indiana, experienced both the hottest and driest temperatures on record for the growingseason months between March through August, shown in figures 8.2 and 8.3. The former shows yearly weather shocks in extreme heat (degree days above 29°C, further described in the following) over the growing season. The latter shows precipitation deviations from average climatic conditions. The decade of poor weather in the 1930s was most accentuated in the two drought years of 1934 and 1936, which brought about the great Dust Bowl, an event of massive wind erosion in states west and south of Indiana. In those years,

232

Michael J. Roberts and Wolfram Schlenker

Fig. 8.2 Shocks in extreme heat in Indiana, 1901–2005 Notes: The graph shows weather shocks (deviations from averages) for degree days above 29°C during the growing season March to August. State- level averages are shown as diamonds. The range of weather shocks among Indiana’s counties is shown as boxplots: The box gives the 25 percent to 75 percent quartile range, the median is shown as a solid line, and whiskers extend to the minimum and maximum. A locally weighted regression of the state averages (bandwidth of ten years) is shown as a black line.

average yields in Indiana were just 27.6 and 25.6 bushels per acre, two of the three worst yields on record for the state during the twentieth century. Note that drought years also showed the largest exposure to extreme heat as temperatures and precipitation are interrelated. Indiana still fared much better than states west and south of Indiana. Iowa harvested just 60 percent of its planted acreage in 1934, an all time low, and Dust Bowl states of Nebraska and Kansas lost nearly all of their corn plantings in these years. It is interesting to note that more recently, and particularly in the last two decades, the weather has been good for corn yields. This is in sharp contrast with what climate models project in the decades to come. Under the slowestwarming scenario (B1) in the Hadley III model, average projected extreme heat for the years 2070 to 2099 is predicted to increase by 103 degree days above 29°C compared to the 1960 to 1989 baseline under the Hadley III model. It is added as a horizontal line in figure 8.2. While the B1- scenario assumes we curb CO2 emissions sharply in the near future, the predicted increase is still worse than the worst of the Dust Bowl years. Average pro-

The Evolution of Heat Tolerance of Corn

233

Fig. 8.3 Precipitation shocks in Indiana, 1901–2005 Notes: The graph shows weather shocks (deviations from averages) for total precipitation during the growing season March to August. State- level averages are shown as diamonds. The range of weather shocks among Indiana’s counties is shown as boxplots: The box gives the 25 percent to 75 percent quartile range, the median is shown as a solid line, and whiskers extend to the minimum and maximum. A locally weighted regression of the state averages (bandwidth of ten years) is shown as a black line.

jected increase in extreme heat under fast- warming A1FI scenario is way off the chart at 310 additional degree days above 29°C. Construction of the weather variables presented in figures 8.2 and 8.3 is further detailed in the appendix. We construct these data from daily individual weather stations in Indiana. Geographical interpolation is achieved by linking it with the PRISM weather data sets, which gives monthly observations on a 2.5  2.5 mile grid for the entire United States. Indiana is the only state in the United States for which the National Climatic Data Center of the National Oceanic and Atmospheric Administration reports having more than three weather stations in the early part of the century. The availability of good, fine- sale weather data is essential for identifying nonlinear weather effects because these effects can be diluted with measurement error or if values are averaged over time and space. The geographical locations of weather stations in Indiana that we use to construct our data set for each twenty- five- year period are shown in figure 8.4. The challenge for a regression model that relates yields to weather out-

Notes: Weather stations used in interpolation are displayed as circles for minimum temperature, triangles for maximum temperature, and diamonds for precipitation.

Fig. 8.4 Weather stations used in Indiana, by time period

The Evolution of Heat Tolerance of Corn

235

comes is in mapping an entire season of temperature and precipitation outcomes to a single yield outcome. We achieve this by assuming temperature effects on yields are cumulative over time and that yield is proportional to total exposure. This implies temperature effects are additively substitutable over time. That is, we sum the daily outcomes associated with each temperature over all days of the growing season. The benefit of assuming additive separability is that it allows us to keep the underlying relationship between temperatures and yields fully flexible. Earlier work has shown that there are three weather variables that give the best out- of- sample predictions of corn yields: (a) total precipitation pit in county i in year t; (b) degree days above 29°C (dd H it ), which captures the harmful effects of high temperatures; and (c) degree days between 10°C and 29°C degrees (ddM it ), which measures the beneficial effects of moderate temperatures (Schlenker and Roberts 2009). Each measure is simply a truncated integral over the temperature distribution within each day and then summed over all days in the growing season, as given in the following. Degree days above 29°C (high temperature measure) are defined as A ugust 31st

dd  H it

∑

∞

∫

j = M arch 1st T =29

(T  29)hitj (T )dT,

where T is temperature (in degrees Celsius) and hitj (T ) is the estimated density of time at each degree during day j in year t in county i. Because the measure is sensitive to geographic variation in temperatures, as wells as variations within and across all days of the growing season, we spend considerable care in estimating hitj (T ). Further details are given in the data appendix. The second temperature measure is degree days between 10°C and 29°C (moderate temperature measure) are defined as A ugust 31st

dd  M it

8.3.2

∑

∞

∫

j = M a r c h 1 s t T =1 0

min(T  10, 19)hitj (T )dT.

Regression Model

In this chapter, we take as given the two temperature measures that our earlier work found to be the best predictor of corn yields in 1950 to 2005. Our focus is to explore how the relationship between yields and weather has changed over the 105 years from 1901 to 2005. We use a flexible restricted cubic spline model that allows temperature and precipitation associations to change smoothly and flexibly over time. Specifically, the regression model is H M H yit  0ddM it  1dd it  fp( pit)  ft(t)  fM(t)  dd it  fH(t)dd it  ft2(t)fp2( pit)  ci  it, H where yit denotes the natural log of yield in county i and year t, dd M it and dd it are the degree day measures described in the preceding, pit is total precipita-

236

Michael J. Roberts and Wolfram Schlenker

tion. The functions fx() are cubic splines of time or precipitation.5 We also include separate intercepts for each county (i.e., fixed effects, denoted ci ) to account for unobserved time- invariant heterogeneity, like soil quality. Because we combine state- level averages before 1929 with county- level averages starting in 1929, we use the corn acreage as weights in the regression equation to make the two sets of aggregation measures comparable. Estimation of restricted cubic spline models is easily done using ordinary least squares. Because the errors within each year are likely correlated in space, we adjust our standard errors to account for this (clustering the errors by year) and possible heteroscedasticity using the Huber-White method. 8.3.3

Results

Figure 8.5 shows the effects of each of the four variables: time, precipitation, ddM, and dd H , while holding all other variables at their median values. These results are characteristically similar to what we found in our earlier work that focused on the period from 1950 to 2005: there is a sharp upward trend in yields over time as shown in the top- left panel. Yields have an inverted-U shape, with rainfall as shown in the top right panel. Yields increase gradually with temperate degree days between 10°C and 29°C as shown in the bottom- left panel. Finally, yields decline sharply with extreme heat, measured as degree days above 29°C, as shown in the bottom- right panel. Because all regressions include county fixed effects, the graph will be shifted up or down by the county- specific intercepts. We hence normalize each graph and display impacts relative to optimal outcome of each variable in question. For example, the top- right panel shows by how much yields decline if precipitation deviates from the optimum for the season. All four panels of figure 8.5 use the same scale on the y- axis to make the contribution of each variable comparable across plots. The time trend is responsible for the largest effect followed by degree days above 29°C. These median- value predictions, however, do not show how these relationships have changed over time. We explore how these relationships change over time in figure 8.6 for precipitation, figure 8.7 for extreme heat dd H , and figure 8.8 for moderate temperatures dd M. Each of these figures plots the relationship of the three weather variables at fifteen points in time.6 The effects of all three weather variables have shifted markedly over time. Figure 8.6 shows that the influence of precipitation continuously vanishes 5. In the baseline model each of the spline functions is approximated using 5 knots, located at the 0.05, 0.275, 0.5, 0.725, and 0.95 quantiles of the empirical distribution of the relevant explanatory variables. For the time trend, knot locations are 1932, 1949, 1967, 1984, and 2001. The early knot in the time trend is due to the fact that we have only state- level observations prior to 1929 and, thus, fewer data points per year than after 1929 when we have county- level observations. To check the stability of the results to specification, we also estimated models with 3, 4, 6, and 7 knots for each spline function in the following. 6. We report the confidence bands obtained from the R package Design after clustering errors by year.

The Evolution of Heat Tolerance of Corn

237

Fig. 8.5 Regression results at median values Notes: The graphs display regression results of one variable while keeping other variables at median outcomes (shown as vertical line in each graph): top- left panel shows yield trend, topright panel shows effect of precipitation, bottom- left panel shows effect of degree days 10°C to 29°C, and bottom- right panel shows effect of degree days above 29°C. Graphs are normalized relative to the optimal outcome of a variable.

over time. Deviations from the optimal precipitation levels have limited effects on yields in 2000. We believe that two explantations are most likely responsible for the fact that yields are no longer directly linked to rainfall during the growing season. First, a lack of precipitation in the growing season might be counterbalanced with irrigation. Continued mechanization of agriculture has led to the gradual expansion of pivot irrigation systems that can provide supplementary water during especially dry periods. While only a minority of corn fields in Indiana have pivot irrigation systems, the ones that do are probably more prone to dryness or have sandier soils. Second, seed companies may have bred increased drought tolerance into corn plant varieties. While climate models vary considerably in their predictions for precipitation changes, with some forecasting increases and others decreases, evidence from weather and yields in Indiana suggest this may be of little economic consideration. The evolution of heat tolerance, displayed in figure 8.7, differs from that of precipitation. Heat tolerance increased until 1960 followed by a decline

Notes: The graphs display the effect of total precipitation during the growing season on log yields at fifteen periods in time. Graphs are normalized relative to the best value of precipitation in each year. Ninety-five percent confidence bands are added.

Fig. 8.6 The evolution of the impact of precipitation on log corn yields

Notes: The graphs display the effect of extreme temperatures during the growing season on log yields at fifteen periods in time. Graphs are normalized relative to the best value of degree days above 29°C. Ninety-five percent confidence bands are added.

Fig. 8.7 The evolution of the impact of extreme heat on log corn yields

Notes: The graphs display the effect of moderate temperatures during the growing season on log yields at fifteen periods in time. Graphs are normalized relative to the best value of degree days 10°C to 29°C. Ninety-five percent confidence bands are added.

Fig. 8.8 The evolution of the impact of moderate heat on log corn yields

The Evolution of Heat Tolerance of Corn

241

Fig. 8.9 The evolution of the marginal impact of extreme heat on log corn yields Notes: The graph displays the marginal effect of extreme temperatures (degree days above 29°C) on log yields, that is, the slope of the regression lines in figure 8.7. Cubic splines with various number of knots are used.

after 1960. Figure 8.9 shows the marginal effect of extreme heat, that is, the slope of the regression line in figure 8.7 over all years in our sample. The negative influence of an additional degree day above 29°C is lowest around 1960 and most damaging in recent years when corn varieties were optimized for maximum average yields. The magnitude of the negative coefficient on dd H is nearly three times as large in 2000 as it is in 1960, and about twice as large in 1901 as compared to 1960. This result is qualitatively insensitive to how many knots we use in the spline once we include at least 4 knots to make the model flexible enough to capture the nonlinearities. Figure 8.10 replicates this analysis for the marginal effect of moderate temperature as measured by degree days between 10°C and 29°C. Estimated slopes in the early years of figure 8.9 should be interpreted with some caution because there are much fewer data points before 1929 as only state- level data are available. Our spline model places more emphasis on subperiods with more data and linearizes the model in the tails of the data. Closer inspection of the data do suggest that much of the increase in heat

242

Michael J. Roberts and Wolfram Schlenker

Fig. 8.10 The evolution of the marginal impact of moderate heat on log corn yields Notes: The graph displays the marginal effect of moderate temperatures (thousand degree days 10°C to 29°C) on log yields, that is, the slope of the regression lines in figure 8.8. Cubic splines with various number of knots are used.

tolerance actually took place between 1940 and 1960, rather than being a steady smooth trend up from 1901. This interpretation would be consistent with the relatively stable farming technologies between 1901 and 1936 and rapid technological progress after 1940. This would also be consistent with Sutch’s historical account of the adoption of hybrid corn. The most interesting and relevant finding that speaks to implications for climate change is the sharp decline in tolerance to extreme heat since 1960. This finding is a powerful counterpoint to the apparent increase in drought tolerance. Under the latest climate change models, a sharp rise in maximum temperatures is predicted to significantly increase the occurrence of temperatures above 29°C. Because degree days above 29°C are a truncated temperature variable, modest shifts in the temperature distribution can have a large relative influence on this temperature measure. For example, a 1°C warming from 29.5°C to 30.5°C triples degree days above 29°C. The historic average number of degree days above 29°C is twenty- five in Indiana. Under the Hadley II model (IS92a scenario), the number is predicted to increase by nineteen at the end of this century. Under the much warmer Hadley III model, degree days above 29°C are projected to increase by 103 under the

The Evolution of Heat Tolerance of Corn Table 8.1

243

Analysis of variance for log yield

Degree days above 29°C All terms Interaction terms with time Degree days 10°C–29°C All terms Interaction terms with time Precipitation All terms Interaction terms with time Time trend All terms Time trend only R2 (all variables)

d.f.

F-statistic

p-value

5 4

18.99 2.82

0.0001 0.0289

5 4

5.07 3.30

0.0003 0.0138

11 7

4.09 2.00

0.0001 0.0624

19 4 0.95

83.80 9.08

0.0001 0.0001

Notes: Table reports F-tests for the joint significance of key explanatory variables and their interactions with time. Our baseline model uses restricted cubic regression splines with 5 knots, which will result in four factors (variables) in the regression equation. The weather variables Degree days above 29°C and Degree days 10°C to 29°C consist of the weather variable (1 degree of freedom [d.f.]) as well as the interactions with the four time factors (4 d.f.). The weather variable Precipitation consists of four factors in the amount of precipitation (4 d.f.) as well as the interaction of the linear time and precipitation term (1 d.f.) and the interaction of precipitation with the three higher order precipitation factors and vice versa (3 d.f. each). Finally, the Time trend consists of four factors and it is interacted with the fifteen terms outlined for the three weather variables. We use both the STATA command mkspline as well as the R-package Design. The point estimates are identical, but the clustering option is implemented differently in both languages. We report the results from STATA, which tend to be more conservative (with the exception of precipitation).

slow- warming B1 scenario. Thus, even under the slowest- warming scenario, typical weather outcomes in the latter part of this century are projected to be far worse than the worst drought years in the historical record, 1934 and 1936 (refer to figure 8.1). Under the fastest- warming A1FI scenario, degree days above 29°C are projected to increase by 310, making the measure in a typical year about 3.5 times worse than the worst year on record. Finally, the relationship between the precipitation and log yield is highly significant, but the interaction between time and precipitation has a p- value of 0.06. For all factors besides precipitation, both the combined effect as well as nonlinear interactions with time, are significant at the 5 percent level, suggesting that the relationship was not stable over the century but has evolved. A summary of significance tests is reported in table 8.1. 8.3.4

Discussion of New Results

The last section extended earlier research on the link between weather and yields by examining how key weather variables are associated with corn yields in Indiana over the time period 1901 to 2005. We use restricted

244

Michael J. Roberts and Wolfram Schlenker

cubic spline regressions to let the effect of precipitation, moderate heat, and extreme heat evolve smoothly over time in a flexible way. Results for each variable, while holding all other variables constant at their median observed outcomes, are comparable to earlier results we obtained for a model using county- level corn yields for all counties east of the 100 degree meridian in the years 1950 to 2005. The median association, however, obscures significant evolution of precipitation and temperature effects over time, effects that we had not examined in our earlier research. The overall influence of precipitation during the growing season has diminished with time.7 We hypothesize that attenuation of precipitation effects stems from increased use of supplemental irrigation and possibly the development of more drought tolerant seed varieties and cropping systems that have increased planting densities and canopy cover of the soil. The evolution of temperature effects looks rather different from that of precipitation effects. The evolution of heat tolerance over time is nonlinear, increasing sharply between about 1940 and about 1960 and then declining. We found corn in Indiana to be most sensitive to extreme heat in the more recent years of our sample. The later decline in heat tolerance might be due to the fact that maximizing corn plants for average yields also makes them more sensitive to suboptimal growing conditions. It is interesting to note that the key turning points in evolution of heat tolerance align almost perfectly with the adoption of double- cross hybrid corn (around 1940) and single- cross hybrid corn (around 1960). It is also notable that, from inspection of Richard Stuch’s figure showing U.S. aggregate corn yields from 1866 to 2002 and our own figure 8.1, yields became noticeably more variable as corn transitioned from double- cross to single- cross varieties, a pattern that could be indicative of greater heat sensitivity. Why did we find relative heat tolerance to be stable in our earlier study and not in this one? We believe there are several interrelated reasons. First, our earlier study began in 1950, a while after first adoption of hybrid corn and growth in heat tolerance, but well before hybrid corn had been universally adopted in all states. Second, we simply split the sample into two subperiods, 1950 to 1977 and 1978 to 2005, while pooling all states east of the 100th median. Because different states adopted hybrid corn at different times and heat tolerance grew and then declined, our regressions would have picked up average heat tolerance in each subperiod. When pooling all states, it is likely that average heat tolerance was about equal in these two subperiods. Note that Indiana was relatively early on the adoption curve for hybrid corn. 7. A cross- validation analysis shows the fine- scale precipitation data to be less accurate than the fine- scale temperature data. Because error in an explanatory variable causes attenuation bias, it is likely that precipitation is more important in reality than our regressions imply. But because the data are likely more accurate in the recent period as compared to the earlier period, attenuation bias cannot explain the general trend of decreasing importance of precipitation.

The Evolution of Heat Tolerance of Corn

245

Our new findings have mixed implications for climate change impacts: on the one hand, sensitivity to extreme heat is highest at the end of the sample, and the one feature all climate models agree on is that these extreme heat events are likely to increase, even though the size of the increase varies tremendously between model and emission scenarios. On the other hand, there was a period between 1940 to 1960 when both heat tolerance and average yields increased at the same time. The question is whether recent increases in yields could only be achieved by making plants less heat resistent or whether future breeding cycles can increase both heat tolerance and average yield at the same time. 8.4

Conclusions

Since the late 1930s when U.S. farmers began using hybrid corn, commercial fertilizers and other modern farming techniques, average crop yields in the United States and around the world have grown tremendously. Today, corn yields in the United States equal more than four times the best yields of the 1930s. Yields of most other staple commodities have more than tripled. Over the same time period, world population grew slightly less than threefold. Higher yields have brought lower commodity prices, which have relieved hunger in less- developed nations and have fed a growing (and likely unhealthful) appetite for meat and processed foods in rich countries. Yield growth has probably also attenuated expansion of cropping areas and deforestation. Recent adoption of genetically modified seeds have a spurred yield gains in developing nations that have adopted them (Qaim and Zilberman 2003) and may hold promise for further yield gains in both developed and developing nations. But global warming now poses a significant threat to crop yields. Crop scientists have long predicted that warming will cause yield declines in tropic and subtropic regions of the world. Climates in these regions are already too warm for optimal growing conditions for most crops, so further warming will not help. More recent evidence suggests warming will also harm yields in more temperate regions where current production is greatest. Our previous statistical analysis of the United States, by far the world’s largest producer and exporter of agricultural commodities, is dismal. Holding growing areas fixed (an important caveat), we predict yield declines of 38 to 46 percent for soybeans and corn between 2070 to 2099 under the Hadley III slow- warming scenario (B1, which presumes sharp reductions in CO2 emissions), and declines of 75 to 82 percent under the Hadley III fast- warming scenario (A1FI, which presumes the fastest growth in CO2 emissions). Projected declines in medium term (2020 to 2049) are also substantial, 18 to 23 percent under the slow- warming scenario and 22 to 30 percent under the fast- warming scenario. The largest driver behind these reductions is the predicted increase in very hot temperatures. It is important to note that these predicted declines are relative to what yields would be without climate

246

Michael J. Roberts and Wolfram Schlenker

change, not what yields are today. They also hold growing locations fixed and do not account for CO2 fertilization, which may increase yields. One way of adapting to warmer climates will be to change the locations where crops are grown. Corn and soybean production is likely to shift northward toward traditional wheat growing regions and wheat (perhaps) to areas that were not previously cropped. Given the world’s currently most productive areas are predicted to be harmed significantly, it is not clear how much of these losses may be mitigated by crop switching. A team of researchers at the International Food Policy Research Institute, led by economist Gerald Nelson, recently developed the most comprehensive analysis to date (Nelson et al. 2009, vii). Their model accounts for yield effects, crop switching, trade, and price effects throughout the world, but takes population and gross domestic product (GDP) as exogenous to agriculture and does not account for sea- level rise, which could be important for rice production in south Asia. They predict that by 2050, calorie availability “will not only be lower than in the no- climate- change scenario—it will actually decline relative to 2000 levels throughout the developing world.” Thus, at present, it would appear that technological solutions, in addition to crop switching, will be necessary to overcome anticipated impacts from global warming. It is in this vein that we have explored historical innovation as it relates to heat tolerance. In particular, we examined the evolution weather effects on corn yields in Indiana and how these effects have changed over time with adoption of new crop varieties and farming techniques. Sensitivity to extreme heat is critical determinant of corn yields. In recent research, we have found this sensitivity to be similar in warmer southern states and cooler northern states. Moreover, we found no evidence that warmer areas have adapted to warmer- than- optimal climates: the crosssection of yields and climate matches the link between yields and weather in a fixed location. In this chapter, we present new evidence that may be somewhat more encouraging. We find that, following the Dust Bowl years—the hottest, driest and lowest- yielding years on record—heat tolerance in corn grew markedly until about 1960. After 1960, however, heat tolerance declined, even though average yields continued their steady rise. At the end of our sample in 2005, corn appears to be less tolerant to extreme heat than it was in the 1930s. The key question is whether plant scientists and seed companies can continue to breed or engineer crops that have both greater yield potential and greater tolerance to extreme heat. At present, these prospects seem uncertain, and greater agricultural productivity investments would seem prudent. The private sector may foresee higher future commodity prices and, thus, engage in these investments on their own. There may also be a role for public sector investments in basic research, particularly because these have been the source of critical innovations in the past. Such innovations have important positive spillovers that can lead to suboptimal private investment.

The Evolution of Heat Tolerance of Corn

247

On the demand side, we believe it important to recognize that global income inequality is a critical obstacle to adaptation. The issue is not so much whether it will be technically feasible to feed the world’s population; we see little doubt that it will be. But when median incomes of the richest nations are hundred times those of the poorest nations, it is easy to see how lower commodity supply combined with, say, a taste for meat in rich countries, could drive prices of staple commodities to the point that the poorest simply cannot afford to survive. Despite the necessity of food, demand response of the poor is larger than that of rich due to a much larger income effect. There is no market failure or malthusian cycle in this story. It’s simply a matter of income inequality. If incomes were not so divergent, prices would simply rise until enough people substituted to a presumably more healthy diet with less meat. The greatest hope is an uncertain one: that technological change will obviate the need for behavioral change.

Appendix Data Appendix This appendix outlines in further detail how we construct our data set. Yield Data Yield data was obtained from the National Agricultural Statistics Service (accessed March 2009). Yearly state- level yields in Indiana are available from 1866 onward.8 County- level yields in Indiana are available starting in 1929.9 We follow the definition of the Department of Agriculture and calculate yields as the ratio of total production divided by area harvested. The traditional definition of yields might overstate actual yields if some fields are not harvested. In a sensitivity check, we define yields as total production divided by all acres planted. Unfortunately, area planted is only available from 1926 onward for state totals and from 1972 for individual counties and, hence, significantly reduces our sample period. The left panel of Figure 8A.1 displays the fraction of the planted area that was harvested in Indiana over time. While there is an upward trend, especially during the 1930s, the right panel shows that the year- to- year variation in yields is similar for each definition of yields. Weather Data Degree days were constructed from daily weather data. We obtained daily observations from the National Climatic Data Center (NCDC) Cooperative 8. See http://www.nass.usda.gov/QuickStats/Create_Federal_All.jsp. 9. See http://www.nass.usda.gov/QuickStats/Create_County_All.jsp.

248

Michael J. Roberts and Wolfram Schlenker

Fig. 8A.1 Fraction of corn area planted that is harvested Notes: The left panel shows the ratio of the corn area harvested to the area planted in Indiana in 1926 to 2005 as diamonds as well as a locally weighted regression with a bandwidth of one decade as solid line. The right panel shows yields under the two different definitions. Production divided by area harvested is shown as diamonds, and production divided by area planted as triangles.

station network.10 The data include daily minimum and maximum temperature as well as precipitation. While the NCDC data has great temporal coverage, we combine it with the PRISM weather data set that provides better spatial coverage.11 The latter gives monthly minimum and maximum temperatures on a 2.5  2.5 mile grid for the United States from 1895 onward. To construct a consistent set of weather data, we followed the following procedure for each twenty- five- year period starting in 1901, 1910, 1920, 1930, . . . , 1980. (i) For each of our three weather variables (minimum and maximum temperature as well as precipitation), we determine the set of stations with a consistent record, which we chose to be stations that moved at most by 2.5 miles during the time period and had at most three missing values in at least 90 percent of the months. (ii) We fill the missing observations at stations with consistent records obtained in step (i) by regressing daily values at each station on daily values at the seven closest stations including half- month fixed effects. We use a linear regression for minimum and maximum temperature and a tobit regression for precipitation, which has several observations at the truncation value of zero. Intuitively, the regression estimates are used to fill the missing values with a weighted average of surrounding stations with nonmissing observa10. See http://ols.nndc.noaa.gov/plolstore/plsql/olstore.prodspecific?prodnumC00447CDR-S0001. 11. See http://www.prism.oregonstate.edu/.

The Evolution of Heat Tolerance of Corn

249

tions to give us a complete weather record at the stations with consistent weather records. (iii) We calculated monthly averages for the stations with consistent records in step (i). (iv) We regress the monthly values at each PRISM grid on the monthly averages at the seven closest weather stations from step (iii) including month fixed effects, again using a linear regression for minimum and maximum temperature and a tobit model for precipitation. The R- squares are generally in excess of 0.999, suggesting that the PRISM data set is a weighted average of individual stations, and we uncovered the weights. (v) We apply the regression results from step (iv) to the daily weather station data from step (ii) to derive daily weather measures at each 2.5  2.5 mile PRISM grid cell. (vi) We fit a sinusoidal curve between the minimum and maximum temperature of each day to calculate degree days accounting for the within day distribution of temperatures (Snyder 1985). We evaluate degree days for each bound between –5°C and 50°C using 1° steps at each 2.5  2.5 mile PRISM grid. Once we have the daily observations on the PRISM grid, we aggregate them spatially. (vii) We obtained the fraction of each PRISM grid cell that is cropland from a one- time LandSat satellite scan in 1992. County- level weather variables are the cropland- weighted average of all PRISM grid cells in a county. (viii) State- level weather data are the weighted average of all county- level measures in step (vii), where the weights are the amount of harvested corn area reported in the yield data. Because harvested corn area is not reported on a county level before 1929, we use the average harvested corn area in each county in the years 1929 to 2005 as weights for years prior to 1929. Finally, we aggregate the data temporally. (ix) We define the growing season as the months March through August and add degree days as well as precipitation for all days in these months. Because total precipitation over the growing season is insensitivity to the within- day and between- day distribution, we use the monthly totals in the PRISM data set. For possibly daily interactions between precipitation and temperature, we use the interpolated daily precipitation data. Because it was impossible to get a sufficiently large set of weather stations that had consistent nonmissing records for the entire sample period 1901 to 2005, we instead derived the measure for twenty- five- year intervals, starting in 1901, 1910, 1920, up to 1980. The results of interpolation series for extreme heat in the state of Indiana (degree days above 29°C) are displayed in figure

250

Michael J. Roberts and Wolfram Schlenker

Fig. 8A.2 Interpolation accuracy, 1901–2005 Notes: The graph shows degree days above 29°C in Indiana for each overlapping twenty- fiveyear interpolation period starting in 1901 to 1925, 1910 to 1935, . . . , until 1980 to 2005.

8A.2. They appear to overlap tightly. One might still wonder whether the state results hide the fact that there are substantial errors in the county- level data that get averaged out. To examine this further, we take the difference of all overlapping series in the county data. The mean absolute difference is 2.2 degree days above 29°C, and the root mean squared prediction error is 3.1 degree days above 29°C, suggesting that the overlapping fit is reasonably close. Our weather data uses the average of all overlapping series.

References Chapin, F. Stuart, Gaius R. Shaver, Anne E. Giblin, Knute J. Nadelhoffer, and James A. Laundre. 1995. Responses of arctic tundra to experimental and observed changes in climate. Ecology 76 (3): 694–711. Deschênes, Olivier, and Michael Greenstone. 2007. The economic impacts of climate change: Evidence from agricultural output and random fluctuations in weather. American Economic Review 97 (1): 354–85. Fisher, Anthony C., W. Michael Hanemann, Michael J. Roberts, and Wolfram Schlenker. Forthcoming. The economic impacts of climate change: Evidence from

The Evolution of Heat Tolerance of Corn

251

agricultural output and random fluctuations in weather: Comment. American Economic Review, forthcoming. Mendelsohn, Robert, William D. Nordhaus, and Daigee Shaw. 1994. The impact of global warming on agriculture: A Ricardian analysis. American Economic Review 84 (4): 753–71. Nelson, Gerald C., Mark W. Rosegrant, Jawoo Koo, Richard Robertson, Timothy Sulser, Tingju Zhu, Claudia Ringler, et al. 2009. Climate change: Impact on agriculture and costs of adaptation. International Food Policy Research Institute Technical Report. Qaim, Martin, and David Zilberman. 2003. Yield effects of genetically modified crops in developing countries. Science 299 (5608): 900–902. Schlenker, Wolfram, W. Michael Hanemann, and Anthony C. Fisher. 2005. Will U.S. agriculture really benefit from global warming? Accounting for irrigation in the hedonic approach. American Economic Review 95 (1): 395–406. ———. 2006. The impact of global warming on U.S. agriculture: An econometric analysis of optimal growing conditions. Review of Economics and Statistics 88 (1): 113–25. ———. 2007. Water availability, degree days, and the potential impact of climate change on irrigated agriculture in California. Climate Change 81 (1): 19–38. Schlenker, Wolfram, and Michael J. Roberts. 2009. Nonlinear temperature effects indicate severe damages to U.S. crop yields under climate change. Proceedings of the National Academy of Sciences 106 (37): 15594–98. Snyder, R. L. 1985. Hand calculating degree days. Agricultural and Forest Meterology 35 (1- 4): 353–58. Sutch, Richard. 2011. The impact of the 1936 Corn Belt drought on American farmers’ adoption of hybrid corn. In The economics of climate change: Adaptations past and present. Eds. Gary D. Libecap and Richard H. Steckel. Chicago: University of Chicago Press.

9 Climate Variability and Water Infrastructure Historical Experience in the Western United States Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

9.1

Introduction

There is a large and rapidly growing literature on climate change (Stern 2007; Nordhaus 2007; Weitzman 2007 and references therein). Agriculture is particularly vulnerable to the impacts of climate change. For the United States, the estimated effects are often mixed, with findings of nonlinearities in key commodity yields beyond threshold temperatures; predictions of higher profitability for U.S. agriculture; and reports of high adjustment costs (Mendelsohn, Nordhaus, and Shaw 1994; Cline 1996; Kelly, Kolstad, and Mitchell 2005; Schlenker, Hanemann, and Fisher 2006; Deschênes and Greenstone 2007; Schlenker and Roberts 2008). These climate change studies generally rely upon contemporary data or simulations. Greater historical perspective, however, would enlighten current debate about the effects of Zeynep K. Hansen is a professor of economics at Boise State University, and a research associate of the National Bureau of Economic Research. Gary D. Libecap is the Donald Bren Distinguished Professor of Corporate Environmental Management and professor of economics at the University of California, Santa Barbara, and a research associate of the National Bureau of Economic Research. Scott E. Lowe is an assistant professor of economics at Boise State University. We thank the National Science Foundation Idaho EPSCoR program and the Boise State University Office of Sponsored Programs for funding. We are indebted to countless colleagues and seminar participants for very helpful comments, including participants in the National Bureau of Economic Research (NBER) Climate Change Past and Present: Uncertainty and Adaptation conference (2009 and 2010), the National Water Resource Association conference (2009), the Economic History Association meetings (2009) and the NBER Development of the American Economy meeting (2010). The work has benefitted from insightful comments from Price Fishback, Sumner La Croix, Sian Mooney, Alan Olmstead, Richard Steckel, Hendrik Wolff, and anonymous reviewers. The opinions and conclusions expressed in this chapter are those of the authors as are any remaining errors.

253

254

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

climate change and future human responses to it. Indeed, the expansion of agriculture across North America in the nineteenth and twentieth centuries encountered greater climatic variation than is predicted under current climate change models (Olmstead and Rhode 2008). Accordingly, analysis of how those conditions were addressed historically and their impacts on crop mixes and agricultural production can provide valuable information for addressing current climate variability. This study adds to the literature on adaptation to climate fluctuation and change. Much academic and policy concern has been focused on the mitigation of potential climate change through international efforts to control greenhouse gas emissions. Examples include the Kyoto Protocol and other national policies to implement cap and trade programs, as well as the shifting of energy production toward less- polluting sources, such as wind and solar power. Adaptation has received somewhat less attention. Yet it is increasingly evident that adaptation must be given more consideration because the stock of greenhouse gasses may result in climate change regardless of mitigation efforts and because of the vulnerabilities of many of the world’s poorest societies. As Nordhaus observed, “mitigate we might; adapt we must” (Pielke 1998, 160). According to the Intergovernmental Panel on Climate Change ([IPCC] 2001), “adaptation refers to adjustments in ecological, social, or economic systems in response to actual or expected climatic stimuli and their effects or impacts. It refers to changes in processes, practices, and structures to moderate potential damages or to benefit from opportunities associated with climate change.” In our study, we examine planned adaptation that involves deliberate policy decisions. Our specific concerns are water resources and the investment in infrastructure dedicated to irrigation that could help to mitigate the effects of more variable precipitation levels and periods of drought on agricultural production. Agriculture is particularly sensitive to changes in water supplies. Water infrastructure investments in the United States in the twentieth century, designed to address semiarid conditions and drought (as well as flooding), provides a natural experiment to assess the impacts of such policies on agricultural production. The land west of the 100th meridian is North America’s driest and has its most variable climate (Lettenmaier et al. 2008). Further, there is an indication of increases in the duration and severity of drought in western regions (Andreadis and Lettenmaier 2006). Concerns about the variability of water supplies in the West are not new—much of the present water supply infrastructure was constructed in the late nineteenth and early to mid- twentieth centuries due to historical demand for agricultural irrigation, flood protection, drinking water, and hydroelectric power. The extent to which this investment assuaged the impacts of climate instability is a focus of this study. Using historical county- level data for five western states, we examine if and how the water supply infrastructure stabilized agricultural

Climate Variability and Water Infrastructure

255

production during droughts (and provided flood protection during periods of increased precipitation). We construct an integrated dataset on major water supply and water infrastructure in the states of North Dakota, South Dakota, Montana, Wyoming, and Idaho. Using the Army Corps of Engineers’ definition of “major dams,” we include all dams that exceed twenty- five feet in height with more than fifteen acre feet in total storage; any dams that exceed six feet in height with more than fifty acre feet in storage; or any dams that maintain a “significant” hazard- mitigation classification. This county- level water infrastructure data set is then spatially linked to topographic characteristics, historical climate data, and historical agricultural data. We obtained data on topographic characteristics from the U.S. Geological Services, agricultural data from the U.S. Census of Agriculture, and climate data from the U.S. Climate Division. The U.S. Climate Division Dataset (USCDD) provides various historical temperature and precipitation measures, including the Palmer Drought Severity Index (PDSI) and other standardized z- scores of temperature and precipitation. The PDSI is a long- term drought measure that is standardized to the local climate so that it shows relative drought and rainfall conditions in a region at a specific time. We use these data and measures to analyze the impact of the water infrastructure on agricultural production, especially during times of drought or excessive precipitation. We find that counties that had major water storage and distribution facilities were generally better able to deal with climatic variability. Farmers with access to more consistent water supplies were more likely to smooth their agricultural production (crop mix and yields) and had a higher likelihood of successful harvest, especially during periods of severe droughts or excessive precipitation. Counties with water supply control were better able to mitigate losses in agriculture relative to similar counties without such infrastructure. Thus, our results indicate that the presence of major water infrastructure has helped to mitigate the damages of periodic droughts and excessive precipitation and will likely continue to do so in the future. 9.2

Origins and Impact of Western Water Infrastructure

During the late nineteenth century, as agricultural settlement of North America moved into the Great Plains and beyond, irrigation expansion and flood control became crucial issues. The agricultural techniques and practices settlers brought with them from the humid East were not applicable in the arid or semiarid West. Institutions such as the 1862 Homestead Act that created small, 160 to 320 acre farms were not appropriate in the region (Libecap and Hansen 2002). As early as the 1870s, John Wesley Powell was promoting the organization of autonomous irrigation districts to increase cooperation among farmers and to cope with the externalities associated with each individual farmer’s decision on water storage and distribution.

256

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

Interest in federal reclamation programs to construct dams and canals to store and distribute water developed after individual, corporate, and state attempts to deliver such infrastructure was found to be inadequate. Many state attempts, such as the 1887 Wright Act of California, faced problems ranging from poor construction to the creation of fraudulent irrigation districts and the huge debts that ensued (Robinson 1979). Most private irrigation projects failed, and those that succeeded were of small size due to problems of free- riding. After much debate and failed attempts to develop water infrastructure in the West, the Federal Reclamation Act was passed in June 1902, and a revolving Reclamation Fund was created to finance water infrastructure projects. The Reclamation Fund was financed through the sales of public lands and through cost- sharing agreements by recipients (Pisani 2002). The title for the water infrastructure remained with the federal government, state, and territorial agencies; local water supply organizations, such as irrigation districts, governed the use and the distribution of water (Robinson 1979). This structure mostly remains in place today. We analyze the impact of this federal and related state and private water infrastructure on agricultural production and flood control over time using historical agricultural and climatic data. The specific research questions we seek to answer are the following:

• Were counties that had major irrigation water supply and distribution infrastructure better able to cope with the problems of short- term climatic variability (either due to natural variability in the hydrologic cycle or due to disruptions of the cycle), relative to those similar counties without such infrastructure? • Did cropping patterns (measured in area of irrigated and harvested land, relative to total agricultural land) change after the construction of irrigation water supply and distribution infrastructure? • Did agricultural productivity (measured in crop- specific tons per acre or bushels per acre) change after the construction of irrigation water supply and distribution infrastructure? • Was the impact of major water infrastructure on agricultural productivity and cropping patterns different (especially important) during periods of climate shocks of severe droughts and excessive precipitation? To address these questions, we examine the variation in agricultural production before and after dam and canal construction at the county level as well as across counties with and without such infrastructure. Counties with access to water infrastructure are likely to experience fewer agricultural production failures after unfavorable climatic conditions, all else equal. We also examine the variation in agricultural production in periods of normal precipitation and in times of droughts and increased precipitation over time.

Climate Variability and Water Infrastructure

9.3

257

Data and Empirical Model

In order to analyze the impact of dams on agricultural activities, we have compiled the most detailed data available on major water infrastructure, historical agricultural productivity, and other relevant data, including the climate and geographic characteristics of our study area. Our empirical strategy focuses on five north- central states: Idaho, Wyoming, Montana, North Dakota, and South Dakota (see figure 9.1). Although there is variation within them, these five states are quite similar in annual temperature trends, precipitation, crops grown, and soil types. However, within these states, the availability of irrigation water varies widely. Our sample includes all counties in these states that are located west of the 100th meridian, generally considered the boundary between the arid West and wetter Great Plains. The counties in our sample, all of which are considered “arid” or “semiarid,” require additional irrigation in order to grow crops in a water- intensive fashion. We have 181 counties and a total of 3,620 observations, measured across twenty years (1900 to 2002) from the agricultural census. The following sections describe our data sources; discuss the mechanisms that were used to assign nonuniform counts and averages to the counties; and present summary statistics on trends in major water infrastructure, agriculture, and climate characteristics. 9.3.1

Major Water Infrastructure

Our primary source for the major water infrastructure is the U.S. Army Corps of Engineers’ National Inventory of Dams (NID). The data include information on the location, owner, year of completion, the primary purposes, capacity, and height characteristics for major dams in the United States.1 The primary purposes of construction include flood control, debris control, fish and wildlife protection, hydroelectric generation, irrigation, navigation, fire protection, recreation, water supply enhancement, and tailings control. Figure 9.1 shows the major dams based on the NID as well as topography types and the average annual precipitation levels. As shown in table 9.1, approximately 22 percent of the water captured by dams in our five- state sample has irrigation listed as the primary purpose, 1. There are approximately 80,000 dams in the Army Corps of Engineers database, 8,121 of which are considered to be “major.” Based on our examination of dam characteristics in the state of Idaho, we do not think that omitted smaller dams were comparable to major dams in providing storage capacity for irrigation or protecting from floods. Specifically, 111 major dams in Idaho have a mean maximum storage of over 155,000 acre feet and a mean height of 115 feet. In contrast, 478 smaller dams have a mean maximum storage of only 687 acre feet and a mean height of 20 feet. Because 1.75 acre feet of water is expected to produce at most 100 bushels of wheat per irrigated acre, a 687 acre feet dam can provide enough water to hydrate only about 392 acres of farmland, producing approximately 40,000 bushels of wheat (Brouwer and Heibloem 1985).

258

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

Fig. 9.1 Major dams, topography types, and average annual precipitation (in.) categories

and 83 percent of the water has irrigation listed as a purpose. Dam construction in the West peaked in the post-WWII period—with the vast majority of dams constructed during the 1950s, 1960s, and 1970s. Over 55 percent of the total dam capacity in the western United States was added during the 1950s and 1960s alone, and over 73 percent was added over that same time frame in our five- state sample (see table 9.2). Because we are concerned with the impact that dams had on agricultural productivity, correctly connecting the available water supply with agricultural demand is essential. In order to do so, all of the major dams in and around our five- state sample have been spatially merged with county- level agricultural census data. Therefore, if a county has a major dam located within its boundaries or nearby, it has been assigned access to the supply of dam water (measured in both volume of water and the number of irrigation dams that the county has access) for all subsequent years after dam completion. Unfortunately, assigning water allocation based solely on the location of the dam itself ignores those counties that are connected to the dam via canals, aqueducts, or major river systems. Many major water infrastructure projects are shared by multiple counties and multiple states. In situations where river systems and aqueducts are many hundreds or thousands of miles in length and have multiple dams on the same river, the water supply

Climate Variability and Water Infrastructure Table 9.1

259

Primary purpose of water provided by dams (total maximum storage measured in acre feet)

Flood control Hydroelectric Irrigation Recreation Water supply Other Total No. of wells in 2009 “Irrigation” listed as a purpose

Idaho

Montana

North Dakota

South Dakota

Wyoming

3,947,643 5,870,818 7,146,890 5,500 235,583 384,290

19,130,823 10,913,527 10,569,892 15,430 107,467 97,176

25,208,840 0 500,860 51,762 0 21,256

23,618,835 0 1,692,602 31,443 439,758 100,970

4,589,966 758,390 7,486,167 0 433,015 361,721

17,590,724

40,834,315

25,782,719

25,883,608

13,629,259

31,400

4,220

860

2,680

1,740

8,483,473

31,186,090

25,000,860

25,736,800

12,691,233

Note: “Other” includes debris control, fish and wildlife, fire protection, tailings, and others.

is assigned to those counties that are downstream from the dam before any subsequent dam. In order to accomplish this adjustment and to create a robustness check for our empirical models, we have contacted and surveyed the operators or management agencies for each of the 475 major dams in our five- state sample. With this survey, we are able to determine the actual allocation of water from each dam to the different counties in our study.2 This survey does not address any temporal changes in the supply of water from major dams, but it does provide a more accurate representation of the spatial distribution of water from major dams.3 9.3.2

Census of Agriculture Data

Historical agriculture data from the U.S. Census of Agriculture have been collected and inputted manually from hard- copy historical agricultural census records. The data include twenty yearly observations for each county in the sample, assembled between 1900 and 2002. The first few years of the agricultural census were decennial and aligned with the population census, but after 1920, the agricultural census was conducted every five years, with 2. A questionnaire was administered via telephone to all major dam operators in Idaho, Montana, Wyoming, North Dakota, and South Dakota. The information collected through these questionnaires includes whether water is withdrawn onsite, downstream, or both; the counties (and the acreages) that dams supply irrigation water to; and the primary method of withdrawing water (pump or canals). 3. We were able to collect data on the total number of wells providing water to agriculture in our sample states (see table 9.1), but the data is measured in 2009 and, therefore, doesn’t reflect the changing availability of well water over time, nor does it represent the volume of well water available. We acknowledge that the availability of well water and the electrification of the farm increased the agricultural potential of many western counties.

1,604,365 2,173,450 131,573 303,378 1,169,090 5,874,917 64,459 5,500,870 134,287 255,707 378,628

17,590,724

⬍1910 1910–1919 1920–1929 1930–1939 1940–1949 1950–1959 1960–1969 1970–1979 1980–1989 1990–2000 Unknown

Total

111

13 14 6 6 10 10 8 15 11 8 10

No.

40,834,315

324,453 1,523,704 691,403 2,953,150 415,850 26,625,161 2,133,834 6,129,722 8,929 24,450 3,659

Storage

Montana

162

7 24 17 36 15 28 15 14 3 1 2

No.

25,782,719

0 2,068 0 529,358 500,860 24,505,318 53,163 189,464 2,206 0 282

Storage

34

0 1 0 11 3 2 6 8 2 0 1

No.

North Dakota

Note: Storage reflects the maximum storage, in acre feet, behind the dams in the given state.

Storage

Idaho

Storage and count of dams

Year completed

Table 9.2

25,883,608

0 186,837 0 25,353 116,794 1,835,649 23,617,135 5,690 5,250 20,900 70,000

Storage

34

0 2 0 3 4 7 4 10 2 1 1

No.

South Dakota

13,629,259

2,895,107 2,760,933 257,250 1,290,735 94,494 4,742,634 749,376 387,056 336,585 115,089 0

Storage

Wyoming

134

12 5 11 12 3 21 20 19 20 11 0

No.

Climate Variability and Water Infrastructure

261

a reorganization in 1978 so that the data were collected at two points before the next decennial population census.4 From the agricultural census data, several different measures of agricultural composition and production have been constructed, including total planted acreage, total failed acreage, total fallow or idle acreage, and total harvested acreage by crop (measured in tons or bushels per acre). We have collected major crop variables that are found in all five of our sample states, including wheat, barley, and hay, as well as major crop variables that are more state or region specific in the west, such as potatoes and corn. Figures 9.2 and 9.3 present the historical trends in agricultural composition and productivity, respectively, in our five sample states. The average trends in agricultural composition presented in figure 9.2 indicate that large decreases in hay acreage were experienced throughout the first twenty- five years of the twentieth century, followed by a general decline in average acreage after 1925. Conversations with research hydrologists justify this decline as a function of hay being a prime input to production for all agriculture prior to the mass availability of internal combustion power, but just an input to livestock agriculture after the arrival of heavy planting and harvesting machinery.5 Wheat acreage increased until the 1950s and then experienced relatively stagnant growth. What becomes very apparent in figure 9.2 is the impact of the mid1930s Dust Bowl era droughts—the spikes in failed and idle acreage and the associated declines in wheat, barley and all of the other major crops. Figure 9.3 presents the average trends in agricultural productivity, as measured by output per acre, per crop. All crops experienced increased productivity throughout the twentieth century, with relatively larger gains made by corn, barley, and wheat. Hay productivity was relatively stagnant, however. Two issues arise when integrating agricultural data into the analysis: the changing shape of counties over time and the changing measures or definitions within the agricultural data set. First, because the agricultural census is provided at the county level, and county boundaries have changed over time, we have normalized the agricultural census data to current 2000 census county boundaries. In order to do so, we multiply the historical census count data (measured in acres or volume of output) by the fraction of the county that lies in the current 2000 census county boundary definition (measured as the percentage of the total land area). In almost all cases, the historical county boundaries were subdivided into current boundaries, with very few modifications.6 4. Our yearly observations include 1900, 1910, 1920, 1925, 1930, 1935, 1940, 1945, 1950, 1955, 1960, 1965, 1970, 1975, 1978, 1982, 1987, 1992, 1997, and 2002. 5. Information is from personal communication between Scott Lowe and Bryce Contor, research hydrologist with the University of Idaho Water Resources Research Institute, November 2009. 6. For example, in Idaho, Alturas County, which existed from 1864 to 1895, was divided over the years into eight separate counties. The transition to eight counties was not instantaneous—

262

Fig. 9.2

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

Historical agricultural composition (% of total acreage)

Second, the crop acreage and harvest data within the census has changed over time. For example, in the early years of the agricultural census, certain forage crops were listed as a single entry, but in later years, the forage crops were split into multiple categories. We are limited by the lowest common denominator in these cases. In the models in which we are interested in individual crops that have been further divided into subcrops, we aggregate so that the unit of measure is consistent across all of the years in our sample. 9.3.3

Climate Data

The U.S. Climate Division Dataset (USCDD) provides averaged climate data based on 344 climatic zones, covering 1895 to present.7 The USCDD includes various temperature and precipitation measures, such as the monthly maximum, minimum, and mean temperatures; total monthly precipitation therefore, the number of counties is often different across different agricultural census periods. By limiting our period of analysis to the 1900 to 2002 agricultural censuses, we avoid much of the redistricting of county areas that took place in the western United States before the twentieth century. Less than 2 percent of the total observations in our sample were affected by county- level redistricting. 7. The five- state sample includes forty- five distinct climate zones in Idaho (ten), Montana (seven), North Dakota (nine), South Dakota (nine), and Wyoming (ten).

Climate Variability and Water Infrastructure

Fig. 9.3

263

Historical agricultural productivity (average per acre)

levels; and the PDSI. The PDSI uses temperature and rainfall information in a formula to determine relative wetness or dryness and is most effective in determining long- term drought—a matter of several months. It uses a 0 as normal, with drought periods shown as negative numbers and flood periods shown as positive numbers.8 The USCDD is zonal in nature, dividing each state into similar climate zones. Unlike monitor- level temperature and precipitation data, the zonal climate data utilize all of the monitor readings within a zone to arrive at zone- averaged temperature and precipitation measures. Unfortunately, the zones reflect topographic and meteorological uniformities and, therefore, don’t conform to sociopolitical county boundaries. For this reason, in order to assign zonal climate data to counties that overlap multiple zones, the zonal climate data must be averaged across the overlapping counties.9 8. The climatic data are from the Area Resource File (ARF). The ARF file is maintained by Quality Resource Systems (QRS) under contract to the Office of Research and Planning, Bureau of Health Professions, within the Health Resources and Services Administration. 9. We use an evenly weighted average—so if a county falls in two zones, the temperature and precipitation values that are assigned that that county would be 50 percent of the first

264

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

9.3.4

Topography and Soil Quality Data

We utilize a general topography classification from the National Atlas (USGS), which identifies counties according to their major landform types. The variable we use is the “Land Surface Form Topography code,” which identifies twenty- one different classifications, including plains, tablelands, open hills, and mountains. This information on topography types allows us to group counties according to similar landforms—and the likelihood that they have land that is suitable for growing certain crops. Similarly, the topography type reflects the likelihood that a county has the potential of having a major dam in place: the major western dams are much more likely to occur in the more mountainous counties, or those with open hills, than they are in tablelands or plains. In addition to landform type, we have collected county- level soil quality (soil type) information from the Natural Resources Conservation Service’s (NRCS) Soil Survey Geographic (SSURGO) database. The SSURGO database provides a national coverage of soil quality measures, according to the nonirrigated Land Capability Classification System (LCC). The LCC includes eight classes of land, derived from a series of soil qualities, including percent slope, erosion potential, rooting depth, drainage class, composition, salinity, growing season, average annual precipitation, and surface textures. The coverage of each LCC in each county is weighted by total area to arrive at a county- averaged LCC that is time invariant. 9.3.5

Econometric Model

In this section, we describe the econometric strategy used to estimate the impact of major water infrastructure on cropping practices and on crop outcomes (such as harvest per acre and crop failure), using the data sources outlined in the previous sections. It is important to note that any empirical exercise that models the impact of water provided by dams on agricultural composition or output may suffer from spurious positive correlation: counties that have the natural amenities present and, thus, anticipate a larger productivity gain from an increased volume of irrigation water may be more likely to invest in a major water infrastructure project. Thus, traditional ordinary least squares (OLS) analyses may be biased. In addition, water supply infrastructure could have brought agricultural production to areas where it was not possible without irrigation and, therefore, introduce a potential endogeneity problem. To overcome these issues, we present two different but related identification strategies. First, we cluster similar counties based on topographic and climatic zone and 50 percent of the second zone. As an example, of the forty- four counties in Idaho, twenty- eight are contained in a single climate zone, nine overlap with two climate zones, five overlap with three climate zones, one overlaps with four climate zones, and one overlaps with five climate zones. As a robustness check, we will run models with only those counties that are contained in a single climate zone.

Climate Variability and Water Infrastructure

265

types (see figure 9.1). Identification, therefore, comes from the withincluster differences in availability of irrigation water over time. In this case, counties within clusters are very similar in agricultural potential (soil type, annual precipitation total, annual temperature average) as well as the potential for having a major water infrastructure project (topography classification), and we estimate the treatment effect of having irrigation water within clusters. To create these clusters, we first exclude all counties in our sample where profitable agricultural production without irrigation would not be possible. The definition of “arability” used includes a minimum average precipitation of at least ten inches per year, sufficiently deep soil with no clay and sand, and no excessive evaporation due to wind and heat during critical stages of plant growth. Assuming that profitable production of agricultural crops is possible, we also cluster counties based on three precipitation classes: low (ten to twelve inches), medium- low (twelve to sixteen inches), medium- high (sixteen to twenty inches), and high (⬎ twenty inches). We further differentiate clusters by the major topography type within which each county falls. These major topography classifications are taken directly from the Land Surface Form Topography Codes from the USGS and include plains, tablelands, plains and hills, open hills, and mountains. The inclusion of a topographic classification into our clusters allows us to control for those counties that have a similar potential for major water infrastructure. We let H itg denote the agricultural outcome in county i in cluster g in year t. Dit is an indicator variable that equals 1 when county i has access to irrigation water provided via a major water infrastructure project in year t, and zero otherwise.10 Our basic econometric model is equation (1): (1)

Hitg ⫽ ␣Dit ⫹ ␤Iit ⫹ Xitϕ ⫹ Zi␮ ⫹ ␪t ⫹ ␦j ⫹ ς g ⫹ ␩it,

where ␣ is the parameter of interest and measures the difference in the dependent variable between counties with and without major water infrastructure projects.11 Iit represents the occurrence of a particularly wet or dry period, depending on the model estimated, using the PDSI. Xit is a vector of controls that vary over time at the county level and includes the annual rainfall and temperature measures.12 Zi is a vector of controls that 10. We begin with a 50,000 acre feet (AF) cutoff to construct the indicator variable for irrigation water provided by dams but relax this cutoff as a robustness check. As expected, as the cutoff level decreases, so does the magnitude of the coefficient on the dam indicator variable. 11. We constrain the major water infrastructure binary variable so that it represents only projects with a dedicated purpose of irrigation. In these cases, dams that were constructed for hydroelectric, recreation, or other purposes without any indication that they will provide irrigation water are not included as dams in our analysis of agricultural impacts. 12. We also include nonlinearities in the precipitation and temperature variables in all of our models. As a robustness check, we alter the months that are used to calculate precipitation and temperature totals, including adding the winter (wetter) months in the year prior to the agricultural census or limiting the climate averages to the summer months during the growing season (to account for dry- cropping or those operations that rely solely on rain and snowfall for crop irrigation). These modifications do not have noticeable effects on our variables of concern.

266

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

are continuous across time but vary at the county level. This vector includes the nonirrigated soil capability classification of the county. ␪t is a year fixed effect, ␦j is a state fixed effect, ς g is the aforementioned panel fixed effect for counties with similar topography and climate, and ␩it is the unobserved error component. We allow Hitg to represent (a) percentage measures of agricultural mix (changes in composition) or (b) the successful harvest per acre before and after a major water infrastructure project is completed.13 To further test whether major water infrastructure provides security during times of drought or abnormally increased precipitation, we utilize a difference- indifference model, presented in equation (2), where we interact Dit with the indicator variables for those periods in which agriculture experienced either a particularly dry or wet period (Iit):14 (2)

Hitg ⫽ ␣Dit ⫹ ␤Iit ⫹ ␥(Dit ⭈ Iit) ⫹ Xitϕ ⫹ Zi␮ ⫹ ␪t ⫹ ␦j ⫹ ςg ⫹ ␩it

Our second strategy to account for the spurious positive correlation follows Duflo and Pande (2007) and takes advantage of the topography classifications of each county as instruments for the likelihood of dam construction. Specifically, we estimate a fixed effects two- stage least squares panel regression of the form (3)

Dit ⫽ ␶(Ti ⭈ ␪t) ⫹ ␤Iit ⫹ Xitϕ ⫹ Zi␮ ⫹ (␪t ⭈ ␦j) ⫹ ␩it.

The notation follows from equation (1), with the following exceptions: first, the topographic classification (Ti ) has been interacted with the year fixed effect (␪t); second, the panel fixed effect is now represented by an interaction between the year fixed effect (␪t) and the state fixed effect (␦j) to arrive at a state- year fixed effect. To generate instruments for Dit, we utilize the parameter estimates from equation (3) to predict the likelihood of a dam being constructed (Dˆit) in our first- stage equation.15 Our final instrumental variables estimation is, therefore, equation (4): (4)

Hit ⫽ ␣Dit ⫹ ␤Iit ⫹ Xitϕ ⫹ Zi␮ ⫹ (␪t ⭈ ␦j) ⫹ ␩it,

using instruments for Dit from the parameter estimates from equation (3).

13. It is worth noting that in most cases, the dependent variable (be it harvest per acre or the composition [percent] of different crops produced) is not biased by the size of the county. In models in which the total land area of the county may affect results, the treatment variable will need to be normalized by the total land area or the size of the county. 14. We use two cutoffs for to construct the drought and flood dummy variables: with a drought condition represented by PDSI ⱕ 2 and a flood condition represented by PDSI ⱖ 2. These cutoffs are considered “moderate to severe droughts/floods” within the PDSI literature. 15. For each of our instrumental variable models, we test the null hypothesis that the excluded instruments are irrelevant in the first- stage regression. The F- statistic allows us to reject the null hypothesis for all models.

Climate Variability and Water Infrastructure

9.4

267

Results

The objective of this study is to examine the impact of major water infrastructure on agricultural composition and productivity. We hypothesize that counties with access to water supply infrastructure are better able to deal with the problems of short- term climatic variability (either due to natural variability in the hydrologic cycle or due to disruptions of the cycle) in terms of smoothing out agricultural production over time relative to those similar counties without such infrastructure. It is important to note that the temperature variability and average levels of precipitation are very similar across the five states in our sample, and unlike more temperate western states, the year- to- year water availability is not a significant factor in determining annual crop mix changes. Because of climate extremes (traditionally represented by very hot summers and very cold winters), the presence of irrigation water is less essential in our sample states than in states that grow crops that are much more dependent on irrigation water. Many of the crops that we analyze are field crops that can be dry- farmed, unlike row crops that require intensive irrigation to produce any output. For these reasons, we anticipate that the impacts of dam- provided irrigation water are likely to be less significant in states in our sample than in those states that grow crops that are much more dependent on it for their agricultural practices. Thus, we expect that our results likely underestimate the true benefit of the major water infrastructure projects, particularly in those states that have more crop variability and that are more dependent on dam- provided irrigation water. Tables 9.3 through 9.9 show results from our clustered sample presented in equations (1) and (2) from section 9.3. Our sample counties are clustered based on similarities in topographic and climatic characteristics. Thus, identification in our models comes from the within- cluster differences in availability of irrigation water over time. Table 9.3 presents the impact of dams on agricultural composition— shares of total cropland dedicated to certain crops as well as fallow and failed crop acres—controlling for precipitation, temperature, soil quality, and state and year fixed effects, measured within clusters of counties. In our sample counties, annual water variability (of dam- provided irrigation water) is not expected to be a significant determinant of crop mix changes. As anticipated, in these models we see that the indicator variable for dams has only a small, insignificant impact on agricultural composition. The indicator variable for droughts, however, which is constructed using PDSI cutoff of “moderate to severe droughts,” shows a shift away from wheat and corn during drought periods and an increase in failed acreage. When we interact the indicator variables for dams and droughts in table 9.4, we see that the counties that have dams experienced a net decrease in wheat acreage (with a net effect of –3.07 percent) but still had 2.46 percent more wheat acreage than similar counties that experienced a drought

3,498 16 0.18 Yes Yes

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Drought dummy

Soil type

Temperature squared

Temperature

0.0038 (0.0305) 0.1494 (0.1024) –0.0406 (0.0291) –0.0417 (0.0503) 0.0004 (0.0006) 0.0412∗ (0.0221) 0.0281 (0.0199)

(1) % hay

3,482 16 0.20 Yes Yes

–0.0018 (0.0197) –0.1671∗∗ (0.0612) 0.0316 (0.0220) 0.0258 (0.0299) –0.0003 (0.0003) –0.0317∗∗∗ (0.0067) –0.0500∗∗∗ (0.0085)

(2) % wheat

3,471 16 0.27 Yes Yes

0.0047 (0.0066) 0.0345 (0.0226) –0.0159∗∗ (0.0064) –0.0045 (0.0201) 0.0000 (0.0002) –0.0014 (0.0031) 0.0032 (0.0054)

(3) % barley

3,359 16 0.25 Yes Yes

0.0009 (0.0022) 0.0005 (0.0051) –0.0031∗∗∗ (0.0010) 0.0161∗∗ (0.0066) –0.0002∗∗ (0.0001) –0.0023 (0.0020) 0.0025∗ (0.0012)

–0.0113∗ (0.0055) –0.0081 (0.0290) 0.0004 (0.0090) –0.0055 (0.0095) 0.0001 (0.0001) –0.0029 (0.0042) –0.0070∗ (0.0036) 3,372 16 0.16 Yes Yes

(5) % potatoes

(4) % corn

Model and dependent variable

Fixed effects models of agricultural composition with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.3

2,415 16 0.52 Yes Yes

–0.0035 (0.0050) 0.1673∗∗ (0.0705) –0.0394∗ (0.0217) 0.0064 (0.0189) –0.0000 (0.0002) –0.0010 (0.0022) 0.0215∗∗∗ (0.0066)

(6) % failed

2,407 16 0.29 Yes Yes

0.0142 (0.0190) 0.0676 (0.0722) –0.0150 (0.0201) 0.0351 (0.0233) –0.0004 (0.0003) –0.0194 (0.0145) –0.0368∗∗∗ (0.0120)

(7) % fallow

Climate Variability and Water Infrastructure

269

without a dam present. Counties that have dams experienced a decrease in failed acreage (a net effect of –1.31 percent) during drought periods, relative to those drought areas without dams present.16 According to these results, dams had very little impact on the agricultural practices during normal precipitation years but had a much more significant impact during drought years. Relative to similar counties that did not have dam- provided irrigation water available, counties with the dams were much more successful during difficult growing years—they were able to plant more of their total acreage, and they lost less of the crop that was planted. The results are economically significant. The average aggregate annual wheat acreage across the five states in our sample was 10.2M acres, and the average aggregate annual failed acreage was 1.8M acres. The coefficient on our drought- dam interaction term from model (2) in table 9.4, a 2.46 percent difference in wheat acreage due to having access to dams during droughts, is equivalent to an average increase of over 251,000 acres dedicated toward wheat, in aggregate across all states, averaged across all years. Similarly, the coefficient on our drought- dam interaction term from model (6) in table 9.4, a 4.41 percent decrease in failed acreage, is equivalent to an average decrease of approximately 80,000 failed acres, in aggregate across all states, averaged across all years. The net effect of having access to dam water during droughts is the 1.31 percent decrease in failed acreage and is equivalent to an average decrease of approximately 24,000 failed acres. Tables 9.5 and 9.6 present similar models to tables 9.3 and 9.4 but measure agricultural productivity as opposed to agricultural composition. Agricultural productivity is measured in bushels per acre for wheat, barley, and corn; in tonnage per acre for hay; and in 100 pounds per acre for potatoes. In these models, we see an across- the- board increase in productivity during drought periods, even when total cropland harvested may be declining. For example, we observe a 0.12 ton per acre increase in hay productivity and a 2.96 bushel per acre increase in wheat productivity. Conversations with water resource managers indicate that these results are not all that surprising—when farmers are faced with decreased water availability, they are very careful with their overall water management, thus applying the water that they do have to their most productive acreage. This application practice results in an increased productivity per acre across the fewer acres that are planted.17 Similar to tables 9.3 and 9.4, the dam- indicator variables have a positive but insignificant effect on agricultural productivity. This is again due to the 16. We estimated our models using all dams as well as irrigation dams only. Results are qualitatively similar although more significant when we use irrigation dams. We also estimated models showing the impact of the dam not only where it is located, but from various distances from the dam and again find similar results. 17. Information is from personal communication between Scott Lowe and Hal Anderson, Administrator of Planning and Technical Services Division, Idaho Department of Water Resources, November 2009.

3,498 16 0.18 Yes Yes

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Drought-dam interaction

Drought dummy

Soil type

Temperature squared

Temperature

0.0004 (0.0325) 0.1523 (0.1028) –0.0416 (0.0293) –0.0421 (0.0500) 0.0004 (0.0006) 0.0412∗ (0.0220) 0.0247 (0.0209) 0.0159 (0.0248)

(1) % hay

3,482 16 0.20 Yes Yes

–0.0071 (0.0208) –0.1627∗∗ (0.0621) 0.0300 (0.0223) 0.0252 (0.0300) –0.0003 (0.0003) –0.0318∗∗∗ (0.0068) –0.0553∗∗∗ (0.0102) 0.0246∗ (0.0120)

(2) % wheat

3,471 16 0.27 Yes Yes

0.0059 (0.0055) 0.0335 (0.0235) –0.0156∗∗ (0.0067) –0.0043 (0.0199) 0.0000 (0.0002) –0.0014 (0.0031) 0.0044 (0.0048) –0.0056 (0.0111)

(3) % barley

3,359 16 0.25 Yes Yes

0.0006 (0.0018) 0.0007 (0.0049) –0.0032∗∗∗ (0.0010) 0.0160∗∗ (0.0066) –0.0002∗∗ (0.0001) –0.0023 (0.0020) 0.0022 (0.0016) 0.0015 (0.0022)

–0.0136∗∗ (0.0062) –0.0060 (0.0290) –0.0003 (0.0091) –0.0057 (0.0094) 0.0001 (0.0001) –0.0029 (0.0042) –0.0093∗ (0.0050) 0.0110 (0.0064) 3,372 16 0.16 Yes Yes

(5) % potatoes

(4) % corn

Model and dependent variable

Difference-in-difference fixed effects models of agricultural composition with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.4

2,415 16 0.53 Yes Yes

0.0055 (0.0044) 0.1584∗∗ (0.0724) –0.0361 (0.0221) 0.0076 (0.0180) –0.0000 (0.0002) –0.0010 (0.0023) 0.0310∗∗∗ (0.0057) –0.0441∗∗∗ (0.0079)

(6) % failed

2,407 16 0.29 Yes Yes

0.0157 (0.0191) 0.0663 (0.0735) –0.0145 (0.0204) 0.0354 (0.0237) –0.0005 (0.0003) –0.0194 (0.0145) –0.0352∗∗∗ (0.0112) –0.0074 (0.0174)

(7) % fallow

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Drought dummy

Soil type

Temperature squared

Temperature

3,506 16 0.53 Yes Yes

0.0123 (0.0480) 0.3182 (0.2458) –0.1098 (0.0732) 0.0603 (0.2465) –0.0004 (0.0029) 0.0338 (0.0480) 0.1209∗∗ (0.0505)

(1) Prod hay

3,420 16 0.60 Yes Yes

0.3328 (0.8814) –6.1633 (4.2880) 0.9087 (1.0289) –2.3920 (3.1610) 0.0299 (0.0374) 1.0415 (0.6582) 2.9577∗∗∗ (0.8035)

(2) Prod wheat

3,413 16 0.66 Yes Yes

0.5809 (0.8567) 4.4790 (3.9326) –2.1465∗∗ (0.9230) –0.4798 (3.8940) 0.0059 (0.0439) 1.3763∗ (0.7562) 2.8479∗∗∗ (0.6823)

(3) Prod barley

2,328 16 0.71 Yes Yes

2.9639 (1.9674) 15.0053 (11.0013) –8.0377∗∗ (3.2000) 0.5813 (5.6545) 0.0018 (0.0634) 0.5997 (0.8414) 7.6389∗∗∗ (1.1996)

(4) Prod corn

Model and dependent variable (output/acre)

Fixed effects models of agricultural productivity with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.5

2,634 16 0.20 Yes Yes

7.2172 (6.1585) 93.7036∗∗ (32.5217) –29.3468∗∗∗ (7.5700) –4.0923 (21.9671) 0.0632 (0.2528) 2.7186 (3.4159) 27.5879∗∗ (11.3440)

(5) Prod potatoes

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Drought-dam interaction

Drought dummy

Soil type

Temperature squared

Temperature

3,506 16 0.53 Yes Yes

–0.0067 (0.0415) 0.3350 (0.2388) –0.1158 (0.0711) 0.0583 (0.2467) –0.0004 (0.0029) 0.0338 (0.0481) 0.1013∗ (0.0533) 0.0903 (0.0528)

(1) Prod hay

3,420 16 0.60 Yes Yes

–0.4286 (0.9418) –5.7519 (4.2002) 0.7523 (1.0107) –2.5230 (3.1334) 0.0314 (0.0370) 1.0307 (0.6582) 2.1855∗∗∗ (0.7126) 3.6065∗∗∗ (1.1923)

(2) Prod wheat

3,413 16 0.66 Yes Yes

0.0188 (0.7529) 4.8524 (3.9769) –2.2842∗∗ (0.9418) –0.5571 (3.9098) 0.0068 (0.0440) 1.3675∗ (0.7552) 2.2760∗∗∗ (0.7209) 2.6113∗ (1.2537)

(3) Prod barley

2,328 16 0.71 Yes Yes

2.8026 (1.9337) 15.1235 (10.8108) –8.0737∗∗ (3.1360) 0.5670 (5.6679) 0.0019 (0.0635) 0.5965 (0.8421) 7.4923∗∗∗ (1.3703) 0.7814 (2.3636)

(4) Prod corn

Model and dependent variable (output/acre)

Difference-in-difference fixed effects models of agricultural productivity with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.6

2,634 16 0.21 Yes Yes

3.0916 (7.8396) 96.1183∗∗∗ (32.4707) –30.1155∗∗∗ (7.7400) –5.0299 (22.2777) 0.0737 (0.2555) 2.6699 (3.4237) 24.1668∗ (12.1120) 17.8331 (17.7115)

(5) Prod potatoes

Climate Variability and Water Infrastructure

273

fact that the presence of irrigation water is likely less essential in our sample states that mostly grow field crops that can be dry- farmed than in states that grow crops that are much more dependent on irrigation water. When we interact the indicator variables for dams and droughts in table 9.6, we see that the presence of dams in areas that are suffering from droughts has a positive, significant impact on wheat productivity (a net effect of ⫹5.80 bushels per acre) and barley productivity (a net effect of ⫹4.89 bushels per acre). These effects also are economically significant. They are roughly 24 percent and 15 percent, respectively, of the annual mean productivity per acre in our sample. In terms of the total harvest, an increase of 5.80 bushels per acre of wheat translates into an average output increase of approximately 334,000 bushels per county, per year.18 At the current commodity price for wheat, this is equal to almost a $1.8M increase in annual revenue, per county (U.S. Department of Agriculture, National Agricultural Statistics Service 2009). To put these results in context, we estimate that the average acres of wheat harvested per county, per year was 57,500 across our study sample. With an average of 23.8 bushels per acre and an average 2009 commodity price of $5 per bushel for wheat, the total revenue per county from wheat is $6.8M per year. Tables 9.7 and 9.8 present similar models to those presented in tables 9.5 and 9.6 but are altered to account for periods of excessive precipitation as opposed to periods of drought. We want to test for the impact of dams on agricultural productivity during periods of potentially damaging flooding. We use a flood condition represented by PDSI ⱖ 2 to create the flood dummy variable indicating “moderate to severe floods/precipitation levels” within the PDSI literature. Increased levels of precipitation, of course, may have other effects, such as a greater opportunity to hold more water behind dams and use additional irrigation to improve the agricultural productivity of those crops that benefit from extra water availability. On the other hand, increased water availability may also increase the use of marginal land for some crops where output per acre is lower. In addition, damaging levels of precipitation and flooding may decrease the overall agricultural productivity levels. We find that periods of excessive precipitation (shown as the flood indicator variable in the tables) result in a significant positive effect on agricultural productivity of wheat and hay, the two dry- farmed crops in our sample. These results are presented in models (1) and (2) in table 9.7. Extra water storage and the availability of more water for irrigation to be applied to lands that are typically cultivated by using dry- cropping techniques is the likely explanation for the positive and significant effect on the total productivity of wheat and hay, ceteris paribus. Precipitation level has a negative impact on wheat productivity. This 18. The mean harvest, measured at the county level between 1900 and 2002, was 57,500 acres of wheat, 41,700 acres of hay, 12,400 acres of barley, 3,300 acres of corn, and 1,370 acres of potatoes. These numbers are larger (with the exception of hay) for the latter half of the twentieth century than the first half, but had more variation across counties in the latter years.

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Flood dummy

Soil type

Temperature squared

Temperature

3,506 16 0.54 Yes Yes

0.0126 (0.0506) 0.0268 (0.2082) –0.0580 (0.0660) 0.0617 (0.2533) –0.0004 (0.0030) 0.0389 (0.0484) 0.1398∗∗∗ (0.0347)

(1) Prod hay

3,420 16 0.60 Yes Yes

0.3072 (0.8782) –13.0083∗∗ (4.6598) 2.1824∗ (1.1178) –2.3434 (3.2984) 0.0298 (0.0390) 1.1521 (0.6787) 2.9364∗∗∗ (0.7315)

(2) Prod wheat

3,413 16 0.66 Yes Yes

0.5481 (0.8637) –0.3792 (4.1653) –1.1781 (0.9004) –0.6577 (3.9774) 0.0081 (0.0449) 1.4650∗ (0.7816) 0.8852 (0.8744)

(3) Prod barley

2,328 16 0.70 Yes Yes

3.2722 (2.1265) 0.7309 (10.7330) –4.8988 (3.0446) 0.2194 (5.5347) 0.0075 (0.0620) 0.7392 (0.8652) 1.9395 (1.5675)

(4) Prod corn

Model and dependent variable (output/acre)

Fixed effects models of agricultural productivity with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.7

2,634 16 0.20 Yes Yes

7.0380 (6.5410) 57.3807∗∗ (26.8012) –21.8682∗∗∗ (6.3280) –5.5828 (21.9204) 0.0816 (0.2531) 3.4508 (3.4181) –1.1331 (8.0507)

(5) Prod potatoes

Note: Robust standard errors in parentheses. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations Geographic-climate clusters R2 Year fixed effects State fixed effects

Flood-dam interaction

Flood dummy

Soil type

Temperature squared

Temperature

3,506 16 0.54 Yes Yes

–0.0021 (0.0493) 0.0182 (0.2100) –0.0550 (0.0667) 0.0639 (0.2528) –0.0004 (0.0030) 0.0387 (0.0485) 0.1267∗∗∗ (0.0361) 0.0535 (0.0316)

(1) Prod hay

3,420 16 0.60 Yes Yes

–0.2770 (0.8708) –13.2862∗∗ (4.7839) 2.2842∗ (1.1547) –2.2708 (3.3171) 0.0290 (0.0392) 1.1438 (0.6806) 2.4195∗∗∗ (0.6928) 2.1364∗∗ (0.9720)

(2) Prod wheat

3,413 16 0.66 Yes Yes

0.1311 (1.0194) –0.6033 (4.3083) –1.0985 (0.9332) –0.5916 (3.9742) 0.0073 (0.0448) 1.4576∗ (0.7816) 0.5073 (0.7736) 1.5742 (1.3151)

(3) Prod barley

2,328 16 0.70 Yes Yes

2.2655 (2.0239) 0.4775 (10.9863) –4.7831 (3.1454) 0.2069 (5.5170) 0.0075 (0.0618) 0.7302 (0.8739) 1.1066 (1.4970) 4.5210∗ (2.1746)

(4) Prod corn

Model and dependent variable (output/acre)

Difference-in-difference fixed effects models of agricultural productivity with geographic-climate clusters

Precipitation squared

Precipitation

Dam dummy

Table 9.8

2,634 16 0.20 Yes Yes

4.1813 (6.5446) 56.1630∗ (26.8280) –21.4047∗∗∗ (6.3653) –5.2821 (22.0397) 0.0778 (0.2547) 3.3650 (3.4402) –4.0288 (8.7669) 12.1472 (8.5258)

(5) Prod potatoes

276

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

result, combined with agricultural composition results presented in tables 9.3 and 9.4 show that the share of total cropland in wheat declines when there is increased precipitation. This finding indicates that some of the most productive lands are shifted away from wheat production during wet seasons to alternative crops, especially more water- intensive potatoes. Although the presence of a dam alone has a limited impact on agricultural productivity in our sample states, when this effect is interacted with the presence of a period of excessive precipitation (and increased likelihood of flooding), we find a positive impact for all of the agricultural productivities studied. Our results, shown in table 9.8, indicate that the increased precipitation- dam interaction effect has a statistically significant and positive impact on the productivity of wheat (a net effect of ⫹4.56 bushels per acre) and corn (a net effect of at least ⫹4.52 bushels per acre). These effects are roughly 19 percent and 13 percent of the annual mean productivity per acre in our sample. The overall increased productivity impact of dams during periods of excessive precipitation may be due to the prevention of damaging flooding when a dam is present, plus the availability of greater irrigation water held behind dams. Table 9.9 presents the results of the empirical models outlined in model (4) of section 9.3. In these models, we use a two- staged fixed effects process, following Duflo and Pande (2007), in which we instrument for the likelihood of a county having a dam, using topographic characteristics of counties (interacted with year effects) in addition to all of the independent variables that we use to explain the variation in crop composition and productivity. This is another econometric strategy we implement to address the likely spurious positive correlation and endogeneity problems we face in our econometric identification: those counties that have the natural characteristics for dam presence and, thus, anticipate a larger productivity gain from dams are also more likely to invest in a major water infrastructure project. The results of instrumental variable approach, presented in table 9.9, are, in general, consistent with our previous results. However, unlike our findings in tables 9.3 through 9.8, the fixed effects instrumental variables results suggest that the presence of a dam has a large, positive impact on all crops and a significant impact on hay productivity (⫹0.94 tons per acre), wheat productivity (⫹7.1 bushels per acre), and barley productivity (⫹22.90 bushels per acre).19 We also continue to find positive impacts of droughts on per acre productivity, but these effects are only marginally significant.20 19. Given our concern with spurious positive correlation, we had expected to find reduced magnitude in the estimated coefficients and possibly improved standard errors when we used a two- staged fixed effects process. We instead find increased coefficient estimates. This may be due to the weakness of our main instrument for dams: topography of counties. For broader work on this topic, we are exploring alternative instruments, such as river gradient, details on reservoir site (wide or narrow valley), county elevation, and length of rivers. 20. As noted earlier, we were able to collect data on the total number of wells providing water to agriculture in our sample states, but the data is measured in 2009 and, therefore, doesn’t re-

3,506 Yes

3,420 Yes

7.0894∗∗ (3.1919) –16.2502∗∗∗ (2.7630) 4.8924∗∗∗ (0.9126) –8.8767∗∗∗ (1.6791) 0.1092∗∗∗ (0.0187) 0.6081 (0.5750)

0.9449∗∗∗ (0.2260) –0.6948∗∗∗ (0.1792) 0.0553 (0.0594) –0.0478 (0.1217) 0.0012 (0.0014) 0.0684∗ (0.0388)

Note: Standard errors in parentheses. Instrument: topography-year interaction variable. ∗∗∗Significant at the 1 percent level. ∗∗Significant at the 5 percent level. ∗Significant at the 10 percent level.

No. of observations State-year fixed effects

Drought dummy

Temperature squared

Temperature

Precipitation squared

Precipitation

(2) Prod wheat

(1) Prod hay

3,413 Yes

22.9035∗∗∗ (4.8980) –11.3826∗∗∗ (3.9622) 3.2137∗∗ (1.3135) 0.4770 (2.7007) 0.0020 (0.0300) 0.7513 (0.8448)

(3) Prod barley

2,328 Yes

7.5803 (8.6688) –2.5440 (6.7786) 0.0489 (2.2179) –4.8688 (3.1910) 0.0617∗ (0.0351) 0.0979 (1.2899)

(4) Prod corn

Model and dependent variable (output/acre)

Fixed effects two-stage least squares models of agricultural productivity

Dam dummy

Table 9.9

2,634 Yes

16.8781 (50.1415) 88.4890∗∗ (38.4334) –37.1447∗∗∗ (12.9879) –34.8418 (21.6511) 0.4318∗ (0.2429) 25.9228∗∗∗ (7.6828)

(5) Prod potatoes

278

9.5

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

Summary and Conclusions

Climate change is a major issue, and much of the scientific focus is on its likely impact and the costs of various paths of mitigation. Less attention has been directed toward adaption—of how policies and markets have addressed past climate shocks. This is vital information because adaptation to climate change can influence both its effects and the costs of addressing it. The empirical evidence on past adaptation policies is limited. This chapter examines the impact of one of the most extensive water infrastructure investments ever made—the “reclamation” systems of the U.S. West—to determine their impact on agricultural production in the face of severe drought or flooding. Water infrastructure is particularly important as a long- term adaptation strategy to climate change because dams and associated canals are among the longest- lasting infrastructures in agriculture, and investment in them differs from most other agricultural adaptation strategies available to farmers (Reilly 1999). All told, our results from the various estimation strategies indicate that the availability of water infrastructure increases agricultural productivity and the likelihood of successful harvest as measured by harvested cropland as a share of total cropland. Although dams, in our sample, may have had little impact on agricultural composition and agricultural productivity during normal precipitation years, when we interact the indicator variables for dams and droughts, however, we find that the counties that have dams experienced a decrease in wheat acreage and a decrease in failed acreage. Relative to similar counties that did not have dam- provided irrigation water, counties with the dams were able to have more of their total acreage planted and lost less of the crop that was planted. Similarly, our results show that the presence of dams in areas that are suffering from droughts has a positive, significant impact on wheat productivity and barley productivity. In addition, our results also indicate that the presence of major water infrastructure helped to mitigate the damages experienced during periods of elevated precipitation. We find that periods of excessive precipitation result in a significant, positive effect on agricultural productivity of wheat and hay, the two dry- farmed crops in our sample. Again, the presence of a dam alone only has a limited impact on agricultural productivity, but when this effect is interacted with the presence of a period of excessive precipitation, we find a positive impact for all agricultural productivities studied. Our results indicate that the productivity impacts on wheat and corn are statistically significant and positive. flect the changing availability of well water over time, nor does it represent the volume of well water available. As a robustness check, we introduce the well availability as both a time invariant indicator variable and as a dummy variable (which is, in turn, instrumented for endogeneity) for all years after 1950, when most western farms uniformly had access to electricity. The inclusion of these well variables slightly reduces the magnitudes of our variables of concern but doesn’t alter their significance, signs, or their general magnitudes.

Climate Variability and Water Infrastructure

279

Analysis of historical data from the western United States, where temperature and precipitation differences relative to the eastern part of the country are greater than those predicted from climate change models, allows for assessment of adaption policies. We present a large- scale time series analysis of historical data on mitigation of the impact of climate change on agricultural production. Because there is evidence that mean changes in temperature may have less impact on agricultural production, our focus is on extremes of precipitation levels that are likely to cause significant negative agricultural outcomes. This approach follows the recommendation of Reilly (1999) and others regarding analysis of the impact of climate change on agricultural production. Our results indicate that historical water supply investments have increased agricultural productivity and improved the likelihood of successful harvest during times of climatic shocks—droughts and floods. These are characteristic of the semiarid West and may become more so if climate projections of greater swings in drought and flooding come to pass.

References Anderson, H. 2009. Personal communication with Scott E. Lowe. November. Administrator of the Planning and Technical Services Division of the Idaho Department of Water Resources. Andreadis, K. M., and D. P. Lettenmaier. 2006. Trends in 20th century drought over the continental United States. Geophysical Research Letters 33, doi:10.1029/ 2006GL025711. Brouwer, C., and M. Heibloem. 1985. Irrigation water management. Training Manual no. 1. Rome, Italy: FAO. Cline, W. R. 1996. The impact of global warming on agriculture: Comment. American Economic Review 86 (5): 1309–11. Contor, B. 2009. Personal communication with Scott E. Lowe. November. Research Hydrologist, Idaho Water Resources Research Institute. Deschênes, O., and M. Greenstone. 2007. The economic impacts of climate change: Evidence from agricultural output and random fluctuations in weather. American Economic Review 97 (1): 354–85. Duflo, E., and R. Pande. 2007. Dams. Quarterly Journal of Economics 122 (2): 601–46. Intergovernmental Panel on Climate Change (IPCC). 2001. Executive Summary. Working Group II, chapter 18. Geneva, Switzerland: IPCC. http://www.grida.no/ publications/other/ipcc_tar/?src⫽/climate/ipcc_tar/wg2/index.htm. Kelly, D. L., C. D. Kolstad, and G. T. Mitchell. 2005. Adjustment costs from environmental change. Journal of Environmental Economics and Management 50: 468–95. Lettenmaier, D. P., D. Major, L. Poff, and S. Running. 2008. The effects of climate change on agriculture, land resources, water resources, and biodiversity in the United States. Chapter 4: Water resources. U.S. Climate Change Sciences Program and the Subcommittee on Global Change Research, Synthesis and Assessment Product no. 4.3 Report.

280

Zeynep K. Hansen, Gary D. Libecap, and Scott E. Lowe

Libecap, G. D., and Z. K. Hansen. 2002. “Rain follows the plow” and dryfarming doctrine: The climate information problem and homestead failure in the Upper Great Plains, 1890–1925. Journal of Economic History 62 (1): 86–120. Mendelsohn, R., W. D. Nordhaus, and D. Shaw. 1994. The impact of global warming on agriculture: A Ricardian analysis. American Economic Review 84 (4): 753–71. Nordhaus, W. D. 2007. A review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature 45 (September): 686–702. Olmstead, A. L., and P. W. Rhode. 2008. Abundance: Biological innovation and American agricultural development. New York: Cambridge University Press. Pielke, R. A., Jr. 1998. Rethinking the role of adaptation in climate policy. Global Environmental Change 8 (2): 159–70. Pisani, D. J. 2002. Water and American government: The Reclamation Bureau, national water policy, and the West, 1902–1935. Berkeley, CA: University of California Press. Reilly, J. 1999. What does climate change mean for agriculture in developing countries? A comment on Mandelsohn and Dinar. World Bank Research Observer 14 (2): 295–305. Robinson, W. W. 1979. Land in California: The story of mission lands, ranches, squatters, mining claims, railroad grants, land scrip, homesteads. Berkeley, CA: University of California Press. Schlenker, W., M. Hanemann, and A. C. Fisher. 2006. The impact of global warming on U.S. agriculture: An econometric analysis of optimal growing conditions. Review of Economics and Statistics 88 (1): 113–25. Schlenker, W., and M. Roberts. 2008. Estimating the impact of climate change on crop yields: The importance of nonlinear temperature effects. NBER Working Paper no. 13799. Cambridge, MA: National Bureau of Economic Research. Stern, N. 2007. The economics of climate change: The Stern Review. New York: Cambridge University Press. U.S. Department of Agriculture, National Agricultural Statistics Service. 2009. Statistics by subject. http://www.nass.usda.gov/Statistics_by_Subject/index.php. Weitzman, M. L. 2007. A review of the Stern Review on the Economics of Climate Change. Journal of Economic Literature 45 (September): 703–24.

10 Did Frederick Brodie Discover the World’s First Environmental Kuznets Curve? Coal Smoke and the Rise and Fall of the London Fog Karen Clay and Werner Troesken

10.1

Introduction

In 1903, at a meeting of the Royal Meteorological Society, a British climatologist named Frederick J. Brodie presented a deceptively simple paper. Using data from the Brixton weather station in London, Brodie graphed the number foggy days per year between 1871 and 1903. His data, reproduced here in figure 10.1, revealed an inverted U- shaped pattern: the annual number of foggy days in London rose during the 1870s and 1880s, reversed trend sometime around 1888 or 1889, and then fell steadily during the 1890s and early 1900s. Brodie attributed the rise and fall of the London fog to variation in the production of coal smoke. During the 1870s and 1880s, Brodie claimed, London businesses and homeowners burned coal with reckless abandon, filling the atmosphere with soot and giving rise to dense and dark fogs. After 1890, however, technological, legal, and social changes enabled, or forced, homeowners and businesses to burn coal more efficiently and cleanly. In particular, the expansion of gas for heating and cooking, and electricity for lighting curtailed domestic sources of smoke, and the London Coal Smoke Abatement Society lobbied local authorities to enforce the Public Health Acts, which required manufacturers to adopt low- smoke technologies (Brodie 1905, 15–20). In this chapter, we evaluate Brodie’s claim using more data than he had access to at the time and modern econometric techniques. Three types of evidence will be considered. First, evidence on foggy days and coal conKaren Clay is an associate professor of economics at Carnegie Mellon University, and a faculty research fellow of the National Bureau of Economic Research. Werner Troesken is a professor of economics at the University of Pittsburgh, and a research associate of the National Bureau of Economic Research.

281

282

Karen Clay and Werner Troesken

Fig. 10.1

Annual number of foggy days in London

Source: See text.

sumption are presented. Both are trending up over time, particularly after 1850. Second, we present a brief history of smoke abatement technologies and the enforcement of pollution control laws during the Victorian and Edwardian eras. This history will help establish a circumstantial case that such actions helped make London a less smoky place. Third, using a procedure we describe as a reverse- event study, we use large and unusual spikes in weekly mortality data to identify the frequency and severity of fog- related events. Fog- related events were severe and persistent episodes of fog that culminated in spikes in mortality, particularly from respiratory diseases. The reverse event study is a centerpiece of the chapter, and the most novel part of our analysis. The results suggest that by the early twentieth century, fogs and the associated spikes in mortality had largely abated. The ostensible rise and fall of the London fog should interest economists and economic historians on at least two levels. First, the debate surrounding Brodie’s paper prefigures current debates about the environmental Kuznets curve (EKC). Second, the basic question Brodie raised is of no small historical moment: how could there have been a meaningful environmental movement in Victorian England? We tend to think of Victorian Britain as a place where the expedience of cholera and child labor trumped the expense of clean water and a decent wage, as a place where environmental and social degradation served as the handmaidens of avarice and economic development. Yet Brodie’s paper suggests something quite different, as does even a cursory look at the political and economic history of smoke abatement.

Did Brodie Discover the First Environmental Kuznets Curve?

283

The chapter also has implications for understanding climate change. First, it shows that industrialization can change climate and that this type of climate change can sometimes be reversed by reducing pollution levels. Second, one can speculate on the effect that current global warming would have had on London had it occurred during the late nineteenth and early twentieth centuries. Global warming would probably have reduced toxic fogs in two ways. First, it would have reduced coal consumption. Second, it would have increased the number of days above 40 degrees. Above 40 degrees, the fog was less likely to form. At the same time, warming would likely have exacerbated other issues such as water quality, which was generally poor and considerably worse in the summer. Very little evidence exists on this tradeoff, but it is important to recognize that there was a trade- off. 10.2

Coal and Fog

Figure 10.2 shows that the number of foggy days in London appears to have risen with per capita coal consumption. The first series, given by the small hollow circles, is the number of foggy days in London from 1730 to 1910. The second series, given by the small black circles, is coal consumption

Fig. 10.2

Coal consumption per capital and foggy days in London, 1700–1925

Sources: The coal data are from the 1899, 1908, and 1919 volumes of The Coal Trade by Saward; the Times (July 23, 1901, 11; December 1, 1913, 26; August 11, 1927, 20); Nature (November 5, 1891, 12); and for earlier years, Mitchell (1988), a source which also provides data on London’s population. The fog data are derived from the sources described in the text, especially Brodie (1891, 1905) and Mossman (1897).

284

Karen Clay and Werner Troesken

per capita. We use imports of coal into London as a proxy for consumption. Nearly all coal imported into London was consumed there. Before 1800 and the Industrial Revolution, the number of foggy days in London is stagnant, hovering around twelve days per year. Coal consumption per head shows only slight growth during the same period. After 1800 and the onset of industrialization, both series begin to rise, and after 1850, the growth is exponential. The foggy days measure rises threefold between 1850 and 1890, increasing from around twenty- five to more than seventy- five days per annum. Similarly, annual coal consumption rises by a factor of 2.5, from one ton per head in 1850 to 2.5 tons in 1890. Things change abruptly for both series around 1890. Foggy days per annum reverses trend and plummets by around 85 percent within a twenty- year interval. Although coal consumption does not reverse trend, it stagnates, showing no growth over the next thirty years. One important concern about the fog data is that it is noninstrumental. That is, the individual collecting the data went outside and made a subjective determination about whether it was foggy. Other series, which are available over shorter time periods, show similar trends in fog. Brodie’s original series was for the Brixton station, which was located five miles south of the Thames. Data is available for a second station, West Norwood, which was another ten miles south. The qualitative patterns are very similar. Fog data compiled by Lempfert (1912) also shows similar patterns. Lempfert’s instrumentally collected data on sunshine, which we will discuss shortly, also support these basic patterns. The renowned nineteenth- century scientist Rollo Russell attributed the length and severity of the fog to coal consumption. According to Russell, the London fog began early in the morning around 6:00 a.m., when the city, or parts of the city, were enveloped by an “ordinary thick white fog.” Soon after this, the city would awaken by lighting “about a million fires.” These fires charged the atmosphere with “carbonaceous particles,” which upon cooling, attached themselves to the spheres of water that constituted the fog. Ordinarily the warmth of the sun would have quickly dissipated the fog, but the smoke and an oily tar that surrounded the spheres of water impaired this process. In these conditions, city residents would not have sunlight until noon.1 In an article in Nature, W. J. Russell (1891, 11) developed a similar line of thought, arguing that coal and sulphur particles induced the formation of fog by offering gaseous water a surface on which to condensate.2 1. This passage quoting Russell is from Hann’s Handbook of Climatology (1903, 78). A similar description is offered in Russell (1906). 2. Russell (1891, 11) wrote, “The dust always present in the atmosphere offers this free surface to the gaseous water, and thus induces its condensation. This specific action of dust varies very considerably, first with regards to its composition, and second with regard to the size and abundance of the particles present. Sulphur burnt in the air is a most active fog producer, so is salt.” For a similar statement regarding the origins and persistence of smoke- laden fogs, see Frankland (1882).

Did Brodie Discover the First Environmental Kuznets Curve?

285

Late twentieth century scientists concur. For example, Brazell (1968, 102) writes, “London fogs are particularly obnoxious because the fog droplets tend to form on minute particles of atmospheric pollution which are usually produced by the combustion of coal, oil, and petrol.” With or without the coal smoke, though, London would have been subjected to much ordinary fog. Two government reports issued in the early 1900s showed that the formation of fog in London was correlated with temperature, humidity, and wind speed. Fogs were much more likely when the temperature in the city was below 40 degrees (no dense fog was observed when temperature was above 40), when humidity was high, and when the winds were calm. Britain’s “latitudinal and continental position” was of special importance because it left the whole country in the path of sequences of “migratory depressions” and anticyclones (Chandler 1965, 35). Anticyclones, and the temperature inversions that accompanied them, played a central role in the propagation of a certain kind of London fog. A short article in the journal Notes and Queries (March 2, 1878, 178) provides the clearest contemporary statement we have found on the significance of anticyclones. We quote it in full: These fogs are not caused by the rarefaction of the air, or by the consumption of gas, nor yet by the hills on the north, nor by the river. The peculiar atmospheric condition termed an anti- cyclone is the real cause of these annoying visitations; the wind is then blowing round a well defined circle, in the centre of which the air is tranquil, and consequently the smoke, condensed vapours, & c., cannot escape as they do when there is a direct onward movement of the wind. The pressure of the atmosphere at such times is almost invariably greatly in excess of the average in the midst of the anti- cyclone, which, by preventing the rise of the smoke, & c., increases the intensity of the fog. Whenever, therefore, an anti- cyclone occurs with London at or near the centre, there must necessarily be a “London fog,” the density of which will be in proportion to the smoke evolved at the time. The same phenomenon may be observed in other places within the anticyclonic circle, but of course in a less degree of density. A 1910 Times of London article argued that air quality had improved significantly. “Visitations” of smoke- laden fogs were “rare” by 1910 and “seldom” continued “without intermission for more than two or three days.” The opening paragraph began with the reporter’s own assessment of the situation, an assessment he clearly believed was unassailable: The decrease in recent years, not only in the frequency but in the intensity of London fog, is a matter which admits no serious question. Persons who have reached middle age well remember the time when dense smoke fogs of the worst possible description were a common feature of the winter season, and lasted not infrequently for a week or more at a stretch.

286

Karen Clay and Werner Troesken

The same article also presented data on the hours of bright sunshine in London, which presumably would have been inversely correlated with the incidence of fog. Sunshine, in contrast to fog, was measured instrumentally using a device known as the “Campbell Stokes Sunshine Recorder.” The recorder consisted of a clear ball that magnified bright sunlight and gradually burned away a piece of cardboard. The sunshine data found in the Times are broadly consistent Brodie’s data on the fog; they suggest that improvements in London’s air began five to ten years earlier, during the early to mid 1880s as opposed to the early 1890s (Times, December 27, 1910, 11).3 Contemporary observers agreed that London was becoming less smoky and foggy, but disagreed as to why. The Times article subscribed to Brodie’s view: The diminution of smoke which has taken place within recent years may be attributed in a large measure to a more vigorous enforcement of the smoke prevention clauses of the Public Health Act, but it has in all probability been materially aided by the increased use of gas fires for both heating and cooking purposes, and also by improved methods of lighting (Times, December 27, 1910, 11). Russell, in contrast, emphasized London’s changing geography and broader weather patterns that were affecting the entire south of England. For Russell, the declining incidence of fog was not unique to London fog, but common to all cities and towns in the region. R. G. K. Lempfert an accomplished climate scientist and the superintendent of the Forecast Division of the Royal Meteorological Office presented evidence on regionwide weather patterns. “It is my object,” he wrote (1912, 23), “to examine the statistics of bright sunshine for London and other large towns to see whether they afford evidence of progressive amelioration or the reverse of the smoke nuisance.” Lempfert’s identification strategy was simple. If London’s atmosphere was becoming more sunny because of purely meteorological phenomena, those same phenomena would have affected surrounding rural areas as well. If, however, London’s atmosphere was improving because of innovations (both regulatory and technological) unique to metropolitan areas, London would have become increasingly sunny relative to the neighboring control areas. Furthermore, because far 3. This was not the first time that the Times argued that London’s atmosphere was becoming cleaner. In an editorial published a few months before Brodie’s paper, the Times (December 24, 1904) wrote, “we think no one whose experience of London extends over many years can entertain the slightest doubt that the fogs of the present day, even the worst of them, are definitely less filthy and less opaque than those of the early or middle Victorian period. The change is commonly, and perhaps right, attributed to the extent to which the production of smoke in the metropolis has been diminished by legislation.” See also, Schlicht (1907, 685), who in an article published in the Journal of the Society of Arts, wrote, “It must be said . . . that in recent years, thanks to admirable efforts of the Coal Smoke Abatement Society, and the exploitation of gas as a substitute for coal by the gas companies, the atmosphere of London is much less offensive than it was twenty- five or thirty years ago.”

287

Did Brodie Discover the First Environmental Kuznets Curve? Table 10.1

Duration of bright sunshine at London stations as a proportion of the duration at neighboring country stations (London)/(country)

Interval 1881–1885 1886–1890 1891–1895 1896–1900 1901–1905 1906–1910

(Kew)/(country)

(London)/(Kew)

Winter

Summer

Winter

Summer

Winter

Summer

.17 .29 .32 .35 .32 .38

.83 .85 .95 .89 .93 .92

.85 .89 .83 .97 .88 .81

.97 .96 1.02 1.03 1.04 .99

.20 .32 .39 .36 .37 .46

.84 .87 .94 .86 .89 .92

Source: Lempfert (1912, 25).

more coal was burnt during the winter months than the summer, if reductions in coal smoke were driving the improvement in London’s atmosphere, one should observe greater relative improvement when we restrict the sample to the winter (Lempfert 1912). Lempfert’s data, which are reproduced in table 10.1, suggest that weather patterns were relatively stable. In first two columns of data, the table expresses the duration of bright sunshine at two London weather stations (Westminster and Bunhill Row) as a proportion of the duration at four nearby “country” stations (Oxford, Cambridge, Marlboro, and Geldeston). Notice that for the first five- year interval, 1881 to 1885, London in the winter enjoys only 17 percent of the sunshine experienced in the control areas; by the last five- year interval, 1906 to 1910, London’s relative sunshine rate has more than doubled, to 38 percent. There is evidence of improvement during the summer—the relative sunshine rate grows from 83 to 92 percent—but the improvement is much less pronounced than that observed during the winter months. The third and forth columns of data perform the same experiment for the weather station at Kew Gardens as the (placebo) treatment. The Kew station resided on the western edge of London (today, about ten miles directly east of Heathrow Airport) and was relatively immune from the smoke problems that plagued the rest of the metropolis.4 Kew shows little relative improvement in the duration of sunshine, in either winter or summer. The final two columns of data compare sunshine at the city stations to that observed at the Kew station. As when the country stations were used as controls, there is evidence that the city stations became increasingly sunny relative to the station at Kew. Again, the improvement is concentrated in the 4. The word “relative” is important. Kew was not entirely immune from the effects of London smoke, and there is a small literature that documents how the gardens were adversely affected by the city’s production of coal smoke. See, for example, J. W. Bean’s article “A Note on Recent Observations of the Smoke Nuisance at Kew Gardens,” presented at Coal Smoke Abatement Conference in 1912.

288

Karen Clay and Werner Troesken

winter months, with the relative sunshine rate rising from 20 to 46 percent during the winter and from 84 to 92 percent during the summer.5 10.3

Victorian Environmentalism

Figure 10.3 plots the natural log of tons of coal imported into London, which is a measure of total coal consumption over time. The observed log is plotted by the empty circles. A vertical reference line is plotted at the year 1890. Figure 10.4 also includes several trend lines that identify changes in slope over different historical intervals. Together with figure 10.1, figure 10.3 shows that coal imports were increasing rapidly overall and in per capita terms. Not until sometime after 1890 does this pattern of increasing growth cease. There is clear evidence that after this point, for the next thirty years, coal consumption is well below trend. The patterns in figures 10.1 and 10.3 seem broadly consistent with Brodie’s explanation of the rise and fall of the London fog. When coal consumption in England grew slowly, and in some absolute sense, was not large, London fogs were much less frequent than they would later prove to be. The data at the end of the period are problematic, however. How could stabilization in the per capita consumption of coal initiate a decline in fog? If increases in coal consumption per capita drove the increase in fog, the converse—that a decline in coal consumption in per capita drove the decrease in fog—would also have to be true, would it not? Moreover, to the extent that smoke density was determined solely by the amount of coal consumed, one should not even bother looking at coal consumption per head; all that should matter is total consumption of coal. By this logic, a reduction in smoke required a reduction in the total amount of coal consumed. We think that three changes reduced the effect of coal, in the absence of declines in consumption. The first change involved dispersing people and smoke over a larger area, in effect, diluting the smoke. The second change involved the Public Health Act. This law empowered police in metropolitan London to fine manufacturers throughout the area for dense smoke emissions. Enforcement of this law encouraged firms to conserve on soft coal. It is important to emphasize that none of this need to have been rational or profit maximizing from a given firm’s perspective. Logic suggests it must have cost firms more to economize on soft coal than the coal was worth; otherwise, they would have adopted smoke abatement technologies and 5. Lempfert (1912, 27–28) offered a caveat regarding instrumental measures of bright sunlight. The glass balls employed in the Campbell-Stokes recorders began to yellow and became less sensitive to sunlight with time. Lempfert argued that this would lead observers to understate increases in the incidence of bright sunlight in London. Unfortunately, it would also have had the same effect in the control counties. Lempfert suggested that, whatever the bias, these effects would have been small and that officials took steps to minimize the resulting measurement errors. Nevertheless, the caution should be noted.

Did Brodie Discover the First Environmental Kuznets Curve?

Fig. 10.3

289

Total coal consumption in London (in logs), 1700–1920

Source: See note to figure 10.2.

practices without the impetus of the Public Health Act. The third change involved households. Complementing the actions of manufacturers, London homeowners also became more cognizant of ways to conserve on coal and adopted gas for cooking and heating or purchased grates and boilers designed to minimize smoke emissions and wasted fuel. 10.3.1

Population Redistribution

As population increases, the amount of coal consumed rises, which, in turn, increases the absolute amount of smoke and smoke density. But there is also a countervailing effect. As population increases, so, too, does the area of the city. New migrants do not only move to previously settled areas of the city; they also take up residence in outlying areas that were previously unsettled. Redistributing population from a densely populated core to a less densely populated periphery would also change the distribution of smoke. Before continuing, we need to more clearly define what is meant by the geographic descriptor “London.” Today, London proper covers only 1.2 square miles; Greater London covers nearly 660 square miles. Thus far, all of our references to London have been to Greater London. We will continue this practice, but will also draw the distinction when appropriate. We did not raise this fine point of urban geography earlier because it was not relevant until now.

290

Karen Clay and Werner Troesken

Fig. 10.4

Population density in London proper and Greater London, 1801–2001

Source: Poore (1883) for nineteenth century; London Statistics various years.

Figure 10.4 shows the reallocation of population in London over time. It plots population density in London proper (the core) and the whole of Greater London from 1811 through 2001. Population density in London proper starts the series at a remarkable 400 persons per hectare and holds steady there until 1861, when it implodes. By 1911, population density in London proper falls to 103.6. Population density for Greater London rises from .6 persons per hectare in 1811 to 1.44 in 1861, after which it begins a more rapid ascent. By 1911, population density in Greater London rises to 3.5. These data suggest that one path to solving London’s smoke problem began to emerge three decades before Brodie’s inflection point around 1890. Poore (1893) indicated that it was not just London proper that experienced an outflow of people. The combined population of all five of London’s central districts—Holborn; The Strand; St. Martins; St. Giles; and the city proper—dropped from a peak of 334,369 in 1861 to 247,140 in 1891.6 By driving people out of the center of the metropolis, the dense, smoky fogs of central London might have contained the seeds of their own demise. An article published in Science more than a century ago explained. Asking why the population density of Paris was so much higher than that of London, the unnamed author wrote (February 19, 1886, 173–74): 6. Poore (1893), it should be noted, believed that the increase in the area of the city was a cause of fogs.

Did Brodie Discover the First Environmental Kuznets Curve?

291

The average density of Paris is more than double that of London, and yet the streets are brighter and cleaner. The question probably turns more upon the prevention of smoke than any thing else. If the fog and gloom could be removed, and free access provided for the sunlight, there is no pleasanter [sic] or healthier place to live than the west end of London; and many who now endure, morning and evening, forty minutes’ journey through choking tunnels, and walk long distances to railway termini, would stay in town if they could be relieved from the depression which is the accompaniment of a murky atmosphere. 10.3.2

Law and Changing Technology

Although London had a long history of antismoke agitation (see Brimblecombe 1987, 10–18, 90–107), the movement began in earnest in 1880 with the creation of the National Smoke Abatement Institution (NSAI). A private institution located mainly in London but with ties throughout the United Kingdom, the NSAI organized exhibitions for inventors and manufacturers to display heating and cooking grates, stoves, steam boilers, smokeless fuels, and other devices designed to mitigate emissions of coal smoke. The group offered prizes for inventions deemed especially promising and created venues through which it sought to instruct manufacturers and homeowners in ways to conserve on coal. It also worked to dispel the widespread notion that coal smoke had disinfectant properties that destroyed airborne pathogens that would have otherwise carried diseases like tuberculosis and influenza (Report of the Smoke Abatement Committee 1883, 183; Nature, February 9, 1888, 356–68; Transactions of the Sanitary Institute of Great Britain, 1887–1888, 301–45). The members of the NSAI, like most participants in the late- nineteenthcentury smoke abatement movement, embraced an early incarnation of the Porter hypothesis, the idea that environmental regulations and controls might spur productivity advances. The NSAI believed that the technologies it recognized and helped introduce to the world would not only make nineteenth- century cities cleaner and less polluted, they would make businesses more efficient and profitable. Arguing that dense smoke represented imperfect combustion and wasted fuel, the advocates of this early Porterism claimed that proper stoking methods, specially designed coal grates, stoking machines, and various types of smoke consumers would enable manufacturers to conserve on coal (e.g., Report from the Select Committee on Smoke Prevention 1843; Report of the Smoke Abatement Committee, 1883, 183; British Medical Journal, August 20, 1908, 615). The idea that proper stoking prevented smoke and saved coal has a long history and was widely accepted, even among producers and manufacturers who were otherwise resistant to smoke abatement. In a paper read before the Royal Sanitary Institute, Caborne (1906) expressed the identical argument. Twenty year earlier in front of the same venue, Fletcher (1887–1888)

292

Karen Clay and Werner Troesken

claimed to have stated the obvious when he discussed the efficacy of proper stoking. A survey conducted by the London Coal Smoke Abatement Society in the early 1900s also supports the idea that factory owners believed this to be true. Surveying sixty- three London factories who had not been cited for excessive smoke in the last six months of 1904, the society elicited thirty- five useful responses. Of the respondents, just over half (thirteen) ascribed “their success in preventing the emission of smoke to careful stoking” (Rideal 1906, 149). Arguably the most important factor driving down pollution in the aftermath of the Public Health Act was the direct effect of having producers switch from soft coal to different varieties of hard coal. In the context of figure 10.2, the data on coal consumption per capita are for the consumption of all kinds of coal not just soft coal. We have found no data source that would allow us to construct separate times series for hard and soft coal. However, of the thirty- five firms responding to the aforementioned survey of the smoke abatement society, all but one used some variety of coke or anthracite. Seventeen firms used Welsh coal, a variety of anthracite (Rideal 1906). When miners in the south of Wales struck during the late 1890s, London factories that had been using Welsh coal were forced to use bituminous. Their furnaces ill- equipped for bituminous coal and unable to stoke their fires properly, the factory owners were “hauled up before sundry magistrates to show cause why they should not abate the smoke nuisance they [were] making” (Booth 1898, 1064; Public Health, August 15, 1898, 373). In the absence of legislation punishing smoke emissions, it is difficult to imagine that manufacturers would have switched from soft to hard coal, which all reports indicate was “much dearer” than soft coal and harder to light (Medical Times, March 11, 1882, 395; Booth 1898; Saward 1914, 156; Reynolds 1882, 167–80). Figure 10.5 plots the relative price of Welsh steam coal and ordinary (Newcastle) bituminous coal over the course of the nineteenth century and early twentieth century. The underlying price series are based on the supply prices at the mine, not delivered prices in London. Over the long term, the relative price of Welsh coal sold at a 20 percent premium throughout the period. This pattern rules out the possibility that the switch from soft to English coal to “smokeless” Welsh coal was driven by a reduction in the relative price of Welsh coal. On the contrary, the trend line indicates a nontrivial uptick in the relative price of Welsh coal sometime after 1885. This pattern is consistent with the hypothesis that demand for smokeless coals increased relative to the demand for ordinary bituminous coal as a result of regulatory or political pressures. As manufacturers switched from soft coal to hard, they also indirectly helped reduce overall coal consumption. Owens (1912, 93–94) provides a useful explanation of this mechanism. Because soot was a “very bad conductor of heat,” when boiler plates became covered in soot, “the rate of heat transfer [was] reduced.” Owens conducted a series of experiments to mea-

Did Brodie Discover the First Environmental Kuznets Curve?

293

Fig. 10.5 Price of steam coal in Swansea (Wales) divided by price of bituminous coal in Newcastle Sources: Price series are from Wright (1905, 409–20). Alternative price series that yield similar patterns can be found in Great Britain Board of Trade (1903, 12–19). Note: Trend line estimated with STATA using a running line smoother, bandwidth of .8, and lowess option.

sure the amount of heat that might have been lost when factory owners and operatives allowed soot to build up on boiler plates. He found that a layer of soot 1/20 of an inch thick reduced the transfer of heat by 15 percent. One way manufacturers could prevent the development of coal soot on boiler plates was to use hard coal intermittently or mixed with the bituminous coal. Because Welsh and anthracite coals generated less soot and tar than bituminous coal, even their intermittent use delayed the formation of soot on boiler plates and enabled workers to generate more heat than they otherwise would have when they used bituminous coal alone. It was not uncommon for manufacturers in turn- of- the- century London to use a mixture of smokeless coal and bituminous coal (Rideal 1906). Aside from the switch from bituminous to Welsh coal among manufacturers, there was also a more voluntary transition taking among consumers. An observer who placed heavy emphasis on the voluntary adoption of new technologies was Sir George Livesey, an officer of a large London gas company and leader of the city’s smoke abatement movement. In a paper presented to the Royal Sanitary Institute, Livesey argued that London had become a less smoky place because more and more consumers, especially those among the working classes, were using gas stoves rather than coal for

294

Karen Clay and Werner Troesken

cooking. The spread of gas cookers among the poor was the result of two pricing strategies adopted by London gas companies: one was the practice of renting stoves to customers; the other was the coin- in- the- slot method of paying for gas. Of the 834,000 households in London that purchased gas from the city’s three major gas companies, 70 percent used gas stoves for cooking, and that number would have been even higher if homes and apartments had had sufficient space for a gas stove and opposition from landlords had not been so strong (Livesey 1906). While one might dismiss Livesey’s arguments and opposition to coal smoke as patently self- serving, ten years before he published his paper, the British Medical Journal (August 22, 1896, 465) published data and evidence very much in keeping with his otherwise partisan observations. Surprisingly the British Medical Journal suggested that it was the wealthy, not the poor, who were slow to adopt gas for cooking.7 10.4

Fog-Related Events: History, Identification, and Health Effects

London’s most famous fog- related event occurred in December 1952 and is documented in William Wise’s popular book Killer Smog: The World’s Worst Air Pollution Disaster. The fog began on December 5 and did not lift for five days. Government officials estimated that there were roughly 4,000 excess deaths because of the fog, mostly due to respiratory complaints such as asthma and bronchitis. 7. The British Medical Journal (August 22, 1896, 465) wrote: A new and unexpected agency is having a most beneficial effect in contributing to the abatement of the smoke nuisance in London. The relative clearness of the London atmosphere within the last twelve months has been plainly apparent, and the smoke cloud which obscures the London atmosphere appears to be progressively lightening. Mr. Ernest Hart, Chairman of the Smoke Abatement Exhibition in London, frequently pointed out that the greatest contributors to the smoke cloud of London were the small grates of the enormous number of houses of the poor, and a great deal of ingenuity had been exhausted with relatively little success in endeavoring to abate this nuisance. The use of gas fires was urgently recommended, but had hitherto been difficult, owing to its cost and want of suitable apparatus. The rapid and very extensive growth of the use of gas for cooking as well as lighting purposes by the working classes, due to the introduction of the “penny in the slot” system, is working a great revolution in the London atmosphere. During the last four years, the South London Gas Company alone has fixed 50,000 slot meters, and nearly 38,000 small gas cooking stoves in the houses of the working man. This movement is still making great progress, and we hope means may be found to extend it to the houses of the more comfortable classes. The enormous improvement in the London atmosphere, and the clearing away of a smoke pall which hangs over London, may then be anticipated. See also Ackermann (1906), who pleaded with London officials to promote the spread of producer gas (a cheap, low- grade type of coal gas appropriate for heating and cooking but not lighting) to replace coal. Martin (1906) presented data to indicate the threshold price at which it would be economical for households to switch from coal to producer gas. Fifteen years before the article in the British Medical Journal, Alfred Carpenter, an independent scientist, proposed heavily taxing coal use in London, thereby encouraging homeowners and manufacturers to adopt gas for lighting, heating, cooking, and mechanical propulsion. See Carpenter (1880).

Did Brodie Discover the First Environmental Kuznets Curve?

295

Although we might know this event best, that does not necessarily mean it was the worst. Earlier fogs lasted longer, and there is evidence that they took a heavier death toll. For example, the aforementioned cattle- show fog of 1873 lasted a week (Brazell 1968, 111). The December fog of 1879 lasted nearly two weeks, darkening London’s skies from December 3 through December 27 (Brazell 1968, 111; Scott 1896). There was also the aptly named “anticyclonic winter” of 1890 to 1891, a two- month interval of almost uninterrupted fog (Brodie 1891). Estimates presented in the following suggest that this event generated 7,405 excess deaths, nearly twice the number of excess deaths observed during the winter of 1952. 10.4.1

History

Imagine London in the late nineteenth century during a fog. Although it is noon, the city is as dark as night. People must uses torches just to see a few yards ahead of them. Horse- drawn carriages cannot move; trains crawl at a snail’s pace, and, in some instances, cease operating. People described the darkness as fog because it was wet and heavy and because it almost always occurred during unusually cold and calm conditions when fog was otherwise common. But it was also more than just fog. The darkness burned the eyes and throat. Deaths from all causes, but especially bronchitis and other acute respiratory diseases, spiked during such dense fogs and immediately after they lifted. First- hand observers blamed the coal smoke trapped in the fog: “There was nothing more irritating than the unburnt carbon floating in the air; it fell on the air tubes of the human system, and formed that dark expectoration which was so injurious to the constitution; it gathered on the lungs and there accumulated” (Times, February 7, 1882, 10). Another observer wrote (Medical Times, March 11, 1882, 395), “After a fog the nostrils are like chimneys, and are lined with a layer of black smut. The expectoration is black from the amount of carbon arrested in the mucus of the air passages. For a day or two after exposure to a smut- laden atmosphere, black phlegm is brought up.” During a cattle show in 1873, the fog was so thick that the Queen’s prize bull dropped dead, as did several other large animals. If all this were not enough, imagine, too, that the fog went on for days at a time and, in some extraordinary cases, weeks. Although a composite of several of the most famous London fogs, these images convey what it was like to experience the most dense and persistent ones.8 8. This composite is based heavily on a report in the Times (February 6, 1882, 7). The quotation is from a talk by Dr. J. M. Fothergill recorded in the Times (February 7, 1882, 10). Other articles in the Times that corroborate this picture, include January 21, 1861, 9; November 19, 1862, 6; and November 26, 1858, 10. That the Queen’s prize bull was killed raises the question about how the monarchy and elite groups responded to the fog. Centuries before the Public Health Act was passed in 1891, the King had barred the burning of coal in London, and wealthy elites were at the forefront of smoke eradication; poorer groups were less concerned or opposed. This cross- sectional pattern is consistent the modern EKC and, in a political

296

Karen Clay and Werner Troesken

After a severe fog in early February of 1882, the Times (February 13, 1882, 10) published the results of the coroner’s inquest into several fog- related deaths. The results provide a window into the physiological mechanisms that made some fogs so deadly. James Smith, aged sixty, was a wheelwright who had been “had been suffering from chest affection some time past.” Although his wife “begged him” not go out in the fog, he went out anyway. When he returned, he was “very ill,” and he died a few days later. The coroner ruled that “the fog had hastened his death very materially, increasing and developing bronchitis to an alarming extent.” Alice Wright, aged sixty- six, went out when the “fog was the thickest, to fetch her mangling.” Twenty minutes later, a passerby found her lying in a passage and ran into the home to search out her daughter. Finding Wright unconscious, the daughter brought her inside and called for a doctor, “who found the poor creature dead.” The postmortem “showed that the fog had brought on effusion on the brain.” William Henry Pepper, aged three months, was the son of a blacksmith. Although he had been a healthy baby, he took ill after he and his mother had ventured into the fog. The coroner concluded that “the child’s lungs” were “too weak to resist the poison which had filtered into them.” “Bronchial pneumonia” set in and “death resulted.” For convenience, we refer to extreme conditions like those just described as a “fog- related event.” Fog- related events were an extreme form of smog. The available evidence indicates that they were associated with anticyclones and temperature inversions (Brodie 1891; Scott 1896; Times, December 13, 1873, 7; Wise 2001, 15–18). As the preceding discussion suggests, a defining features of a fog- related event is an unusually large number of deaths, especially from acute respiratory diseases. One of the clearest contemporary statements on the spike in death rates that accompanied fog- related events comes from a short article in the British Medical Journal. The article described an event that occurred in February of 1880 (February 14, 1880, 254): If one or two weeks during the cholera epidemic of 1849 and 1854 be excepted, the recorded mortality in London last week was higher than it has been at any time during the past forty years of civil registration. No fewer than 3,376 deaths were registered within the metropolis during the week ending Saturday, showing an excess of 1,657 upon the average number in the corresponding week of the last ten years. Of these deaths, most were attributable to respiratory diseases, particularly bronchitis: The excess of mortality was mainly referred to diseases of the respiratory organs, which caused 1,557 deaths last week, against 559 and 757 in the two preceding two weeks, showing an excess of 1,118 upon the corrected economy sense, sounds plausible. To the extent that the costs of smoke eradication were born disproportionately by the poor, one would expect that the wealthy segments of society to have been the primary advocates environmental improvements.

Did Brodie Discover the First Environmental Kuznets Curve?

297

weekly average. The fatal cases of bronchitis, which had been 531 in the previous week, rose to 1,223 last week. The last time the weekly death rate in the metropolis had approached these levels was during the cattle- show fog of 1873. 10.4.2

Identification

Ideally, we would like to have a complete history of all of the fog- related events in London during the late nineteenth and early twentieth century. If we possessed such a history, and that history was independent of Brodie’s fog observations, we could look at the frequency and severity of fog- related events before and after 1890. If Brodie’s data were correct, we would expect to observe increasing frequency and severity in the years leading up to 1890 and decreasing frequency and severity in the years following. The difficultly, however, is that there is no formal history of fog- related events in London that claims to be comprehensive in any sense. The question, then, is how to construct such a history. One approach might be to return to the records of the Royal Meteorological Office (which Brodie used) and search for extended periods of dense fog. This approach is problematic: there were periods of persistent and dense fog in London that were not associated with unusually high mortality and would not fall under our definition of a fog- related event. As an alternative strategy, one might scour the Times of London and other contemporary news outlets for articles that described phenomena consistent with fog- related events. Besides the high search costs and subjectivity of this approach, one could never be certain of identifying all relevant events. The British Medical Journal article discussed in the preceding suggests an econometric strategy for identifying events. The article observed that deaths spiked during the weeks associated with fog- related events and that these spikes were large and uncommon. Our identification strategy is an event study in reverse: we use spikes in the data to predict events, as opposed to using events to predict spikes in the data. To implement this strategy, we proceed as follows. We first collect weekly data on deaths in London during the late nineteenth and early twentieth centuries and calculate the weekly crude death rate. Using a simple regression framework—the weekly crude death rate is regressed against a few control variables—we then estimate a predicted death rate for the each of the weeks in our sample. From there, it is a simple matter to calculate a residual death rate for each week. The value of the residual establishes a necessary, but not a sufficient, condition to qualify as an event week. In particular, to be coded as an event week, the week must satisfy the following three conditions: (c.1) The residual death rate for the week must exceed .1. The choice of .1 as a threshold is not entirely arbitrary. Residuals above .1 are rare—they are 1.1 percent of the total sample—and as shown in the following, they are probably generated by a different set of forces than those below the threshold.

298

Karen Clay and Werner Troesken

(c.2) There must be supporting qualitative evidence in the Times indicating the presence of unusually dense and persistent fog during the week or sometime in the preceding week. (c.3) The preponderance of evidence must suggest that the spike of was not caused by something other than fog, such as an epidemic disease like cholera or influenza. If the weight of the evidence suggests an epidemic disease was present and important, the week is coded as a nonevent week. The online catalogues of the British Medical Journal (at PubMed Central) and the Times are searched for such evidence. As reported in the following, there are thirty- two weeks out of the sample that satisfy all three of these requirements and are coded as event weeks. The data on deaths per week are gathered from the Weekly Returns of the Registrar General, which recorded deaths for several of England’s largest cities including London. For years when the published volumes of the Registrar General are unavailable, the Times is consulted. After 1888, the Times regularly summarized the weekly returns of the Registrar General.9 Death rates are calculated as deaths per 1,000 persons and are constructed using interpolated population data from Mitchell (1988, 673). Note that we calculate and report a true weekly death rate, not an annualized weekly death rate. The sample period covers the fifty- five- year interval extending from 1855 to 1910, yielding a maximum possible sample size of 2,860 (52  55). There are, however, twenty- five weeks for which data are unavailable. Dropping these from the sample, 2,835 observations remain. Basic plots suggest that toxic fog was having an adverse effect on the health of Londoners, even as mortality rates were falling. Figure 10.6 plots the annual crude death rate in London, which is falling steeply over time. Figure 10.7 plots deaths from respiratory diseases and bronchitis in London from 1850 through 1920. Of the two series, we believe the bronchitis series is superior because bronchitis is more closely correlated with fog and inorganic pathogens. Pneumonia and tuberculosis, which have bacterial and viral origins, have weaker connections (Lawther 1959; Schaefer 1907; White and Shuey 1914). We include the latter series only because one might question the ability of nineteenth- century physicians to distinguish among the three diseases. The difference between figures 10.6 and 10.7 is quite striking. The crude death rate is falling, yet the deaths from respiratory diseases are high up to 1880, when they begin to fall. Bronchitis deaths appear to be increasing until the late 1870s and remain high through 1890, when they begin to fall. We estimate the following model: (1)

dkt    wk1  yt2  3 N  ekt,

9. England’s death registration system began in 1836 and included provisions that fined those who failed to report deaths.

Fig. 10.6

Crude death rate in London, 1840–1915

Sources: Annual Reports of the Registrar General (England and Wales), various years; and London Statistics, various years.

Fig. 10.7

Deaths from respiratory diseases in London, 1850–1920

Sources: Annual Reports of the Registrar General (England and Wales), various years; and London Statistics, various years. Note: The category “all respiratory” includes pneumonia, phthisis (tuberculosis), and bronchitis.

300

Karen Clay and Werner Troesken

where dkt is the overall death rate in London in week k (k  1, 2, . . . , 52) and year t (t  1, 2, . . . , 55); wk is a vector of week dummies; yt is a vector of year dummies; N is an overall time trend (N  1, 2, . . . , 2,860); and ekt is a random error term.10 Table 10.2 identifies the thirty- two weeks in the 1855 to 1910 interval that satisfy the conditions necessary to be designated as an event week. The first two columns indicate the year and the week of the year in which the event took place. The final column, labeled documentation, provides citations to the month, day, and page of the Times corroborating the presence of a fog- related event. If the event has already been described in the secondary literature on the London fog, citations to representative secondary sources are provided. The third column, labeled “known,” indicates whether the secondary literature already describes the weeks as involving fog- related events. Of the thirty- two event weeks, twenty- two are part of a sequence of continuous weeks. Events 1 and 2 involve the fourth and fifth weeks of 1855, events 3 and 4 involve the forty- seventh and forty- eighth weeks of 1858, and so on. The longest sequences are five weeks (events 22 to 26) and four weeks (events 29 to 32). The last event occurs in the second week of 1900. After that time, fog- related events cease for the period covered by our sample, which as said in the preceding ends in 1910. The results are robust to changes in our criteria that define event weeks. If, for example, we drop conditions (c.2) and (c.3), the same basic patterns and substantive conclusions emerge. The results are also unchanged if we lower or raise the residual threshold by a modest amount. It is notable how few of the event weeks we identify have been identified by the extant literature. For the period between 1855 and 1872, the reverse event study yields ten previously unknown events. The descriptions of these events found in the Times suggest the procedure is onto something. Describing events 6 and 7, the Times (January 21, 1861, 9) observed: Last Thursday week, when the whole of the metropolis was enveloped in a dense fog, large numbers of person [sic] were stuck down as if shot. Dr. Lotheby, in his report to the city . . . says ‘the quantity of organic vapour, sulphate of ammonia, and finely divided soot in the atmosphere was unprecedented.’ The Times (November 19, 1862, 6) characterized the fogs associated with event 8 this way: “There was a dense fog on Tuesday night, and on Thursday 10. Ideally, it would be desirable to analyze data on cause- specific death rates for respiratory ailments such as bronchitis and pneumonia. Although we believe these data are available, the source in which they are reported is not easily obtained. We hope to acquire this source and analyze cause- specific death rates in subsequent research. Nevertheless, the anecdotal evidence presented elsewhere in the chapter indicates that spikes in mortality were driven by respiratory diseases.

Did Brodie Discover the First Environmental Kuznets Curve? Table 10.2

301

Fog-related events identified by reverse event study, 1855–1910

Year

Week

Event no.

Knowna

1855 1855 1858 1858 1859 1861 1861 1862 1864 1871 1873 1874 1874 1874 1875 1879

4 5 47 48 52 3 4 48 3 50 51 48 49 52 1 51

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

No No No No No No No No No No Yes No No No No Yes

1880

5

17

Yes

1880 1882 1890

6 6 2

18 19 20

Yes No Yes

1890 1892 1892 1892 1892 1892 1893 1894 1899 1899 1900 1900

3 1 2 3 4 5 49 2 51 52 1 2

21 22 23 24 25 26 27 28 29 30 31 32

Yes Yes Yes Yes Yes Yes No No Yes Yes Yes Yes

Documentation (Times unless otherwise indicated) 1/15, 10; 2/1, 5 2/5, 10 11/17, 10 11/26, 10 12/19, 6; 12/20, 10; 12/21, 12; 12/30, 9; 1/8, 6 1/11, 9; 1/16, 10 1/21, 10; 1/23, 12 11/15, 11; 11/17, 10; 11/19, 6; 12/3, 12 1/13, 12; 1/22, 12 11/18, 9; 11/19, 8; 12/20, 6 12/19, 10; Brazell (1968, 111) 11/23, 5 12/2, 10 12/25, 7; 12/28, 5 1/1, 11; 1/8, 9 12/17, 8; Scott (1896); Nature (November 5, 1891, 13) British Medical Journal (February 14, 1880, 254); Nature (November 5, 1891, 13) Same as above 2/3, 6; 2/6, 7; 2/7, 10; 2/13, 10 Nature (November 5, 1891, 14–15); Brodie (1891) Same as above Nature (November 5, 1891, 14–15) Same as above Same as above Same as above Same as above 11/23, 8; 11/30, 8 1/8, 6; 1/15, 6 12/1, 11; 12/2, 5; 12/4, 13; 12/13, 11 12/23, 12; 12/28, 5 1/3, 8; Brazell (1968, 111) Same as above

Sources: See text and final column of table. a “Known” indicates whether the secondary literature already describes the weeks as involving fog-related events.

afternoon fog prevailed of a density that has not been equaled for several years.” Of events 3 and 4, the Times (November 26, 1858, 10) said, “it has been several years since we have seen so dense a fog.” Events 12 through 15 are also missed by the current literature. Occurring one year after the cattle show fog, these events might have been overshadowed by their immediate

302

Karen Clay and Werner Troesken

predecessor, but newspaper accounts suggest an anticyclone and a series of dense and persistent fogs extending over weeks.11 Of the previously unidentified fog of 1882 (event 19), the Times (February 6, 1882, 7) wrote: By general consent the fog which prevailed over a great part of the metropolis during Saturday and Saturday night was one of the densest ever experienced. It was attended with all the usual inconvenience and incidents, intensified to an unprecedented degree. Trains were delayed, fog signals were heard in rapid succession on the railways, street lamps were lighted, street traffic was impeded and gradually suspended, many tramcars ceased to run, and businesses everywhere carried on by artificial light. In the streets, torches and lamps did not much expedite locomotion. Market carts failed to reach Covenant garden until many hours after they were due. Last, it is important to point out that our procedure misses no fog- related event suggested by the existing secondary literature. Figure 10.8 plots a measure of relative deviation, the residual death rate divided by the predicted rate, for event and nonevent weeks. Event week residuals are given by black triangles; nonevent week residuals by small, empty circles. The least severe events increased weekly death rates by 25 to 30 percent. The most severe events could double death rates, increasing them by 75 to 100 percent. To calculate the residuals, we estimate equation (1) using only nonevent weeks. The difference between the predicted and observed death rate equals the residual. The triangles are consonant with Brodie’s data on the incidence of fog. The size of the residuals increase in the years leading up to 1891 and are more frequent before 1891 than after. There is a large and unusual spike in the nonevent week residuals in the eighth, ninth, tenth, and eleventh weeks of 1895. This is the result of an influenza epidemic (Times, March 9, 1895, 5). Note that the nonevent residuals fall below –.100 only twice, when they reach –.111 and –.103. To the extent that we expect symmetry in the structure of the error term, one might plausibly argue that any residual greater than .1 is generated by a different process than that which produces the nonevent residuals. Figure 10.9 provides a look at the absolute effects of fog- related events. Plotting the excess number of deaths associated with event weeks, it shows increasing severity before 1891 and declining severity and frequency thereafter. The excess deaths associated with the fog- related events of the 1850s and 1860s numbered around 500, but by 1891, these deaths neared 2,000. It is significant that fog- related events cease after 1900. Taken together, the results in table 10.2 and in figures 10.8 and 10.9 support Brodie’s conten11. On December 2, 1874, the Times quoted one observer as saying, “the most dense fog I ever saw in this locality.” An editorial in the Times (January 8, 1875) attributed the large number of deaths in the metropolis to cold and variable temperatures, but this observation, combined with the numerous reports of dense fog over a long period, suggest an anticyclone.

Fig. 10.8

Percentage deviation from predicted weekly crude death rate, 1855–1910

Source: See text.

Fig. 10.9

Excess number of deaths during fog-related events

Source: See text.

304

Karen Clay and Werner Troesken

tion that London’s atmosphere was worsening before 1890 and improving thereafter. One concern with the foregoing analysis is what demographers term “harvesting.” In the case of fog- related events, suppose that fogs killed only the frailest and most sickly individuals, people who would have died within a few days or weeks of the fog, whether the fog had ever occurred. If so, the preceding data would overstate the significance of fog- related events. There is anecdotal evidence to support this hypothesis. Witness, for example, the coroner inquests of James Smith, Alice Wright, and William Henry Pepper, discussed previously. To address this concern, we estimate the following variant of equation (1) using the full sample of event and nonevent weeks: (2)

dkt    wk1  yt2  3 N  ψ0 F0  ψ1F1  ψ2 F2  . . .  ψ12 F12  ekt,

where, F0 is a dummy variable that assumes a value of one for all event weeks, and zero otherwise; F1 is a dummy variable that assumes a value of one for the week immediately following an event week, and zero otherwise; F2 is a dummy variable that assumes a value of one for two weeks after an event week, and zero otherwise; and so on down to F12, which assumes a value of one for twelve weeks after an event week, and zero otherwise. All other variables in equation (2) have the same definitions as in equation (1). Figure 10.10, which plots estimated coefficients on F0 through F12, suggests that deaths were not merely being rearranged. The black diamonds indicate a statistically significant coefficient at the 1 percent level; the empty circles indicate insignificant coefficients. The average fog- related event increased the weekly death rate by a statistically significant .16 points. In the first and second weeks after the event, the death rate remained a statistically significant .02 to .03 points above normal. Except for week four, all subsequent weeks are indistinguishably different from zero in terms of statistical significance. As for magnitudes, the point estimates are usually positive and are always very close to zero. Only weeks five, nine, and eleven fall below zero, to –.002, –.003, and –.007, respectively. These patterns are inconsistent with the hypothesis that fog- related events merely rearranged deaths, only causing people to die a few weeks or days earlier than they otherwise would have. Equation (1) control for week and year fixed effects but not for rainfall and temperature. To the extent that fogs were associated with low temperatures, and low temperatures were associated with excess deaths, this estimating procedure imparts an upward bias to the effects of fogs. Although we have not been able to code and analyze the weather data that would allow us to expressly control for such concerns, we have located anecdotal evidence to suggest that, absent dense fogs, extremely low temperatures did not have

Did Brodie Discover the First Environmental Kuznets Curve?

Fig. 10.10 events

305

Deviation from predicted death rate in weeks following fog-related

Source: See text.

the same large effects on mortality.12 Intense cold alone raised the death rate but not by the magnitudes we estimate for fogs and cold (Times, January 16, 1867, 6; December 14, 1897, 11; and February 4, 1879, 3). Also, for variation in temperatures to explain the inverted U- shaped pattern we find in fog- related events reported in figures 10.3 to 10.5, temperatures in England and London would have had to follow that same pattern. We have examined temperature data for the whole of Central England from a variety of sources (e.g., Manley 1974) and find little evidence of this. Finally, in those cases where fog and cold struck London simultaneously, the fog was unique to London but the cold was not; the latter affected surrounding areas as well. If it had been the cold causing the excess deaths, the spike in death rates would have occurred for all areas with cold. But when we consult the Weekly Reports of the Registrar General and various accounts in the Times, both sources indicate that while the cold was a general event (affecting all cities and towns around London), the fogs and the spikes in mortality were not; they occurred only in London. Similarly, in his study of 12. It is our intention to use these data in subsequent research, but compiling the weather data and merging it the mortality data we use here is beyond the scope of this chapter.

306

Karen Clay and Werner Troesken

the relationship between temperatures and the death rate in London, Dines (1894) argued that one should not be quick to attribute increased mortality during the winter quarter to reduced temperatures, but rather to increased crowding and a heavily polluted atmosphere. Ideally, we would like to construct a panel of British cities, some of which were subject to smoke abatement efforts and some of which were not. With such a data set, we could perform a more formal and standard fixed effects estimation of the effects of smoke control on respiratory diseases. While we have identified a data source that would enable us to construct such a panel, we are unable at this stage compile and analyze the data. That is left for a later paper. 10.5

Concluding Remarks

We motivated this chapter with the simple question: did Frederick Brodie discover the world’s first environmental Kuznets curve? The evidence presented in the chapter, though not conclusive, suggests that he did. The strongest single piece of evidence is the reverse event study, which shows that fog- related events—defined as unexplained spikes in weekly mortality rates—rose steadily in frequency and severity in the years leading up to 1891 and declined in frequency and severity thereafter. Furthermore, if one looks at sunshine measures, which in contrast to fog was measured instrumentally, there is evidence that sunshine rates in London were in decline relative to neighboring areas before 1891 and in ascension in the years following. In addition, qualitative evidence in the form of first- hand testimonials are presented throughout the paper to indicate that contemporary observers believed that London’s atmosphere grew cleaner and more breathable in the wake of the Public Health Act. There is also some noisy evidence to suggest that annual bronchitis rates in London rose and fell with the incidence of fogs. Although we do not yet have the capacity to perform a full- blown analysis of panel data, evidence from other cities, particularly Glasgow, suggests that what was happening London was not unique: bronchitis rates grew increasingly frequent and severe with the rise of coal and subsided only with reductions in coal smoke. There are at least three plausible mechanisms through which Londoners might have successfully curtailed their production of coal smoke, or at least dissipated the smoke. First, the preceding data indicate the population of Greater London redistributed itself, as central districts became less densely populated and outlying districts more so. If smoky fogs formed only when smoke density rose above a certain threshold, redistributing population might have reduced the number of fogs experienced by the metropolis as a whole. Second, with passage of the Public Health Act in 1891, businesses were fined for failing to consumer their own smoke, or otherwise generating excess amounts of smoke. In response, manufacturers switched from

Did Brodie Discover the First Environmental Kuznets Curve?

307

bituminous coal from Newcastle and elsewhere in England to Welsh steam coal, a harder coal that generated much less smoke, though it was about 20 percent more expensive than ordinary soft coal. Businesses also adopted more efficient stoking and firing procedures so that they economized on coal and minimized emissions of unburnt soot and tar. Third, the expansion of gas for cooking and heating helped undermine demand for coal among homeowners, who otherwise burned much of the coal in London. If one were to plot real wages for unskilled laborers in London against foggy days per annum the typical EKC would emerge. At low levels of development, pollution (as proxied by fog) rose with real wages, but at some real- wage threshold, the correlation reversed itself and rising wages were associated with reductions in foggy days per year. It is difficult to say what the EKC means in this context. Perhaps it reflects an income effect, or perhaps the correlation between income and pollution is spurious, the result of a threshold effect for pollution. Ordinary citizens and voters tolerate pollution as long as it is below some level but begin actively lobbying for improvement once it crosses some threshold. The difficulty with the latter interpretation is that smoke abatements efforts in London had a long history, going back hundreds of years—though it was eventually repealed, the first statute prohibiting the burning of bituminous of coal in London was passed in 1273 (Martin 1906). It seems more likely that the metropolis had to reach some level income and technological advancement before it had the capacity to effectively deal with the coal smoke problem. This, however, is mere speculation. Fully resolving this question, and the other issues raised in the preceding, we leave for future research.

References Ackerman, A. S. E. 1906. The distribution of producer gas as a means of alleviating the smoke nuisance. Journal of the Royal Society for the Promotion of Health 27:80–84. Booth, W. H. 1898. Smoke Prevention. Journal of the American Society of Naval Engineers 10:1064–68. Brazell, J. H. 1968. London weather. London: Her Majesty’s Stationery Office. Brimblecombe, P. T. 1987. The big smoke: A history of air pollution in London since medieval times. Cambridge, UK: Cambridge University Press. Brodie, F. J. 1891. Some remarkable features in the winter of 1890–91. Quarterly Journal of the Royal Meteorological Society 17:155–67. ———. 1905. Decrease of fog in London during recent years. Quarterly Journal of the Royal Meteorological Society 31:15–28. Caborne, W. F. 1906. Stoking and smoke abatement. Journal of the Royal Sanitary Institute 27:142–48. Carpenter, A. 1880. Town smoke and town fog. Westminster Review 61:136–57. Chandler, T. J. 1965. The climate of London. London: Hutchinson of London.

308

Karen Clay and Werner Troesken

Dines, W. H. 1894. On the relation between the mean quarterly temperature and the death rate. Quarterly Journal of the Royal Meteorological Society 20:173–79. Frankland, E. 1882. Climate in town and country. Nature, August 17. Great Britain Board of Trade. 1903. Report on wholesale and retail prices in the United Kingdom in 1902, with comparative statistical tables for a series of years. Ordered by the House of Commons. London: Her Majesty’s Stationery Office. Hann, J. 1903. Handbook of climatology. Part I. General climatology. New York: MacMillan. Lawther, P. J. 1959. Chronic bronchitis and air pollution. Journal of the Royal Society for the Promotion of Health 1:4–9. Lempfert, R. G. K. 1912. Sunshine records: A comparison of sunshine statistics for urban and rural stations. In Papers read at the smoke abatement conferences, 23–28. Westminster, UK: Coal Smoke Abatement Society. Livesey, G. 1906. Domestic smoke abatement. Journal of the Royal Society for the Promotion of Health 27:57–63. Manley, G. 1974. Central England temperatures: Monthly means 1659 to 1973. Quarterly Journal of the Royal Meteorological Society 100:389–405. Martin, A. J. 1906. Smoke prevention and coal conservation. Journal of the Royal Society for the Promotion of Health 27:85–107. Mitchell, B. R. 1988. British historical statistics. London: Cambridge University Press. Mossman, R. C. 1897. The non- instrumental meteorology of London, 1713–1896. Quarterly Journal of the Royal Meteorological Society 23:287–94. Owens, J. S. 1912. Wasteful power production: With special reference with waste due to smoke. In Papers read at the smoke abatement conferences, 89–94. Westminster, UK: Coal Smoke Abatement Society. Poore, G. V. 1893. Light, air, and fog. Transactions of the Sanitary Institute 14: 13–41. Report from the Select Committee on Smoke Prevention. 1843. Great Britain. Parliament. House of Commons. Report of the Smoke Abatement Committee. 1883. London. Reynolds, M. 1882. Stationary engine driving: A practical manual. London: Crosby Lockwood. Rideal, S. 1906. The abatement of smoke in factories. Journal of the Royal Sanitary Institute 27:149–51. Russell, Rollo. 1906. The artificial production of persistent fog. Journal of the Royal Sanitary Institute 27:152–59. ———. 1912. Smoke and fog. In Papers read at the smoke abatement conferences, 21–22. Westminster, UK: Coal Smoke Abatement Society. Russell, W. J. 1891. Town fogs and their effects. Nature, November 5. Saward, F. E. 1899, 1908, and 1914. The coal trade: A compendium of valuable information relative to coal production, prices, transportation, etc. New York: The Coal Trade Journal. Schaefer, T. W. 1907. The contamination of our cities with sulfur dioxide, the cause of respiratory disease. Boston Medical and Surgical Journal 157:106–10. Schlicht, P. 1907. The production of coke and its application in domestic fires. Journal of the Society of Arts 684–98. Scott, R. H. 1896. Notes on some of the differences between fogs, as related to the weather systems which accompany them. Quarterly Journal of the Royal Meteorological Society 22:41–65. White, W. C., and P. Shuey. 1914. The influence of smoke on acute and chronic lung infections. In Papers on the influence of smoke on health, ed. O. Klotz and W. C.

Did Brodie Discover the First Environmental Kuznets Curve?

309

White, 164–73. Pittsburgh: Mellon Institute of Industrial Research and School of Specific Industries. Wise, W. 2001. Killer smog: The world’s worst air pollution disaster. Lincoln, NE: iUniverse.com publication. Wright, C. D. 1905. Coal mine labor in Europe. Washington, DC: GPO.

11 Impacts of Climate Change on Residential Electricity Consumption Evidence from Billing Data Anin Aroonruengsawat and Maximilian Auffhammer

11.1

Introduction

Forecasts of electricity demand are of central importance to policymakers and utilities for purposes of adequately planning future investments in new generating capacity. Total electricity consumption in California has more than quadrupled since 1960, and the share of residential consumption has grown from 26 percent to 34 percent (Energy Information Administration [EIA] 2008). Today, California’s residential sector alone consumes as much electricity as Argentina, Finland, or roughly half of Mexico. The majority of electricity in California is delivered by three investor- owned utilities and over a hundred municipal utilities. On a per capita basis, California’s residential consumption has stayed almost constant since the early 1970s, while most other states have experienced rapid growth in per capita consumption. The slowdown in growth of California’s per capita consumption coincides with the imposition of aggressive energy efficiency and conservation programs during the early 1970s. The average annual growth rate in per capita consumption during 1960 to Anin Aroonruengsawat is a lecturer in economics at Thammasat University. Maximilian Auffhammer is an associate professor of agricultural and resource economics at the University of California, Berkeley, and a faculty research fellow of the National Bureau of Economic Research. We would like to thank Guido Franco and the Public Interest Energy Research (PIER) Program at the California Energy Commission for generous funding of this work. We thank the University of California Energy Institute (UCEI) and the investor- owned utilities of California for letting us gain access to the billing data at the UCEI data center. We also thank Severin Borenstein and Koichiro Ito for helping us understand the electricity data. We gratefully acknowledge constructive comments by Gary Libecap and Rick Steckel, Olivier Deschênes, one anonymous referee, seminar participants at the Energy and Resources Group and the UCEI. All findings and remaining errors in this study are those of the authors.

311

312

Anin Aroonruengsawat and Maximilian Auffhammer

1973 was approximately 7 percent and slowed to a remarkable 0.29 percent during 1974 to 1995. Growth rates during the last decade of available data have increased to a higher rate of 0.63 percent, and this difference in growth rates is statistically significant. California’s energy system faces several challenges in attempting to meet future demand (California Energy Commission [CEC] 2005). In addition to rapid population growth, economic growth and an uncertain regulatory environment, the threat of significant global climate change has recently emerged as a factor influencing the long- term planning of electricity supply. The electric power sector will be affected by climate change through higher cooling demand, lower heating demand, and potentially stringent regulations designed to curb emissions from the sector. This chapter simulates how the residential sector’s electricity consumption will be affected by different scenarios of climate change. We make three specific contributions to the literature on simulating the impacts of climate change on residential electricity consumption. First, through an unprecedented opportunity to access the complete billing data of California’s three major investor- owned utilities, we are able to provide empirical estimates of the temperature responsiveness of electricity consumption based on microdata. Second, we allow for a geographically specific response of electricity consumption to changes in weather. Finally, we explore socioeconomic and physical characteristics of the population, which help explain some of the variation in temperature response. The chapter is organized as follows: section 11.2 reviews the literature assessing the impacts of climate change on electricity consumption. Section 11.3 describes the sources of the data used in this study. Section 11.4 contains the econometric model and estimation results. We simulate the impacts of climate change on residential electricity consumption in section 11.5. Section 11.6 explores the heterogeneity in temperature response, and section 11.7 concludes. 11.2

Literature Review

The historical focus of the literature forecasting electricity demand has been on the role of changing technology, prices, income, and population growth (e.g., Fisher and Kaysen 1962). Early studies in demand estimation have acknowledged the importance of weather in electricity demand and explicitly controlled for it to prevent biased coefficient estimates as well as wanting to gain estimation efficiency (e.g., Houthakker and Taylor 1970). Simulations based on econometrically estimated demand functions had, therefore, focused on different price, income, and population scenarios, while assuming a stationary climate system. The onset of anthropogenic climate change has added a new and important dimension of uncertainty over future demand, which has spawned a small academic literature on climate change impacts estimation, which can be divided into two approaches.

Impacts of Climate Change on Residential Electricity Consumption

313

In the engineering literature, large- scale bottom- up simulation models are utilized to simulate future electricity demand under varying climate scenarios. The advantage of the simulation model approach is that it allows one to simulate the effects of climate change given a wide variety of technological and policy responses. The drawback to these models is that they contain a large number of response coefficients and make a number of specific and often untestable assumptions about the evolution of the capital stock and its usage. The earliest impacts papers adopt this simulation approach and suggest that global warming will significantly increase energy consumption. Cline (1992) provides the earliest study on the impacts of climate change in his seminal book The Economics of Climate Change. The section dealing with the impact on space cooling and heating relies on an earlier report by the U.S. Environmental Protection Agency (1989). That study of the potential impact of climate change on the United States uses a utility planning model developed by Linder, Gibbs, and Inglis (1987) to simulate the impact on electric utilities in the United States and finds that increases in annual temperatures ranging from 1.0°C to 1.4°C (1.8°F to 2.5°F) in 2010 would result in demand of 9 percent to 19 percent above estimated new capacity requirements (peak load and base load) in the absence of climate change. The estimated impacts rise to 14 percent and 23 percent for the year 2055 and an estimated 3.7°C (6.7°F) temperature increase. Baxter and Calandri (1992) provide another early study in this literature and focus on California’s electricity use. In their study, they utilize a partial equilibrium model of the residential, commercial, agriculture, and water pumping sectors to examine total consumption as well as peak demand. They project electricity demand for these sectors to the year 2010 under two global warming scenarios: a rise in average annual temperature of 0.6°C (1.1°F—low scenario) and of 1.9°C (3.4°F—high scenario). They find that electricity use increases from the constant climate scenario by 0.6 percent to 2.6 percent, while peak demand increases from the baseline scenario by 1.8 percent to 3.7 percent. Rosenthal, Gruenspecht, and Moran (1995) focus on the impact of global warming on energy expenditures for space heating and cooling in residential and commercial buildings. They estimate that a 1°C (1.8°F) increase in temperature will reduce U.S. energy expenditures in 2010 by $5.5 billion (1991 dollars). The economics literature has favored the econometric approach to impacts estimation, which is the approach we adopt in the current study. While there is a large literature on econometric estimation of electricity demand, the literature on climate change impacts estimation is small and relies on panel estimation of heavily aggregated data or cross- sectional analysis of more microlevel data. The first set of papers attempts to explain variation in a cross section of energy expenditures based on survey data to estimate the impact of climate change on fuel consumption choices. Mansur, Mendelsohn, and Morrison (2008) and Mendelsohn (2003) endogenize fuel choice, which is usually assumed to be exogenous. They find that

314

Anin Aroonruengsawat and Maximilian Auffhammer

warming will result in fuel switching toward electricity. The drawback of the cross- sectional approach is that one cannot econometrically control for unobservable differences across firms and households, which may be correlated with weather or climate. If that is the case, the coefficients on the weather variables and corresponding impacts estimates may be biased. Instead of looking at a cross section of firms or households, Franco and Sanstad (2008) explain pure time series variation in hourly electricity load at the grid level over the course of a year. They use data reported by the California Independent System Operator (CalISO) for 2004 and regress it on a population- weighted average of daily temperature. The estimates show a nonlinear impact of average temperature on electricity load and a linear impact of maximum temperature on peak demand. They link the econometric model to climate model output from three different global circulation models (GCMs) forced using three quasi- official scenarios based on the Intergovernmental Panel for Climate Change (IPCC) Special Report on Emissions Scenarios (SRES) to simulate the increase in annual electricity and peak load from 2005 to 2099. Relative to the 1961 to 1990 base period, the range of increases in electricity and peak load demands are 0.9 percent to 20.3 percent and 1.0 to 19.3 percent, respectively. Crowley and Joutz (2003) use a similar approach where they estimate the impact of temperature on electricity load using hourly data in the Pennsylvania, New Jersey, and Maryland interconnection. Some key differences, however, are that they control for time- fixed effects and define the temperature variable in terms of heating and cooling degree days. They find that a 2°C (3.6°F) increase in temperature results in an increase in energy consumption of 3.8 percent of actual consumption, which is similar to the impact estimated by Baxter and Calandri (1992). Deschênes and Greenstone (2007) provide the first panel data- based approach to estimating the impacts of climate change on residential total energy consumption, which includes electricity, natural gas, and oil as the main nonrenewable sources of energy. They explain variation in U.S. statelevel annual panel data of residential energy consumption using flexible functional forms of daily mean temperatures. The identification strategy behind their paper, which is one we will adopt here as well, relies on random fluctuations in weather to identify climate effects on electricity consumption. The model includes state fixed effects, census division by year fixed effects, and controls for precipitation, population, and income. The temperature data enter the model as the number of days in twenty predetermined temperature intervals. The authors find a U- shaped response function where electricity consumption is higher on very cold and hot days. The impact of climate change on annual electricity consumption by 2099 is in the range of 15 percent to 30 percent of the baseline estimation or 15 to 35 billion (2006 US$). The panel data approach allows one to control for differences in unob-

Impacts of Climate Change on Residential Electricity Consumption

315

servables across the units of observation, resulting in consistent estimates of the coefficients on temperature. The current chapter is the first study using a panel of household level electricity billing data to examine the impact of climate change on residential electricity consumption. Through a unique agreement with California’s three largest investor- owned utilities, we gained access to their complete billing data for the years 2003 to 2006. We identify the effect of temperature on electricity consumption using within- household variation in temperature, which is made possible through variation in the start dates and lengths of billing periods across households. Because our data set is a panel, we can control for household fixed effects, month fixed effects, and year fixed effects. The drawback of this data set is that the only other reliable information we have about each individual household is price and its five- digit zip code location. 11.3 11.3.1

Data Residential Billing Data

The University of California Energy Institute (UCEI) jointly with California’s investor- owned utilities established a confidential data center, which contains the complete billing history for all households serviced by Pacific Gas and Electric (PG&E), Southern California Edison, and San Diego Gas and Electric (SDG&E) for the years 2003 to 2006. These three utilities provide electricity to roughly 80 percent of California households. The data set contains the complete information for each residential customer’s bills over the four- year period. Specifically, we observe an ID for the physical location, a service account number, bill start date, bill end date, total electricity consumption (in kilowatt- hours [kWh]) and the total amount of the bill (in $) for each billing cycle as well as the five- digit zip code of the premises.1 Only customers who were individually metered are included in the data set. For the purpose of this chapter, we define a customer as a unique combination of premise and service account number. It is important to note that each billing cycle does not follow the calendar month, and the length of the billing cycle varies across households with the vast majority of households being billed on a twenty- five to thirty- five- day cycle. While we have data covering additional years for two of the utilities, we limit the study to the years 2003 to 2006 to obtain equal coverage. Hereafter, we will refer to this data set as “billing data.” Figure 11.1 displays the zip codes we have data for, which is the majority of the state. Due to the difference in climate conditions across the state, California is 1. The premise identification number does not change with the occupant of the residence. The service account number, however, changes with the occupant of the residence.

316

Anin Aroonruengsawat and Maximilian Auffhammer

Fig. 11.1 Observed residential electricity consumption 2003–2006 and National Oceanic and Atmospheric Administration cooperative weather stations Note: The map displays five- digit zip codes with available geographic boundaries.

divided into sixteen building climate zones, each of which require different minimum efficiency building standards specified in an energy code. The climate zones are depicted in figure 11.2.2 We expect this difference in building standards to lead to a different impact of temperature change on electricity consumption across climate zones. We will, therefore, estimate the impact 2. The California climate zones shown are not the same as what one would commonly call an area like desert or alpine climate. The climate zones are based on energy use, temperature,

Impacts of Climate Change on Residential Electricity Consumption

Fig. 11.2

317

California Energy Commission building climate zones

Source: California Energy Commission.

of mean daily temperature on electricity consumption separately for each climate zone. We later empirically explore the sources of this variation in section 11.6. We assign each household to a climate zone via their five- digit zip code through a mapping, which we obtained from the California Energy Commission. weather, and other factors. They are essentially a California Energy Commission (CEC)- defined geographic area that has similar climatic characteristics. Each climate zone has a representative city. These are for each of the climate zones: (1) Arcata, (2) Santa Rosa, (3) Oakland, (4) Sunnyvale, (5) Santa Maria, (6) Los Angeles, (7) San Diego, (8) El Toro, (9) Pasadena, (10) Riverside, (11) Red Bluff, (12) Sacramento, (13) Fresno, (14) China Lake, (15) El Centro, and (16) Mount Shasta.

318

Anin Aroonruengsawat and Maximilian Auffhammer

The billing data set contains 300 million observations, which exceeds our ability to conduct estimation using standard statistical software. We, therefore, resort to sampling from the population of residential households to conduct econometric estimation. We designed the following sampling strategy. First, we only sample from households with regular billing cycles, namely twenty- five to thirty- five days in each billing cycle which have at least thirty- five bills over the period of 2003 to 2006.3 We also removed bills with an average daily consumption less than 2 kWh or more than 80 kWh. The reason for this is our concern that these outliers are not residential homes but rather vacation homes and small- scale “home based manufacturing and agricultural facilities.” Combined with the fact that our data does not contain single- metered multifamily homes, our sampling strategy is likely to result in a slight under representation of multifamily and smaller single- family homes. These are more likely to be rental properties than larger single- family units. Our results should be interpreted keeping this in mind.4 From the population subject to the preceding restrictions, we take a random sample from each zip code, making sure that the relative sample sizes reflect the relative sizes of the population by zip code. We draw the largest possible representative sample from this population given our computational constraints. For each climate zone, we test whether the mean daily consumption across bills for our sample is different from the population mean and fail to reject the null of equality, suggesting that our sampling is indeed random, subject to the sample restrictions discussed above. We proceed with estimation of our models by climate zone, which makes concerns about sampling weights mute. Figure 11.3 displays the spatial distribution of 2006 consumption shares across zip codes. Finally, California has a popular program for low- income families— California Alternate Rates for Energy (CARE)—where program- eligible customers receive a 20 percent discount on electric and natural gas bills. Eligibility requires that total household income is at or below 200 percent of federal poverty level. For the first set of models, we exclude these households from our sample. We then explore the robustness of our simulations by including these households in a separate simulation. The concern here is that omitting these smaller homes with lower HVAC saturation rates may lead to an overestimation of impacts. No single zip code is responsible for more than 0.5 percent of total consumption. Table 11.1 displays the summary statistics of our consumption 3. With the regular billing cycle, there should be forty- eight bills for the households in our sample during the period 2003 to 2006. 4. After removing outlier bills, we compared the population average daily consumption of bills with billing cycles ranging from twenty- five to thirty- five days to the average daily consumption of bills for any length. The average daily consumption by climate zone in the subset of bills we sample from is roughly 1/10th of a standard deviation higher than the mean daily consumption of the complete population including bills of any length.

Impacts of Climate Change on Residential Electricity Consumption

Fig. 11.3 zip code

319

Share of total residential electricity consumption for 2006 by five-digit

sample by climate zone. There is great variability in average usage across climate zones, with the central coast’s (zone 3) average consumption per bill at roughly 60 percent that of the interior southern zone 15. The average electricity price is almost identical across zones, at thirteen cents per kWh. 11.3.2

Weather Data

To generate daily weather observation to be matched with the household electricity consumption data, we use the Cooperative Station Dataset published by National Oceanic and Atmospheric Administration’s (NOAA) National Climate Data Center (NCDC). The data set contains daily observations from more than 20,000 cooperative weather stations in the United States, the U.S. Caribbean Islands, the U.S. Pacific Islands, and Puerto Rico. Data coverage varies by station. Because our electricity data cover the state of California for the years 2003 to 2006, the data set contains 370 weather

1,459,578 2,999,408 3,200,851 4,232,465 2,621,344 2,970,138 3,886,347 2,324,653 3,067,787 3,202,615 4,106,432 3,123,404 3,827,483 4,028,225 2,456,562 3,401,519

No. of observations

31,879 65,539 69,875 92,294 57,123 64,145 85,169 50,373 66,231 70,088 90,245 68,342 84,493 88,086 54,895 74,644

No. of households

Summary statistics

550 612 469 605 504 529 501 583 632 700 795 721 780 714 746 589

Mean 354 385 307 362 317 334 327 364 389 416 455 420 464 413 532 409

S.D.

Usage per bill per billing cycle (Kwh)

0.13 0.13 0.13 0.13 0.13 0.13 0.15 0.14 0.13 0.14 0.13 0.13 0.13 0.13 0.13 0.13

Mean 0.03 0.03 0.02 0.03 0.03 0.03 0.04 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.03 0.02

S.D.

Average price per billing cycle ($/Kwh)

34.5 36.0 42.0 40.5 42.0 48.5 47.0 49.5 48.0 35.5 28.5 38.5 36.6 32.0 34.5 22.5

1 37.5 39.0 44.3 42.8 44.3 50.4 48.9 51.5 50.3 39.0 32.8 40.8 39.3 35.0 37.8 26.5

5

54.7 55.5 57.0 57.8 58.8 62.0 61.5 63.3 63.0 61.0 54.8 58.5 59.0 57.5 63.8 52.3

50

77.0 77.5 75.0 81.4 76.0 78.0 77.5 80.6 81.0 81.8 84.3 84.0 87.8 91.3 97.0 83.0

95

80.0 80.5 78.0 85.5 78.5 81.0 80.0 83.3 83.5 84.5 87.0 87.0 90.0 95.0 99.5 86.5

99

Percentiles daily mean temperature distribution in sample (degree Fahrenheit)

Notes: The table displays summary statistics for residential electricity consumption for the non-CARE sample used in the estimation. S.D.  standard deviation.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

Zone

Table 11.1

Impacts of Climate Change on Residential Electricity Consumption

321

stations reporting daily data. In the data set, we observe daily minimum and maximum temperature as well as total daily precipitation and snowfall. Because the closest meaningful geographic identifier of our households is the five- digit postal zip code, we select stations as follows. First, we exclude any stations not reporting data in all years. Further, we exclude stations reporting fewer than 300 observations in any single year and stations at elevations more than 7,000 feet above sea level, which leaves us with 269 “valid” weather stations.5 Figure 11.1 displays the distribution of these weather stations across the state. While there is good geographic coverage of weather stations for our sample, we do not have a unique weather station reporting data for each zip code. To assign a daily value for temperature and rainfall, we need to assign a weather station to each zip code. We calculate the Vincenty distance of a zip code’s centroid to all valid weather stations and assign the closest weather station to that zip code. As a consequence of this procedure, each weather station on average provides data for approximately ten zip codes. Because we do not observe daily electricity consumption by household, but rather monthly bills for billing periods of differing length, we require a complete set of daily weather observations. The NCDC data have a number of missing values, which we fill in using the following algorithm. First, we calculate the Vincenty distance of each zip code’s geographic centroid to all qualifying weather stations. We then identify the ten closest weather stations to each centroid, provided that each is less than fifty miles from the monitor. Of these stations, we identify the “primary station” as the closest station reporting data for at least 200 days a year. We fill in missing values by first regressing, for observations in which the primary weather station was active, the relevant climate weather variable for the primary station onto the same variable for the remaining nine closest stations. We use the predicted values from that regression to replace missing values. Following this step, primary station observations are still missing whenever one of the remaining nine closest stations is also missing an observation. To estimate the remaining missing values, we repeat the preceding step with the eight closest stations, then the seven closest, and so on. To check the performance of our algorithm, we conduct the following experiment. First, we select the set of data points for which the primary weather station has an observation. We then randomly set 10 percent of the temperature data for this station to missing. After applying the algorithm described in the preceding to this sample, we compare the predicted temperature data to the observations we had set aside. Even for observations in which a single additional weather station was used to predict a missing temperature, the correlation coefficient between actual 5. The cutoff of 300 valid days is admittedly arbitrary. If we limit the set of weather stations to the ones providing a complete record, we would lose roughly half of all stations. We conducted robustness checks using different cutoff numbers, and the results are robust.

322

Anin Aroonruengsawat and Maximilian Auffhammer

and predicted temperatures exceeds 0.95. Plotting the actual and predicted series against each other provides an almost perfect fit. We, therefore, feel confident that our algorithm provides us with a close representation of the true data generating process for missing weather observations. We end up with a complete set of time series for minimum temperature, maximum temperature, and precipitation for the 269 weather stations in our sample. For the remainder of our empirical analysis, we use these patched series as our observations of weather.6 There is an important caveat to using daily weather data when studying households’ response to climate change. By using daily weather shocks, we implicitly estimate individuals’ response to changed daily temperatures. While climate change will affect daily temperatures on average, it is a more long- run process and should be thought of as the long- run moving average of weather. The estimated impacts for this reason may, on the one hand, be too high if individuals have lower cost options in the long run and relocate to cooler climates. The estimated impacts based on daily weather, on the other hand, may be too low if individuals adapt in the sense that areas that do not currently cool using electricity start seeing a high degree of air conditioner penetration. The overall sign of the bias is not clear. Unfortunately, it is not clear whether the perfect counterfactual to study this problem exists. One would require randomly assigned climate (not weather) to study this issue. This randomization would affect technology adoption. Electricity demand, in turn, is determined at the daily level by fluctuations in weather around a long- run trend. The second caveat is that it would be preferable to have a weather index, which counts all relevant dimensions of weather, such as minimum and maximum temperature, humidity, solar radiation, and wind speed and direction. Unfortunately, these indicators are not available for the vast majority of stations at the daily level. One could, however, estimate a response function using such an index for locations that have sufficient data. We leave this for future research. 11.3.3

Other Data

In addition to the quantity consumed and average bill amount, all we know about the households is the five- digit zip code in which they are located. We purchased sociodemographics at the zip- code level from a firm aggregating this information from census estimates (zip- codes.com). We only observe these data for a single year (2006). The variables we will make use of are total population and average household income. The final sample used for estimation comprises households in zip codes that make up 81 percent of California’s population. Table 11.2 displays summary statistics for all zip 6. We also tried an inverse distance weighting algorithm for filling in missing data, and the results are almost identical.

239 239 239 239 239 239 239

n 19.83 2.66 39.52 200.08 36.92 1,081.45 69.66

Mean (not in sample) 20.86 0.60 19.39 177.33 7.34 1,526.95 130.12

S.D.

Summary statistics for zip codes in and out of sample

Note: S.D.  standard deviation. ∗∗∗Significant at the 1 percent level.

Population Household size Household income House value Median age Elevation Land area

Variable

Table 11.2

1,325 1,325 1,325 1,325 1,325 1,325 1,325

n 20.39 2.79 48.32 234.90 36.85 439.63 68.05

Mean (in sample)

20.67 0.60 21.53 177.51 7.50 737.94 140.45

S.D.

0.56 0.14∗∗∗ 8.80∗∗∗ 34.83∗∗∗ –0.07 –642∗∗∗ –1.61

Difference

324

Anin Aroonruengsawat and Maximilian Auffhammer

codes in California with registered residential population, broken down by whether we observe households in a given zip. We observe households for 1,325 zip codes and do not observe households for 239 zip codes. The 239 zip codes are not served by the three utilities, which provided us with access to their billing data. Table 11.2 shows that the zip codes in our sample are more populated, have larger households, are wealthier, and are at lower elevations. There seems to be no statistically significant difference in population, median age, or land area. Taking these differences into consideration is important when judging the external validity of our estimation and simulation results. Finally, we will explore which observable characteristics of households are consistent with differences in the temperature repose function. We use the year 2000 long form census data for the state of California to calculate indicators of observable characteristics of the average household or structure in that zip code. We obtain measures of the share of households using gas or electricity as heating fuel, year the average structure was built, the percent of urban households, and the percent of rental properties. 11.4

Econometric Estimation

As discussed in the previous section, we observed each household’s monthly electricity bill for the period 2003 to 2006. Equation (1) shows our main estimating equation, which is a simple log- linear specification commonly employed in aggregate electricity demand and climate change impacts estimation (e.g., Deschênes and Greenstone 2007). k

(1)

log(qit)  ∑ pDpit  Zit  i  m  y  εit p =1

log(qit) is the natural logarithm of household i’s electricity consumed in kWhs during billing period t. For estimation purposes, our unit of observation is a unique combination of premise and service account number, which is associated with an individual and structure. We thereby avoid the issue of having individuals moving to different structures with more or less efficient capital or residents with different preferences over electricity consumption moving in and out of a given structure. California’s housing stock varies greatly across climate zones in its energy efficiency and installed energy consuming capital. We estimate equation (1) separately for each of the sixteen climate zones discussed in the data section, which are also displayed in figure 11.2. The motivation for doing so is that we would expect the relationship between consumption and temperature to vary across these zones as there is a stronger tendency to heat in the more northern and higher altitude zones and a stronger tendency to cool, but little heating taking place, in the hotter interior zones of California. The main variables of interest in this chapter are those measuring temperature. The last five columns of table 11.1 display the median, 1st, 5th,

Impacts of Climate Change on Residential Electricity Consumption

325

90th, and 95th percentile of the mean daily temperature distribution by climate zone. The table shows the tremendous differences in this distribution across climate zones. The southeastern areas of the state, for example, are significantly hotter on average yet also have greater variances. Following recent trends in the literature, we include our temperature variables in a way that imposes a minimal number of functional form restrictions in order to capture potentially important nonlinearities of the outcome of interest in weather (e.g., Schlenker and Roberts 2006). We achieve this by sorting each day’s mean temperature experienced by household i into one of k temperature bins.7 In order to define a set of temperature bins, there are two options found in the literature. The first is to sort each day into a bin defined by specific equidistant (e.g., 5°F) cutoffs. The second approach is to split each of the sixteen zones’ temperature distributions into a set of percentiles and use those as the bins used for sorting. The latter strategy allows for more precisely estimated coefficients because there is guaranteed coverage in each bin. The equidistant bins strategy runs the risk of having very few observations in some bins, and, therefore leading to unstable coefficient estimation, especially at the extremes. There is no clear guidance in the literature on which approach provides better estimates, and we, therefore, conduct our simulations using both approaches. For the percentile strategy, we split the temperature distribution into deciles yet break down the upper and bottom decile further to include buckets for the 1st, 5th, 95th, and 99th percentile to account for extreme cold or heat days. We, therefore, have a set of fourteen buckets for each of the sixteen climate zones. The thresholds for each vary by climate zone. For the equidistant bins approach, we split the mean daily temperature for each household into a set of 5° bins. In order to avoid the problem of imprecise estimation at the tails due to insufficient data coverage, we require that each bin have at least 1 percent of the data values in it for the highest and lowest bin. The highest and lowest bins in each zone therefore contain a few values that exceed the 5° threshold. For each household, bin definition and billing period we then counted the number of days the mean daily temperature falls into each bin and recorded this as Dpit. The main coefficients of interest to the later simulation exercise are the ps, which measure the impact of one more day with a mean temperature falling into bin p on the log of household electricity consumption. For small values, ps interpretation is approximately the percent change in household electricity consumption due to experiencing one additional day in that temperature bin. Zit is a vector of observable confounding variables which vary across billing periods and households. The first of two major confounders we observe at the 7. We use mean daily temperature as our temperature measure. This allows a flexible functional form in a single variable. An alternate strategy we will explore in future work is separating the temperature variables into minimum and maximum temperature, which are highly correlated with our mean temperature measure.

326

Anin Aroonruengsawat and Maximilian Auffhammer

household level are the average electricity price for each household for a given billing period. California utilities price residential electricity on a block rate structure. The average price experienced by each household in a given period is, therefore, not exogenous because marginal price depends on consumption (qit). Identifying the price elasticity of demand in this setting is problematic, and a variety of approaches have been proposed (e.g., Hanemann 1984; Reiss and White 2005). The maximum likelihood approaches are computationally intensive and given our sample size cannot be feasibly implemented here. More important, however, we do not observe other important characteristics of households (e.g., income) that would allow us to provide credible estimates of these elasticities. For later simulation, we will rely on the income specific price elasticities provided by Reiss and White (2005), who used a smaller sample of more detailed data based on the national level Residential Energy Consumption (REC) survey. We have run our models by including price directly, instrumenting for it using lagged prices, and omitting it from estimation. The estimation results are almost identical for all three approaches, which is reassuring. While one could tell a story that higher temperatures lead to higher consumption and, therefore, higher marginal prices for some households, this bias seems to be negligible given our estimation results. In the estimation and simulation results presented in this chapter, we omit the average price from our main regression.8 The second major time varying confounder is precipitation in the form of rainfall. We calculate the amount of total rainfall for each of the 269 weather stations, filling in missing values using the same algorithm discussed in the previous section. We control for rainfall using a second- order polynomial in all regressions. The i are household fixed effects, which control for time invariant unobservables for each household. The ϕm are month- specific fixed effects, which control for unobservable shocks to electricity consumption common to all households. The y are year fixed effects, which control for yearly shocks common to all households. To credibly identify the effects of temperature on the log of electricity consumption, we require that the residuals conditional on all right- hand side variables be orthogonal to the temperature variables, which can be expressed as E(εitDpit | D–pit, Zit, i, ϕm, y)  0. Because we control for household fixed effects, identification comes from within- household variation in daily temperature after controlling for shocks common to all households, rainfall, and average prices. We estimate equation (1) for each climate zone using a least squares fitting criterion and a clustered variance covariance matrix clustered at the zip code.9 Figure 11.4 plots the estimated temperature response coefficients 8. The full set of estimation results are available upon request from the authors. 9. Clustering along the time dimension would be desirable but due to the temporal nesting structure of the billing dates not possible to our knowledge. We also used the White sandwich variance covariance matrix, which yielded smaller standard errors than the ones obtained from clustering by zip.

Fig. 11.4 Estimated climate response functions for California Energy Commission climate zones 1–16 Notes: The panels display the estimated temperature slope coefficients for each of the fourteen percentile bins (solid) and the equidistant bins (dashed) against the midpoint of each bin. The plots were normalized using the coefficient estimate for the 60 to 65 temperature bin. The title of each panel displays the name of a representative city for that climate zone.

328

Anin Aroonruengsawat and Maximilian Auffhammer

for each of the climate zones against the midpoints of the bins for the percentile and equidistant bin approaches. The coefficient estimates are almost identical, which is reassuring. We do not display the confidence intervals around the estimated coefficients. The coefficients are so tightly estimated that for visual appearance, displaying the confidence intervals simply makes the lines appear thick. From this figure, several things stand out. First, there is tremendous heterogeneity in the shape of the temperature response of electricity consumption across climate zones. Many zones have almost flat temperature response functions, such as southern coastal zones (5, 6, and 7). Other zones display a very slight negative slope at lower temperatures, especially the northern areas of the state (1, 2, and 11), indicating a decreased consumption for space heating as temperatures increase. California’s households mostly use natural gas for space heating, which explains why for most areas we do not see a steeper negative slope at lower temperatures. This is consistent for a lower share of homes using electricity for heat in California (22 percent) than the national average (30 percent). Further, many of these electric heaters are likely located in areas with very low heating demand, given the high cost of using electricity for space heating compared to using natural gas. While there is use of electricity for heating directly, a significant share of the increased consumption at lower temperatures is likely to stem from the operation of fans for natural gas heaters. On the other end of the spectrum, for most zones in the interior and southern part of the state, we note a significant increase in electricity consumption in the highest temperature bins (4, 8, 9, 10, 11, 12, 13, and 15). We further note that the relative magnitude of this approximate percent increase in household electricity consumption in the higher temperature bins varies greatly across zones as indicated by the differential in slopes at the higher temperatures across zones. We now turn to simulating electricity consumption under different scenarios of climate change using these heterogeneous response functions as the underlying functional form relationship between household electricity consumption and temperature. 11.5

Simulations

In this section, we simulate the impacts of climate change on electricity consumption under two different Special Report on Emissions Scenarios (SRES). We calculate a simulated trajectory of aggregate electricity consumption from the residential sector until the year 2100, which is standard in the climate change literature. To simulate the effect of a changing climate on residential electricity consumption, we require estimates of the climate sensitivity of residential electricity consumption as well as a counterfactual climate. In the simulation for this section, we use the estimated climate response parameters shown in

Impacts of Climate Change on Residential Electricity Consumption

329

figure 11.4. Using these estimates as the basis of our simulation has several strong implications. First, using the estimated p parameters implies that the climate responsiveness of consumption within climate zones remains constant throughout the century. This is a strong assumption because we would expect that households in zones that currently do not require cooling equipment may potentially invest in such equipment if the climate becomes warmer. This would lead us to believe that the temperature responsiveness in higher temperature bins would increase over time. On the other hand, one could potentially foresee policy actions, such as more stringent appliance standards, which improve the energy efficiency of appliances such as air conditioners. This would decrease the electricity per cooling unit required and shift the temperature response curve downward in the higher buckets. As is standard in this literature, the counterfactual climate is generated by a GCM. These numerical simulation models generate predictions of past and future climate under different scenarios of atmospheric greenhouse gas (GHG) concentrations. The quantitative projections of global climate change conducted under the auspices of the IPCC and applied in this study are driven by modeled simulations of two sets of projections of twenty- first century social and economic development around the world, the so- called A2 and B1 storylines in the 2000 Special Report on Emissions Scenarios (SRES; Intergovernmental Panel on Climate Change [IPCC] 2000). The SRES study was conducted as part of the IPCC’s Third Assessment Report, released in 2001. The A2 and B1 storylines and their quantitative representations represent two quite different possible trajectories for the world economy, society, and energy system and imply divergent future anthropogenic emissions, with projected emissions in the A2 being substantially higher. The A2 scenario represents a “differentiated world,” with respect to demographics, economic growth, resource use, energy systems, and cultural factors, resulting in continued growth in global CO2 emissions, which reach nearly 30 gigatons of carbon (GtC) annually in the marker scenario by 2100. The B1 scenario can be characterized as a “global sustainability” scenario. Worldwide, environmental protection and quality and human development emerge as key priorities, and there is an increase in international cooperation to address them as well as convergence in other dimensions. A demographic transition results in global population peaking around midcentury and declining thereafter, reaching roughly 7 billion by 2100. Economic growth rates are higher than in A2 so that global economic output in 2100 is approximately one- third greater. In the B1 marker scenario, annual emissions reach about 12 GtC in 2040 and decline to about 4 GtC in 2100. We simulate consumption for each scenario using the National Center for Atmospheric Research Parallel Climate Model 1 (NCAR). These models were provided to us in their downscaled version for California using the Bias Correction and Spatial Downscaling (BCSD) and the Constructed Analogues (CA) algorithms (Maurer and Hidalgo 2008). There is no clear

330

Anin Aroonruengsawat and Maximilian Auffhammer

guidance in the literature as to which algorithm is preferable for impacts estimation. We, therefore, provide simulation results using both methods. To obtain estimates for a percent increase in electricity consumption for the representative household in zip code j and period t  h, we use the following relation:

(2)

⎛ k ⎞ exp ⎜ ∑ ˆ p j Dp j,t + h ⎟ qj,th ⎝ p =1 ⎠  k qj,t ⎛ ⎞ exp ⎜ ∑ ˆ p j Dp j,t⎟ ⎝ p =1 ⎠

We implicitly assume that the year fixed effect and remaining right- hand side variables are the same for period t  h and period t, which is a standard assumption made in the majority of the impacts literature. Figure 11.5 shows the change in the number of days spent in each 5° bin of the temperature distribution from 1980 to 1999 to 2080 to 2099 using the NCAR Paral-

Fig. 11.5 Change in number of days in each 5-degree temperature bin for 2080– 2099 relative to 1980–1999 for six selected California cities and Intergovernmental Panel for Climate Change Special Report on Emissions Scenarios A2 (black) and B1 (white) using the National Center for Atmospheric Research Parallel Climate Model with the constructed analogues downscaling method

Impacts of Climate Change on Residential Electricity Consumption

331

lel Climate Model (PCM) forced by scenarios A2 and B1 for six selected California locations. A clear upward shift of the temperature distribution is apparent for all six locations. For locations with upward sloping temperature response functions, this entails increases in electricity consumption due to more days spent in higher temperature bins. Inspecting these graphs for all major urban centers in California, in addition to the six displayed here, confirms the pattern emerging from figure 11.5. The areas with the steepest response functions at higher temperature bins happen to be the locations with highest increases in the number of high and extremely high temperature days. While this is not surprising, this correspondence leads to very large increases in electricity consumption in areas of the state experiencing the largest increases in temperature, which also happen to be the most temperature sensitive in consumption—essentially the southeastern parts of the state and the Central Valley. The first simulation of interest generates counterfactuals for the percent increase in residential electricity consumption by a representative household in each zip code. We feed each of the two climate model scenarios through equation (2) using the 1980 to 1999 average number of days in each temperature bin as the baseline. Figure 11.6 displays the predicted percent increase in per household consumption for the periods 2020 to 2039, 2040 to 2059, 2060 to 2079 and 2080 to 2099 using the NCAR PCM model forced by the A2 scenario using the percentile bins. Figure 11.7 displays the simulation results for the SRES forcing scenario B1. Changes in per household consumption are driven by two factors: the shape of the weather- consumption relationship and the change in projected climate relative to the 1980 to 1999 period. The maps show that for most of California, electricity consumption at the household level will increase. The increases are largest for the Central Valley and areas in southeastern California, which have a very steep temperature response of consumption and large projected increases in extreme heat days. Simulation results for this model and scenario suggest that some zip codes in the Central Valley by the end of the century may see increases in household consumption in excess of 100 percent. The map also shows that a significant number of zip codes are expected to see drops in household level electricity consumption—even at the end of the current century. It is important to keep in mind that the current projections assume no change in the temperature electricity response curve. Specifically, the current simulation rules out an increased penetration of air conditioners in areas with currently low penetration rates (e.g., Santa Barbara) or improvements in the efficiency of these devices. The projected drops essentially arise from slightly reduced heating demand. We conduct a simulation in the following, which addresses this concern. Figure 11.7 displays the simulated household increase in electricity consumption by zip code for the lower emissions scenario B1. The maps display an almost identical spatial pattern yet a smaller overall increase in consumption.

332

Anin Aroonruengsawat and Maximilian Auffhammer

Fig. 11.6 Simulated increase in household electricity consumption by zip code for the periods 2020–2039 (a), 2040–2059 (b), 2060–2079 (c), and 2080–2099 (d) in percent over 1980–1999 simulated consumption. National Center for Atmospheric Research Parallel Climate Model forced by Intergovernmental Panel for Climate Change Special Report on Emissions Scenario A2.

While changes in per household consumption are interesting, from a capacity planning perspective, it is overall consumption that is of central interest from this simulation. We use the projected percent increase in household consumption by zip code and calculate the weighted overall average increase, using the number of households by zip code as weights, in order to arrive at an aggregate percent increase in consumption. The top panel of table 11.3 displays these simulation results for aggregate consumption. Predicted aggregate consumption across all zip codes in our

Impacts of Climate Change on Residential Electricity Consumption

333

Fig. 11.7 Simulated increase in household electricity consumption by zip code for the periods 2020–2039 (a), 2040–2059 (b), 2060–2079 (c), and 2080–2099 (d) in percent over 1980–1999 simulated consumption. National Center for Atmospheric Research Parallel Climate Model forced by Intergovernmental Panel for Climate Change Special Report on Emissions Scenario B1.

data set ranges from an 18 percent increase in total consumption to 55 percent increase in total consumption by the end of the century. To put this into perspective, this represents an annual growth rate of aggregate electricity consumption between 0.17 percent and 0.44 percent, if all other factors are equal. These growth rates accelerate from period to period as the number of extreme heat days predicted from the GCMs increases in a slightly nonlinear fashion. For the first twenty- year period, the simulated annual growth rates range from 0.10 percent per year to 0.29 percent per

334

Anin Aroonruengsawat and Maximilian Auffhammer

Table 11.3

Simulated percent increase in residential electricity consumption relative to 1980–2000 for the temperature only, price ⴙ temperature and population growth ⴙ temperature (%) Equidistant BCSD Price increase

A2

2000–2019 2020–2039 2040–2059 2060–2079 2080–2099

0 0 0 0 0

5 5 15 24 48

2000–2019 2020–2039 2040–2059 2060–2079 2080–2099 2000–2019 2020–2039 2040–2059 2060–2079 2080–2099

B1

Percentile CA

A2

BCSD B1

CA

A2

B1

A2

B1

6 6 17 28 55

3 9 11 17 21

5 7 17 28 50

3 8 10 16 20

0 30 30 30 30

High price  temperature scenario 5 2 5 3 6 –6 –3 –5 –4 –5 3 –2 3 –2 6 11 3 11 2 15 33 6 29 4 39

3 –3 –1 5 9

5 –4 5 15 35

3 –3 –1 4 7

0 0 0 0 0

Population  temperature scenario 17 13 16 14 18 31 34 33 34 32 48 41 50 41 52 66 52 68 51 72 113 65 113 65 124

14 35 42 55 70

16 34 53 73 123

15 35 42 54 70

Temperature only scenario 2 5 3 8 7 8 9 17 10 15 28 16 18 50 20

Notes: Equidistant and Percentile pertain to bin type. BCSD  bias correction and spatial downscaling; CA  constructed analogues. A2 and B1 represent the Intergovernmental Panel for Climate Change scenarios.

year. Because these simulations hold population constant, the correct comparison of these growth rates for the current simulation is, therefore, one with current growth in per capita household electricity consumption for California. Figure 11.8 depicts historical per capita electricity consumption since 1960 (EIA 2008). The average annual growth rate in per capita consumption during 1960 to 1973 was approximately 7 percent and slowed down to a remarkable 0.29 percent during 1974 to 1995. Growth rates during the last decade of available data have increased to a higher rate of 0.63 percent, and this difference in growth rates is statistically significant. The estimates from our simulation are lower than this growth rate and for the 2000 to 2019 period suggest that 26 percent to 60 percent of this growth may be due to changing climate. All of the results presented in the chapter so far have excluded CARE customers from the estimation sample. One potential concern is that these households live on fewer square feet, are more likely to be renting, have lower average use and lower HVAC saturation rates. This would suggest

Impacts of Climate Change on Residential Electricity Consumption

Fig. 11.8

335

California residential per capita electricity consumption

Source: Author’s calculations based on EIA (2008) SEDS data.

that the temperature response for these households is potentially lower than for the households in the full sample. The number of CARE households in California is large. The SCE reports over 1 million customers on CARE, which is roughly one- quarter of residential accounts. For PG&E and SDG&E, the share of accounts is roughly 20 percent. We, therefore, separately sample from only the CARE households by zip code, adopting the same sampling restrictions as in the non-CARE sample. We then estimate temperature response functions by climate zone, which are slightly less steep in the higher temperature bins. We then conduct the simulations for the CARE households separately. To obtain an estimate of the overall impacts, when we include CARE, we weight impacts for each zip code by the share of CARE to non-CARE households in that zip code. Table 11.4 reports these results for the Bias Correction Spatial (BCS) downscaling algorithm and equidistant bin simulations. As suspected, the CARE households are slightly less affected by higher temperatures, yet the overall weighted average is very close to the simulations presented in table 11.3. 11.5.1

Temperature and Price Simulations

The assumed flat prices from the previous section should be considered as a comparison benchmark. It is meaningful and informative to imagine climate change imposed on today’s conditions. It is worth pointing out,

336

Anin Aroonruengsawat and Maximilian Auffhammer

Table 11.4

Simulated percent increase in residential electricity consumption relative to 1980–2000 for California Alternate Rates for Energy (CARE) and non-CARE households (%) Non-CARE

2000–2019 2020–2039 2040–2059 2060–2079 2080–2099

CARE

Weighted

Price increase

A2

B1

A2

B1

A2

B2

0 0 0 0 0

5 5 15 24 48

2 8 9 15 18

4 4 12 20 39

2 6 8 12 15

5 5 14 23 46

2 7 9 14 17

Notes: For this table, an equidistant bins approach was used, as well as the BCSD downscaling. A2 and B1 represent the Intergovernmental Panel for Climate Change scenarios.

however, that real residential electricity prices in California have been, on average, flat since the early- mid 1970s spike. In this section, we will relax the assumption of constant prices and provide simulation results for increasing electricity prices under a changing climate. While we have no guidance on what will happen to retail electricity prices twenty years or further out into the future, we consider a discrete 30 percent increase in real prices starting in 2020 and remaining at that level for the remainder of the century. This scenario is based upon current estimates of the average statewide electricity rate impact by 2020 of AB 32 compliance combined with natural gas prices to generators within the electric power sector. These estimates are based on analysis commissioned by the California Public Utilities Commission. This scenario represents the minimum to which California is committed in the realm of electricity rates. This scenario could be interpreted as one assuming very optimistic technological developments post- 2030, implying that radical CO2 reduction does not entail any cost increases, or as a California and worldwide failure to pursue dramatic CO2 reductions such that California’s AB 32 effort is not expanded. To simulate the effects of price changes on electricity consumption, we require good estimates of the price elasticity of demand. In this chapter, we rely on the estimates of mean price elasticity provided by Reiss and White (2005). Specifically, they provide a set of average price elasticities for different income groups, which we adopt here. Because we do not observe household income, we assign a value of price elasticity to each zip code based on the average household income for that zip code. Households are separated into four buckets, delineated by $18,000, $37,000, and $60,000 with estimated price elasticities of –0.49, –0.34, –0.37, and –0.29, respectively. It is important to note that these price elasticities are short- run price elasticities. These are valid if one assumes a sudden increase in prices, as we do in this chapter. To our knowledge, reliable long- term price elasticities based on microdata for California are not available, but in theory, they are larger than

Impacts of Climate Change on Residential Electricity Consumption

337

the elasticities used in this chapter. The second panel in table 11.3 presents the simulation results under the scenarios of climate change given a sudden persistent increase in electricity prices in the year 2020. Given the sizable assumed price elasticity estimates, it is not surprising that the simulated increases in residential electricity consumption for the first period after the price increase are roughly 6 percent to 12 percent lower than the predicted increases given constant prices. For the NCAR model under both considered forcing scenarios the path of electricity consumption under this price scenarios returns to levels below its 1980 to 2000 mean for the 2020 to 2040 period, given this assumed price trajectory. By the end of the century, we still observe significant increases in electricity demand for the higher forcing scenario (A2). It is important to note that these effects are conditional on the estimated price elasticities being correct. Smaller elasticities would translate into price- based policies, such as taxes or cap and trade systems, being less effective at curbing demand compared to standards. 11.5.2

Temperature and Population

California has experienced an almost sevenfold increase in its population since 1929 (Bureau of Economic Analysis [BEA] 2008). California’s population growth rate over that period (2.45 percent) was more than twice that of the national average (1.17 percent). Over the past fifty years, California’s population has grown by 22 million people to almost 37 million in 2007 (BEA 2008). To predict what the trajectory of California’s population will look like until the year 2100, many factors have to be taken into account. The four key components driving future population are net international migration, net domestic migration, mortality rates, and fertility rates. The State of California provides forecasts fifty- five years out of sample, which is problematic because we are interested in simulating end- of- century electricity consumption. The Public Policy Institute of California has generated a set of population projections until 2100 at the county level. For illustration purposes, we use their “low” series, where population growth slows as birth rates decline, migration out of the state accelerates, and mortality rates show little change. This low series is equivalent to a 0.18 percent growth rate and results in a population 18 percent higher than today’s. Projections are available at the county level and not at the zip code level. We, therefore, assume that each zip code in the same county experiences an identical growth rate. The bottom panel of table 11.3 displays the simulated aggregate electricity consumption given the “low” population growth scenarios. This table holds prices constant at the current level. It is not surprising to see that population uncertainty has much larger consequences for simulated total electricity consumption compared to uncertainty over climate or uncertainty over prices. The simulations for the low forcing scenario B1 and the low population growth scenario show 65 percent to 70 percent increase in residential electricity consumption. If we consider the A2 forcing, the predicted low population

338

Anin Aroonruengsawat and Maximilian Auffhammer

average increase in consumption is a 118 percent increase. The source of this disproportionate increase in overall consumption from a relatively modest increase in population the predicted increases in population in areas with steeper response functions (e.g., the Central Valley). 11.6

Adaptation

The major finding in the chapter so far is the heterogeneity in temperature response of residential electricity consumption across climate zones. While geographic location clearly plays an important role in determining this responsiveness, we wish to study whether there are household or structure characteristics, which help explain some of this difference in temperature response. We, therefore, construct a statewide sample by sampling 10 percent of the households from each of the sixteen climate zone- specific data sets used in the preceding. We restrict ourselves to non-Care customers in this exercise. We construct 10 percentile temperature bins, where the cutoffs are at every 10th percentile of the California- wide temperature distribution for the years 2003 to 2006. The smaller number of bins and percentile approach guarantee that there are enough observations in the extreme bins at meaningful cutoff points. We then slice the preceding data set along several dimensions in order to see whether the temperature response varies with certain variables of interest from the census 2000 Summary File 3 (SF 3). Specifically, for each indicator, we divide this sample into two groups, a “low group” and a “high group,” based on the value of the variables of interest. The following are the variables of interest and percentiles used in estimation: 1. Percentage of household using electricity as heating fuel. • Low group: households in zip code with this variable 30 percent • High group: households in zip code with this variable  60 percent 2. Percentage of household using gas as heating fuel. • Low group: households in zip code with this variable 40 percent • High group: households in zip code with this variable  60 percent 3. Percentage of households in an urban area. • Low group: households in zip code with this variable 40 percent • High group: households in zip code with this variable  60 percent 4. Median year of structure built. • Low group (older building): zip codes with median year of structure built 1959 • High group (newer building): zip codes with median year of structure built 1979

Impacts of Climate Change on Residential Electricity Consumption

Fig. 11.9

339

Temperature response for households by major heating fuel

For each variable of interest, we estimate the same models as previously, while making sure that we are making a fair comparison across groups. For our regressions, we, therefore, limit the sample for both groups to those households with median household income between 40 to 60 percent of the distribution of census 2000 zip- code- level median household income. For each variable of interest, we plot the estimated coefficients for each temperature bin against their midbin temperature. Each of the graphs has two sets of lines, one for “low group” (thin lines) and the other one for “high group” (thick lines). We also plot the 95 percent confidence intervals for each group. Figure 11.9 plots the response functions for households in zip codes with a high penetration of electricity as the major heating fuel against the response functions for households from zip codes with a low penetration of electricity of a heating fuel. The difference is drastic and

340

Anin Aroonruengsawat and Maximilian Auffhammer

Fig. 11.10

Temperature response for households by year built and urban location

statistically significant. The zip codes using electricity as the major source of heat have significantly higher electricity consumption at low temperatures, while the low penetration zip codes have an almost flat response. The following panel displays the figure for natural gas. It is switched, which is not surprising, given that electricity and natural gas are the two major heating fuels in California. In the top panel, it is also noteworthy that households with higher electric heating have a drastically higher temperature response at high temperatures. Figure 11.10 displays the temperature response functions for older houses versus newer houses in the top panel. At the low- temperature spectrum, newer houses seem to require more electricity to heat compared to older houses. At the high end of the temperature spectrum, older and newer houses appear to have an almost identical temperature response. The bottom panel of figure 11.10 displays the temperature response for houses located in mostly urban zip codes versus the temperature response of households located in mostly rural zip codes. The difference is quite drastic, with rural households having an almost flat temperature response function and urban households having the typical U- shaped response. This finding is due to the fact that much of the Central Valley and the greater Los Angeles area are considered urban.

Impacts of Climate Change on Residential Electricity Consumption

11.7

341

Conclusions

This study has provided the first estimates of California’s residential electricity consumption under climate change based on a large set of panel microdata. We use random and, therefore, exogenous weather shocks to identify the effect of weather on household electricity consumption. We link climate zone specific weather response functions to a state of the art downscaled global circulation model to simulate growth in aggregate electricity consumption. We further explore the household characteristics potentially responsible for the heterogenous temperature response of consumption. There are two novel findings from this chapter. First, simulation results suggest much larger effects of climate change on electricity consumption than previous studies. This is largely due to the highly nonlinear response of consumption at higher temperatures. Our results are consistent with the findings by Deschênes and Greenstone (2007). They find a slightly smaller effect using national data. It is not surprising that impacts for California, a state with a smaller heating demand (electric or otherwise), would be bigger. Second, temperature response varies greatly across the climate zones in California—from flat to U- shaped to hockey stick- shaped. This suggests that aggregating data over the entire state may ignore important nonlinearities, which combined with heterogeneous climate changes across the state may lead to underestimates of future electricity consumption.

References Baxter, L. W., and K. Calandri. 1992. Global warming and electricity demand: A study of California. Energy Policy 20 (3): 233–44. Bureau of Economic Analysis (BEA). 2008. Regional economic accounts. Washington, DC: BEA. http://www.bea.gov/regional/spi/default.cfm?seriessummary. California Energy Commission (CEC). 2005. Integrated energy policy report. Sacramento, CA: CEC. Cline, W. R. 1992. The economics of global warming. Washington, DC: Institute for International Economics. Crowley, C., and F. Joutz. 2003. Hourly electricity loads: Temperature elasticities and climate change. In Proceedings of the International Association for Energy Economics 23rd North American Conference. Mexico City, Mexico, October 20. Deschênes, O., and M. Greenstone. 2007. Climate change, mortality and adaptation: Evidence from annual fluctuations in weather in the U.S. MIT Department of Economics Working Paper no. 07- 19. Energy Information Administration (EIA). 2008. State energy data system. Washington, DC: EIA. Fisher, F. M., and C. Kaysen. 1962. A study in econometrics: The demand for electricity in the US. Amsterdam: North-Holland.

342

Anin Aroonruengsawat and Maximilian Auffhammer

Franco, G., and A. Sanstad. 2008. Climate change and electricity demand in California. Climatic Change 87:139–51. Hanemann, W. 1984. Discrete/continuous models of consumer demand. Econometrica 52:541–61. Houthakker, H. S., and L. D. Taylor. 1970. Consumer demand in the United States: Analyses and projections. Cambridge, MA: Harvard University Press. Intergovernmental Panel on Climate Change (IPCC). 2000. Emissions scenarios. Cambridge, UK: Cambridge University Press. Linder, P., M. Gibbs, and M. Inglis. 1987. Potential impacts of climate change on electric utilities. Report no. 88- 2. Albany, NY: New York State Energy Research and Development Authority. Mansur, E., R. Mendelsohn, and W. Morrison. 2008. Climate change adaptation: A study of fuel choice and consumption in the U.S. energy sector. Journal of Environmental Economics and Management 55 (2): 175–93. Maurer, E. P., and H. G. Hidalgo. 2008. Utility of daily vs. monthly large- scale climate data: An intercomparison of two statistical downscaling methods. Hydrology and Earth System Sciences 12:551–63. Mendelsohn, R. 2003. The impact of climate change on energy expenditures in California. In Global climate change and California: Potential implications for ecosystems, health, and the economy, ed. T. Wilson, L. Williams, J. Smith, and R. Mendelsohn, appendix II. Consultant report no. 500- 03- 058CF to the Public Interest Energy Research Program. Sacramento, CA: California Energy Commission. Reiss, P. C., and M. W. White. 2005. Household electricity demand revisited. Review of Economic Studies 72:853–83. Rosenthal, D., H. Gruenspecht, and E. Moran. 1995. Effects of global warming on energy use for space heating and cooling in the United States. Energy Journal 16:77–96. Schlenker, W., and M. Roberts. 2006. Nonlinear effects of weather on corn yields. Review of Agricultural Economics 28 (3): 391–98. U.S. Environmental Protection Agency. 1989. The potential effects of global climate change on the United States, ed. Joel B. Smith and Dennis Tirpak. Washington, DC: Environmental Protection Agency.

Contributors

Anin Aroonruengsawat Faculty of Economics Thammasat University 2 Prachan Road Bangkok 10200, Thailand Maximilian Auffhammer Department of Agricultural and Resource Economics 207 Giannini Hall University of California, Berkeley Berkeley, CA 94720- 3310 Karen Clay Heinz College Carnegie Mellon University 5000 Forbes Avenue Pittsburgh, PA 15213 Price V. Fishback Department of Economics University of Arizona Tucson, AZ 85721 Jonathan F. Fox Department of Economics University of Arizona Tucson, AZ 85721

Michael Haines Department of Economics 217 Persson Hall Colgate University 13 Oak Drive Hamilton, NY 13346 Zeynep K. Hansen Department of Economics Boise State University 1910 University Drive Boise, ID 83725- 1620 Trevor Kollmann Department of Economics University of Arizona Tucson, AZ 85721 John Landon-Lane Department of Economics Rutgers University 75 Hamilton Street New Brunswick, NJ 08901- 1248 Gary D. Libecap Bren School of Environmental Science & Management 4412 Bren Hall University of California, Santa Barbara Santa Barbara, CA 93106

343

344

Contributors

Scott E. Lowe Department of Economics Boise State University 1910 University Drive Boise, ID 83725- 1620 Alan L. Olmstead Department of Economics University of California, Davis Davis, CA 95616 Robert S. Pindyck Sloan School of Management Room E62- 522 MIT 100 Main Street Cambridge, MA 02139 Paul W. Rhode Department of Economics 205 Lorch Hall University of Michigan 611 Tappan Street Ann Arbor, MI 48109- 1220 Michael J. Roberts Department of Agricultural and Resource Economics North Carolina State University Box 8109 Raleigh, NC 27695- 8109 Hugh Rockoff Department of Economics Rutgers University 75 Hamilton Street New Brunswick, NJ 08901- 1248

Wolfram Schlenker Department of Economics and School of International and Public Affairs Columbia University 420 West 118th Street, MC 3323 New York, NY 10027 Richard H. Steckel Department of Economics 410 Arps Hall Ohio State University 1945 North High Street Columbus, OH 43210- 1172 Richard Sutch Department of Economics University of California, Riverside Riverside, CA 92507 Melissa Thomasson Department of Economics Miami University Oxford, OH 45056 Werner Troesken Department of Economics University of Pittsburgh Pittsburgh, PA 15260 Martin L. Weitzman Department of Economics Harvard University Littauer 313 Cambridge, MA 02138

Author Index

Ackerman, A. S. E., 294n7 Allen, M., 32n6, 35 Alley, W. M., 10 Alston, J. M., 188n45 Alston, L. J., 88 Andreadis, K. M., 254 Andrew, A. P., 75 Andrews, D. G., 35 Antonovsky, A., 151 Archer, D., 39 Baker, M. B., 32, 32n6 Baker, O. E., 171n4 Ball, C. R., 177n13, 177n16, 177n17 Barrett, S., 58n5 Baxter, L. W., 313, 314 Beeson, P. E., 151 Bergey, D. H., 135 Bernanke, B. S., 83n3 Bernstein, J., 151 Betrán, J., 201n7, 206n15 Bleakley, H., 135 Bogart, E. L., 75 Bogue, A. G., 76, 77, 78, 206n15 Booth, W. H., 292 Bordo, M. D., 75 Boss, A. D., 184n35 Bowman, M. L., 181n33 Boyer, J. G., 56 Brannon, Y. S., 151 Brazell, J. H., 285, 295 Brimblecombe, P. T., 291

Brodie, F. J., 281, 283, 295, 296 Brouwer, C., 257n1 Buller, A. H. R., 177n15, 184n34 Caborne, W. F., 291 Calandri, K., 313, 314 Calomiris, C. W., 75 Cardwell, V. B., 203n11 Carleton, M. A., 178n18, 178n22 Carlson, M., 75 Carlton, M. A., 172 Carpenter, A., 294m7 Carter, S. B., 202, 209 Castleberry, R. M., 202n9, 203n11 Chan, Y. C., 136n10 Chandler, T. J., 285 Chapin, F. S., 227 Chay, K., 151 Clark, J. A., 177n16, 177n17 Clemow, F. G., 136 Clifford, W. B., 151 Cline, W. R., 253, 313 Cohen, W., 158 Collier, J. W., 204 Collings, G. H., 188n45 Coman, K., 74 Constantine, J. H., 188n45 Cook, E. R., 11, 11n3 Cooley, T. F., 103 Craig, L. A., 128n12, 133n5, 151, 172n7 Creighton, C., 133 Crossley, B. W., 181n33

345

346

Author Index

Crowley, C., 314 Crum, C. W., 202n9, 203n11 Culver, J. C., 210n18, 210n20, 210n21, 211, 212, 213, 214, 215, 216 Cunfer, G., 132n2 Dasgupta, P., 40, 58 David, P. A., 201n4 Davis, J. H., 75, 92 Davis, L. E., 86 Debreu, G., 58n6 DeCabuim, S. J., 103 Dehejia, R., 151 Dell, M., 55, 55n3 Deschênes, O., 109, 133, 228n1, 253, 314, 324, 341 Dieter, L., 29n3 Dietz, S., 56 Dines, W. H., 306 Doti, L. P., 82 Dowell, A. A., 204n12 Duflo, E., 266, 276 Durost, D. D., 202n9, 209, 215n24 Duvick, D. N., 202, 203, 206n15, 210, 216 Fernandez-Cornejo, J., 216 Ferrie, J. P., 140 Field, C. B., 169n1 Fishback, P. V., 128n12, 140, 151, 152, 158 Fisher, A. C., 228, 229, 253 Fisher, F. M., 312 Fite, G. C., 188n46 Fitzgerald, D., 212, 212n23, 219 Fogel, R. W., 151 Fox, J., 141 Franco, G., 314 Frankland, E., 284n2 Frazier, I., 176n9 Friedman,, M., 15, 74, 75 Fuglie, K., 216 Galloway, P. R., 128n12, 133n5 Gardner, J., 158 Gibbs, M., 313 Goodrich, C., 176n10, 179n26 Greenstone, M., 109, 133, 151, 227n1, 253, 314, 324, 341 Griliches, Z., 18, 199, 201, 201n4, 201n6, 204, 205n14, 216, 217 Gross, N., 220

Gruenspecht, H., 313 Haines, M. R., 128n12, 133n5, 151, 152, 172n7 Hake, K. D., 185n41 Hall, B., 201n5 Hamilton, D. E., 101 Handy, R. B., 184n38 Hanemann, W. M., 229, 229n4, 253, 326 Hanes, C., 75, 92 Hansen, J., 30n4 Hansen, Z. K., 82, 132n2, 178n25, 195n1, 255 Hargreaves, M. W. M., 179n27 Hart, J. F., 185n41, 188n46 Hays, W. M., 184n36 Heal, G., 57n6 Heibloem, M., 257n1 Heim, R. R., Jr., 9 Hendrickson, L. G., 184n35 Herrick, G. W., 135 Hewitt, C. G., 135 Hibbard, B. H., 178n21 Hope, C. W., 56n4 Hornbeck, R., 195n1 Horrace, W. C., 158 Houthakker, H. S., 312 Humphreys, M., 135 Hyde, J., 210n18, 210n20, 210n21, 211, 212, 213, 214, 215, 216 Inglis, M., 313 James, J. A., 86 Jenkins, M. T., 199, 213 Jesness, O. B., 204 Jevons, W. S., 15, 73, 74 Johnson, P. R., 202n9 Johnson, W. D., 176n11 Jones, B. F., 55, 55n3 Joutz, F., 314 Jugenheimer, R. W., 208 Kantor, S., 151, 152, 158 Kaplan, G. A., 151 Karl, T. R., 105 Kashiwagi, T., 139 Kawachi, I., 151 Kaysen, C., 312 Kelly, D. L., 253 Kennedy, B. P., 151

Author Index Kerby, T. A., 185n41 Keynes, J. M., 15, 74 Klippart, J. H., 170n3 Kolstad, C. D., 253 Komlos, J., 128n12, 133n5 Kousky, C., 12 Kramer, R. A., 101 Kriström, B., 58n6 Krull, C. F., 202n9, 203n11 Krusic, P. J., 11 Landon-Lane, J., 92 Lange, F., 101, 187n44 Larson, D. F., 107, 123 Lauszus, D., 140 Lawther, P. J., 298 Lee, H., 213 Lempfert, R. G. K., 284, 286, 287, 288n5 Lettenmaier, D. P., 254 Libecap, G. D., 2, 82, 132n2, 178n25, 195n1, 255 Linder, P., 313 Lleras-Muney, A., 151 Lord, R., 206n16, 211n22 Lucas, R. E., Jr., 58n6 Luekel, R. W., 178n21, 178n23 Lytle, C. D., 136 MacCallum, F. O., 139 MacNutt, J. S., 134n7 Maddala, G. S., 106 Mäler, K.-G., 58n5 Malin, J. C., 177n14, 178n19, 178n21 Mansur, E., 313 Martin, A. J., 307 Martin, J. H., 177n16, 177n17 Mathews, O. R., 178n21, 178n23 Matthews, M. S., 103 May, E., 212, 213, 214 McDonald, J. R., 139 McIntyre, S., 5 Mendelsohn, R., 48n1, 56n4, 229n4, 253, 313 Miller, R. C., 81 Mitchell, B. R., 283, 298 Mitchell, G. T., 253 Mitchell, W. C., 74 Mitchener, K. J., 75 Monath, T. P., 157 Moore, H. L., 74 Moran, E., 313 Morrison, W., 313

347

Mossman, R. C., 283 Mundlak, Y., 107, 123 Nelson, G. C., 227, 246 Nishiura, H., 139 Nordhaus, W. D., 12, 25n1, 28, 28n2, 48n1, 52, 56n4, 229n4, 253 Norrie, K. H., 177n15 Officer, L. H., 103 Olken, B. A., 55, 55n3 Olmstead, A. L., 12, 17, 101, 152, 176n10, 177n12, 187n43, 187n44, 199, 201n8, 206n16, 254 Ortiz, R., 170n2, 171 Owens, J. S., 292 Pachauri, R. K., 3 Palmer, W. C., 9, 10 Pande, R., 266, 276 Persson, U. M., 25n1, 27, 28 Pielke, R. A., Jr., 254 Pigou, A. C., 20, 74 Pindyck, R. S., 48, 49n2, 66, 68 Pisani, D. J., 19, 256 Poehlman, J. M., 185n39, 185n40 Pond, G. A., 184n35 Poore, G. V., 290 Prothrow-Stith, D., 151 Qaim, M., 245 Quisenberry, K. S., 177n14 Ramirez, C. D., 75 Redenius, S. A., 86, 86n4 Redish, A., 75 Reichart, T. A., 137 Reilly, J., 278, 279 Reisinger, A., 3 Reiss, P. C., 326 Reitz, L. P., 177n14 Reyes, J. W., 151 Reynolds, M., 292 Rhode, P. W., 12, 17, 75, 92, 101, 152, 176n10, 177n12, 187n43, 187n44, 201n8, 206n16, 254 Rideal, S., 292, 293 Roberts, M. J., 225, 235, 253, 325 Robinson, J. L., 199 Robinson, W. W., 256 Rockoff, H., 75, 92

348

Author Index

Roe, G. H., 32, 32n6, 35 Rogers, J. S., 204 Rogers, L., 133 Rosenau, M. J., 134 Rosenthal, D., 313 Rostapshova, O., 12 Routh, C. H. F., 134n7 Ruhm, C. J., 151 Runge, C. A., 201n7, 206n15 Russell, W. A., 203n11 Russell, W. J., 284, 284n1, 284n2 Ryan, B., 220 Sagripanti, J.-L., 136 Salmon, S. C., 178n20, 178n21, 178n23, 178n24 Sanstad, A., 314 Saward, F. E., 292 Schaefer, T. W., 298 Schlenker, W., 225, 229, 229n4, 235, 253, 325 Schlesinger, A., Jr., 211 Schlesinger, M. E., 32n6 Schlict, P., 286n3 Schubert, S. D., 195 Schwartz, A. H., 15, 74, 75 Schweikart, L., 82 Scott, R. H., 296 Sedgwick, W. T., 134n7 Shaw, D., 229n4, 253 Shaw, L. H., 202n9, 209 Shuey, P., 298 Sleper, D. A., 185n39, 185n40 Smil, V., 203n10 Smiley, G., 86 Smith, C. W., 201n7, 206n15 Smith, V. H., 188n45 Snowden, K., 78 Snyder, R. L., 249 Sokolov, A. P., 65, 169n1 Spinzig, C., 133 Sprague, G. F., 201, 204, 204n13, 215 Sprague, O. M. W., 75

Steckel, R., 128n12, 133n5, 151, 180n29, 180n30, 181n31, 181n32 Stephens, S. G., 185n39 Stern, N., 57, 253 Sterner, T., 25n1, 27, 28 Stokey, N., 54 Sutch, R. C., 202n9, 203n10, 229, 230 Sylla, R., 86 Taylor, L. D., 312 Thomasson, M., 146, 151n11 Thornwaite, C. W., 9 Thorp, W. L., 101 Tol, R. S. J., 56 Treber, J., 146, 151n11 Troesken, W., 136n8, 139, 140, 151 Trotter, P. S., 101 Troyer, A. F., 184n35, 184n36, 206n16 Turner, J., 188n45 Unstead, J. F., 171n4 Urban, N., 210, 210n18 Ward, T., 177n15 Ware, J. O., 184n37, 184n38, 185n42, 188n45 Webb, W. P., 176n9 Weidmann, M., 136n10 Weiss, T., 128, 133n5, 151, 172n7 Weitzman, M. L., 31n5, 41n10, 41n11, 43, 43n12, 48n1, 51, 52, 253 Whipple, G. C., 135 White, M. W., 326 White, W. C., 298 Wigley, T. M. L., 32n6 Wise, W., 294, 296 Wright, C. D., 293 Wu, S., 106 Yusuf, S., 136 Zilberman, D., 245 Zinsser, H., 136 Zuber, M. S., 199

Subject Index

Page numbers followed by the letters f and t indicate figures and tables, respectively. adaptation, 254; Californian residential electricity consumption and, 338– 41 adverse weather. See droughts; extreme weather agriculture: climate change and, 225– 26; economic historians on how weather affects, 73– 76; historical data for, 259– 62; weather, financial markets, and, 15; western water supplies and, 254– 55; western water supply infrastructure and, 254– 55 American agricultural development, hallmarks of, 171– 72 Andrew, A. Piatt, 75 anticyclones, London fogs and, 285 bank equity: drought and, 15; effect of drought on rates of return to, 89– 95 bank failures: drought and, 15; national, in Kansas (1875– 1910), 79– 80t banking systems, local: effect of extreme weather on, 84– 87, 89– 95 BAU (business as usual) scenarios, 49 birth rates, infant mortality and, 152 Bogart, Ernest Ludlow, 75 bovine tuberculosis (BTB), infant mortality and, 152 branch banking, weather- related agricultural shocks and, 75 Brodie, Frederick J., 281– 82

BTB. See bovine tuberculosis (BTB), infant mortality and Burman, Erik, 8 business as usual (BAU) scenarios, 49 California: residential electricity consumption of, 311– 12; total electricity consumption in, 311. See also electricity consumption; residential electricity consumption California Alternate Rates for Energy (CARE), 318, 334– 35 cap and trade systems, 2 carbon dioxide levels, 29 carbon taxes, 2 CARE. See California Alternate Rates for Energy (CARE) Celsius, Anders, 8 climate: literature on weather, mortality and, 133– 39; weather, noninfant mortality rates, and, 152– 55 climate change, 253– 54; adaptability of American economy to, 12– 13; agriculture production and, 225– 26; capacity to adapt to, 5; challenges of representing damages from, 23– 24; costs of, 14– 17; crises and dealing with, 2; economics of, 31– 32; effects of, on North America by end of twenty- first century, 169– 70; impact of, on mortality rates,

349

350

Subject Index

climate change (continued ) 16– 17; impact of, on residential electricity consumption, 20– 21; implications of study for understanding, 283; importance of international collective action to address, 2; industrialization and, 283; methods of discounting disutilities of, 39– 42; modeling impact of warming in, 14; reasons for unpredictability of, 32– 35; uncertain economic implications of, 13– 14 climate change policy: damages model for, 47– 50; multiplicative vs. additive net utility function and, 25– 29 climate extremes, structural uncertainty about, 29– 32 climate feedback, defined, 35 climate forcing, defined, 35 climate sensitivity, 24– 25, 30; “dismal proposition” for fat- tailed infinite- variance, 42– 44; reasons for unpredictability of, 32– 35 climatic challenges, evidence of adaptation to, 17– 19 climatic conditions, variable: government policy and adaptation to, 19– 21 coal consumption, London fogs and, 283– 88; illustration of (1700– 1910), 283f Coman, Katherine, 74 constant relative risk aversion (CRRA) utility function, 49, 57– 58 corn. See hybrid corn; maize Corn Belt: corn production in, during 1930s, 196, 196f cotton: adaptation and, 185; changing distribution of U.S. production of, 184– 88; distribution of U.S. production of, 189– 90t; Mexican stocks of, 186– 87; types of, in antebellum period, 186 Crookes, Sir William, 171n4 crop harvests: causes of, 100– 101, 102t; data for, 103– 6; droughts and, 101– 2; global warming and, 245– 46 crop prices, weather shocks and, 16 CRRA. See constant relative risk aversion (CRRA) utility function death rates. See mortality rates dendochronology, 10– 11 DICE. See Dynamic Integrated model of Climate and the Economy (DICE) displaced gamma distribution, 50– 51

droughts: bank equity and, 15; bank failures and, 15; defined, 102n3; development of ideas for measuring, 9– 10; effect of, on farm income, 87– 88; effects of, on crop harvests, 100; impact of, on adoption of hybrid corn, 217– 19; local weather and, 100– 102. See also Dust Bowl; extreme weather dry- farming techniques: for wheat, 178– 79 Dust Bowl, 195, 195n1; weather- driven financial distress during, 81– 84. See also droughts Dynamic Integrated model of Climate and the Economy (DICE), 28, 52 economic development, mortality rates and, 139– 40 electricity consumption: economics literature on, 313– 14; engineering literature on, 313; impacts of climate change on residential, 20– 21; literature review of forecasting, 312; panel data- based approach to estimating, 314– 15; time series studies of, 314; total, in California, 311. See also residential electricity consumption environmentalism, Victorian, 282, 288– 94. See also London fogs environmental Kuznets curve (EKC), 20, 282 evapotranspiration, 9 exponential loss function, 52 extreme weather: effect of, on farm income, 84– 88; effect of, on farm mortgage foreclosures, 88– 89; effect of, on local banking systems, 84– 87, 89– 95; empirical model of farm commodity prices and, 109– 12; results of, model and farm commodity prices, 112– 27. See also droughts Fahrenheit, Daniel, 8 farm commodity prices: data for, 103– 6; empirical model of adverse weather and, 109– 12; results of model of extreme weather and, 112– 27; transportation costs and, 106– 9 farm income: effect of drought on, 87– 88; effect of extreme weather on, 84– 87 farm mortgage foreclosures: effect of drought on, 88– 89; effect of extreme weather on, 84– 87 Federal Reclamation Act (1902), 256

Subject Index feedbacks, climate: defined, 30n4, 35 financial markets, agriculture, and weather, 15 flies, mortality and, 135 fog- related events: defined, 296– 97 fogs. See London fogs forcings, climate, 35 Friedman, Milton, 74, 75 global warming, 131; crop yields and, 245– 46; dynamic aggregative model of, 35– 39; London fogs and, 283; modeling impact of, 52– 57. See also climate change Gore, Al, 131 government policies, adaption to climate conditions and, 19– 21 Great Depression: data and estimation for impact of temperature on mortality rates during, 141– 46; impact of climate change and weather on mortality rates during, 16– 17; mortality rates during, 132– 33 greenhouse gases (GHG), 2; levels of, and climate sensitivity, 29– 31; overall weather response to, and challenges to American economy, 11 greenhouse gases (GHG) abatement policy, case for, 60 Green Revolution, 230 growth rates: impact of global warming on, 53– 55; willingness to pay and, 59– 60 harvests, factors contributing to good or bad, 101– 2 heat- tolerant corn: discussion of results for yields of, 243– 45; regression model for yields of, 235– 36; results of model for yields of, 236– 43. See also hybrid corn; maize Hi-Bred Seed Company, 210– 12 Horsely, Reverend, 8 hybrid corn: adoption of, 18; adoption of, during 1930s, 199– 204; depressed prices of corn during Great Depression and adoption of, 209– 10; drought of 1930s and, 214– 15; Dust Bowl and, 195– 99; Henry A. Wallace and, 210– 14; history of development of, 208– 10; illustration of impact of drought on adoption of, 217– 19; inventors of, 199; Iowa corn yield tests and superiority of, 204– 6; role of research and success in adop-

351

tion of, 215– 17; vigor of, 206– 10. See also heat- tolerant corn; Indiana; maize; Wallace, Henry Agard IAMs. See integrated assessment models (IAMs) ice core drilling, 29 Indiana: data for corn yields and weather in, 230– 35, 247– 50; discussion of results of model for heat tolerance of corn yields in, 243– 45; regression model for heat tolerance of corn yields in, 235– 36; results of model for heat tolerance of corn yields in, 236– 43. See also hybrid corn; maize industrialization, climate change and, 283 infant mortality: annual fluctuations in temperature, precipitation, and, 146– 50; birth rates and, 152; bovine tuberculosis and, 152; economic activity and, 151; influence of public health education and prevention on temperature and, 140– 41; temperature changes and, 134– 35. See also mortality rates information, mortality rates and, 139– 40 instrument readings, for measuring weather, 6– 8 integrated assessment models (IAMs), 28– 29, 49, 55n4, 56 Intergovernmental Panel on Climate Change (IPCC), 3 Iowa corn yield tests, 204– 6, 207t, 211– 12 Jevons, William Stanley, 73– 74 Jones, Donald F., 199 Kansas: early production of wheat in, 178; national bank failures in (1875– 1910), 79– 80t; weather- driven financial distress in nineteenth century, 76– 81 Keynes, John Maynard, 74 Klippart, John, 170– 71, 170n3, 171n4 Kuznets curve. See environmental Kuznets curve (EKC) Lempfert, R. G. K., 284, 286– 88, 288n5 Leonardo da Vinci, 10 local weather, drought and, 100– 102 London fogs, 20; adoption of gas and reduction of, 289, 293– 94; adoption of hard coal and, 292– 93; anticyclones and, 285; coal consumption (1700– 1910)

352

Subject Index

London fogs (continued ) and, 283– 88, 283f; effects of laws and changing technologies on, 291– 94; evaluation of evidence on, 281– 82; evaluation of mortality data and, 282; famous, 294– 95; global warming and, 283; history of, 295– 97; identification of, 297– 98; plotting deaths from, 298– 306; population redistribution and reduction of, 288, 289– 91; production of coal smoke and, 281; Public Health Act and reduction of, 288– 89; reasons economists and economic historians should be interested in, 282– 83; Victorian environmentalism and, 288– 94 maize: changing distribution of U.S. production of, 180– 84; distribution of U.S. production of, 182– 83t; evolution of heat- tolerant, 18– 19, 227– 29; growing of, 206– 7. See also heat- tolerant corn; hybrid corn; Indiana malaria, mortality and, 135 Marquis wheat, 176– 78 mercury thermometer, 8 methane levels, 29 Mitchell, Wesley Clair, 74– 75 Moore, Henry Ludwell, 74 mortality rates: climate, weather, and noninfant, 152– 55; data and estimation for impact of temperature on, during Great Depression, 141– 46; economic development and, 139– 40; flies and, 135; during Great Depression, 132– 33; impact of climate and weather on, 16– 17; information and, 139– 40; literature on climate, weather, and, 133– 39; malaria and, 135; mosquitoes and, 135; rainfall and, 139; respiratory diseases and, 136; typhoid and, 135; water- borne diseases and, 134– 35; weather and, 139– 41. See also infant mortality mosquitoes, mortality and, 135 national meteorological services, 8 National Smoke Abatement Institution (NSAI), 291– 92 North American Drought Atlas, The, 11 Oklahoma, weather- driven financial distress during Dust Bowl in, 81– 84. See also Dust Bowl

Palmer Drought Severity Index (PDSI), 9– 10, 255; rates of return and, 15– 16 Pigou, A. C., 74 Pioneer Hi-Bred Seed Company, 18, 210– 14 pluviometer (rain gauge), 8 price effects, transportation costs and, 106– 9 probability density function (PDF), 24– 25, 30 Public Health Act (UK), 20, 288– 89 public health education, influence of, on temperature and infant mortality, 140– 41 public health technologies, mortality rates and, 141 rainfall, mortality and, 139 rain gauge (pluviometer), 8 rates of return, Palmer Drought Severity Index and, 15– 16 Reclamation Fund, 256 Red Fife wheat, 176– 77 residential electricity consumption, 20– 21; adaptation and, 338– 41; alternative data sources for, 322– 23; of California, 311– 12; econometric estimation of, 324– 28; household data for, 315– 19; simulations of, 328– 35; temperature and population growth and, 337– 38; temperature and price simulations of, 335– 37; weather data for, 319– 22. See also electricity consumption respiratory diseases, mortality and, 136 Russell, Rollo, 284 Schwartz, Anna J., 74, 75 seed types, new: adoption of, 18 smoke abatement technologies, 282 Sprauge, O. M, 75 sunspot theory, 73– 74 temperature changes, 3– 5; data and estimation for impact of, on death rates during Great Depression, 141– 46; infant mortality and, 134– 35; influence of public health education and prevention on infant mortality and, 140– 41; model of, 50– 52; mortality and, 134– 39 temperature damages, 25– 29 transportation costs, price effects and, 106– 9

Subject Index tree rings, study of, 10– 11 Turkey- type wheat, 177 typhoid, mortality and, 135 United States: adaptability of economy of, to climate change, 12– 13; agricultural production of, 225– 26; meteorological services of, 8 U.S. Climate Division Dataset (USCDD), 255, 262 Victorian environmentalism, 282. See also London fogs Wallace, Henry Agard, 18, 199, 210– 14, 210n20, 210n21 warming. See global warming water- borne diseases, mortality and, 134– 35 weather: agriculture, financial markets, and, 15; climate, noninfant mortality rates, and, 152– 55; impact of, on mortality rates, 16– 17; literature on climate, mortality, and, 133– 39; local, and drought, 100– 102; methods of measuring, 6– 11; mortality rates and, 139– 40 weather patterns: changes in, 3– 5; systematic study of, 8 weather shocks, crop prices and, 16

353

western water supplies, agriculture and, 254– 55 western water supply infrastructure: agriculture and, 254– 55; climate data for, 262– 63; data for, 257– 61; econometric model of, 264– 67; historical agriculture data for, 259– 62; origins and impact of, 255– 56; results of model of, 267– 78; state attempts of, 256; topography and soil quality data for, 264 wheat: changing U.S. geographic production of, 172– 76; distribution of U.S. production of, 174– 75t; dry- farming techniques for, 178– 79; early production of, in Kansas, 178; innovations, 176– 77; Marquis, 176– 78; production of winter vs. spring, 179– 80; Red Fife, 176– 77; Turkey- type, 177; U.S. production of, 172 wheat culture, 170– 71; westward movement of, 176 willingness to pay (WTP), 48– 49, 59n7; direct impact of, 58– 59; growth rate impact of, 59– 60; as measure of demand side of policy, 60; modeling, 57– 58; modeling implication for, 67– 68; modeling results for, 60– 66; policy implications for, 68– 69