Compensation Policies within Firms : Evidence from Linked Employer-employee Data 9781846638350, 9781846638343

Citation preview

ijm cover (i).qxd

08/05/2008

11:16

Page 1

ISSN 0143-7720

Volume 29 Number 1 2008

International Journal of Manpower An interdisciplinary journal on human resources, management & labour economics

Compensation policies within firms: evidence from linked employer-employee data Guest Editors: Ana Rute Cardoso and Chiara Monfardini

www.emeraldinsight.com

International Journal of Manpower

ISSN 0143-7720 Volume 29 Number 1 2008

Compensation policies within firms: evidence from linked employer-employee data Guest Editors Ana Rute Cardoso and Chiara Monfardini

Access this journal online _______________________________

2

Editorial advisory board _________________________________

3

INTRODUCTION Compensation policies within firms: evidence from linked employer-employee data Ana Rute Cardoso and Chiara Monfardini___________________________

4

Who pays for performance? Erling Barth, Bernt Bratsberg, Torbjørn Hægeland and Oddbjørn Raaum _

8

Market power, dismissal threat, and rent sharing: the role of insider and outsider forces in wage bargaining Anabela Carneiro and Pedro Portugal_______________________________

30

Worker churning and firms’ wage policies Pedro S. Martins________________________________________________

48

The worsening of wage expectations in Italy: a study based on administrative data Elena Giarda ___________________________________________________

Access this journal electronically The current and past volumes of this journal are available at:

www.emeraldinsight.com/0143-7720.htm You can also search more than 175 additional Emerald journals in Emerald Management Xtra (www.emeraldinsight.com) See page following contents for full details of what your access includes.

64

CONTENTS

www.emeraldinsight.com/ijm.htm As a subscriber to this journal, you can benefit from instant, electronic access to this title via Emerald Management Xtra. Your access includes a variety of features that increase the value of your journal subscription.

Structured abstracts Emerald structured abstracts provide consistent, clear and informative summaries of the content of the articles, allowing faster evaluation of papers.

How to access this journal electronically

Additional complimentary services available

To benefit from electronic access to this journal, please contact [email protected] A set of login details will then be provided to you. Should you wish to access via IP, please provide these details in your e-mail. Once registration is completed, your institution will have instant access to all articles through the journal’s Table of Contents page at www.emeraldinsight.com/0143-7720.htm More information about the journal is also available at www.emeraldinsight.com/ ijm.htm

Your access includes a variety of features that add to the functionality and value of your journal subscription:

Our liberal institution-wide licence allows everyone within your institution to access your journal electronically, making your subscription more cost-effective. Our web site has been designed to provide you with a comprehensive, simple system that needs only minimum administration. Access is available via IP authentication or username and password. Emerald online training services Visit www.emeraldinsight.com/training and take an Emerald online tour to help you get the most from your subscription.

Key features of Emerald electronic journals Automatic permission to make up to 25 copies of individual articles This facility can be used for training purposes, course notes, seminars etc. This only applies to articles of which Emerald owns copyright. For further details visit www.emeraldinsight.com/ copyright Online publishing and archiving As well as current volumes of the journal, you can also gain access to past volumes on the internet via Emerald Management Xtra. You can browse or search these databases for relevant articles. Key readings This feature provides abstracts of related articles chosen by the journal editor, selected to provide readers with current awareness of interesting articles from other publications in the field. Non-article content Material in our journals such as product information, industry trends, company news, conferences, etc. is available online and can be accessed by users. Reference linking Direct links from the journal article references to abstracts of the most influential articles cited. Where possible, this link is to the full text of the article. E-mail an article Allows users to e-mail links to relevant and interesting articles to another computer for later use, reference or printing purposes.

Xtra resources and collections When you register your journal subscription online you will gain access to additional resources for Authors and Librarians, offering key information and support to subscribers. In addition, our dedicated Research, Teaching and Learning Zones provide specialist ‘‘How to guides’’, case studies and interviews and you can also access Emerald Collections, including book reviews, management interviews and key readings. E-mail alert services These services allow you to be kept up to date with the latest additions to the journal via e-mail, as soon as new material enters the database. Further information about the services available can be found at www.emeraldinsight.com/alerts Emerald Research Connections An online meeting place for the world-wide research community, offering an opportunity for researchers to present their own work and find others to participate in future projects, or simply share ideas. Register yourself or search our database of researchers at www.emeraldinsight.com/connections

Choice of access Electronic access to this journal is available via a number of channels. Our web site www.emeraldinsight.com is the recommended means of electronic access, as it provides fully searchable and value added access to the complete content of the journal. However, you can also access and search the article content of this journal through the following journal delivery services: EBSCOHost Electronic Journals Service ejournals.ebsco.com Informatics J-Gate www.j-gate.informindia.co.in Ingenta www.ingenta.com Minerva Electronic Online Services www.minerva.at OCLC FirstSearch www.oclc.org/firstsearch SilverLinker www.ovid.com SwetsWise www.swetswise.com

Emerald Customer Support For customer support and technical help contact: E-mail [email protected] Web www.emeraldinsight.com/customercharter Tel +44 (0) 1274 785278 Fax +44 (0) 1274 785201

EDITORIAL ADVISORY BOARD Professor David J. Bartholomew London School of Economics, UK Professor Derek Bosworth Manchester Business School, University of Manchester, UK Professor Martin Carnoy School of Education, Stanford University, USA Professor Morley Gunderson University of Toronto, Canada Professor Thomas J. Hyclak Lehigh University, Bethlehem, USA Professor Susan E. Jackson Rutgers University, New Jersey, USA Professor Harish C. Jain McMaster University, Canada Professor Geraint Johnes Lancaster University Management School, Lancaster University, UK Professor Meni Koslowsky Department of Psychology, Bar-Ilan University, Israel Professor Thomas Lange Auckland University of Technology, New Zealand

Professor Lord Richard Layard Centre for Economic Performance, London School of Economics, UK

Editorial advisory board

Professor John Mangan University of Queensland, Brisbane, Australia Professor Franc¸ois Rycx Department of Applied Economics (DULBEA), Free University of Brussels (ULB), Belgium

3

Professor Stephen L. Mangum Ohio State University, Ohio, USA Professor David Sapsford Chairman Economics Division, University of Liverpool, UK Professor P.J. Sloane Department of Economics, University of Wales, Swansea Professor Zhong-Ming Wang School of Management, Zhejiang University, China Professor Klaus F. Zimmerman IZA – Institute for the Study of Labor, Bonn, Germany

International Journal of Manpower Vol. 29 No. 1, 2008 p. 3 # Emerald Group Publishing Limited 0143-7720

The current issue and full text archive of this journal is available at www.emeraldinsight.com/0143-7720.htm

IJM 29,1

4

INTRODUCTION

Compensation policies within firms: evidence from linked employer-employee data Ana Rute Cardoso IZA Bonn, Bonn, Germany, and

Chiara Monfardini University of Bologna, Bologna, Italy Abstract Purpose – The purpose of this paper is to introduce this special issue on compensation policies within firms while using evidence from linked employer-employee data. Design/methodology/approach – The paper looks at the use of linked employer-employee data over time and how this has enabled progress in the understanding of the functioning of the labour market as the arena where labour supply and demand interact, under the mediation of labour market institutions and regulations. Findings – The example of issues that have been covered using linked employer-employee data, generating new insights, could be extended and it is continuously being updated. Originality/value – The articles collected in this special issue provide some fine examples of recent work on the field of linked employer-employee data. Keywords Data analysis, Compensation, Employment contracts Paper type General review

International Journal of Manpower Vol. 29 No. 1, 2008 pp. 4-7 q Emerald Group Publishing Limited 0143-7720 DOI 10.1108/01437720810861976

Within a decade, use of linked employer-employee data has enabled striking progress in our understanding of the functioning of the labour market as the arena where labour supply and demand interact, under the mediation of labour market institutions and regulations. Having in the late 1980s asked the question “Does the new generation of labor economists know more than the older generation?”, Freeman (1989, p. 319) asserted “the main conclusion I reach is that while, labor economists are more knowledgeable of labor supply issues, we do not know more about firm behavior, labor demand and the overall functioning of the markets”. Use of linked employer-employee data by the research community during the last decade has fortunately rendered Freeman’s assertion to some extent obsolete nowadays (that does not, of course, exclude the possibility that the statement became outdated exactly because it had an impact mentoring the research that meanwhile developed). Use of linked employer-employee data has indeed led to successive accomplishments towards the identification of the microeconomic underpinnings of income distribution, unemployment, and growth. Initial studies relied on cross-sectional data and one of the first challenges came from the puzzling rise in wage inequality that took place in several countries during the 1980s and 1990s. Linked employer-employee data enabled a shift in the emphasis of the analysis from the worker attributes – namely wage differences across schooling levels, or the gender

wage gap – to the role of employers and their wage policies. Similarly, unemployment could be analysed as the outcome, not strictly of certain worker attributes – schooling, age and gender, for example – but also of employers’ policies and characteristics. Longitudinal-linked employer-employee data tracking firms and/or workers over time enabled a jump into the study of dynamics, to look at a whole new set of issues: the nature of adjustment of employment and wages at the micro level to shocks such as technological progress, changing patterns of trade and growing internationalization of the economies; the role of labour market regulations in such adjustment; job and worker flows at the firm level as determinants of macro fluctuations and changes in employment composition; labour demand and the substitutability across factors of production; worker careers, in terms of wage and job changes; micro determinants of the functional distribution of income; retirement decisions in a framework of ageing population in developed countries; vacancies, job search and matching in the labour market; the effectiveness of matching policies; the impact of firm fortunes (plant closures, in particular) on workers wages, the occurrence and duration of unemployment and successive employment spells; the impact of minimum wage policies. Another challenge was to open the black box to look inside the firm. The analysis of human resources management policies was enriched by the availability of comparable data on a large number of firms, replacing case studies of a few firms. New trends in company policies have been analysed, concerning in particular hiring, firing, promotions, wages and payment schemes, training, technology adoption, and how they impact on profitability, on one hand, and on workers’ careers and worker flows, on the other. The example of issues that have been covered using linked employer-employee data, generating new insights, could be extended and it is continuously being updated. The articles collected in this special issue provide some fine examples of recent work on the field. The first two contributions in this volume provide new theoretical insights and empirical evidence on the wage formation mechanism in different European labour markets. The paper by Barth, Bratsberg, Hægeland and Raaum proposes an agency model to explain the firms’ choice between fixed and performance-related pay schemes and test its main predictions with two repeated cross sections of Norwegian establishment survey data. Consistent with the theoretical model, they find that performance pay schemes are more widespread in firms where employees have a higher autonomy in defining their working tasks. On the other hand, this compensating scheme is found to be less likely in small firms, and in presence of both centralized and local collective bargaining. Thanks to the two periods of observation available, the authors are also able to show that, net of changes in a set of labour market characteristics like industry structure and bargaining institutions, performance-related pay experienced a positive trend in Norway in the recent years. With a similar research approach Carneiro and Portugal test empirically, in the second paper of our collection, the prediction of the insider-outsider theory that firms with high labour turnover costs pay higher wages, due to the increased negotiation power gained by insider workers. The collective bargaining model proposed describes how firm’s wages are shaped by firm-specific characteristics, outside factors and insiders’ bargaining power. The empirical investigation is performed on a panel

Introduction

5

IJM 29,1

6

dataset of large Portuguese firms. Even though this market is characterized by high level of centralized negotiation, the estimated insider weight turns out to be similar to that of economies with more decentralized bargaining. New empirical evidence is found on the role of labour adjustment costs, showing that higher risks of being laid off weaken insiders’ power and results in lower wages. The insiders’ power effect is also found to be asymmetric, revealing downward wage rigidity. The paper by Martins looks at the relationship between turnover costs and wages from a personnel economics perspective, proposing an instrumental variable method to identify the causal impact of wage policies on the amount of worker churning experienced by the firm. The general identification strategy relies on estimating the impact of wages using the exogenous source of variation represented by the share of workers paid by collective bargaining contracts. The estimation is performed on a matched employer-employee panel dataset collecting the whole population of firms in Portugal. The idea that wages are a choice variable and therefore endogenous is supported by the results of the author. Adopting the instrumental variable method the estimated impact on churning turns from negative to non-negative. A possible interpretation of this finding it that workers’ effort might be weakly responsive to wages and firms consequently replace priced-out workers with more skilled new hires. Wages are back to the role of dependent variable in the final paper in this issue, although at the individual rather than at the firms’ level. Giarda’s study analyses jointly the role played by age, supply effect and macroeconomic conditions on the wage determination process, using a large panel of Italian wages administrative data. The rather long time span covered by the dataset makes it possible to derive robust evidence on the worsening of lifetime wage expectations of younger generations. They benefit from higher entry wages but face flatter wage-age profile. A supply factor such as the relative size of the active population by age group is found to be negatively correlated with wages. Turning to the conditioning factors of the macro-economic environment in which they are determined, wages are found to be negatively affected by regional unemployment rate, but positively influenced by real union wages. This collection of papers illustrates the potential from use of linked employer-employee data. Nevertheless, the current limitations to data access that still frequently hamper the progress of empirical research should also be mentioned. While the need for micro data has been identified, the discussion on the feasibility and conditions for granting wider access to these data by the scientific community is still evolving slowly. Ethical issues involved in accessing and analyzing micro data (such as confidentiality or restriction of data use to specific scientific projects) are a major concern of data producers and often turn into a source of miscommunication with the scientific community, impeding or delaying data access. Hopefully, the quality of the scientific results generated on linked employer-employee data and the respect for the ethical issues involved will prompt wider collection and dissemination of this type of data. We would like to conclude this introduction thanking Adrian Ziderman for his invitation to collaborate in the editorial process of this volume and his support throughout its preparation. The papers in this special issue have been examined through a peer review system. We are grateful to all the reviewers for their comments and suggestions that helped the authors to improve their early versions.

Reference Freeman, R.B. (1989), “Does the new generation of labor economists know more than the older generation?”, in Freeman, R.B. (Ed.), Labour Markets in Action: Essays in Empirical Economics, Harvester Wheatsheaf, New York, NY, pp. 317-42. About the Guest Editors Ana Rute Cardoso is Senior Research Associate at the Institute for the Study of Labor (IZA Bonn), where she is Deputy-Director of the Research Program “The Future of Labor”. She received her PhD in economics in 1997 at the European University Institute (Florence). She is currently on leave from the University of Minho, Portugal. She is an elected member of the executive committee of the European Association of Labour Economists (EALE). Ana’s research interests include earnings dispersion and mobility, employer behaviour and the impact of labour market institutions. Ana Rute Cardoso is the corresponding author and can be contacted at: [email protected] Chiara Monfardini is Associate Professor of Econometrics at the Faculty of Economics of the University of Bologna since 2005. She obtained her PhD in Economics at the European University Institute (Florence). She is member of the Center for Household, Income, Labour and Demographic Economics (CHILD) and of the Health, Econometrics and Data Group (HEDG). Her research and teaching field is microeconometrics, with particular focus on limited dependent variable models.

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints

Introduction

7

The current issue and full text archive of this journal is available at www.emeraldinsight.com/0143-7720.htm

IJM 29,1

Who pays for performance? Erling Barth Institute for Social Research, University of Oslo, Oslo, Norway and IZA, Bonn, Germany

8

Bernt Bratsberg Frisch Centre for Economic Research, University of Oslo, Oslo, Norway and Kansas State University, Manhattan, Kansas, USA

Torbjørn Hægeland Statistics Norway and Frisch Centre for Economic Research, University of Oslo, Oslo, Norway, and

Oddbjørn Raaum Frisch Centre for Economic Research, University of Oslo, Oslo, Norway Abstract Purpose – The purpose of this paper is to improve our understanding of why some firms tie compensation to worker performance as well as the variation in type of performance pay system across firms. Design/methodology/approach – The study first presents a theoretical framework that motivates n empirical study of performance-related pay. The data are based on Norwegian establishment surveys from 1997 and 2003. The empirical analysis addresses determinants of adoption of performance pay systems. Findings – Performance-related pay is more prevalent in firms where workers of the main occupation have a high degree of autonomy in how to organise their work. Performance pay is also more widespread in large firms, but is less common in highly unionised firms and in firms where wages are determined through centralised bargaining. Results show that performance pay is on the rise in Norway, even after accounting for changes in industry structure, bargaining regime, and union density. Finally, it is found that the incidence of performance-related pay relates positively to product-market competition and foreign ownership. Originality/value – The paper provides new empirical evidence on the use of performance-related pay. The results support an interpretation of incentive pay as motivated by agency problems, and provide new evidence on the relationship between payment schemes and institutions such as unions and bargaining framework. Keywords Performance related pay, Compensation, Payments, Profits, Norway Paper type Research paper

1. Introduction Why do different firms choose different pay schemes? Following the seminal work by Holmstro¨m and Milgrom (1987), agency problems are typically cited as the explanation why some firms tie compensation to performance. Consider, for example, the textbook case of Lazear (1995), where output depends on both worker effort and some stochastic International Journal of Manpower Vol. 29 No. 1, 2008 pp. 8-29 q Emerald Group Publishing Limited 0143-7720 DOI 10.1108/01437720810861985

This research has received financial support from the Norwegian Research Council, grant no. 150666/510. The authors are grateful to John Dagsvik, Kristine Nergaard, Hege Torp, the Guest Editors of the special issue, and an anonymous referee for helpful comments.

factor. When it is costly or impossible to directly observe effort and sort out the influence of the stochastic factor, the firm may benefit from implementing an incentive pay scheme in order to motivate workers to supply effort. If workers are risk averse, however, the uncertainty associated with the stochastic factor will reduce the merits of incentive-based schemes as more uncertainty imposes a greater risk on workers. This observation has motivated a substantial body of empirical studies that examine whether or not there is a trade-off between risk and use of incentive schemes (see, e.g. the summary in Prendergast, 1999). As emphasised by Prendergast (2002), these studies have by and large not had much success in finding evidence of such a trade-off. Prendergast argues that the lack of clear empirical evidence stems from a failure of the literature to recognise the association between uncertainty and allocation of responsibility. In uncertain settings, firms seek to delegate responsibility to workers. In turn, when responsibility is delegated, firms use incentive pay schemes to constrain worker discretion. This gives rise to a second, and positive, effect of uncertainty on the use of incentives. A prediction is that output-based incentive pay schemes are more likely to be observed when there is considerable employee discretion over work tasks. In this paper, we investigate the relationship between worker discretion over tasks and the use of performance-related pay. We first develop a simple theoretical framework, focusing on the firm’s choice between a fixed pay system where the firm monitors worker effort, and a remuneration scheme with a variable pay component that is proportional to observed individual output. High monitoring costs will induce the firm to transfer authority to its employees and permit worker discretion over what tasks to spend time on. In this case, pay for performance is the optimal remuneration scheme. As in Prendergast’s model, an important empirical implication of the framework is that performance-related pay is more likely to be used when worker autonomy over tasks is high. In the empirical analyses, we use data from two Norwegian establishment surveys, from 1997 and 2003, to test the hypothesis of a positive relationship between autonomy of the main occupational group in terms of defining work tasks and the incidence of performance-related pay. Salas-Fumas (1993) provides an early analysis of the relationship between incentives and supervision with respect to compensation of managers. Using 1998 WERS data, Belfield and Marsden (2003) investigate the relationship between performance pay, monitoring environments, and establishment performance. They argue that it is the combination of pay systems and monitoring environments that drives organisational outcomes. A recent study of performance pay by Foss and Laursen (2005), using Danish establishment data, finds evidence of a positive relationship between delegation and environmental uncertainty. In the present paper, we move on to investigate the relationship between allocation of responsibility and performance-related pay. We also analyse to what extent worker autonomy is associated with different types of performance pay, such as traditional piece rates, profit sharing and group bonuses, and new forms of individual performance-related pay. In many European countries, including Norway, wage setting has traditionally been dominated by negotiations between worker unions and employer associations. A fixed hourly wage has been the predominant type of pay. Internationally, the empirical literature displays some divergence with respect to the relationship between unionism and the incidence of performance-related pay. While Brown (1990) and Heywood et al.

Who pays for performance?

9

IJM 29,1

10

(1997) find less use of performance-related pay in unionised establishments, Booth and Frank (1999) conclude that union status increases coverage of performance pay. Collective bargaining and union influences over decisions may affect the firm’s motives for using performance-related pay in several ways. First, if some expectation regarding worker effort is part of the bargaining settlement, unions may reduce monitoring costs simply because it is easier to enforce effort rules using the trade union as a self-disciplining device. Second, union bargaining over wages may act as a rent-sharing device, and thus reduce the motive to provide other high-powered incentives. Third, unions may be expected to oppose performance-related pay schemes if measurement of output is in part left to management’s discretion. Unions are likely more supportive of well-defined, and easily measured, piece rates, than of merit pay based on individual assessments using, perhaps, subjective criteria. In our empirical analyses, we therefore distinguish between bargaining levels in order to sort out the effects of bargaining regime and unionism on performance pay. As observed by Brown (1990), Ortin-Angel and Salas-Fumas (1998), and Parent (2002), among others, there are substantial differences in the use of performance-related pay across industries, institutional settings, and other firm characteristics. In an international comparison, Brown and Heywood (2002, p. 261) find that “combinations of performance pay methods differ by country, and the recent emphasis and growth of such methods is far from uniform”. In the empirical analyses, we check whether any trend in the incidence of performance pay in the Norwegian data can be due to changes in industry structure and bargaining institutions by including industry as well as bargaining level and union density at the establishment as explanatory variables in the empirical model. Two other underlying developments may add to the explanation of trends in use of performance-related pay systems. One development is increased product-market competition, arising both from international integration as well as from deregulation policies. Increased competition in the product market is likely to yield greater uncertainty for the firm, which according to the Prendergast model will trigger more delegation of tasks within the firm and thus greater reliance on performance pay. Increased competition may also create a stronger relationship between effort and profits (Schmidt, 1997; Raith, 2003). In line with this argument, Cun˜at and Guadalupe (2005) find stronger performance sensitivity of executive pay with higher product-market competition. The other development is skill-biased technological change, which adds to the knowledge intensity of production. Brown (1990) argues that in high-skilled jobs, worker output is more sensitive to differences in worker quality compared to jobs requiring less skill. A similar argument applies to effort. Effort-sensitive jobs are more likely to benefit from performance-related pay, particularly in settings where the choice between work tasks is delegated to workers. We thus include measures of product-market competition and the level of human capital at the establishment in the empirical analyses. We also investigate the association between foreign ownership and performance-related pay in order to test the notion that increased globalisation and imported management practices may have boosted the incidence of performance pay in Norwegian establishments. A significant, although not very large, literature has investigated the relationship between performance-related pay and various measures of establishment performance.

Several papers report from case studies of particular firms (see, e.g. Lazear, 2000; Bandiera et al., 2005), but there are also examples of studies using representative samples of workers, such as Booth and Frank (1999) using BHPS for the UK, and Parent (2002) providing evidence for the USA based on the NLSY. Typically, studies find a positive effect of incentive schemes on firm performance indicators such as wages and productivity. In this study we do not aim at assessing the effect of performance-related pay on establishment performance, but rather at testing hypotheses related to the agency model of the choice of method of pay. It is worth noting that a positive relationship between performance-related pay and performance indicators is consistent with both the agency model of Holmstro¨m and Milgrom (1987) and the selection model of Lazear (1995, 2002). In our view, the agency and selection models do not represent competing explanations of performance pay, but rather separate mechanisms that are likely to be present in the labour market at the same time. Evidence in favour of one of these models cannot be used as evidence against the other. While we provide a test of the agency model, our data do not permit a good test of the merits of the selection model. In the next section we present a simple theoretical model for the firm’s choice between fixed and performance-related pay schemes. Section 3 presents our data, while results are reported in Section 4. The final section concludes. 2. Theoretical background We present a simple theoretical framework as a basis for the discussion of why pay systems differ across firms. By relating compensation to an output-based performance measure, the firm gives workers incentives to supply effort. When the performance measure is subject to shocks, the firm has to compensate risk-averse workers. Our starting point is a simple setting along the lines of Lazear (1995, Ch. 2). The firm chooses one of two pay systems. With performance-related pay, the remuneration of a worker consists of fixed component and a share of firm revenues, as in the Holmstro¨m and Milgrom (1987) model. With a fixed-pay system, the total pay is independent of revenues (i.e. the worker share is zero). Effort is unobservable unless the firm implements a costly monitoring technology. Output is assumed to be observable. With performance-related pay, the firm exploits the incentives embedded in revenue sharing to raise effort, while monitoring is used under fixed pay to ensure that the worker supplies a given level of effort. Our focus is on the firm’s choice: which pay scheme –fixed pay (FP) or performance-related pay (PRP) – maximises expected profits? Technology and market conditions are the simplest possible, with worker i’s contribution to revenues equal to the value of her observable skills (ai), effort (ei), and the outcome of a random event (1i);
 yi ¼ ai þ ei þ 1i ; 1i , N 0; s 2 : ð1Þ With PRP, workers are paid a fixed wage, wi, and a “bonus.” To reward effort, the performance-related bonus is set proportional to the worker’s observed contribution to revenues (net of the observable skill component, ai), i.e. the bonus equals bðei þ 1i Þ. The firm cannot, without costs, distinguish between effort and (bad) luck. Instead of PRP, the firm may choose FP and invest in some monitoring technology to verify that workers supply a desired level of effort, e
. 0. To simplify the exposition,

Who pays for performance?

11

IJM 29,1

we assume that this effort level is the same for all workers in the firm. Monitoring costs, M, are given by: M ¼ M ðe
Þ ¼ nle
;

12

ð2Þ

where n is the number of workers and l . 0. Higher effort requires more intensive monitoring and l reflects the marginal monitoring cost per worker. Ex post worker utility is given by:    U i ¼ 2exp 2a wi þ bðei þ 1i Þ 2 cðei Þ ; ð3Þ where a reflects the level of worker risk aversion, wi denotes the fixed wage component, b $ 0, and costs of supplying effort in money terms are given by: cðei Þ ¼

e2i ; pi . 0: 2pi

ð4Þ

When 1i is drawn from a normal distribution, expected utility is given by:   1 E ðU i Þ ¼ 2exp 2aFi ; where Fi ¼ wi þ bei 2 cðei Þ 2 b 2 as 2 : 2

ð5Þ

Effort costs may be influenced by both job characteristics and individual talent. The parameter 1/pi is the slope of the marginal cost function of supplying effort. A high pi may reflect talent or ability, implying that additional revenue requires little extra effort on part of the worker. Effort costs (or, rather, the value of pi) can also be determined by the particular task or job to be done. For simplicity, we will ignore worker heterogeneity and assume that effort costs are the same for all workers within a given firm. (Hence we drop the subscript in the following.) These costs may, however, differ across firms according to the type of production. Some firms have tasks where workers easily (i.e. high p) can increase output through extra effort (e.g., by reducing duration of breaks, work longer hours, do extra work at home, etc.). Other firms have jobs with less scope to do so. With PRP, the optimal effort (e *) is chosen independently by each worker and determined by equality between marginal return and marginal cost of effort, i.e. b ¼ e * =p. To retain workers in the firm, total pay must match opportunities elsewhere. Again ignoring worker heterogeneity, we write the outside option, X, for an individual worker as given by: 1 X ¼ g p þ a; g $ 0: 2

ð6Þ

When effort costs reflect ability, the parameter g captures that the outside option is more favorable for more productive workers. When p reflects job characteristics, g will be low since these are specific to the actual worker-firm match. Observable skills, a, also affect outside options. According to equation (6), the outside option is thus assumed to be increasing in both “effort productivity” and observable skills. We consider a profit-maximising firm that determines its wage policy by comparing the two alternatives. With performance-related pay, the firm decides on the fixed wage component and the share parameter. The share parameter is set to give the correct

incentives for workers to provide effort and the fixed wage component is set to match outside options. With a fixed pay system, the firm invests in a monitoring technology, sets an optimal “effort standard,” and fixes the wage level to ensure that worker utility matches that of the outside option. Performance-related pay With PRP, the firm’s expected profits are given by:     E PPRP ¼ n eð1 2 bÞ þ a 2 nw;

13 ð7Þ

which the firm maximises with respect to b and w, subject to: F ¼ X ðindividual outside optionÞ

ð8aÞ

e * ¼ bp ðindividual optimal effortÞ:

ð8bÞ

It is straightforward to show that, with PRP, the optimal wage policy will be given by: 0 , b* ¼

p , 1; p þ as 2

ð9Þ

similar to Holmstro¨m and Milgrom (1987) and Lazear (1995), and that compensation becomes: W PRP ¼ w þ b * e þ b * 1 ¼

 b *2
p þ as 2 þ b * 1 þ X: 2

ð10Þ

The optimal share parameter, b *, is decreasing in a (degree of risk aversion), s 2 (variance of random shocks that separate effort from observed production), and 1/p (slope of marginal effort costs). The worker receives her outside option, a share of the random event, and is compensated for the risk associated with PRP as well as the (optimal) effort supplied. The expected firm profits are then given by: h i n
 b * 2 g p: ð11Þ E PPRP ¼ 2 Fixed pay With FP, the expected profits of the firm are given by:   E PFP ¼ nðe
þ aÞ 2 nw 2 nle


ð12Þ

which are maximised with respect to e
and W, subject to: F ¼ X ðindividual outside optionÞ:

Who pays for performance?

ð13Þ

It follows directly from the first-order conditions that the optimal common effort level is determined by:

IJM 29,1

e
¼ 1 2 l; p

ð14Þ

where the marginal effect of increased effort on revenues net of monitoring costs (i.e. 1 2 l) is equal to the marginal cost of supplying effort (
e=p). Pay is given by the fixed wage, determined by the outside option constraint (F ¼ X):

14 W FP ¼

e
2 þ X: 2p

ð15Þ

The FP wage is the sum of the outside option and a compensation for the effort costs associated with the common effort level. The firm’s expected profits are given by:   n  E PFP ¼ ð1 2 lÞ2 2g p: ð16Þ 2 The optimal wage policy Comparing the two alternative pay regimes, it is straightforward to show that:     ð17Þ E PPRP . E PFP , b * . ð1 2 lÞ2 or e * . ð1 2 lÞ
e: Profits under PRP are higher if and only if the optimal effort supplied individually by workers is higher than the optimal common effort level, net of monitoring costs, set by the firm in the FP regime. It follows that there exists a critical value of marginal ~ This critical value is monitoring cost, l~ . 0, where the firm chooses PRP when l . l. determined by risk aversion, effort costs, and the dispersion of productivity shocks: rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi p l~ ¼ 1 2 : ð18Þ p þ as 2 Note that the choice of pay system is independent of outside options (a and g). The predictions of the model can be summarised as follows. According to equation (18), PRP is more likely when: . the marginal cost function of effort is flat ( p is high); . marginal monitoring costs per worker are high (l is high)[1]; . worker risk aversion is low (a is low ¼. b* is large ¼. e* is high); and . there is little noise in the output signal (s 2 ! 0 ) b * ! 1 ) e * is high). We have no direct empirical counterparts to the parameters in the theoretical model, but the theory predicts several patterns to be expected in the data. In firms where employees perform their tasks autonomously, monitoring costs are likely to be high and the prevalence of performance-related pay is high. Individual pay for performance is more likely when output is highly sensitive to variations in effort. In light of our model, where revenues equal efforts (plus shocks and observable skills), high sensitivity with respect to effort can be interpreted in terms of low effort costs (a high p), as an increase in effort costs will be associated with a large increment in revenues. If the productivity of a high-skilled worker is particularly sensitive to effort, we would

expect a greater propensity of performance pay in firms with many high-skilled workers. High-skilled employees typically perform individual or autonomous tasks that add to the attractiveness of a performance-pay scheme. We extend this discussion about theoretical predictions when we present our empirical results in section 4. 3. Data sources, samples, and variable construction The core of our data material consists of the Norwegian Flexibility Survey from 1997 and the Norwegian Work and Establishment Survey from 2003. Both surveys were carried out as computer assisted telephone interviews with either the manager or the chief personnel officer of the establishment. In both surveys, random, but stratified (with respect to establishment size, age and sector), samples were drawn from the population of Norwegian establishments with more than ten employees. The survey instruments included questions concerning standard establishment characteristics, their main products and markets, employees, recruitment and training practices, use of external labour, compensation policies and wage determination, employer-employee cooperation, etc. Questions concerning employees typically related to the “main occupational group” at the establishment[2]. In addition, the survey data were matched with detailed data about the establishment and all its employees taken from various administrative registers. The register data are annual and cover the period 1995-2003. The response rates of the surveys were 76 per cent in 1997 and 77 per cent in 2003. The net samples consist of 2,130 establishments in 1997 and 2,358 in 2003. Of these, 1,154 establishments are represented in both surveys. In the present study, we focus on the private sector. This leaves us with 1,556 establishments with valid data on key variables in 1997 and 1,426 in 2003. Of these, 775 establishments are represented in both surveys. Performance-related pay Both establishment surveys contained questions about performance-related pay. Unfortunately, these questions were not identical in the two surveys. In 1997, respondents were asked whether or not “the main occupational group receives any pay through incentive pay systems, bonuses, or profit sharing?” In 2003, the survey instrument instead included separate questions about six different forms of performance-related pay: (1) A: individual and group piece-rates. (2) B: commissions. (3) C: group bonuses. (4) D: profit sharing. (5) E: individual bonuses. (6) F: individual performance assessments. Respondents were also asked to estimate the share of total wages associated with each type of performance pay. It seems reasonable to assume that respondents who in 2003 answered affirmative on the use of at least one the five former pay types (A-E) would have answered “yes” to the 1997 question. It is not obvious, however, how establishments with type F,

Who pays for performance?

15

IJM 29,1

16

“individual performance assessments,” would have interpreted the 1997 question. In addition, it is not clear whether the answers refer to permanent or variable elements of compensation. In the empirical analyses, we therefore use three alternative definitions of performance-related pay in 2003: (1) Strict definition. Answered “yes” on at least one of the types A-E. (2) Medium definition. Answered “yes” on at least one of the types A-F. If “yes” on F only, its share of total wages must be at least 3 per cent. (3) Wide definition. Answered “yes” on at least one of the types A-F. In the next section, we also report results from analyses based on 2003 data where we distinguish between different types of performance pay, classifying types A and B as “Traditional schemes,” C and D as “Group-based schemes,” and E and F as “Individual-based schemes.” Other important firm characteristics Among other questions, managers were asked to what extent (very large, large, some, or no) employees are free to organise their own work. If the answer is large or very large, we classify the establishment as having a high degree of employee autonomy (dummy variable). The exact wording of the response categories of the autonomy question was, however, not completely identical in the two surveys. We define the establishment as an export establishment if the manager reports their main product market to be outside of Norway. Similarly, the establishment is defined as being exposed to high product market competition if the manager states that the degree of competition is “very large” or “quite large,” as opposed to “quite small” or “very small.” We also use information from the manager interview about wage determination at the establishment; whether or not workers in the main occupational group are covered by individual or collective agreements, and whether or not collective agreements are negotiated at the central or local level, or both. We collect information about the union density at the establishment from the manager surveys. If not available in the survey data, we computed densities from data on individual payments of union membership dues identified through registers and aggregated to establishment level. It should be noted that information on foreign ownership is not available in 1997. In the estimations, we therefore impute the 1997 value using 2003 data for the establishments that are observed both years. For the other establishments, we set the variable to zero, and include a dummy variable indicating that information on foreign ownership is missing. Our sample is restricted to the private sector. Because of the reorganisation of former government monopolies, establishments within postal services and the national telecommunications company (Telenor) were classified as belonging to the public sector in 1997 and to the private sector in 2003. Table I shows summary statistics for our sample, separately by year and by use of performance-related pay. Except for workforce characteristics and union density, all variables are dummy variables; hence the numbers reflect the share of establishment observations with this characteristic. The table shows that the share of firms with performance-related pay is around 43 per cent in the 1997 sample. In 2003, the share is 46, 55, or 61 per cent, depending on how we define performance-related pay. The

0.7185 0.1909 0.8824 N/A 0.2012 0.1980 0.3650 0.1930 0.2365 0.5360 0.2275 0.5070 0.0212 0.1887 0.1497 0.0733 0.1317 0.1703 0.0630 0 0.0315 0.1041 0.0386 0.0289 0 1,556

0.4274

0.3560

0.1923 0.2667 0.2324

SDb 0.4642 0.5533 0.6115 0.5891 0.1971 0.8219 0.2454 0.1879 0.2129 0.3607 0.1919 0.1732 0.6017 0.2251 0.5507 0.0372 0.1585 0.1417 0.0673 0.0968 0.1438 0.0659 0.0344 0.0323 0.1262 0.0477 0.0484 1 1,426

n

2003 (2)

0.3738

0.2256 0.2678 0.2373

SDb

0.6348 0.1675 0.8109 0.1837 0.2225 0.1927 0.3896 0.2226 0.1492 0.5661 0.2847 0.5735 0.0190 0.1957 0.1420 0.0445 0.0753 0.1747 0.0897 0.0229 0.0295 0.0818 0.0700 0.0550 0.4169 1,528 0.3624

0.2005 0.2781 0.2451

Establishments without performance pay (3) n SDb

0.6795 0.2215 0.8982 0.2953 0.1657 0.2182 0.3350 0.1608 0.2662 0.5688 0.1651 0.4800 0.0392 0.1506 0.1499 0.0977 0.1568 0.1396 0.0378 0.0096 0.0344 0.1492 0.0144 0.0206 0.5426 1,454

0.3726

0.2168 0.2522 0.2189

Establishments with performance pay (4) n SDb

Note: a Foreign ownership not available in 1997 sample; means in columns (3) and (4) refer to 2003 sample; b For continuous variables. In (3) and (4), establishments are classified according to the medium 2003 definition of performance-related pay

Performance pay Strict definition 2003 Medium definition 2003 Wide definition 2003 Autonomy Exports High competition Foreign ownership a Fewer than 20 employees Share college Share females Share part-time Individual bargaining (omitted) Local union bargaining Central union bargaining Union density Oil, mining, energy Non-durables (omitted) Durables Construction Wholesale Retail, hotels, restaurants Transportation Post and telecom Finance and real estate Business services Health and social services Education, personal service 2003 observation Observations

n

1997 (1)

Who pays for performance?

17

Table I. Sample descriptive statistics, by year and use of performance-related pay

IJM 29,1

18

fraction of establishments with high employee autonomy is lower in 2003 than in 1997. This may reflect differences in wording of the question in the two surveys. What is clear from the table is that worker autonomy is more prevalent among establishments with performance-related pay. Establishments with performance-related pay tend to be larger, have higher shares of college-educated workers, and have lower shares of female and part-time workers. Interestingly, union density and the incidence of local bargaining is higher in the 2003 sample than in the 1997 sample. Firms with performance pay have lower union densities and are less likely to set wages through centralised bargaining only. In Figure 1, we display the sample proportions of performance-related pay for each bargaining regime, separately by year. The figure shows the same pattern across bargaining regimes as in the table, with less performance pay the more centralised bargaining. Importantly, the figure also illustrates that the use of performance-related pay increased between 1997 and 2003, regardless of the type of wage-setting regime[3]. 4. Empirical results Changes in the use of performance pay We begin the empirical analysis with a closer examination of trends in performance pay over the sample period. A first look at the data indicates that the use of performance-related pay in the private sector of Norway increased from 1997 to 2003. Table II, panel A, shows that this conclusion holds regardless of which definition of performance-related pay we use in the 2003 data. Using the strict definition, the increase is 3.7 percentage points. Using the medium or wide definition, the increase is 12.6 or 19.4 percentage points, respectively. As was evident in Table I and Figure 1, however, there are large differences in the use of performance-related pay across industries and wage bargaining regimes. Changes over time in industry structure and wage bargaining regimes might therefore explain the observed changes in use of performance-related pay. To address this issue, we also include industry dummies and information on wage bargaining regimes in the

Figure 1. Performance pay by bargaining regime

0.0182

0.0191 0.0298 0.0314 0.0323 0.0586 0.0328 0.0388 0.0365 0.0361 0.0414 0.0617 0.0568 0.0374 0.0345 0.0449

0.0369 * * 0.0010 22,046.8 4.1 0.0431 0.0686 * * * 20.0577 * 20.1469 * * * 20.0858 * * * 0.1041 * 0.0402 0.2453 * * * 0.2062 * * * 0.0097 20.1287 * * * 20.2490 * * * 0.1127 * * 0.1556 * * * 20.3332 * * * 20.2219 * * * 0.0834 21,878.0 341.7 0.0000

Strict definition Standard errors

0.1615 * * * 20.0614 * * 20.1813 * * * 20.0950 * * * 0.2217 * * * 0.0632 * 0.2294 * * * 0.1889 * * * 0.0286 20.1544 * * * 20.1895 * * 0.1141 * * 0.1626 * * * 20.2979 * * * 20.2073 * * * 0.0919 21,876.2 379.7 0.0000

0.1259 * * * 0.0115 22,042.4 47.3 0.0000 0.0191 0.0304 0.0321 0.0327 0.0534 0.0328 0.0378 0.0359 0.0365 0.0426 0.0699 0.0563 0.0367 0.0420 0.0489

0.0182

Medium definition n Standard errors 0.0180

0.0189 0.0307 0.0325 0.0329 0.0524 0.0327 0.0375 0.0356 0.0365 0.0434 0.0743 0.0552 0.0362 0.0449 0.0530

0.2202 * * * 20.0653 * * 20.2100 * * * 20.1012 * * * 0.2134 * * * 0.0632 * 0.2058 * * * 0.1735 * * * 0.0341 20.1670 * * * 20.1176 0.1337 * * 0.1614 * * * 20.2893 * * * 20.1430 * * * 0.1016 21,855.7 419.7 0.0000

Wide definition Standard errors

0.1841 * * * 0.0246 22,014.7 101.6 0.0000

n

Notes: There are 2,982 observations. The table lists estimated marginal effects on the probability of performance pay. Models are estimated using the “dprobit” command in Stata 9 (StataCorp, 2005). For continuous variables, the marginal effect is evaluated at the mean values of explanatory variables. For dummy variables, the marginal effect is computed as the increment in probability from a discrete change, holding other explanatory variables constant at their mean value. Reference groups are non-union bargaining and non-durables manufacturing; * Statistically significant at the 0.10 level; * * at the 0.05 level; * * * at the 0.01 level

A. Specification 1 (only 2003 dummy) 2003 observation Pseudo-R 2 Log likelihood Chi-squared (1) p-value B. Specification 2 2003 observation Local bargaining Central bargaining only Union density Oil, mining, energy Durables Construction Wholesale Retail, hotels, restaurants Transportation Post and telecom Finance and real estate Business services Health and social services Education, personal services Pseudo-R 2 Log likelihood Chi-squared (15) p-value

n

Who pays for performance?

19

Table II. Incidence of performance-related pay 1997 and 2003, pooled cross-section probit regressions

IJM 29,1

20

probit regressions (see Table II, panel B). Controlling for such factors, we find that the increase in the use of performance-related pay is even stronger than what the changes in unconditional averages tell us. As in panel A, the estimated change in the use of performance-related pay from 1997 to 2003 depends on which definition we use in the 2003 data. The estimated increase is 6.9, 16.2 or 22.0 per cent, if we use the strict, medium or wide definition, respectively. These empirical patterns imply that changes in industry structure and wage bargaining regimes actually contributed to a decline in the use of performance-related pay in the period from 1997 to 2003. As Table I revealed, union density increased over the sample period. There has also been and an increase in collective agreements with local bargaining at the expense of regimes without collective agreements. Using the numbers for 1997 and 2003 from Table I and the coefficients for the medium definition in Table II, we find that changes in unionisation and wage bargaining regimes contributed to a decline of 0.8 percentage point in the period. Similarly, changes in industry structure contributed to a decline of 1.7 percentage points. The impact of bargaining regime on the incidence of performance-related pay appears substantial. Establishments with central bargaining only are less likely to have performance-related pay; using the middle definition, the probability of performance pay is 21 percentage points lower than in establishments with individual agreements only. In establishments where there is local collective bargaining, the probability is around six per cent lower than in firms without any collective agreement. Even conditional on wage bargaining regime, the use of performance-related pay is lower in establishments with a high share of unionised employees. Using the medium definition, an increase in the union membership rate of 50 percentage points reduces the probability of performance-related pay by 4.8 percentage points. There are also significant differences in the use of performance-related pay across industries. Construction, wholesale trade, oil, mining and energy, and business services are the industries where performance pay is most prevalent. Private-sector health services, education, transportation, and post and telecommunications have the smallest incidences. However, the picture varies somewhat with respect to definition of performance-related pay. For example, the oil, mining and energy industry appears to have relatively more performance pay if we apply the medium or wide definition rather than the narrow definition. Thus, individual performance assessments appear to be an important form of performance pay in the oil industry. The same applies to the post and telecommunications industry. The indication is that there may be substantial differences across industries, not only with respect to the prevalence of performance pay, but also what type of performance pay they use. We return to this issue towards the end of this section. Determinants of performance-related pay Having established that there has been an increase in the use of performance-related pay in recent years in Norway, even within industries and wage-bargaining regimes, we now turn to the determinants of use of performance-related pay. A clear prediction from the theoretical framework is that when it is costly to observe worker effort and workers have autonomy over tasks, establishments are, all else equal, more likely to choose performance-related pay. We extend the model specification from Table II by adding further establishment characteristics to the list of explanatory variables.

Because we now are concerned with the statistical strength of relationships between firm characteristics and performance pay, we use a random-effects probit model to account for the fact that an establishment may contribute two observations to the pooled sample. The random effect will capture any serial correlation of error terms from the same establishment. Estimation is based on the “xtprobit” command in Stata 9; see StataCorp (2005) for details on the likelihood function and numerical algorithm. Table III presents separate estimation results for the three alternative definitions of performance-related pay. Consistent with the theoretical model, we find that establishments where employees have a high degree of autonomy in organising their own work are significantly more likely to have performance-related pay. In firms with a high degree of worker autonomy, it may be more costly to monitor worker effort; hence they are more likely to use performance pay. The difference in probabilities of performance pay between firms with “high autonomy” and “low autonomy” ranges from 4.7 to 6.9 percentage points, depending on the exact 2003 definition of performance-related pay. Product market conditions appear to be important for the choice between fixed or performance-related pay. Firms that face high competition in their product markets or export their main product have significantly higher incidences of performance-related pay than other firms. Firms that are exposed to competition in the product market may need to have a stronger focus on productivity than firms with market power. This may be an explanation of why performance pay is more common in such firms. Foreign ownership is positively related to the use of performance pay, even after controlling for bargaining regime as well as product-market competition and production for export markets. The estimated effect is strongest when we use the strict definition (11.3 percentage points) and smallest if we use the wide definition of performance pay (6.2 percentage points). The finding is consistent with the notion that performance-related pay might be a management practice imported from abroad[4]. It is also interesting to note that performance-related pay is much less common in smaller establishments. In small firms, it is easier, all else equal, to observe how hard individual employees work, i.e. it is cheaper to implement a monitoring technology and choose fixed pay, than in large firms. Consequently, a lower incidence of performance pay in small establishments is consistent with the main prediction from the theoretical model. Relative to larger firms (20 or more employees), we find that smaller establishments are 14-15 percentage points less likely to use performance-related pay schedules. Looking at employee characteristics, the only finding that is statistically significant across all definitions is that performance-related pay is less common in establishments with a high share of part-time employees. A 10 percentage point increase in the share of part-time workers is associated with a 3 percentage point lower probability of performance-related pay. Certain types performance-related pay can be more difficult to implement when there are large differences between employees with respect to hours worked. In terms of the theoretical model, in firms with a large part-time workforce, random events may contribute to a larger part of the variation in output and consequently the effort under the optimal sharing rule will be lower than under fixed pay (with monitoring). In the theoretical model, the costs of supplying effort (determined by p) play a central role. In some jobs it is easier, and less costly, to increase effort in a way that

Who pays for performance?

21

Table III. Determinants of use of performance-related pay; random-effects probit regressions 0.0628 * * 0.0914 * * * 0.1206 * * * 0.0826 * * 2 0.1523 * * * 0.0603 0.0859 2 0.3148 * * * 2 0.0671 * 2 0.1911 * * * 2 0.1568 * * * 0.2127 * * * 0.3549 * * * 218,120.8 2,440.8 0.0000 0.0265 0.0347 0.0356 0.0354 0.0318 0.0737 0.0695 0.0790 0.0393 0.0424 0.0431 0.0277

0.0688 * * * 0.0921 * * * 0.1054 * * * 0.0622 * 2 0.1378 * * * 0.1286 * 0.1139 * 2 0.2875 * * * 2 0.0635 * 2 0.2117 * * * 2 0.1535 * * * 0.2756 * * * 0.2954 * * * 2 17,990.9 2,720.4 0.0000

0.0257 0.0330 0.0350 0.0337 0.0312 0.0710 0.0664 0.0748 0.0379 0.0416 0.0412 0.0271

Wide definition Standard n errors

Notes: The sample consists of 2,982 observations of 2,207 establishments. The table lists estimated marginal effects on the probability of performance pay. Models are estimated using the “xtprobit” command in Stata 9 (StataCorp, 2005). “rho” denotes the proportion of the overall error variance that is attributed to the establishment components. See note to Table 2 for calculations of marginal effects. Regressions also include indicators for industry as well as an indicator for missing data on foreign ownership; * Statistically significant at the 0.10 level; * * at the 0.05 level; * * * at the 0.01 level

0.0267 0.0359 0.0349 0.0368 0.0313 0.0747 0.0713 0.0808 0.0396 0.0415 0.0438 0.0273

Medium definition Standard n errors

22

Autonomy 0.0471 * Exports 0.1131 * * * High competition 0.1254 * * * Foreign ownership 0.1135 * * * Fewer than 20 employees 20.1532 * * * Share of employees with college education 20.0307 Share females 0.0710 Share part-time 20.2929 * * * Local bargaining 20.0686 * Central bargaining 20.1639 * * * Union density 20.1542 * * * 2003 observation 0.0969 * * * rho 0.4022 * * * Log likelihood 218,060.9 Chi-squared (25) 2,270.4 p-value 0.0000

Strict definition Standard n errors

IJM 29,1

increases output than in others. This will typically be in jobs where discretion over tasks is high. Following Brown (1990), it is likely that the productivity of high-skilled workers is more sensitive to effort, either because of their inherent or acquired characteristics or because they are assigned to jobs where it is easier to influence output through effort. This should imply a higher incidence of performance-related pay in establishments with a large share of highly educated workers. Table III reveals a mixed picture. We find a positive and weakly significant effect only when we use the wide definition of performance-related pay, where we include individual performance assessments even when they have a minor impact on total wages. The results with respect to bargaining regime and union density uncovered in Table II hold even when we include more establishment characteristics: the further away from the individual level wages are set, and the higher the union density, the smaller is the incidence of performance pay. Unions may have preferences against high wage inequality, also within firms. If performance-related pay results in greater wage inequality within firms, as found in Barth et al. (2006), and unions have some influence on the choice of pay system, this may explain the negative association. Unions are also likely to oppose wage systems that leave parts of the performance assessment at the discretion of management. Further, wage bargaining may act as a substitute for performance-related pay, as local bargaining may act as a profit sharing device. From the theoretical model, we find that more risk-averse employees imply less performance pay. If membership in a trade union is perceived as insurance against fluctuating wages, a high union density may reflect that workers in the firm on average are more risk averse. The theoretical framework predicts that increased risk aversion will raise the compensation for the uncertainty embedded in performance pay systems and thereby make fixed pay relatively more favourable to the firm. It is also likely that unions effectively reduce costs of monitoring effort. In a bargaining context, unions may share the interest of the employer in terms of monitoring effort of workers, and unions may have more efficient means of policing effort through peer control, group pressure, etc. In the estimations in Table IV, we also control for industry. The results are very similar to those in Table III, and are not reported in the table. Traditional, group-based, and individual-based forms of pay So far we have only discussed the determinants of use of performance-related pay in general. However, the discussion of results using the three alternative definitions indicated that there may be important differences in the effects of firm characteristics across types of performance pay. As the 2003 survey separated between several different types of performance pay, we are also able to study how different establishment characteristics influence the choice of specific forms of performance pay. Table IV reports the results from analyses where we distinguish between “traditional” (i.e. piece rates and commissions), “group-based” (profit sharing and group bonuses), and “individual-based” (individual bonuses and performance assessments) performance pay schemes. Because firms can combine two or more forms of performance pay, regression errors are likely correlated across equations. In order to account for any cross-equation correlation of regression errors, we base estimates on multivariate probit regressions. For this purpose, we employ the simulated maximum estimator developed by Cappellari and Jenkins (2003) in their Stata “mvprobit” module.

Who pays for performance?

23

Table IV. Determinants of use of traditional, group-based, and individual-based forms of performance pay; multivariate probit regressions 0.0994 0.1365 0.1544 0.1072 0.1304 0.2984 0.2525 0.2847 0.1548 0.1769 0.1559 0.4580 0.1865 0.1892 0.1977 0.2213 0.2216 0.3519 0.2632 0.1896 79.6437 0.2708 0.2513 0.0516

0.1076 0.0900 0.2322 * * 0.2471 * * * 20.2365 * * 20.1704 0.3197 21.0184 * * * 20.2048 * 20.3374 * * 20.1578 20.1503 20.0877 0.1136 0.1862 0.1470 20.7424 * * * 20.7600 * * * 0.5536 * * 0.0243 20.8203 * * * 20.6572 * * * 20.4028 * * 0.0725b 0.3299

0.0796 0.1028 0.1112 0.0850 0.1038 0.2084 0.2096 0.2517 0.1228 0.1422 0.1289 0.2149 0.1343 0.1706 0.1527 0.1737 0.2018 0.2932 0.2231 0.1551 0.2824 0.2389 0.1934 0.0453

0.2537 * * * 0.0185 0.1042 0.1062 20.2925 * * * 0.7146 * * * 0.1857 20.5961 * * * 20.1046 20.4699 * * * 20.2885 * * 0.6615 * * * 0.0484 0.1099 0.4318 * * * 0.4249 * * 20.2498 0.0276 0.3062 0.3138 * * 20.3113 20.2115 20.6163 * * * 0.2398 * * *c 0.3552

n

0.0782 0.1026 0.1026 0.0854 0.1009 0.1974 0.1989 0.2277 0.1184 0.1378 0.1257 0.2073 0.1386 0.1761 0.1554 0.1698 0.1883 0.2361 0.2223 0.1536 0.2287 0.2080 0.1863 0.0504

Individual (3) Standard errors

Notes: a equation nos (1) and (2); b equation nos (2) and (3); c equation nos (3) and (1); Sample size is 1,426 (2003 data only). Coefficients reflect changes in the inverse cumulative standard normal distribution, z. Estimation is based on the mvprobit module by Cappellari and Jenkins (2003). The (joint) log likelihood value is 22,061.0; and the likelihood ratio test for the joint significance of the explanatory variables yields the chi-squared (69 degrees of freedom) statistic of 450.0 (p-value ¼ 0:0000). The row labeled “Correlation” lists the estimated correlation coefficient between errors of the equations. The likelihood ratio test for the joint significance of the three correlation coefficients yields the chi-squared (3 degrees of freedom) statistic of 30.228 (p-value ¼ 0:0000). See text for definitions of forms of pay; * Statistically significant at the 0.10 level; * * at the 0.05 level; * * * at the 0.01 level

0.0626 0.0515 0.3881 * * 0.1891 * 2 0.2184 * 2 1.5292 * * * 0.1627 0.0223 0.2269 0.0120 2 0.3083 * * 2 0.5905 2 0.2968 1.2153 * * * 0.0540 2 0.1867 0.0173 2 0.3243 0.9626 * * * 0.8031 * * * 2 3.8308 0.4846 * 2 1.470 * * * 0.1315 * *a 0.1437

n

Group (2) Standard errors

24

Autonomy Exports High competition Foreign ownership Fewer than 20 employees Share of employees with college education Share females Share part-time Local bargaining Central bargaining Union density Oil, mining, energy Durables Construction Wholesale Retail, hotels, restaurants Transportation Post and telecom Finance and real estate Business services Health and social services Education, personal services Constant Correlation
wðXbÞ

n

Traditional (1) Standard errors

IJM 29,1

The table shows that a high degree of worker autonomy is particularly associated with a higher probability of individual-based pay schemes. In the table, the coefficients refer to changes in the value of Z, where Z has a standard normal distribution. In order to evaluate the marginal effect of explanatory variables on the probability that the firm adopts a performance pay scheme, we rescale the coefficient estimate with the value of the standard normal density function evaluated at the predicted mean of the Z-variable[5]. As such, evaluated at sample means of the explanatory variables, the estimated effect of workplace autonomy on individual-based performance pay is 9.0 percentage points (0:2537*0:3552; the scale factor is reported in the last row of the table). Establishments with high product market competition and foreign ownership are more likely to have traditional and group-based schemes. The finding in Table III that establishments with a highly educated workforce may be slightly more likely to have performance-related pay, masks large differences with respect to the different types of pay. In fact, such establishments are less likely to have traditional schemes than fixed pay, but more likely to have individual-based schemes. This pattern may reflect that monitoring problems associated with output as well as effort are more important for this group, thus favouring individual-based forms of pay for performance over other forms. A high share of part-time workers reduces the use of non-traditional pay schemes. A high union density rate is associated with less use of all three forms of performance pay. Note, however, that union density effect on group-based schemes is not statistically significant and is smaller in size than those of the two pay types, indicating that collective preferences for pay equality is particularly important when union membership is high. Firms with central bargaining only are less likely to use the non-traditional pay schemes. Finally, the negative effect of local bargaining that we found in Table III is driven by less use of group-based schemes. This is consistent with the view that local bargaining acts as a profit-sharing mechanism and may substitute for group-based performance-related pay systems. Figure 2 summarises the patterns of use of pay method according to bargaining

Who pays for performance?

25

Figure 2. Forms of performance pay by bargaining regime, 2003

IJM 29,1

26

regime, showing that the incidences of non-traditional pay schemes are less prevalent in establishments with union bargaining. In Table IV, the estimated industry coefficients show that there are large differences across industries in what types of performance pay that is used. We see that the high incidence of performance pay in the wholesale and oil and energy sectors is driven by their use of the individual-based schemes. The construction, finance and business services industries mainly use traditional schemes, reflecting their larger use of group piece-rates in construction and commissions in finance and business services. 5. Conclusions Our theoretical analysis of pay systems emphasises the role of costs involved in monitoring worker effort in combination with standard factors embedded in the agency model like risk aversion, uncertainty, and the sensitivity of output to effort. Theory predicts that the choice of performance-related pay schemes is positively associated with delegation of decisions over tasks. Using data from two Norwegian employer surveys, we find that the use of performance-related pay is positively associated with autonomy of the main occupational group in terms of defining work tasks. In our analyses, the positive association remains even after we include extensive controls for workforce and establishment characteristics. Worker autonomy has the strongest positive effect on individual-based pay schemes such as individual bonuses and performance assessments. On the other hand, we find no indication that worker autonomy has any impact on the incidence of group bonuses or profit sharing. The observation that the incidence of performance-related pay is higher with autonomous employees is consistent with an agency model interpretation of performance pay, and lends support to the hypothesis of Prendergast (2002) about a positive relationship between incentive pay and delegation of tasks. Our empirical results also suggest that the relationship is economically significant. We find that employees in firms where the main occupational group enjoys considerable freedom in choosing how to organise their own work, have a six percentage points higher incidence of performance-related pay than employees in firms with less freedom to choose how to organise one’s work. Likewise, workplace autonomy is associated with an increase in the incidence of individual-based performance pay schemes of nine percentage points. Evaluated at sample means, autonomy is estimated to raise the likelihood of performance pay by 13 per cent, and that of individual-based pay of 27 per cent. We also find, in line with previous literature, a higher incidence of pay for performance in larger establishments (see, e.g. Brown, 1990; Foss and Laursen, 2005). We find that collective bargaining reduces the incidence of performance pay. In particular, centralised bargaining over wages has a strong negative effect. Adding local bargaining diminishes the negative effect of collective bargaining. It is worth noting that local bargaining in effect may act as a profit sharing device, thus providing a substitute measure for other profit sharing schemes. This interpretation is consistent with the observation that local bargaining has a larger negative impact on group-incentive arrangements than on individual-based performance-related pay. In addition to the effects of bargaining level, union density has by itself a negative effect on pay for performance. There are several reasons why unions might oppose pay-for-performance schemes. In light of our model, it is likely that unions make monitoring of effort less costly. In a bargaining context, the union may share the

interest of the employer in terms of monitoring effort levels, and the union may have more efficient means, such as peer control and group pressure, to enforce effort rules. Unions also tend to oppose wage systems that lead to increased wage dispersion, and might be expected to dislike wage systems that tie pay to individual performance assessments at the discretion of management. Our empirical results reveal that a more powerful union in terms of establishment membership does not reduce the likelihood of group bonuses or profit sharing. It turns out that product-market competition is associated with a higher probability that the firm employs performance pay schemes. This effect is largest for the traditional types of performance-related pay. We find a positive association between the educational attainment of employees and use of individual-based types of performance pay. At the same time, the use of traditional piece rates is significantly lower in firms with a high fraction of college graduates. We interpret this pattern as follows: It is likely that the quality and effort of high-skilled workers have larger impacts on productivity than the quality and effort of other groups of workers. If this is the case, paying for performance has a greater effect on output for high- than for low-skilled workers. On the other hand, educational attainment of the workforce is negatively associated with traditional performance-related schemes, which typically are tailored towards blue-collar jobs. Finally, we find no significant linkage between educational attainment and group-based incentives schemes. Even when controlling for a full set of explanatory variables, the data reveal a significant underlying growth trend in use of performance-related pay in Norwegian private-sector establishments. Higher prevalence of performance-related pay over the sample period may reflect what Brown and Heywood (2002) describe as an “accelerating nature of experimentation and change in payment methods.” If this is true, there exists both a great deal of uncertainty among management about optimal methods of pay, as well as quite some leverage in terms of what types of payment schemes that prevail in the market at the same time. Notes 1. Relaxing the implicit assumption that marginal and average monitoring costs are equal, higher marginal monitoring costs will reduce the optimal effort level and thereby firm profits, while higher average monitoring costs will have a direct, negative effect on profits. In either case, PRP is more likely the higher are monitoring costs. 2. During the survey, managers were first asked about the main product or service of the establishment, and then asked to name the main occupation involved in processing that product/service. In the data, responses to the product or service question correspond closely with the standard industry classification of the establishment available from registers. Responses to the main occupation question also adhere to standard occupational classifications. To illustrate, the most frequently listed occupations within the ship-building and construction industries (to name two of the largest three-digit industries in the data) are “production workers,” “metal workers,” “carpenters,” and “construction workers.” 3. The figure uses values from the medium definition of performance pay in 2003. The alternative definitions also indicate increases for all bargaining regimes. To illustrate, using the strict definition the 2003 proportions are 64.0 (individual bargaining), 45.5 (local union), and 35.5 (central union).

Who pays for performance?

27

IJM 29,1

28

4. We are, however, unable to rule out any reverse effect – that foreign investors seek out firms with performance pay schemes. Moreover, foreign ownership and performance pay may both be influenced by a third and unobserved firm characteristic. Estimated effects of other explanatory variables are hardly affected if we drop the foreign ownership variable from the models. 5. This follows from ›F=›x ¼ ð›F=›zÞð›z=›xÞ ¼ fð›z=›xÞ, where F denotes the cumulative standard normal distribution function and f the standard normal density function.

References Bandiera, O., Barankay, I. and Rasul, I. (2005), “Social preferences and the response to incentives: evidence from personnel data”, Quarterly Journal of Economics, Vol. 120 No. 3, pp. 917-62. Barth, E., Bratsberg, B., Hægeland, T. and Raaum, O. (2006), “Performance pay and within-firm wage inequality”, paper presented at the Oslo Workshop on Employer Surveys, Institute for Social Research, Oslo, January. Belfield, R. and Marsden, D. (2003), “Performance pay, monitoring environments, and establishment performance”, International Journal of Manpower, Vol. 24 No. 4, pp. 452-71. Booth, A. and Frank, J. (1999), “Earnings, productivity, and performance-related pay”, Journal of Labor Economics, Vol. 17 No. 3, pp. 447-63. Brown, C. (1990), “Firm’s choice of method of pay”, Industrial and Labor Relations Review, Vol. 43 No. 3, pp. 165S-82S. Brown, M. and Heywood, J.S. (Eds) (2002), Paying for Performance: An International Comparison, M.E. Sharpe, Armonk, NY. Cappellari, L. and Jenkins, S.P. (2003), “MVPROBIT: stata module to calculate multivariate probit regression using simulated maximum likelihood”, available at: http://ideas.repec.org/c/ boc/bocode/s432601.html ˜ Cunat, V. and Guadalupe, M. (2005), “How does product market competition shape incentive contracts?”, Journal of the European Economic Association, Vol. 3 No. 5, pp. 1058-82. Foss, N.J. and Laursen, K. (2005), “Performance pay, delegation and multitasking under uncertainty and innovativeness: an empirical investigation”, Journal of Economic Behavior and Organization, Vol. 58 No. 2, pp. 246-76. Heywood, J., Siebert, W.S. and Wei, X. (1997), “Payment by results systems: British evidence”, British Journal of Industrial Relations, Vol. 35 No. 1, pp. 1-22. Holmstro¨m, B. and Milgrom, P. (1987), “Aggregation and linearity in the provision of intertemporal incentives”, Econometrica, Vol. 55 No. 2, pp. 303-28. Lazear, E.P. (1995), Personnel Economics, MIT Press, Cambridge, MA. Lazear, E.P. (2000), “The use of performance measures in incentive contracting”, American Economic Review, Vol. 90 No. 2, pp. 415-20. Lazear, E.P. (2002), “Performance pay and productivity”, American Economic Review, Vol. 90 No. 5, pp. 1346-61. Ortin-Angel, P. and Salas-Fumas, V. (1998), “Agency-theory and internal-labor-market explanations of bonus payments: empirical evidence from Spanish firms”, Journal of Economics and Management Strategy, Vol. 7 No. 4, pp. 573-613. Parent, D. (2002), “Performance pay in the United States: its determinants and effects”, in Brown, M. and Heywood, J.S. (Eds), Paying for Performance: An International Comparison, M.E. Sharpe, Armonk, NY, Ch. 2.

Prendergast, C. (1999), “The provision of incentives in firms”, Journal of Economic Literature, Vol. 37 No. 1, March, pp. 7-63. Prendergast, C. (2002), “The tenuous trade-off between risk and incentives”, Journal of Political Economy, Vol. 110 No. 5, pp. 1071-102. Raith, M. (2003), “Competition, risk and managerial incentives”, American Economic Review, Vol. 93 No. 4, pp. 1425-36. Salas-Fumas, V. (1993), “Incentives and supervision in hierarchies”, Journal of Economic Behavior and Organization, Vol. 21 No. 3, pp. 315-31. Schmidt, K.M. (1997), “Managerial incentives and product market competition”, Review of Economic Studies, No. 64 No. 2, April, pp. 191-213. StataCorp (2005), Stata Statistical Software: Release 9, StataCorp LP, College Station, TX. About the authors Erling Barth is Research Director at the Institute for Social Research, Oslo and Professor at the Department of Economics, University of Oslo. He is also a Research Fellow at IZA, Bonn. He earned his PhD in Economics from the University of Oslo. His research interests include education, productivity, immigration, the wage structure, gender wage differentials, labour mobility, firms’ behaviour and the impact of labour market institutions. Erling Barth is the corresponding author and can be contacted at: [email protected] Bernt Bratsberg is Senior Research Fellow at the Ragnar Frisch Centre for Economic Research, University of Oslo, and Professor of Economics at Kansas State University. He received his PhD in Economics from University of California, Santa Barbara. Bratsberg has published in the areas of immigration, wage inequality and unions. Torbjørn Hægeland is Head of Research, Labour Market and Firm Behaviour, at Statistics Norway and Scientific Adviser at The Ragnar Frisch Centre for Economic Research, University of Oslo. He holds a PhD in Economics from University of Oslo, Norway. His research fields are mainly empirical labour economics and economics of education. Oddbjørn Raaum is a Senior Research Fellow at The Ragnar Frisch Centre for Economic Research, University of Oslo. He holds a PhD in Economics from University of Oslo, Norway. His research areas and academic publications cover topics like wage formation, labour market policies, education economics, labour market performance among immigrants and the impact of family background on adult socio-economic outcomes.

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints

Who pays for performance?

29

The current issue and full text archive of this journal is available at www.emeraldinsight.com/0143-7720.htm

IJM 29,1

30

Market power, dismissal threat, and rent sharing The role of insider and outsider forces in wage bargaining Anabela Carneiro Faculdade de Economia da Universidade do Porto and CETE, Porto, Portugal, and

Pedro Portugal Banco de Portugal and Universidade Nova de Lisboa, Lisboa, Portugal Abstract Purpose – The purpose of this study is to investigate to what extent the existence of high labor adjustment costs has some influence on the process of wage negotiation. In particular, it aims to analyse if the risk of being laid off has any impact on insiders’ bargaining power and, consequently, on their wage claims. Design/methodology/approach – A collective bargaining model that closely follows those developed by Nickell et al. and Bentolila and Dolado is adopted and a longitudinal panel of large Portuguese firms from all sectors over the 1993-199 period is used. Findings – The results reveal that firms where insider workers appear to have more bargaining power tend to pay higher wages. In particular, we found that the threat of dismissal tends to weaken insiders’ bargaining power and, consequently, to depress wages. Research limitations/implications – In future research an attempt should be made to measure directly the labor turnover costs. Originality/value – This paper presents robust empirical evidence using micro-data for individual firms that support one of the predictions of the insider-outsider theory that wages will be higher in sectors (firms) with high labor turnover costs. Keywords Pay bargaining, Market forces, Dismissal, Rents, Portugal Paper type Research paper

1. Introduction It is widely agreed that labor turnover costs (LTC) give insider workers market power that they can exploit to their own advantage. In fact, insiders’ positions are protected

International Journal of Manpower Vol. 29 No. 1, 2008 pp. 30-47 q Emerald Group Publishing Limited 0143-7720 DOI 10.1108/01437720810861994

The authors would like to thank Juan Dolado and two anonymous referees for helpful comments and suggestions. They would also like to thank Manuel Arellano for kindly supplying them with the DPD program. The authors gratefully acknowledge the Departamento de Estatı´stica do Ministe´rio do Trabalho e da Solidariedade Social for allowing them to use the data from the Social Audit. Financial support from Fundac¸a˜o para a Cieˆncia e Tecnologia (FCT) under research grant POCTI/ECO/35147/99 is also warmly acknowledged. The usual disclaimer applies. CETE is a Research Centre supported by Fundac¸a˜o para a Cieˆncia e a Tecnologia, Programa de Financiamento Plurianual through the Programa Operacional Cieˆncia, Tecnologia e Inovac¸a˜o (POCTI)/Programa Operacional Cieˆncia e Inovac¸a˜o 2010 (POCI) of the III Quadro Comunita´rio de Apoio, which is financed by FEDER and Portuguese funds.

by LTC which give them some labor market power in the process of wage negotiation. The insider-outsider approach relies on the assumption that wages are set through bargaining, not between firms and the whole labor force, but rather between firms and their workers. In this context wages might be widely influenced by firm’s internal conditions rather than by external conditions and it should be expected that the greater the hiring and firing costs, the more the insider wage will depend on the “inside factors” relative to the “outside factors”. Furthermore, it should also be expected that in labor markets with high job security and/or high adjustment costs the threat of dismissal is relatively stronger, because mean unemployment duration tends to be longer. Relying on the distinction between insider and outsider workers, the insider-outsider theory of wage formation aims to explain why wages may be set above their market-clearing levels[1]. The insider-outsider explanation is based on the idea that the level of wages is primarily determined by the currently employed workers (the so-called “insiders”), with unemployed (the “outsiders”) playing little or no role in the process of wage bargaining. Furthermore, this approach attempts to explain why unemployed workers do not compete for existing jobs by offering to work at jobs for which they are qualified at a wage lower than that currently being paid to incumbents. Lindbeck and Snower (1986) showed that the existence of costs associated with insider-outsider turnover might explain why firms do not replace their high-wage insiders with low-wage outsiders. Accordingly, involuntary unemployment can arise due to the existence of LTC such as hiring, training, and firing costs or the costs generated by the disincentive to cooperate with outsiders, that make it costly for the firm to replace an insider worker with an unemployed worker[2]. The rents associated with these labor market frictions give some bargaining power to insiders in the process of wage setting. Based on this theoretical framework, in the 1990s, the econometric models of wage determination started to include measures of the firm’s profits or financial performance as explanatory variables. This literature has focused directly on rent-sharing models (see, among others, Nickell et al., 1994; Holmlund and Zetterberg, 1991; Abowd and Lemieux, 1993; Blanchflower et al., 1996; Hildreth and Oswald, 1997). These studies used panel data at both firm or industry levels, and estimated a number of versions of the wage equation with rents per worker included. Although they used different models of collective bargaining, the results of these studies indicate, in general, that changes in profitability are shown to feed through into long-run changes in wages. A branch of this literature, to which this paper is more closely related, has been focusing on the relative importance of insider versus outsider forces in wage determination. These studies show that firm specific factors, as well as general labor market conditions, play an important role on the process of wage determination, although the weight attached to firm-specific factors varies considerably across countries. Indeed, the results suggest that firm-specific factors are relatively unimportant in the Nordic Countries, of some importance in Britain, West Germany, Spain and The Netherlands, and highly important in Australia, Canada and the USA[3]. Using an approach inspired in Nickell et al. (1994), the main contribution of this study is to investigate to what extent the existence of high labor adjustment costs has some influence on the process of wage negotiation. In particular, we analyze if the risk of being laid off has any impact on insiders’ bargaining power and, consequently, on

Forces in wage bargaining

31

IJM 29,1

32

their wage claims. For this purpose, we use a representative dataset with 820 large Portuguese firms from all sectors over the period 1993-1999. In fact, we believe that Portugal is a well suited case to better identify the impact of a dismissal threat on wages, since Portugal may be classified as an extreme case of employment protection (see OECD, 1999; Blanchard and Portugal, 2001; Vareja˜o and Portugal, 2006). The stricter Portuguese legislation is associated with lower turnover (high adjustment costs) in the labor market, with both jobs and unemployment spells tending to last longer. A labor market characterized by a high employment protection with high adjustment costs and lower turnover tends to create bargaining power for currently employed workers that they can exploit to their own advantage. Of course, it can be argued that if turnover costs give bargaining power to workers, and thus the possibility to extract some rents, firms may try to recoup those rents from the insiders. However, firms are generally unable to pass these costs on fully to their insiders because firms do not incur these costs until they replace their insiders with new entrants. Even if firms could extract some of these rents by imposing lump sum payments to insiders upon voluntary quitting or firing “without cause”, such fees are usually illegal and incentive-incompatible (Lindbeck and Snower, 2002). Two additional objectives also drive the investigation. One is to evaluate the role and weight of insider forces in wage determination. The Portuguese industrial relations system presents some contrasting features. While on the one hand, the role of massive wage-setting mechanisms and the existence of extension mechanisms point to a centralized bargaining system, on the other hand, the scattered nature of union organization, the possibility opened to employers to bargain at the firm level, and the presence of a significant wage cushion, highlight aspects of decentralization that may grant employers some room for maneuver to set wages. In fact, whatever the wage floor agreed upon for each category of workers at the collective bargaining table, firms are free to pay higher wages, and they often deviate from that benchmark, adjusting to firm-specific conditions (Cardoso and Portugal, 2005). The second is to test the existence of asymmetric effects in wage adjustments, i.e., to test the extent to which wages in Portugal are more responsive to insider variables in the face of rising demand than in the face of declining demand. This paper will be organized as follows. The theoretical framework is presented in Section 2. The data and estimation method are described in Section 3. The empirical results are summarized in Section 4. Section 5 concludes. 2. The theoretical model The model used here as a basis for the estimations is a collective bargaining model that closely follows those developed by Layard et al. (1991) and Nickell et al. (1994), including the extension proposed by Bentolila and Dolado (1994), who consider a firm that employs two types of workers: under permanent and temporary contracts. In these types of models it is presumed that wages are determined through negotiations between the firm and the union at the first stage. At the second stage, firm sets price, output, and employment levels. In Portugal, negotiations between unions and firms, at the industry level, play an important role in the determination of wages. Hence, modeling the process of wage formation at the firm level using a bargaining approach seems to be an appropriate choice for the Portuguese case.

The basic idea underlying these models is that product market power generates rents that can be captured by employees in the form of higher wages. The possibility that workers share monopoly rents depends on workers’ bargaining power and may take place even in the absence of unions. In this framework, it can be shown that the average wage in the ith firm depends on firm-specific factors, outside factors, the firm’s market power and workers’ bargaining power[4,5]: 

 wit ¼ c0 þ l ð pit þ yit 2 nit Þ þ 1 2 a þ g Dnpit 2 ð1 2 lÞfit ð1Þ þ ð1 2 lÞ½w
t 2 c1 U t
þ c2 MSit þ c3 bit where wit is the log real wage in firm i, pit is the firm’s output price (in logs), yit is the firms output (in logs), nit is the number of employees (in logs), with (pit þ yit 2 nit ) being the firm’s revenue per employee (in logs). Dnpit measures the change in permanent employment and it is the proportion of temporary workers. w
t is the log of the outside real wage, Ut is the aggregate unemployment rate, MSit is the firm’s market share and bit measures workers’ bargaining power. We assume, as in Bentolila and Dolado (1994), that workers’ bargaining power, b, is a function of a set of variables that are related to the firms financial situation ( f ) and the proportion of temporary workers (f). Replacing b as a linear function of fit and fit and introducing dynamics in the model with the inclusion of lagged wages, equation (1) may be rewritten as[6]:   wit 2 w
t ¼ c0 þ a1 ðwit21 2 w
t21 Þ þ ð1 2 a1 Þ l ð pit þ yit 2 nit 2 w
t Þ

 ð2Þ þ 1 2 a þ g Dnpit þ ½d 2 ð1 2 lÞ
fit 2ð1 2 lÞc1 U t þ c2 MSit þ c3 f it : The interpretation of equation (2) is straightforward. The firm’s average wage per employee depends on previous wages, on firm-specific factors such as revenue per employee, the change in the number of insiders, the market share and workers’ bargaining power, and on outside factors such as the unemployment rate. The parameter l may be termed the “insider weight”, i.e., the long-run elasticity of firm wages relative to firm revenue per employee. 3. Data and estimation method 3.1. The data Our basic data source is the Social Audit survey (Balanc¸o Social ) and includes a panel of Portuguese firms, with at least 100 employees, from all sectors, over the period 1993-1999. The Social Audit survey (SA) is gathered annually by the Portuguese Ministry of Employment. When it was first introduced (1986), it covered state-owned firms only. Since then its coverage has spread first to firms with at least 500 employees, and since 1992 to firms with at least 100 employees. Responding to this survey is mandatory. On average, 2,040 firms respond to the survey each year, corresponding to a total of 772,000 workers. In fact, the SA is characterized by a very high degree of coverage of large firms in Portugal. Each year, a respondent firm reports data on a large variety of topics concerning the workforce composition and labor costs. This is organized in six major areas: (1) firm’s characteristics; (2) employment;

Forces in wage bargaining

33

IJM 29,1

34

(3) (4) (5) (6)

labor costs; occupational safety; vocational training; and fringe benefits.

The collection of firm’s characteristics includes information about location, economic activity (SIC codes), legal setting, employment, number of establishments and production (value-added). The existence of a unique identification number for each firm allows us to create a longitudinal panel of firms. The employment branch, which is the largest in this survey, collects detailed information about workers’ attributes. This includes information about gender, age, skills, schooling, tenure, hours of work, etc. Total employment is also decomposed by type of contract and skill level, which allows one to obtain the number of permanent and temporary workers at the end-of-year count. The information about labor costs includes annual base-wage, regular paid benefits and bonuses, irregular benefits and bonuses, costs with vocational training, and other fringe benefits. One of the main advantages of this data set, besides its coverage and longitudinal nature, is the availability of information on both the firm’s and workers’ characteristics. The possibility of controlling the skill composition of the workforce over the years as well as the possibility of computing workers’ flows constitutes an important advantage of this data set. The information about employment by type of contract is equally important. In order to complement the information available in the SA survey, we will also use the data contained in the Quadros de Pessoal survey (QP). QP is an annual mandatory employment survey collected by the Portuguese Ministry of Employment that covers nearly all establishments with wage earners[7]. In each year every establishment with wage earners is legally obliged to fill in a standardized questionnaire. Reported data cover the establishment itself (location, economic activity and employment), the firm (location, economic activity, employment, sales and legal framework) and each of its workers (gender, age, education, skill, occupation, tenure, earnings and duration of work). Currently, the data set collects information on around 250,000 firms and 2.5 million employees. The information from QP about wages will be used in order to compute the outside wage. There are two main reasons to believe that QP can provide a reliable measure of the outside wage. The first is its coverage and reliability. By law, the questionnaire is made available to every worker in a public space of the establishment. This requirement facilitates the work of the services of the Ministry of Employment that monitor compliance of firms with the law (e.g., illegal work). Indeed, the administrative nature of the data and its public availability imply a high degree of coverage and reliability. The second is that the information on earnings is very complete. It includes the base wages (gross pay for normal hours of work), seniority payments, regular benefits, irregular benefits, and overtime pay. Since the data on value-added are not available in the SA, the information on sales from QP will be used instead. This is possible because the identification code of firms in the SA and the QP data sets is the same. The sales variable will be used to compute a measure of productivity and a measure of market share. Thus, nominal productivity

will be defined as annual sales per employee[8]. The market share is obtained by the ratio between the firms sales and total (five-digit) sector’s sales. Neither the SA nor the QP data sets have information on union density. In fact, there are no micro-data in Portugal with information on the number of workers who are members of a trade union. Even though the SA survey includes information about profits and financial costs these data are not available. In order to overcome these difficulties, some proxies were used to measure workers bargaining power. As pointed out by Lindbeck and Snower (2002), “. . . the insider-outsider theory is not just about labor unions. Any employee whose position is protected by labor turnover costs is an insider of sorts, regardless of whether he belongs to a union”. As shown before, since Portugal is characterized by a stricter employment legislation with higher firing costs and low flows in and out of unemployment, it appears that there is some scope for the existence of insider power beyond the one that might result from the behavior of unions. Hence, we include the labor utilization rate within the firm and the layoff rate as measures of insiders’ bargaining power. These two variables may be viewed by insider workers as a signal of the firms risk of illiquidity. As initially suggested by Gregory (1986), it is probably the labor utilization rate within the firm that is particularly important for wage negotiations, rather than the labor utilization rate within the economy. The labor utilization rate may affect wages in two ways. First, higher labor utilization rates within the firm increase the probability of job retention of an insider worker raising their power of negotiation. Second, as the labor utilization rate increases, the threat of a strike becomes more credible to the firm, raising the workers bargaining power. The labor utilization rate (lur) is defined as the ratio between the total number of hours actually worked in the year and the maximum annual potential of hours worked[9]. Higher labor utilization rates within the firm will induce workers to demand higher wages, ceteris paribus. The layoff rate (layoff) is measured as the ratio of the total number of involuntary separations in the year (of permanent and temporary workers) by the firm’s average employment in the year. In order to minimize the endogeneity problems both variables were lagged by one year. When wages are largely set in the interest of the insiders, as predicted by the insider-outsider theory, it should be expected that layoff rates have a negative impact on wages, since higher layoff rates threaten the jobs of the insiders. Graafland (1992), using aggregate data for The Netherlands, showed that the layoff rate has a significant negative influence on wages. In order to control for the aggregate outside labor market conditions, we decided to include a set of time dummies and the regional unemployment rate. The regional unemployment rate is defined 5 at the level of NUTS II[10]. The outside wage per employee will be defined by region (NUTS III), excluding the firms own wage[11]. A precise definition of all variables is presented in Appendix 1. The sample was limited to firms for which data are held for at least four consecutive years and with no missing values in the explanatory variables. To minimize the effects of outliers, we also excluded from the sample those firms whose (real) sales increase more than five times or decrease to less than one-fifth from one year to the next. After these restrictions, we obtained an unbalanced panel of 820 firms and 5,150 observations, representing a total of about 276,000 workers[12]. In Table I some

Forces in wage bargaining

35

Table I. Basic characteristics of the data (1993-99) 0.016 0.029 0.922 0.075 378.9 59,054.9 5,947.5 4,382.0 689

0.002 0.024 0.928 0.098 371.2 61,269.7 5,946.8 4,585.2 723

1994

Note: a Annual real values (in Euros); CPI deflator (base ¼ 1991)

Employment growth rate Layoff rate Labor utilization rate Market share Firm size Sales per employeea Wages per employeea Outside wage per employeea Number of firms

1993 2 0.003 0.027 0.928 0.095 361.2 62,571.3 6,062.0 4,534.0 759

1995 0.003 0.029 0.932 0.091 352.9 63,045.7 6,212.1 4,660.4 820

1996

0.014 0.028 0.930 0.095 367.7 69,658.8 6,495.8 4,791.9 782

1997

2 0.006 0.029 0.931 0.097 391.2 71,804.4 6,662.8 5,004.4 718

1998

36

Variables

2 0.001 0.027 0.927 0.101 409.6 74,594.0 6,700.3 4,820.7 659

1999

IJM 29,1

selected variables are reported in order to characterize the sample over the 1993-1999 period. As can be seen in Table I, the firms in the sample have an average size of 376 employees and 9.3 percent of market share. Over the 1993-1999 period, employment growth rates changed between a positive value of 1.6 percent in 1993 to a negative one of 0.1 percent in 1999, with involuntary separations (layoffs) representing around 2.8 percent of average employment. On average, the labor utilization rate is around 93 percent. Between 1993 and 1999 real average wages in the firm grew, on average, at an annual rate of 2.0 percent, whereas the outside wage in the region grew at an annual rate of 1.6 percent. The average annual growth rate of real sales per employee was 4 percent. 3.2. Estimation method In the light of the previous discussion the empirical counterpart of equation (2) is given by:

wijt 2 w
jt ¼ a0i þ a1 wijt21 2 w
jt21 þ a2 pijt þ yijt 2 nijt 2 w
jt þ a3 Dnpijt þ a4 fijt þ a5 layoffijt21 þ a6 lurijt21 þ a7 MSijt21 þ a8 U jt þ 1ijt

ð3Þ

where i ¼ firm, j ¼ region and t ¼ time, a0i is a firm fixed effect and 1ijt is an idiosyncratic error term. Equation (3) constitutes the basis of the empirical analysis. The dynamic linear model of equation (3) is an autoregressive fixed effects model. In the presence of such a model, it is well known that the ordinary least squares (OLS) estimator is inconsistent. A conventional way to tackle this problem is to use an instrumental variables estimation method. The application of the generalized method of moments (GMM) estimator suggested by Arellano and Bond (1991) overcomes these difficulties, producing consistent estimates. The GMM estimator identifies the parameters of the model under the assumption of lack of serial correlation in the error terms, and as this assumption is essential for the consistency of the estimator, a test of autocorrelation, developed in Arellano and Bond (1991), will be reported. The empirical model will be estimated using the system (SYS) GMM estimator proposed by Arellano and Bover (1995) and Blundell and Bond (1998). The SYS-GMM estimator uses lagged first differences as instruments for equations in levels in addition to the usual lagged levels as instruments for equations in first-differences. This option is justified by the fact that the SYS-GMM estimator can dramatically improve the performance of the traditional first-differences (DIF) GMM estimator when the autoregressive parameter is moderately high and the number of time-series observations is moderately small. Indeed, the SYS-GMM estimator has superior properties in terms of small sample bias and root mean squared error, especially for persistent series (see Blundell and Bond, 1998)[13]. Recent empirical studies have reported some problems with the estimation of dynamic panel data models using the DIF-GMM estimator in cases of highly persistent regressors, which imply weak correlation of lagged levels with subsequent first differences. Mairesse and Hall (1996) and Blundell and Bond (1998), for example, showed that when the panel data are characterized by a large sample of firms observed

Forces in wage bargaining

37

IJM 29,1

over a small number of time periods, standard GMM estimators, which eliminate unobserved firm-specific effects first-differencing, have been found to produce unsatisfactory results. Blundell and Bond (2000) and Blundell et al. (2000) applied the SYS-GMM estimator to panel production functions for the USA and also showed that the use of the SYS-GMM estimator not only improves the precision of the regression coefficient estimates, but also reduces the finite sample bias.

38 4. Empirical results 4.1. Measuring insider power and insider forces The DIF and SYS-GMM two-step estimates of equation (3) for the unbalanced panel of 820 firms for the period 1994-1999 are displayed in Table II[14]. In order to control for the skill composition of the firm’s workforce, each specification includes a set of controls for workers’ skills. Thus, five levels of education (omitted category is basic school and less than basic school) and six levels of qualifications (omitted category is apprentices) were added to equation (3)[15]. As hinted above, our preferred parameter estimates correspond to the SYS-GMM estimator. In fact, compared with the DIF-GMM estimator, the SYS-GMM estimator yields more reasonable parameter estimates. In both regressions presented in Table II, the test statistics reported verify the critical assumption of no second-order serial correlation (m2 test) and the validity of the instruments (Sargan test) at the conventional levels of significance. Furthermore, the Dif-Sargan test for the validity of the additional level moment conditions used by the SYS estimator do not reject their validity at the 10 percent level. Thus, hereinafter our discussion of the estimation results will be based on the SYS-GMM approach[16]. The SYS-GMM results report a value of the insider weight (l) of 18 percent, estimated with precision[17]. This value is considerably higher than those obtained for other European Countries such as Spain and the UK using firm-level data (see Appendix 2, Table AI). In fact, the short-run effect of nominal productivity on wages is strong and significant (coefficient estimate of 0.143), suggesting that in Portugal wages are highly responsive to the firm’s performance. This is also consistent with one of the predictions of the insider-outsider theory that the greater the hiring and firing costs, the more the insider wage will depend on the “inside factors” relative to the “outside factors”. Other evidence is revealed by the results. First, market share exerts a positive and significant impact on wages, suggesting that monopoly power generates monopoly rents that are captured by the employees in the form of higher wages. Second, with respect to the dismissal threat variables, we obtain the expected signs for the coefficients on both the labor utilization rate and on the layoff rate. A 1 percent increase in lur raises wages, in the short run, by 0.32 percent. Hence, workers in firms with higher labor utilization rates have higher insider power and, thus, earn more. A 1 percent increase in the layoff rate decreases wages, in the short-run, by 0.022 percent[18]. This finding seems to suggest that when the employment perspectives of employed workers worsen, they tend to restrain wage demands[19]. Another interpretation is possible if the layoff rate is viewed as a proxy for labor adjustment costs. In firms with high (low) adjustment costs the risk of being fired is lower (higher) and thus insider workers are in a better position to extract rents in the form of higher wages. In fact, besides the high dismissal costs

Explanatory variables Wages lagged (wijt21 2 w
jt21 ) Nominal productivity (pijt þ yijt 2 nijt 2 w
jt ) Growth permanent employment (Dnpijt ) Proportion of temporary employees (fijt) Layoff rate (layoffijt2 1) Labor utilization rate (lurijt2 1) Market share (MSijt2 1) Regional unemployment rate (Ujt) Education levels Preparatory and lower secondary Upper secondary College Others Qualification levels Top executives Intermediary executives Supervisors Highly skilled and skilled professionals Semi-skilled and unskilled professionals Constant Time dummies Wald test of joint significance ( p-value) Sargan ( p-value) Dif-Sargan ( p-value) m1 ( p-value) m2 ( p-value) NT

DIF-GMM

SYS-GMM

0.064 (1.5) 0.124 * * (2.3) 2 0.032 * * (2 2.0) 2 0.005 (2 0.1) 2 0.002 (2 0.6) 2 0.160 * * * (2 1.9) 0.005 (1.1) 2 0.068 (2 1.4)

0.227 * (7.4) 0.143 * (6.1) 20.096 * (25.8) 20.019 (20.6) 20.022 * (25.3) 0.318 * * (2.5) 0.018 * (4.3) 20.123 * (25.9)

0.004 (0.6) 2 0.001 (2 0.2) 0.027 * (3.2) 0.008 (1.6)

0.010 (21.6) 0.009 (1.5) 0.063 * (7.6) 0.001 (0.2)

0.010 (2 1.1) 0.005 (0.9) 0.010 (1.4) 0.009 (1.5) 2 0.006 (2 1.1) 0.034 * (4.6) Yes * 44.5 (0.0) 85.6 (0.073) 2 4.3 (0.0) 1.5 (0.144) 3,510

Forces in wage bargaining

39

0.007 (0.7) 0.014 * * (2.4) 0.012 * * (2.2) 0.021 * (3.5) 20.003 (20.4) 20.377 * (23.0) Yes * 2,114.8 (0.0) 110.9 (0.099) 25.3 (0.44) 2 4.7 (0.0) 1.7 (0.088) 4,330

Notes: Dependent variable: wages (wijt 2 w
jt ); Subscript i denotes firm, j refers to region and t denotes time;all variables, except growth permanent employment and the proportion of temporary employees, are in logs; heteroskedasticity-consistent t-statistics in parentheses; *, * *, * * * denote significant, at 1, 5 and 10 percent, respectively; the variables treated as endogenous are: (wijt21 2 w
jt21 ), ð pijt þ yijt 2 nijt 2 w
jt ÞDnpijt ) and fijt; instruments include: the exogenous variables in equation (3), wijt22 :::wijt26 , (ð p þ y 2 nÞijt22 :::ð p þ y 2 nÞijt29 , npijt22 :::npijt26 , fijt22 :::fijt26 , Dwijt21 :::Dwijt25 , Dð p þ y 2 nÞijt21 :::Dð p þ y 2 nÞijt28 , Dnpijt21 :::Dnpijt25 , and Dfijt21 :::Dfijt25

Table II. GMM estimates of wage equation (1994-1999) Measuring insider power and insider forces

IJM 29,1

40

that Portuguese employers have to bear, conditions in which a termination of contract is admissible are also regulated quite strictly. These factors appear to work together to strengthen the bargaining position of incumbent workers and their power to claim for higher wages. Third, the regional unemployment rate has a negative and significant impact on wages. The elasticity of wages with respect to the regional unemployment rate is 2 0.123, which is a value that is in accordance with previous estimates (see, for instance, Blanchflower et al., 1990; Nickell et al., 1994). This result reveals that outsiders’ forces have an important role in wage determination in the sense that they affect the alternative options to the bargaining parties. Fourth, there is no evidence of membership hysteresis effect when the insiders are measured by the number of permanent employees, contrary to the result obtained by Bentolila and Dolado (1994) for Spain[20]. In fact, the estimate of the coefficient on the permanent employment change (Dnp) is negative and statistically different from zero. This result is not too surprising since in Portugal, contrary to Spain, unemployment rates in the last decade remained at fairly low levels (5 to 6 percent) with wages exhibiting a high aggregate wage flexibility. Thus, it is not expected that current wages depend inversely on past employment. Finally, a small negative effect of the proportion of temporary employees on average wages was found, although not statistically different from zero. On balance, the results presented in this section show that firms where insider workers have more labor market power tend to pay higher wages, ceteris paribus. In particular, in firms with low layoff rates and high rates of labor utilization within the firm, workers seem to extract rents in the form of higher wages. 4.2. Testing for asymmetric insider effects Insider effects may be more important in expanding firms when compared to declining ones, and similarly, firms’ wages may be more responsive to insider variables in good rather than in bad times. Such asymmetric insider effects imply downward wage rigidity, and will tend to put more pressure on employment when times are bad. Even though there is some empirical evidence showing that wage adjustments are asymmetric (see, for instance, Nickell and Wadhwani, 1990; Blanchflower, 1991; Holzer and Montgomery, 1993; Johansen, 1996), this issue remains unsettled. In this section we test the extent to which wages in Portugal are more responsive to insider variables in the face of rising demand than in the face of declining demand. The main problem associated with the implementation of any test of asymmetry is that demand is not observed. In order to construct a measure of expected product demand we use the average growth rate of real sales over the last three years as a proxy[21]. We then interact a dummy that takes the value one for positive rates of sales growth in the last three years (zero otherwise) with lagged wages and with nominal productivity. The SYS-GMM estimates for the full sample are reported in Table III. As can be seen from Table III, the coefficient of interaction term between the sales growth dummy and nominal productivity is positive and statistically different from zero, suggesting that when sales are expected to grow the impact of productivity on wages is higher. The interaction term between the average growth rate of sales and lagged wages is also statistically significant and negative, suggesting that when demand is expected to rise, the impact of last periods wage is reduced. Thus, the

SYS-GMM Explanatory variables Wages lagged (wijt21 2 w
jt21 ) Nominal productivity (pijt þ yijt 2 nijt 2 w
jt ) Employment growth (Dnpijt) Proportion of temporary employees (fijt) Layoff rate (layoffijt2 1) Labor utilization rate (lurijt2 1) Market share (MSijt2 1) Regional unemployment rate (Ujt) Interaction terms Wages lagged*Sales growth dummy Nominal productivity*Sales growth dummy Constant Time dummies Wald test of joint significance ( p-value) Sargan ( p-value) m1 ( p-value) m2 ( p-value) NT

0.367 * (7.1) 0.145 * (5.8) 20.093 * (25.6) 20.024 (20.7) 20.022 * (25.4) 0.341 * (2.7) 0.017 * (4.0) 20.128 * (26.1) 20.243 * (24.1) 0.020 * (3.3) 20.462 * (3.7) Yes * 2,116.1 (0.0) 100.3 (0.261) 2 5.3 (0.0) 1.4 (0.153) 4,330

Notes: Dependent variable: wages (wijt 2 w
jt ); see notes to Table II; each regression includes five educational levels and six qualification levels; Sales growth dummy =1 if sales growth . 0; 0 otherwise

asymmetry test seems to reveal that wages in Portugal are less responsive to productivity when demand is expected to decline and subject to greater inertia under these same circumstances. This evidence is quite similar to that obtained by Nickell and Wadhwani (1990). 5. Conclusion This study investigates wage determination at the firm level using a longitudinal panel of large Portuguese firms. The main empirical findings are the following. First, insider forces such as revenue per employee and market share have a significant impact on wage determination. After controlling for the skill mix of the workforce, the full sample estimates imply a long-run insider weight of 18 percent, which is comparable with estimates reported for economies characterized by a decentralized system of wage negotiation despite the fact that the system of wage bargaining in Portugal is very centralized and heavily regulated. Nevertheless, this

Forces in wage bargaining

41

Table III. SYS-GMM estimates of wage equation (1994-1999) testing for asymmetric insider effects

IJM 29,1

42

study reveals that there is room for firm maneuvering. In fact, wages seem to be very sensitive to firm-specific conditions, i.e., firm wages are significantly affected by firm performance, at least during period under scrutiny. This indication is in line with one of the predictions of insider-outsider theory, which states that the higher the firing and hiring costs, the higher the weight attached to insider forces. Second, the idea that wages will be higher in sectors (firms) with high labor adjustment costs found some empirical support. Most notably, the results revealed that a threat of dismissal tends to weaken insiders bargaining power and, consequently, to depress wages. Third, outside labor market conditions measured by the regional unemployment rate also play an important role in wage determination. The regional unemployment level in the economy has an influence on the negotiated wage through the probabilities of finding a job. Thus, the negative and significant impact of the regional unemployment rate on wages suggests that workers are more inclined to accept wage moderation when the probabilities of finding a job worsen. Finally, some evidence was found in favor of the existence of asymmetric insider effects. That is, real wages in Portugal seem to exhibit some downward real wage rigidity. Notes 1. See, for example, Lindbeck and Snower (1985, 1986, 1988) and Solow (1985). 2. For a description of this type of costs see Lindbeck and Snower (1986). 3. For a summary of the insider weight estimates obtained in these studies see Appendix 2, Table AI. 4. To derive equation (1), see Appendix A of Bentolila and Dolado (1994). 5. The benefit replacement ratio is omitted from equation (1) since the figures for this aggregate variable are virtually constant over the period of analysis and it does not seem reasonable to include it as an explanatory variable. 6. In order to ensure that the long-run homogeneity assumption in both the inside and outside factors is verified ½l þ ð1 2 lÞ ¼ 1
, all nominal variables such as wit, wit2 1 and
(pit þ yit 2 nit ) are measured as deviations from outside wage (w). 7. Public administration and household servants are excluded. 8. It should be noted that in each year information on sales lagged by one, two and three years is also available. 9. The information about the maximum annual potential of hours worked is provided directly by the firms. 10. At NUTS II mainland Portugal is split into five geographical areas. 11. At NUTS III mainland Portugal is split into 28 geographical areas. 12. Permanent employment represents around 82 percent of total employment. 13. For completeness, the results using the DIF-GMM estimator are also reported. 14. The equations are estimated using DPD98 (Dynamic Panel Data software) written by Arellano and Bond (1998). 15. The Wald test of joint significance of the education and qualification levels rejects the hypothesis that the coefficients are all equal to zero. Moreover, the SYS-GMM results revealed that controlling for worker’s skills reduces the effect of nominal productivity on

16. 17. 18. 19. 20.

21.

wages by around 5 percentage points, suggesting that there might exist a positive correlation between workers skills and nominal productivity. These same estimates are reported in Appendix 3, Table AII for the manufacturing sector. Overall, the results are qualitatively similar to the ones obtained for the full sample. The long-run value of the insider weight is calculated by dividing the nominal productivity coefficient (the short-run coefficient) by one minus the coefficient on the lagged wages. This result is reinforced by a composition effect of reverse sign that can emerge if we assume that temporary workers are the first ones to be fired because of their lower firing costs. Blanchflower (1991) obtained a similar result using microeconomic data on individuals for the UK. In this context, a membership hysteresis effect arises from insider power if employment growth raises wages. In fact, given current membership, the lower the last periods employment (i.e., the higher employment growth is) the more protected from losing their jobs the insiders will be. In order to minimize endogeneity problems, (and since in year t we have information on sales in t 2 1, t 2 2, and t 2 3) we used the average growth rate of real sales between t 2 1 and t 2 3.

References Abowd, J. and Lemieux, T. (1993), “The effects of product market competition on collective bargaining agreements: the case of foreign competition in Canada”, Quarterly Journal of Economics, Vol. 108 No. 4, pp. 983-1014. Arellano, M. and Bond, S. (1991), “Some tests of specification for panel data: monte carlo evidence and an application to employment equations”, Review of Economic Studies, Vol. 58 No. 2, pp. 277-97. Arellano, M. and Bond, S. (1998), “Dynamic panel data estimation using DPD98 for Gauss: a guide for users”, mimeo, Economics Department, American University, Washington, DC. Arellano, M. and Bover, O. (1995), “Another look at the instrumental variable estimation of error-components models”, Journal of Econometrics, Vol. 68 No. 1, pp. 29-51. Bentolila, S. and Dolado, J. (1994), “Labour flexibility and wages: lessons from Spain”, Economic Policy, Vol. 9 No. 18, pp. 53-99. Blanchard, O. and Portugal, P. (2001), “What hides behind an unemployment rate: comparing Portuguese and US labor markets”, American Economic Review, Vol. 91 No. 1, pp. 187-207. Blanchflower, D. (1991), “Fear, unemployment and pay flexibility”, Economic Journal, Vol. 101 No. 406, pp. 483-96. Blanchflower, D., Oswald, A. and Garrett, M. (1990), “Insider power in wage determination”, Economica, Vol. 57 No. 226, pp. 143-70. Blanchflower, D., Oswald, A. and Sanfey, P. (1996), “Wages, profits and rent-sharing”, Quarterly Journal of Economics, Vol. 111 No. 1, pp. 227-51. Blundell, R. and Bond, S. (1998), “Initial conditions and moment restrictions in dynamic panel data models”, Journal of Econometrics, Vol. 87 No. 1, pp. 115-43. Blundell, R. and Bond, S. (2000), “GMM estimation with persistent panel data: an application to production functions”, Econometric Review, Vol. 19 No. 3, pp. 321-40. Blundell, R., Bond, S. and Windmeijer, F. (2000), “Estimation in dynamic panel data models: improving on the performance of the standard GMM estimators”, in Baltagi, B. (Ed.), Advances in Econometrics, Volume 15: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, Vol. 15, JAI Elsevier Science, Amsterdam, pp. 53-91.

Forces in wage bargaining

43

IJM 29,1

44

Cardoso, A. and Portugal, P. (2005), “Contractual wages and the wage cushion under different bargaining settings”, Journal of Labor Economics, Vol. 23 No. 4, pp. 875-902. Forslund, A. (1994), “Wage setting at the firm level -insider versus outsider forces”, Oxford Economic Papers, Vol. 46 No. 2, pp. 245-61. Graafland, J. (1992), “Insiders and outsiders in wage formation: the Dutch case”, Empirical Economics, Vol. 17 No. 4, pp. 583-602. Gregory, R. (1986), “Wages policy and unemployment in Australia”, Economica, Vol. 53 No. 210, pp. S53-S74. Hildreth, A. and Oswald, A. (1997), “Rent-sharing and wages: evidence from company and establishment panels”, Journal of Labor Economics, Vol. 15 No. 2, pp. 318-37. Holmlund, B. and Zetterberg, J. (1991), “Insider effects in wage determination – evidence from five countries”, European Economic Review, Vol. 35 No. 5, pp. 1009-34. Holzer, H. and Montgomery, E. (1993), “Asymmetries and rigidities in wage adjustments by firms”, Review of Economics and Statistics, Vol. 75 No. 3, pp. 397-408. Johansen, K. (1996), “Insider forces, asymmetries, and outsider ineffectiveness: empirical evidence for Norwegian industries 1966-87”, Oxford Economic Papers, Vol. 48 No. 1, pp. 89-104. Layard, R., Nickell, S. and Jackman, R. (1991), Unemployment: Macroeconomic Performance and the Labour Market, Oxford University Press, Oxford. Lever, M. and van Werkhooven, J. (1996), “Insider power, market power, firm size and wages: evidence from Dutch manufacturing industries”, Labour Economics, Vol. 3 No. 1, pp. 93-107. Lindbeck, A. and Snower, D. (1985), “Explanations of unemployment”, Oxford Review of Economic Policy, Vol. 1 No. 2, pp. 34-59. Lindbeck, A. and Snower, D. (1986), “Wage setting, unemployment, and insider-outsider rela-tions”, American Economic Review, Vol. 76 No. 2, pp. 235-9. Lindbeck, A. and Snower, D. (1988), “Cooperation, harassment, and involuntary unemployment: an insider-outsider approach”, American Economic Review, Vol. 78 No. 1, pp. 167-88. Lindbeck, A. and Snower, D. (2002), “The insider-outsider theory: a survey”, IZA Discussion Paper No. 534, IZA, Bonn. Mairesse, J. and Hall, B. (1996), “Estimating the productivity of research and development in French and US manufacturing firms: an exploration of simultaneity issues with GMM methods”, in Wagner, K. and van Ark, B. (Eds), International Productivity Differences and Their Explanations, Elsevier Science, Amsterdam, pp. 283-315. Nickell, S. and Kong, P. (1992), “An investigation into the power of insiders in wage determination”, European Economic Review, Vol. 36 No. 8, pp. 1573-99. Nickell, S. and Wadhwani, S. (1990), “Insider forces and wage determination”, Economic Journal, Vol. 100 No. 401, pp. 496-509. Nickell, S., Vainiomaki, J. and Wadhwani, S. (1994), “Wages and product market power”, Economica, Vol. 61 No. 244, pp. 457-73. Organization for Economic Cooperation and Development (OECD) (1999), OECD Employment Outlook, OECD, Paris. Solow, R. (1985), “Insiders and outsiders in wage determination”, Scandinavian Journal of Economics, Vol. 87 No. 2, pp. 411-28.

Teulings, C. and Hartog, J. (1998), Corporatism or Competition? Labour Contracts, Institutions and Wage Structures in International Comparison, Cambridge University Press, Cambridge. Vareja˜o, J. and Portugal, P. (2006), “Employment dynamics and the structure of labor adjusment costs”, IZA Discussion Paper No. 1922, IZA, Bonn. Wulfsberg, F. (1997), “An application of wage bargaining models to Norwegian panel data”, Oxford Economic Papers, Vol. 49 No. 3, pp. 419-40. Further reading Blanchard, O. and Summers, L. (1986), “Hysteresis and the European problem”, NBER Macro-economics Annual, MIT Press, Cambridge, MA. Gottfries, N. and Horn, H. (1987), “Wage formation and unemployment persistence”, Economic Journal, Vol. 97 No. 388, pp. 877-84. Appendix 1. Variables: definition and source . Average employment. Defined as the mean between the number of workers at the beginning of the year and the number of workers at the end of the year; Social Audit. . Wages. Annual real labor cost (base wage þ regular paid benefits and premiums) divided by average employment; Social Audit. . Nominal productivity. Annual sales at constant prices divided by average employment; Quadros de Pessoal and Social Audit. . Market share. Total sales in each firm divided by total sales in the sector defined at five digits according to Portuguese Classification of Economic Activities (CAE); Quadros de Pessoal. . Growth permanent employment. Measured as the annual rate change in the total number of permanent employees; Social Audit. . Proportion of temporary employees. The number of temporary employees in the end-of-year count as a proportion of total employment in the end-of-year count; Social Audit. . Labor utilization rate. The ratio between the number of total hours actually worked and the maximum annual potential of worked hours; Social Audit. . Layoff rate. Total number of workers (permanent and temporary) who left the firm involuntarily over the year divided by average employment; Social Audit. . Outside wage. Aggregate real wage by region (defined at NUTS III) excluding the firm’s own wage (per employee); Quadros de Pessoal. . Regional unemployment rate. Defined at the level of NUTS II; Employment Survey – (Instituto Nacional de Estatstica (INE). . Education. Five educational levels were defined (proportion of workers) – primary and less than primary (the omitted category), preparatory and lower secondary, upper secondary, college and others (a residual category); Social Audit. . Qualification. Six qualification levels were defined (proportion of workers) – top executives, intermediary executives, supervisors, skilled and highly skilled professionals, semi-skilled and unskilled professionals and apprentices (the omitted category); Social Audit. . Price deflator. Consumer Price Index (1991 ¼ 100); Consumer Price Index – INE.

Forces in wage bargaining

45

IJM 29,1

Note: all variables, except growth permanent employment and the proportion of temporary employees, are in logs; data sources are in italics. Appendix 2. Previous research

46

l Firm-level data (manufacturing) Nickell and Wadhwani (1990)

0.08-0.15

Nickell et al. (1994)

0.15-016

Bentolila and Dolado (1994) Forslund (1994) Wulfsberg (1997) Industry-level data Holmlund and Zetterberg (1991)

0.05-0.07 0.07

Nickell and Kong (1992)

0.07-0.12 0.00-0.03 0.03-0.04 0.03-0.04 0.00-0.01 0.00-0.00 0.12-0.15 0.04-0.10 0.48-0.49 0.30-0.38 0.10-0.11 0.04-0.04 0.07-0.20 0.03-0.17 0.33-0.38 0.22-0.25 0.20-0.22 0.19-0.23 0.02-0.50a

Johansen (1996)

0.16-0.25

Teulings and Hartog (1998)

Table AI. Estimates of l, the long-run elasticity of firm (industry) wages with respect to firm (industry) revenue per employee

0.11

Lever and van Werkhooven (1996)

0.12-0.15 (all firms) 0.197 (large firms) 0.007 (small firms)

Country UK (1975-1982; 1972-1986) (219 firms) UK (1975-1986) (814 firms) Spain (1985-1988) (1,167 firms) Sweden (1984-1988) (128 firms) Norway (1976-1988) (7,323 firms) Sweden (1965-1985) (28 industries) Norway (1965-1982) (27 industries) Finland (1965-1985) (28 industries) Germany (1965-1985) (25 industries) USA (1965-1985) (28 industries) The Netherlands (1965-1985) (13 industries) Japan (1970-1980) (25 industries) Canada (1972-1985) (27 industries) Australia (1975-1985) (21 industries) UK (1961-1985) (14 industries) Norway (1966-1987) (117 industries) The Netherlands (1974-1986) (68 industries)

Notes: a l for each industry; in Holmlund and Zetterberg (1991) and Teulings and Hartog (1998), for each country, the first range for l refers to trend productivity and the second to industry relative price

Forces in wage bargaining

Appendix 3. GMM estimates: manufacturing firms

Explanatory variables Wages lagged (wijt21 2 w
jt21 ) Nominal productivity (pijt þ yijt 2 nijt 2 w
jt ) Growth permanent employment (Dnpijt) Proportion of temporary employees (fijt) Layoff rate (layoffijt2 1) Labor utilization rate (lurijt2 1) Market share (MSijt2 1) Regional unemployment rate (Ujt) Constant Time dummies Wald test of joint significance ( p-value) Sargan ( p-value) m1 ( p-value) m2 ( p-value) NT

DIF-GMM

SYS-GMM

0.011 (0.3) 0.105 * * (2.1) 2 0.013 (2 1.1) 0.109 * * (2.0) 2 0.002 (2 0.6) 2 0.140 (2 1.3) 0.003 (0.6) 0.011 (0.2) 0.028 (3.9) Yes * 35.5 (0.0) 83.3 (0.101) 2 2.9 (0.0) 0.8 (0.439) 2,164

0.252 * (8.5) 0.183 * (8.1) 20.018 * * (22.4) 20.021 (20.5) 20.020 * (25.2) 0.516 * (4.0) 0.010 * * (2.0) 20.098 * (25.4) 20.570 * (24.9) Yes * 1,817.4 (0.0) 117.4 (0.045) 2 3.5 (0.0) 1.2 (0.245) 2,673

Note: Dependent variable: wages (wijt 2 w
jt ); see notes to Table II; each regression includes five educational levels and six qualification levels

About the authors Anabela Carneiro studied economics at the Faculty of Economics, University of Porto, where she received her PhD in 2003. She is currently an Assistant Professor at the Faculty of Economics, University of Porto, teaching Labor Economics to undergraduates and Microeconometrics to the Masters and Doctoral Courses. She is also a research member of CETE. Her main area of interest is applied labor economics, focusing on issues of wage determination, worker displacement and immigration. Anabela Carneiro is the corresponding author and can be contacted at: [email protected] Pedro Portugal studied economics at the University of South Carolina, where he received his PhD in 1991. He is currently a senior researcher at the Bank of Portugal and Full Professor at the Faculdade de Economia da Universidade Nova de Lisboa. He is also an elected member of the executive committee of the European Association of Labour Economists (EALE). His main area of interest is applied labour economics, focusing on issues of microeconomics of unemployment, unemployment compensation, wage bargaining, job security, and worker displacement.

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints

47

Table AII. GMM estimates of wage equation (1994-1999) measuring insider power and insider forces

The current issue and full text archive of this journal is available at www.emeraldinsight.com/0143-7720.htm

IJM 29,1

Worker churning and firms’ wage policies

48

Queen Mary, University of London, London, UK, IZA, Bonn, Germany and CEG-IST, Lisbon, Portugal

Pedro S. Martins

Abstract Purpose – The purpose of this paper is to provide empirical evidence of a causal nature about the relationship between wages and churning (“excessive” worker turnover). Design/methodology/approach – Matched employer-employee panel data from Portugal, covering the period 1986-2000 are used in the study. Econometric methods are also used, including random effects tobit models, fixed effects and instrumental variables. Findings – Unlike in previous research (which typically does not consider causal relationships), the paper presents evidence that wages do not necessarily decrease the amount of churning. If employers are forced to increase pay, they may respond by hiring different workers. Detailed evidence about the nature of job and worker flows and churning levels across industries is presented. Research limitations/implications – Future research should examine the paths of workers whose wages are affected by collective bargaining. Practical implications – The paper provides additional evidence that effort may not be particularly sensitive to wages in some industries/occupations. The should be a better understanding of role of wages in personnel policies. Originality/value – This paper is probably the first that seeks to examine the causal relationship between wages and churning. The results will be of interest to labour economists and human resource management experts. Keywords Employee turnover, Pay policies, Portugal Paper type Research paper

1. Introduction One of the several striking stylised facts that has emerged from the literature on job flows (Davis et al., 1996) is the large extent of worker churning (the excess of worker reallocation with respect to job reallocation). Contrary to what one may expect, even firms that exhibit stable levels of employment typically display high levels of hires and separations. Moreover, firms that are increasing the size of their workforce can exhibit a considerable number of separations, while hirings also tend to coexist with separations in firms that are cutting employment. For instance, in a well-known study, Burgess et al. (2000) document substantial levels of worker churning, at 12.1 per cent in manufacturing and 22.8 per cent in non-manufacturing. These churning levels represent, respectively, 61.9 per cent and 70.4 per cent of total worker flows, implying that most hirings and separations that occur within firms do not translate into changes in firm size (Hamermesh et al., (1996); International Journal of Manpower Vol. 29 No. 1, 2008 pp. 48-63 q Emerald Group Publishing Limited 0143-7720 DOI 10.1108/01437720810862001

The author would like to thank, without implicating, Harald Dale-Olsen, Pedro Portugal, the guest editors (Ana Rute Cardoso and Chiara Monfardini) and a referee for their very useful comments and Banco de Portugal for logistical assistance.

Albaek and Sorensen (1998); Abowd et al. (1999); Burgess et al. (2001); and Ilmakunnas and Maliranta (2005), present additional evidence of churning)[1]. The importance of churning in most firms’ employment policies raises important questions about the quality of the matching between employers and employees that is achieved by the labour market. For instance, since hirings or separations entail costs for both employers and employees, and low average tenure levels are typically detrimental to the acquisition of productivity-enhancing firm-specific skills, one may expect that the market would adjust in such a way that churning would not be such an important phenomenon. For instance, as firms age and gain experience, they would presumably be able to improve their hiring and retention practices, in order to minimise churning. In any case, it is important to acknowledge the role of wages upon the amount of churning present in the labour market. For instance, one of the mechanisms that may generate efficiency wages (Akerlof and Yellen, 1986) involves the reduction in turnover costs achieved by increasing pay. However, little is known in both the labour economics and the human resource management literatures about the specific trade-offs between wage and churning costs faced by firms. More specifically, there is apparently no evidence about whether a random firm that were to increase its wages would typically achieve a reduction in churning. Indeed, while there is plenty of evidence about the association between different HRM variables and different firm performance indicators, there is a disappointing lack of evidence about causal relationships (Guest et al., 2003 (some exceptions include Lazear, 1995, and Bandiera et al., 2005)). This paper seeks to fill a part of such large gap in our understanding of the causal impact of personnel policies by focusing specifically on the relationship between churning and pay (see also Barth and Dale-Olsen, 1999; Dale-Olsen, 2006). Our analysis is implemented by considering long and detailed Portuguese matched employer-employee panel data and different identification methods, including instrumental variables. After describing evidence about job and worker flows and churning for the entire economy (including the services sector, unlike many related papers), we present regression results that explain the role of wages upon the variability of churning levels across and within firms over time. As in other studies, we find, at first, that churning is negatively related to wages. This result still holds when considering only within-firm time variation. However, because the wages paid by firms are, at least to some extent, a choice variable, such estimates cannot necessarily be interpreted as causal parameters. Once we try to tackle such non-randomness, we obtain our main result: by instrumenting wages using variation in wages driven by collective bargaining contracts, we find that the relationship between wages and churning is actually non-negative: either significantly or insignificantly positive, depending on the estimates. The remaining of the paper is as follows: section 2 presents the data and some statistics about job and worker flows. Section 3 describes the results, based on different estimation methods and samples. Finally, section 4 offers a short summary. 2. Data and descriptive statistics We employ the “Quadros de Pessoal” (Personnel Records) data set, an annual census of all firms based in Portugal that have at least one employee. The census, conducted by the Ministry of Employment, requires that each firm provides detailed information

Worker churning and firms’ wage policies 49

IJM 29,1

50

about itself (size, industry, region, age, sales, etc.) and also about each one of its employees (gender, age, schooling, tenure, wages, etc.). Moreover, each firm and each employee is assigned a unique identifier, allowing one to follow them over time. In this paper, we use data from all years from 1986 to 2000 (except 1990, when only firm-level data are available). “Quadros de Pessoal” can thus be characterised as a matched employer-employee panel data set covering a very large share of the entire labour market over a relatively long period. (The groups of workers not covered are the unemployed, the civil servants, the self-employed and the informal workers.) The data used here include information about 2.4 million firms-year and almost 30 million workers-year. Moreover, because the main goals of the census are to check compliance with employment law and to provide statistical evidence about the labour market, great care is devoted to data quality[2]. All flow variables are defined in the way that has become standard in the literature, as described in Davis et al. (1996). Each rate is constructed by dividing a given flow (job creation, for instance) by the average employment of the firm over the two periods analysed. For instance, the job creation rate is defined as JC ¼ ðLt 2 Lt21 Þ=ð0:5ðLt þ Lt21 ÞÞ, if Lt . Lt21 , or 0, if Lt , Lt21 , in which Lt stands for the firm size in period t. Similarly, the job destruction rate is defined as JD ¼ ðLt21 2 Lt Þ=ð0:5ðLt þ Lt21 ÞÞif Lt , Lt21 , or 0, if Lt . Lt21 , while the net job creation rate (NJCR) corresponds to JC 2 JD and the job reallocation rate (JR) is JC þ JD. The hiring rate is H ¼ Hiringst;t21 =ð0:5ðLt þ Lt21 ÞÞ, in which Hiringst,t2 1 denotes the number of workers present in the firm in period t but not in period t 2 1, while the separation rate is S ¼ Separationst;t21 =ð0:5ðLt þ Lt21 ÞÞ, in which Separationst,t2 1 denotes the number of workers present in the firm in period t 2 1 but not in period t. Finally, the worker reallocation rate (WR) is H þ S, and the churning rate (CR) is defined as WR 2 JR. All descriptive statistics and regression results are carried out by weighting each firm-level observation by its average employment. We follow Blanchard and Portugal (2001) (see also Vareja˜o, 2003) in the computation of worker flows and classify as (new) hirings all workers whose date of hiring was subsequent to the census date of the previous period (March, up to 1993, and October, from 1994 onwards). Since DL ; H 2 S, in which L is firm size, H is hirings and S separations, the number of separations can be defined as H 2 DL, the difference between the total number of hirings in that firm-year and the change in firm size[3]. Table I presents some descriptive statistics. Given that they refer to the entire population of Portuguese firms with at least one employee, they may deserve some particular attention. The first two columns refer to all firms in all years (a total of 2,899,846 firms-year, including interpolations[4]). The remaining four columns refer to 1987 and 2000 (given that 1986 is the first year in our data, it is not possible to compute job and worker flows variables for that year), which include, respectively, 128,754 and 301,607 firms. Average firm size is 1,032 – in other words, given that our statistics are weighted by firm size, an average worker has 1,031 colleagues in her firm. 3 per cent of all firms-year are new firms, while 5 per cent are in their last year in the data. About 9 per cent of firms are foreign owned (defined as when at least 50 per cent of the equity of the firm is held by foreign investors). 39 per cent of the workforce is female and their average schooling attainment is 6.7 years. They have 24 years of experience and

Variable Firm size First year Last year Foreign firm Female Schooling Experience Tenure Months since last promotion Job creation rate Job destruction rate Net job creation rate Hiring rate Separation rate Worker reallocation rate Monthly pay Hourly pay Binding wages

All firms Mean Std dev. 1,032.0 0.03 0.05 0.09 0.39 6.70 24.08 99.65 56.30 0.14 0.12 0.02 0.23 0.22 0.46 738.22 4.54 0.18

3,521.0 0.18 0.21 0.29 0.31 2.38 7.03 71.20 49.19 0.37 0.34 0.54 0.40 0.36 0.52 446.55 3.40 0.23

1987 Mean Std dev. 1,715.5 0.04 0.04 0.09 0.34 5.81 24.99 111.24 56.34 0.13 0.09 0.04 0.21 0.17 0.38 595.45 3.56 0.19

5,155.3 0.19 0.19 0.28 0.29 1.92 6.44 66.57 41.19 0.38 0.31 0.51 0.40 0.32 0.49 309.45 2.14 0.21

2000 Mean Std dev. 688.5 0.04 0.00 0.11 0.42 7.57 23.55 89.46 58.22 0.15 0.18 -0.03 0.26 0.27 0.52 813.95 5.10 0.18

2,285.8 0.19 0.00 0.31 0.32 2.64 7.43 69.32 50.18 0.39 0.49 0.67 0.41 0.44 0.57 490.96 3.29 0.25

Notes: Author’s calculations based on “Quadros de Pessoal” data. “Firm size” refers to the number of workers in each firm; “first year” and “last year” are dummy variables taking value one when the firm appears for the first or last time in the data, respectively; “foreign firm” is a dummy variable taking value one if at least 50 per cent of the equity of the firm is held by foreign investors; “schooling” refers to the years of schooling attained by each worker; “tenure” is measured in months; “monthly” and “hourly pay” are measured in 2004 euros; “binding wages” refer to the percentage of the workforce in each firm that is paid the wage determined by collective bargaining (this is measured adapting the approach of Cardoso and Portugal (2005)), considering only the worker’s occupation or his/her occupation and industry (see main text for more details). Job and worker flow rates follow the standard definition in the literature (more details in the main text)

almost 100 months of tenure, while their last promotion occurred, on average, 56 months before the date of the census. Average job creation rate is 14 per cent, while job destruction rate is 12 per cent, resulting in a net job creation rate of 2 per cent. Hiring and separation rates are almost twice that, at 23 per cent and 22 per cent. It is also noticeable all four rates increase significantly from 1987 to 2000, particularly the job destruction and the separation rates (from 9 per cent to 18 per cent and from 17 per cent to 27 per cent, respectively). Figure 1 presents the distribution of net job creation rates in 2000 (similar figures could be presented for the remaining years). As documented in other studies, we find a large concentration around zero, the modal category, and at the two spikes at 2 2 and 2 (firm deaths and births, respectively). Except for those two extreme cases, very few firms (almost none) exhibit changes in employment of less than (greater than) the smaller spikes at 2 2/3 (2/3). Coming back to Table I, one can calculate churning rates at an average of about 20 per cent, corresponding therefore to almost half of the worker reallocation rate. Moreover, we also find that the distribution of the churning rates is highly skewed. Out of the 2.3 million firms-year for which churning can be calculated, churning is greater than zero in only about 26 per cent of the cases (not weighting by size). From Figure 2,

Worker churning and firms’ wage policies 51

Table I. Descriptive statistics

IJM 29,1

52

Figure 1. Distribution of net job creation rates, 2000, all firms

Figure 2. Distribution of churning rates, 2000, all firms

which presents the distribution of churning rates in 2000, we can also see that only a very limited number of firms exhibit churning rates above 50 per cent. Moreover, gross monthly pay is on average e738 (2004 prices), ranging from e595 in 1987 to e814 in 2000. Hourly pay also increases, at a slightly higher rate. The last row refers to the variable that will be used as an instrument later in the paper. The variable is the percentage of workers in each firm that earn a base wage that is the modal wage in the workers’ occupation (defined at the four-digit level). On average, about 18 per cent of workers in each firm (again, weighted by firm size) earn those modal wages. Table II presents more detailed results about the flows of jobs and workers across firms. In the first table, the firms are divided into their sector of activity (agriculture, manufacturing and services)[5]. As in other studies, the services sector exhibits much higher rates of job creation and destruction and of worker hirings and separations – and therefore the services sector also exhibits higher rates of job and worker reallocation. For instance, worker reallocation is 51.5 per cent in services and only 39.2 per cent in manufacturing. However, we find that the agriculture sector is characterised by even higher job and worker reallocation rates. Finally, another important result is that, while churning is much higher in services than in manufacturing, its share in total worker reallocation is very similar in the two sectors, at about 50 per cent[6].

Worker churning and firms’ wage policies 53

3. Results In this section we present the main results of the paper. These results involve the estimation of a simple reduced-form linear equation that relates the churning rate of a firm in a given year with the average wage paid by that firm to its workers plus a number of controls. These controls include different measures of firm heterogeneity that may help explaining churning rates, including firm and time fixed effects. The equation is as follows: C it ¼ bW it þ X it d þ gi þ dt þ 1it :

ð1Þ

Cit is the churning rate of firm i in period t, Wit is the average wage, and Xit are firm-year characteristics (the percentage of women in the workforce, average education and average experience of the workers, the levels of sales and of equity and log firm size), g and d are firm and time fixed effects, respectively. As suggested above, the firm

Job creation Job destruction Net job creation Job reallocation Hiring Separation Worker reallocation Worker churning Worker churning/worker reallocation

Agriculture

Sector Manufacturing

Services

0.198 0.194 0.005 0.392 0.340 0.351 0.691 0.265 0.439

0.109 0.113 20.004 0.222 0.188 0.203 0.392 0.163 0.518

0.166 0.123 0.044 0.289 0.275 0.241 0.515 0.218 0.502

Notes: Job and worker flow rates follow the standard definition in the literature (more details in the main text)

Table II. Job and worker flows rates, 1986-2000, by sector

IJM 29,1

54

fixed effects are of particular importance, as otherwise one would have to assume that the wage policy of the firm is chosen randomly in order to interpret the b coefficient as the causal impact of wages upon churning. The estimation method also deserves some discussion. Given the large number of zeros in the dependent variable, a tobit model would be appropriate. However, to the best of our knowledge, an estimator for this model with fixed effects has not yet been developed. We are also not aware of tobit models, even with random effects, that could incorporate instrumental variables. Because of those technical constraints, our first approach is to employ a random effects tobit model. We then adopt a standard OLS model, first excluding firm fixed effects, and then including them. Finally, given that we are also concerned about the endogeneity of the within-firm time variation of wages, we also consider a model in which we instrument the change in wages. As a robustness test, we also replicate the previous models considering only firms that have a positive level of churning, including with a specification based on log churning and log wages. Our instrument is the share of workers being paid collective bargaining contractual wages (henceforth SWCW). The choice of this variable is motivated by the very interesting work by Cardoso and Portugal (2005), who study the implications of collective wage bargaining upon wage determination. In order to derive their results, Cardoso and Portugal (2005) assume that bargained wages – basically an industry/occupation specific minimum wage – can be defined as the modal wage for each worker’s job category, an assumption that receives support from the sample of jobs for which they collect wage information from collective agreements. Our specific approach underpinning our adoption of the instrument involves arguing first that the greater the SWCW, the lower the average wages paid by the firm. This is straightforward because those contractual wages are wage floors – firms are free to pay above those agreed levels and most workers do indeed earn more). More important, we also argue that the SWCW does not affect directly the level of churning of the firm. In other words, it is not the share of workers that determines their mobility decisions, it is instead the wages themselves. This assumption corresponds to the exclusion restriction. To sum it up, all the effect from SWCW upon churning in each firm is assumed to take place through the wages paid by the firm; the share is only a predictor, which is also influenced by factors outside the control of the firm, namely the bargaining process that takes place between different employers association and different unions. Moreover, as suggested before, we also expect the relationship between the instrument and wages to be negative: controlling for other variables, the greater the SWCW, the lower the average wage paid by the firm. Table III presents the first set of results. All equations consider all firms (including those with zero churning rates). The first column is based on a tobit model with random effects. We find the expected significantly negative relationship between average pay and churning rates. The following three columns are based on OLS. Column 2 ignores the possible endogeneity of wages across firms while column 3 ignores the possible endogeneity of wage differences within firms, over time. In both cases, we again obtain very significant, negative coefficients of wages. For instance, the result of column 3 indicates that for each e1,000 extra that each worker earns per month, there is a decline of 1.7pp in that firm’s churning rate. This negative

1 Average pay Fixed effects IV R-squared No. obs.

2 0.029 * [0.000]

931,093

2 2 0.076 * [0.000] 0.119 931,093

3 2 0.017 * [0.000] £ 0.571 931,093

4 0.320 * [0.000] £ £ 2 0.022 931,093

Notes: Dependent variable: churning rate; years used: 1987-2000, except 1990. Column 1 is based on a random effects tobit model. Column 2 is based on pooled OLS. The coefficient of the instrumental variable in the first-stage equation is 2 0.067 (t-ratio: 2 9.44), with a partial R 2 of 0.0022 and an F-statistic of 89.05. Controls used in all regressions: year dummies, percentage of women, average education, average experience, sales, equity and log firm size. The explanatory variable is measured in thousands of euros (2004 prices). Only firms present three or more years over 1987-2000 are included in the analysis. Robust standard errors. * p , 0:001

Worker churning and firms’ wage policies 55

Table III. Impact of wages on churning (all firms)

relationship is consistent with the pattern documented in Figure 3, which depicts the level of churning and average pay across all industries in 2004: the fitted line, weighted by industry size, is clearly downward sloping. However, in column 4, when we instrument wage differences with the shares of workers that are paid collective bargaining wages, we find that the coefficient remains

Figure 3. Churning rates and average pay per industry, 2000

IJM 29,1

56

statistically significant, but it is now positive. At 0.32, the coefficient indicates that there is an increase of 32pp in a firm’s churning rate for each e1,000 increase of average pay. Although the contribution of the instrument in the first stage is not particularly strong (a partial R 2 of 0.0022), the instrument is significantly negative, as expected from our discussion above. Overall, the result suggests that, contrary to previous findings, if firms pay higher wages (because of exogenous reasons), then churning may actually increase. Roughly speaking, a doubling of wages would lead to the doubling of churning. In order to test the robustness of this (surprising) result, we repeat the analysis of the three last columns of Table III but considering only firms-year in which churning is positive (see Table IV). This selection of the sample is motivated by the fact that, as mentioned before, a considerable number of firms do not exhibit churning. In any case, we still find that the same pattern of results emerges, with significantly negative coefficients in the first two specifications (those without instruments) and a positive coefficient (although now insignificant) when wages are instrumented. Finally, we consider the same sample of firms-years (those with positive levels of churning) but now adopting a log-log specification. The same pattern emerges again, with negative elasticities for the first two specifications and an insignificant, but also positive coefficient when wages are instrumented. In both cases, the instruments are significantly negative in the first-stage equations. 3.1. Interpretation Our finding of a non-negative causal relationship between wages and churning (which is even significantly positive in some specifications) may at first seem counterintuitive. Presumably, if a firm were to increase the wages paid to its workers due to forces outside the firm’s control, then workers would be less interested in leaving their jobs. Therefore, churning – understood as “replacement hiring” – , would fall. This would then generate a negative relationship between pay and churning. Dependent variable Average pay

1 20.106 * * [0.006]

2 Churning rate 20.025 * * [0.005]

3

Table IV. Impact of wages on churning (only firms with positive levels of churning)

5 Log churning rate

20.511 * * [0.027] £ 0.186 413,246

0.694 413,246

6

0.164 [0.123]

Log average pay Fixed effects IV R-squared No. obs.

4

£ £ 0.057 387,468

0.312 413,246

20.120 * [0.038] £ 0.710 413,246

0.167 [0.845] £ £ 0.104 387,468

Notes: Dependent variable: churning rate or log churning rate; years used: 1987-2000, except 1990. The coefficient of the instrumental variable in the first-stage equation of column 3 is 20.064 (t-ratio: 25.59), with a partial R 2 of 0.0017 and an F-statistic of 31.26. For column 6, the instrumental variable in the first-stage equation coefficient is 20.055 (t-ratio: 26.72), with a partial R 2 of 0.0016 and an F-statistic of 45.09. Controls used in all regressions: year dummies, percentage of women, average education, average experience, sales, equity and log firm size. The explanatory variable is measured in thousands of euros (2004 prices). Only firms present three or more years over 1987-2000 are included in the analysis. Robust standard errors. * p , 0:01; * * p , 0:001

In our view, one aspect that may be missing in such analysis concerns the behaviour of firms once their workers become better paid due to exogenous forces. According to some efficiency wage models, wage increases may pay for themselves, but only if workers’ effort increase more than proportionately. However, if that were the case, one would presumably expect that firms would be keen to implement those pay rises in the first place, before being forced to do so by virtue of the increase in collective bargaining contractual wages. Moreover, even if firms were not pursuing optimal pay policies, it is still not obvious that collective bargaining wage increases would move them in that direction. Since the outside options of workers earning the minimum wage would necessarily also increase (i.e. the minimum wage in other firms has also increased, due to the collective nature of bargaining), then the incentive for those workers to put in more effort will probably be relatively weak. Our explanation for the non-negative relationship between wages and churning is therefore that workers’ effort may not be sufficiently responsive to wage increases, particularly in the firms most affected by our instrument (a local average treatment effect). Such firms are those that are paying lower wages, so that their wage bill goes up significantly when collective bargaining increases minimum wages. Moreover, if effort does not increase at least in a commensurate way to the increase in pay, then employers may very well prefer to dismiss those workers whose wages exceed their productivity and replace them with more skilled workers. To the extent that not all dimensions of skill are captured in our regression controls, this process of replacement can easily prevent worker churning rates from falling. Such non-decrease of churning will then generate the non-negative association between wages and churning documented in our study, when using instrumental variables. Of course, it would be desirable to present additional empirical evidence about this interpretation in future research. One possible way of achieving that would be to follow the individuals whose pay is most affected by collective bargaining (i.e. those workers that earn the industry/occupation minimum wage). One could then examine whether such workers are also more likely to leave their jobs once their wages are increased by virtue of new collective bargaining agreements, as suggested by our interpretation. 4. Summary Our study examined the relationship between churning and wages, using matched data covering the entire population of Portuguese firms between 1986 and 2000. We believe this relationship deserves attention from the points of view of both labour economics and human resource management: While economists may be concerned about the waste of resources that may occur when workers are being constantly replaced in firms, HRM experts may find it helpful to draw on stronger quantitative evidence about how firms may ensure that employer-employee matches are long-lasting. Given that we are interested in the causal relationship between the two variables, churning and wages, we used within-firm time variation (an improvement upon studies that focus on cross-section variation only). Moreover, we also exploit what we argue is a source of exogenous variation in wages: the share of workers in each firm being paid collective bargaining wages. Using that instrument, we find that the standard negative correlations between wages and churning become a non-negative causal link. We also argue that this finding may be explained by a relatively weak responsiveness of effort

Worker churning and firms’ wage policies 57

IJM 29,1

58

to wages in the firms affected by the instrument and the consequent replacement by employers of priced-out workers by more skilled new hires in such a way that churning does not fall. Notes 1. See also Martins (2006a) for evidence that churning does not only correspond to “replacement hiring”, the most common interpretation in the literature: events involving “job upgrading” or “job downgrading” (i.e. firms changing the profile of their workforces in terms of skills and/or occupations) also explain part of the “excessive” worker turnover (with respect to job flows). 2. Although we do believe data quality is indeed of high standards, there are inevitably some missing or incorrect observations, particularly for smaller firms and/or in the earlier years of the census, when computers were not so widespread. 3. One problem with this approach is that any underestimation of H (for instance, because of missing or incorrect hiring dates) will also lead to the underestimation of S. However, from a more detailed analysis of this issue (see Martin, 2006b), we believe this problem is not particularly relevant in our study. 4. A closer inspection of the data indicated that some firms do not report their information in some years. In order to avoid biases in terms of the artificial definition of too many new and closing firms, the firm size of those firm-years (less than 10 per cent of the firms-year used) was interpolated. 5. Electricity and Gas, and Construction have been included in the Services sector. The agriculture sector also includes Forestry and Fishing. 6. See also the Appendix for a more detailed analysis of differences in these variables across industries. References Abowd, J., Corbel, P. and Kramarz, F. (1999), “The entry and exit of workers and the growth of employment: an analysis of French establishments”, Review of Economics and Statistics, Vol. 81 No. 2, pp. 170-87. Akerlof, G. and Yellen, J. (1986), Efficiency Wage Models of the Labor Market, Cambridge University Press, Cambridge. Albaek, K. and Sorensen, B. (1998), “Worker flows and job flows in Danish manufacturing, 1980-91”, Economic Journal, Vol. 108 No. 451, pp. 1750-71. Bandiera, O., Barankay, I. and Rasul, I. (2005), “Social preferences and the response to incentives: evidence from personnel data”, Quarterly Journal of Economics, Vol. 120 No. 3, pp. 917-62. Barth, E. and Dale-Olsen, H. (1999), “Employer’s wage policy and worker turnover”, in Haltiwanger, J., Lane, J., Spletzer, J., Theeuwes, J. and Troske, K. (Eds), The Creation and Analysis of Linked Employer-Employee Data, Elsevier, North Holland, Amsterdam. Blanchard, O. and Portugal, P. (2001), “What hides behind an unemployment rate: comparing Portuguese and US labor markets”, American Economic Review, Vol. 91 No. 1, pp. 187-207. Burgess, S., Lane, J. and Stevens, D. (2000), “Job flows, worker flows, and churning”, Journal of Labor Economics, Vol. 18, pp. 473-502. Burgess, S., Lane, J. and Stevens, D. (2001), “Churning dynamics: an analysis of hires and separations at the employer level”, Labour Economics, Vol. 8 No. 1, pp. 1-14. Cardoso, A.R. and Portugal, P. (2005), “Contractual wages and the wage cushion under different bargaining settings”, Journal of Labor Economics, Vol. 23 No. 4, pp. 875-902.

Dale-Olsen, H. (2006), “Fringe attraction: compensation policies, worker turnover and establishment performance”, mimeo, ISF, Oslo. Davis, S., Haltiwanger, J. and Schuh, S. (1996), Job Creation and Destruction, MIT Press, Cambridge, MA. Guest, D., Michie, J., Conway, N. and Sheehan, M. (2003), “Human resource management and corporate performance in the UK”, British Journal of Industrial Relations, Vol. 41 No. 2, pp. 291-314. Hamermesh, D., Hassink, W. and van Ours, J. (1996), “Job turnover and labor turnover: a taxonomy of employment dynamics”, Annales d’Economie et de Statistique, Nos 41/42, pp. 21-40. Ilmakunnas, P. and Maliranta, M. (2005), “Worker inflow, outflow, and churning”, Applied Economics, Vol. 37 No. 10, pp. 1115-33. Lazear, E. (1995), Personnel Economics, MIT Press, Cambridge, MA. Martins, P. (2006a), “Worker churning and within-firm job flows”, mimeo, Queen Mary, University of London, London. Martins, P. (2006b), “Inter-firm worker mobility: some stylised facts”, mimeo, Queen Mary, University of London, London. Vareja˜o, J. (2003), “Job and worker flows in high adjustment cost settings”, Portuguese Economic Journal, Vol. 2 No. 1, pp. 37-51. Further reading Lane, J., Stevens, D. and Burgess, S. (1996), “Worker and job flows”, Economics Letters, Vol. 51 No. 1, pp. 109-13. Appendix Table AI decomposes information about job and worker flows and churning rates into two-digit industries for all three sectors. The results below support the findings of considerable heterogeneity in job flows, an important result of Davis et al. (1996). For instance, while net job creation rates in manufacturing are negative over the 1986-2000 period (although at only 2 0.4 per cent), specific industries can exhibit very different patterns. On the one extreme, firms in the “Petroleum and Natural Gas” industry have employment declines of 103 per cent, and those in “Coke and Refined Petroleum” see their number of jobs falling by 12.2 per cent; on the other extreme, the “Recycling” and “Motor Vehicles” industries increase their jobs by 10.5 per cent and 3.2 per cent. Moreover, job reallocation rates also tend to be high although, again, they vary considerably across industries. These results are further evidence of the dynamic nature of labour markets and the constant reshuffling of jobs across firms, within or not the same industries. Table AII presents worker flows across all two-digit industries, over the 1986-2000 period. A striking result from this table is the considerable dispersion of worker reallocation rates across industries. They range between 171 per cent and 162 per cent for “Petroleum and Natural Gas” and “Mining” to as little as 13.1 per cent and 16.9 per cent in “Water” and “Financial Intermediation”, respectively. Churning rates also vary considerably across industries. They range from 0 per cent and 3.1 per cent in “Petroleum and Natural Gas” and “Mining” to as much as 43.1 per cent and 39.5 per cent in “Fishing” and “Other Business Activities” (which includes, for instance, “Cleaning Services” firms), respectively. However, as suggested before, in the comparison between the manufacturing and the services sectors, there is much less dispersion across industries in the ratios between churning and worker reallocation rates (the coefficient of variation drops to about

Worker churning and firms’ wage policies 59

IJM 29,1

60

Table AI. Job and worker flows rates, 1986-2000, by industry

Industry Agriculture Forestry Fishing Mining Petroleum and natural gas Mining of uranium Mining of metal ores Other mining and quarrying Food products and beverages Tobacco products Textiles Wearing apparel Leather, luggage, and footwear Wood and products of wood and cork Paper and paper products Publishing, printing Coke, refined petroleum Chemicals Rubber and plastics products Other non-metallic mineral products Basic metals Fabricated metal products Machinery and equipment Office, accounting and computing machine Electrical machinery Radio, television, communication equipment Medical, precision, optical instruments Motor vehicles Other transport equipment Furniture Recycling Electricity, gas Water Construction Sale, maintenance of vehicles; fuel Wholesale trade Retail trade Hotels and restaurants Land transport Water transport Air transport Auxiliary transport; travel agencies Post and telecommunications Financial intermediation Insurance and pension funding Activities auxiliary to finance Real estate Renting of machinery and equipment Computer and related activities Research and development

Job creation

Job destruction

Net job creation

0.204 0.244 0.124 0.015 0.000 0.082 0.062 0.142 0.098 0.029 0.070 0.144 0.120 0.122 0.070 0.112 0.003 0.072 0.101 0.103 0.072 0.136 0.103 0.215 0.132 0.134 0.080 0.121 0.081 0.126 0.217 0.091 0.039 0.202 0.141 0.152 0.194 0.192 0.103 0.089 0.042 0.125 0.110 0.059 0.075 0.215 0.284 0.185 0.294 0.178

0.194 0.208 0.181 1.529 1.029 0.054 0.153 0.118 0.112 0.066 0.106 0.127 0.109 0.133 0.104 0.097 0.125 0.115 0.100 0.101 0.129 0.119 0.105 0.215 0.102 0.136 0.079 0.090 0.125 0.115 0.112 0.128 0.042 0.152 0.111 0.120 0.136 0.141 0.092 0.133 0.030 0.117 0.120 0.062 0.083 0.117 0.176 0.112 0.127 0.110

0.010 0.036 2 0.057 2 1.513 2 1.029 0.029 2 0.091 0.025 2 0.014 2 0.038 2 0.036 0.017 0.011 2 0.011 2 0.034 0.015 2 0.122 2 0.043 0.001 0.002 2 0.058 0.017 2 0.002 0.000 0.029 2 0.002 0.001 0.032 2 0.044 0.012 0.105 2 0.037 2 0.002 0.049 0.030 0.031 0.058 0.050 0.011 2 0.044 0.013 0.008 2 0.010 2 0.003 2 0.009 0.097 0.108 0.074 0.167 0.068

Job reallocation 0.398 0.452 0.306 1.544 1.029 0.136 0.215 0.260 0.210 0.095 0.176 0.271 0.228 0.255 0.173 0.209 0.129 0.186 0.201 0.204 0.201 0.255 0.208 0.430 0.234 0.270 0.159 0.211 0.206 0.241 0.329 0.219 0.081 0.354 0.252 0.272 0.329 0.333 0.194 0.222 0.072 0.242 0.230 0.121 0.158 0.332 0.460 0.297 0.421 0.287 (continued)

Industry Other business activities Public administration and defence Education Health and social work Sewage and refuse disposal, sanitation Activities of membership organisations Recreational and sporting activities Other service activities Extra-territorial organisations

Industry Agriculture Forestry Fishing Mining Petroleum and natural gas Mining of uranium Mining of metal ores Other mining and quarrying Food products and beverages Tobacco products Textiles Wearing apparel Leather, luggage, and footwear Wood and products of wood and cork Paper and paper products Publishing, printing Coke, refined petroleum Chemicals Rubber and plastics products Other non-metallic mineral products Basic metals Fabricated metal products Machinery and equipment Office, accounting and computing machines Electrical machinery

Job creation

Job destruction

Net job creation

Job reallocation

0.241 0.156 0.139 0.144 0.343 0.119 0.128 0.210 0.146

0.134 0.098 0.088 0.074 0.067 0.109 0.100 0.160 0.149

0.107 0.058 0.051 0.069 0.275 0.010 0.028 0.050 2 0.002

0.374 0.255 0.227 0.218 0.410 0.227 0.227 0.370 0.295

.

Worker Hiring Separation reallocation

Worker churning

Worker churning/worker reallocation

0.338 0.366 0.340 0.032 0.000 0.076 0.117

0.342 0.339 0.421 1.592 1.714 0.114 0.211

0.680 0.705 0.761 1.624 1.714 0.190 0.328

0.247 0.218 0.431 0.031 0.000 0.054 0.115

0.423 0.336 0.609 0.036 0.000 0.071 0.539

0.247

0.236

0.483

0.216

0.484

0.199 0.083 0.126 0.233

0.227 0.120 0.171 0.222

0.426 0.202 0.297 0.455

0.213 0.108 0.119 0.168

0.579 0.507 0.512 0.526

0.207

0.204

0.411

0.172

0.556

0.204 0.126 0.195 0.024 0.122

0.226 0.171 0.191 0.149 0.183

0.430 0.297 0.386 0.173 0.305

0.164 0.121 0.172 0.040 0.119

0.478 0.517 0.529 0.400 0.475

0.187

0.196

0.382

0.176

0.548

0.188 0.130 0.234 0.184

0.196 0.196 0.224 0.194

0.384 0.326 0.458 0.377

0.177 0.120 0.194 0.168

0.545 0.437 0.507 0.515

0.276 0.222

0.288 0.210

0.564 0.432

0.113 0.193

0.339 0.509 (continued)

Worker churning and firms’ wage policies 61

Table AI.

Table AII. Worker flows rates, 1986-2000, by industry

IJM 29,1 Industry

62

Table AII.

Radio, television, communication equip. Medical, precision, optical instruments Motor vehicles Other transport equipment Furniture Recycling Electricity, gas Water Construction Sale, maintenance of vehicles; fuel Wholesale trade Retail trade Hotels and restaurants Land transport Water transport Air transport Auxiliary transport; travel agencies Post and telecommunications Financial intermediation Insurance and pension funding Activities auxiliary to finance Real estate Renting of machinery and equipment Computer and related activities Research and development Other business activities Public administration and defence Education Health and social work Sewage and refuse disposal, sanitation Activities of membership organisations Recreational and sporting activities Other service activities Extra-territorial organisations

Worker Hiring Separation reallocation

Worker churning

Worker churning/worker reallocation

0.187

0.223

0.409

0.123

0.421

0.146 0.181 0.131 0.215 0.308 0.129 0.062 0.358

0.148 0.169 0.181 0.209 0.212 0.100 0.068 0.313

0.294 0.350 0.312 0.424 0.519 0.229 0.131 0.671

0.135 0.143 0.102 0.163 0.181 0.031 0.054 0.301

0.553 0.510 0.407 0.481 0.459 0.370 0.535 0.517

0.231 0.242 0.291 0.359 0.153 0.189 0.101

0.211 0.225 0.248 0.312 0.156 0.249 0.097

0.442 0.467 0.539 0.671 0.309 0.438 0.198

0.178 0.189 0.194 0.318 0.119 0.211 0.127

0.496 0.496 0.432 0.552 0.507 0.599 0.693

0.187

0.196

0.383

0.143

0.314

0.159 0.075

0.171 0.094

0.330 0.169

0.100 0.064

0.594 0.518

0.085

0.127

0.213

0.074

0.395

0.289 0.399

0.202 0.286

0.491 0.685

0.137 0.193

0.317 0.372

0.317

0.254

0.571

0.259

0.514

0.412 0.284 0.440

0.247 0.231 0.348

0.659 0.516 0.788

0.222 0.224 0.395

0.425 0.527 0.543

0.255 0.233 0.216

0.199 0.195 0.154

0.455 0.429 0.370

0.171 0.196 0.148

0.522 0.548 0.512

0.504

0.224

0.728

0.302

0.497

0.171

0.180

0.351

0.121

0.395

0.226 0.318

0.212 0.268

0.438 0.586

0.205 0.181

0.551 0.374

0.202

0.193

0.395

0.068

0.308

Worker churning and firms’ wage policies 63

Figure A1. Job reallocation and worker reallocation, by industry, 2000

half), indicating that the two rates are positively correlated (an unweighted Pearson correlation of 18 per cent). This relationship between churning and worker reallocation rates can also be inferred from Figure A1, which plots job and worker reallocation rates by industry in 2000. Since churning is the difference between the two, Figure A1 suggests that churning increases with worker reallocation.

About the author Pedro Martins is a Senior Lecturer in Economics at the School of Business and Management of Queen Mary, University of London. He is also affiliated with IZA, Bonn, and CEG-IST, Lisbon. Pedro earned his PhD in Economics from the University of Warwick and his “Licenciatura” from the Universidade Nova de Lisboa. His research interests are in labour, education and personnel economics and in issues regarding globalisation. Pedro Martins can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints

The current issue and full text archive of this journal is available at www.emeraldinsight.com/0143-7720.htm

IJM 29,1

The worsening of wage expectations in Italy: a study based on administrative data

64

Elena Giarda PROMETEIA, Bologna, Italy and Department of Statistics, University of Bologna, Bologna, Italy Abstract Purpose – This paper aims to build on existing studies on the relationship between individual wages, age and experience, and provide new evidence on the determinants of wages in Italy. Design/methodology/approach – Wage-age profiles, which include cohort variables to capture generational differences in wages and are characterised by a changing-over-time structure, are estimated by fixed and random effects panel regressions. The analysis exploits a longitudinal dataset of administrative data on wages for the period 1985-1999. Findings – This paper shows that wage to age profiles for different cohorts of workers are not stable over time: although younger generations of Italian workers are benefiting from higher starting wages than older generations, they face the prospect of lower growth of future earnings. It also confirms the existence of a significant supply effect: the bigger the cohort relative to the active population, the smaller the cohort’s gain in terms of wage levels. Finally, it captures the dependence of individual wages on aggregate labour market conditions: individual wages are shown to be negatively related to the unemployment rate and positively related to the union wage index. Research limitations/implications – Although the paper does not propose a novel theoretical approach to individual wage analysis, it demonstrates the benefits of a more integrated empirical analysis of individual wages. Practical implications – The empirical findings suggest that it would be possible and useful to integrate the changing age profiles of individual wages with the estimation and projections of Italian aggregate industry and service sector average wages. Originality/value – The paper provides new evidence on the determinants of the dynamics of individual wages through the estimation of time-varying wage to age profiles of workers in the Italian industry and service sectors. Keywords Pay, Young adults, Older workers, Italy Paper type Research paper

1. Introduction and literature review The aim of this paper is to test the hypothesis of time-varying wage to age profiles and the dependence of individual wages on aggregate labour market indicators, for the

International Journal of Manpower Vol. 29 No. 1, 2008 pp. 64-87 q Emerald Group Publishing Limited 0143-7720 DOI 10.1108/01437720810862010

A previous version of the paper entitled “The worsening of wage expectations in the Italian industry sector: a study based on administrative data” was published as PROMETEIA’s Working Paper No. 0601. The author would like to thank Giovanni Bruno, Claudio Lucifora and Angelina Mazzocchetti for valuable suggestions. The paper also benefited from comments of participants at the XX National Conference of Labour Economics (University of Rome La Sapienza, September 2005) and the IV Brucchi Luchino Workshop (Catholic University of Milan, December 2005). The author acknowledges the comments of two anonymous referees, which resulted in a major revision of the paper. All remaining errors are the author’s responsibility.

Italian labour market. The study combines three different aspects of the determinants of individual wages that are usually considered separately: the dependence of individual real wages on age; the impact of supply-side effects; and the relevance of macroeconomic variables such as the unemployment rate and union wages[1]. The path-breaking studies by Mincer (1958), in the field of human capital, stressed how the flow of future individual incomes depends on the education levels and career choices of workers. In these studies, experience (or age, as a proxy for experience) is the driving force of wage formation. This applies both at time of entry to the job market and throughout the workers’ professional career, when experience is acquired through learning-by-doing and on-the-job training. Following Mincer’s preliminary studies, several authors studied the dynamics of individuals’ wages, emphasising the dependence of wages on age. Creedy and Hart (1979) were among the first to extend the analysis of individual wage-age profiles by including a wage-cohort structure. More recently, in a paper on Canadian earnings, Beaudry and Green (2000, p. 919), analysed weekly earnings by means of cohort averages, taken from survey and census data, in 13 alternate years in the period 1971-1993. They suggest, by extending the analysis to age and cohort interactions, that “recent cohorts . . . experience slower wage growth as they age than earlier cohorts did”. The notion that wage to age profiles may not be invariant over time is also addressed in studies dealing with the estimation of returns to education. As far back as 1979, Psacharopoulos and Layard argued that the gradient of the wage-age profiles was positively correlated with educational attainment (Psacharopoulos and Layard, 1979). Brunello and Comi (2000), in a study that utilises individual cross-sectional data from 11 European countries, from the early 1980s to the early 1990s, show that the slope of the wages to age profiles depends on the investment in education. An important strand in the literature has investigated the effects of supply factors on individual wages, testing the hypothesis that larger cohorts, resulting from demographic cycles, have a negative effect on wages. Korenman and Neumark (2000) provided some new evidence on this hypothesis, confirming Berger’s (1985) results. Using the same data as Welch (1979), but with less restrictive model specifications, Berger concluded that cohort size had the effect of “slowing” earnings growth at the beginning of a worker’s career. Stapleton and Young (1988) found that workers belonging to larger cohorts needed to adjust their educational attainment substantially in order to maximise their lifetime income. Card and Lemieux (2001) explained the wage gap between college and high school in the USA, UK and Canada as equating with changes in the relative supply of highly educated workers across age groups. Fertig and Schmidt (2003) show that the evidence related to cohort size effects is weaker for Europe than for the USA. Brunello and Lauer (2004) found that the impact of cohort size and educational composition of the labour force on wages is relatively larger in southern than in the northern Europe. Brunello et al. (2000) studied ten European countries and came to the conclusion that institutional factors play an important role in determining the evolution of wage differentials among cohorts. Naticchioni et al. (2006, p. 1) use repeated cross-sections for the years 1993-2004 (Bank of Italy data on Italian wages) and find that “returns to education decline over time almost uniformly across the wage distribution”. A parallel field of research, bridging micro and macroeconomics, has concentrated, in Card’s (1995, p. 785) words, on “the examination of the role that local unemployment

Worsening of wage expectations 65

IJM 29,1

66

plays in pay determination”. Card was referring to the seminal work of Blanchflower and Oswald (1994), in which the “the wage-curve” is defined as the relationship between individual wages and unemployment. They estimate an elasticity of individual wages to unemployment of 2 0.10 for the USA and many other countries, including Italy. Baltagi and Blien (1998), for example, found an elasticity of 2 0.07 for Germany, and Bell et al. (2002) estimated a percentage response of wages to changes in unemployment of 2 0.025 for the UK. Very recently, and using administrative data for Italy, Devicienti et al. (2006) estimated an elasticity of 2 0.041 for females and of 2 0.029 for males, for the period 1993-1999. Of course, many scholars have researched this field, and we are not aiming to provide an exhaustive list of examples, as estimation of the wage-curve per se is not the main focus of this study. Lucifora and Origo (1999) stressed the relevance of institutional factors on individual wages. They raise the issue of union versus non-union wages, with respect to wage relativities and wages growth. A similar position is taken in Manacorda (2004), who studies the impact of wage indexation mechanisms (the scala mobile) on Italian wages in the period 1977-1993. Union wages are to be considered a conditioning factor of the macro-economic environment in which individual wages are determined. Regarding Italy specifically, Lucifora and Rappelli (1995), Lucifora and Vignocchi (1997), and Brugiavini and Peracchi (2003) use micro-data drawn from the National Institute of Social Security (INPS) administrative archives, to test the hypothesis of wages being dependent on age. In Lucifora and Rappelli’s (1995) study, the cohort dummies are all shown to be positive and increasing: younger cohorts benefit from higher wages at entry to the labour force than older ones. Better job qualifications, larger sized companies and location in northern regions were also found to generate higher wage levels. Women face lower entry wages than men, though, as Lucifora and Vignocchi (1997) demonstrate, the gap between women’s and men’s entry wages is progressively decreasing. Brugiavini and Peracchi (2003), in introducing age at degree 3, illustrate that the prospects for earnings growth are less rosy for women. The specification of their model also includes variables such as the imputed years of contribution (estimated on the basis of the Bank of Italy’s survey on households’ income and wealth) to account for on-the-job experience gained outside the INPS social security programme. Biagi (2003), based on Beaudry and Green’s (2000) methodology, analyses synthetic cohort average wages, through repeated cross-sections of survey data. He provides some evidence of a time-varying wage to age structure for male workers and of the evening out of wage to age profiles among more recent generations. The modest statistical significance of the interaction between age and cohort effects prompted us to investigate the issue further. Varying wage to age profiles are at the basis of our study, which focuses on workers in the Italian industry and service sectors. The paper provides new evidence on the determinants of the dynamics of individual wages exploiting a longitudinal dataset of administrative data on wages for the period 1985-1999. Although it does not propose a novel theoretical approach to individual wage analysis, it demonstrates, building on existing studies, the benefits of a more integrated empirical analysis of individual wages. First, it tests the assumption that the age profiles of individual wages are time-dependent: due to a variety of factors – the most prominent perhaps being the progressive slowdown of the Italian economy – younger generations of Italian workers face a lowering of their life-time wage expectations. Second, it investigates the effects

of supply factors, such as the changing age structure of the active population, on individual wages. Third, it searches for evidence on the effects of macroeconomic conditions in the labour market by testing the relevance, in the individual wage equation, of variables such as unemployment rate and real union wages index. The work is organised as follows. Section 2 provides a description of the dataset used in the analysis. Section 3 presents ordinary least squares (OLS) and random effects estimations of the cohort model, which includes changing-over-time wage to age profiles. In addition to the usual control variables, the model specification includes number of years of social security contributions as a proxy for on-the-job experience. Section 4 extends the analysis by testing the effects of cohort size (as an indicator of supply-side effects), and of aggregate labour market variables (regional unemployment rates and union wage index). Section 5 concludes with a summary of the main findings.

Worsening of wage expectations 67

2. The data The dataset used in this paper is the Work Histories Italian Panel (WHIP)[2], a database built out of INPS administrative archives, which contains the work histories of a sample of workers from 1985 to 1999. The paper concentrates on industry and service sector workers, aged 18-60, and born between 1925 and 1981. The number of individuals in this extracted sample is 99,825 (84.7 per cent of the original dataset), among which 38,714 are females (38.8 per cent of the total) and 61,111 are males (61.2 per cent of the total). Table I presents the cohort composition. For those individuals with more than one employment spell within the same year, the total yearly wage is computed by adding together all intra-year wages[3]. As far as individual characteristics (work area, sector, qualifications, etc.) are concerned, those that prevailed for the longest in any intra-year spell were applied. Mean annual nominal and real wages for full-time workers in the sample (Table II) show high variability, with growth rates of around 2 per cent in real terms at the end of the eighties and negative growth rates in the second half of the nineties. Females’ average weekly real wages for the entire 15-year period are consistently lower than males’ (Table III). The average wage by age class from 1985 to 1999 remains relatively stable over time for young workers, but shows significant increases for older workers (Table III). Weekly real wages in the industry sector are an average 3.6 per cent higher than in the service sector, with a percentage difference between the two sectors of as much as 17.5 per cent for white collar workers (Table IV). Earnings in northern Italy Cohort

Year of birth

Individuals

%

Observations

%

1 2 3 4 5 6 7 8 9 Total

1925-1935 1936-1940 1941-1945 1946-1950 1951-1955 1956-1960 1961-1965 1966-1970 1971-1981

6,383 5,763 6,170 8,147 8,459 11,027 15,557 17,085 21,234 99,825

6.4 5.8 6.2 8.2 8.5 11.0 15.6 17.1 21.3 100

27,612 43,732 57,430 81,950 78,208 92,026 119,056 111,283 78,593 689,890

4.0 6.3 8.3 11.9 11.3 13.3 17.3 16.1 11.4 100

Table I. Sample composition by cohort

IJM 29,1

68

Table II. annual nominal and real wages of full-time workers

Table III. Weekly real wages of full-time workers

Year

Mean

Nominal wages Annual growth rates Median of mean wages

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

7,962 8,537 9,210 9,909 10,879 11,684 12,825 13,493 14,139 14,570 15,140 15,662 16,260 16,517 16,993

8,062 8,587 9,262 9,913 10,683 11,329 12,432 12,978 13,507 14,098 14,607 15,011 15,864 16,142 16,747

1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999

7.21 7.88 7.59 9.79 7.40 9.76 5.20 4.79 3.05 3.91 3.45 3.81 1.58 2.89

Mean

Real wages Annual growth rates Median of mean wages

15,304 15,494 15,850 16,192 16,699 16,732 17,045 17,136 17,222 17,018 16,609 19,445 16,812 16,781 16,993

15,494 15,585 15,940 16,197 16,398 16,223 16,522 16,482 16,452 16,466 16,024 15,762 16,403 16,400 16,747

Women

Men

Total

18-25

26-30

31-35

273 273 281 283 298 299 308 310 313 308 303 302 308 314 318

372 372 381 385 399 400 407 411 409 402 393 389 397 396 401

340 340 348 352 366 367 375 378 378 372 364 361 368 369 374

259 255 261 261 272 274 278 278 277 269 261 262 265 265 267

316 315 321 321 335 332 338 338 337 328 322 316 323 325 329

348 347 356 360 373 376 384 384 381 370 365 358 365 367 374

Age class 36-40 41-45 378 380 390 394 411 407 414 413 409 403 399 396 403 402 408

388 394 413 419 433 435 447 446 440 432 423 416 423 423 430

1.24 2.30 2.16 3.13 0.20 1.87 0.53 0.50 2 1.18 2 2.41 2 0.98 2.23 2 0.19 1.26

46-50

51-55

56-60

389 398 411 423 442 449 454 467 458 451 443 442 450 448 455

391 394 407 422 441 445 460 465 463 467 465 456 473 466 463

410 404 421 428 446 446 455 464 466 474 447 447 454 459 464

are higher than in the rest of the country, by as much as 90 euros per week, corresponding to a difference of 33.7 per cent compared to the Isles. Also workers employed in larger companies benefit from higher real wages (Table V). The average real wage[4] for full-time workers in the sample increases over time in line with the same pattern for the National Accounts (NA) average wage (Figure 1). Before turning to the analysis, we need to point to some limitations of our dataset. For example, it provides no information on education level, and does not follow private employees enrolled in other social security schemes. It does not provide information on reasons why workers leave the archive: this can be due to periods of unemployment, mobility towards and from self-employment and public employment, or retirement. As

a result, there may be some problems of attrition, as discussed by Cappellari (2000). However, controlling for the related sample selection bias, via the Heckman (1979) procedure, would be problematic because of the limited information available within WHIP. Following Cappellari’s (2000, p. 667) judgment that “the use of a dataset with an unbalanced design partially mitigates the problem”, the decision was taken to ignore this problem. Annual wages are also characterised by a top coding at 60,000 euro, but no specific treatment is given to this problem of censoring because the censored observations account for less than 0.6 percent of the sample size.

Apprentice Blue collar White collar White collar (senior) Manager Total

Firm dimension 0-9 10-19 20-199 200-999 $1,000 All

Industry

Services

Total

196 323 476 840 1,220 368

199 315 405 802 1,183 356

197 321 440 826 1,210

North-West

North-East

308 345 397 456 470 393

295 310 361 415 425 345

Geographical area Centre South 278 305 345 434 547 378

248 271 312 398 464 303

Isles

All

248 274 299 394 479 294

285 314 364 435 489

Worsening of wage expectations 69

Table IV. Weekly real wages by job qualifications and sector of activity of full-time workers

Table V. Weekly wages by firm dimension and area of work (full-time workers)

Figure 1. Sample, union and national accounts real wages

IJM 29,1

70

3. The cohort structure of wage to age profiles The analysis of the determinants of individual wages is based on panel data and includes cohort variables to capture generational differences in wages. The present paper starts by considering the cohort effects as treated in models such as Lucifora and Rappelli (1995) and Lucifora and Vignocchi (1997). It then takes into account suggestions presented in Beaudry and Green (2000) for Canada, and Biagi (2003) for Italy, on the interaction between age and cohort effects, to analyse the determinants of individual wages in the industry and service sectors in Italy. Their model specification includes an interaction variable defined as the product of age and cohort. Our analysis improves on these studies in several aspects. It applies to both male and female workers. It uses a large sample of individual longitudinal administrative data, rather than cohort average values of real wages, taken from cross-sectional survey data. It covers a period of 15 consecutive years from 1985 to 1999, instead of seven waves of cross-sections on alternate years from 1984 to 1995 (Biagi, 2003), or 13 in the period 1971 to 1993 (Beaudry and Green, 2000). In terms of model specification, the interaction between age and cohorts is defined on an individual basis, rather than on cohort averages. The use of individual data allows us to account for individual control variables, such as geographical area, job qualifications, sector of employment, firm size and part-time work. Furthermore, it allows us to test the interaction of the age-cohort structure with the propositions derived from the wage-curve literature, and the effects of union wages on individual earnings. The hypothesis to be tested in this section is that wage to age profiles – affected by the sluggish growth of the Italian economy and by the slowdown in average wage growth rates – may present a slope of decreasing value for younger generations of workers. This section first presents (sub-section 3.1) estimates of a base cohort model, and then proposes and tests (sub-section 3.2) a specification of the changing wage-age structure over time. 3.1. The base model Traditional econometric models of individual wages express the level of individual wages over time as a non-linear (most frequently, quadratic) function of age, and a linear function of control variables such as gender, sector, job location and job qualifications. They also take account of education levels. Given that the WHIP dataset does not include information on education, our model specification includes the variable “number of years of social security contributions”. With some approximation, this variable can be seen as a proxy for on-the-job training and work experience. Individual wages equations estimated on panel data show the relevance of cohort effects. Cohort dummy variables provide estimates of how and by how much the entry wage varies as it moves along the different cohorts of workers. Taking account of all these variables, the base cohort model can be shown as: ln wit ¼ a0 þ a1 ageit þ a2 age2it þ

9 X

bj dcoj þ d0 lcontri þ d1 ptimei

j¼2

ð1Þ

þ d2 secti þ d3 sizei þ d4 areai þ d5 qualit þ 1it where i indexes the individual and t the time. Wages, age and qualifications[5] are individual time-varying variables and are denoted by it. The equation is estimated separately for men and women.

The log of real weekly wages[6] is thus a function of: . A second-degree age polynomial. . Cohort dummies (dcoj), where cohorts refer to workers born in the same five or ten year periods. Older and younger individuals are classified in ten-year cohorts, from 1925 to 1935 and from 1971 to 1981 respectively. All other cohorts cover periods of five years (for a percentage composition of the sample by the nine cohorts, see Table I). . A set of control variables[7]: indicating whether the worker is a part-timer (ptime), the sector of employment (sect), the company size (size), the geographical location of the job (area), and the job qualifications[8] (qual). . The log for the total years of social security contributions[9] paid to the INPS during the 15-year period under analysis (lcontr). In estimating the model the reference individual is assumed to be a full-time white-collar worker belonging to cohort 1, working in a company of size 20-199 employees, in northern Italy and in the industry sector. The constant term in the regression is thus the value that defines the base wage of this representative worker, and the coefficients associated with cohort dummies measure the effect on wages linked to the different times of entry into the labour force. The OLS robust estimates[10] of the model for males (Table VI) show satisfactory goodness of fit statistics with the R-squared of 0.462 in line with the values obtained in similar analyses. However, for women the R-squared is only 0.267. The estimated coefficients confirm most of our a priori expectations[11]: wages are higher in the north, in bigger companies, for better qualifications, and in the industry sector (the dummy associated with services is significant and negative). The number of years of social security contributions has a positive sign in the equation, indicating higher wages for those workers with more on-the-job experience. In line with the existing literature, cohort coefficients are higher for younger workers[12]. Figure 2 depicts the estimated wage-age profiles for men (which are similar for women): each cohort has a different intercept derived from the sum of the common constant term and the specific cohort coefficient. This defines the entry wage for each cohort. 3.2. The model with a time-varying wage to age structure The first three terms in equation (1) define a concave wage to age profile that shifts along the different cohorts with a concavity that is assumed to be invariant over the different cohorts. A stable wage to age profile gradient over time somewhat contradicts the widespread opinion (or prejudice), that the wage growth prospects of today’s young workers are bleaker than their parents’. The reduction in the growth rate of the Italian economy since the mid 1980s has been accompanied by a decline in the share of wages in national income, and by a slow-down in average wage dynamics. Between 1985 and 1999, both the industry and service national average (NA) wage and the sample average wage show a declining rate of growth. In the first part of the sample, from 1985 to 1991, the annual average growth of real wages was around 1.8 per cent; in the second part it became negative, about 2 0.1 per cent. These trends prompt questions about the reality of the assumption of stable wage to age profiles that is implicit in the specification of equation (1) with its constant values

Worsening of wage expectations 71

IJM 29,1

72

Table VI. Base cohort model

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers const.

Menb

Coef.

t

Coef.

t

0.02123 21.66E-04 20.01898 20.00624 0.00936 0.03290 0.04903 0.06830 0.10378 0.16128 0.10597 0.08799 20.05027 20.15092 20.03528 20.04840 20.12383 20.44946 20.29665 0.96227 5.04322

25.82 214.22 22.37 20.79 * 1.12 * 3.64 5.13 6.92 10.34 15.87 60.41 34.94 224.51 253.11 219.92 227.16 56.61 2109.14 2174.24 29.08 320.58

0.03392 2 2.99E-04 2 0.01860 2 0.01570 0.00302 0.00989 0.03146 0.03508 0.04760 0.10469 0.12300 2 0.00074 2 0.01231 2 0.14373 2 0.00842 2 0.10399 0.12850 2 0.51178 2 0.30503 0.82009 5.01163

61.30 2 40.47 2 5.42 2 4.09 0.73 * 2.13 6.39 6.97 9.04 19.14 87.89 2 0.14 2 9.72 2 91.36 2 7.30 2 82.53 105.22 2 142.14 2 249.95 250.63 468.13

Notes: a Number of obs ¼ 246,419; F(21, 128,356Þ ¼ 4323:64; Prob . F ¼ 0:000; R-squared ¼ 0:2665; Number of obs ¼ 440,910; F(21, 283,514Þ ¼ 22,127.2; Prob . F ¼ 0:000; R-squared ¼ 0:4621; * Not significant at the 95 per cent level

b

Figure 2. Base cohort model: men

for coefficients a1 and a2. An alternative assumption is that younger cohorts of wage earners still enjoy higher real wages than older generations (in line with the predictions of the traditional model), but may face less optimistic prospects of income growth. In such a situation, where the careers of new entrants in the labour market develop at

slower annual growth rates than in the past, the assumption of an over-time invariant gradient of the wage to age relation should be rejected. To test this alternative hypothesis, the wage to age relation described by the first terms in equation (1) is modified to generate the possibility of a different gradient for each cohort (Beaudry and Green, 2000; Biagi, 2003). This is done by introducing a set of variables into the model, defined as “multiplicative age-cohort dummies”, obtained from the product of age by each cohort dummy. The expectation is that multiplicative dummies will generate flatter gradients for the wage to age relation, with the passage of time (from older to younger cohorts). In the revised specification of the model, the log of individual weekly real earnings is thus regressed on the second-degree age polynomial and the already defined set of control variables, with the addition of the set of multiplicative age-cohort dummies. The new model is: ln wit ¼ a0 þ a1 ageit þ a2 age2it þ

9 X j¼2

bj dcoj þ

9 X

gj agedcojt þ d0 lcontri

j¼2

þ d1 ptimei þ d2 secti þ d3 sizei þ d4 areai þ d5 qualit þ 1it

ð2Þ

where the variables agedcojt ¼ dcoj *ageit are the multiplicative age-cohort dummies. Taken together, the cohort dummies and the multiplicative age-cohort dummies provide measures of both the “additive” cohort effect and the different response of wages to age among the different cohorts, as can be seen more clearly in this re-specification of equation (2): ln wit ¼

9
9
X  X  a0 þ bj dcoj þ a1 þ gj dcoj ageit þ a2 age2it þ etc: j¼2

j¼2

where the gradient of the wage to age relation is given, for each cohort j, by a1 þ gj þ 2a2 age. The new hypothesis requires a smaller value of the parameter gj for individuals of the same age belonging to younger cohorts. Table VII reports the OLS robust estimates by sex for the parameters of the new wage profiles in equation (2). As in the base model, goodness of fit is satisfactory for men, with an R-squared of 0.463, but less so for women, with an R-squared of 0.267. The signs of the control variables are consistent with expectations[13]. The pattern of the coefficients associated with cohort dummies shows a regular increase for both male and female workers, confirming the traditional finding that younger cohorts receive higher entry wages. As for the multiplicative dummies, the results of the separate equations for males and females illustrate that their coefficients are all negative (except agedco2, which is non-significant) and progressively increasing in absolute value, as we move from older to younger generations, with the gradient between wage and age becoming flatter for younger cohorts. Both additive and multiplicative cohort coefficients are pair-wise statistically different from zero[14]. The new wage to age profiles depicted in Figure 3, clearly show that the slopes of each segment become progressively more gradual as we move from older to younger generations, for instance from cohort 6 (older workers born between 1956 and 1960) to cohort 9 (younger workers born between 1971 and 1981).

Worsening of wage expectations 73

IJM 29,1

74

Table VII. Multiplicative cohort model

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers const.

Menb

Coef.

t

Coef.

t

0.10387 29.31E-04 20.01126 0.46388 0.76038 1.06064 1.35713 1.67250 1.8881 2.12468 0.00005 20.00903 20.01509 20.02165 20.02914 20.03848 20.04536 20.05357 0.10490 0.08813 20.05015 20.15110 20.03407 20.05873 0.11446 20.45319 20.29719 0.95994 2.82506

17.89 219.64 20.08 * 3.23 5.13 6.83 8.29 9.68 10.57 11.63 0.02 * 23.42 25.46 27.32 29.06 210.83 211.86 213.17 59.65 34.99 224.46 253.21 219.21 231.64 51.11 2103.43 2174.46 29.03 14.27

0.10405 2 8.61E-04 0.60086 0.78932 1.07697 1.38402 1.58556 1.70115 1.84281 2.08454 2 0.01114 2 0.1473 2 0.02047 2 0.02778 2 0.3289 2 0.03645 2 0.04115 2 0.04938 0.12296 0.00041 2 0.01218 2 0.14375 2 0.00803 2 0.11129 0.12141 2 0.51147 2 0.30481 0.82048 2.85595

29.22 2 27.91 9.98 12.71 16.46 18.91 19.62 19.39 19.82 21.61 2 10.14 2 12.80 2 16.63 2 19.30 2 19.69 2 19.04 2 19.27 2 21.14 87.74 0.08 * 2 9.61 2 91.39 2 6.97 2 85.49 96.19 2 131.01 2 249.89 252.37 26.27

Notes: a Number of obs ¼ 246,419; F(29, 128,348Þ ¼ 4323:64; Prob . F ¼ 0:000; R-squared ¼ 0:2665; Number of obs ¼ 440; 910; F(29, 283,506Þ ¼ 15; 930:61; Prob . F ¼ 0:000; R-squared ¼ 0:4681; *Not significant at the 95 per cent level b

The cohort structure of the model is confirmed by the estimation of a random effects model (which takes account of unobserved individual heterogeneity), as demonstrated by the results in Table VIII. Coefficients of additive cohorts increase from older to younger generations, and those of multiplicative cohorts decrease accordingly. Number of years of social security contributions has a positive impact on earnings levels for both women and men, and area and qualifications dummies are in line with our expectations[15]. Finally, the estimated individual specific standard deviation is 0.287 for women and 0.211 for men, and the idiosyncratic standard deviation is 0.327 and 0.307 for women and men respectively. 4. Individual wages and aggregate labour market indicators This section provides some evidence on the relevance of aggregate labour market conditions on the dynamics of individual wages. The model described in Section 3.2 is

Worsening of wage expectations 75

Figure 3. Multiplicative-cohort dummy model: men

extended to test the effects on individual wages of changes in the age structure of the active population, and the effects of unemployment rate and union wages. The purpose of this section is thus to test the robustness of the wage to age profiles through more detailed model specifications. The aim is not to provide a full-fledged integration of the microanalysis of wage determination with labour market conditions; it is rather to suggest the usefulness of such integrations and possible further development of individual wages analysis. 4.1. Age structure of the active population In the model described in equation (2) none of the explanatory variables is related to labour market conditions. Individual wages and cohort effects are assumed to be independent of changes in the age structure of the population. In light of the changes that have occurred over the last 30 years in the age structure of the active population in Italy, this assumption is not appropriate. Significant studies on the topic, such as those by Welch (1979) and Berger (1985), suggest that supply-side effects are measurable by the relative size of the population by age groups (cohort size). A larger active population in a given age class relative to the total active population, may generate a downward pressure on the wages of workers in that class. In other words, when the size of a cohort increases, it may have a depressive effect on the wages of individuals in that cohort; when it decreases, it may have an expansionary effect on wages. The yearly cohort size, separately computed for men and women, is given by: 1 CS ¼ ln

9

N ðk22Þt þ 29 N ðk21Þt þ 39 N kt þ 29 N ðkþ1Þt þ 19 N ðkþ2Þt N ð18260Þt



k ¼ 18; :::; 60 t ¼ 1985; :::; 1999

ð3Þ

where the numerator is the moving average of individuals in adjacent age groups (with annual age classes indexed by k) and the denominator is the total population aged 18-60. The trend in average size of age cohorts from 1985 to 1999 (Figure 4) shows the progressive ageing of the male Italian labour force and the changing structure of the active population. The pattern is similar for women.

IJM 29,1

76

Table VIII. Multiplicative-cohort model (random effects)

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers const. sigma_u sigma_e rho

Menb

Coef.

z

Coef.

z

0.11718 2 9.54E-04 0.36987 0.89808 1.31962 1.64791 1.97818 2.34474 2.53183 2.63880 2 0.00704 2 0.01700 2 0.02549 2 0.03255 2 0.04064 2 0.05133 2 0.05716 2 0.05983 0.11028 0.19283 2 0.06123 2 0.15652 2 0.04779 2 0.02093 0.03636 2 0.35091 2 0.19292 0.50092 2.08177 0.32742 0.28728 0.56502c

24.19 224.35 2.96 7.35 10.50 12.48 14.22 16.06 16.75 17.04 23.08 27.62 210.99 213.05 214.96 217.26 217.86 217.43 48.51 78.13 215.40 233.92 218.44 210.99 14.34 273.62 267.48 19.51 12.46

0.12636 29.98E-04 0.68491 1.10678 1.51602 1.86846 2.08851 2.27365 2.37430 2.50878 20.01214 20.02028 20.02829 20.03676 20.04218 20.04754 20.05064 20.05384 0.12352 0.11996 20.02971 20.13649 20.02315 20.04413 0.04462 20.35443 20.16355 0.45833 1.96407 0.30676 0.21070 0.67945c

50.78 2 46.90 15.67 24.99 31.91 35.42 36.00 36.18 35.73 36.19 2 15.45 2 25.28 2 32.18 2 36.05 2 35.73 2 35.19 2 33.70 2 32.15 71.21 35.77 2 12.38 2 54.07 2 14.51 2 37.43 36.49 2 104.57 2 90.61 81.79 25.46

Notes: a Number of obs ¼ 246,419; Number of groups ¼ 38; 686; R-sq. within ¼ 0:0853; between ¼ 0:2542; overall ¼ 0:2384; b Number of obs ¼ 440,910; Number of groups ¼ 61,077; R-sq. within ¼ 0:1461; between ¼ 0:4212; overall ¼ 0:4208; c Fraction of Variance due to u_i

4.2. Macroeconomic factors Studies on the determinants of individual wages, such as Lucifora and Rappelli (1995), and Lucifora and Vignocchi (1997), suggest taking account of temporal effects by means of time dummies. The rationale for year dummies in the individual wage model is that they would remove the mis-specification resulting from lack of consideration of the effects of macroeconomic development on the short-term dynamics of individual wages. However, time dummies, age and cohorts taken together generate over-identified non-estimable models. Could this be dealt with, any system of time dummies would not anyway provide a direct identification of the macroeconomic events that affect the behaviour of individual wages. The economic explanation of what has happened is left to the researcher’s interpretation of the cyclical history of the

Worsening of wage expectations 77

Figure 4. Population cohort size: men

economy. Macroeconomic events affect the dynamics of individual wages and thus the model specification has to be controlled “for common across-cohorts business-cycle fluctuations” (Biagi, 2003, p. 8). The problem associated with contemporaneously identifying age, cohort and year effects has been brought to the attention of researchers by Heckman and Robb (1985). As suggested by Burbidge et al. (1997, p. 8) “one way to resolve the problem is to model the effect of one or more of the independent influences (cohort, year or age) as functions of other variables rather than simply including them as fixed effects or trends”. This suggests the inclusion of macro variables in the estimating equations, such as unemployment rate and union wages.

4.3. Unemployment rate Over the 15-year history of individual wages examined in the paper, the unemployment rate has shown marked cyclical behaviour, with wide regional differences (Figure 5). It is not unreasonable to assume that the worsening of the overall condition of the labour market might have affected individual wages and the wage to age relation. Blanchflower and Oswald (1994, 1995, 2005), in estimating the wage curves for a variety of countries, including Italy, found a consistent negative relationship between individual wages and regional unemployment rates. In applied macroeconomics studies wage dynamics and unemployment are obviously interdependent. The risk of simultaneous equation bias in individual wage equations has been questioned. Blanchflower and Oswald (1995, p. 158) argue that “attractive though such thinking is theoretically, little support for it can be found empirically”. The reason being that instrumental variable techniques have not been shown to significantly modify the value of estimated wage to unemployment elasticities. In their words: “Unemployment apparently has the characteristics of a predetermined variable” (Blanchflower and Oswald, 1995, p. 158). To be consistent with the literature and avoid introducing collinearity, we utilise regional unemployment rates[16] in our model, which includes age and cohort dummies among the regressors.

IJM 29,1

78

Figure 5. National and regional unemployment rates, 1985-1999

4.4. Union wages Individual wages are affected by changes in union wages that occur as a result of the quasi-automatic adjustments to inflation and of periodic national level wage negotiations. Such effects can be thought of as almost independent of individual behaviour and apply to all workers. The union wage level is determined by bargaining at national level and is independent of the age structure of the work force. Furthermore, it is closely linked – with lags of varying lengths - to inflation. A positive correlation (r ¼ þ0:38) exists between the sample average wage and the union wages index. The real union wage index (Figure 1) increases up to 1992 due to the prevailing indexation of wages to inflation, which created a mechanism for real wage resistance in case of monetary shocks and changes in indirect taxation (Fabiani et al., 1998). It falls in the period 1993 to 1996 as an effect of the 1993 labour cost agreement, and starts to move upwards again in subsequent years. Overall, it shows fluctuations around a stable mean and, therefore, our expectation is that it is more likely to interfere with the constant term of equation (2) than with the cohort dummies. 4.5. Estimation On the basis of the arguments developed so far, the model is revised to include labour market conditions. The new specification thus adds cohort size, regional unemployment rates and real union wages to the explanatory variables in equation (2): ln wit ¼ a0 þ a1 ageit þ a2 age2it þ

9 X

bj dcoj þ

j¼2

9 X

gj agedcojt þ d0 lcontri

j¼2

þ d1 ptimei þ d2 secti þ d3 sizei þ d4 areai þ d5 qualit þ q1 CSit þ q2 ln unempt þ ln wcontrt þ 1it

ð4Þ

where the variables in the first two lines of the equation are those defined in Section 3, CSit is cohort size as defined in equation (3), ln unempt is the log of unemployment rates for Italian macro regions and ln wcontrt is the log of the real union wage index. The model is estimated separately for the two sexes using OLS and taking account of the different levels of aggregation of the dependent variable (individual wages) and the newly introduced annual macro variables, to correct for heteroskedasticity and avoid a standard errors bias[17] (as addressed by Moulton, 1986). Table IX presents the estimated coefficients of the model. The estimates of additive and multiplicative cohort dummies are statistically significant (with the exception of cohorts 2 and 3 for women). As with the model in equation (2), additive dummies increase from older to younger generations;

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers cs ln_unemp ln_wcontr const.

Worsening of wage expectations 79

Menb

Coef.

t

Coef.

t

0.07806 27.17E-04 20.13810 0.18233 0.42419 0.60843 0.84273 1.010982 1.25828 1.42136 0.00232 20.00383 20.00843 20.01227 20.01784 20.02507 20.02910 20.03430 0.10597 0.08824 20.04165 20.012437 20.03447 20.05696 0.15578 20.45585 20.29702 0.96319 20.16306 20.02087 0.92601 20.49939

11.72 213.10 21.17 * 1.46 * 2.84 3.96 4.68 5.64 6.24 7.20 1.06 * 21.67 * 22.99 24.09 24.72 25.77 26.33 27.23 20.04 8.17 23.49 23.87 25.04 29.60 18.15 242.10 261.11 31.39 23.79 20.99 * 5.55 20.63 *

0.07275 2 6.06E-04 0.45732 0.45580 0.66344 0.83461 0.96056 1.01832 1.08035 1.23823 2 0.00857 2 0.00855 2 0.01223 2 0.01639 2 0.01918 2 0.02031 2 0.02173 2 0.02648 0.12344 0.00179 0.00860 2 0.07679 2 0.00841 2 0.11044 0.12116 2 0.51479 2 0.30482 0.81977 2 0.16548 2 0.05345 0.72396 0.66205

10.74 2 9.78 6.14 5.56 6.02 6.40 7.03 6.93 7.00 7.11 2 6.49 2 5.81 2 5.67 2 6.15 2 6.63 2 6.25 2 6.04 2 5.75 15.04 0.14 * 1.49 * 2 3.30 2 1.44 * 2 36.70 38.10 2 47.51 2 45.45 31.93 2 5.47 2 3.64 7.79 1.27 *

Notes: a Number of obs ¼ 246,419; R-squared ¼ 0:2687; Number of clusters ðyearÞ ¼ 15; Robust standard errors; b Number of obs ¼ 440,910; R-squared ¼ 0:2687; Number of clusters ðyearÞ ¼ 15; Robust standard errors; * Not significant at the 95 per cent confidence level

Table IX. Model with regional unemployment rate and union wage

IJM 29,1

80

multiplicative dummies are negative and progressively increasing in absolute value. This confirms the robustness of the changing-over-time cohort structure of individual wages[18]. The coefficients associated with cohort size are always statistically significant and, as expected, with a negative sign. The associated elasticities are around 2 0.16 for both females and males. The elasticity coefficients of the union wage index are 0.93 and 0.72, for women and men respectively, and are statistically different from zero. The quasi-unitary elasticity of individual to union wages has the effect of keeping the constant term of the equation close to zero and making it statistically non-significant. The high constant term (which has no clear economic meaning), which characterises the simpler models presented in this and some of the studies discussed in Section 1, is now substituted by a significant relation for a macroeconomic variable relevant to the labour market[19]. The unemployment rate is not statistically significant in the female equation or the two older cohorts in either their additive or multiplicative forms. For men, the estimated elasticity coefficient of regional unemployment rate is significant and equal to 2 0.053. This figure is between the values for the UK estimated by Card (1995) and Bell et al. (2002), of 2 0.082 and 2 0.025 respectively. It is also well below Blanchflower and Oswald’s (1995, 2005) result for the USA and other countries, including Italy, of 2 0.10. It should be noted that our model includes the union wage index, which might interact with the unemployment rate. To be consistent with previous empirical work on wage-curves, we estimated a model that includes only cohort size and unemployment, as follows: ln wit ¼ a0 þ a1 ageit þ a2 age2it þ

9 X j¼2

bj dcoj

9 X

gj agedcojt þ d0 lcontri

j¼2

þ d1 ptimei þ d2 secti þ d3 sizei þ d4 areai þ d5 qualit þ q1 CSit þ q2 ln unempt þ 1it :

ð5Þ

In this case, the estimated elasticity coefficients of regional unemployment rates for men and women are 2 0.072 and 2 0.046 respectively (Table X), and are statistically significant. The estimated value from the male equation compares well with Blanchflower and Oswald’s (1995, 2005) and Card’s (1995) estimates referred to above. The elasticity estimated in the female equation is rather low, but is comparable with Bell et al.’s (2002) estimate of 2 0.025 for the UK. For Italy, Devicienti et al. (2006), using the WHIP dataset, estimated an elasticity of 2 0.041 for females and 2 0.029 for males for the period 1993-1999. Equations (4) and (5) were estimated with a random effects specification, and to account for the different levels of aggregation among the variables[18] (Moulton, 1986). Estimates (Table XI) for the male equation of model (4) generally confirm the previous results, but with a slight deterioration in the decreasing pattern of the multiplicative-cohort coefficients for the two younger cohorts (8 and 9) and loss of significance of the two area dummies. In the female specification, the results show an improvement, with the unemployment elasticity coefficient statistically significant and

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers cs ln_unemp const.

Menb

Coef.

t

Coef.

t

0.08696 27.95E-04 20.08106 0.27851 0.5316 0.76699 1.02491 1.31211 1.47922 1.64902 0.00125 20.00572 20.01111 20.01574 20.02205 20.03006 20.03492 20.04061 0.10610 0.08828 20.03171 20.09262 20.03407 20.05825 0.11444 20.45543 20.29708 0.96245 20.15126 20.04586 3.53431

6.46 26.65 20.68 * 1.97 2.90 3.35 3.67 4.27 4.54 4.76 0.57 * 22.22 22.99 23.33 23.55 24.17 24.40 24.46 19.33 8.24 22.29 22.75 24.92 28.66 16.32 241.43 260.89 31.47 23.24 22.13 8.42

0.08010 26.69e-04 0.50522 0.53747 0.77149 0.966144 1.11269 1.8861 1.26474 1.42493 20.00945 20.01014 20.01445 20.01923 20.02265 20.02444 20.02649 20.03150 0.12349 0.00161 0.01598 20.05305 20.00830 20.11104 0.12067 20.51403 20.30491 0.81956 20.15967 20.07231 3.80443

6.29 2 5.81 6.72 5.35 4.67 4.68 4.78 4.62 4.49 4.53 2 7.05 2 5.60 2 4.34 2 4.42 2 4.40 2 4.08 2 3.82 2 3.72 15.03 0.12 * 2.05 2 1.86 * 2 1.42 * 2 33.08 34.69 2 47.58 2 45.53 31.85 2 4.92 2 3.95 9.98

Notes: a Number of obs ¼ 246,419; R-squared ¼ 0:2679; Number of clusters ðyearÞ ¼ 15; Robust standard errors; b Number of obs ¼ 440,910; R-squared ¼ 0:4636; Number of clusters ðyearÞ ¼ 15; Robust standard errors; * Not significant at the 95 per cent confidence level

elasticity estimated at 2 0.038. The elasticities between union and individual wages are confirmed (Table XI): 0.88 for women and 0.70 for men. The estimation of equation (5) produces unemployment rate elasticities (Table XII) that compare with those estimated via OLS: 2 0.063 for women and 2 0.096 for men. The estimations of equations (4) and (5) thus confirm the robustness of the cohort structure discussed in the previous section. In both the male and female equations, the wage to age profiles of individual wages remain substantially the same as estimated in equation (2) and depicted in Figure 3, with the cohort and multiplicative dummies having the correct sign and dimension. Thus, the profile estimates inclusive of the change over time in wage to age gradients, are robust to the introduction of aggregate labour market indicators.

Worsening of wage expectations 81

Table X. Model with regional unemployment rate

IJM 29,1

82

Table XI. Model with regional unemployment rate and union wage (random effects)

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers cs ln_unemp ln_wcontr const. sigma_u sigma_e rho

Menb

Coef.

z

Coef.

z

0.08199 26.75E-04 0.13435 0.48663 0.79496 0.99004 1.22661 1.52938 1.64292 1.65189 20.00287 20.00964 20.01526 20.01927 20.02448 20.03218 20.03478 20.03398 0.11270 0.19288 20.04508 20.10681 20.04839 20.01874 0.03823 20.35380 20.19248 0.50543 20.27790 20.03811 0.87856 20.57864 0.32698 0.28683 0.56513c

10.21 211.52 0.97 * 2.84 4.11 5.04 5.60 6.39 6.67 6.40 21.13 * 23.09 24.26 25.18 25.69 26.58 26.68 25.82 12.60 9.51 22.57 22.84 26.02 22.35 3.39 233.59 215.14 13.37 26.83 22.14 6.71 21.00 *

0.08141 2 6.53E-04 0.39123 0.57766 0.83527 1.01436 1.11950 1.23073 1.24068 1.25964 2 0.00696 2 0.01082 2 0.01497 2 0.01955 2 0.02140 2 0.02325 2 0.02246 2 0.02168 0.12562 0.12251 0.00124 2 0.03886 2 0.02397 2 0.04296 0.04380 2 0.35700 2 0.16257 0.45443 2 0.32391 2 0.07693 0.70103 0.50121 0.30652 0.20983 0.68091c

12.67 2 12.27 4.88 5.53 7.14 7.64 7.69 8.25 7.69 7.37 2 5.13 2 5.83 2 7.20 2 7.95 2 7.66 2 7.51 2 6.43 2 5.22 10.36 5.67 0.12 * 2 1.41 * 2 3.33 2 3.91 4.16 2 26.87 2 10.69 53.02 2 7.27 2 6.14 8.31 1.06

Notes: a Number of obs ¼ 246,419; Number of groups ¼ 38,686; R-squ. within ¼ 0:0881; between ¼ 0:2556; overall ¼ 0:2395; b Number of obs ¼ 440,910; Number of groups ¼ 61,077; R-squ. within ¼ 0:153; between ¼ 0:4208; overall ¼ 0:42; c Fraction of variance due to u_i; * Not significant at the 95 per cent confidence level

5. Summary and conclusions This paper builds on the results of existing studies on the individual wage to age relation. It confirms that the younger generations of workers in the Italian labour market have higher entry wages and that they can expect increasing earnings in the future. However, it shows that wage-age profiles cannot be considered invariant over time, and are characterised by progressively smaller gradients as one moves from older to younger generations. As a consequence, the expectations for younger generations of workers are of flattening wage-age profiles: higher entry wages are compensated for by

Womena age age2 dco2 dco3 dco4 dco5 dco6 dco7 dco8 dco9 agedco2 agedco3 agedco4 agedco5 agedco6 agedco7 agedco8 agedco9 lcontr ptime d_centre d_south d_serv d_small d_large d_apprentices d_bluecollars d_managers cs ln_unemp const. sigma_u sigma_e rho

Menb

Coef.

z

Coef.

z

0.09161 2 7.5E-04 0.20546 0.59910 0.94654 1.17194 1.43380 1.75832 1.88969 1.91179 2 0.00420 2 0.01181 2 0.01836 2 0.02318 2 0.02919 2 0.03773 2 0.04114 2 0.04110 0.11256 0.19299 2 0.03520 2 0.07553 2 0.04766 2 0.02084 0.03594 2 0.35330 2 0.19269 0.50451 2 0.26454 2 0.06269 3.20738 0.32732 0.28707 0.56523c

6.85 26.38 1.58 * 3.63 4.62 5.17 5.24 5.65 5.73 5.52 21.76 * 24.01 24.82 25.21 24.99 25.42 25.48 24.84 12.18 9.57 21.73 * 21.81 * 25.86 22.45 2.94 233.71 215.14 13.20 26.20 23.09 7.68

0.08935 27.18E-04 0.4582 0.68040 0.96773 1.17084 1.29747 1.42737 1.45089 1.47714 20.00816 20.01277 20.01764 20.02284 20.02536 20.02792 20.02774 20.02742 0.12541 0.12229 0.00884 20.01476 20.02360 20.04404 0.04280 20.35609 20.16280 0.45365 20.31283 20.09595 3.50722 0.30680 0.21004 0.68089c

7.09 2 6.60 5.27 5.31 5.55 5.50 5.23 5.24 4.86 4.57 2 5.56 2 5.77 2 5.43 2 5.45 2 4.97 2 4.66 2 4.04 2 3.5 10.06 5.68 0.69 * 2 0.43 * 2 3.28 2 4.21 3.90 2 27.44 2 10.74 51.24 2 7.55 2 5.53 8.68

Notes: a Number of obs ¼ 246,419; Number of groups ¼ 38,686; R-squ. within ¼ 0:0866; between ¼ 0:2548; overall ¼ 0:2388; b Number of obs ¼ 440,910; Number of groups ¼ 61,077; R-squ. within ¼ 0:1514; between ¼ 0:4202; overall ¼ 0:42; c Fraction of variance due to u_i; * Not significant at the 95 per cent confidence level

reduced expectations of future earnings growth. These results hold in a variety of model specifications, including in the presence of macroeconomic variables. The results provide evidence of the relevance of aggregate labour market indicators. In fact, individual wages are shown to be: . Negatively related to cohort size: in the age categories of wage earners where the ratio of the active to total population is higher, individual wages tend to be lower. . Negatively affected by the unemployment rate prevailing in the Italian macro regions, with elasticities of magnitudes of between 2 0.046 and 2 0.063 for

Worsening of wage expectations 83

Table XII. Model with regional unemployment rate (random effects)

IJM 29,1 .

84

women, and between 2 0.053 and 2 0.096 for men, depending on the model specification, but in line with those found in similar studies on wage-curves. Positively influenced by the dynamics of union wages. The elasticity of individual earnings to union wages is close to 1 (0.7 and 0.8), with the interesting result that the union wage index of equation (5) absorbs the variance explained by the constant term in the model of equation (2). It thus gives some economic meaning to the (so far unexplained) high constant term that characterises the simpler model of individual wages.

The results presented in this paper suggest that it would be possible and useful to integrate the changing age profiles of individual wages with the estimation and projections of Italian aggregate industry and service sector average wages. Two sets of countervailing factors would be at work here. On the one hand, productivity gains at the macro level would generate progressively increasing entry wages. On the other hand, the declining wage-age profiles would determine a lower growth in individual wages throughout the workers’ life cycle than under time invariant wage expectations. This paper leaves open for future research investigation of the reasons why individual wage expectations in the Italian labour force are undergoing such significant changes.

Notes 1. “Union wages” are wages determined at national level through collective bargaining. 2. WHIP is built and distributed by LABORatorio Riccardo Revelli, Moncalieri – Italy. The section of WHIP used in the paper is a linked employer-employee database, that combines information on private sector employees with information on the companies that employ them. WHIP is based on a systematic random sampling of approximately one in every 180 individuals of the INPS individual files (the sampling is based on dates of birth). The original sample contains 117,785 individuals (41,347 females and 76,438 males, with a percentage composition of 35 and 65 per cent respectively) working in all branches of the private sector, for a total of 937,152 observations over the period 1985-1999. The fact that individuals who are shown as having had more than one employment spell in the same year, reflects that they took up new employment within that year or that they had two part-time jobs; or perhaps is due to the precision and speed at which INPS registers job movers in its archives. 3. Wages are net of social security contributions payable by the employer, but gross of income tax and social security contributions payable by the employee. 4. As a deflator of nominal wages we used the earnings conversion coefficients provided by INPS and utilised to compute social security benefits (Coefficiente di rivalutazione quota A). 5. Qualifications are considered time-dependent as career progression can be automatic, regardless of the worker’s ability. 6. The choice of weekly wages is justified by the risk of an under-reporting bias for daily wages. 7. Variable sect distinguishes between the industry and the service sectors; variable size distinguishes among companies with up to 19 employees, 19-200 employees, and more than 200 employees; variable area captures regional effects (North, Centre, and South and Isles); variable qual distinguishes among qualifications: apprentices, blue collar workers, white collar workers and managers.

8. The category “white collar” also includes senior officers (“quadri” in Italian), which in the dataset are identifiable only from 1997. 9. Years of contribution refers to the period of enrolment with the INPS only; the dataset does not provide information on how long or whether an employee was a member of another social security scheme. 10. Estimates are obtained using the Huber-White estimator (Huber, 1967; White, 1980). 11. The only exception being the part-time dummy, which has a positive sign in the women’s equation, where we would expect a negative contribution. In the men’s equation it has a negative sign, although it is not statistically significant. 12. A Wald test on the pair-wise equality of cohort coefficients was performed for both equations. All pairs of adjacent coefficients were significantly different from zero: for women the F-statistic was 150.7 (p-value ¼ 0:000), for men it was 156.9 (p-value ¼ 0:000). 13. Again, with the exception of the control variable part-time, which is positive for both sexes and non-significant for men. 14. A Wald test of the pair-wise equality of multiplicative cohort coefficients was performed on both the women’s and men’s equations. All pairs of adjacent coefficients were significantly different from zero: for women the F-statistic was 46.3 (p-value ¼ 0:000), for men it was 61.9 (p-value ¼ 0:000). For additive cohort dummies the F-tests were 60.9 (p-value ¼ 0:000) and 77.3 (p-value ¼ 0:000) respectively. 15. Again, the sign of the control variable part-time is an exception. 16. The disaggregation of the unemployment rate is at the macro-regional level: North, Centre, and South of Italy. Consistent series of national and regional unemployment rates over the period 1985-1999 were provided by ISTAT (Gatto et al., 2001). 17. Estimates were obtained using the Huber-White (Huber, 1967; White, 1980) estimator of variance as implemented by Rogers (1993). 18. The parameters of the cohort structure estimated in equations (2) and (4) compare favourably with the estimates provided in Biagi (2003), where a large number of the estimated cohort coefficients was not statistically significant, possibly as a consequence of the small number of degrees of freedom. In fact, by using average wages by cohort, the number of observations is necessarily very small. 19. Estimation of model (4) by OLS and random effects, with the inclusion of only cohort size and union wages (among the macro variables), produced satisfactory results for both sexes in terms of the cohort structure, with only cohorts 2 and 3 not statistically significant in the female equation. Individual to union wages elasticity is 1.00 and 0.92 for females and males respectively. As expected the constant term of both equations is not significant.

References Baltagi, B.H. and Blien, U. (1998), “The German wage curve: evidence from the IAB employment sample”, Economic Letters, Vol. 61 No. 2, pp. 135-42. Beaudry, P. and Green, D. (2000), “Cohort patterns in Canadian earnings: assessing the role of skill premia in inequality trends”, Canadian Journal of Economics, Vol. 33 No. 4, pp. 907-36. Bell, B., Nickell, S. and Quintini, G. (2002), “Wage equations, wage curves and all that”, Labour Economics, Vol. 9 No. 3, pp. 341-60. Berger, M.C. (1985), “The effect of cohort size on earnings growth: a re-examination of the evidence”, The Journal of Political Economy, Vol. 93 No. 3, pp. 561-73.

Worsening of wage expectations 85

IJM 29,1

86

Biagi, F. (2003), “Characterizing cohort patterns for Italian wages”, paper presented at the workshop Mercato del lavoro, istituzioni e distribuzione dei redditi, Universita` degli Studi del Piemonte Orientale, Novara, 29-30 May. Blanchflower, D.G. and Oswald, A.J. (1994), The Wage Curve, MIT Press, Cambridge, MA. Blanchflower, D.G. and Oswald, A.J. (1995), “An introduction to the wage curve”, The Journal of Economic Perspectives, Vol. 9 No. 3, pp. 153-67. Blanchflower, D.G. and Oswald, A.J. (2005), “The wage curve reloaded”, Working Paper No. 11338, NBER, Cambridge, MA. Brugiavini, A. and Peracchi, F. (2003), “Social security wealth and retirement decisions in Italy”, Labour, Vol. 17, special issue. Brunello, G. and Comi, S. (2000), “Education and earnings growth: evidence from 11 European countries”, IZA Discussion Paper No. 140, IZA, Bonn. Brunello, G. and Lauer, C. (2004), “Are wages in Southern Europe more flexible? The effects of cohort size on European earnings”, IZA Discussion Paper, No. 1299, IZA, Bonn. Brunello, G., Comi, S. and Lucifora, C. (2000), “The college wage gap in 10 European countries: evidence from two cohorts”, IZA Discussion Paper No. 228, IZA, Bonn. Burbidge, J.B., Magee, L. and Robb, A.L. (1997), “Cohort, year and age effects in Canadian wage data”, Working Paper No. 13, Canadian International Labour Network, Hamilton. Cappellari, L. (2000), “The covariance structure of Italian male wages”, The Manchester School, Vol. 68 No. 6, pp. 659-84. Card, D. (1995), “The wage curve: a review”, Journal of Economic Literature, Vol. 33 No. 2, pp. 785-99. Card, D. and Lemieux, T. (2001), “Can falling supply explain the rising return to college for younger men? A cohort-based analysis”, The Quarterly Journal of Economics, Vol. 116 No. 2, May, pp. 705-46. Creedy, J. and Hart, P.E. (1979), “Age and the distribution of earnings”, The Economic Journal, Vol. 89 No. 54, pp. 280-93. Devicienti, F., Maida, A. and Pacelli, L. (2006), “The resurrection of the Italian wage curve”, Working Paper No. 52, Laboratorio R. Revelli, Moncalieri. Fabiani, S., Lovarno, A., Oneto, G. and Sestito, P. (1998), “Risultati e problemi di un quinquennio di politica dei redditi: una prima valutazione quantitativa”, Working Paper No. 329, Bank of Italy, Rome. Fertig, M. and Schmidt, C.M. (2003), “Gerontocracy in motion? European cross-country evidence on the labour market consequences of population ageing”, IZA Discussion Paper No. 956, IZA, Bonn. Gatto, R., Gennari, P. and Massarelli, N. (2001), La ricostruzione e il riallineamento delle serie storiche delle forze di lavoro. Gennaio 1984-Luglio 1992, ISTAT, Rome. Heckman, J.J. (1979), “Sample selection bias as a specification error”, Econometrica, Vol. 47 No. 1, pp. 153-61. Heckman, R. and Robb, J. (1985), “Using longitudinal data to estimate age, period and cohort effects in earnings equations”, in Mason, W.M. and Feinberg, S.E. (Eds), Cohort Analysis in Social Research beyond the Identification Problem, Springer-Verlag, New York, NY. Huber, P.J. (1967), “The behaviour of maximum likelihood estimates under nonstandard conditions”, Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability, Vol. 1, University of California Press, Berkeley, CA.

Korenman, S. and Neumark, D. (2000), “Cohort crowding and youth labour markets: a cross-sectional analysis”, in Blanchflower, D. and Freeman, R. (Eds), Youth Employment and Joblessness in Advanced Countries, NBER, Chicago University Press, Chicago, IL. Lucifora, C. and Origo, F. (1999), “Alla ricerca della flessibilita´: un’analisi della curva dei salari in Italia”, Rivista Italiana degli Economisti, No. 1. Lucifora, C. and Rappelli, F. (1995), “Evoluzione delle retribuzioni nel ciclo di vita: un’analisi su dati longitudinali”, Lavoro e relazioni industriali, No. 3. Lucifora, C. and Vignocchi, C. (1997), Le carriere retributive: un’analisi con microdati, in CNEL, Rapporto sulle retribuzioni e sul costo del lavoro, Giuffre` Editore, Milan. Manacorda, M. (2004), “Can the Scala Mobile explain the fall and rise of earnings inequality in Italy? A semiparametric analysis, 1977-1993”, Journal of Labour Economics, Vol. 22 No. 3, pp. 585-613. Mincer, J. (1958), “Investments in human capital and personal income distribution”, Journal of Political Economy, Vol. 56, pp. 281-302. Moulton, B.R. (1986), “Random group effects and the precision of regression estimates”, Journal of Econometrics, Vol. 32 No. 3, pp. 385-97. Naticchioni, P., Ricci, A. and Rustichelli, E. (2006), “Return to education and changes in the wage structure: is there an unskilled bias technological change in Italy?”, paper presented at the XXI National Conference of Labour Economics, University of Udine, Udine, 14-15 September. Psacharopoulos, G. and Layard, R. (1979), “Human capital and earnings: British evidence and a critique”, Review of Economic Studies, Vol. 46 No. 3, pp. 485-503. Rogers, W.H. (1993), “Regression standard errors in clustered samples”, Stata Technical Bulletin, No. 13, pp. 16-19, (reprinted in Stata Technical Bulletin Reprints, No. 3, pp. 88-94). Stapleton, D.C. and Young, D.J. (1988), “Educational attainment and cohort size”, Journal of Labour Economics, Vol. 6 No. 3, pp. 330-61. Welch, F. (1979), “Effects on cohort size on earnings: the baby boom babies’ financial bust”, Journal of Political Economy, Vol. 87 No. 2, pp. S65-S97. White, H. (1980), “A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity”, Econometrica, Vol. 48 No. 4, pp. 817-38. About the author Elena Giarda is a researcher at PROMETEIA – Associazione per le Previsioni Econometriche, Bologna, Italy, and a PhD student in Statistics at the University of Bologna, Italy. Her current research interests include labour economics, income distribution, and microeconometrics. Elena Giarda can be contacted at: [email protected]

To purchase reprints of this article please e-mail: [email protected] Or visit our web site for further details: www.emeraldinsight.com/reprints

Worsening of wage expectations 87