TH-IE WO)RLD- AX|- -ECOOMI li CA'vk Volume 13 January 1999 Number 1 ;~ ~ ~~~~~~I 3 194 A SYMPOSIUM ISSUE ON EFFICIENT PUBLIC SECTOR DOwNSIZING l Public Sector Downsizing: An Introduction g1 IlMartin Rama -, Cross-Country Evidence on Public Sector Retrenchment John Haltiwanger and Manisha Singh The Efficient Mechanism for Downsizing the Public Sector l Doh-Shin Jeon and Jean-Jacques Laffont Earnings and Welfare after Downsizing: Central Bank Employees in Ecuador Martin Rama and Donna MacIsaac Matching Severance Payments with Worker Losses in the Egyptian Public Sector L a Ragui Assaad The Algerian Retrenchment System: | I A Financial and Economic Evaluation Elizabeth Ruppert Labor Earnings in One-Company Towns: i I Theory and Evidence from Kazakhstan Martin Rama and Kinnon Scott ISSN 0258-6770 THE WORLD BANK ECONOMIC REVIEW EDITOR Moshe Syrquin CONSULTING EDITOR Sandra Gain EDITORIAL BOARD Kaushik Basu, Cornell University and Universitv of Delhi Stijn Claessens Francois Bourguignon, DELTA, Paris David Dollar Carmen Reinhart, University of Maryland Gregory K. Ingram Mark R. Rosenzweig, Universitv of Pennsylvania Martin Ravallion L. Alan Winters, University of Sussex John Williamson The World Bank Economic Review is a professional journal for the dissemination of World Bank- sponsored research that informs policy analyses and choices. It is directed to an international readership among economists and social scientists in government, business, and international agencies, as well as in universities and development research institutions. The Review emphasizes policy relevance and opera- tional aspects of economics, rather than primarily theoretical and methodological issues. It is intended for readers familiar with economic theory and analysis but not necessarily proficient in advanced mathematical or econometric techniques. Articles will illustrate how professional research can shed light on policv choices. Inconsistency with Bank policy will not be grounds for rejection of an article. Articles will be drawn primarily from work conducted by World Bank staff and consultants. Before being accepted for publication by the Editorial Board, all articles are reviewed by two referees who are not mem- bers of the Bank's staff and one World Bank staff member; articles must also be recommended by at least one external member of the Editorial Board. The Review may on occasion publish articles on specified topics by non-Bank contributors. Any reader interested in preparing such an article is invited to submit a proposal of not more than two pages in length to the Editor. The views and interpretations expressed in this journal are those of the authors and do nor necessar- ilv represent the views and policies of the World Bank or of its Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility whatsoever for any consequences of their use. When maps are used, the bound- aries, denominations, and other information do nor imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. Comments or brief notes responding to Review articles are welcome and will be considered for publica- tion to the extent that space permits. Please direct all editorial correspondence to the Editor, The World Bank Economic Review, The World Bank, Washington, D.C. 20433, U.S.A. The World Bank Economic Review is published three times a year (januarv, May, and September) by the World Bank. Single copies may be purchased at $12.95. Subscription rates are as follows: Individuals Institutions 1-year subscription US$30 US$50 2-year subscription US$55 US$95 3-year subscription US$70 US$130 Orders should be sent to: World Bank Publications, Box 7247-7956, Philadelphia, PA 19170-7956, U.S.A. Subscriptions are available without charge to readers with mailing addresses in developing coun- tries and in socialist economies in transition. Written request is required every two years to renew such subscriptions. (© 1999 The International Bank for Reconstruction and Development / THE WORLD BANK 1818 H Street, N.W., Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America ISBN 0-8213-4361-0; ISSN 0258-6770 Material in this journal is copyrighted. Requests to reproduce portions of it should be sent to the Office of the Publisher at the address in the copyright notice above. The World Bank encourages dis- semination of its work and will normally give permission promptly and, when the reproduction is for noncommercial purposes, without asking a fee. Permission to make photocopies is granted through the Copyright Clearance Center, Suite 910, 222 Rosewood Drive, Danvers, MA 01923, U.S.A. This journal is indexed regularly in Current Contents/Social & Behavioral Sciences, Index to Interna- tional Statistics, Journal of Economic Literature, Public Affairs Information Service, and Social Sci- ences Citation Index". It is available in microform through University Microfilms, Inc., 300 North Zeeb Road, Ann Arbor, MI 48106, U.S.A. THE WORLD BANK ECONOMIC REVIEW Volume 13 January 1999 Number 1 A SYMPOSIUM ISSUE ON EFFICIENT PUBLIC SECTOR DOWNSIZING THIS SYMPOSIUM ISSUE draws on papers originally prepared for the World Bank conference on Public Sector Retrenchment and Efficient Compensation Schemes, held in November 1996. The articles in the symposium were refer- eed in the usual way. The Editorial Board invited Martin Rama, who coordi- nated the conference, to write an introduction to the symposium. Public Sector Downsizing: An Introduction 1 Martin Rama Cross-Country Evidence on Public Sector Retrenchment 23 John Haltiwanger and Manisha Singh The Efficient Mechanism for Downsizing the Public Sector 67 Doh-Shin Jeon and Jean-Jacques Laffont Earnings and Welfare after Downsizing: 89 Central Bank Employees in Ecuador Martin Rama and Donna MacIsaac Matching Severance Payments with Worker Losses 117 in the Egyptian Public Sector Ragui Assaad The Algerian Retrenchment System: A Financial 155 and Economic Evaluation Elizabeth Ruppert Labor Earnings in One-Company Towns: 185 Theory and Evidence from Kazakhstan Martin Rama and Kinnon Scott THE WORLD BANK ECONOMIC REVIEW, VOL. 13, NO. 1: 1-22 Public Sector Downsizing: An Introduction Martin Rama Authorities throughout the developing world are turning to downsizing in an effort to reduce budget deficits and address the inefficiencies engendered by state-led develop- ment strategies. Because large-scale involuntary dismissals are often politically diffi- cult, a voluntary approach to reductions in public sector employment is increasingly popular among developing-country governments, multilateral organizations, and do- nor countries. This article (and, more generally, the research project on Public Sector Retrenchment) attempts to sketch a protocol for public sector downsizing that takes into account the costs and benefits for the workers and the economy. After reviewing the international experience with downsizing, the article addresses five questions: how to identify the redundant workers, how to predict their losses from separation, how to design compensation and assistance packages, how to assess the financial and economic returns to downsizing, and how to deal with downsizing in one-company towns. Based on the answers to these questions, a decision tree is proposed. Public sector downsizing is not a final goal of economic policy, but economic reforms may require mass layoffs. State-led development strategies left a legacy of bloated bureaucracies and overstaffed public enterprises. Severe labor redun- dancies characterize transition economies, where the shift from plan to market requires millions of workers to be relocated out of the public sector. In Latin America and South Asia, decades of protective policies led to the proliferation of white elephants and sick industries. All over the world, technological progress is making natural monopolies disappear, thus confronting formerly somnolent utilities with harsh competition. Increasingly, authorities are correcting the em- ployment excesses from past patronage and cronyism as more modern and demo- cratic ways replace traditional and authoritarian ones. The extent of labor redundancies could make politically unfeasible any seri- ous downsizing, especially based on involuntary dismissals. Hence, a voluntary approach to reductions in public sector employment is increasingly popular among developing-country governments, multilateral organizations, and donor coun- tries. The voluntary approach offers severance pay to encourage the redundant workers to quit, thus overcoming their resistance to downsizing, restructuring, and privatization. In many developing countries, buying out the redundant work- Martin Rama is with the Development Research Group at the World Bank. This article summarizes the findings of a research project on Public Sector Retrenchment and Efficient Compensation Schemes, supported by the Research Committee of the World Bank through grant RPO 679-51. ©D 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 1 2 THE WORLD BANK ECONOMIC REVIEW. VOL. 13, NO. I ers is in fact the only way to bypass the legal obstacles to the dismissal of public sector employees. In addition, it seems fair to compensate those who may suffer from a change in the rules of the game. In this respect, severance pay resembles the lump-sum transfers that characterize textbook analyses of economic policy. If public sector downsizing increases efficiency, it should be possible to com- pensate redundant workers in the public sector and still make a net gain for the economy. Public sector downsizing could thus qualify as a Pareto-optimal reform. Until recently, one of the main obstacles to the implementation of the volun- tary approach to public sector downsizing was its cost. Sometimes downsizing requires the relocation of hundreds of thousands of workers to the private sec- tor, with the average compensation, retraining, and redeployment package amounting to several thousand dollars per worker. A single downsizing opera- tion may therefore cost hundreds of millions of dollars. However, some of the countries that most desperately need public sector downsizing are strapped for cash. Recent changes in the attitude of multilateral organizations toward mass retrenchment have significantly softened this constraint. In March 1996 the World Bank modified its operational rules to allow lending for severance pay aimed at restructuring the public sector. The International Monetary Fund also favors public sector downsizing because it could allow a more durable reduction in government expenditures than cuts in the wages of civil servants, which are not sustainable in the long run. And regional organizations such as the Inter-American Development Bank will lend for severance pay. As a result, many developing coun- tries and transition economies have plans for public sector downsizing in prepara- tion or already in execution. While the gains from downsizing are potentially large, the chances of mishan- dling it are considerable too. It is quite obvious that many workers in the public sectors of developing countries contribute little to aggregate output or welfare, if anything at all. The issue is whether the use of severance pay packages really helps to relocate these workers to more productive activities. The frequently observed "revolving door" syndrome, whereby separated workers are subse- quently rehired, suggests that some downsizing operations lead to the departure of the few who make the public sector work. The amount and nature of the assistance provided to separated workers may be inadequate. The authorities usually set up severance pay packages in an arbi- trary way. Typically, they use a rule of thumb involving salary and perhaps seniority in the public sector (see Nunberg 1994, Kikeri 1997, and Kouame 1997). For instance, separated workers receive two years of salary, one month of salary per year of service, or some other combination of these two variables. But the resulting amount may bear little relation to the loss these workers expe- rience as a result of their separation. Some of them clearly suffer, whereas others become net winners. Moreover, the authorities may spend large amounts of re- sources on ineffective retraining and redeployment programs. Thus, in practice, public sector downsizing may diverge a lot from the envisioned Pareto optimality. Rama 3 Assessing the returns to downsizing operations is also difficult. The typical assessment compares the savings in terms of public sector wages with the cost in terms of severance pay packages, retraining, and redeployment programs. Over- staffing is only one among several distortions characterizing the public sector, however, so the analysis should not use financial returns to measure economic returns. Public sector wages usually differ from private sector wages and there- fore are not a good indicator of the opportunity cost of labor. Furthermore, the analysis should not ignore the externalities from mass retrenchment. The most obvious externalities arise in the context of one-company towns, which may easily become ghost towns after downsizing takes place. Public sector downsizing leads to fiscal externalities, too, because it reduces the equilibrium level of gov- ernment expenditures and hence the burden from distortionary taxes. This symposium issue provides a conceptual framework for analyzing the effects of downsizing operations. In addition to this introduction, the issue con- tains six articles produced for a recently completed World Bank research project on Public Sector Retrenchment and Efficient Compensation Schemes.' This in- troduction presents the common thread underlying those articles, as well as other articles prepared for the project but not included in this issue. It also draws lessons from other studies on public sector downsizing, including the studies by Svejnar and Terrell (1991), Diwan (1994), Fiszbein (1994), Kikeri (1997), and Lindauer and Nunberg (1994). Most of the previous research dealt with rela- tively narrow downsizing issues. The Public Sector Retrenchment project repre- sents the first systematic attempt to address downsizing from a variety of per- spectives, ranging from public economics to labor economics to mechanism design. I. A MIXED RECORD The World Bank indirectly supported more than 40 attempts to downsize the public sector in developing countries between early 1991 and late 1993. The units targeted included government administration, state-owned enterprises, and, in the context of post-conflict demobilization, the military. Haltiwanger and Singh (this issue) review these downsizing operations. The average downsizing operation led to the separation of 125,000 workers at a cost of $400 million, including $87 million in severance pay. The downsizing operations varied con- siderably. For instance, the smallest one affected 247 public sector workers, compared with more than 1.6 million workers for the largest one. Downsizing reduces public sector expenditures, particularly the public sector wage bill. When the present value of this reduction is higher than the up-front cost in terms of severance pay and enhanced safety nets, downsizing has positive financial returns. A more precise assessment has to be based on economic re- turns. Downsizing also reallocates workers across sectors. When aggregate wel- 1. For more information see http://www.worldbank.org/research/projects/downsize. A complete version of most of the research papers produced for the project can be downloaded from this Web site. 4 THE WORLD BANK ECONOMIC REVIEW, VOL. 13, NO. 1 fare increases as a result of this reallocation, downsizing has positive economic returns. Haltiwanger and Singh evaluate the financial returns of 15 of the surveyed operations based on the number of years it would take to recover their direct financial costs, using a 10 percent annual discount rate. This indicator is called the break-even period. Measured by this indicator, the downsizing operations performed remarkably well. The average break-even period was two years and four months, and it exceeded four years in only one case. Few investment projects display such high financial returns. As an indirect measure of economic returns, Haltiwanger and Singh calculate the percentage of the displaced workers who were subsequently rehired by the restructured units. Rehires indicate a poorly handled downsizing process. In the best case, they imply that workers who were essential to the operation of the restructured units were mistakenly considered redundant. In the worst case, they suggest that workers who had no intention of leaving the public sector were able to cash in golden handshakes. Haltiwanger and Singh found rehiring in 40 percent of the operations for which the required information was available. More than 10 percent of the sepa- rated workers were rehired in half of these cases. If anything, these results un- derestimate the extent of labor misallocation. The fact that 60 percent of the operations display no rehiring does not imply that essential workers did not leave. Moreover, rehiring provides no information on another type of error, which consists of retaining redundant public sector workers. Haltiwanger and Singh find that for every dollar spent on severance pay, on average more than two dollars were spent on safety net enhancements. Almost two-thirds of the downsizing operations surveyed included some enhancement of the safety net, such as early retirement programs, counseling and placement services, or wage subsidies. Retraining was a feature in more than half of the operations. Are these programs worth their cost? Probably not, based on previ- ous assessments of the effectiveness of vocational education programs in devel- oping countries (see Middleton, Ziderman, and Van Adams 1993). Usually, the same government agencies that get low grades in the evaluations of vocational education programs end up in charge of the retraining component of downsizing operations. In the context of public sector downsizing, a more relevant assess- ment is provided by a micro-econometric study of the effects of active labor market programs on employment and earnings in the Czech Republic, Hungary, Poland, and Turkey (Fretwell, Benus, and O'Leary 1998). This study finds that some of the active labor market programs made it more likely for subsets of workers to find jobs after separation. But others did not, and the programs do not appear to have a significant impact on the labor earnings of those who did find jobs. Two of the studies prepared for the Public Sector Retrenchment project sug- gest that a considerable amount of resources has been wasted on active labor market programs. An evaluation of the public sector downsizing operation imple- Rama S mented by Spain in the 1980s showed its limited ability to relocate workers to alternative industries, in spite of its extremely large retraining program (Campa 1996). This failure was partly due to retraining being focused on the update of previous skills, rather than on the acquisition of new ones. A case study of downsizing in the Central Bank of Ecuador found that only 12 percent of the displaced workers took the retraining courses that were offered, in spite of these courses being free of charge (Rama and MacIsaac this issue). II. ADVERSE SELECTION Some of the difference in productivity among public sector workers is asso- ciated with observable worker characteristics, such as educational attainment or occupation. But part of it is unobservable. Although productivity differ- ences also exist in the private sector, they seem to be exacerbated in the public sector. Jeon and Laffont (this issue) analyze how policymakers should take into ac- count these unobservable differences when deciding the extent and composition of public sector downsizing. They apply the tools of mechanism design to exam- ine the optimal manner for reducing public sector employment. They show that the optimal composition of the layoffs depends on the nature of the public sec- tor unit and on the prospects workers face after separation. For instance, the public sector unit could produce a socially valuable service, such as basic health or economic management, and the labor market could be tight, so that the pros- pects for workers with low productivity would be relatively good. In this case, the optimal downsizing policy would be to retain all of the workers with high productivity and to retrench some of those with low productivity. Conversely, the public sector unit could produce a service with little social value, such as steel or direct credit allocation, and the unemployment rate could be high, leav- ing little hope of finding a job for workers with low productivity. In this case, the optimal downsizing policy would retrench all of the workers with high pro- ductivity and retain those with low productivity. Standard voluntary separation programs usually lead to the departure of the workers with high productivity, because those workers have the best prospects outside the public sector. It follows that standard voluntary separation programs are not appropriate in all circumstances. For instance, these programs would lead to the wrong composition of layoffs when applied to public sector units that produce valuable services and operate in a tight labor market. The fact that these programs were used quite systematically in the past could be one of the reasons why so many separated workers were rehired in the aftermath of downsizing operations. Before the use of voluntary separation programs became common, govern- ments used other methods to cut expenditures. Several analysts have criticized those efforts, saying that they had an adverse impact on the effectiveness of governments (see Van Ginneken 1991 and Colclough 1997). By compressing 6 THE WORLD BANK ECONOMIC REVIEW, VOL. 13, NO. I the pay scale, the argument goes, budget cuts encouraged skilled workers to leave, thus jeopardizing the ability of governments to deliver on basic services. There is little doubt that better outcomes could have been achieved by getting rid of genuinely redundant public sector workers and offering better wages and working conditions to the others. However, standard voluntary separation pro- grams can lead to the departure of skilled public sector workers, too, much the same as budget cuts did in the past, and in substantially larger numbers. Several alternatives have been proposed to standard voluntary separation pro- grams for cases where adverse selection is a serious concern. In a study prepared for the Public Sector Retrenchment project, Levy and McLean (1997) analyze the merits and demerits of some of those alternatives. Randomization is one of them. Each public sector worker would face a probability of losing the job equal to the estimated percentage of redundant workers. This alternative would make the com- position of separated workers resemble that of those who remain in the public sector (Diwan 1994). For units producing socially valuable services in a tight labor market, randomization thus represents an unambiguous improvement com- pared with standard voluntary separation programs. However, Levy and McLean show that closing down those units, or leaving them untouched, could be prefer- able to a randomized downsizing and that, in general, randomization is not an efficient mechanism for downsizing the public sector. In the presence of unobservable information, the efficient mechanism for downsizing has to lead the workers to reveal their productivity. Jeon and Laffont show how to implement this mechanism by means of a menu of wage and sever- ance pay pairs. Each pair is associated with a different probability of separation. If the probability of separation is equal to one, the pair can be interpreted as a standard severance pay offer. If it is equal to zero, it can be viewed as a typical, open-ended public sector contract. All of the workers choosing the first pair are retrenched, whereas all those choosing the second one are retained. For pairs in between these two extremes, some of the workers are retrenched whereas others keep their jobs. If the menu is appropriately designed, workers should choose the pairs associated with their socially optimal probability of separation. For instance, if the overstaffed public sector unit produces a valuable service and operates in a tight labor market, workers with low productivity should choose a pair associated with a strictly positive probability of separation. However, setting up the right menu might be difficult in practice, so that other, simpler devices for identifying workers with low productivity in the pub- lic sector should be used as well. In their cross-country survey of downsizing operations, Haltiwanger and Singh (this issue) find that the targeting of separa- tions significantly reduces the probability of subsequent rehiring. The targeting mechanism can include such simple devices as chasing ghost workers (workers on the payroll, but not working). The experience of the Central Bank of Ecua- dor, analyzed by Rama and MacIsaac (this issue), is also interesting in this re- spect. After a first, disastrous attempt to downsize using voluntary separation programs, the Central Bank decided to classify all of its personnel in three cat- Rama 7 egories: those who were essential for its functioning, those who were clearly redundant, and those for whom it was difficult to tell. The classification was based on the nature of the worker's unit and on the worker's occupation and educational attainment. Essential workers did not have the option to leave, clearly redundant workers did not have the option to stay, and the rest of the workers were proposed a voluntary separation program. In light of the article by Jeon and Laffont, the menu approach has potentially higher payoffs for the third group of workers. III. LOSSES FROM SEPARATION The welfare loss a separated public sector worker experiences can be disag- gregated into three components. The first one is the present value of the result- ing change in earnings, including bonuses and other cash benefits. Except for highly skilled workers, salaries in the public sector tend to be higher than labor earnings out of it. Moreover, it may take a long time for some of the separated workers to find a new job, and earnings can be close to zero during that period. For simplicity, it is assumed here that displaced workers do not withdraw from the labor force after separation. The logic would be similar in case of with- drawal, except that reservation wages (rather than labor earnings out of the public sector) should be considered when assessing the present value of the earn- ings loss. The second component is the present value of the loss in nonwage benefits. Public sector jobs usually provide health coverage and old-age pension, among other benefits. In most developing countries, the jobs available to separated public sector workers do not carry such benefits. The third component is other, more intangible losses from separation. For instance, effort levels tend to be lower in the public sector than out of it, whereas job security is almost invariably higher. The possibility of taking bribes or using government facilities for private pur- poses also falls into this category. In preparation for a downsizing operation, policymakers should assess the welfare loss that separated workers might experience. This assessment may help predict the cost of the downsizing operation in terms of severance pay, if work- ers are fully compensated, or the harshness of their resistance to downsizing, if they are not. The Public Sector Retrenchment project used three empirical strat- egies to assess the workers' welfare loss and to link it to a variety of observable characteristics of the workers. These characteristics include salary and seniority in the public sector, which are the two variables most commonly used when designing severance pay packages. But they also include gender, age, education level, and province of residence, among others. The first and more direct strategy is to interview separated workers, and to ask them to evaluate subjectively the change in their well-being. Rama and MacIsaac (this issue) apply this strategy to the study of downsizing in the Cen- tral Bank of Ecuador. The subjective evaluation of well-being depends on the Table 1. Determinants of Losses from Separation Welfare loss Earnings loss Argentina, Ecuador, Egypt, Turkey, Ecuador, Ghana, Slovenia, Turkey, Worker white-collar central Bank public sector cement and central Bank civil formal labor cement and characteristic employees employees workers oil workers employees servants force oil workers Public sector wage +a 0 n.a. 0 0 0 n.a. + Seniority in the job n.a. + ? 0 + 0 + b 0 Educational level 0 - + - - 0 0 - Total work experience ? + - + 0 0 + b - Female n.a. 0 + 0 + + - 0 Married - 0 n.a. 0 + 0 n.a. 0 Number of dependents ? + n.a. + 0 0 n.a. 0 Source Robbins Rama and Assaad Tansel Rama and Alderman, Orazem, Tansel (1996) MacIsaac (this issue) (1997) MacIsaac Canagarajah, Vodopivec, (1997) (this issue) (this issue) and Younger and Wu (1996) (1995) Note: Statistically significant signs are indicated by + or-, while 0 indicates a nonsignificant coefficient and ? indicates a change in sign across specifications or groups of workers. When the variable was not included in the analysis, n.a. is reported. a. The coefficient is positive as a result of an implicit restriction imposed by the chosen specification. b. Almost all work experience was under the self-management system that characterized Yugoslavia until the late 1980s. Rama 9 amount of compensation received at the time of separation. Rama and MacIsaac use information on the subjective change in well-being and the amount of com- pensation received to infer the welfare loss from separation. A second empirical strategy relies on the welfare loss predicted by public sec- tor workers before separation. Robbins (1996) applies this strategy to voluntary separation programs in several banks and government agencies in Argentina. Some of the workers in these banks and agencies accepted the severance pay package they were offered, whereas others rejected it. Robbins uses this infor- mation to infer the amount of severance pay that would have made workers with specific individual characteristics indifferent between accepting or rejecting the offer. That amount is an indicator of the welfare loss expected by each worker. The third empirical strategy assumes that there is a stable relationship be- tween the welfare loss and the earnings loss. Assaad (this issue) applies this strategy to the case of workers in Egyptian public sector enterprises. Using data from a national household survey, Assaad compares the present value of earn- ings for workers in and out of the public sector. This comparison shows that some public sector workers earn less, over their work life, than similar workers out of the public sector. If these workers do not voluntarily quit, it is probably because they derive other benefits from their jobs. The gap in earnings ob- served for the most disadvantaged group of public sector workers can thus be used to infer the value of nonwage and other intangible benefits from public sector jobs. Assaad assumes that the loss of nonwage and intangible benefits is the same proportion of the salary for other public sector workers. Under the hypothesis that a stable relationship exists between welfare losses and earnings losses from displacement, studies dealing with the latter can be expected to provide information on the former. Several articles have analyzed the impact of observable characteristics of separated public sector workers on their earnings losses. Some of the studies use data on earnings out of the public sector from surveys of separated workers. For example, Rama and MacIsaac (this issue) refer to Central Bank employees in Ecuador. In another study pre- pared for the Public Sector Retrenchment project, Tansel (1997) assesses the welfare losses of blue-collar workers in cement and petrochemical factories in Turkey. Alderman, Canagarajah, and Younger (1996) and Younger (1996) look at the case of civil servants in Ghana. Orazem, Vodopivec, and Wu (1995) ana- lyze how earnings changed in Slovenia with the transition to a market economy.2 Table 1 summarizes the findings of the available studies on the determinants of welfare and earnings losses from displacement in developing countries and transition economies. The table reports the signs of the impact of workers' char- acteristics on their welfare and earnings losses. 2. Other empirical studies on public sector downsizing in developing countries, such as the one by Mills and Sahn (1996) for Guinea, or the one by London Economics (1996) for Zambia, do not estimate the impact of individual characteristics on displacement losses. For a comparison with losses from job displacement in industrial countries, see Hamermesh (1989); Topel (1990); Jacobson, LaLonde, and Sullivan (1993); and Fallick (1995). 10 THE WORLD BANK ECONOMIC REVIEW, VOL. 13, NO. I Several regularities emerge from table 1. First, it appears that the wage level in the public sector is a poor predictor of welfare losses, at least as long as other observable characteristics of the workers are taken into account. Second, with the exception of Egypt, where government hiring and compensation policies strongly distort the payoffs to schooling, the loss from displacement is usually smaller the higher the educational level of the workers. Third, while higher se- niority in a public sector job may lead to larger losses from displacement, there is no clear link between total work experience and losses from displacement. And fourth, female workers and those with bigger families may be bound to suffer more from displacement. IV. COMPENSATION The rationale for compensating displaced workers stems from the welfare loss they may experience as a result of separation. However, compensation may contradict the broader objectives of economic policy reform in developing coun- tries and transition economies. Many efforts by multilateral organizations and donor countries focus on reorienting public expenditures toward the most needy. Efforts to tilt the budgetary process in favor of the poor may conflict with the willingness to lend generous amounts of money to finance severance pay pack- ages for workers who are not poor, even after separation (see London Econom- ics 1996). This perspective would justify full compensation only when legal or political constraints make it absolutely necessary. This conclusion may not be valid, however, when public sector workers share a significant portion of their earnings with their extended families, as is the case in some Sub-Saharan African countries. Private transfers imply that for each public sector job suppressed, several households are bound to experience a welfare loss. And some of those households were probably poor even before downsizing. Private transfers also suggest that the displaced public sector workers are likely to share the compen- sation they receive with their less fortunate relatives. In the extended family setting, compensating the displaced workers may therefore reduce the adverse impact of downsizing on poverty. While the decision to fully compensate displaced workers should be made on a case-by-case basis, there are clearly no circumstances that would justify over- compensation. This is, however, a bias of downsizing operations based on vol- untary separations. Workers who are offered less than full compensation will prefer to stay in the public sector, whereas those who are offered more than full compensation will accept the offer and leave. Therefore mistakes in the direc- tion of excessively low compensation will not reduce severance pay costs, whereas mistakes in the direction of excessively high compensation will materialize. A poor tailoring of compensation packages may exacerbate this second type of mistake. The Public Sector Retrenchment project leads to a five-step procedure to tai- lor assistance to the displaced workers when full compensation is needed. This Rama I 1 procedure relies on the third empirical strategy used to estimate losses from displacement because this is the only one that can be applied before any re- trenchment has taken place. Fiszbein (1994) used a simpler version of this pro- cedure for Sri Lanka. Assaad (this issue) refined it in his analysis of downsizing in the Egyptian public sector. The five-step proposal in this section also draws from lessons of experience in several countries. The first step estimates an earnings function for workers who are out of the public sector. The data should come from individual records in labor force or living standards surveys. The right-hand-side variables in the earnings function include individual characteristics that are also observable for workers in the pub- lic sector. Ideally, the variables should be exactly the same as those in the records the public sector has about its own employees. The left-hand-side variable mea- sures the labor earnings of all individuals working out of the public sector, in- cluding the self-employed and those in the informal sector. More sophisticated analyses of earnings out of the public sector could also try to infer the reservation wage of unemployed or economically inactive individuals. Whatever the sophis- tication of the analysis, this first step aims to predict the value of earnings for the public sector workers who are bound to be separated. The second step calculates the present value of the earnings loss that public sector workers experience when they lose their jobs. This calculation compares, for each worker to be separated, his or her public sector salary with the earnings (or reservation wages) estimated in step one. The difference between the two is discounted over the duration of the contract the worker has with the public sector. In most cases this duration is the number of years to retirement. The third step assesses the loss in benefits. In many developing countries, the most important component of this loss concerns old-age pension. An actuarial calculation of the present value of the forgone old-age benefits can be used to quantify this loss. As a simpler alternative, the calculation could rely on the present value of past contributions to social security, plus accrued interest when applicable. Whatever the chosen approach, previous experience with downsizing suggests that the loss in benefits needs to be dealt with separately. Explicitly canceling outstanding social security obligations is important to avoid misun- derstandings (or opportunistic behavior) that can eventually lead to legal and political wrestling. The fourth step evaluates workers' loss of other, more intangible benefits. This analysis focuses on groups of public sector workers for whom the sum of the earn- ings loss estimated in step two and the benefits loss estimated in step three is sub- stantially negative. If these workers stay in the public sector, it is because they derive some other benefits from their jobs. The monetary value of these other ben- efits is at least equal to the sum of the earnings and benefits losses. The ratio be- tween this monetary value and the public sector salary can be used to infer the intangible benefits enjoyed by other, less disadvantaged public sector workers. Finally, the fifth step develops a simple formula to calculate compensation based on a few observable characteristics of public sector workers. Under the 12 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 assumption that the losses related to old-age pension are settled separately, the problem is to identify a relatively small set of information that can be used to predict the loss of earnings and intangible benefits in a convenient and noncon- troversial way. For instance, making compensation depend on individual char- acteristics such as gender would not be legally or socially admissible in some countries. Other characteristics, such as marital status or the number of depen- dents, may be subject to manipulation in countries where common law mar- riages and extended families are widespread. The compensation formula developed in step five differs from the typical rules of thumb used to design severance pay packages in two important ways. First, the information set may or may not include salary and seniority in the public sector, depending on how useful these two variables are to predict the loss of earnings and intangible benefits. Second, the coefficients multiplying these two variables, as well as the other variables in the relevant information set, are not arbitrary. They are the coefficients of a regression explaining the predicted loss of earnings and intangible benefits as a function of observable characteristics of public sector workers. For state-owned enterprises in Egypt, Assaad (this issue) finds that a well- tailored package could reduce the total compensation cost by 31 percent, com- pared with the best-performing rule of thumb. In the case of the Central Bank of Ecuador, Rama and MacIsaac (this issue) find that well-tailored compensa- tion could have reduced the cost of voluntary separations by 19 percent. Given that the average downsizing operation surveyed by Haltiwanger and Singh (this issue) spent $87 million in compensation payments, the potential savings from the proposed procedure could be substantial. Moreover, the procedure would be more fair, because it would provide more assistance to those who lose more. V. RETURNS TO DOWNSIZING Downsizing operations involve spending a considerable amount of resources in the short run in order to reap some gain in the longer run. Consequently, the decision to undertake a downsizing operation should consider its payoff, much the same as an investment decision. The most common approach focuses on financial returns, that is, on the impact of downsizing on the budget deficit. This approach can be justified when downsizing is part of a broader adjustment pro- gram. However, the decision should also consider the economic returns to downsizing-that is, its impact on aggregate output or welfare. The first and most obvious financial gain from downsizing results from the cut in the wage bill. In government administration, this cut directly reduces bud- get expenditures. The budgetary impact may be smaller in state-owned enter- prises if their wage bill is only partially subsidized by the budget. A second fi- nancial gain results from the reduction in long-term liabilities as separated workers lose all or some of their entitlement to old-age pensions. A third potential gain is Rama 13 the increase in privatization prices when downsizing is done in preparation for privatization and contributes to its success. The upfront cost is the amount of resources spent in compensation, retraining, and other redepolyment programs for separated workers. An assessment of financial returns should focus on the consolidated budget, not just on the budget of the overstaffed unit. Examples abound where rede- ployment programs simply shift the fiscal burden to another government body. Consider, for instance, the social services provided by state-owned enterprises in many transition economies. Often taxpayers pay for the cost of these services under the form of explicit or implicit subsidies. When downsizing reduces the number of beneficiaries of these services, it also reduces the burden to taxpayers. However, there is no such reduction if downsizing leads to a mere transfer of these services to central or local governments. Another example is provided by redeployment programs that allow the redundant workers to take another pub- lic sector job elsewhere. This fiscal illusion may be particularly severe when downsizing affects entitle- ments to old-age pension and other social security benefits. For example, Carneiro and Gill (1997) show that in Brazil the savings from downsizing are substan- tially smaller for the consolidated government than for the individual states. The pension benefits granted to the displaced workers increase the long-term liabili- ties of the federal government. As a result of the implicit transfer of obligations, budget savings are 15 to 25 percent lower than it appears at first glance. Economic returns focus on aggregate output or welfare, rather than on budget revenue and expenditures. Downsizing affects aggregate output or welfare in two different ways: it reduces the equilibrium level of taxes, and it relocates public sector workers to activities where they are supposedly more productive. The value to society of a lower taxation level should not be confused with the financial returns to downsizing. If transfers were costless, financial returns would be totally irrelevant when assessing economic returns. In practice, however, when the government raises one additional dollar of revenue, there is a net loss to society due to the inefficiencies created by distortionary taxation. This loss, known as the marginal tax burden, can be quite large in developing countries. In India, for example, it was estimated at around 0.8, which means that 80 cents of out- put are lost per dollar of revenue raised (Ahmad and Stern 1987). The overall assessment of a downsizing operation depends on whether it uses financial or economic returns. As an example, consider a downsized unit that cannot be privatized and whose wage bill is entirely paid for by the budget. Assume that public sector jobs entail no intangible benefits and that compensa- tion is tailored so as exactly to offset the loss in salaries and benefits that the separated workers are bound to experience. Because the potential earnings of at least some of the separated workers are positive, the government spends less in compensation than it saves in salaries and benefits. Financial returns are thus positive. But this conclusion is not warranted on economic grounds. If many 14 THF WORLD BANK ECONOMIC REVIEW, VOL 13, NO I separated workers end up unemployed, downsizing may reduce aggregate out- put. More specifically, economic returns to downsizing are more likely to be positive when productivity in the public sector is low, when potential earnings out of it are high, and when the marginal tax burden is large. Adverse selection can dramatically affect the economic returns to downsizing, without modifying its financial returns much. If the retrenched workers were genuinely redundant, their productivity in the public sector would probably be very low. In the limit, if productivity in the public sector were equal to zero, the economic returns would just be a multiple of the financial returns, so that finan- cial returns and economic returns would be perfectly correlated. However, if the retrenched workers were essential for the operation of a unit producing a so- cially valuable service, productivity in the public sector could actually be quite large. Good civil servants can contribute to society more than they cost. If they leave, public sector downsizing may have negative economic returns, in spite of positive (and possibly high) financial returns. There have been several attempts to evaluate the economic returns to public and private sector retrenchment, including those by Jenkins and Montmarquette (1979), Svejnar and Terrell (1991), and Brander and Spencer (1994). Ruppert (this issue) estimates the financial and economic returns to the downsizing of state-owned enterprises in Algeria. She shows that the program is financially sound under a wide range of assumptions; however, its impact on aggregate output depends very much on the efficiency or inefficiency of these state-owned enterprises. Ruppert considers two extreme cases. In one of them, the public sector pro- ductivity of the retrenched workers is equal to zero, and economic returns are high. At the other extreme, state-owned enterprises are on their labor demand curve, so that the marginal productivity of labor is equal to the public sector salary. But the public sector salary is probably much higher than labor earnings in the private sector in Algeria. The unemployment rate there is 28 percent, which suggests that searching for a good job (mainly in the public sector) is much more attractive than taking a bad job (mainly in the informal sector). As a result of this earnings gap, there is a tradeoff between the reduction in dis- tortionary taxation from downsizing and the fall in aggregate output. Ruppert shows that under plausible assumptions, economic returns can be either positive or negative. The possible discrepancy between financial returns and economic returns il- lustrates the second-best principle. The initial situation of the public sector unit to be downsized is one where several distortions and imperfections prevail, in- cluding overstaffing. Most likely the public sector unit is also characterized by a distorted pay scale, compared with the private sector. And the public sector unit is at least partially financed out of taxes that create distortions and reduce ag- gregate output. Downsizing operations usually only tackle one of these distor- tions and imperfections, namely overstaffing. Thus downsizing may not result in improved economic efficiency. Rama 15 VI. ONE-COMPANY TOWNS Public sector downsizing may affect the rest of the economy not only through its fiscal impact, but also through its direct impact on private sector output. The one- company-town setting provides an illustration of this productive externality. The main feature of this setting is the large share of jobs in a particular region (the town, for short) provided by the public sector unit to be downsized (the company). As a result, many of the other jobs in the town also depend on employment and wage levels in the company. For instance, the company's employees are probably the most important customers of the town's private shops. A drastic reduction in employment in the company is therefore likely to depress private sector activity in the town in a very Keynesian way. It follows that the level of earnings and produc- tivity out of the public sector cannot be taken as given. Productive externalities from downsizing may also occur at the national level. Mass retrenchment programs, affecting a substantial fraction of the urban labor force, can increase unemployment rates over long periods of time. For instance, in some Sub-Saharan African countries, where the public sector represents a large share of the modern economy, downsizing may depress economic activity in the short run. Externalities like those arising in the one-company-town setting provide a justi- fication for retaining some of the redundant workers. Limiting the extent of downsizing certainly entails a cost to the rest of society, which has to pay for these redundancies in the form of higher taxes or lower social expenditures. But retrenching the redundant workers entails a cost to the population that depends on it. The optimal extent of downsizing involves a tradeoff between these two costs. Rama and Scott (this issue) evaluate the potential externalities from downsizing in the one-company towns of Kazakhstan, a country that has dozens of them. Admittedly, Rama and Scott's estimates represent an upper bound for other countries, where labor mobility between towns is probably higher. The one- company towns of Kazakhstan are quite isolated, both for geographic and for institutional reasons. On the geographic side, Kazakhstan has a relatively small population but is the ninth largest country on earth. On the institutional side, the housing market is poorly developed, and the unemployed need to hold a local passport to draw their benefits. Consequently, the externalities from downsizing could be stronger in Kazakhstan than elsewhere. Rama and Scott's results suggest that the externality from downsizing can be substantial: retrenching the equivalent of 1 percent of the town's population would reduce the average labor earnings of the town by roughly 1.5 percent. Completely shutting down the average company, in turn, would reduce those earnings by more than 11 percent. These results imply that compensation should not be restricted to the retrenched workers and that transfers of resources to the community as a whole are justified. Whatever the compensation strategy adopted, the calculation of the economic returns to downsizing needs to take this produc- tive externality into account. 16 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Figure 1. A Downsizing Decision Tree Q1: Is pnvatization No advisable? Yes Q3: Is / Q2: Is adverse selection A < overstaffing a serious __ an obstacle to concern ? Yes privatization? Yes| N o |lNo Use targeting Assess Privatize and "menus" percentage without to identify of redundant prior redundancies workers downsizing Q4: Is Predict predicted loss of p loss higher redundant N than legal workers No compensation? Set compensation Yes based on predicted Yes loss Set up training QS: Is full and other I compensation redeployment of workers services needed? Settle old-age N pension liabilities Let workers choose their Pay legal mix of cash \ compensation and services Assess economic returns to downsizing Rama 17 VII. A PRACTICAL GUIDE TO DOWNSIZING Many things can go wrong in a downsizing operation, as the discussion in the previous sections shows. But these warnings should not be seen as an encourage- ment for inaction. One of the studies prepared for the Public Sector Retrenchment project, by Basu, Fields, and Debgupta (1996), deals with the consequences of not addressing the overstaffing problem. In India firms employing more than 100 work- ers may seek government permission for any retrenchments they wish to make. But these applications seldom succeed, and in the end the firms are often declared sick and are required to continue functioning on the basis of government subsidies. Basu, Fields, and Debgupta show that this seemingly protective legislation not only reduces economic efficiency but also may harm the workers it aims to protect. This is because it reduces labor demand, thus lowering the equilibrium wage level. This section integrates the main lessons about downsizing into a simple blue- print for action. Figure 1 presents the blueprint in the form of a decision tree, intended to assist policymakers in developing countries and task managers in multilateral organizations and donor countries. For simplicity, figure 1 assumes that there are no productive externalities from downsizing, like those foresee- able in one-company towns. Question 1 in this decision tree refers to the appropriate private sector counterfactual to public sector downsizing (see Devarajan, Squire, and Suthiwart- Narueput 1995). In some cases, the choice is not whether to downsize, but whether to have the government or the private sector manage the downsizing operation. Discussing whether an agency or enterprise should be privatized (question 1 in figure 1) is clearly beyond the scope of this article. The answer involves effi- ciency considerations and public interest issues that policymakers need to evalu- ate carefully in each case. This section addresses whether downsizing should precede privatization when the latter is advisable (see question 2). If the government does not downsize prior to privatization, the new private owners have to deal with labor redundancies. Because of the ensuing differences in the extent of labor shedding, in the amount of compensation, and in the privatization price of the enterprise, assessing the net gains from downsizing prior to privatization may be difficult. But a net loss is likely. The total number of displaced workers may be larger when downsizing is managed by the government prior to privatization than when it is left to the new owners. Kikeri (1997) reports examples from various countries where the new owners kept the labor force more or less intact. Based on a more systematic comparison of employment patterns across Polish firms during the transition to a market economy, Frydman and others (1997) show that employment cuts were larger in state-owned enterprises than in otherwise similar privatized firms. Furthermore, when the government is in charge of downsizing, it may sepa- rate the wrong workers from their jobs at an excessively high cost. A wrong composition of the separations is possible because governments usually are not good at managing human resources. If they could make the right decisions re- garding whom to retain and whom to lay off and, in addition, if they were able 18 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 to deliver on those decisions, the rationale for privatization would be seriously weakened. An excessively high cost of separation is likely because governments can shift part of the downsizing costs to the taxpayer, for instance in the form of early retirement programs, while in principle the new owners cannot. The temp- tation to resort to golden handshakes should therefore be stronger when downsizing takes place prior to privatization. Case studies suggest that this has happened in practice (see Galal and others 1994). Unnecessary downsizing costs cannot be recovered through a higher privatization price of the state-owned enterprise. At the theoretical level, the privatization price would of course increase every time a redundant worker is separated from his or her job. But getting rid of workers who are not redundant would not increase that price. And even for the genuinely redundant workers, the increase would be equal to the amount of resources the new owners would have spent to secure their separation, not to the amount of resources actually spent (directly and indirectly) by the government. Therefore, the net proceeds from downsizing prior to privatization can be expected to be negative. At the empirical level, some evidence suggests that the increase in privatization prices resulting from downsizing prior to privatization may not be worth its cost. In a study of the determinants of auction prices for 263 Mexican enter- prises privatized between 1983 and 1992, L6pez-de-Silanes (1997) finds that downsizing had a marginal impact on privatization prices. The effect was actu- ally insignificant in one specification and only weakly significant in another. If the significant estimate is taken literally, a 5 percent reduction in employment prior to privatization increases the price of the enterprise by 6 percent. Given the substantial cost of prior restructuring policies, L6pez-de-Silanes draws the key lesson that governments should not do too much: they should simply sell. Financial returns to downsizing prior to privatization are most likely nega- tive, but what about economic returns? Most likely, aggregate welfare would increase if the new managers of the privatized firms made the downsizing deci- sions. Discrepancies arise between financial and economic returns because of the presence of several distortions and imperfections, in addition to overstaff- ing, including distorted pay scales in the public sector and deadweight losses from taxation. The new management has no interest in keeping distorted pay scales, so wages in the privatized enterprise should move closer to the alternative earnings of the workers. And privatization usually eliminates the soft budget constraint, so the fiscal externality disappears, too. Downsizing may be justified prior to privatization, however, if it enhances the credibility of the reform process (see Vickers and Yarrow 1991 and World Bank 1995). The ability to overcome labor resistance and trim employment could signal the government's commitment to privatization. This signal, in turn, would reduce the uncertainty faced by potential investors, thus making privatization possible. If the government took no action to overcome the opposition of those who stand to lose from privatization, chances are there would be no bids for the enterprise to be privatized. Rama 19 Question 3 in figure 1 concerns the adverse selection problem. The willing- ness of multilateral organizations and donor countries to lend substantial amounts of money for severance pay may create a bias in favor of voluntary separation schemes. Moreover, buying out the workers is a simple and convenient way to defuse the opposition to public sector reform. But severance pay creates an in- centive for the most productive workers to leave the public sector and for the least productive ones to stay. This may not be socially desirable in the case of public sector units producing valuable services. In this case, it is better to target separations based on the observable characteristics of the workers and to use a menu approach to deal with their unobservable differences. Questions 4 and 5 in figure 1 concern assistance for separated workers. What- ever the combination of cash, retraining, and redeployment services, the govern- ment should not spend more than the amount needed to buy out the redundant workers. The expected losses from separation thus provide a benchmark against which to judge both the existing laws on compensation and the ad hoc packages proposed in the context of downsizing. Analysts can use labor market data to predict the expected losses based on observable characteristics of the workers, such as education, seniority, and gender. Workers' predicted losses may be higher than their legal compensation. In this case, given that public sector workers are rarely poor, even after separation, it is preferable to apply the law. Alternatively, the predicted losses may be lower than legal compensation. In Guinea-Bissau, for instance, severance pay for civil servants is set at 10 months of salary per year of service. The average compensa- tion package thus amounts to 9.6 years of salary. Using the procedure proposed in section IV, the average loss from separation would be 4.2 years of salary (Chong and Rama 1998). In a case like this, it is advisable to try to modify the laws governing severance pay. Retraining and other redeployment programs deserve special attention when assessing the cost of the assistance to be provided to separated workers. More resources may be spent on this component than on direct compensation. How- ever, the evidence on the effectiveness of retraining and other redeployment pro- grams is mixed at best. If these programs are to be part of the downsizing opera- tion, a safeguard should be introduced to minimize the potential waste of resources. Basically, separated workers should be allowed to choose between enrolling in any of the programs offered and cashing in the equivalent of their marginal cost. This demand-driven approach would make it more difficult for vocational training programs and other (often ailing) government agencies to divert resources from the downsizing operation. VIII. CONCLUDING REMARKS Public sector downsizing may become a major reform endeavor in developing countries in the coming years, much the same as trade liberalization and finan- cial liberalization were in the past two decades. These reform efforts can sub- 20 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I stantially improve economic efficiency. But the risks are considerable, too. The comparison between the successful experience with trade liberalization and the more mixed record with financial liberalization shows the importance of a well- designed reform protocol. This article and, more generally, the Public Sector Retrenchment project represent an attempt to sketch a protocol for public sector downsizing. This attempt should be interpreted with great caution. Some of the findings and policy recommendations may need to be adjusted, and some are possibly wrong. More research and experimentation are certainly needed to move in the direc- tion of a more comprehensive and reliable protocol. Forthcoming downsizing operations provide an ideal opportunity to test and evaluate some of the hy- potheses and recommendations. A strong interaction of policy and research in the immediate future could contribute to the success of public sector downsizing endeavors over the longer run. But even a carefully designed protocol could prove ineffective if the mecha- nisms that led to overstaffing in the first place remain unchallenged. The equilib- rium level of public sector employment is probably determined by political forces operating in the context of a particular institutional setting. Temporary incen- tives to downsize, such as financing for severance pay packages, will probably fail to modify that equilibrium. The institutional setting itself would need to evolve in the direction of increased professionalism and efficiency. Downsizing operations should therefore be part of a broader effort to reform and modernize the public sector, not just isolated endeavors. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Ahmad, Ehtisham, and Nicholas Stern. 1987. "Alternative Sources of Government Rev- enue: Illustrations from India, 1979-80." In David Newbery and Nicholas Stern, eds. The Theory of Taxation for Developing Countries. New York: Oxford University Press. Alderman, Harold, Sudharshan Canagarajah, and Stephen Younger. 1996. "A Com- parison of Ghanaian Civil Servants' Earnings before and after Retrenchment." Jour- nal of African Economies 4(2):259-88. Basu, Kaushik, Gary S. Fields, and Shub Debgupta. 1996. "Retrenchment, Labor Laws, and Government Policy: An Analysis with Special Reference to India." Presented at the conference on Public Sector Retrenchment and Efficient Compensation Schemes, World Bank, Washington, D.C. Processed. Brander, James A., and Barbara J. Spencer. 1994. "Trade Adjustment Assistance: Wel- fare and Incentive Effects of Payments to Displaced Workers." Journal of Interna- tional Economics 36(May):239-61. Campa, Jos6 Manuel. 1996. "Public Sector Retrenchment: Spain in the 1980s." Stern School of Business, New York University, New York. Processed Carneiro, Francisco, and Indermit Gill. 1997. "Effectiveness and Financial Costs of Voluntary Separation Programs in Brazil: 1995-1997." Economic Notes 25. Coun- Rama 21 try Department I, Latin America and the Caribbean Regional Office, World Bank, Washington, D.C. Processed. Chong, Alberto, and Martin Rama. 1998. "A Compensation Package for Separated Public Sector Workers in Guinea-Bissau. " Development Economics Department, World Bank, Washington, D.C. Processed. Colclough, Christopher, ed. 1997. Public Sector Pay and Adjustment: Lessons from Five Countries. London: Routledge. Devarajan, Shantayanan, Lyn Squire, and Sethaput Suthiwart-Narueput. 1995. "Reviving Project Appraisal at the World Bank." Policy Research Working Paper 1496. Policy Research Department, World Bank, Washington, D.C. Pro- cessed. Diwan, Ishac. 1994. "Public Sector Retrenchment and Severance Pay: Nine Proposi- tions." In Shahid A. Chaudhry, Gary J. Reid, and Waleed H. Malik, eds., Civil Ser- vice Reform in Latin America and the Caribbean: Proceedings of a Conference. World Bank Technical Paper 259. Washington, D.C. Fallick, Bruce. 1995. "A Review of the Recent Empirical Literature on Displaced Work- ers." Finance and Economics Discussion Series 95-14. Federal Reserve Board, Wash- ington, D.C. Processed. Fiszbein, Ariel. 1994. "An Opportunity Cost Approach to Redundancy Compensation: An Application to Sri Lanka." Estudios de Economia: Special Issue on Labor Eco- nomics in Less Developed Countries. 115-26. Fretwell,David,Jacob Benus, and Christopher J. O'Leary. 1998. "Evaluating the Impact of Active Labor Market Programs: Results of Cross-Country Studies in Europe and Central Asia." Europe and Central Asia Regional Office, World Bank: Washington, D.C. Frydman, Roman, Cheryl W. Gray, Marek Hessel, and Andrzej Rapaczynski. 1997. "Private Ownership and Corporate Performance: Some Lessons from Transition Economies." Policy Research Working Paper 1830. Policy Research Department, World Bank, Washington, D.C. Processed. Galal, Ahmed, Leroy Jones, Pankaj Tandoon, and Ingo Vogelsang. 1994. Welfare Con- sequences of Selling Public Enterprises: An Empirical Analysis. New York: Oxford University Press. Hamermesh, Daniel. 1989. "What Do We Know about Worker Displacement in the U.S.?" Industrial Relations 28(1):51-59. Jacobson, Louis, Robert LaLonde, and Daniel Sullivan. 1993. "Earnings Losses of Dis- placed Workers." American Economic Review 83(4):685-709. Jenkins, Glenn P., and Claude Montmarquette. 1979. "Estimating the Private and So- cial Opportunity Cost of Displaced Workers." Review of Economics and Statistics 619(3):342-53. Kikeri, Sunita. 1997. Privatization and Labor: What Happens to Workers When Governments Divest? World Bank Technical Paper 396. Washington, D.C. Kouame, Auguste-Tano. 1997. "Workers' Severance Pay Packages in Privatization of State-Owned Enterprises in Bangladesh, Pakistan, and Sri Lanka: Options for the Future." Internal Discussion Paper. South Asia Regional Office, World Bank, Wash- ington, D.C. Processed. Levy, Anat, and Richard McLean. 1997. "Optimal and Sub-optimal Retrenchment Schemes: An Analytical Framework." Policy Research Department, Rutgers Univer- sity, New Brunswick, N.J. Processed. 22 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Lmdauer, David L., and Barbara Nunberg, eds. 1994. Rehabilitating Government: Pay and Employment Reform in Africa. A World Bank Regional and Sectoral Study. Washington, D.C. London Economics. 1996. "The Impact of Privatisation on Labour in Africa." Prepared for the Africa Regional Office, World Bank, Washington, D.C. Processed. L6pez-de-Silanes, Florencio. 1997. "Determinants of Privatization Prices." Quarterly Journal of Economics 112(4):965-1025. Middleton, John, Adrian Ziderman, and Arvil Van Adams. 1993. Skills for Productiv- ity: Vocational Education and Training in Developing Countries. New York: Oxford University Press. Mills, Bradford, and David Sahn. 1996. "Life after Public Sector Job Loss in Guinea." In David Sahn, ed., Economic Reform and the Poor in Africa. Oxford: Clarendon Press. Nunberg, Barbara. 1994. "Experience with Civil Service Pay and Employment Reform: An Overview." In David L. Lindauer and Barbara Nunberg, eds., Rehabilitating Gov- ernment: Pay and Employment Reform in Africa. A World Bank Regional and Sectoral Study. Washington, D.C. Orazem, Peter, Milan Vodopivec, and Ruth Wu. 1995. "Worker Displacement during the Transition: Experience from Slovenia." Policy Research Working Paper 1449. Policy Research Department, World Bank, Washington, D.C. Processed. Rama, Martin. 1997. "Efficient Public Sector Downsizing." Policy Research Working Paper 1840. Policy Research Department, World Bank, Washington, D.C. Processed. Robbins, Donald. 1996. "Public Sector Retrenchment and Efficient Severance Pay Schemes: A Case Study of Argentina." Harvard Institute for International Develop- ment, Cambridge, Mass. Processed. SSvelnar, Jan, and Katherine Terrell. 1991. "Reducing Labor Redundancy in State-Owned Enterprises. " Policy Research Working Paper 792. Policy Research Department, World Bank, Washington, D.C. Processed. Tansel, Aysit. 1997. "Public Sector Retrenchment and the Impact of Labor Shedding Programs on Workers in Turkey." Department of Economics, Middle East Technical University, Ankara. Processed. Topel, Roberr. 1990. "Specific Capital and Unemployment: Measuring the Costs and Consequences of Job Loss." Carnegie-Rochester Conference Series on Public Policy 33(0):181-224. Van Ginneken, Wouter. 1991. Government and Its Employees: Case Studies of Devel- oping Countries. Aldershot, Hants, U.K.: Avebury. Vickers, John, and George Yarrow. 1991. "Economic Perspectives on Privatization." Journal of Economic Perspectives 5(spring): 111-32. World Bank. 1995. Bureaucrats in Business: The Economics and Politics of Govern- ment Ownersbip. A World Bank Policy Research Report. New York: Oxford Univer- sity Press. Younger, Stephen. 1996. "Labor Market Consequences of Retrenchment for Civil Ser- vants in Ghana." In David Sahn, ed., Economic Reform and the Poor in Africa. Ox- ford: Clarendon Press. THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 23-66 Cross-Country Evidence on Public Sector Retrenchment John Haltiwanger and Manisha Singh This article reports the results from a survey of public sector employment retrench- ment episodes across a wide variety of developing and transition economies. The in- formation collected and analyzed is primarily from internal World Bank documents and in-depth interviews with World Bank staff having operational information about experiences in specific countries. Using the information collected on 41 retrenchment programs across 37 countries, the article analyzes the relationships between the fac- tors leading to retrenchment, the scope and nature of retrenchment, and the methods used to accomplish the retrenchment. The discussion of methods includes an analysis of the mix of involuntary and voluntary employment reduction programs, the com- pensation schemes offered, and the extent of targeting of specific types of workers. Although relevant quantitative information is limited, the article also attempts to evalu- ate the outcome of the programs on several dimensions. The most striking findings relate to analysis of the factors leading a significant fraction of programs to rehire workers separated from the public sector (thereby defeating the programs' objective). In addition, the article relates program characteristics to calculated summary financial payback indicators and to the nature of the labor market adlustment. The retrenchment of public sector employment is an increasingly pervasive phe- nomenon. Many industrial, developing, and transition economies have recently faced the issue of downsizing their public sector employment. The reasons un- derlying the downsizing vary considerably across countries. For some, it is a general move toward a more market economy; for others, it is a reduction in the role of the military or an attempt to reduce a bloated bureaucracy; for others, it is sparked by a fiscal crisis necessitating a severe cutback in government spend- ing; and, finally, for some, it is a combination of some or all of these. Given the pervasiveness of the phenomenon, there is no shortage of studies of individual countries or episodes. Although country-specific studies are quite use- John Haltiwanger and Manisha Singh are with the Department of Economics at the University of Maryland. This is the revised version of a paper presented at the World Bank conference on Public Sector Retrenchment and Efficient Compensation Schemes held in November 1996. The authors are grateful to many officials at the World Bank for providing data and their time for extensive interviews on specific public sector retrenchment programs. They thank participants at an informal seminar held at the World Bank in January 1996, participants at the conference in November, Martin Rama, Peter Fallon, and the anonymous referees for helpful comments. Catherine Buffington and Andrew Figura provided excellent research assistance C 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 23 24 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 ful, it is difficult to draw common lessons from them, and there is relatively little analysis comparing experiences across countries. Important exceptions are Sveinar and Terrell (1991), which provide a useful comparison across six countries for retrenchment in the transport sector, and Lindauer and Nunberg (1994), which provide a book-length discussion of civil service reform. Our study has benefited substantially from the insights in both of these publications. In this article, we compile and analyze information on the recent retrenchment experiences across a large number of countries and retrenchment episodes. We take a broad view of public sector employment and associated retrenchment. Public sector employ- ment includes civil service workers, military personnel, and employees of public sector enterprises. Comparing and contrasting the experiences across countries are formidable tasks. Beyond the usual problems of cross-country comparisons, no central da- tabase has compiled the requisite information that permits such analysis. Ac- cordingly, we compiled and collected our own information from myriad sources, but primarily from internal World Bank documents and interviews of World Bank staff having operational information about retrenchment experiences in individual countries. (Section II provides details regarding the collection of in- formation; a complete listing of individual country summaries is available from the authors upon request.) Our analysis is based on 37 countries and 41 retrenchment programs. Most of the programs for which we could collect detailed information commenced in the early 1990s, and many of the programs are ongoing (which makes analysis of the ultimate success of the programs difficult). The gross number of workers separated in all the programs considered exceeds S million, with a somewhat smaller net reduction in employment. The discrepancy reflects the fact that some programs exhibited significant rehiring of separated workers or hiring of new workers by the public sector. In the analysis that follows, we look closely at the characteristics of programs that involved significant rehiring and new hires. Al- though new hires may be part of a coherent plan to restructure the public sector work force, rehiring is clearly not an indicator of success. The converse is not true; that is, the absence of rehiring is not synonymous with success. Also, a restructuring could involve separating employees en masse and subsequently rehiring those with necessary specific skills or rehiring workers on temporary terms or both. No program in our sample intended such purposive rehiring. Evaluating the costs and benefits of individual programs in a more compre- hensive fashion is inherently difficult. In principle, the requisite information in- cludes the compensation packages offered, the relative productivity and wages of the displaced workers in the public and private sectors, and the adjustment costs borne directly by the affected workers and the entire economy. Measure- ment of much of this-particularly the nature of the adjustment costs-is well beyond the scope of available information. Nevertheless, we can document con- siderable details about the magnitude and forms of compensation used in re- trenchment programs and the wage bill savings, and we can characterize some Haltiwanger and Stngh 25 of the other aspects that are relevant for evaluation. To summarize the available information on costs and benefits, we calculate a simple financial break-even indicator measured as the number of years required for the present value of financial costs to equal the present value of financial gains. We also examine refinements of this indicator that attempt to incorporate the productivity gains generated from the retrenchment. In turn, we relate these summary financial indicators to other program characteristics. Section I provides a brief presentation of the underlying conceptual frame- work for characterizing the private and social costs and benefits of a public sector retrenchment program. The analysis is deliberately simple and borrows heavily from the existing literature. Our primary objective is to provide guid- ance about the type of information that is required to compare and analyze alternative programs and to discuss the interaction between the conceptual and measurement issues that must be confronted. Section II outlines the methodol- ogy used to collect information for the individual countries and to compile the information in a systematic fashion. Section III presents the analysis of the infor- mation collected as well as some detailed discussion of individual countries to illustrate the patterns that emerge. Section IV offers concluding remarks. I. CONCEPTUAL FRAMEWORK AND MEASUREMENT ISSUES A simple conceptual framework helps us to organize the information we have collected and provides some guidance on the type of factors relevant for com- paring and analyzing retrenchment programs across countries. Although the dis- cussion in this section borrows heavily from the existing literature, our charac- terization of the relevant issues emphasizes some points that are neglected in the literature. For example, many of the conceptual issues discussed in this section are discussed more formally and with greater attention to the range of relevant issues in Diwan (1993a, 1993b) and Rama (1997). Because our ultimate objec- tive is to quantify the relevant measures using data across countries, we focus on the interaction between conceptual and measurement issues. The analysis considers six relevant private and social incentives for the re- trenchment decision regarding the marginal worker. Although we do not char- acterize all relevant sources of heterogeneity explicitly, differences across work- ers in the following present values may reflect differences in ability, experience, skills, horizons, discount rates, and mobility costs. Pprivate is the present discounted value of the worker's productivity in the private sector. Ppubhic is the present discounted value of the worker's productivity in the public sector. Cind,vidual IS the present discounted value of adjustment costs borne by the individual in relocat- ing from the public to the private sector (for example, job search costs, reloca- tion costs, and time spent unemployed). Csoc,ai is the present discounted value of adjustment costs borne by society for the individual to relocate from the public to the private sector (that is, adjustment costs borne by the individual plus spillover effects such as congestion effects). Wpubh1c is the present discounted value of the 26 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 earnings of the worker in the public sector. Wpr,>ate is the present discounted value of the earnings of the worker in the private sector. The discussion here and the subsequent analysis focus primarily on the reallo- cation of workers from the public to the private sector taking into account the relevant productivity, wages, and costs of labor market adjustment. Public sec- tor downsizing also likely involves the reallocation of capital. Even though our focus is on the reallocation of workers, the implications of capital reallocation and its interaction with worker reallocation deserve further attention. Also, ex- plicit modeling of the distinction between the present discounted value of earn- ings and adjustment costs would require incorporating the fact that transitions from a public sector job to a private sector job may involve several transitions and, accordingly, several spells of unemployment. As emphasized by Hall (1995), a potentially important explanation of persistence in unemployment rate dy- namics is that separations tend to beget further separations. The principle of social optimality considered here is that all resources should be allocated to their highest-valued use (net of relocation costs given the existing allocation of resources). Thus it is socially optimal to relocate the marginal worker to the private sector if: (1) Pprivate - Csocial > Ppublic- An individual worker will choose to stay in a public sector job as long as: (2) Wptbhhc > Wprlvate- Cndividual- If workers are paid more than their marginal product in the public sector (that is, WpubliC > PpUblhC) then it is easy to imagine that inequalities 1 and 2 can both hold simultaneously. That is, it is in society's interest for the marginal worker to relocate, but it is not privately optimal. It may be that individual workers and the government discount the future differently because of differential access to capital markets. Such differential access to capital markets is, in principle, consistent with the specification and associated discussion considered here, but explicit consideration of the role of capital markets deserves more attention. For example, unemployment benefits and other forms of worker safety net assistance are often justified as a form of social insurance in the face of imperfections in the capital market. Because en- hancements of the worker safety net are often part of a retrenchment program, explicit consideration of the role of differential capital market access iS quite relevant. Several additional factors are potentially important in evaluating the optimality of a retrenchment program. One factor is the expenditures incurred to pay the wages of public workers or, alternatively, the compensation packages offered to induce workers to relocate voluntarily. If taxes are not distortionary and there is no distortionary rent-seeking behavior, these expenditures should be viewed as transfers and thus should not affect efficiency. Under these strong assump- tions, these terms should not appear in inequality 1. However, in the presence of Haltiwanger and Singh 27 distortionary taxes (including the inflation tax) and distortions from rent- seeking behavior, such transfers yield efficiency losses. In addition, an excessive wage bill or fiscal crisis may imply that retaining redundant workers may ad- versely affect other government services. To incorporate these effects in a modified version of inequality 1, suppose that there is a distortions-based loss function L(.) that is an increasing function of the transfers to workers. Taking into account these distortionary losses in inequality 1, retrenchment is optimal if: (3) Pprlvate - Csoc1a, - L(COMP) > Ppublc- L(Wpubhc) where COMP reflects the present discounted value of the compensation or other assistance programs that accompany the retrenchment. A related issue is whether the nature of the distortions from the transfers depends on the timing of the program as well as on the net present value of the transfers. For example, rents in the form of artificially high public sector wages may have dynamic distortionary effects beyond those associated with a one-time transfer. Policymakers face the problem (shared in our analysis) of measuring the com- ponents in inequalities 1, 2, and 3. Many of the components are difficult to measure or difficult to observe in the presence of worker heterogeneity (outside opportunities and individual productivity). Among the most difficult to assess are the adjustment costs. Factors influencing adjustment costs include the degree of labor market flexibility (barriers to adjustment in terms of hiring and firing costs), the worker safety net, informal compared with formal sector develop- ment, and the method, scope, and speed of retrenchment. Social adjustment costs will be greater than private adjustment costs to the extent that there are spillover or externality effects of worker displacement. The spillover and externality effects include the impact of changes in the demand for final goods due to reduced consumption expenditures by affected workers, con- gestion externalities among job searchers, and social disruption (such as na- tional strikes). For further discussion of the congestion externalities among job seekers, see, for example, Blanchard and Diamond (1989, 1990). For further discussion of the idea that employment and earnings losses by workers in one sector may generate spillover effects in other sectors by changing the demand for final goods, see, for example, Cooper and Haltiwanger (1996). The method, scope, and speed of retrenchment potentially influence these spillover effects. In addition, the concentration of the retrenchment in a local community may imply important local spillover effects even if there are no economywide spillover effects. The nature and magnitude of these adjustment costs are very difficult to mea- sure, and we have relatively little quantifiable information on adjustment costs that we can use in our comparison across countries. The most direct evidence available is from recent studies of privatization in transition economies for which the massive restructuring implies that adjustment costs take center stage (see, for example, the studies in Commander and Coricelli 1994). Even in these studies, 28 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I information on the nature of the adjustment process is often quite limited. In general, and particularly in the transition economies, characterizing these ad- justment costs is at the heart of the debate regarding gradualism and the big push in efforts to reduce the role of the public sector in the economy. Another difficult measurement and conceptual problem is evaluating the pre- and post-retrenchment productivity of workers in the public sector. Although there may be widespread agreement that some public sector workers are redun- dant, quantifying this is extremely difficult. The nature of the measurement prob- lems in evaluating pre- and post-retrenchment productivity will vary depending on the type of public sector employment (civil service, military, or public sector enterprises producing goods and services). Further, some of the need for re- trenchment may reflect more of a need for restructuring than for simply downsizing. For example, low morale and low productivity of public sector employees may reflect poor organizational structure or wage compression. Thus a program may involve changes in the organizational structure and the wage structure that appear to be financially expensive but are well motivated by such factors. Simple financial indicators that do not adequately incorporate such fac- tors should not be used to evaluate the need for or success of a program. In terms of inequality 3, this discussion implies that the public sector productivity term needs to be interpreted broadly. That is, retrenching the marginal worker will yield the direct loss of the worker's output (if any) but may also yield pro- ductivity gains from the reorganization that accompanies downsizing. Beyond the problems of measurement and assessment, it is often politically necessary to implement the retrenchment scheme through a voluntary program. This necessitates the provision of some incentive payment in the form of sever- ance pay or pensions (denoted INCENTIVE below) such that the following in- equality would hold for an individual worker: (4) INCENTIVE > Wpubhc- (Wprivate - Cindnixdual)- In principle, with complete information about worker heterogeneity, it is pos- sible to design an optimal incentive scheme for each worker (see Diwan 1993a, 1993b and Rama 1997 for a formal analysis). One well-recognized problem that immediately emerges is that if a simple, common incentive package is of- fered to all workers, heterogeneity across workers implies that only those with the lowest rent will depart. It is important to emphasize that some aspects of the selection process may be unfavorable, while others may be advantageous. For example, in environments with wage compression, a common package offered to a wide class of workers implies that the most skilled and capable workers will leave. Individuals with better outside opportunities or low adjustment costs are more likely to leave. Other things being equal, this voluntary aspect of selection will be favorable. The implication of this discussion is that an optimal incentive scheme must be individually tailored to reflect worker heterogeneity in rents and adjustment costs but that this is difficult to implement in practice in the face of imperfect infor- Haltiwanger and Singh 29 mation. This tension is at the center of the debate about the pros and cons of voluntary and involuntary retrenchment programs. As noted, voluntary pro- grams have the advantage that they yield some favorable voluntary selection of workers with good outside opportunities or low adjustment costs without re- quiring the policymaker to have complete information. However, workers with good outside opportunities are likely the most productive public sector workers, so that the voluntary selection may be adverse. The problem with adverse selec- tion is apt to be especially severe in environments with wage compression. Levy and McLean (1996) discuss optimal schemes using different types of severance contracts or auctions. Involuntary programs potentially permit specific targeting of groups of work- ers on observable characteristics but do not take advantage of favorable volun- tary selection on unobservable characteristics. Further, involuntary schemes may also yield high adjustment costs. For example, mass involuntary layoffs may involve substantial private and social adjustment costs, and in this case the na- ture of the safety net takes on particular importance. Finally, the political will (ability) to undertake and sustain schemes with an involuntary component may be lacking. Putting these pieces together suggests that our comparison and analysis of alternative programs should consider the following factors. First, we must con- sider the factors leading to retrenchment (fiscal crisis, overstaffing, low morale as a result of wage compression) because they may provide insights about the relationship between wages and productivity in the private and public sectors. Second, several factors influence the adjustment costs, including the scope and speed of retrenchment, the mechanism used (involuntary or voluntary), the use of targeting (for example, based on skill or age), and the nature of labor market flexibility and the safety net. Many of these same factors are important (espe- cially the method used and the nature of targeting) in terms of the need for individual tailoring of plans given worker heterogeneity. Finally, the magnitudes of the financial costs (WpUbl,c and INCENTIVE) are relevant for characterizing the magnitude of the transfers. II. DATA COLLECTION The survey of public sector retrenchment programs across developing and transition countries was carried out on two dimensions. First, we collected inter- nal World Bank documents relating to macroeconomic and public sector adjust- ment in these countries. We drew up a preliminary portrait of the issues relating to each country using various World Bank documents.' We used a variety of external sources to supplement these documents. Second, we interviewed World 1. The World Bank documents included Staff Appraisal Reports, Memoranda and President's Recommendations, President's Reports, supervision memoranda, Project Completion Reports, Implementation Completion Reports, Project Performance Audit Reports, Operations Eva[uation StudLes, Country Economic Memoranda, and sector and other economic reports. 30 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Bank officials associated with the retrenchment programs to obtain more direct information and assessment. The interviews, conducted over more than eight weeks during the summer of 1995, typically yielded further acquisition of rel- evant documents. The countries, retrenchment programs, and associated offi- cials were selected using the World Bank's electronic management information system and the adjustment lending database called ALCID. Appendix table A-1 lists the surveyed countries and programs. The projects and loans listed in the World Bank's management information system go back to the mid-1980s. The selection of programs was mostly restricted to this list. In general, the movement of World Bank staff away from early programs reduced the usefulness of such interviews. The interview transcripts, documents, reports, and articles relating to each program were synthesized and summarized using a uniform outline. The infor- mation presented relates to four broad aspects characterizing retrenchment, the nature of labor turnover and institutions, cost-benefit analyses in financial and economic terms, and monitoring and evaluation of the program. A complete listing of the individual country summaries is available from the authors upon request.2 The information collected and assimilated ranges from quantitative informa- tion about scope, speed, and financial costs and benefits to qualitative informa- tion about the factors precipitating the retrenchment, as well as characteristics of the program such as the methods used and the productivity gains observed in the public sector. The quality and completeness of the information collected vary substantially across programs. This blend of quantitative and qualitative information yields an analysis that is primarily descriptive-an attempt at sum- marizing the basic facts and drawing out observable patterns. Not surprisingly, a host of institutional and idiosyncratic factors appear to be important. These institutional and idiosyncratic factors serve as an additional reminder to inter- pret the analysis of basic patterns with caution. III. SURVEY AND ANALYSIS OF CROSS-COUNTRY EVIDENCE We begin the analysis by characterizing some basic facts about the programs for which we gathered information. Tables 1 and 2 present summary statistics about the programs surveyed (summary information on each program is pro- vided in the cross-country appendix tables). The analysis is based on 41 pro- grams in 37 countries. The World Bank has not historically provided direct as- sistance for public sector employment retrenchment programs. (As of February 1997, it allowed lending for severance payments provided they entail no organi- zation closure.) However, under the umbrella of a comprehensive assistance package, an agreement with a country often stipulates that the country use do- mestic counterpart funds for specific purposes. Under such umbrella agreements, 2 The detailed country summaries total more than 200 pages, including extensive references to documents, papers, and other sources beyond those cited here. Haltiwanger and Singh 31 public sector retrenchment may be part of the overall package. In our survey, about 65 percent of the programs had this umbrella connection to World Bank financing. Table 1 shows that total separations are greater than the total reduction in employment, reflecting rehiring and new hires. Most programs did not experi- ence rehiring of the same workers who had departed, but a nontrivial fraction (20 percent) experienced significant rehiring. New hires were also relatively rare (about 13 percent of programs). Some programs represented only good inten- tions, with no workers actually separated. One of the programs included in our survey is Hungary's massive public sector retrenchment program. This program is a clear outlier on a number of dimensions, but we felt it was important to include large programs among the transition economies as points of contrast. In table 1, the method of employment reduction is divided into three basic categories: involuntary-hard refers to layoffs; involuntary-soft refers to employ- ment reductions generated by strict enforcement of rules such as mandatory retirement and the removal of ghost workers (workers on the payroll who do not exist, although someone is collecting the paycheck); and voluntary refers to programs in which employment reductions were achieved through workers who quit voluntarily (for example, early retirements). Table 1 shows that involuntary-hard reductions dominate total reductions in employment. How- ever, this is primarily driven by the massive involuntary reductions in Eastern Europe. By contrast, the use of voluntary and involuntary-soft measures is the prevalling pattern in Latin America, Asia, and Africa. Table 2 indicates that the total financial costs of the retrenchment programs were large (exceeding $12 billion), with the average program having a total cost of $400 million. The total cost is driven in part by enormous increases in pen- sion and safety net expenditures in Poland. From all accounts, the increases for Poland reported in appendix table A-2 are real, but they skew totals and aver- ages somewhat. However, much of the subsequent analysis is not sensitive to outliers on this dimension. That is, much of the analysis is based on the percent- age of programs that exhibit various characteristics. The financial costs of the programs took a variety of forms, including sever- ance payments, higher pension liabilities, and enhancements to the worker safety net. The worker safety net is a shorthand term for the various worker assistance programs, including unemployment benefits, job search assistance, training, and relocation assistance designed to aid in the process of worker reallocation. As is clear from the minimum and the maximum in table 2, the scope and mix of packages varied considerably across countries. As the patterns across continents show, the transition economies in Eastern Europe relied heavily on the worker safety net. Countries in the Southern Hemisphere and Asia relied much more heavily on direct compensation to workers, such as severance pay and enhanced pensions. The latter pattern is particularly striking for severance pay, which was 92 percent of total costs in Asia, 74 percent in Latin America, and 51 percent in Africa. Table 1. Scope of Retrenchment: Number of Workers, 1990s By instrument Employment Involuntary- Involuntary- Region reduction hard soft Voluntary Separations Rehires New hires Total All cases 4,710,566 1,178,394 752,809 577,188 5,033,875 282,307 41,002 Africa 892,062 104,035 140,896 97,878 909,705 15,900 1,743 Asia 233,311 0 11,000 172,711 255,311 3,000 19,000 Europe 2,853,559 883,259 132,000 105,000 2,853,559 0 0 Latin America 731,634 191,100 468,913 201,599 1,015,300 263,407 20,259 Median All cases 32,900 0 1,800 4,800 37,500 0 0 Africa 10,061 1,922 6,861 3,696 10,061 0 0 Asia 21,925 0 () 14,373 23,425 0 0 Europe 172,959 42,000 0 17,500 172,959 0 0 Latin America 26,000 0 0 4,599 56,409 0 0 Average All cases 117,764 39,280 27,882 18,037 125,847 7,058 1,025 Africa 59,471 10,404 14,090 7,529 60,647 1,060 116 Asia 29,164 0 2,200 28,785 31,914 375 2,375 Europe 407,651 176,652 33,000 26,250 407,651 0 0 Latin America 73,163 19,110 58,614 22,400 101,530 26,341 2,026 Minimum All cases 0 0 0 0 247 0 0 Africa 247 0 0 0 247 0 () Asia 6,500 0 0 0 6,500 0 0 Europe 35,000 0 0 0 35,000 0 0 Latin America 0 0 0 0 2,034 0 0 Maximum All cases 1,661,000 547,300 424,095 112,000 1,661,000 163,059 19,000 Africa 541,200 57,000 75,000 59,810 547,200 7,500 1,743 Asia 69,466 0 7,000 69,466 69,466 3,000 19,000 Europe 1,661,000 547,300 132,000 70,000 1,661,000 0 0 Latin America 405,995 100,000 424,095 112,000 424,095 163,059 18,100 Note Values are based on 41 retrenchment programs in 37 countries. Information about worker separations by instrument is incomplete in several cases. Source: Authors' calculations based on individual country summaries (see appendix tables for cross-country tabulations) Table 2. Scope of Retrenchment: Financial Costs, 1 990s Financial costs (millions of dollars) Financial benefits (millions of dollars) Severance Enhanced Safety Cost per Annual wage Region Total payments pension net worker Total bill savings Total All cases 12,074 2,708 6,098 3,268 n.a. 4,206 2,834 Africa 500 255 20 225 n.a. 673 96 Asia 1,400 1,281 71 48 n.a. 179 -153 Europe 8,583 0 5,588 2,995 n.a. 1,846 1,446 Latin America 1,591 1,172 419 0 n.a. 1,508 1,445 Median All cases 42 13 0 0 1,085 17 10 Africa 20 13 0 0 1,476 10 8 Asia 50 25 0 24 1,040 9 9 Europe 454 0 25 859 616 923 723 Latin America 280 112 0 0 4,735 222 222 Average All cases 402 87 244 192 4,142 168 123 Africa 38 20 2 32 3,630 61 11 Asia 200 183 14 24 3,651 26 -22 Europe 2,146 0 1,397 998 3,817 923 723 Latin America 265 167 70 0 5,695 302 289 Minimum All cases 0 0 0 0 0 -157 -157 Africa 2 0 0 0 320 0 0 Asia 0 0 0 0 0 -157 -157 Europe 6 0 0 6 24 298 298 Latin America 2 0 0 0 0 -15 -15 Maximum All cases 7,669 1,140 5,538 2,130 17,108 1,548 1,148 Africa 200 80 20 200 13,166 542 30 Asia 1,188 1,140 71 48 17,108 350 83 Europe 7,669 0 5,538 2,130 14,012 1,548 1,148 Latin America 530 425 418 0 16,000 1,063 1,000 n.a. Not applicable Note Values are based on 41 retrenchment programs in 37 countries The percentages computed from this table (for example, severance as a percentage of total cost) may not match the percentages given in table 5 that are based on a count of programs. Source Authors' calculations based on individual country summaries (see appendix tables for cross-country tabulations). 34 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 A wide variety of circumstances led to the retrenchment programs across coun- tries. The most prominent factors leading to retrenchment were fiscal crises and a general effort to reduce the size of government in the economy. However, in some cases, the compelling factors appeared to be structural problems with the type and mix of government workers. Wage compression among public sector workers leading to morale and staffing problems was a common complaint. Overstaffing, including the problem of ghost workers, was another common complaint. Finally, the downsizing of the military played a prominent role as well. Not surprisingly, in many programs multiple factors led to the retrenchment. The differences in the factors generating retrenchment are closely linked to the mix of packages used across countries and continents. For some countries, the retrenchment episode was viewed as a one-time event correcting a perceived relatively narrow problem in a particular public sector agency or enterprise (for example, ghost workers or productivity and morale problems due to wage com- pression). These one-time-event cases typically used some form of direct com- pensation (severance and enhanced pension) and accordingly were more likely to use voluntary methods of retrenchment. By contrast, in some countries (for example, the transition economies), the public sector retrenchment was part of a fundamental change in the role of the public sector in the economy. In these cases, the programs typically paid much greater attention to institutional changes (unemployment benefits, relocation assistance, training assistance) that acted as the safety net for worker reallocation. Much of the detailed information collected involves the method used to re- duce employment, the nature of the compensation provided, and the degree of targeting. Tables 3, 4, and 5 provide summary information on these characteris- tics. Many programs (but less than half) used both involuntary and voluntary reduction methods; relatively few programs used all three methods (table 3). Although voluntary programs appear to be the most popular in terms of the Table 3. Distribution of Employment Reduction Methods in Retrenchment Programs, 1990s Percentage Percentage Type of method of programs of workers Involuntary-hard (layoffs) 41.5 47 0 Involuntary-soft (enforcement of rules, removal of ghost workers) 65.0 30.0 Involuntary (removal of ghost workers) 22.5 3.3 Voluntary 77.5 23.0 Both voluntary and Involuntary 42.5 19.2 All methods 17.5 13.4 Note: Values are based on 41 retrenchment programs in 37 countries The percentage of workers may differ from those computable from table 1 on account of missing values that are subsumed in the totals in table 1. Missing values are excluded here Source. Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). Haltiwanger and Singb 35 Table 4. Distribution of Compensation and Transition Assistance in Retrenchment Programs, 1990s Type of assistance Percentage of programs Severance payment 68.3 Pension enhancement 29.3 No direct compensation 14.6 Safety net 63.4 Safety net (training) 53.7 Unknown 12.2 Note: Values are based on 41 retrenchment programs in 37 countries. Source. Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations) percentage of programs with a voluntary component, most reductions in em- ployment were achieved through involuntary methods. The inclusion of Hun- gary in our 41 programs has a clear influence on employment-weighted results. For the most part, we present unweighted results so that outliers such as Hun- gary do not play a disproportionate role. The removal of ghost workers played a relatively minor supporting role in the employment reductions but was quite prevalent in the African countries in our survey. Table 4 indicates that the primary form of direct compensation to workers was a severance payment. Relatively few programs offered no direct compensa- tion. Most programs involved some enhancement of the overall safety net in- tended to assist unemployed workers and workers attempting to relocate. An important component of the safety net enhancement was often the stipulation of some form of training assistance. As discussed in section II, the underlying theory suggests that designing a program that targets specific individuals or groups of individuals is likely to be important to avoid adverse selection and to promote favorable selection. We attempt to summarize the many different approaches to targeting used in prac- tice by classifying programs into three admittedly crude groups: skill-biased, age-biased, and neutral. Skill-biased programs restrict the program along de- tailed occupational or skill groupings. For example, the restructuring of the work force in the tax administration in Peru was based on new and continuing work- ers passing a written test. Age-biased programs focus on voluntary retirement Table 5. Distribution of Targeting in Retrenchment Programs, 1990s Type of targeting Percentage of programs Skill-biased 53.7 Age-biased 51.2 Neutral 19.5 Unknown 24 Note. Values are based on 41 retrenchment programs in 37 countries Source. Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). 36 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I (and associated pension enhancements) as the means for retrenchment. Neutral programs offer a simple, common package to a wide class of workers with little or no attempt to target specific groups. Some programs are classified as both skill-biased and age-biased if they use a mix of packages that induce selection on both types of criteria. Table 5 indicates that targeting on the basis of skills or age (in the latter case primarily through early retirement programs) was important, but not pervasive. Almost 20 percent of the programs did not use targeting of workers. As we will see in the following sections, the nature of targeting is closely related to a variety of measures of success of the program. Rehires, New Hires, and Other Program Characteristics Tailoring a program to individual characteristics using some form of target- ing is likely to be important to avoid losing the most productive or essential workers. The absence of targeting may yield excessive losses of workers in key areas or with key skills and thus will necessitate either rehiring some of the same workers who departed (a sure sign of poor targeting) or hiring new workers. In many programs, workers are rehired by the same organization and unit. In some cases, they are rehired by a different branch of the public sector. Using the infor- mation collected, we examine the simple bivariate relationships between the like- lihood of rehires and new hires and other program characteristics, including the nature of targeting. Table 6 examines the link between rehiring and program characteristics. The most striking results involve the role of targeting and the type of compensation offered. Programs with rehiring were much less likely to use skill or age target- ing. For example, only 25 percent of programs with rehiring also used skill tar- geting compared with more than 60 percent of programs without rehiring. The flip side of this is that more than 60 percent of the programs with rehiring did no targeting compared with less than 10 percent of programs without rehiring. These findings are consistent with the view that programs that fail to target yield severe adverse selection problems, with the most critical workers departing and creating the subsequent need to rehire these same workers. Some other interesting patterns emerge that are large in magnitude but not statistically significant at conventional levels given the relatively small sample size of 41 observations on individual programs. Specifically, programs with re- hiring were also less likely to have a safety net component and, in particular, some type of training component. The potential inference here is that the pres- sure to rehire may be greater if efforts are not made to assist workers in the transition to the private sector. Table 7 examines the analogous relationships between new hires and pro- gram characteristics. The largest and most significant results again involve the nature of targeting and the type of compensation assistance. Programs with new hires are all age-biased with associated pension enhancements and do not in- clude safety net components. It is not clear, of course, that evidence of new Haltiwanger and Singh 37 Table 6. Relationship between Rehiring and the Characteristics of Retrenchment Programs, 1 990s Program Percentage of programs Percentage of characteristic with rehiring other programs Employment reduction method Involuntary-hard 50.0 39.4 Involuntary-soft 62.5 65.6 Voluntary 62.5 81.3 All methods 12.5 18.8 Nature of targeting Skill-biased 25.0* 60.6 Age-biased 12.5** 60.6 Neutral 62.5'* 9.1 Compensation or assistance Severance payment 75.0 66.7 Pension enhancement 37.5 27.3 No direct compensation 12.5 15.2 Safety net 50.0 66.7 Safety net (training) 37.5 57.6 The difference between programs with and without rehiring iS statistically significant at the 10 percent level. - The difference between programs with and without rehiring is statistically significant at the 5 percent level. Note: Values are based on 41 retrenchment programs in 37 countries. Each value is the percentage of programs with or without rehiring that also has the indicated program characteristic. Source: Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). hiring should be interpreted or treated in the same manner as rehiring. Rehiring the same workers who departed is much more likely to indicate problems with the program, while hiring new workers may involve some intended restructur- ing of the work force. Rehiring the same workers who were separated could be part of a coherent plan of restructuring. For example, it might be difficult politi- cally to target specific workers for retrenchment. Given constraints on the abil- ity to target workers, the optimal strategy might be to induce separations en masse and then to rehire specific workers. We found no evidence for such pur- posive rehiring. The results on age targeting and pension enhancements suggest that for some programs the intent was to replace older, incumbent workers with new workers. Summary Financial Indicators We calculate a summary financial indicator that measures the simple, finan- cial break-even period reflecting the number of years required before a program breaks even on financial costs and benefits. Financial costs are the present dis- counted value of the compensation package or safety net expenses incurred. 38 THE WORLD BANK ECONOMIC REVIEW, VOI 13, NO I Table 7. Relationship between New Hires and the Characteristics of Retrenchment Programs, 1990s Program Percentage of programs Percentage of programs characteristic with new hires without new hires Employment reduction method Involuntary-hard 0W 47.2 Involuntary-soft 80 62.9 Voluntary 80 77 1 All methods 0 20.0 Nature of targeting Skill-biased 40 55.6 Age-biased 1001- 44.4 Neutral 0 22.2 Compensation or assistance Severance payment 60 69.4 Pension enhancement 60 25.0 No direct compensation 0 16.7 Safety net 0.X 72.2 Safety net (training) 0 61.1 The difference between programs with and without new hires is statistically significant at the 5 percent level. Note- Values are based on 41 retrenchment programs in 37 countties. Each value is the percentage of programs with or without new hires that also has the indicated program characteristic Source- Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). This present discounted value is, in principle, calculated over an infinite hori- zon. In practice, for many of the cases, the financial costs are front-loaded with severance payments, making this present discounted value easy to calculate. How- ever, when costs are in the form of a continuing safety net or pension liabilities, we generate appropriate present discounted value measures of the financial costs. (For some programs, the evaluation of the program by World Bank staff had already yielded such calculations.) Financial benefits are measured as the present discounted value of the wage bill savings from the retrenched workers. The an- nual wage bill savings are assumed to be constant over time and equal to the wage bill savings relevant at the time of implementation of the program. These assumptions permit direct calculation of the break-even period as the number of years for the present discounted value of wage bill savings to equal the overall present discounted value of financial costs. To achieve comparability of measures across countries, we use a common discount rate of 10 percent. Under these assumptions, the break-even period is a scale-free financial indicator that permits comparing and contrasting the finan- cial costs and benefits across programs. Information was sufficient to calculate the break-even measure for about 40 percent of the programs. Haltiwanger and Szngh 39 Across the programs surveyed, 22 percent yielded net losses, essentially im- plying an infinite break-even period. The presence of net losses reflects in part the impact of rehires or new hires or of rising compensation for retained work- ers. Accordingly, some programs (for example, the Peruvian tax administration authority, SUNAT) yielded no wage bill savings as a result of rehires, new hires, or an increase in compensation. For the programs for which we can calculate a finite break-even period, the average break-even period is 2.27 years with a median of 1.82 years, a maximum of 10 years, and a minimum of 0 years. Programs with immediate break-even periods (O years) are those with involuntary reduc- tions without direct compensation or other assistance programs. This simple financial indicator demonstrates the financial viability of the re- trenchment programs. Many of the programs yielded relatively rapid financial payoffs. However, as discussed in section II, we would ideally like to quantify the present discounted value of all costs and benefits from retrenching public sector workers and evaluate programs accordingly. Where does this simple fi- nancial indicator fit into such a calculation? To address this question, it is useful to return to inequality 3. For purposes of discussion, suppose we treat the loss function in inequality 3 as a simple linear function such that a dollar of expendi- tures on public sector programs yields a dollar in distortionary losses. Then, inequality 3 can be written as (5) Wpublic - COMP + Pprlvate - Csocial - PpublLc > 0. Thus the measure of the break-even period essentially provides an evaluation of the program based on only the first two terms in inequality 5. A comprehen- sive evaluation, even in these crude terms, requires calculating all of the terms in inequality 5. Unfortunately, data limitations imply that we cannot measure all of the com- ponents in inequality 5, and thus we cannot literally evaluate programs on this basis. However, for some programs we can go a few steps further. In particular, for some programs we can estimate the wages that retrenched workers receive in the private sector. The assumptions that workers are paid their marginal prod- uct in the private sector, ignoring adjustment costs, and that productivity of the redundant (retrenched) workers is zero in the public sector permit a crude imple- mentation of the terms in inequality 5 using this additional information. Where possible, we calculate a modified break-even measure (denoted for labeling pur- poses as the economic payback period) using this additional information. The economic payback period is the number of years required for the benefits from the reduced public sector wage bill to be such that the present discounted value of costs equals the present discounted value of benefits. For this calculation, the costs and benefits for all components other than the public sector wage bill are calculated over an infinite horizon. In addition to measuring benefits based on retrenched workers' earnings in the private sector, for programs with measur- able increases in productivity in the public sector (for example, SUNAT in Peru), we include such effects in the benefits. The average calculated economic pay- 40 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 back period is 2.1 years, and the median is 0.8 year across the programs for which we can generate the measure. (Sufficient information is available to calcu- late the payback period indicator for 20 percent of the programs.) We calculate comparable summary financial and related indicators across programs so that we can relate them to other program characteristics. Table 8 characterizes the relationship between programs with net financial losses and other program characteristics. Programs with a net financial loss are much more likely to include a voluntary component than programs without financial losses. Interestingly, programs with net financial losses are more likely to use targeting, particularly age targeting through pension enhancements. This result indicates that targeting may be expensive, at least measured in simple, financial terms. Table 9 provides a related look at the relationship between the magnitude of the break-even period and other program characteristics. Given that these are at best crudely calculated, we divide programs into two groups-those in the lower and upper tail of the distribution relative to the median break-even period (which is 1.82 years). The results in table 9 reinforce those in table 8. Programs with relatively high break-even periods are more likely to involve a voluntary compo- nent, more likely to use targeting, and more likely to use direct compensation (through either severance payments or pension enhancements). Table 8. Relationship between Net Losses and the Characteristics of Retrenchment Programs, 1990s Program Percentage of programs Percentage of characteristic with net losses other programs Employment reduction method Involuntary-hard 44.4 40.6 Involuntary-soft 50.0 68.8 Voluntary 87.5 75.0 All methods 25.0 15.6 Nature of targeting Skill-biased 66.7 50.0 Age-biased 88.9*£ 40.6 Neutral 0.04 25.0 Compensation or assistance Severance payment 66.7 68.8 Pension enhancement 55.6** 21.9 No direct compensation 22.2 12.5 Safety net 55.6 65.6 Safety net (training) 55.6 53.1 The difference between programs with and without net losses is statistically significant at the 10 percent level " The difference between programs with and without net losses is statistically significant at the 5 percent level Note: Values are based on 41 retrenchment programs in 37 countries. Each value is the percentage of programs with or without net losses that also has the indicated program characteristic. Source: Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). Haltiwanger and Singh 41 Table 9. Relationship between Financial Break-Even Periods and the Characteristics of Retrenchment Programs, 1990s Percentage of programs Percentage of programs Program with below-median with above-median characteristic break-even periods break-even periods Employment reduction method Involuntary-hard 50.0 33.3 Involuntary-soft 66.7 70.6 Voluntary 33.3 > 94.1 All methods 0.0 23.5 Nature of targeting Skill-biased 0.0 66.7 Age-biased 16.7 66.7 Neutral 66.7 5.6 Compensation or assistance Severance payment 66.7 83.3 Pension enhancement 16.7 38.9 No direct compensation 33.3 11.1 Safety net 66 7 50.0 Safety net (training) 50 0 50.0 * * The difference between programs with break-even periods below and above the median value is statistically significant at the 5 percent level Note: Values are based on 41 retrenchment programs in 37 countries. Each value is the percentage of programs with a break-even period below or above the median value that also has the indicated program characteristic. Source: Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). Some of these patterns change substantially when we examine the distribu- tion of the payback period. As seen in table 10, once we incorporate even crude information about productivity gains into the calculation, we no longer find that targeting implies a high payback period. Not surprisingly, we still find that programs with high calculated payback periods are more likely to involve direct compensation and in particular severance pay. To sum up, simple calculations of financial break-even periods indicate that many of the programs had relatively rapid financial payoffs. By contrast, a non- trivial fraction were financial losers. However, caution must be used in inter- preting such financial indicators as measures of the relative success of programs. There are many relevant components of the private and social costs and benefits not captured by these measures. Even the limited attempt made to incorporate relevant economic costs and benefits indicates that inferences based on narrowly defined financial indicators may be quite misleading. Highlights of Individual Programs We summarize the experiences of a select set of individual programs with three objectives in mind. First, consideration of individual programs provides a 42 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Table 10. Relationship between Payback Periods and the Characteristics of Retrenchment Programs, 1990s Percentage of programs Percentage of programs Program with below-median with above-median characteristic payback periods payback periods Employment reduction method Involuntary-hard 60 18.2 Involuntary-soft 80 81.8 Voluntary 20* 100.0 All methods 0 18.2 Natuire of targeting Skill-biased 20 54.5 Age-biased 40 36.4 Neutral 40 27.3 Compensation or assistance Severance payment 60* 100.0 Pension enhancement 40 18.2 No direct compensation 40' 0.0 Safety net 60 45.5 Safety net (training) 60 36.4 - The difference between programs with payback periods below and above the median value is statistically significant at the 5 percent level. Note. Values are based on 41 retrenchment programs in 37 countries Each value is the percentage of programs with a payback penod below or above the median value that also has the indicated program characteristic. Source: Authors' calculations based on individual country summaries (see appendix tables for cross- country tabulations). more comprehensive view of the scope and heterogeneity across programs. Sec- ond, the discussion of individual programs permits illustration of some of the key patterns across programs. Third, our cross-country appendix tables and re- lated analyses admittedly leave out a lot. Part of the reason for this is lack of data suitable for quantifying effects in a summary fashion. For example, quanti- fying the nature of productivity gains associated with retrenchment as well as the ad]ustment costs is difficult both in principle and in practice, given the avail- able data. In general, measuring productivity in the service sector, especially in the public sector, is difficult; most work uses quantitative or technical measures (see Griliches 1992). However, some cases do provide qualitative information about these issues. Here we highlight programs in Peru, Argentina, Uganda, Ghana, India, and Hungary. Key characteristics are presented in table 11. In the discussion, the ordering of the country/program cases reflects a rough attempt to select cases that highlight the role of targeting, productivity gains, the worker safety net, and labor market adjustment. This ordering is only approximate because all of these issues are present for every program. In table 11, entries marked with a Table 11. Key Characteristics in Selected Retrenchment Programs, 1990s Argentina Peru Uganda Federal Civil Civil Characteristic government Railroads Ghana Hungary India service SUNATa service Military Targeting Skill-biased No No Yes Yes No No' Yes* Yes' Yes Age-biased Yes Yes Yes Yes Yes No" Yes* Yes" No Neutral - - - - Yes - - - Reduction method Involuntary-hard No No No Yes No Yes No Yes* Yes Involuntary-soft Yes Yes Yes - No Yes Yes Yes" Yes Voluntary No Yes Yes - Yes Yes Ycs Yes" Yes Rehires No No No No No Yes No No No New hires Yes No No No No No Yes No No Financial indicators Break-even period 0.41 * 1.56# 1.82* Nct loss Net loss 2.6 Net loss' Net loss' 2.7 Payback period - - 1.66" - - - 0.0023- 1.2 Productivity gains Organizational Yes * Yes' No - - - Yes - No (quantitative) (quantitative) (monetary) Worker - Yes" - - - - - Yes* (monetary) (monetary) Safety net provision No No Yes4 Yes* Yes* No No Yes Yes (most workers (most workers reemployed) reemployed) - Not available * A key issue that motivates our inclusion of the program for special discussion. a. The Peru tax administration authority Source: Authors' calculations based on individual country summaries (see appendix tables for cross-country tabulations). 44 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I star are key issues that motivate our inclusion of the program for special discussion. PERU. Two programs were initiated as part of a broader exercise of fiscal austerity and adjustment in Peru. Problems included the underfunded pension system, the posting of ghost employees by the regions to enhance employment- based federal transfers, and the erosion in public salaries, making it difficult to hire and retain qualified workers. With the support of external donors, Peru initiated two labor adjustment pro- grams in 1991, one for the civil service and one for SUNAT. These were completed in 1993 and 1992, respectively (World Bank 1994b). The civil service program used all three involuntary and voluntary employment reduction methods, sepa- rating about 250,000 workers over three years. Induced departures used both lump-sum severance and pension enhancements. We found little evidence of tar- geting by skill. Targeting by age was implicit in the use of pension enhancements to induce voluntary separation. Poor targeting aggravated the shortage of human resources, with many of the most qualified staff leaving. The poor targeting and accompanying shortages probably led to the significant rehiring of separated workers. Further, federal government staff reductions were offset by increases in employment by regional governments, mostly by rehiring erstwhile federal staff. In total, 163,000 of the originally retrenched workers were rehired. Severance packages of about $1,000 were provided to less than half (112,000) of the work- ers separated. This limited the direct financial losses associated with the signifi- cant rehiring. Our simple measure of the break-even period for this program is 2.6 years, more than the median of 1.82 years for all the programs studied. How- ever, this measure does not reflect the loss of productivity associated with shuf- fling the same workers in and out of the same positions. In contrast, the other program in Peru appears to have been a model of good targeting. The SUNAT (tax authority) program also used a mix of voluntary and involuntary reduction methods. Voluntary separations came with an enhanced pension. Involuntary workers were selected on the basis of a written test. Thus targeting was worker-specific and objectively determined. Two-thirds of the work force (2,034 workers) was separated. Subsequently, SUNAT hired 1,309 new work- ers, again based on a written test. Because SUNAT established objective levels of productivity and competence, little basis remained for rehiring (skilled but severance-induced) separated workers. Rehiring was barred for 10 years, and none was found. The average salary for affected workers increased from $50 per month to $1,000 per month. Deflation of wages using the consumer price index instead of the exchange rate produces an increase of similar magnitude; the in- stantaneous increase drops from 20 to 18.6 times. Tax collections more than doubled, and so did SUNAT's revenues (2 percent of tax collections). These were insufficient to cover the salary increase, and the scheme incurred a net financial loss of $47 million in present value terms. However, the entire increase and improvement in tax collections can be interpreted as a substantial gain for the Haltiwanger and Singh 45 government (and, in principle, the economy).3 Incorporating the tax collection gains into our calculation of the payback period yields a payback period of only 0.002 year. This highlights the importance of evaluating program performance along multiple dimensions, evaluating not just organization-level financial costs and benefits but also the broader impact of the program. Neither of these two programs emphasized job search, relocation, training, or other forms of assis- tance to reduce the adjustment costs for the retrenched workers. ARGENTINA. With external assistance from the World Bank, Argentina under- took retrenchment programs for the federal administration and the railroads (as part of a broader program for public enterprises) during 1990-92 and 1991-94, respectively. In a case study on public sector retrenchment in Argentina since 1990, Robbins (1996) examines issues of adverse selection and post-retrenchment worker status using micro data. Given chronic public sector deficits and endemic inflation, extensive public sector reform was considered essential. Wages accounted for 70 percent of federal expenditure, excluding interest payments and transfers. The level of employee redundancy was estimated at 50 percent (World Bank 1995a). For this reason, staff restructuring was an important component of overall reform. The federal administration program retrenched 400,000 workers. No explicit targeting by worker is found. Involuntary separations implied some implicit tar- geting in the selection of redundant workers. Some evidence indicates an age bias. Also, targeting by function was attempted in that the federal government's role was restricted to the provision of security, social services, and economic management. Rehiring was barred by law, and none is found, despite the large scale of retrenchment and the absence of targeting. We find no evidence of rehir- ing in the source reports (World Bank 1994d, 1995a, 1995c); however, some anecdotal evidence indicates the occurrence of rehiring. The railroads program separated about 73,000 workers using both voluntary and involuntary reduction methods. An important feature of Argentinean pro- grams is the heterogeneity in compensation amounts. In federal administration, the average severance payment per worker was $3,000, amounting to a one- time cost of $425 million (World Bank 1995a). In railways, the average sever- ance per worker was $12,000, or a one-shot payment of $360 million (World Bank 1995c). Savings in the annual wage bill was expected to be $1 billion in the federal service and $238 million in the railroads. The calculated break-even pe- riods are 0.4 year and 1.6 years, respectively, in the federal administration and the railroads, with the difference reflecting the higher compensation per worker in railroads. Although difficult to measure, restructuring in both railways and federal ad- ministration apparently was characterized by substantial productivity gains. 3. This discussion neglects the welfare costs (for example, reduction in hours and output) due to increased enforcement of tax collections 46 THE WORLD BANK ECONOMIC REVIENW, VOL 13, NO I Railways experienced an increase in freight miles per worker and passenger miles per worker of about six to seven times. Federal administration experienced a 50 percent cut in processing time. UGANDA. In Uganda, high fiscal deficits combined with inadequate pay levels and wage compression led to a civil service labor restructuring program in 1992- 94. Concomitantly, the military reduced its size in line with peacetime national priorities and conducted a labor restructuring program in 1992-95. Both were partly supported by external financing. The civil service program employed all three reduction methods, restricting the package offer to a fraction of the workers separated. Separations were targeted by skill and tenure. Some critical jobs and personnel were "ring-fenced," that is, protected from retrenchment. Such workers did not have the option of obtaining enhanced benefits associated with voluntary separation. The gross number of separations was approximately 150,000. The average compensation per worker was $320, or about twice the annual gross domestic product (GDP) per capita (Uganda Ministry of Public Service 1994). This total expense of $16 million constituted a measured net financial loss because there was no saving on the wage bill. The wage bill increased due to salary revisions for the remaining civil service staff and was intended to improve service perfor- mance. Interviews with program advisers revealed that the salary hike was in- tended to achieve parity between the civil service average wage and the living wage in Uganda so as to motivate the civil service staff and generate productiv- ity increases. Although an evaluation of the civil service staff performance was not available at the time of our data collection, measurement indicators to evaluate the performance improvements were planned by the Economic Development Institute of the World Bank. The military program was a postwar retrenchment exercise. It used both vol- untary and involuntary reduction methods to reduce strength by 44,000 em- ployees. Skill targeting was based on soldier performance and military require- ments. Average compensation was $955, provided partly in kind and distributed over several months. The calculated break-even period is 2.7 years. In Uganda, some measurement of post-retrenchment earnings for veterans is available. Ac- tual post-separation earnings for the military programs were up to 71 percent of GDP per capita (World Bank 1995b). Including these earnings of a total $5.4 million per year in Uganda's case yields a calculated payback period of 1.2 years. The rarely observed actual data on post-separation earnings are interesting, par- ticularly because the earnings were lower than the average. Although this evi- dence is too idiosyncratic to draw general inferences, it highlights the difficulty in generating the appropriate cost-benefit calculations. Using average earnings in the private sector as a proxy (even when it is available) is obviously inappro- priate in this case. An interesting feature of the military program was the safety net nature of the compensation, which enabled veterans to reintegrate into civilian society. Assis- Haltiwanger and Singh 47 tance in kind was tailored to the intended post-separation occupation; for in- stance, farmers were provided land and farm implements. Furthermore, the as- sistance was known to be temporary (six months), creating an incentive for the veteran to hasten adjustment and thereby limiting program costs. GHANA. Motivated by a very large public sector wage bill as a fraction of total government expenditures, the government of Ghana initiated a program for the civil service in 1987 with partial financing from the World Bank. The program concluded in 1992. It was driven by voluntary departures induced by additional severance and complemented by involuntary-soft measures-removing ghost workers and enforcing retirement age. Voluntary departure, with compensation, was conditional on employment being noncritical to the performance of the unit. The government thus sought to protect necessary skills. More than three-fourths of the total 73,000 workers separated received the package. Average compensation was $700. The formula of two months of base pay for each year of uninterrupted service exceeded the legally required four months of base pay in most cases. Anecdotal evidence indicated some rehiring in subvented organizations (attached to ministries but not covered under the budgetary process). The calculated break- even period is 1.8 years. Despite a generous package offer, the break-even period is relatively low. The government planned extensive training through the Program of Action to Mitigate the Social Costs of Adjustment (PAMSCAD) to assist workers mak- ing a transition to the private sector. However, assistance for job search and placement and for retraining was lacking. PAMSCAD offered courses in entre- preneurial development and provided inputs, including land for potential farm- ers. In rural areas, retrenched workers were also eligible to participate in food- for-work schemes. A sample survey of the retrenched workers finds that about 10 percent of them quit the labor force (Alderman, Canagarajah, and Younger 1996). Of the rest, 97 percent were reemployed by the second year, about 20 percent in formal sector wage jobs and the rest in self-employment or informal sector jobs. INDIA. Fiscal and external payments deficits led India to initiate a stabilization and structural adjustment program in 1991. A program to support public sector labor restructuring was required. This retrenchment support program was directed at public enterprises declared to be sick (with several years of accumulated losses). A voluntary retrenchment program in the sick public sector textile firms separated about 70,000 workers in 1993-94. The average cost per worker was about $17,000. The formula used was 30 days of wages for each year worked, compared with the legally required 15 days of wages for each year of permanent service. The scheme incurred a net loss of $276 million in present value terms, given the exorbitant compensation. The high compensation reflects in part the effects of a rigid law against retrenchment and closure in India (see Basu, Fields, and Debgupta 1996). Given the incentives, the voluntarily departing workers may have had 48 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I long tenures, increasing the severance amount granted. No explicit targeting mechanism by tenure or age is found. The maximum skill level among the retrenched workers was that of a supervisor. Results from a sample survey indicate that all retrenched workers remained in the labor force, and 80 percent were reemployed (PA Consulting Group 1993). Of the total surveyed, 32 percent were in wage jobs, one-fourth of them in the same industry, that is, textiles. The rest (48 percent) were self-employed. HUNGARY. In Hungary, reform and stabilization programs aimed at a transition to a competitive economic system led to mass dismissals in state enterprises during 1990-92. About 1.7 million workers constituting 8.7 percent of the country's labor force were separated. This is extremely large compared with the scope of other programs (0.2 percent of the labor force in Ghana and 0.01 percent in India). The annual wage saving amounted to $298 million but was exceeded by the increase in the annual safety net expenditures of $858 million. The exercise was clearly financially costly but reflected a fundamental change in the structure of the economy. A comprehensive set of institutions designed to act as a safety net for worker reallocation is very common in Western economies, and Hungary was obviously attempting to follow that model. Surveys indicate that retrenched workers remained in the labor force (Com- mander and others 1994). Most of them were reemployed in the private sector, in trade and service industries. The average private manufacturing wage was found to be about 70 percent of the state firm wage (indicating the size of the rent in public sector firms). The nature of the labor market adjustment appeared to change as the transition process proceeded and accelerated. Early on, many of the retrenched workers left voluntarily and left the labor force (for example, through voluntary retirement). Others left voluntarily and transited directly to other jobs. However, as involuntary layoffs increased, a larger fraction of the retrenched workers entered unemployment. Moreover, the available evidence suggests that unemployed workers had an increasingly difficult time finding jobs. Early in the transition, in February 1991, the outflow rate from unemployment implied a steady-state duration of unemployment of 7 months. By November 1992, the implied steady-state duration was 50 months. Aggregate Factors Another means of evaluating the programs is to look at relevant macroeco- nomic indicators prior to and during the program, including indicators of fiscal posture. Examining such macro indicators provides a rough independent check of the factors that led to a retrenchment episode. Further, evaluating macro indicators prior to and during the program provides another means of evaluat- ing the impact of the program. In many cases, the actual retrenchment program is too small by itself to have important macro effects. Exceptions are the pro- grams in Eastern Europe (for example, Hungary) in which there was massive public sector downsizing. However, even in cases where the specific program Haltiwanger and Singh 49 analyzed was too small by itself to have macro effects, the program may have been part of a larger effort to restructure and downsize the public sector. The results that follow can be interpreted in this light. For each of the programs surveyed, we acquired data on the unemployment rate, the real GDP growth rate, the ratio of the current budget deficit to GDP, the ratio of domestic debt to GDP, the ratio of foreign debt to GDP, and the ratio of government spending to GDP. Several sources were used for this purpose, includ- ing the Penn World Tables (Summers and Heston 1991), the cross-country labor market database generated by Rama (1995), World Bank (various issues), and International Monetary Fund (various issues). To remove the influence of idio- syncratic effects across countries, all variables are characterized in terms of de- viations from country-specific means. Country-specific means for each variable are calculated from data for 1980-92. The data are not complete for all vari- ables and all countries over this entire horizon. The unemployment rate series is probably the worst in this regard. Considering deviations from country-specific means mitigates the problems with limited data availability. However, appro- priate caution should be used in considering these results (particularly for unem- ployment), with the additional usual caveats that considerable measurement er- ror is likely in the aggregates available by country. Table 12 presents the patterns of these variables for the five years prior to initiation of the retrenchment pro- gram in each country and during the program. For example, for a specific coun- try, if the retrenchment program began in 1990 and ended in 1992, the prior year calculations reflect data for 1980-89 and the calculations during the pro- gram reflect data averaged over 1990-92. Unemployment rises and GDP growth rates fall in the years leading up to a program (table 12). The unemployment rate appears to fall somewhat during programs, while GDP growth remains below average. These patterns are roughly consistent with the view that retrenchment programs are often precipitated by economic crises. The evidence on fiscal indicators is somewhat mixed at first glance. The government share of GDP rises steadily prior to the onset of a pro- gram and falls during the program. Somewhat surprisingly, there is no accom- panying pattern for the deficit-to-GDP ratio or the debt-to-GDP ratio. For both of these domestic fiscal indicators, there is an improvement in the fiscal posture preceding the program. However, in part this appears to reflect the rising for- eign debt-to-GDP ratio that precedes a program and continues during a program. Putting these pieces together is consistent with the view that these retrenchment programs are often part of a more general austerity program that is supported in part by foreign assistance. Missing Pieces: Measurement and Characterization of Adjustment Costs For evaluation, one of the biggest missing pieces is associated with the con- ceptual and measurement problems involved in quantifying private and social adjustment costs. Although quantification is beyond the scope of readily avail- able information, a few qualitative observations can be made regarding the cross- Table 12. Conditions before and during Retrenchment Programs: Deviations from Country-Specific Means, 1990s Five years Four years Three years Two years One year During Indicator before program before program before program before program before program program Unemployment rate -0.01 -0.26 -0.05 1.09 0.87 -0.18 Real GDP growth rate'" -0.05 0.82 -0.70 -3.25 -0.41 -2.08 Deficrt/GDP* 0.012 -0.005 -0.008 -0.010 -0.012 -0.034 Domestic debt/GDP * -0.036 -0.049 -0.041 -0.036 -0.063 -0.129 Foreign debt/GDP -0.050 -0.058 0.026 0.055 0.098 0.140 Government spendmg/GDP * 0.020 0.021 0.021 0.014 0.001 0.007 * * Indicates that the F-test for the null hypothesis that all of the coefficients in the row are equal is rejected at the 5 percent level. Note: Values are based on 41 retrenchment programs in 37 countries. The deficit numbers are equal to expenditures less revenues and ignore any grants and/ or loans. Source: Authors' calculations based on a variety of sources including the Penn World Tables (Summers and Heston 1991), World Tables (World Bank, various issues), International Finance Statistics (International Monetary Fund, various issues), and Rama (1995). See text for methodology. Haltiwanger and Singh 51 country variation in labor market adjustment costs. Specifically, qualitative evi- dence suggests that it is important to consider the impact of the formal and informal sectors. The discussion that follows on the formal and informal labor market is inadequate on a number of dimensions and does not reflect the rich body of research that has been conducted on the role of formal, informal, and rural sectors for labor market dynamics in developing countries (see Mazumdar 1989 for an overview of this research). Our point here is simply to highlight the potential importance of these considerations for cross-country differences in the nature of labor market adjustment. In countries in which the informal sector plays a large role (for example, Ghana), many of the retrenched public sector workers found employment relatively quickly, albeit rarely in the formal wage sector. By contrast, in the transition economies, retrenched workers from public sector enterprises faced increasing difficulty in leaving unemployment because of difficulties in finding employment in the formal wage sector. Stated very roughly, these observations suggest the following potentially im- portant difference across countries in the relevant labor market adjustment. One set of countries seems to yield relatively quick adjustment, with the informal sector absorbing retrenched public sector workers. Other countries seem to yield long adjustment, with retrenched workers from public sector enterprises experi- encing long spells of unemployment while seeking jobs in the formal private sector. These differences are further reflected in the mix of policies observed across countries. For example, in Africa and to some extent in Latin America, many of the programs surveyed involved direct compensation and relatively few changes in the formal worker safety net. By contrast, in transition economies, the resources for retrenched workers were devoted primarily to the worker safety net. This contrast is probably overstated in several respects. First, employment in the informal sector may involve substantial underemployment. Second, workers reported to be officially unemployed are arguably often engaged in some unre- ported work in the informal sector. Nevertheless, even this rough characteriza- tion highlights the potentially important role of differences across countries in the structure of economies and, in turn, the differences in the nature of labor market adjustment costs. We have learned a great deal from industrial economies (and some transition economies) about labor market flexibility and adjustment by examining the flow of workers and jobs (see, for example, Davis, Haltiwanger, and Schuh 1996). For example, some industrial economies, including the United States and Canada, exhibit high rates of flow into and out of unemployment, while others, including France and Italy, exhibit low rates (see OECD 1993, 1996). The variation in the rates of flow into and out of unemployment translate into striking differences in the importance of long-term unemployment across countries. In the United States the percentage of long-term unemployed (more than 12 months) in the early 1990s was about 6 percent of total workers, while the equivalent figure in Italy was 71 percent. Even this limited and indirect evidence yields two immediate 52 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I inferences for the analysis of public sector retrenchment. First, the low outflow rates from unemployment and the accompanying importance of long-term un- employment in some countries imply that the magnitude of the adjustment costs are potentially very large. Second, the variation across countries is substantial, implying that evaluation of a specific program in a specific country will depend critically on the nature of the labor market adjustment in that country. IV. CONCLUDING REMARKS This survey and analysis of cross-country experiences with retrenchment re- flects the gathering and processing of a wealth of information about individual programs. The nature of the compensation packages, the employment reduction methods used, the nature of targeting, the worker safety net components of the program, the financial costs and benefits, and the nature of adjustment (includ- ing the presence of significant rehiring of the same workers) have been docu- mented and examined in detail. One of the primary objectives and thus contri- butions of this article has been to gather all of this information into one place in an easily accessible manner. Several interesting findings emerge from analysis of the basic facts. Theory suggests the importance of individually tailoring programs to account for worker heterogeneity and to avoid adverse selection problems. Consistent with this view, we found that programs were much less likely to exhibit problems with rehiring if they used targeting on the basis of skills and age, multiple methods of employ- ment reduction, and a combination of compensation packages that included enhancements of the safety net for assisting the reallocation of workers. We also found that programs with this multidimensional approach and with targeting tended to be financially expensive. However, both quantitative and qualitative information suggests that there is a potentially large payoff in productivity gains and in lower adjustment costs. Our analysis of the financial indicators revealed considerable heterogeneity in financial viability across the programs. Many of the programs had a rapid finan- cial payoff in the sense that the wage bill savings from retrenchment quickly covered the financial costs of the programs resulting from worker compensation and assistance. Alternatively, a nontrivial fraction of the programs were clear financial losers. However, we emphasize that although simple financial indica- tors are of obvious interest, they are inappropriate for evaluating the relative success of programs because they omit many of the potentially relevant private and social costs and benefits. Unfortunately, many of the relevant private and social costs and benefits are difficult to quantify, given data (and conceptual) limitations. Especially difficult to quantify are the individual and economywide (social) adjustment costs that are incurred as part of restructuring programs. The data required to assess these adjustment costs are generally not available, and this is accompanied by difficult conceptual issues. This problem has both micro and macro dimensions. On the Haltiwanger and Singh 53 micro side, little effort has been made to collect data systematically on the post- separation experiences of retrenched workers. On the macro side, characteriz- ing and understanding the nature of these adjustment costs requires understand- ing of the myriad factors (for example, institutions) that affect the processes of labor market adjustment within individual countries. The large observed differ- ences across even industrial economies provide prima facie evidence in support of the need for this type of information. In countries with massive public sector retrenchment episodes, the labor market adjustment process is endogenous to the retrenchment program itself. Given the importance of the issues, more re- sources should be devoted to developing the data necessary to evaluate these adjustment costs. Table A-1. Public Sector Retrenchment, by Country and Program, 1990s (number of workers) By instrument Region, Time Employment Involuntary- Involuntary- New Number country, and case period reduction hard soft Voluntary Separations Rehires hires Africa 1 Benin Civil Service 1991-94 10,061 1,040 5,721 3,300 10,061 0 0 2 Burkina Faso Joint Railway Line 1994a 1,585 - - - 1,585 0 0 3 Cameroon Civil Service 1989-94 6,500 2,804 - 3,696 6,500 0 0 4 Cape Verde Public Enterprises 1992-97 247 0 - 247 247 0 0 5 Central African Republic Civil Service 1987-90 1,100 0 2,275 725 3,000 1,900 0 6 Congo CGvil Service 1995a 8,000 - 8,000 0 8,000 0 0 7 Ethiopia Military 1991-95 541,200 - - 0 547,200 6,000 0 8 Ghana Civil Service 1987-92 73,810 0 14,000 59,810 73,810 0 0 9 Kenya Civil Service 1994-97 30,800 - 25,000 5,800 30,800 0 0 10 Malawi Civil Service 1994a - - - - - - 11 Mauritania Public Enterprises 1990-94 1,900 - - - 1,900 0 0 12 Namibia Military 1991-94 49,500 57,000 0 0 57,000 7,500 0 13 Senegal Civil Service 1989-91 4,357 0 1,800 4,300 6,100 0 1,743 14 Sierra Leone Civil Service 1992-97 27,452 20,852 1,100 6,000 27,952 500 0 15 Uganda Civil Service 1992-94 91,339 11,339 75,000 5,000 91,339 0 0 16 Uganda Military 1992-95 44,211 11,000 8,000 9,000 44,211 0 0 Asia 17 Bangladesh Jute Sector Public Enterprises 1994-95 22,250 0 4,000 21,250 25,250 3,000 0 18 Cambodia Civil Service 1995/961 50,000 - - - 50,000 0 0 19 China Shenyang Region Reform 1994-2002 7,000 - 7,000 0 7,000 0 0 20 India Public Enterprises 1993-94 69,466 0 0 69,466 69,466 0 0 21 Lao PDR CivIl Service 1989-93 21,600 - - 21,600 0 0 22 Pakistan Public Enterprises 1991-93 7,495 0 0 7,495 7,495 0 0 23 Pakistan Sindh Region Reform 1993-97 6,500 0 - 6,500 6,500 0 0 24 Sri Lanka Civil Service 1991-92 49,000 0 0 68,000 68,000 0 19,000 Europe 25 Albania Public Enterprises 1992 253,000 253,000 0 0 253,000 0 0 26 Hungary Public Enterprises 1990-92 1,661,000 - - - 1,661,000 0 0 27 Kazakhstan Public Enterprises 1993 172,959 40,959 132,000 - 172,959 0 0 28 Macedonia Public Enterprises 1988-93 112,000 42,000 0 70,000 112,000 0 0 29 Poland Public Enterprises 1990-93 547,300 547,300 0 0 547,300 0 0 30 Russian Federation Coal Sector 1996b 72,300 - - - 72,300 0 0 31 Turkey Public Enterprises 1993-94 35,000 0 - 35,000 35,000 0 0 Latin America 32 Argentina Public Enterprises 1991-94 72,818 0 42,818 30,000 72,818 0 0 33 Argentina Federal Administration 1990-92 405,995 0 424,095 0 424,095 0 18,100 34 Bolivia Mining-Public Corporation 1991-94 4,251 0 0 4,599 4,599 348 0 35 Brazil Civil Service 1990-91 0 100,000 0 0 100,000 100,000 0 36 Chile Civil Service and Parastatal Organizations 1973-77 91,100 91,100 0 0 91,100 0 0 37 Colombia Tourism and Transport Ministry 1990-92 12,000 0 0 12,000 12,000 0 0 38 Ecuador Civil Service 1992-94 40,000 0 0 40,000 40,000 0 0 39 MexIco SOCEFI-Mmistry of Trade and Industry 1989-92 4,150 0 2,000 3,000 5,000 0 850 40 Peru Civil Service 1991-93 100,595 0 - 112,000 263,654 163,059 0 41 Peru SUNAT-Tax Collecting Authority 1991-92 725 0 - - 2,034 0 1,309 - Not available. Note: In several cases, the number of workers separated by instrument does not add up to total separations, owing to partial availability of information. a. Ongoing. b. Proposed. Source: Authors' calculations. s6 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Table A-2. The Financial Costs of Public Sector Retrenchment, by Country and Program, 1990s Financial costs (millions of dollars) Cost per Region, Total Severance Enhanced Safety worker Number country, and case cost payments pension net (dollars)a Africa 1 Benin Civil Service 21 21 0 0 6,424 2 Burkina Faso Joint Railway Line - - - - - 3 Cameroon Civil Service 7 7 0 - 1,997 4 Cape Verde Public Enterprises 3 2 0 0 10,260 5 Central African Republic Civil Service - - - - - 6 Congo Civil Service - - - - 7 Ethiopia Military 200 0 0 200 365b 8 Ghana Civil Service 42 42 0 - 700 9 Kenya Civil Service 20 0 20 - 3,448 10 Malawi Civil Service 20 20 0 0 - 11 Mauritania Public Enterprises 9 9 0 - 4,910c 12 Namibia Military 38 13 - 25 658 13 Senegal Civil Service 80 80 - 0 13,166c 14 Sierra Leone Civil Service 2 2 0 0 353 15 Uganda Civil Service 16 16 - - 320d 16 Uganda Military 42 42 0 - 955c Asia 17 Bangladesh Jute Sector Public Enterprises 56 56 0 - 2,621 18 Cambodia Civil Service 50 50 0 - 1,000' 19 China Shenyang Region Reform 0 0 0 - 0 20 India Public Enterprises 1,188 1,140 - 48 17,108 21 Lao PDR CIVII Service 10 10 - - 470' 22 Pakistan Public Enterprises 25 25 0 - 3,318 23 Pakistan Sindh Region Reform - - - - 24 Sri Lanka Civil Service 71 0 71 0 1,040 Europe 25 Albania Public Enterprises 6 0 0 6 241 26 Hungary Public Enterprises 859 0 0 859 517b 27 Kazakhstan Public Enterprises - - - - 28 Macedonia Public Enterprises 50 0 50 - 714 29 Poland Public Enterprises 7,669 0 5,538 2,130 14,012b 30 Russian Federation Coal Sector - - - - 31 Turkey Public Enterprises Haltiwanger and Singh 57 Table A-2 (continued) Financial costs (milhlons of dollars) Cost per Region, Total Severance Enhanced Safety worker Number country, and case cost payments pension net (dollars)a Latin America 32 Argentina Public Enterprises 360 360 0 0 12,000 33 Argentina Federal Administration 425 425 0 0 1,002' 34 Bolivia Mining-Public Corporation 74 74 0 0 16,000 35 Brazil Civil Service - - - - - 36 Chile Civil Service and Parastatal Organizations - 0 - - 0 37 Colombia Tourism and Transport Ministry - - - -- 38 Ecuador Civil Service 200 200 0 - S,000 39 Mexico SOCEFI-Ministry of Trade and Industry - - - - 40 Peru Civil Service 530 112 418 0 4,735 41 Peru SUNAT-Tax Collecting Authority 2 1 1 0 1,131' - Not available. a. Amount of severance per (voluntarily retrenched) worker. b. Cost per worker; workers separated involuntarily. c. Average severance amount per worker (for all retrenched workers) d. Average severance using all workers (excluding the 42,000 ghost workers). Source; Authors' calculations. Table A-3. The Financial Benefits of Public Sector Retrenchment, by Country and Program, 1990s (millions of dollars) Region, Wage bill Wages Number country, and case Total savings (annual) Separations Rebires New bires Other' Africa 1 Benin Civil Service 0 0 2 Burkina Faso Joint Railway Line - - 3 Cameroon Civil Service - - - - - - 4 Cape Verde Public Enterprises - - - - - - 5 Central African Republic Civil Service 8 8 6 Congo Civil Service - - - 00 7 Ethiopia Military 542 - - - - 542 8 Ghana Civil Service 24 24 24 - - - 9 Kenya Civil Service 6 6 6 - - 10 Malawi Civil Service 10 10 10 - - 11 Mauritania Public Enterprises - - - - 12 Namibia Military 34 - 13 Senegal Civil Service 30 30 14 Sierra Leone Civil Service 1 1 1 15 Uganda Civil Service 0 0 - - - - 16 Uganda Military 17 17 17 0 0 Asia 17 Bangladesh Jute Sector Public Enterprises 18 18 - - 0 - 18 Cambodia Civil Service -126 -126 18 - 144 - 19 China Shenyang Region Reform 3 3 3 0 0 - 20 India Public Enterprises 83 83 83 0 0 - 21 Lao PDR Civil Service 9 9 - - - 22 Pakistan Public Enterprises 350 18 18 - _ 332 23 Pakistan Sindh Region Reform - - - - 24 Sri Lanka Civil Service -157 -157 Europe 25 Albania Public Enterprises - - 26 Hungary Public Enterprises 298 298 - - - - 27 Kazakhstan Public Enterprises - - 28 Macedonia Public Enterprises - - - - 29 Poland Public Enterprises 1,548 1,148 1,148 - - 400 30 Russian Federation Coal Sector - - - - 31 Turkey Public Enterprises - - - - Latin America 32 Argentina Public Enterprises 237 237 237 - - 0 33 Argentina Federal Administration 1,063 1,000 - - - 63 34 Bolivia Mining-Public Corporation - - - - - 0 35 Brazil Civil Service 0 0 - - - 0 36 Chile Civil Service and Parastatal Organizations - - - - - - 37 Colombia Tourism and Transport Ministry - - - - - 38 Ecuador Civil Service - - - - 39 Mexico SOCEFI-Ministry of Trade and Industry - - - - - 40 Peru Civil Service 222 222 - - - - 41 Peru SUNAT-Tax Collecting Authority -15 -15 1 0 16 - - Not available. a. Includes program-specific gains such as a reduction in arms imports in Ethiopia and release of real estate in Pakistan. For details, see individual country summaries (which are available on request from the authors). Source: Authors' calculations. Table A-4. Performance Indicators for Public Sector Retrenchment Programs, by Country and Program, 1990s Region, country, Break-even Net financial Payback period Net economic Number and case period (years) gain or loss (years) gain or loss Africa 1 Benin Civil Service - -21 2 Burkina Faso Joint Railway Line - - - - 3 Cameroon Civil Service - - - - 4 Cape Verde Public Enterprises - - 5 Central African Republic Civil Service - - 6 Congo Civil Service 7 Ethiopia Military 0 40 342a 0.30 365a 8 Ghana Civil Service 1.82 1.66 9 Kenya Civil Service 3.60 - 3 02 10 Malawi Civil Service 2.00 - 11 Mauritania Public Enterprises - - 12 Namibia Military 1.10 48b 0.40 115b 13 Senegal Civil Service - -409 14 Sierra Leone Civil Service 1.87 - 15 Uganda Civil Service - -16 16 Uganda Military 2.70 - 1.2(0 Asia 17 Bangladesh Jute Sector Public Enterprises 3.40 - 18 Cambodia Civil Service - -1436 19 China Shenyang Region Reform - 29 0.00 93 20 India Public Enterprises - -276 21 Lao PDR Civil Service 1.10 - 22 Pakistan Public Enterprises 1 44 501' 23 Pakistan Sindh Region Reform 0.00 24 Sri Lanka Civil Service - -1,802 Europe 25 Albania Public Enterprises - - 26 Hungary Public Enterprises - -561 27 Kazakhstan Public Enterprises - - 28 Macedonia Public Enterprises 29 Poland Public Enterprises - -6,121 30 Russian Federation Coal Sector 31 Turkey Public Enterprises Latin America 32 Argentina Public Enterprises 1.56 - 33 Argentina Federal Administration 0.41 - 34 Bolivia Mining -Public Corporation 10.00d - 10 O,1 35 Brazil Civil Servicc 36 Chile Civil Service and Parastatal Organizations - - 37 Colombia Tourism and Transport Ministry 38 Ecuador Civil Service 39 MexIco SOCEFI -Ministry of Trade and Industry - 40 Peru Civil Service 2.60 - - 41 Peru SUNAT -Tax Collecting Authority - -47 0 002 - Not available. a. Including savings in arms imports. b For three-year program duration, defense expenditure savings reversed thereafter. c Including privatization proceeds. d. Not calculated (World Bank 1994a). Source: Authors' calculations Table A-5. Economic Costs and Benefits of Public Sector Retrenchment Programs, by Country and Program, 1990s (millions of dollars) Economic costs Economic benefits Region, country, Nonfinanczal Nonfinancial (production rise) Number and case Total (production lost) Total Organization Worker Africa 1 Benin Civil Service 21 - - - 2 Burkina Faso Joint Railway Line 3 Cameroon Civil Service 7 - 4 Cape Verde Public Enterprises 3 - - 5 Central African Republic Civil Service - - 8 6 Congo Civil Service - - - 7 Ethiopia Military 200 0 565 - 23 8 Ghana Civil Service 42 - 26 2 9 Kenya Civil Service 20 - 7 - 1 10 Malawi Civil Service 20 - 10 - 11 Mauritania Public Enterprises 9 - 0 - 12 Namibia Military 38 - 101 0 67 13 Senegal Civil Service 80 - 30 - 14 Sierra Leone Civil Service 2 - 1 - 15 Uganda Civil Service 16 - 0 - 16 Uganda Military 42 - 23 - 5 Asia 17 Bangladesh Jute Sector Public Enterprises 56 - 18 - 18 Cambodia Civil Service 50 - -126 - 19 China Shenyang Region Reform 0 0 15 12 20 India Public Enterprises 1,188 - 83 - 21 Lao PDR Civil Service 10 - 9 22 Pakistan Public Enterprises 25 - 18 23 Pakistan Sindh Region Reform - - - 24 Sr I anka Civil Service 71 - -157 Europe 25 Albania Public Enterprises 6 - 0 26 Hungary Public Enterprises 859 - 298 27 Kazakhstan Public Enterprises - - - 28 Macedonia Public Enterprises 50 - - 29 Poland Public Enterprises 7,669 - 1,148 30 Russian Federation Coal Sector - - - 31 Turkey Public Enterprises - - - Latin America 32 Argentina Public Enterprises 360 - 237 Quantitative rises 33 Argentina Federal Administration 425 - 1,000 Quantitative rises 34 Bolivia Mining-Public Corporation 74 - - - 35 Brazil Civil Service 0 - 0 0 0 36 Chile Civil Service and Parastatal Organizations 37 Colombia Tourism and Transport Ministry 38 Ecuador Civil Service 200 - 39 Mexico SOCEFI-Ministry of Trade and Industry 40 Peru Civil Scrvice 530 - 222 - 41 Peru SUNAT-Tax Collecting Authority 2 0 933 947 - Not available. Source: Authors' calculations. 64 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Included here are some references from the approximately 200 pages of individual country case summaries that may not be cited in the text. Alderman, Harold, Roy S. Canagarajah, and Stephen Younger. 1996. "A Comparison of Ghanaian Civil Servants' Earnings before and after Retrenchment." Journal of African Economies 4(2):259-88. Basu, Kaushik, Gary Fields, and Shub Debgupta. 1996. "Retrenchment, Labor Laws, and Government Policy: An Analysis with Special Reference to India." Paper pre- sented at the conference on Public Sector Retrenchment and Efficient Compensation Schemes, World Bank, Washington, D.C. Processed. Bertola, Giuseppe. 1990. "Job Security, Employment, and Wages." European Economic Review 34(4, June):381-402. Bertola, Giuseppe, and Richard Rogerson. 1997. "Institutions and Labor Reallocation." European Economic Review 41(6, June) :1147-71. Blanchard, Olivier, and Peter Diamond. 1989. "The Beveridge Curve." Brookings Pa- pers on Economic Activity 1:1-60. .1990. "The Cyclical Behavior of Gross Flows of Workers in the United States." Brookings Papers on Economic Activity 2:85-155. Blanchard, Olivier, Simon Commander, and Fabrizio Coricelli. 1994. "Unemployment and Restructuring in Eastern Europe." In Simon Commander and Fabrizio Coricelli, eds., Unemployment, Restructuring, and the Labor Market in Eastern Europe and Russia. EDI Development Studies. Washington, D.C.: World Bank. Commander, Simon, and Farbrizio Coricelli, eds. 1994. Unemployment, Restructuring, and the Labor Market in Eastern Europe and Russia. EDI Development Studies. Wash- ington, D.C.: World Bank. Commander, Simon, Janos Kollo, Cecilia Ugaz, and Balacs Vilagi. 1994. "Hungary." In Simon Commander and Fabrizio Coricelli, eds., Unemployment, Restructuring, and the Labor Market in Eastern Europe and Russia. EDI Development Studies. Washing- ton, D.C.: World Bank. Cooper, Russell, and John Haltiwanger. 1996. "Evidence on Macroeconomic Comple- mentarities." Review of Economics and Statistics 78(1, February):78-93. Davis, Steven J., John Haltiwanger, and Scott Schuh. 1996. Job Creation and Destruc- tion. Cambridge, Mass.: MIT Press. Diwan, Ishac. 1993a. "Public Sector Retrenchment and Efficient Severance Pay Schemes." World Bank, Policy Research Department, Washington, D.C. Processed. . 1993b. "Public Sector Retrenchment and Severance Pay: Nine Propositions." World Bank, Policy Research Department, Washington, D.C. Processed. Griliches, Zvi, ed. 1992. Output Measurement in the Service Sectors. NBER Studies in Income and Wealth 56. Chicago: University of Chicago Press. Hall, Robert E. 1995. "Lost Jobs." Brookings Papers on Economic Activity 1:221- 56. International Monetary Fund. Various issues. International Finance Statistics. Washing- ton, D.C. Haltiwanger and Singh 65 Lazear, Edward. 1990. "Job Security Provisions and Employment." Quarterly Journal of Economics 105(3, August):699-726. Levy, Anat, and Richard McLean. 1996. "Optimal and Suboptimal Retrenchment Schemes: An Analytical Framework." Paper presented at the conference on Public Sector Retrenchment and Efficient Compensation Schemes, World Bank, Washing- ton, D.C. Processed. Lindauer, David, and Barbara Nunberg, eds. 1994. Rehabilitating Government: Pay and Employment Reform in Africa. A World Bank Regional and Sectoral Study. Washington, D.C.: World Bank. Mazumdar, Dipak. 1989. Microeconomic Issues of Labor Markets in Developing Coun- tries: Analysis and Policy Implications. EDI Seminar Paper 40. Washington, D.C.: World Bank. Mills, Bradford, David E. Sahn, Edward E. Walden, and Stephen D. Younger. 1993. "Public Finance and Employment: An Analysis of Public Sector Retrenchment Pro- grams in Ghana and Guinea." World Bank, Policy Research Department, Washing- ton, D.C. Processed. OECD (Organisation for Economic Co-operation and Development). 1993. Employment Outlook (July). Paris. . 1996. Job Creation and Loss: Analysis, Policy and Data Development. Paris. PA Consulting Group. 1993. "Social Safety Net Adjustment Credit Technical Assis- tance-Final Report." Report submitted to the Government of India's Industrial De- velopment Ministry and the World Bank. Auckland, New Zealand. Processed. Rama, Martin. 1994. "Flexibility in Sri Lanka's Labor Market." Policy Research Work- ing Paper 1262. World Bank, Policy Research Department, Washington, D.C. Pro- cessed. .1995. "Labor Market Cross-Country Database: Glossary." World Bank, Policy Research Department, Washington, D.C. Processed. . 1997. "Efficient Public Sector Downsizing." World Bank, Policy Research De- partment, Washington, D.C. Processed. Robbins, Donald. 1996. "Public Sector Retrenchment: A Case Study of Argentina." Paper presented at the conference on Public Sector Retrenchment and Efficient Com- pensation Schemes, World Bank, Washington, D.C. Processed. Summers, Robert, and Alan Heston. 1991. "The Penn World Table (Mark 5): An Ex- panded Set of International Comparisons, 1950-198 8." Quarterly Journal of Eco- nomics 106(2, May):327-68. Svejnar, Jan, and Katherine Terrell. 1991. "Reducing Labor Redundancy in State-Owned Enterprises." Policy Research Working Paper 792. World Bank, Policy Research De- partment, Washington, D.C. Processed. Uganda Ministry of Public Service. 1994. Management of Change, Context, Vision, Objectives, Strategy, and Plan: Civil Service Reform. Kampala. World Bank. 1994a. "Bolivia: Structural Reforms, Fiscal Impact, and Economic Growth." Report 13067-B). Washington, D.C. Processed. .1994b. "Peru Public Expenditure Review." Report 13190-PE. Washington, D.C. Processed. .1994c. "Shenyang Industrial Reform Project Appraisal Report." Report 12415- CHA. Washington, D.C. Processed. 66 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I .1994d. World Development Report 1994: Infrastructure for Development. New York: Oxford University Press. . 1995a. "Argentina Public Sector Reform Project Completion Report." Report 14781. Washington, D.C. Processed. . 1995b. "From Swords to Ploughshares: The Uganda Demobilization and Rein- tegration of Ex-combatants in Uganda." Washington, D.C. Processed. . 1995c. "Public Enterprise Adjustment Loan Audit Report." Report 14809. Washington, D.C. Processed. . Various issues. World Tables. Baltimore, Md.: The Johns Hopkins University Press. THE WORLD BANK ECONOMIC REVIEW, VOL 13. NO 1 67-88 The Efficient Mechanism for Downsizing the Public Sector Doh-Shin Jeon and Jean-Jacques Laffont This article analyzes the efficient mechanism for downsizing the public sector, focus- ing on adverse selection in productive efficiency. Each worker is assumed to have two type-dependent reservation utilities: the status quo utility in the public sector before downsizing and the utility that the worker expects to obtain by entering the private sector. The efficient mechanism consists of a menu of probability (of remaining in the public sector) and transfer pairs that induces self-selection. A worker's full cost is defined by the sum of production cost in the public sector and reservation utility in the private sector. It is optimal to start by laying off the agents with higher full cost. When the public sector before downsizing is discriminating as the differential of private in- formation about productive efficiency suggests, there are countervailing incentives. This makes the size of downsizing smaller under asymmetric information than under complete information. Inefficient public sectors in developing countries exhibit considerable labor re- dundancy. Hence, downsizing constitutes a natural step for every public sector reform in developing countries. However, downsizing is subject to adverse selec- tion problems (Diwan 1994 and Rama 1997). For example, consider a simple downsizing mechanism that gives generous severance pay to every worker who leaves the public sector. Suppose that efficient workers have better job opportu- nities in the private sector than inefficient workers. Then, it may happen that only the efficient workers leave the public sector. If the consequence of this brain drain is so serious that it disrupts the public sector, downsizing may result in an increase rather than a decrease of inefficiency in the public sector. In this article, we consider adverse selection in workers' productive efficiency. The type of worker, which represents the worker's production cost in the public sector, is the worker's private information. Alternatively, the type can be inter- preted as disutility of effort. We assume that there are two types of workers, those with high and those with low production cost in the public sector. We call the type with low production cost efficient and the type with high production Doh-Shin Jeon is with the Groupe de Recherche en Economie Math6matnque et Quantitative at the Unlversit6 de Toulouse I in France, and Jean-Jacques Laffont is with the Institut d'Economte IndustrLelle and the Atelier de Recherche Quantitative Appliquee au Developpement Economique at the Universit6 de Toulouse 1. The authors thank Martin Rama and an anonymous referee for helpful comments and suggestions. (C 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 67 68 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I cost inefficient. We analyze mechanisms for voluntary downsizing when each worker has two type-dependent reservation utilities. The first represents the worker's status quo utility level in the public sector before downsizing; the sec- ond represents the worker's expected utility on entering the private sector. Thus our work is closely related to the literature on mechanism design under type- dependent reservation utility (see Lewis and Sappington 1989, Maggi and Rodriguez-Clare 1995, and Jullien 1997). Because we focus mainly on downsizing public sectors in developing coun- tries, we assume that monitoring in the public sector is so inefficient that the quantity produced by a worker cannot be a controllable instrument. We do not envision a reform of the incentive system in the public sector. Indeed, it is often the political unfeasibility of such a reform that leads to downsizing. Hence the principal, the benevolent regulator of the public sector, or the government has two instruments: the probability that a worker will remain in the public sector and the monetary transfer to the worker. In our model, the stochastic element of the mechanism is the key feature that allows the government to induce self- selection. After analyzing the efficient downsizing mechanism, we show how the government can implement it through a menu of probability, wage, and sever- ance pay triplets without causing any worker to regret having participated in the downsizing procedure. Kahn (1985) analyzes optimal severance pay when there is asymmetric infor- mation about a worker's outside productivity. His framework is different from ours, however, in that he assumes complete information about on-the-job pro- ductivity. Lazear (1995) studies efficient severance pay and efficient layoff rule in a context in which concerns about firm-specific human capital make firms adopt upward-sloping age-earnings profiles. But he does not consider adverse selection. Our work is more closely related to studies about downsizing under adverse selection. Diwan (1994) studies adverse selection in workers' productivity. He does not consider self-selection mechanisms and thus proposes randomization as the optimal response. By contrast, we show that self-selection can be achieved, and that, as a consequence, randomization is, in general, a suboptimal solution. Levy and McLean (1996) consider adverse selection in workers' disutility of effort. They assume that workers' productivity in the public sector and their alternative wages are type-dependent. However, they restrict the government's instrument to one severance pay that is a scalar multiple of the status quo wage. In our work, we do not restrict the government's instruments: it can use a menu of probability and transfer pairs. Furthermore, we show that the government can implement an efficient downsizing mechanism that will not cause regret on the part of the workers. Rama (1997) studies adverse selection in workers' aversion to effort. He con- siders a self-selection mechanism that combines a fixed-term contract with sev- erance pay such that hard-working employees want to switch to the fixed-term contract, whereas lazy ones prefer severance pay. But he does not analyze the Jeon and Laffont 69 efficient downsizing mechanism. Moreover, as we discuss in section VII, fixed- term contracts can be regarded as an alternative to stochastic mechanisms if concerns about favoritism on the part of public officials make a transparent implementation of the latter difficult. Section I gives a self-contained summary of the main results derived in this article. Section II presents our model of an efficient mechanism for downsizing the public sector. Section III analyzes the benchmark case of the efficient downsizing mechanism when there is complete information about workers' pro- ductive efficiency. Section IV analyzes the efficient downsizing mechanism when there is asymmetric information about workers' productive efficiency. Section V shows how the efficient mechanism derived in section IV can be implemented without regret through a menu of probability, wage, and severance pay triplets. Section VI relaxes some of the main assumptions and describes extensions of the analysis. Section VII discusses the relevance of key assumptions and suggests directions for further research. I. MAIN RESULTS The efficient downsizing mechanism consists of a menu of contracts that in- duces self-selection. Each contract, composed of a probability of staying in the public sector and a monetary transfer, is conceived such that all the efficient workers (respectively, all the inefficient workers) choose the contract designed for the efficient type (respectively, the inefficient type). Self-selection can be achieved because each worker attaches a different value to the possibility of keeping a job in the public sector according to the worker's productive effi- ciency in the public sector and job opportunities in the private sector. Whether the regulator should start to lay off the efficient workers or the inef- ficient workers depends on each type's full cost of staying in the public sector. Each type's full cost is equal to the production cost in the public sector plus the opportunity cost. The opportunity cost is given by the utility that the worker expects to obtain by leaving the public sector and trying to find a job in the private sector. For example, if the probability of finding employment in the pri- vate sector is low and the cost of searching for a job is high, the opportunity cost will be small. It is always optimal to start laying off the workers with high full cost. This implies that there may exist cases in which it is optimal to start laying off the efficient workers. To determine which type has high full cost, we need information about the differential in terms of productive efficiency between the efficient type and the inefficient type, in the public sector and in the private sector. We also need infor- mation about the nature of the incentive schemes that map efficiency levels into informational rents in both sectors. For example, if workers' efficiency is gen- eral (that is, not specific to a sector) and if the private sector's reward structure is more sensitive to workers' efficiency, then it is optimal to start laying off the efficient workers. 70 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 The extent of downsizing is determined by comparing the social marginal gain with the social marginal cost of keeping a worker in the public sector. When the social value of public production is low, it is optimal for the regulator to lay off all the workers with high full cost and to keep only a proportion of the workers with low full cost. When the social value is high, it is optimal to keep all the workers with low full cost and to lay off a proportion of the workers with high full cost. Under complete information, the social marginal cost of keeping a worker is equal to the private marginal cost (given by the full cost defined above) multi- plied by 1 + X, where X(> 0) represents the shadow cost of public funds. Thus the higher is the negative impact of distortionary taxation, the larger the size of downsizing should be. Under asymmetric information, workers may obtain an informational rent. This makes downsizing more costly than under complete information and af- fects the social marginal cost of keeping a worker in the public sector. In this article, we consider the case in which the public sector before downsizing is discriminating as the differential of information suggests: the differential in terms of status quo utility between the two types is equal to the differential in terms of productive efficiency in the public sector. In this case, there are always countervailing incentives. In particular, when the social value of public produc- tion is low, the efficient downsizing mechanism requires that all the workers with high full cost leave the public sector, obtaining a positive informational rent, while the workers with low full cost stay with a positive probability and do not obtain any rent. An important consequence of countervailing incentives is that the social marginal cost of laying off a worker with low full cost is higher under asymmetric information than under complete information. Hence, asym- metric information reduces the size of downsizing when the social value of pub- lic production is low. The efficient mechanism can be implemented through a menu of probability, wage, and severance pay triplets without causing regret on the part of the work- ers. The government proposes a menu composed of two triplets, and each worker chooses one triplet after having accepted the offer. By choosing a triplet, a worker is committed to respect the random outcome associated with the probability specified in the triplet. If the outcome is to stay in the public sector (respectively, to leave the sector), the worker should stay (respectively, leave), receiving the wage (respectively, the severance pay) specified in the triplet. The menu can be designed in such a way that each worker will choose the triplet designed for that worker's type and that none of the workers will regret having participated in the downsizing procedure regardless of whether they stay or get laid off. II. THE MODEL We consider an economy composed of two sectors: the public sector and the private sector which represents the rest of the economy. The government (the Jeon and Laffont 71 benevolent regulator of the public sector) wants to downsize the public sector. There is a continuum of workers of mass 1 employed in the public sector before downsizing. We denote the set of workers in the public sector by I. A worker in the public sector who gets laid off after downsizing may try to find a job in the private sector. We consider voluntary downsizing in the sense that downsizing should be done through a process that induces voluntary participation of the workers in the public sector. In other words, workers have the right to stay in the public sector with their current status and cannot be laid off against their will. This situation corresponds to reality because most countries forbid mandatory lay- offs in the public sector by law. We assume that monitoring in the public sector is so inefficient that the quan- tity produced by a worker cannot be a controllable instrument. In particular, we assume that the quantity produced by each worker is normalized to 1 both be- fore and after downsizing. In other words, the government does not revise the incentive schemes within the public sector simultaneously with downsizing. Thus the efficient downsizing mechanism that we analyze is conditional on not chang- ing incentive schemes in the public sector. The production level of the public sector after downsizing will equal the mass of workers remaining in the sector. To focus on adverse selection in productive efficiency, we assume that all the workers are homogeneous except in terms of their production cost. In reality, discrimination based on some observable and verifiable characteristics is pos- sible. The efficient downsizing mechanism will be discriminatory if there exists some correlation between the characteristics and productive efficiency. The model denotes worker A,'s type, with i e I, by 0,, which represents the worker's mar- ginal cost of production in the public sector. 0, is independently and identically distributed. It takes the value 0 with probability v and the value 6 with probabil- ity 1 - v. We denote the difference between the two values by AO _= 0 - 0 > 0. 0, is private information known only by A,. The distribution of 0, is common knowl- edge. We call type O the efficient type and type 6 the inefficient type. A, has two type-dependent reservation utilities. The first is the utility that A, obtains by staying in the public sector, rejecting the government's offer of the downsizing mechanism. This reservation utility is expected to be high because, in general, public sector wages are large in view of the workers' productivity and because nonwage benefits are generous in public sectors. We denote it by UP(0,), where the superscript p indicates the public sector. In particular, UV denotes the efficient type's reservation utility and Uf denotes the inefficient type's reserva- tion utility. The second reservation utility is the one that A, expects to obtain by leaving the public sector and entering the private sector. We denote it by U'(0,), where the superscript m indicates market. When laid off by the public sector, a worker may either find a job in the private sector (probably after incurring some search cost) or remain unemployed. U'(O,) is an expected type-dependent utility that takes into account various factors reflecting labor market conditions. In particu- 72 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 lar, Um denotes the efficient type's reservation utility and Um denotes the ineffi- cient type's reservation utility. Without loss of generality, U't is normalized to 0. A, has two choices: stay in the public sector, rejecting the government's offer, or accept the offer. Hence, A,'s utility, denoted by U,, is given as follows: U = UP (0,) if staying in the public sector and rejecting the government's offer lU(O,) if accepting the offer. The government maximizes social welfare, denoted by W, defined as follows: W S(q)-(1+) Et, + U, where S(Q) represents the social surplus generated by public production, which is a function of the total quantity produced by the public sector, denoted by q; X(> 0) represents the shadow cost of public funds; and t, is the monetary transfer from the government to worker A,. S(-) is increasing and strictly concave with S'(0) = °°. X is strictly positive because distortionary taxation inflicts a cost of (1 + X) dollars on taxpayers in order to levy 1 dollar for the government. According to the revelation principle, we can restrict our attention to the set of direct revelation mechanisms (see, for example, Gibbard 1973, Green and Laffont 1977, and Myerson 1979). A downsizing mechanism is then defined by: {p, (6), t,()} with H, e {H, H} where p,( ) represents the probability of keeping A, in the public sector, t,(-) is the monetary transfer from the government to Al, and 8, is A,'s report to the govern- ment about A,'s type. We note that because there is no aggregate uncertainty about type, the probability and the transfer for A, depend only on A,'s report. The government offers to each worker the opportunity to participate in a random exercise that entails a payment and a probability of remaining in the public sector. The worker may or may not participate. A worker who does not participate keeps the reservation utility UP( ). A worker who does participate makes a commitment to respect the outcome of the random exercise. In particu- lar, if the outcome tells the worker to leave the public sector, the worker must try to find a job in the private sector, expecting the reservation utility U'( ). For expositional simplicity, we introduce the following notations: P, (0) =P, P, (0)=p t, (0) =t t, (0) = t. Jeon and Laffont 73 Given a total quantity produced by the public sector q, we denote the social marginal cost of keeping a worker in the sector under complete information by MCc(q). We introduce the following assumptions. ASSUMPTION 1: S'(1) < MCc(1) ASSUMPTION 2: UP - UP = AO. Assumption 1 says that, under complete information, if the government keeps all the workers in the public sector, the social marginal surplus of keeping a worker is smaller than the social marginal cost of keeping a worker. Thus as- sumption 1 justifies the necessity of downsizing. Assumption 2 states that the reservation utility that a worker obtains by staying in the public sector (refusing the government's offer) is higher for the efficient type than for the inefficient type and that the difference between the two reservation utilities is exactly equal to the cost differential between the two types. Assumption 2 implies that the public sector before downsizing is discriminating as the differential of informa- tion suggests. Thus asymmetric information about workers' efficiency also ex- ists within the public sector. In fact, UP(O) is determined by three factors: nonwage benefits, wages, and production costs. Suppose that every worker in the public sector before downsizing receives the same nonwage benefits and the same wage regardless of type. Then, the difference between UP(Q) and UP(6) should be equal to the cost differential between the two types (AO) because we normalized the quantity produced by each worker to 1. We define A,'s full cost as 0, + Ut(0,). The efficient type's full cost is of 0 + Ut; the inefficient type's full cost is if 0. The difference in full cost between the inefficient and efficient types is A&f - - Hf The efficient type becomes the low-cost type if AO - Um = (AOf) > 0; it becomes the high-cost type if AO - U' < 0. Whether the efficient type is the low-cost type depends crucially on the opportunity cost Utm. Because we normalized the inefficient type's opportu- nity cost to 0, UL' represents the difference in terms of opportunity cost between the two types. In the following sections, we distinguish three cases depending on the value of AO - U: AO > U'n AO < U, and AO = Um. III. BENCHMARK: THE COMPLETE INFORMATION CASE As a benchmark, we consider the case in which there is complete information about cost parameter O,. For downsizing to be an issue, it is natural to assume that ULP(O) 2 Um(0) for all 0 E {0, 6}. For the downsizing mechanism to induce participation, it should satisfy the following individual rationality constraints: (1) U(O) - pO + (1 - p)Utn 2 UP for the efficient type (2) U(H) -t - > UP for the inefficient type. 74 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Expected social welfare, denoted by EW, is defined as follows: EW _ S[vp + (1- v)p] - (1 + k)[vt + (1- v)t] + vU + (1 - v)U where U_ U(f) and U- U(H). Hence, the government's program under complete information, denoted by P', is given by: {maxEW pC p, p,t,t subject to constraints 1 and 2. It is useful to rewrite the equation for expected welfare in terms of workers' utilities and full costs as follows: EW = S[vp + (1- v)p] - X[vU + (1- v)U] - (1 + X)[VpO + (1- V)pW ] + (1 + k)vUm. From this expression, it is costly for the government to give positive utility levels to workers (beyond the utility derived from the public sector) as long as the shadow cost of public funds is positive. Using the above expression for expected welfare, the government's program in terms of full costs can be written as follows: [maxEW p,p,tlt pc,J subject to t-pOE fUP+Aef t - f > UP From the efficient type's individual rationality constraint in P", we can rein- terpret the efficient type as the type whose marginal cost is equal to Of and whose reservation utility in the public sector is equal to UP + AHf, ignoring the reserva- tion utility in the private sector. And we can do the same for the inefficient type. From the first-order conditions with regard to the probabilities, we can define the following social marginal costs of keeping a worker in the public sector under complete information: MC' (1 + k)ftf for the efficient type C-c _ (1 + X)6f for the inefficient type where the superscript c indicates complete information. The social marginal cost of keeping a worker is equal to the worker's private marginal cost (given by the worker's full cost) multiplied by (1 + X). The shadow cost of public funds inter- Jeon and Laffont 75 venes because the government must resort to distortionary taxation in order to raise the necessary money. The social marginal cost of keeping an efficient worker is smaller than that of keeping an inefficient worker if and only if AO - U' = AOf > 0 holds. We characterize the optimal downsizing mechanism under complete informa- tion in proposition 1. PROPOSITION 1. Under assumptions 1 and 2, the efficient downsizing mechanism under complete information {[p , p C, t t} is characterized as follows. 1. Utility. Every worker accepts the government's offer and obtains a utility level equal to the status quo utility level in the public sector [U(01) = UP(O,)]. 2. Probability. We can distinguish five cases according to the value of AO - Um and the social value of public production: Social value of public production AO > Um AO < U" AO = Um Low O < p < 1 p*= 0 - < p- < 1 vp* + (1 - V)p High p= 1 0 < <1 p* <1 0 < p < 1 PROOF. The proof is trivial and is omitted. For the formal exposition, see appendix A. Under complete information, the government offers the minimal utility levels necessary to induce the workers to accept the downsizing mechanism, that is, the status quo utility levels in the public sector. Figure 1 shows how to determine the optimal number of workers to keep in the public sector when Ao > U_. If the social value of public production is low- that is, if S'(.) is small-the number is determined by the point where the social marginal surplus of keeping a worker is equal to MCC. If the social value of public production is high-S'( ) is large-the number is determined by the point where the social marginal surplus is equal to MCC. In the general case in which AO is different from U_m if the social value of public production is low, the government lays off all the high-cost workers and keeps, with a random mechanism, a proportion of the low-cost workers. In prac- tice, secondary personal characteristics known to the government may be used to choose the set of workers laid off, rather than using a random mechanism. If the social value of public production is high, the government keeps all the low- cost workers and lays off a proportion of the high-cost workers. In the special case in which AO is equal to _Um the social marginal cost of keeping an efficient worker is equal to that of keeping an inefficient worker. Hence, the government is only interested in the total number of workers to keep. Given the total num- 76 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Figure 1 The Optimal Vumber of Workers under Complete Information, AO > U' Social marginal i () surplus S'( ), social marginal cost MC MC" VP* V V+ (1-V)P# 1 Number of workers ber, the government does not care about the choice between the efficient and inefficient workers. Hence, in this case, randomization within the whole set of workers (p* = p - = p*, t0 = f *) is an optimal solution. Appendix A gives a formal characterization of the efficient downsizing mecha- nism, including the optimal transfers. If the government keeps a worker, it pays the status quo utility level plus the production cost. If it lays off the worker, it pays the status quo utility level minus the worker's reservation utility in the private sector. IV. THE ASYMMETRIC INFORMATION CASE In this section, we assume that there is asymmetric information about cost parameter O,. We distinguish three scenarios, depending on the value of AO - Urn: AO > Utm AO < Um, and AO = U'. Because AO UV is generic and the second scenario is perfectly symmetric to the first one, we will focus on the first scenario. For the downsizing mechanism to induce truth telling, it should satisfy the following incentive compatibility constraints: (3) t - pO + (1-p)UM Žt-p+(1- p)U for the efficient type (4) t - 5O > t - pe for the inefficient type. Jeon and Laffont 77 The government's program under asymmetric information, denoted by pa, is given by: fmaxEW pa p,p,t,t lsubject to constraints 1, 2, 3, and 4. There is no loss of generality in considering full participation. Any mechanism that induces the participation of only one type is equivalent to the mechanism with full participation in which the contract for the excluded type, denoted by He, is given by {p(Oe) = 1, t(Oe) = EP(0e) + Oel. pa can be written equivalently in terms of full costs as in the following program, denoted by pal: maxEW p,p,t,t subject to paw t -p PO2 U + o pal - ~f>u t-pof >U -pO Under asymmetric information, workers may obtain informational rents in the downsizing mechanism. A,'s informational rent is defined as the difference between the utility level when the worker accepts the government's offer [U(H,)] and the worker's status quo utility level [UP(O,)]. Because rents are costly to the principal, the optimal probabilities under asymmetric information will be deter- mined by the best tradeoff between efficiency and rent extraction. We examine now the government's program (Pa') in the case in which the efficient type is the low-cost type: AO > Um. Figure 2 shows intuitively which constraints are binding in pat. The solid line starting from point A is the ineffi- cient type's indifference curve keeping utility equal to UV. The other solid line, with a less steep slope, is the efficient type's indifference curve keeping utility equal to UP + AO. Suppose that the social value of public production is low, such that under complete information the optimal probabilities are given by 0 < P >< 1 and V = 0. Thus the efficient downsizing mechanism under complete informa- tion consists of two contracts represented by points A and B for the inefficient and efficient types, respectively. Suppose now that there is asymmetric information and that the government still offers contracts A and B. Then the inefficient type will choose B instead of A because utility is greater than UP with contract B. Thus the offer of (A, B) is not incentive compatible. Restoring incentive compatibility is possible by offering A' instead of A. The inefficient type is now indifferent between A' and B. How- 78 THE WORI D BANK ECONONMIC REVIEW, VOL 13, NO 1 Figure 2 Countervalling Incentives unzder Aswpnmetric Information, AO > fff t =U + i p Government A i f U /+O6fp transfer to the worker, t AX 0 P P p Probabilty vp ever, this result requires a high transfer to the inefficient type. Thus the govern- ment can improve social welfare by increasing a little bit the probability of keep- ing the efficient type in order to reduce the inefficient type's informational rent. Therefore, the efficient downsizing mechanism under asymmetric information will consist of A" for the inefficient type and B" for the efficient type. The ineffi- clent type is indifferent between the two contracts and obtains a positive rent. The efficient type strictly prefers B" to A" and obtains no rent. Hence, there are countervailing incentives: the inefficient type's incentive compatibility constraint and the efficient type's individual rationality constraint are binding.' From the first-order conditions with regard to the probabilities, we can define the following social marginal costs of keeping a worker in the public sector under asymmetric information: MC -(1 + )Of - 1 i - v) Af < MC for the efficient type MCa (1 + )Of = MC for the inefficient type where the superscript a indicates asymmetric information. Asymmetric informa- tion makes downsizing more costly because of the rent that the government 1 The problem differs from a classical principal-agent model with adverse selection in which the status quo utility levels are identical for the two types, leading to a nonnegative rent for the efficient type. Here, the efficient type's status quo utility level is higher by AOf, making any contract ensuring this level attractive for the inefficient type. Jeon and Laffont 79 gives up to the inefficient type.' The fact that this rent is decreasing in p makes the social marginal cost of keeping an efficient worker smaller under asymmet- ric information than under complete information. In fact, the difference between MCC and MC' is equal to [X(1 - v)]/vAOf, implying that the rent extraction effect is increasing in the shadow cost of public funds (X), in the proportion of the inefficient workers to the efficient workers (1 - v)/v, and in the marginal impact of the increase of p on the inefficient type's rent (AOf). However, asymmetric information does not change the social marginal cost of keeping an inefficient worker because p does not affect rent extraction. We characterize the efficient downsizing mechanism under asymmetric infor- mation in proposition 2. PROPOSITION 2. Suppose that AO > Ur holds. Under assumptions 1 and 2, the efficient downsizing mechanism under asymmetric information {p X, p* -, t* t'* } is characterized as follows. 1. When the social value of public production is low: a. Utility. Every worker accepts the government's offer. The efficient type's utility level is equal to the status quo utility level, while the inefficient type's utility level is greater than the status quo utility level. b. Probability. All the inefficient workers get laid off, and a proportion of the efficient workers is retained. The size of downsizing is smaller under asymmetric information than under complete information. 2. When the social value of public production is high, the optimal complete information outcome characterized in proposition 1 can be achieved. PROOF. For the formal exposition and proof, see appendix B. When the social value of public production is low, the government lays off all the inefficient workers and keeps a proportion of the efficient workers. But the government keeps more efficient workers under asymmetric information than under complete information. Figure 3 illustrates how asymmetric information reduces the size of downsizing when the social value of public production is low: pi < p -. When the social value of public production is high, the government keeps all the efficient workers and lays off a proportion of the inefficient work- ers. In this case, the proportion of the inefficient workers being laid off is not affected by asymmetric information. In fact, the government achieves the opti- mal complete information outcome because the inefficient type obtains no rent when p- - = 1. The transfer to the efficient type is the same as under complete information, while the transfer to the inefficient type is equal to the transfer under complete information plus the worker's informational rent. Of course, the transfers are functions of (p * *, p - *) instead of(p *, p *); see appendix B. 2. The inefficient type's rent is given by (1 - p)AOJ. 80 THE axORlD BANK ECONOMIC REVIEW, NOL 13, NO I Figure 3 The Optimal Nuxmber of Workers lnderAsymmetric Information, AS> U, Social marginal S'() surplus S'( ) social marginal cost MC \MCC w MC MC' vp** V v + (1 -V)pc* Number of workers Because the case in which the inefficient type is the low-cost type (AO < Um) is perfectly symmetric to the case AO > Um, we just state the results for this case in proposition 3. For a more detailed analysis of this case and the proof of propo- sition 3, see Jeon and Laffont (1999). PROPOSITION 3. Suppose that AO < Ur holds. Under assumptions I and 2, the efficient downsizing mechanism under asymmetric information is characterized as follows. 1. When the social value of public production is low: a. Utility. Every worker accepts the government's offer. The inefficient type's utility level is equal to the status quo utility level, while the efficient type's uttlity level is greater than the status quo utility level. b. Probability. All the efficient workers get laid off and a proportion of the inefficient workers is retained. The size of downsizing is smaller under asymmetric information than under complete information. 2. When the social value of public production is high, the optimal complete information outcome characterized in proposition 1 can be achieved. PROOF. The proof is omitted because of the perfect symmetry with the case AO > Utm. Consider now the special case in which AO = Um holds. In this case, there is no difference between the efficient type and the inefficient type, not only in terms of Jeon and Laffont 81 the full cost but also in terms of the reservation utilities both in the public sector and in the private sector. Because there is only one type, asymmetric informa- tion about 0, does not matter and the optimal complete information outcome can be achieved, as stated in proposition 4. PROPOSITION 4. Suppose that AO = Um holds. Suppose that there is asymmetric information about 0 . Under assumptions 1 and 2, the optimal complete information outcome characterized in proposition 1 can be achieved. PROOF. The proof is trivial and is omitted. We note that the randomization analyzed in the complete information case is still an optimal solution in the asymmetric information case. V. IMPLEMENTATION WITHOUT REGRET The optimal revelation mechanism obtained in section IV can be implemented with a nonlinear transfer mechanism t(p) as usual, offering different transfers for different probabilities of remaining in the public sector. In this section, we show that the efficient downsizing mechanism can be implemented through a menu of probability, wage, and severance pay triplets. Indeed, we are interested in the implementation in which no worker regrets having participated in the downsizing procedure, regardless of whether the worker stays or gets laid off. We denote the triplet of probability, wage, and severance pay designed for type 0 by [p(0), w(0), s(0)], with 0 e {Q, 01. The sequence of implementation is as follows. The government offers each worker the menu of triplets [p(0), w(6), s(O)]. For those workers who have accepted one of the triplets, the outcome of the random mechanism associ- ated with the probability specified in the triplet gets realized. If the outcome is to leave the public sector, the worker should leave, receiving the severance pay specified in the triplet. If the outcome is to remain in the sector, the worker should remain in the sector, receiving the wage specified in the trip- let. The wage should be interpreted in a broad sense; it comprises all the monetary and nonmonetary benefits that a worker obtains by working in the public sector. The optimal menu for implementing the efficient downsizing mechanism char- acterized in section IV is given by: p(A) = p * 8(0), w(o) = w(O) = UP (0) + 0 s(H) = UP(0)-U' (0) + R(6) with H E {H, H1 where R(0) represents type 0's informational rent under the efficient mechanism. In the equilibrium under the optimal menu, each worker will accept the offer and choose the triplet designed for that worker's type. Moreover, the menu sat- 82 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 isfies ex post individual rationality in the sense that every worker obtains at least the status quo utility level in the public sector. Hence, no worker will regret having participated in the downsizing procedure regardless of whether the out- come is to stay or to get laid off. This result can be achieved because type- dependent wages and severance payments give the government more degrees of freedom than do transfers that are not conditional on the outcome of the ran- dom mechanism. We observe from the optimal menu that only those leaving the public sector may obtain an informational rent and that those remaining in the sector receive the same wage regardless of type. Moreover, this wage is equal to the wage before downsizing. The wage before downsizing, denoted by w, is defined im- plicitly from the status quo utility level by the following equation: UP(O) = wP _ Ofor all 0 e{O,6}. Here we show how the efficient downsizing mechanism can be implemented through the optimal menu. (See Jeon and Laffont 1999 for a complete case-by- case analysis.) Consider the case in which the efficient type is the low-cost type: AO >LT. Suppose that the social value of public production is low such that the optimal probabilities are given by 0 < p'*< 1 and p F> = 0. In this case, the optimal menu is given by: w(O) = U i+ - s(O) = U -U s(O) = UP + (1 -p *)(AO- Un)- Because 0 * 0, w(O) can take any value. A proportion of the efficient work- ers remains in the public sector, receiving a wage equal to their status quo utility level in the public sector plus their production cost. The rest of the efficient workers leave the sector and are offered a severance pay equal to their status quo utility level in the public sector minus their reservation utility in the private sector. Hence, every efficient worker gets zero ex ante and zero ex post rent. The inefficient workers always get laid off and are offered a severance pay larger than their status quo utility level in the public sector. Hence, they obtain a posi- tive informational rent. We note that the severance pay for the efficient type is larger than the one for the inefficient type: s(O) - s(O) > 0. VI. EXTENSIONS In this section, we relax the main assumptions and investigate how our analy- sis can be extended. When Discrimination before Downsizing is Less Than the Differential of Information We have considered only the case in which the public sector before downsizing discriminates as the differential of information suggests. Even if this is a good benchmark, the public sector before downsizing is likely to discriminate less than the differential of information suggests: AO > UP - UP > max {O, Um - UP}. Jeon and Laffont 83 This case corresponds to the stylized fact that low-ability workers tend to be overpaid and high-ability workers tend to be underpaid in public sectors of de- veloping countries, compared with their peers in the private sector. Hence, it will be interesting to examine this more general case. We fix the level of LJP and investigate the impact of the variation in the level of UP on the efficient downsizing mechanism. The optimal probabilities under com- plete information are not affected by the change in the level of UP because the full costs are independent of it. We investigate below the impact on the optimal probabilities under asymmetric information in the case in which the efficient type is the low-cost type (AO > Um). Suppose that the social value of public production is so low that the optimal probabilities under complete information are given by 0 < p-'< 1 and p- = 0. In figure 2, a decrease in the level of LTP will shift the efficient type's indiffer- ence curve (which keeps the worker's utility equal to the status quo utility level) downward without affecting the slope. Thus there will be three cases. First, if U is large enough, the inefficient type's incentive compatibility constraint is bind- ing and the size of downsizing will be smaller under asymmetric information than under complete information: p* < p". Second, if UP is small enough, the efficient type's incentive compatibility constraint will be binding. In this case, the efficient type obtains a positive informational rent, but the size of downsizing will not be affected by asymmetric information: p = p * *.3 Third, if LP is inter- mediate, both incentive compatibility constraints are slack, and the optimal com- plete information outcome can be implemented: p'F = p*'. Hence, countervailing incentives exist only when UP is large enough. Mandatory Downsizing Here we look at mandatory downsizing instead of voluntary downsizing. The latter may favor workers who have already been favored by entering the public sector because the government must compensate the workers for their status quo utility level in the public sector to induce their participation. Under manda- tory downsizing, the utility that worker A, expects to obtain by rejecting the government's offer is given by U"'(O,) instead of UV(O,). It is easy to see that the optimal probabilities under complete information are the same regardless of whether downsizing is voluntary or mandatory. We study the optimal probabilities under asymmetric information, maintaining the as- sumption that the public sector before downsizing is discriminating as the differ- ential of information suggests. Consider the case in which the efficient type is the low-cost type (AO > LI'). Suppose that the social value of public production is high, such that the optimal probabilities under complete information are given by p*> = 1 and 0 < p *< 1. In figure 2, under mandatory downsizing, the indiffer- ence curves for the efficient and inefficient types have the same constant, given 3 However, if the social value of public production is high, such that the optimal probabilities under complete information are given by pa = 1 and 0 < p I < 1, the size of downsizing will be larger under asymmetric information than under complete information: p ' - < p -. 84 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I by (p = 0, t = 0). Hence the efficient type's incentive compatibility constraint is binding, and the size of downsizing is larger under asymmetric information: F**< p-. Under mandatory downsizing, the monetary transfer is reduced for every worker by at least UP. However, the government gives up a positive rent to the efficient type and thus introduces a downward distortion in p i Other cases can be analyzed in the same way. Several Types The analysis can be extended to the case in which there are several types: the cost parameter in the public sector 0 can take several values. Consider the regu- lar case in which full cost [6 + Ut(6)] is monotone in 0. Under complete informa- tion, the social marginal cost of keeping a worker in the public sector will be monotone in 0. Thus there will be a cutoff type such that the government will keep this type with a positive probability and will lay off (respectively, keep) with probability 1 all the types whose full cost is larger (respectively, smaller) than the cost of the cutoff type. We analyze what happens under asymmetric information, maintaining the assumption that the public sector before downsizing is discriminating as the differential of information suggests. Consider the case in which [6 + UZ(0)] is increasing in 6. In figure 2, the indifference curve that keeps the individual ratio- nality constraint binding will have an increasing slope and a decreasing constant in 6. Every curve will pass through the point [p = 1, t = 0 + UJP(O)], where [0 + UTP()] represents the wage before downsizing and has the same value for all 0. Let 6C denote the cutoff type under complete information. Suppose that under complete information we have 0 < p *(Oc) < 1. Let tr (0c) be the transfer that makes the cutoff type's individual rationality constraint bind- ing. Then, under asymmetric information, every type whose cost parameter is larger than OC will have a strictly positive rent by taking the contract [p*(6%), t*(Oc)]. Hence, there are countervailing incentives. Under certain regularity con- ditions, the social marginal cost of keeping a worker under asymmetric informa- tion will be increasing in 0, but it will be smaller than it would be under com- plete information. Consequently, under asymmetric information, the size of downsizing is smaller: either the cost parameter of the cutoff type under asym- metric information is larger than OC, or it is unchanged and the probability of keeping the same cutoff type under asymmetric information p**(0c) is larger than p*(0C). Risk Aversion We can relax the assumption of risk neutrality and incorporate risk aversion in our model. Then in the downsizing mechanism, the transfer to a worker should depend not only on type but also on whether the worker gets laid off. The utility 4. Under mandatory downsizing, rent is defined not with regard to UP( ) but with regard to U-( ) Jeon and Laffont 85 that a worker expects to obtain by entering the private sector [Um'()] should take into account the uncertainty related to finding a job. In the special case in which risk aversion is represented by ex post individual rationality constraints, the efficient downsizing mechanism is given by the case analyzed in section V. In this case, risk aversion has an effect only on the indi- vidual rationality constraints and not on the incentive compatibility constraints. However, in general, if workers have risk aversion, they will have incentives to take the contract without randomness (for example, to stay with probability 1 in the public sector). Thus the government should pay a risk premium to induce workers to choose a random contract. VII. CONCLUDING REMARKS We have assumed that the quantity produced by each worker in the public sector is not a controllable instrument and is normalized to 1 regardless of type. This assumption can be relaxed if the public sector has some information about workers' productive efficiency. In this case, the quantities before downsizing may vary according to the information possessed by the public sector. This rein- forces the relevance of the case in which the public sector before downsizing is discriminating less than the differential of information suggests. One important question is what is the optimal way to use this information for downsizing. In general, only direct superiors have information about their subordinates' pro- ductive efficiency. Moreover, those superiors might be engaged in favoritism or collusive behavior. Hence, it will be necessary to design an incentive scheme that induces superiors to reveal their information and deters them from engaging in favoritism or collusive behavior. The stochastic element is a main feature of the efficient downsizing mecha- nism analyzed in our article. However, if the integrity of public officials is in doubt because of favoritism or corruption, the implementation of stochastic mechanisms may be difficult. In this case, we need to find an alternative that is more transparent and less prone to administrative arbitrariness. From this point of view, fixed-term contracts seem to have desirable features, but they will be much more costly. Another interesting extension is to incorporate type-dependent externalities. For example, if the labor market is not competitive, the reservation utility in the private sector cannot capture all the social surplus that a worker generates in the private sector. Once we have type-dependent externalities, they will affect the first-order conditions. Hence, the conditions determining which type of workers the regulator should start to lay off will change. In some countries, downsizing in the public sector is accompanied by retrain- ing programs for laid-off workers. If the outcome of retraining depends on workers' productive efficiency in the public sector, retraining can be a useful supplementary instruIment for solving the adverse selection problem. Hence, it 86 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 would be interesting to integrate retraining as an element of the downsizing mechanism. Finally, although we mentioned the brain-drain effect in the introduction, we did not explicitly analyze it in our model. Explicit consideration of this effect will make it less likely that it is optimal to start laying off efficient workers. APPENDIX A. THE EFFICIENT DOWNSIZING MECHANISM UNDER COMPLETE INFORMATION The Formal Characterization The efficient downslzing mechanism under complete information tp *, p, _ , t *1 is characterized as follows. 1. The two individual rationality constraints are binding. 2. The probabilities are given as in proposition 1. 3. The transfers are obtained from the binding individual rationality con- straints: t. = p* ((U + 0) + (1-p )(U -U ) t = p(UP +9)+(1 -T*)Up. First-Order Conditions The first-order conditions determining the optimal probabilities are given as follows. CASE 1. When AO > Ur holds. a. If the social value of public production is low: MCC =S'(vp-)) UP. It is satisfied because we have AO - Um > 0. The efficient type's incentive compatibility constraint is satisfied if the follow- ing inequality holds: UP + AO>U P+ (l-p: -) (AO-U ) +ip*AO3+ (1-pi )U"' It is satisfied because we have p ** 2 p * * and AG - Un > 0. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Diwan, Ishac. 1994. "Public Sector Retrenchment and Severance Pay: Nine Proposi- tions." In Shahid A. Chaudhry, Gary J. Reid, and Waleed H. Malik, eds., Civil Ser- vice Reform in Latin America and the Caribbean: Proceedings of a Conference. World Bank Technical Paper 259. Washington, D.C.: World Bank. Gibbard, Allan. 1973. "Manipulation of Voting Schemes: A General Result." Econometrica 41(July):587-601. Green, Jerry R., and Jean-Jacques Laffont.1977. "Characterization of Satisfactory Mecha- nisms for the Revelation of Preferences for Public Goods." Econometrica 45(March):427-38. Jeon, Doh-Shin, and Jean-Jacques Laffont. 1999. "The Efficient Mechanism for Downsizing the Public Sector." Policy Research Working Paper. World Bank, Wash- ington, D.C. Forthcoming. Processed. Jullien, Bruno. 1997. "Participation Constraints in Adverse Selection Models." Institut d'Economie Industrielle, Toulouse (IDEI). Processed. Kahn, Charles M. 1985. "Optimal Severance Pay with Incomplete Information." Jour- nal of Political Economy 93(June):435-51. Lazear, Edward. 1995. Personnel Economics. Cambridge, Mass.: MIT Press. Levy, Anat, and Richard McLean. 1996. "Optimal and Suboptimal Retrenchment Schemes: An Analytical Framework." Policy Research Department, Rutgers Univer- sity, New Brunswick, N.J. Processed. Lewis, Tracy R., and David E. M. Sappington. 1989. "Countervailing Incentives in Agency Problems." Journal of Economic Theory 49(December):294-313. Maggi, Giovanni, and Andres Rodriguez-Clare. 1995. "On Countervailing Incentives." Journal of Economic Theory 66(June):238-63. Myerson, Roger. 1979. "Incentive Compatibility and the Bargaining Problem." Econometrica 47(January):61-73. Rama, Martin. 1997. "Efficient Public Sector Downsizing." Policy Research Working Paper 1840. Policy Research Department, World Bank, Washington, D.C. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 89-116 Earnings and Welfare after Downsizing: Central Bank Employees in Ecuador Martin Rama and Donna MacIsaac This article measures the earnings and welfare losses experienced by displaced employ- ees of the Central Bank of Ecuador. It links these losses to individual characteristics such as gender, education, seniority, and salary in the public sector. Data are from a survey of displaced employees that included subjective evaluations of well-being in addition to information on activity and earnings. The welfare losses of separated em- ployees are not highly correlated with their earnings losses, partly because some of them (especially women) withdrew from the labor force after separation. Earnings and welfare losses also vary depending on the nature of displacement, which was voluntary for roughly half the employees and involuntary for the rest. Overall, the losses were larger for employees with less education and more seniority, but not necessarily larger for employees with higher salaries. However, compensation for displacement was based on a rule of thumb that involved only salary and seniority and was applied across-the- board. For those employees who left voluntarily, the resulting compensation package was, on average, about 20 percent higher than the welfare loss. The article derives the implications of these findings for the design of assistance programs for displaced workers and, more specifically, for the tailoring of compensation packages to their individual characteristics. Public sector downsizing is an increasingly important component of economic reform, both in industrial and in developing countries. Nowhere is the need to trim public sector employment more obvious than in formerly planned econo- mies. But it is also considerable in Latin American countries, where a long his- tory of populist economic policies has led to bloated bureaucracies and over- staffed state-owned enterprises. A good case in point, the Central Bank of Ecuador (BCE) had 5,800 relatively well-paid employees in the early 1990s. While their work is now done by some 2,700 (better-paid) employees, many other public sector agencies and firms around the world have yet to start moving in this direction. Martin Rama is with the Development Economics Research Group at the World Bank, and Donna Maclsaac is with Wells Fargo. This article was written for the research project Public Sector Retrenchment and Efficient Compensation Schemes, supported by the Research Committee of the World Bank through grant 679-5 1. The authors are especially grateful to Constanza Calder6n They are also grateful to Julia Gonzalez for her skillful assistance with the field interviews and to Bruce Fallick, Florencio L6pez-de- Silanes, John Newman, staff at the research department of the Central Bank of Ecuador, and three referees for their helpful comments. (© 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 89 90 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I One of the main obstacles to public sector downsizing is the large welfare loss that displaced workers may experience. Typically, some of these workers spend long periods unemployed or out of the work force. Others move to activities characterized by lower pay, less income security, and higher levels of effort than the jobs they used to hold. The extent of their welfare loss depends on the nature of the alternatives to their public sector job, the value they attach to income security, and their aversion to work effort. Not being able to predict these losses, or to defuse the resistance to downsizing, governments sometimes offer overly generous severance pay packages. Many potential losers are thus transformed into winners, and public sector downsizing ends up being a very expensive endeavor. In this article we measure the individual losses experienced by displaced BCE employees and link these losses to their observable characteristics. Although the exercise focuses on a relatively narrow group of public sector workers, its impli- cations go beyond their particular case. To date, there have been few studies on this topic, and policymakers willing to assess the amount of resources needed to compensate displaced public sector workers, or to predict the harshness of resis- tance to downsizing in the absence of such compensation, have little on which to build. Moreover, these studies seldom refer to white-collar workers in middle- income countries, a group that will face significant job losses in the coming years.1 Because these workers tend to be highly vocal and politically influential, compensation issues will certainly be of paramount importance in their case. This article should not serve to justify the complete compensation of dis- placed workers. Many public sector employees get rents from their jobs, and these rents are paid for by the rest of society. On fairness grounds, it would seem that they should bear some of the cost of adjustment. Moreover, public sector employees are rarely among the poorest segments of the population. Therefore, a government concerned about poverty alleviation could find better uses for its scarce resources than the complete compensation of this particular group. But to compensate displaced employees even partially, the government would need in- formation about the cost of complete compensation. And this information may be relevant for predicting the fierceness of the resistance to restructuring. Fi- nally, in some countries public sector employees cannot legally be fired, which implies that the government has to "buy" them out. In those countries, an accu- rate estimate of the losses from displacement may help to reduce the cost of downsizing. Our main contribution in this article is methodological. Attempts to estimate the losses experienced by displaced workers usually involve a comparison of 1 Alderman, Canagaralah, and Younger (1996) analyze the case of government employees in Ghana, most of whom moved to agriculture after being retrenched Tansel (1997) focuses on blue-collar workers in Turkey. Assaad (in this issue) compares workers in and out of the Egyptian public sector, taken as a whole. Only the study by Robbins 1 1996) considers a group of workers similar to ours. Other studies of downsizing are descriptive in nature, with no attempt to quantify the loss experienced by individual workers Rama and MacIsaac 91 earnings in two different states, such as in and out of the public sector or before and after separation. But as important as the change in earnings may be, it rep- resents only one aspect of the welfare loss. The disutility from lesser income security and (possibly) higher effort levels, as well as the utility from additional leisure time (if any), also need to be taken into account. We use discrete mea- sures of earnings losses and welfare losses reported directly by the displaced employees, in addition to more traditional measures of earnings losses. We use regression analysis to estimate the link between these measures and the observ- able characteristics of the employees. These characteristics include salary and seniority in the public sector, which are the two main attributes considered in the rules of thumb on which compensation is usually based, but they are not restricted to these two variables. BCE gave about half of the displaced employees the choice of staying in their jobs and forced the rest to resign. Thus, self-selection is a key problem in evalu- ating the link between the losses ensuing from downsizing and the observable characteristics of the employees. Most likely, those who accept a severance pay package and leave the public sector are also, other things being equal, those who face the best outside alternatives or have the lowest aversion to risk and effort. Using the losses experienced by these workers to predict the losses of those who remain in the public sector may be misleading. We deal with this problem by replicating the analysis of welfare losses separately for BCE employees who had a choice and for those who did not. The differences observed between these two groups provide an indication of the extent and nature of the biases resulting from self-selection. Section I narrates the downsizing process, including the mechanisms set up by BCE to compensate the displaced employees, as well as to decide which among them would be forced to resign and which would have the choice of staying. Section II describes the survey we implemented to gather information on the activity, earnings, and welfare of displaced employees from the Quito branch of BCE in the 15 to 18 months following their separation. Section III analyzes how displaced employees allocated the compensation they received and evaluates the returns on the investments they made with it. Section IV retraces employees' activity and earnings history after separation; it shows that unemployment rates were quite low, but the quality of the new jobs was low as well. Sections V and VI construct and analyze indicators of the actual and perceived earnings loss experienced by displaced BCE employees. Section VII focuses on indicators of the welfare loss and systematically links them to the observable characteristics of the displaced employees. Section VIII draws policy implications from the analy- sis, and section IX concludes. I. THE DOWNSIZING PROCESS In the early 1990s it became apparent that BCE was overstaffed. Part of the growth in employment was due to political patronage, which had been particu- 92 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I larly intense during the 1980s. But bad management was to blame as well. Over the years, management did not restructure or suppress poorly functioning units; instead, it replicated them by creating other units with similar tasks. Moreover, financial liberalization reduced the scope for some of the activities traditionally performed by BCE, such as those related to credit allocation. Gradually, the management of BCE came to accept that a reduction in personnel was warranted. In 1991, it decided that the work force had to be cut by half. This target was intended to signal management's resolve to address the overstaffing problem. It was not set as the result of a careful analysis of the functions of BCE and the productivity of its staff. The reduction in personnel took place over two distinct phases. A first round in 1992 led to the departure of 1,400 employees on a strictly voluntary basis. At that time, there was no appropriate legal framework for the downsizing. The compensation mechanism was set up by the monetary authority (Junta de Regulaci6n Monetaria, or JRM) itself. This mechanism was based on two vari- ables: salary and seniority. More specifically, the compensation offered to those willing to resign was 1.7 times their average monthly salary in 1992, multiplied by their number of years of service in BCE. The resulting figure could not fall below 5 million sucres or rise above 50 million sucres (Ecuador's currency is the sucre). This amounted to a range of $3,600 to $36,000 per employee in 1992 U.S. dollars. At that time, income per capita in Ecuador was about $1,400 per year. Over a period of slightly less than three months, BCE offered these packages to all employees below retirement age. It gave a special committee discretionary power to reject applications by employees that it considered essential. Chaos ensued, as most employees thought the compensation offered was too generous to last and applied for severance. Work effort dropped, as many started planning their impending exit. Still, the trade union representing BCE employees fought to improve the severance pay package, while employees above retirement age lobbied for access to it. The latter group saw its efforts rewarded less than one week after the mechanism was announced, under the condition of relin- quishing the entitlement to an old-age pension. But later on, many of those who benefited from the severance pay package sued BCE to get their pensions paid anyway. Last but not least, employees whose applications were rejected (mainly tellers) managed to mobilize some politicians to gain access to the package. Yet, not all politicians were supportive. Congress questioned the legitimacy of the mechanism set up by the JRM, considering it an abuse of government funds. Electoral competition placed the downsizing process under even more severe criticism. A parliamentary commission was created to investigate the is- sue, and the chair of BCE faced the possibility of a trial. The Ministry of Finance, in turn, succeeded in taxing severance payments as income, which led yet an- other group of former employees to sue BCE. In the meantime, labor productivity declined further, due to the lack of selectivity of the reduction in personnel. When the government changed, the chair of BCE was finally dismissed. Rama and Macisaac 93 The second phase of the downsizing took place in 1994 in a very different context. Under the new, more market-oriented government, parliament passed a law for "modernization of the state." This law provided a framework for the implementation of compensation packages. Moreover, the management of BCE decided that the second phase of the downsizing had to be based explicitly on the role and functions defined by the law for "monetary regulation and state banking" of May 1992. Thus reducing the overlap between units and the exist- ing duplication of effort became a deliberate goal of the exercise. It was finally agreed that BCE had to return to its size prior to the great expansion of the 1980s, which amounted to taking some 1,500 extra employees off the payroll. In practice, the second phase commenced with a classification of all personnel in three groups, labeled A, B, and C. BCE could not afford to lose employees in group A; they were therefore denied access to the new severance pay package. Employees in group C were clearly redundant. BCE could not legally fire them, however, so they were informed, on an individual basis, that new laws under preparation would eliminate job security for government employees and that next time the compensation package offered to them would be much less gener- ous. This threat was very successful, in that all of the employees in group C agreed to leave BCE. Finally, employees in group B were given the choice to stay in BCE or to take the severance pay and leave. We refer to employees in group B as those who "had a choice" and to employees in group C as those who "had no choice." The JRM set compensation at twice the average monthly salary in 1993, mul- tiplied by the number of years of service with BCE. The resulting figure could not fall below 10 million sucres or exceed 400 times the national minimum wage, which effectively capped individual compensation at 26 million sucres. Sever- ance pay thus ranged from $4,800 to $12,400 per employee (in 1993 dollars). From the outset, the government agreed not to tax severance pay as income. The total cost of the operation was on the order of 35 billion sucres (roughly $17 million in 1993), and its financial payback period was estimated at 20 months. As before, the trade union and some politicians pressured the management of BCE, but overall the process went much more smoothly than in the first phase. Compensation was the main component of the 1994 downsizing program, but not the only one. The program included a training mechanism to facilitate the transition out of the public sector. Courses on microenterprise management and financial investments were offered free of charge to any resigning employee willing to take them. Other courses were organized at the request of some of the parting employees who wanted to become insurance agents in the private sector. For those staying, BCE overhauled the pay structure and working conditions and introduced job descriptions and performance evaluations. BCE sold 9 of its 14 buildings in order to regroup personnel physically. Reportedly, the morale of those who stayed with BCE improved. 94 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I II. A SURVEY OF DISPLACED EMPLOYEES To assess the change in earnings, employment status, and welfare experi- enced, we interviewed a relatively large proportion of displaced BCE employ- ees. Our survey focused on employees who had left BcE during the second phase of downsizing. When this research project started, employees who had left during the first phase had been out of BCE for more than two years already. Given the time needed to implement and run the survey, we were concerned that limited recall capabilities would lead to a distorted picture of employment and earnings histories. More important, unlike the second phase, the first phase was not characterized by a mix of voluntary and involuntary job separations, and this mix is key to controlling for self-selection. Due to budget constraints, we decided to survey the displaced employees who used to work in the Quito branch of BCE only. For this group of employees, the target was the entire population. The management of BCE kindly provided us with copies of its own files for the 1,380 employees who left in February and March 1994, including 455 from the Quito branch. These records include information on gender, age, and years of experience both while working in BCE and before joining. They also include total earnings in 1992 and 1993. The management of BCE helped us to identify the appropriate person to run the survey. This person was the former social worker of the institution, who had been classified into the C group during the second phase of downsizing. One of the survey questionnaires filled was therefore hers. The personal knowledge she had of many employees of the Quito branch was crucial in securing a high rate of response. To minimize the risk that her per- sonal involvement could unwittingly influence the responses, multiple-choice questions were used whenever possible. Unfortunately, in the time elapsed since their resignation, many former BCE employees had moved, including 38 who could not be found, 31 who had left Quito, and 15 who had migrated out of the country. One of the former employ- ees had died, and four others did not want to cooperate with the survey. Because the responses to three of the questionnaires were quite incomplete, in the end we got a sample of 366 displaced employees, of whom 359 answered all of the questions asked. Roughly half of the surveyed employees were male, and three-quarters were married either legally or by common law. The average age was 37 years, and the average family size was slightly larger than five. Average schooling was 14 years, as almost two-thirds of the interviewees had received at least some postsecondary education. In fact, a mere 4 percent had not finished high school. Average expe- rience was 11 years, with almost 70 percent of the sample reporting no work experience before joining BCE. An astounding 98 percent of the interviewees had been members of the trade union while still working for BCE. Finally, about 53 percent of the displaced employees declared that they had no choice but to leave BCE, while the rest could have stayed, had they wished to do so. Rama and Maclsaac 95 III. THE COMPENSATION PACKAGE The compensation package, as perceived by BCE employees who left in 1994, included up to three components. The most important was the severance pay established by the JRM, based on salary and seniority, but it was not the only one. BCE reimbursed all of the surveyed employees for past contributions to its own old-age pension fund. Many of them also recovered their contributions to the general unemployment insurance mechanism (Fondo de Reserva) managed by the Ecuadorian social security system. The top portion of table 1 indicates how many of the interviewees had access to each of these three components and the importance of each component in monetary terms. (All monetary figures in table 1 and hereafter are measured in millions of sucres at constant June 1995 prices. At that time, the exchange rate was 2,500 sucres per U.S. dollar.) Strictly speaking, only severance pay represents a compensation for job loss. The other two components of the package, it could be argued, actually belong to the workers and only change hands at the time of their resignation. But this would be so only if two conditions held true. First, pensions and unemployment benefits would have to be based on actual contributions plus accrued interest. And second, BCE employees would have unrestricted access to credit. Under these two assumptions, the net wealth of displaced employees would not be affected when they cashed in their past contributions. However, neither of the two as- sumptions is likely to be true. It is usually agreed that salaried workers have limited access to credit, and there is no reason to believe that Ecuador is an Table 1. The Compensation Package and Its Use among Former Employees of the Central Bank of Ecuador, 1994-95 Number of former Average value (millions of Source and use of funds employees sucres at June 1995 prices) Components of the package (1994) Basic severance pay package 363 30.2 Contributions to old-age pension 363 18.6 Contributions to unemployment insurance 170 2.4 Total 363 51.0 Use of the package (1994-95) Current spending (food, clothing, etc.) 178 3.6 Physical investment (car, house, etc.) 252 21.5 Repayment of debts 85 2.6 Transfer to other family members 18 0.7 Financial investment 160 11.9 Launching or expansion of business 140 10.5 Total 363 51.0 Note- The average yearly salary before separation was 22.1 million sucres in June 1995 prices All BCE employees received more than one type of compensation, and most spent it on more than one item. Individual amounts received and spent may not add up to the total due to rounding. Source: Authors' calculations. 96 THE WORLD BANK ECONOMIC REVIEW, VOL 13. NO I exception in this respect. More important, Ecuadorian social security programs are characterized by a very tenuous link between contributions and benefits. There are several reasons for the weakness of this link. To begin with, indi- vidual pensions in Ecuador are based on age, years of experience, and salary at retirement, rather than on past individual contributions. Moreover, even the average pension may bear little resemblance to the average contribution because the Ecuadorian social security system is not financially viable. It is therefore likely that benefits will be eroded as the population grows older. (The BCE pen- sion fund is independent of the social security system, but pressures could mount to merge them as the general old-age pension program runs out of resources.) The discounted loss from relinquishing an old-age pension today is therefore much smaller than the present value of the contributions that would be needed to gain access to that pension. As regards the unemployment insurance mecha- nism, any money deposited in it by BCE employees was likely to be considered lost, too, given the lifetime job security employees enjoyed. Because this third component is minor compared with the other two, the decision whether to count it as compensation should not affect the analysis. The lower portion of table 1 describes how interviewees used their compensa- tion packages. The breakdown refers to the sum of all cash payments they re- ceived as a result of their resignation, including severance pay, past contribu- tions to the old-age pension, and, for some, past contributions to the unemployment insurance mechanism. The first column of the table indicates how many of the surveyed workers spent at least part of their package on each of a series of items. The second column reports the average spending by item. Because of the long period elapsed since displacement, the resulting breakdown probably reflects the equilibrium allocation, rather than a point in the transition toward equilibrium. Physical investment, including the purchase of a car, a house, or an apartment, or the expansion of a previously owned house or apartment, was by far the pre- ferred investment of displaced BCE employees. Almost 70 percent of the sample favored this type of spending, which amounted on average to more than 40 percent of the compensation package (table 1). A similar fraction of the sample either made financial investments or canceled previous debts, but the amount spent, on aver- age, was smaller (less than 30 percent of the compensation package). The propor- tion of those who used some of their compensation package to start or expand an independent activity was considerable. For those who incurred this type of expen- diture, the average spending was roughly 27.2 million sucres (10.5 x 363 / 140), equivalent to more than one year of salary at BCE. Finally, about half of the interviewees used part of the compensation to pay for consumption expenditures. For this group, the average consumption spending amounted to 7.3 million sucres (3.6 x 363 / 178), which is roughly a quarter of the income they would have earned since the separation date, had they not resigned. The survey asked displaced employees to report their earnings from invested compensation. Because of the way the question was phrased, the answers do not Rama and Maclsaac 97 include imputed rent on any housing acquired or built for own use or on any other physical assets purchased, or capital gains or losses. Consequently, these answers refer only to cash returns. For financial investments, an adjustment was needed to infer real returns from the reported nominal returns because inflation was high in Ecuador at the time of the survey. The calculation is made under the assumption of a monthly increase in consumer prices of 2 percent. This figure is based on the actual inflation observed in the years surrounding the downsizing of BCE. Because such an adjustment would require heroic assumptions in the case of diversified portfolios, returns are evaluated for former employees who made either physical or financial investments, but not both. Figure 1 shows the distribution of real returns on invested compensation. The upper graph in the figure represents the rate of return on physical assets for those interviewees who made no financial investments. The lower graph corre- sponds to the rate of return on financial assets for those who made no physical investments. The distribution remains roughly unchanged if small investments (less than 1 million sucres at June 1995 prices) are set aside. The reported rates of return correspond to a 12-month period, under the assumption that the monthly returns observed at the time of the interview remain constant over time. As a result of this assumption, the variance of the observed returns is most likely overestimated. Although the average rate of return on invested compensation is high, the median rate is quite moderate and is definitely insufficient to sustain household consumption. It is actually zero in the case of physical investments, which im- plies that the latter were mostly made for the individual's own use. It is 17.5 percent in the case of financial investments. Using 10-point intervals, the major- ity of those who actually get earnings from their investments lies in the 10-20 percent interval. The income stream from a 10 percent return on the average compensation package represents less than a quarter of the average salary be- fore separation. Although not negligible, this figure clearly suggests that the compensation packages would not allow displaced BCE employees to finance their consumption expenditures out of rents only. Their ability to work remains key for them to make a living. IV. ACTIVITY AND EARNINGS AFTER DISPLACEMENT The retrenched employees scarcely used the support mechanisms implemented by BCE to facilitate the job transition. Only 12 percent of them took any of the training courses offered, although these were provided free of charge. Similarly, only one interviewee reported having received help from BCE in the search for a new job. That was the social worker identified by BCE management to assist us with the survey. It appears that the BCE trade union was not instrumental in facilitating the transition either, given that none of the interviewees received support from it. Displaced BCE employees were therefore on their own after they resigned. 98 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Figure 1. Annual Rates of Return on Invested Compensation of Former Employees of the Central Bank of Ecuador, 1995 Observations Physical investments 100 90 80 - 60 - 50 - 40 - 30 - 20 - 10 - 0~~~ Annual rate of return (percent) Financial investments Observations 20 16 14 12 10 8 6 4 2 0§ I Annual rate of return (percent) Source Authors calculations Rama and Maclsaac 99 Table 2 reports the activities they performed in the months following their resignation. Because multiple answers were allowed, the monthly total in the table exceeds the sample size. A slight upward trend in the gap between the two indicates that multiple activities, and, more specifically, multiple jobs, became increasingly prevalent over time. A more significant trend concerns the public- private split in the jobs taken. Many interviewees kept doing work for BCE after separation, in one way or another, but almost all of this work was finished about eight months later. The number of interviewees holding salaried jobs in the private sector increased accordingly. The expansion of self-employment was even larger, with more than half of the sample moving to this category within one year of displacement. The new jobs seem to be characterized by their relatively low quality, rather than by their scarcity. The unemployment rate, measured as the displaced em- ployees who were unemployed as a proportion of those who either had a job or were actively searching for one, never exceeded 15 percent. It was down to 10 percent six months after separation and to only 6.5 percent by the end of the period. The relatively low unemployment rate was not due to a buoyant economy. (The growth rate of output was 4.3 percent in 1994 and 2.3 percent in 1995, implying almost stagnant output per capita, given how fast population grows in Ecuador.) But not all the new jobs were in compliance with labor laws. By six months after separation, less than a quarter of those employed were entitled to the "teen" salaries, which are the quintessential mandated benefit in Ecuador (see MacIsaac and Rama 1997). The thirteenth to sixteenth salaries are pay- ments to be made to the worker at different times in the year, on top of the twelve monthly salaries. An equally small fraction of the employed was enrolled with social security, and barely 5 percent of those at work were unionized. These figures are hardly surprising in a context where most of the interviewees were self-employed, but they contrast sharply with the characteristics of former BCE jobs. We calculated annual labor earnings after separation based on data corre- sponding to the month when the interview took place. The calculation included both cash earnings and payments in kind but excluded earnings from invested compensation. We added the benefits mandated by law to the resulting figure when the interviewees declared that they were entitled to "teen" salaries. Con- versely, contributions to the social security system were subtracted for those who were enrolled in it. The rationale for this subtraction is the weakness of the link between contributions and benefits, as discussed in section II. In the end, the average labor earnings of those who had a job amounted to 12.3 million sucres per year (at June 1995 prices), which is roughly 55 percent of what they used to make at BCE. The survey also asked displaced BCE employees to compare their total house- hold earnings in each of the months after separation with the household earn- ings they had when they worked for BCE. In what follows, we identify this sub- jective assessment as the perceived earnings. More specifically, all interviewees Table 2. Activity Following Separation from the Central Bank of Ecuador, by Month after Separation, 1994-95 (percentage of former employees) Activity, 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Private employee 9.7 10.2 11.3 12.9 15.2 16.5 17.5 20.1 21.5 21.2 24.2 24.8 27.0 27.9 29.4 Public employee 12.7 14.6 13.2 10.7 9.4 8.3 4.7 0.6 0.3 0.3 0.0 0.0 0.3 0.3 0.3 Self-employed 28.0 29.2 34.7 38.6 41.0 41.6 43.5 47 4 47.4 48.5 52.1 52.6 51.5 51.0 50.7 Unpaid family worker 5.2 5.5 6.3 6 9 6.3 6.6 6.9 6.9 6.9 6.3 4.4 4.4 3.9 3.6 4.2 Unemployed 9.7 9 4 9.4 9.6 9.1 8 5 8.3 8.0 7.7 7.2 7.7 7.7 7.7 6 7 5.9 Study 6.0 6.9 6.6 6.1 5.5 5.5 6.1 6.1 6.1 6.3 6.9 7 2 6.9 6.1 6.6 Home care 26.1 25.9 26.2 26.7 26.2 26.7 27.1 27.0 26.7 26.2 24.5 24.8 24.8 24.8 23.1 Retirement 0.7 0 6 0.6 0.6 0.6 0.6 0 6 0.8 0.8 0.8 1.1 1.1 1.1 1.1 0.7 Rest, travel, and other 18.7 15.2 10.5 8.5 7.7 5 8 6.4 6.6 5.5 6.3 4.7 3.3 2.5 2.2 1.7 Total 116.8 117.4 118.7 120.7 120.9 120.1 121.1 123.4 122.9 123.1 125.6 125 9 125.6 123.7 122.7 Sample size (number of individuals) 268 363 363 363 363 363 361 363 363 363 363 363 363 359 286 Note. The total exceeds 100 percent due to multiple answers by some of the interviewees. The sample size is smaller at the beginning of the period because several interviewees saw month 1 as a transition point and did not answer questions about it. The sample declines at the end of the period because the first interviews were carried out during month 15. Source: Authors' calculations. Rama and MacIsaac 101 were asked to rank their monthly household earnings from 1 (much lower than when in BCE) to 5 (much higher), with 3 corresponding to the status quo. For those households with no earnings at all in a particular month, a 0 was reported. The answers to this question thus provide an indication of the overall earnings distribution over the 15 months following the downsizing. Table 3 summarizes them. Few interviewees had no household earnings at all in any particular month after separation. There were, however, some changes in the level of those earn- ings over time. The proportion with much higher household earnings than be- fore declined very quickly in the months following separation. This trend could reflect a selective memory, whereby past times look better as they become more distant, but it is more likely to result from the gradual completion of public sector jobs reported in table 2. Indeed, the proportion of those who declared much higher household earnings than when they worked for BCE dropped by half in the first eight months after separation, which is roughly the time period over which pending assignments with BCE were finalized. The proportion with household earnings similar to those they had before increased steadily over time. By the end of the period considered, the weighted average of the answers was 2.7, which is not too far from the status quo. The perceived earnings loss thus appears to be smaller than the actual earnings loss. The question remains whether earnings 15 months after separation, either actual or perceived, provide an adequate measure of future earnings. For in- stance, studies suggest that displaced U.S. workers experience a drop in earnings at first, but a gradual recovery over time (see Hamermesh 1989; Topel 1990; Jacobson, LaLonde, and Sullivan 1993). In the case of BCE employees, the analy- sis in section VI indicates a similar trend in earnings, at least for those who had no choice, but this recovery takes place at declining rates and is negligible by the time of the survey. More important, the recovery can be expected to be partial anyway because pay at BCE was probably out of line with private sector pay to begin with. An important rent component in BCE salaries would make it difficult for displaced employees to find similar deals subsequently. Therefore we assume that the losses from displacement had materialized fully by the time of the survey. V. THE ACTUAL EARNINGS Loss The most straightforward indicator of the loss experienced by displaced em- ployees is the change in their annual earnings, excluding returns from invested compensation. This indicator could be criticized on the grounds that earnings before separation do not provide an appropriate counterfactual. A case could be made that the appropriate comparison is with the earnings these employees would have received had there been no downsizing. If the situation prior to downsizing was unsustainable, it could be argued, earnings would have declined in any event. Alternatively, if the situation was sustainable, some of these employees would Table 3. Earnings Following Separation Compared with Previous Earnings from the Central Bank of Ecuador, by Month after Separation, 1994-95 (percentage of former employees) Relative earningsa 1 2 3 4 5 6 7 8 9 10 11 12 13 14 1.5 Much higher (5) 19.5 16.3 13.5 12.2 10.8 10.8 11.3 10.7 10 5 9.9 8.3 8.0 8.9 8.1 7.4 Higher (4) 21.0 22.9 23.4 23.8 23.3 23.2 23.4 22.9 22.4 22.1 21.9 22.2 21.2 21.1 22.0 Similar (3) 18.4 18.2 2(0.1 213 21.6 21.8 22.0 242 24.4 25.4 26.9 27.1 26.7 26.4 24.5 Lower (2) 26.2 27.5 27.5 27.6 28 5 27.3 27.8 27.3 27.7 27.6 26.3 26 0 25.6 26.7 28.7 Much lower (1) 11.2 12.1 12 4 12.2 12.5 13.0 11.3 10.5 10.5 11.0 12.7 12.7 13 1 13.2 12.1 No earnings (0) 3.7 3.0 3.0 3.0 3.3 3.9 4.1 4.4 4.4 3 9 3.9 3.9 4.5 4.5 5.3 Total 100.0 1(0.0 100.0 100.0 100.0 100.0 100.0 100.0 100 0 100.0 100.0 100.0 100.0 100.0 100.0 Sample size (number of individuals) 267 363 363 362 361 362 363 363 361 362 361 361 359 356 282 Note The sample size is smaller at the beginning of the period because several interviewees saw month 1 as a transition point and did not answer questions about it. The sample declines at the end of the period because the first interviews were carried out during month 15. a. The value in parentheses is the rank given to the category of relative earnings. Source: Authors' calculations. Rama and MacIsaac 103 have gotten pay raises and promotions in the 15 months elapsed since separa- tion. More generally, the appropriate comparison would be between the lifetime earnings profile after separation from BCE and the corresponding earnings pro- file in the case of no separation. But this comparison would require heroic as- sumptions, so that it is safer to stick to observed earnings before and after separation. We define the actual earnings loss of individual i (AEL,) as: (1) AEL, = log E' -log E' where EGis the average annual salary at BCE in the two years preceding separa- tion and EP represents annual labor earnings 15 to 18 months after separation. This indicator overestimates the loss in earnings when public sector employees hold multiple jobs, including moonlighting and daylighting activities. But mul- tiple job holding was not all that common among the relatively well-paid BCE employees. Note that when EG and Ep are not too far apart, AEL, can be inter- preted as a percentage change in earnings. Both EG and EP, are likely to depend on individual characteristics such as gender, education, and experience. Denot- ing the observable characteristic h of individual i as Xh,, we assume the follow- ing Mincerian specifications: (2) log EG = (XG + onG XI, + + aGk, + UG (3) log EP = c4P + cX1X, + . .. + c Xk, + u, where the u terms capture the effect of unobservable characteristics on earnings. Some relevant individual characteristics, such as talent or aversion to effort, are unfortunately unobservable. We assume that the effects of these characteris- tics on earnings before and after separation are correlated and write this hypoth- esis as: (4) uP= p UG +rE, where E, is a stochastic disturbance. Replacing above yields: (5) AEL, = O + fG log EG + P1X1, +. * * + fkXk, - with p, = p _ -Pifj G and G = I -p. The actual earnings loss is therefore a function of the set of individual character- istics usually included in Mincerian equations, augmented to include annual earn- ings before separation. Estimating equation 5 for displaced BCE employees would provide valuable information on the individual characteristics most likely to be associated with a significant earnings loss. However, this estimation raises an econometric prob- lem: almost half of the displaced employees had no labor earnings at the time of the survey. From a mathematical point of view, the earnings loss AEL, is not defined in their case because it would involve the log of zero. We address this 104 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 difficulty by using the percentage change in earnings (Ec - E G) / Eras a proxy for AEL,. A more fundamental concern is that the lack of earnings does not just represent another level of EP but rather reflects either unemployment or with- drawal from the labor force. It can therefore be argued that the change in earn- ings experienced by those who are out of a job does not belong to the linear equation above. To address this concern, we correct for self-selection by esti- mating the earnings loss conditional on having a job. Table 4 reports the results obtained when estimating equation 5 with and without a correction for self-selection.2 The correction is based on the Heckman selection model. Due to the limited number of observations available, we apply this correction to the full sample only. The coefficients of the underlying selec- tion model (not reported in the table) indicate that women, married employees, and those with more seniority in BCE are less likely to hold a job after separation than their otherwise identical fellows. However, the data reject the existence of a self-selection bias, as shown by the similarity of the coefficients in the first two columns in table 4 and by the lack of significance of the X value in the second column. This outcome is not surprising in a small sample like ours, but it leads us to focus on the results obtained without a correction for self-selection.3 Education is an individual characteristic that reduces the earnings loss signifi- cantly, according to table 4. Everything else being equal, the loss falls by roughly 3 percent for each additional year of schooling. But after controlling for educa- tion, the loss is not significantly affected by the level of salaries at BCE. The implicit value of parameter p is thus close to 1. By contrast, the earnings loss increases with work experience at BCE. In the Mincerian framework, work expe- rience could of course be seen as an indicator of age. However, experience out of BCE does not increase the earnings loss, which suggests that more than a mere age effect is involved. One possible interpretation of the gap between the two experience coefficients is that longer careers at BCE are associated with heavier investments in specific skills that have low returns in the private sector. Another possibility is that working for an overstaffed public sector agency entails a de- preciation of human capital. Finally, the earnings loss is larger for female and married employees, and it could be larger for those with more dependents as well. The effects of individual characteristics on the actual earnings loss have roughly the same signs for all displaced employees regardless of whether they had the 2. The statistical significance of the coefficients in table 4 is probably underestimated. This is because log EG is by assumption correlated with the other explanatory variables. In practice, the adjusted R2 coefficient of a regression of log EG on all the X5, variables is 0.69. However, the discussion in the text implies that omitting log EG from the regression would yield biased coefficients on the X5, variables. 3. The lack of adequate identifying variables further diminishes the relevance of the results obtained with a correction for self-selection. It is indeed difficult to claim that any of the variables in our survey are highly correlated with participation in the labor force but not with the extent of the earnings loss of those who do participate. For instance, the number of dependents is typically assumed to be an identifying variable in other studies dealing with labor force participation and earnings. But the results in tables 4 and 5 suggest that the number of dependents has a direct effect on earnings. Rama and Maclsaac 105 Table 4. Determinants of the Actual Earnings Loss of Former Employees of the Central Bank of Ecuador Full sample Ordinary least Heckman Had Had a Variable squares selection model no choicea choice' Independent term 1.3496 -1.3883 2.5765 -0.2610 (0.359) (-0.261) (0.709) (-0.037) Log of wage at BCE -0.0664 0.0799 -0.1293 0.0072 (-0.283) (0.242) (-0.570) (0.016) Schooling (years) -0.0324** -0.0414* -0.0251* -0.0370 (-2.471) (-2.106) (-1.954) (-1.503) Experience at BCE (years) 0.0404*a 0.0420' 0.0298* 0.0532* (2.550) (1.736) (1.943) (1.784) Previous experience (years) 0.0155 0.0253 0.0146 0.0182 (1.420) (1.506) (1.393) (0.882) Married (yes = 1) 0.1760* 0.1323 0.0670 0.2460 (1.789) (0.875) (0.685) (1.413) Female (yes = 1) 0.3294`- 0.2469" 0.2604*' 0.3782*' (4.041) (1.733) (3.115) (2.595) Number of dependents 0.0438 0.0655* 0.0275 0.0614 (1.571) (1.664) (1.025) (1.151) Had a choice (yes = 1) -0.1815k -0.2951V (-2.303) (-2.529) X value 0.0642 (0.224) Adjusted R2 0.1147 0.0851 0.1051 F value (X2 value for Heckman) 6.80 27.21 3.51 3.82 Sample size 359 359 190 169 *Significant at 10 percent I a Significant at 5 percent. Note: The table compares 1995 with 1992-93 at June 1995 prices. The dependent variable, actual earnings loss, is the wage at the Central Bank of Ecuador (BCE) minus earnings after separation as a proportion of the wage at BCE. t values are reported in parentheses. a Estimated using ordinary least squares. Source: Authors' calculations. choice of staying with BCE (table 4). But when we split the sample between vol- untary and involuntary separations, the hypothesis that the D coefficients multi- plying observable characteristics are the same for the two groups is rejected. Other things being equal, the loss is larger for those who were forced to leave, as shown by the negative and statistically significant coefficient on the dummy variable for choice. Taken at face value, this coefficient implies that the earnings loss of employees who had a choice is roughly 18 percentage points smaller than that of those who had no choice. VI. THE PERCEIVED EARNINGS Loss We use subjective assessments of household earnings after separation to con- struct a second indicator of the loss experienced by displaced BCE employees. As 106 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I discussed in section IV, these assessments yield a slightly different (more posi- tive) view of earnings after downsizing, compared with the actual figures. This difference could merely reflect measurement error in either or both of the two earnings indicators. In this case, subjective assessments allow us to evaluate the robustness of the results obtained based on actual earnings figures. But subjec- tive assessments may capture more accurately the employees' sentiments follow- ing downsizing. Thus this second indicator of the earnings loss may be as rel- evant as the first one, in spite of being less precise. We define the perceived earnings loss of household i in month t after separa- tion (PELZ,) based on the employee's assessment of household earnings in that month (PHE,,), as follows: (6) PEL, = 3 - PHE,1. By definition, PHE,t is equal to 0 in the case of no earnings and equal to 5 when current earnings are much higher than the salary at BCE. The case where PHE,7 = 3 corresponds to the status quo. As a result, PEL,, is an integer varying from 3 for very large losses to -2 for large gains, with 0 reflecting no change. Note that because of the way it is defined, PEL,t measures the loss relative to household earnings while still working for BCE and therefore bears some similarity to AEL,. By analogy with the analysis in section V, we assume the following relation- ship between PEL,t and the observable characteristics of individual i: (7) PEL,, = yo + yG log E + ylXI, + + yk XkX + yTJ t + yT2 t + v,. The right-hand-side variables in this relationship are the same ones used to ac- count for the actual earnings loss, plus a time trend and its square (t and t2, respectively). Indeed, many monthly observations are now available per em- ployee, so that it becomes possible to analyze how earnings losses evolve over time. However, the coefficients on the common regressors are not strictly com- parable across the two relationships because PEL,7 is not defined in the same way as AEL,. As a result, there is no direct link between the coefficients in the equation above and the ox coefficients in the Mincerian equations for earnings in and out of the public sector (equations 2 and 3). This means, in particular, that yG is not equal to 1 - p any more. Table 5 reports the estimation results for the determinants of the perceived earnings loss. Given the rejection of the self-selection model in the previous sec- tion, and given also the fact that only 5 percent or less of the employees report no household earnings in any particular month, the estimation was carried out by ordinary least squares. The coefficients in table 5 have a higher significance than those in table 4 due to the panel data aspect of the analysis. Not surpris- ingly, the signs of most of the coefficients are the same in the two tables. As before, the earnings loss is smaller for employees with more education. It is larger for those with longer careers at BCE, for female and married employees, and for those with more dependents, as well. The loss also appears to be larger for those who had no choice. There is, however, an important difference be- Rama and MacIsaac 107 tween the two tables. In table 5 there is a significantly negative association be- tween the perceived earnings loss and the salary while at BCE. This means that, everything else being equal, employees at the top of the BCE hierarchy lost less, in relative terms, than those at the bottom. For the sample as a whole, table 5 displays a gradual recovery in household earnings over time. However, trends are different depending on whether the employees had a choice. For those who had a choice, the coefficients on the time trend and its square are statistically insignificant. Their household earnings were thus fairly stable after separation. For those who had no choice, household earn- ings were lowest at separation and then increased gradually. If the obtained coefficients are taken literally, household earnings stabilized roughly one year Table 5. Determinants of the Perceived Earnings Loss of Former Employees of the Central Bank of Ecuador Full sample Without dummy With dummy Had Had a Variable for choice for choice no choice choice Independent term 20.7893*X 20.7380 * 19.0599"t 24.5693* ' (13.401) (13.368) (9.016) (10.711) Log of wage at BCE -1.1671'* -1.1625-* -1.0586'" -1.4025 *x (-12.062) (-12.012) (-8.028) (-9.812) Schooling (years) -0.0850*' -0.0848** x 0.0745*8 -0.0920'* (-15.649) (-15.605) (-9.912) (-11.486) Experience at BCE (years) 0.0391X 0.0390- 0.0249* 0.06294* (5.958) (5.940) (2.789) (6.456) Previous experience (years) 0.0071 0.0073 0.0068 0.0114* (1.581) (1.618) (1.105) (1.701) Married (yes = 1) 0 1629** 0.1562'* -0.0194 0.3462** (4.010) (3.830) (-0.326) (6.091) Female (yes = 1) 0.5859*r 0.5895'" 0.5259** 0.6854'- (17.379) (17.456) (10.757) (14.451) Number of dependents 0.1106X 0.1105 0.1416'* 0.0564" (9.588) (9.577) (9.004) (3.245) Time (in months) -0.0378** -0.0370'- -0.0469** -0.0249 (-2.498) (-2.446) (-2.282) (-1.114) Time squared 0.0014 0.0013 0.0020* 0.0006 (1.619) (1.555) (1.693) (0.434) Had a choice (yes = 1) -0.0570* (-1.741) Adjusted R2 0.1675 0.1678 0.1590 0.1892 F value 123.01 111.05 62.59 66.41 Sample size 5,458 5,458 2,934 2,524 Significant at 10 percent. * r Significant at 5 percent. Note: The table compares 1995 with 1992-93 at June 1995 prices The dependent variable is the perceived monthly loss in household earnings; it is an integer that can take a value between 3 (very large losses) and -2 (large gains). The estimates were obtained by ordinary least squares. t values are reported in parentheses Source: Authors' calculations. 108 THE WORLD BANK ECONOMIC REVIEW, VOL 13. NO I after separation. These different trends suggest that those who left voluntarily already had an alternative out of the public sector at the time of separation, whereas the rest of the employees were relatively unprepared to cope with their new situation. The time trends remain roughly unchanged if fixed-effect estimates are used (results are not reported here). However, most of the coefficients on the Xh vari- ables become statistically insignificant. Fixed-effects estimates are obtained by replacing the perceived earnings loss variable PEL,z with its deviation with re- spect to the individual mean (PEL, - PEL1). It follows that individual character- istics have an impact on the magnitude of the perceived earnings loss but not on its variation over time. VII. THE IMPLICIT WELFARE Loss The change in well-being experienced by displaced employees has dimensions that are not necessarily captured by changes in earnings, either actual or per- ceived. The total loss can be higher than suggested by the corresponding earn- ings loss, if more effort is needed to achieve the same living standard as before. It can also be higher if the variability of earnings increases as a result of displace- ment, or if the prospects for earnings growth over time are worse out of BCE than in it. The total loss can be lower, however, if the observed earnings loss reflects a deliberate withdrawal from the labor force. Last but not least, the impact of the earnings loss on well-being is mitigated by the compensation received. For all of these reasons, an evaluation of the consequences of displacement would be partial if it focused on earnings only. Rather than try to correct the earnings losses for these other dimensions of the change in well-being, in this article we generate a direct measure of the wel- fare loss ensuing from displacement. This measure is constructed from a subjec- tive assessment of well-being after separation. Of course, subjective assessments of this kind raise perennially thorny questions about the comparability of utility levels across individuals. But in spite of their limitations, they have been used fruitfully before. For instance, there is a growing literature on the links between happiness and employment status relying on indicators of psychological distress as a measure of disutility (see, for instance, Clark and Oswald 1994; Korpi 1997). In the present context, subjective evaluations of well-being can be used to iden- tify the individual characteristics that are more strongly associated with distress after job separation. Specifically, the surveyed employees were requested to evaluate their well- being relative to the time when they worked for BCE. The questionnaire explic- itly listed a series of elements that needed to be taken into account when answer- ing this question. The list included labor earnings and satisfaction at work if applicable, leisure time available, the possible implications of displacement for labor force participation and migration of any members of the household, and the amount of compensation received. The answer, hereafter identified as the Rama and MacIsaac 109 perceived well-being of the household (PHW1), is an integer ranging from 1 (when well-being is much lower than before separation) to 5 (when it is much higher). The case where PHW, = 3 corresponds to the status quo. By analogy with the perceived earnings loss, the perceived welfare loss of household i (PWL, ) is defined as: (8) PWL, = 3 - PHW,. By construction, PWL, is an integer varying from 2 (for a large loss) to -2 (for a large gain), with 0 reflecting no change. This variable can be linked to individual characteristics like those considered in the analysis of the determinants of the actual earnings loss. But the perceived loss of well-being is also affected by the amount of compensation received (C,). The following specification can there- fore be used: (9) PWL, = 60 + 3G log EG% + c log C, + 61 Xl + * * * + + kXk, + where coefficient 6c is expected to be negative. Unlike the actual or perceived earnings, which are calculated or reported based on monetary figures, the level of well-being cannot be measured objectively. Consequently, the distance between adjacent values of PHW, is not necessarily constant. For instance, the gap in household well-being between 1 and 2 could be much larger, or much smaller, than the gap between 4 and 5. The problem is the same with PWL,. An ordered logit regression is therefore warranted to esti- mate the coefficients in equation 9. In practice, however, this estimate yields almost identical values for all of the distances between adjacent levels of PWL2, while reducing the significance of the estimated coefficients compared with an ordinary least squares estimate. Table 6 reports the results obtained using ordinary least squares. The statisti- cal significance of the coefficients may be more seriously underestimated than in the previous two sections, due to the inclusion of log C, among the regressors. Severance pay was based on salary and seniority in BCE, so that the correlation among right-hand-side variables is potentially very high. Fortunately, there were upper and lower bounds on the amount of severance, and the latter was only one of the components of the compensation package. Still, the adjusted R2 of a regression between log C, and the other regressors is 0.82. The main difficulty in interpreting the coefficients in table 6 stems from the link between perceived well-being and the amount of compensation received. Because separated employees were compensated, their net welfare loss underes- timates the impact that displacement, on its own, had on their well-being. This impact is called the implicit welfare loss in what follows. Note that from a policy perspective, it is relevant to know how the implicit welfare loss varies with the individual characteristics of the employees. This information could help in the design of assistance strategies for displaced employees elsewhere. Knowing how the net welfare loss varies with individual characteristics is less relevant because their observed effects depend on the rules used to set up the compensation pack- 110 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Table 6. Determinants of the Perceived Loss of Well-Being of Former Employees of the Central Bank of Ecuador Full sample Without dummy With dummy Had Had a Variable for choice for choice no choice choice Independent term 18.5153 '{ 18.1507* 14.8293t 20.7632' k (3.642) (3 593) (2.180) (2.665) Log of wage at BCE -0.4945 -0 4910 -0.2872 -0.7764 (-1.415) (-1.414) (-0.646) (-1.384) Schooling (years) -0.0841 -0.08331 -0.0824'* -0.0798*^ (-5.139) (-5.126) (-3.784) (-3.156) Experience at BCE (years) 0.0711' 0.0694k 0.0621X 0 0726* (2.895) (2.841) (1 927) (1.863) Previous experience (years) 0.0249' 0.0251' 0 0128 0.0377 (1 689) (1.716) (0 665) (1.635) Married (yes = 1) 0.0100 -0.0178 -0.1786 0.1141 (0.082) (-0.146) (-1 031) (0.638) Female (yes = 1) -0.0757 -0.0613 -0.0140 -0.1370 (-0.746) (-0.607) (-0.099) (-0.898) Number of dependents 0.0800'- 0.0794' 0.0641 0.1022' (2 301) (2.299) (1.405) (1.851) Log of compensation received -0.5723' -0.5477: -0.5411 -0 4492 (-1.896) (-1.825) (-1.401) (-0 912) Had a choice (yes = 1) -0.2328'- (-2.385) Adjusted R2 0.1521 0 1633 0.1225 0.1524 F value 9.029 8 765 4.298 4.777 Sample size 359 359 190 169 Significant at 10 percent - * Significant at 5 percent Note: The table compares 1995 with 1992-93 at June 1995 prices. The dependent variable is perceived loss of well-being; it is an integer that can take a value between 2 (large losses) and -2 (large gains). The estimates were obtained by ordinary least squares t values are reported in parentheses. Source: Authors' calculations. ages, and the specifics of those rules are likely to vary from one downsizing operation to another. Fortunately, the implicit welfare loss of separated employees can be inferred from their perceived welfare loss. A conceptual experiment may be useful here. Imagine that there is an individual whose compensation package C, exactly off- sets the impact of displacement on well-being, so that there is no perceived wel- fare loss (that is, PWL, = 0). For this particular individual, the amount of com- pensation received provides a monetary measure of the implicit welfare loss. Of course, there may not be many individuals in the sample for whom PWL, = 0, so that this measure is generally not available. But the estimated values of the 6 coefficients allow us to predict the amount of compensation that would have been needed to keep the level of well-being of the separated employees unchanged, given their observable characteristics. In analytical terms, we define the implicit welfare loss of individual z (IWL,) Rama and MacIsaac 111 as the value of log C, for which the predicted value of the perceived welfare loss PWL, is equal to 0, which implies: (10) IWL, = 00 + OG log E+ 01Xj, + * . * + ok Xk, with 0,= -6, I - In this equation 8, stands for the estimated value of 8,. Table 7 reports the result- ing 0 coefficients. Note that their interpretation is similar to that of the P and y coefficients in the expressions of the actual and perceived changes in earnings, respectively. But their absolute values are not strictly comparable because the left-hand-side variables are different. As before, more-educated employees fare better than their otherwise identical fellows. If interpreted literally, the coefficients in table 7 mean that the implicit welfare loss decreases by roughly 15 percent for each additional year of school- ing. The loss is larger for those employees with more experience, particularly at BCE. It is also larger for those with more dependents. And it could be about 40 percentage points larger for those who had no choice. But the implicit welfare loss does not appear to be influenced by the employee's salary or gender. The lack of significance of the gender variable is worth stressing. The analysis of the determinants of earnings losses, in tables 4 and 5, suggests that women were more severely affected by downsizing than men. But the results in table 7 implic- itly confirm that the fall in female earnings was due to a deliberate decision to withdraw from the labor force. Table 7. Determinants of the Implicit Welfare Loss of Former Employees of the Central Bank of Ecuador Full sample Without dummy With dummy Had Had a Variable for choice for choice no choice choice Independent term 32.3498 -'* 33.1403** 27.4056*" 46.2278", Log of wage at BCE -0 8650 -0.8965 -0.5308 -1.7287 Schooling (years) -0.1469*' " -0.1521*- -0.1523*.* -0.1776*- Experience at BCE (years) 0.1242' 0.1266- 0.1147*' 0.1616*' Previous experience (years) 0.0435- 0.0459- 0.0237 0.0840 Married (yes = 1) 0.0175 -0.0325 -0.3301 0.2541 Female (yes = 1) -0.1326 -0.1119 -0.0259 -0.3051 Number of dependents 0.1398- 0.1450 0.1184 0.2274* Had a choice (yes = 1) -0.4251*k Significant at 10 percent. ' Significant at 5 percent. Note- The table compares 1995 with 1992-93 at June 1995 prices. The explained variable is the log of compensation required for no perceived loss of well-being. The calculations are based on the coefficients in table 6 The implicit welfare loss is defined as the log of compensation for which the predicted value of the perceived loss of well-being is equal to zero. The significance of the corresponding coefficients in table 6 is reproduced here. Source: Authors' calculations. 112 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Ex post regret at having left BCE provides yet another way of assessing the welfare loss from displacement, at least in principle. All of the surveyed employ- ees were asked whether, in retrospect, they would have preferred to stay with BCE. Their answer to this question can be assumed to be a function of the same observable characteristics considered before. It can also be expected to depend on the amount of compensation received. If such a function could be estimated, it would be possible to predict, for each individual in the sample, the amount of compensation that would have achieved ex post indifference. Such an amount, in turn, would provide another measure of the implicit welfare loss. A binary dependent variable model was thus estimated for the ex post regret variable, with the right-hand-side variables being the same as for the perceived loss of well-being. Unfortunately, the overall fit of the regression is poor, except in the case of the employees who had a choice. The estimated coefficients are therefore not reliable enough to make inferences about the amount of compensation that would have left displaced BCE employees indifferent between staying and leaving. VIII. POLICY IMPLICATIONS In spite of the small sample size, our analysis of the changes in earnings and welfare experienced by displaced BCE employees has potential implications for downsizing operations elsewhere. These implications refer to the design of cost- effective ways to support displaced workers. To some extent, our analysis could also be interpreted as an ex post evaluation of downsizing at BCE. But such an evaluation would be intrinsically partial, for two reasons. First, when the downsizing operation was decided on, the management of BCE did not have the kind of hindsight that our survey gives us. It would certainly be unfair to judge the decisions it made using information that was not available at that time. And second, our analysis can only tell us how the displaced employees fared after separation. It cannot tell us which among all BCE employees should have been separated in the first place. A warning from our analysis concerns the potential of training programs to facilitate the redeployment process. In the case of BCE, only 12 percent of the displaced employees took any of the courses offered, although these were pro- vided free of charge. This finding is similar to those from other studies (see Lindauer and Nunberg 1994) and runs against the belief that training programs must be an important component of downsizing operations. These programs tend to be costly and may become easy prey for ailing training agencies. If sepa- rated workers shun them, resources could be better spent on things they value, namely, direct compensation. Other policy implications from our analysis are related to the appropriate design of compensation packages. Our analysis highlights the links between the observable characteristics of public sector workers and the extent of the losses they experience as a result of displacement. The ability to predict those losses Rama and MacIsaac 113 may be particularly important in cases where involuntary separations are pre- cluded by law. It may also be important when compensation is required due to fairness considerations or as a matter of political feasibility. No valid reasons exist, however, for overcompensating displaced workers. The results in this ar- ticle could therefore be useful in mitigating some of the overcompensation bias that characterizes downsizing programs based on voluntary job separations. A result obtained consistently across the study is the negative association be- tween schooling and losses from displacement. Employees with higher levels of education experience smaller earnings and welfare losses than their otherwise identical fellows. Severance pay packages that fail to take schooling into ac- count may therefore lead to a public sector brain drain. This possibility can only be exacerbated when the amount of compensation offered is a multiple of the public sector salary.4 Our results indicate that after controlling for other indi- vidual characteristics, the losses from displacement (in relative terms) are at best independent of the level of salaries in the public sector. If anything, there is a negative association between salaries and relative losses. A compensation pack- age that increases proportionally with the salary is more attractive for employ- ees who are at the top of the public sector ladder than for those at the bottom, whereas redundancies tend to be more prevalent at the bottom. However, the results show that the losses from displacement do increase with seniority in the public sector, as is assumed by the standard rules of thumb for compensation. Another result that deserves attention concerns gender. At first glance, women lose more than men from displacement. As a percentage of public sector salaries, their earnings losses appear to exceed those of men by about 30 points. But the gap is partly explained by the withdrawal of many female employees from the labor force after separation. Employees who withdraw from the labor force ex- perience a drop in earnings, but not necessarily a large reduction in welfare. Table 8 reports a low correlation between actual earnings loss and implicit wel- fare loss, particularly for employees who had a choice. The implicit welfare loss, in fact, does not vary significantly with gender. Because compensation packages aim at offsetting welfare losses rather than earnings losses, our results suggest that the packages should be similar for men and women. A final policy implication of our analysis is that a poor tailoring of compensa- tion to individual characteristics can be both unfair and expensive. As shown in table 8, the correlation coefficient between the compensation received and the actual earnings loss is low, particularly in the case of displaced employees who had a choice. The correlation coefficient between the compensation received and the implicit welfare loss is even lower. In the context of involuntary separa- tions, these low correlation coefficients imply that some employees are penal- ized (or rewarded) much more than others by the downsizing process. The per- ceived unfairness of this process can, in turn, create resistance to further public sector restructuring. In a context of voluntary separations, moreover, the low 4. Whether education and other individual characteristics can actually be used to tailor compensation packages depends, of course, on legal, political, and social considerations specific to each country. 114 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Table 8. Correlation between Loss Indicators for Former Employees of the Central Bank of Ecuador Actual Perceived Implicit Indicator earnings loss earnings loss welfare loss Had no choice Actual compensation received 0.4448 -0.0161 0.3134 Actual earnings loss 1.0000 0.3822 0.2540 Perceived earnings loss 0.3822 1.0000 0.1937 Had a choice Actual compensation received 0.3627 0.0005 0.2229 Actual earnings loss 1.0000 0.3186 0.1506 Perceived welfare loss 0.3186 1.0000 0.1697 Note: The table compares 1995 with 1992-93 at June 1995 prices. Actual compensation received includes returned contributions to old-age pensions and to unemployment insurance, in addition to severance pay. The perceived earnings loss is the individual average over the months following separation. The implicit welfare loss is calculated based on the coefficients in table 7. Source: Authors' calculations. correlation between compensation and losses may substantially increase the cost of downsizing operations. Assume for a moment that employees have accurate expectations about their prospects out of the public sector. If this is so, those for whom compensation exceeds the expected welfare loss will leave the public sec- tor, while the others will stay. The former mistake will thus lead to overspend- ing, while the latter will not achieve any savings because the separation offer will be rejected. Table 9 compares the actual cost of BCE downsizing with the cost of paying each displaced employee the equivalent of his or her implicit welfare loss. The two costs turn out to be similar for the employees who had no choice. But the cost of downsizing could have been smaller in the case of those who had a choice, had the compensation packages been better tailored. On average, the actual compensation received by these employees, including refunds of past contributions to social se- Table 9. The Cost of Alternative Compensation Rules (average per employee) In millions As a multiple As a percentage of sucres at of the wage of actual Indicator June 1995 prices at the BCE compensation paid Had no choice Actual compensation received 50.89 2.26 100.0 Implicit welfare loss 55.63 2.47 109.3 Had a choice Actual compensation received 51.58 2.29 100.0 Implicit welfare loss 41.78 1.86 81.0 Note: The implicit welfare loss was calculated based on the coefficients in table 7. Actual compensation received includes returned contributions to old-age pensions and to unemployment insurance, in addition to severance pay. Source: Authors' calculations Rama and MacIsaac 115 curity, amounted to 2.3 years of salary. Note that 2.3 years is the average payback period of downsizing operations across the episodes surveyed by Haltiwanger and Singh (this issue). Our estimates suggest that an average of 1.9 years would have been enough to compensate appropriately for the ensuing welfare losses. The total cost of the voluntary component of the downsizing operation would then have been about 20 percent lower than it actually was. IX. CONCLUSIONS Public sector downsizing will almost certainly become a major policy endeavor in the coming years. In the case of developing countries, efforts at trimming the payroll will affect highly organized and vocal groups of public sector workers, thus giving a prominent role to compensation issues. Yet, relatively little is known about the nature and extent of the losses these workers may experience. Given the paucity of information, and given also the need to defuse the potential resis- tance to restructuring, governments around the world may be tempted by the "golden handshake" approach. A problem with this approach is that it may exacerbate the adverse selection problem that plagues most downsizing pro- grams, whereby competent and hard-working employees leave the public sector, while those who are a real drag on efficiency stay. But even if the only separated workers are those who are genuinely redundant, golden handshakes could still significantly increase the cost of public sector restructuring. Typically, a major downsizing program has a price tag in the hundreds of millions of dollars. If ignorance about the losses that displaced workers may experience leads to a "mere" 10 percent premium on the compensation offered, the total cost may increase by several tens of millions of dollars. All of this suggests that a careful assessment of the losses that displaced workers may experience is warranted. In this article we analyzed one particular episode of downsizing, but the find- ings may have implications for other cases as well. First, we showed that in spite of their popularity, training programs may be ineffective in assisting displaced employees. Second, we showed that both earnings and welfare losses are corre- lated with a variety of observable individual characteristics, such as education or the number of dependents, and not just with salary and seniority in the public sector, which are the two variables considered by the standard rules of thumb used for compensation. These rules implicitly assume that the loss from dis- placement increases with seniority, which is supported by our analysis. But they also assume that the loss decreases with salary, which is rejected. Finally, we showed that the savings from better-tailored compensation can be substantial, particularly in the case of public sector workers who are offered the option of staying in the public sector. The particular episode we analyzed was quite remarkable, in the sense that overspending was moderate in the case of the employees who had a choice and slightly negative in the case of those who had no choice. It is probable that the management of BCE, rather than luck, deserves the credit for this relatively good 116 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 performance. Unlike the first phase of the downsizing process, the second phase was indeed characterized by attention to details. And it would appear that the selection process to distinguish the employees who were dispensable from those who were essential might have been successful as well. But it would be a mistake to assume that the second phase of BCE downsizing can be easily replicated else- where. In the absence of careful preparation, the standard downsizing operation is more likely to resemble the first phase. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Alderman, Harold, Sudarshan Canagarajah, and Stephen Younger. 1996. "A Compari- son of Ghanaian Civil Servants' Earnings before and after Retrenchment." Journal of African Economies 4(2):259-88. Clark, Andrew E., and Andrew J. Oswald. 1994. "Unhappiness and Unemployment." Economic Journal 104(424):648-59. Hamermesh, Daniel S. 1989. "What Do We Know about Worker Displacement in the U.S.?" Industrial Relations 28(Winter):51-59. Jacobson, Louis S., Robert J. LaLonde, and Daniel G. Sullivan. 1993. "Earnings Losses of Displaced Workers." American Economic Review 83(4):685-709. Korpi, Tomas. 1997. "Is Utility Related to Employment Status? Employment, Unem- ployment, Labor Market Policies, and Subjective Well-Being among Swedish Youth." Labour Economics 4(2):125-47. LImdauer, David L., and Barbara Nunberg, eds. 1994. Rehabilitating Government: Pay and Employment Reform in Africa. World Bank Regional and Sectoral Study. Wash- ington, D.C.: World Bank. MacIsaac, Donna, and Martin Rama. 1997. "Do Labor Market Regulations Affect La- bor Earnings in Ecuador?" Journal of Labor Economics 15(3, pt. 2):136-65. Robbins, Donald. 1996. "Public Sector Retrenchment and Efficient Severance Pay Schemes: A Case Study of Argentina." Harvard Institute for International Develop- ment, Cambridge, Mass. Processed. Tansel, Aysit. 1997. "Public Sector Retrenchment and the Impact of Labor Shedding Programs on Workers in Turkey." Department of Economics, Middle East Technical University, Ankara. Processed. Topel, Robert. 1990. "Specific Capital and Unemployment: Measuring the Costs and Consequences of Job Loss." Carnegie-Rochester Conference Series on Public Policy 33(0):181-214. THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 117-53 Matching Severance Payments with Worker Losses in the Egyptian Public Sector Ragui Assaad Severance pay programs can reduce political opposition and minimize the social costs of labor redundancies. In Egypt, only voluntary programs are feasible because legal limitations preclude layoffs and strong organized labor groups oppose any weakening of job security protections. A common problem with voluntary severance programs, however, is that they tend to overpay workers relative to the welfare losses they expe- rience from displacement. This article estimates the losses that public sector workers would incur if they were displaced from their jobs and simulates several voluntary severance schemes to deter- mine how well the schemes match compensation payments to these estimated losses. It provides a fairly strong argument for looking at the structure of opportunity costs and wage profiles when designing severance programs It shows that significant overpay- ment can be avoided by matching compensation payments to the expected losses of workers. It also provides a method for estimating these losses from standard labor force surveys that are available in most countries. Public sector restructuring and privatization have caused substantial labor re- trenchment in many countries, and displaced workers often suffer important welfare losses. Governments have used a variety of mechanisms to address the political and social costs of large-scale displacements, including employment guarantees, retraining programs, job search assistance for displaced workers, and severance pay. A recent survey has shown that severance pay programs are one of the most effective methods for reducing political opposition and minimiz- ing the social costs of labor redundancies (Kikeri 1996). In some cases, like the case of Egypt examined here, only voluntary programs are feasible because legal limitations preclude layoffs and strong organized labor groups oppose any weak- ening of job security protections. A common problem with voluntary severance programs, however, is that they tend to overpay workers relative to the welfare losses the workers experience from displacement. Therefore the programs can be quite expensive (Rama in this issue). Some overpayment is inevitable in voluntary programs because per- Ragui Assaad is with the Humphrey Institute of Public Affairs at the University of Minnesota. This is the revised version of a paper presented at the World Bank conference on Public Sector Retrenchment and Efficient Compensation Schemes held in November 1996. The author thanks Ishac Diwan, Martin Rama, partcipants in the World Bank conference on Public Sector Retrenchment and Efficient Compensation Schemes, and three anonymous referees for many useful comments and suggestions, as well as Hin Kin Lam for research assistance. C) 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 117 1 18 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I fect matching of compensation to the unobserved worker-specific losses is im- possible and undercompensation would result in insufficient exits. Public au- thorities can reduce the extent of overpayment by using observable worker char- acteristics to match the compensation amount to individual-specific losses. This article compares the earnings of workers in and out of the public enterprise sector, while taking account of differences in nonwage benefits and nonrandom sector selection. It relates worker losses to observable characteristics such as seniority, age, years of overall labor market experience, educational attainment, and gender. It assesses how well alternative redundancy pay formulas typically used in severance programs match compensation payments to these estimated losses. I calibrate the parameters for several formulas to achieve a given rate of exit at a minimum cost per exiting worker. Finally, I analyze the effect of the choice of formula on the composition of exiters and stayers. Minimizing compensation payments to workers is not the only desirable fea- ture of voluntary severance programs. Severance programs also may need to achieve a certain mix of qualifications and occupations among workers who remain in the public sector so as to minimize the need to rehire workers. Sever- ance programs that simply achieve a given exit rate at minimum cost generally do not yield the desired mix. For instance, in the Egyptian case, workers with less than secondary education have significantly lower displacement costs than workers with secondary or university education. The lowest-cost program would therefore aim to exit all the less-educated workers first. This is unlikely to be desirable from the point of view of labor force composition. A well-designed program would need to set separate target rates for the redundancy of workers in different educational or occupational categories and to achieve these rates by offering different compensation packages to workers in each category. Determining the rate of redundancy for specific categories of workers in Egyp- tian state-owned enterprises (SOES) is beyond the scope of this study. I simply assume a 30 percent rate that applies equally to workers in three broad educa- tional categories. Within each category, workers who opt to accept the severance package will self-select based on their relative prospects in the private sector. This article estimates the welfare losses to displaced workers in the government and in SOES in Egypt. It then uses these individual-specific estimates to assess how well standard severance pay formulas minimize the overpayment to work- ers. Standard formulas typically index payments on seniority, the worker's wage at separation, years of denied service, and in some cases age (see Nunberg 1994 and Kikeri 1996). To get quantitative estimates of the extent of overpayment, I perform simulations on the SOE workers in the October 1988 labor force sample survey. The simulations attempt to achieve a given exit rate at the lowest fiscal cost by attempting to exit the workers with the lowest welfare losses first. A worker is assumed to exit voluntarily if the payment offered exceeds the worker's displacement losses. The parameters of the standard severance pay formulas are optimized to achieve the desired exit rate at the lowest possible fiscal cost using each formula. The analysis then compares the performance of the different for- Assaad 119 mulas in minimizing costs. The simulations also generate the composition of exiters and stayers in terms of observable characteristics. The goal of exiting first the workers with the lowest welfare losses from dis- placement, the main measure of performance here, may in fact worsen the ad- verse selection problem. Workers who have the best reemployment prospects in the private sector tend to be the first to accept the severance package. They also are likely to be the most motivated and productive in the public sector. There are ways to reduce such adverse selection, but all of them involve some increase in the cost of the program. For instance, the severance payment can be raised to achieve more voluntary exits than required. The right to accept the severance package can then be rationed to prevent the exit of some workers. This can be done either by vetoing the exit of workers whose performance exceeds a certain standard or, if this is deemed unfair, through some sort of randomization pro- cess. The random allocation would not prevent all high-quality workers from exiting, but it would make sure that some of them remain (Diwan 1993a, 1993b). Other mechanisms to make workers reveal private information about themselves have been proposed, such as sealed auctions and menus (Levy and McLean 1997; Rama in this issue). Here I focus on the overpayment problem and abstract from the issue of adverse selection. Other important issues not addressed in this article relate to the design of voluntary severance schemes. These include the form that compensation should take, whether it should be a lump-sum payment, an annuity, or an in-kind pay- ment such as retraining assistance, or whether workers should be able to choose from a menu of options. I also do not deal with issues relating to the timing and speed of the retrenchment program, the need for reforms in public sector pay and management after retrenchment takes place, or ways to reduce deadweight losses from the fact that some workers would leave anyway. Such deadweight losses are likely to be small in Egypt because historically the public sector has had a very low rate of turnover. Most workers who wish to work in the private sector simply engage in moonlighting while remaining on the public sector pay- roll. The value that tolerance for moonlighting adds to public sector employ- ment is implicitly taken into account in my estimate of the nonwage benefits of public employment. Section I outlines the estimation strategy pursued in this article. Section II develops and implements the methodology for estimating worker-specific dis- placement losses. Section III runs simulations of the alternative severance pay schemes to see how well they match compensation payments to the estimated worker losses. I. AN OPPORTUNITY COST APPROACH TO ESTIMATING THE LOSSES OF DISPLACED WORKERS The literature has proposed several empirical strategies to relate the losses of displaced workers to individual-specific characteristics. These approaches gen- 120 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 erally involve a comparison of workers' earnings before and after displacement (Alderman, Canagarajah, and Younger 1996; Younger 1996; Tansel 1997; and Rama and MacIsaac in this issue) or a comparison of the characteristics of stayers and leavers after retrenchment has taken place (Robbins 1997). Both of these approaches must be done ex post and therefore require previous experience with public sector retrenchment for the country in question. The approach pursued in this article produces estimates of anticipated losses from displacement by com- paring the earnings of public sector workers with the opportunity cost of their labor in the private sector. For some female workers, the appropriate compari- son may be between their public sector wages and their reservation wage for market work, which may be higher than their private sector wage. Estimates of welfare losses based on wage comparisons across the two sectors may thus over- state the losses of female workers. In making comparisons, I take into account the nonrandom selection of workers into the public sector and the potential difference in the nonwage attributes of public and private sector jobs. The advantage of this approach is that it relies on data from standard labor force surveys that are commonly available in many countries. I use data from the October 1988 Egyptian Labor Force Sample Sur- vey to estimate the potential welfare losses for workers who could be displaced by privatization. Welfare losses due to displacement can be classified into three parts: loss of earnings due to transitional unemployment while searching for a private sector job, permanent loss of earnings associated with moving to lower-paying jobs in the private sector, and loss of nonwage benefits associated with a public sector job, including intangible benefits like greater job security and lower levels of effort (Rama in this issue). In voluntary severance programs, where workers are given some flexibility in choosing the timing of exit within a fairly broad window, it is safe to assume that most job search occurs while workers are still in public sector jobs. Under such circumstances, transition costs would be fairly small compared with permanent losses. For simplicity, this analysis neglects transition costs. The analysis estimates the permanent loss of earnings due to displacement by comparing the expected earnings of workers in the public sector with those of similar workers in the private sector. I estimate selectivity-corrected earnings equations for workers in and out of the public sector. I then calculate differences in discounted streams of earnings from the time of displacement until retirement as a function of observed worker characteristics such as seniority, overall labor market experience, education, and gender. In theory, there could be downward shifts in the private sector wage schedule as a result of supply shocks from large-scale exits from the public sector. I ne- glect such supply effects because exits are likely to take place over a fairly long period of time, so that the impact on total labor supply is limited. The total number of public enterprise workers to be retrenched under the 30 percent re- dundancy assumption (about 400,000), if spread out over several years, is fairly small compared with the annual increment to the labor force in Egypt (about Assaad 121 half a million workers). In addition, any wage impact on the private sector is likely to be short-lived. Some of the permanent losses from displacement result from the higher ex- pected unemployment and greater job instability in the private sector that the worker will experience over the long run. In fact, a sizable fraction (34 percent) of workers in the private sector sample are employed only intermittently, whereas none of the public sector workers are. Earnings equation estimates (not shown here) indicate that workers employed intermittently earn 50 percent less per year than private sector workers employed regularly. I assume that displaced public sector workers are as likely to end up in intermittent employment as are similar workers currently in the private sector. Subject to this assumption, earn- ings equation estimates that include intermittent workers in the private sector sample but do not correct for such a status automatically incorporate potential earnings losses due to employment instability. Workers considering whether to accept a compensation package would look at differences in both the pecuniary and nonpecuniary attributes of jobs in the public and private sectors. To estimate the value that workers would place on such differences, I make four assumptions. First, I assume that workers would not enter the public sector unless their anticipated lifetime compensation in that sector, including these nonpecuniary rewards, was at least as high as what they could get in the private sector. Because relative wages in the public and private sectors could have shifted since some workers made their decisions to join the public sector, I limit this part of the analysis to relatively young public sector workers (age 35 and under). Second, I assume that public sector workers face a uniform discount rate of 5 percent. Third, because queuing for public sector jobs is ubiquitous, I assume that most public sector workers receive higher lifetime compensations in that sector than the opportunity cost of their labor in the private sector. However, for at least a marginal group of workers, these rents are close to zero. To identify the marginal group of workers who receive no rents, I classify the sample of public sector workers into 12 groups based on observable characteristics-gender, edu- cation, and whether they work in SOEs or the government. I then compare the ratio of private to public discounted lifetime monetary earnings for each of the groups. The group with the highest ratio in favor of the private sector is as- sumed to be the marginal group that dissipates its rents first. By equating total lifetime compensation in the public and private sector career paths for this mar- ginal group, I obtain an estimate of the ratio of total compensation to monetary compensation. In calculating the earnings stream in each career path, I take into account the time spent in the public sector job queue. The employment guarantee for gradu- ates that has been in effect in Egypt since 1964 entitles graduates of vocational secondary schools, technical institutes (equivalent to two-year colleges), and universities to a government job after a certain waiting period (see Assaad 1997). As of 1988, the last cohort of graduates to be offered government appointments 122 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I under the employment guarantee was the 1982 cohort for university graduates and the 1981 cohort for secondary and technical institute graduates. An as- sumption that queuing is costly does not mean that applicants have to remain unemployed while queuing, but it does mean that they earn less than workers who are not queuing. Because workers with less than secondary education are not guaranteed public sector jobs, I assume that they do not engage in costly queuing for such jobs. (See Assaad 1997 for a more detailed discussion of the workings of the public sector queue.) Fourth, I assume that nonwage benefits in the public sector are proportional to monetary earnings and that the constant of proportionality is invariant to worker characteristics and is equal across government and SOE employment. The assumption that the value of the nonwage attributes of public sector jobs is proportional to monetary earnings can be justified as follows. The most impor- tant nonwage aspects of public sector jobs are the higher probability of receiving a retirement pension and paid vacations and the lower effort required relative to private sector jobs. The benefits derived from these nonwage job attributes are therefore either directly related to monetary remuneration or depend on the value of a worker's time, which relates them indirectly to wages. Because some nonwage benefits, such as health insurance, are clearly independent of wages, in a sensitivity analysis, I consider the assumption that the value of the nonwage aspects of public sector jobs over a worker's lifetime is a constant absolute amount for all workers rather than a constant multiple of monetary compensation. Because the estimation of the ratio of nonwage benefits to earnings depends on several assumed parameters, including the length and cost of queuing, the discount rate, and the age cutoff used to identify new entrants, I conduct exten- sive analyses of the sensitivity of the estimates to changes in these parameters. The need to rely on the observed heterogeneity in the sample to identify the marginal group of workers, however, makes it impossible to test the assumption that the ratio is invariant to worker characteristics. I do entertain the possibility that nonwage benefits are constant across workers and see the extent to which such a pattern of benefits across workers alters the simulation results conducted in section III. II. ESTIMATION OF DISPLACEMENT LosSES This section uses estimates of selectivity-corrected earnings equations to ob- tain the expected earnings profiles of workers in the government, SOEs, and the private sector. It estimates nonwage benefits in the public sector and individual- specific displacement losses. It then analyzes the sensitivity of the loss estimates to the estimate of nonwage benefits. Expected Wages The earnings equation estimates used to predict wages are reduced-form equa- tions based on a standard Mincerian model. I correct for selectivity using a stan- Assaad 123 dard Heckman-type two-stage model with a multinomial logit selection rule that predicts the probability of selection into the government, SOES, and the private sector.1 Appendix table A-1 shows the selection equation estimates. The selec- tion equation includes several variables on the worker's parental background and marital status for all workers, as well as the number of children and employ- ment characteristics of male family members for female workers. The exclusion of these variables from the earnings equations helps identify these equations in the second stage. Assaad (1997) provides a more detailed discussion of the iden- tification issue in a similar context. The earnings equations include a tenure variable to account for seniority- based wage-setting rules in the government and SOEs. Tenure is calculated as the time since joining the public sector for government and SOE workers and the time since the last job change for private sector workers. This definition takes into account the fact that public sector workers can transport their seniority level across public sector jobs. Experience is calculated as the time since entry into the labor market and may therefore include a period of un- employment at entry. Because of the way in which data are collected, educa- tion is specified as the attainment of specific educational credentials rather than years of schooling. In the subsequent analysis, level-one workers have less than secondary education, level-two workers are graduates of general and vocational secondary schools and technical institutes, and level-three workers are university graduates. Occupation is not taken into account explicitly, but a rough division between blue-collar and white-collar employment is implied by the education variables. In Egypt, individuals educated up to and including the preparatory level can be assumed to be engaged in blue-collar occupations. Those educated at the tech- nical institute, university, and general secondary levels can be assumed to be white-collar workers. Because vocational secondary education can lead to either blue- or white-collar occupations, I use information on occupation to classify vocational school graduates into blue- and white-collar workers. Finally, re- gional dummy variables take into account regional differences in the cost of living and institutional wage-setting rules. The data are obtained from the October 1988 round of the Egyptian Labor Force Sample Survey, which was a special round designed to collect much more detailed information than the standard survey. In particular, it included a spe- cial module on earnings, which is the source of the earnings data used here. The earnings module gathered data on earnings net of payroll taxes and deductions in the reference year. An attempt was made to get data on earnings in kind, but the quality of that data seems quite poor. The annual earnings of intermittent workers were estimated by asking about the number of months worked in each 1. The standard errors of the wage equations are adjusted for the inclusion of the predicted sample selection terms. See Lee (1983) for more details on the multinomial logit selection model 124 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I of four quarters, the number of days worked per month in each quarter, and the daily wage rate.2 While the overall survey was administered to a stratified random sample of 10,000 households, parental background information comes from a module that was ad- ministered to a randomly selected subsample of 5,000 households. The data used here are limited to that subsample. The earnings equation estimates are also limited to nonagricultural wage workers between the ages of 18 and 59, the age group that is likely to have regular employment in the government and SOES. Table 1 provides descriptive statistics on the variables used in the earnings equations. Table 2 shows the earnings equation estimates for males and females in the government, SOES, and the private sector. While the earnings-experience profile has the usual concave shape in all three sectors, the profile has significantly more curvature in the private sector. Tenure (or seniority) is a significant deter- minant of earnings in all three areas. The returns to education are similar in the government and SOEs and are significantly higher than they are in the private sector (see Assaad 1997). While wages are roughly equal for males and females in the public sector, there is a large gender wage gap in the private sector. There- fore, female public sector workers are likely to have significantly higher dis- placement losses than their male counterparts. An examination of the selection terms reveals negative selection into govern- ment employment for both males and females. This is consistent with the opera- tion of the queue, which results in adverse selection into the government. By contrast, there is positive selection into the private sector, consistent with the operation of a more competitive labor market. There is no significant selectivity into SOEs. Figure 1 shows the expected earnings profile of a male worker in a white- collar occupation displaced from the government or an SOE after 15 years of tenure. The figure is based on actual parameter estimates from the earnings equa- tions. In drawing an earnings profile, I assume that general labor market experi- ence is transferable to the private sector but returns to seniority are not. How- ever, the worker is assumed to begin accumulating seniority once the move to the private sector takes place. The adjusted government and SOE profiles shown in the figure include an estimate of the value of nonwage job attributes. The permanent losses experienced by workers displaced from the government or an SOE are therefore equal to the area between the relevant adjusted public sector curve and the private sector curve after taking discounting into account. Nonwage Benefits in the Public Sector As described in section I and the appendix, nonwage benefits are estimated indirectly by calculating the ratio of total compensation to monetary compensa- 2 The earnings module was designed by Mohaya Zaytoun, professor of economics at El-Azhar University, Cairo, Egypt. Overall technical direction for the special round of the Labor Force Sample Survey was provided by Nader Fergany, managing director of the Al-mishkar Center for Research and Training, Cairo, Egypt. Assaad 125 Figure 1 Estimated Earnings Profiles for a Worker in a State-Owned Enterprise in Egypt Earnings (Egyptian pounds per year) 7,000 Adjusted state-owned 6,000 enterprise 5,000 Adjusted government 5,000- 4,000 3,000 Actual, state-owned enterprise Actual, government 2,000 , . - ' ' . Actual, private sector 1,000 15 20 25 30 35 Experience (years) Note These estimated earnings profiles are for a male wvhite-collar worker with secondary education who is displaced after 15 years of experience in the public sector The adjusted profiles for government and state-owned enterprises include an estimate of the value of nonwage job attributes Source Author's calculations tion (Tj) that sets lifetime rents to zero for a marginal group of public sector work- ers but yields positive rents for all other workers. Lifetime rents are defined here as the excess public sector compensation the worker obtains over a lifetime above his or her opportunity cost in the private sector. To find the estimated ratio of total compensation to monetary compensation, , I first calculate a parameter l,for a 12-cell classification of public sector workers who are recent entrants to the labor market. The parameter rj, is the multiple of the discounted lifetime earnings in the public sector career path that equates these earnings with the earnings the worker could have obtained in a private sector career path.3 The group with the highest mean f1, is the marginal group, and its rl, is equated to rj. The 12 cells are obtained by classifying workers by gender, educational level, and government or SOE affili- ation. I could have used a finer classification that also classifies workers by region or urban-rural status, but that would have given cells with a very sparse number of observations for which estimates would be unreliable. The base scenario has five assumptions. First, the discount rate is constant (p = 0.05). Second, the length of the queuing process (tj is seven years for level- 3. In calculating the discounted lifetime earnings in the public sector path, I include a queuing period for level two and level three, when a worker obtains a specified fraction p of the private sector earnings. Table 1. Means and Standard Deviations for Variables in the Earnings Equations for Nonagricultural Wage Workers in Egypt, 1988 Males Females State-owned Private State-owned Private Variable Government enterprises sector All Government enterprises sector All Log annual earnings 7.09 7.41 6 89 7.08 6.81 7.00 6.32 6.73 (0.675) (0.592) (1.005) (0.828) (0.722) (0.954) (1.149) ((.895) Experience (years) 18.9 20.6 13.7 17 2 9.89 11.9 6.88 9.48 (11.6) (11.0) (10.2) (11.3) (8.31) (8.74) (8.21) (8.48) Tenure (years) 13.8 14.7 4.80 10.5 9 17 11.( 3.64 8.17 (10.7) (10.3) (7.39) (10.5) (8.18) (8.38) (7.01) (8.33) Educational attainment Illiterate (reference category) 0.124 0.185 0 302 0.206 0.031 0.075 0.308 0.099 (0.330) (0.388) (0.459) (0.404) (0.172) (0.263) (0.462) (0.298) Read and write 0.146 0.236 0.195 0.184 (.010 0.037 0.055 0.023 a, (0 353) (0.425) (0.397) (0.388) (0.097) (0.191) (0.229) (0 151) Primary 0.064 0.092 0.084 0 078 0.006 0.075 0.049 0.025 (0.245) (0.289) (0.277) (0.268) (0.076) (0.264) (0.217) (0.155) Preparatory 0.040 0.074 0.101 0.071 0 017 0.047 0.077 0.035 (0.197) (0.262) (0.301) (0.257) (0.130) (0.212) (0.267) (0.183) General secondary 0.033 0.028 0.048 0.038 0.017 0.056 0.038 0.027 (0.179) (0.166) (0.213) (0.190) (0 130) (0.231) (0.193) (0.163) Vocational secondary, all 0 232 0 197 0.170 0.201 0.469 0.467 0.264 0.423 (0.422) (0.398) (0.376) (0.401) (0.499) (0.499) (0.441) (0.494) Vocational secondary, blue collar 0.016 0.062 0.113 0.063 0.004 0.093 0.077 0.032 (0.124) (0.241) (0.317) (0.243) (0.062) (0.292) (0.267) (0.176) Vocational secondary, white collar 0.217 0.136 0.057 0 138 0.466 0.374 0.187 0.391 (0.412) (0.343) (0.232) (0.345) (0.499) (0.486) (0.391) (0.488) Technical institute 0.077 0.030 0.024 0.047 0.134 0.037 0.033 0.099 (0.267) (0.171) (0.153) (0.211) (0.341) (0.191) (0.179) (0.298) University and above 0.283 0.158 0.076 0.177 0 316 0.206 0 176 0.270 (0.451) (0.365) (0 265) (0.381) (0.465) (0.406) (0.382) (0.444) Region of residence Greater Cairo (reference category) 0.231 0.387 0.376 0.320 0.307 0.533 0.544 0.390 (0.422) (0.487) (0.484) (0.467) (0.461) (0.499) (0.498) (0.488) Alexandria and Suez Canal 0.085 0.197 0.125 0.124 0.151 0.215 0.159 0.162 (0.280) (0.398) (0.331) (0.330) (0.359) (0.413) (0.367) (0.368) Urban Lower Egypt 0.145 0.137 0.148 0.144 0.224 0.150 0.110 0.189 (0.352) (0.344) (0.355) (0.352) (0.417) (0.358) (0.314) (0.391) Urban Upper Egypt 0.171 0.046 0.086 0.112 0 151 0.019 0.049 0.111 (0.377) (0.209) (0.280) (0.315) (0.359) (0.136) (0.217) (0.314) Rural Lower Egypt 0.216 0.165 0.176 0.190 0.138 0.047 0.099 0.117 (0.412) (0.372) (0.381) (0.392) (0.345) (0.212) (0.299) (0.322) Rural Upper Egypt 0.152 0.067 0.090 0.110 0.029 0.037 0.038 0.032 (0.359) (0.250) (0.286) (0.313) (0.167) (0.191) (0.193) (0.176) Job-related variables Intermittent employment 0.373 0.148 (0.484) (0.356) Work outside establishments 0.386 0.187 (0.487) (0.391) Selection term (X) 0.701 1.231 0.670 0.800 0.389 1.382 0.839 0.621 (0.420) (0.334) (0.481) (0.484) (0.335) (0.422) (0.576) (0.540) Number in sample 1,089 568 1,050 2,707 522 107 182 811 Number in population (thousands) 1,897 1,018 1,904 4,820 866 193 366 1,395 Note: All variables except log annual earnings, experience, tenure, and the selection term are dummy variables. Standard deviations are in parentheses. Source: Author's calculations based on survey data. Table 2. Selectivity-Corrected Earnings Equation Estimates for Nonagricultural Wage Workers in Egypt, 1988 Males Females State-owned Private State-owned Private Variable Govern7nent enterprises sector Government enterprises sector Constant 6.447-' 6.528 "x' 6.116'" 6.079- 5.713 5.706' (38.48) (37 11) (52.22) (26.12) (10.08) (26 80) Experlenice 0 047 0.039 0.084"- 0 039' 0.131 . 0.075 (8.83) (5.28) (8.19) (2.99) (3.25) (2.78) Experience2 /100 -0.081' "' -0.057 . -0.199: ' -0.116' -0.232 -0.279 (-7.79) (-3.91) (-9.26) (-3 91) (-2.41) (-3.12) Tenure 0.021"" 0 013 ''' 0.023 '-' 0.043 -0.001 0 045 (8 77) (4.47) (5.55) (4 22) (-0.03) (2.55) Educational attainmenta Read and write 0.072 0.165"' 0.024 0.414 -0.339 0.309 (l 24) (2.69) (0.31) (1.58) (-0.75) (0.89) Primary 0 084 0.168' ' 0.086 0 128 -0.041 0.215 (1.10) (2.16) (0.77) (0.40) (-0.10) (0.58) Preparatory 0.222-'" 0.284' -0.217 0.110 0.492 -0.467 (2.52) (3 29) (-0.20) (0.51) (1.16) (-1.49) General secondary 0 215" 0.671"' -0.460-"' 0.432" 0.849"" 0.075 (2.15) (5.47) (-3.19) (2.01) (2.09) (0.18) Vocational secondary, blue collar 0.536"" 0.631 '" 0.056 -0.799' 0 963""" -0.074 (3.94) (6.64) (0.42) (-1.89) (2.64) (-0.21) Vocational secondary, white collar 0.324"' 0.532- t" 0.148 0.262 0.693 -0.331 (433) (7.13) (0.95) (1.63) (2.30) (-1.02) Technical institute 0.298'X'X 0.615 '" -0.114 0.133 0.081 -0 530 (3.00) (5.05) (-0.52) (0.73) (0.19) (-1.05) University and above 0.586""' 1.062' x 0.578"'¢ 0.591 '-' 0.720a- 0.785* (6.50) (14.73) (3.44) (3.56) (2 34) (2.57) Region of residenceb Alexandria and Suez Canal -0.071 -0.201'>** 0.005 0.027 0.094 -0.092 (-1.18) (-3.68) (0.06) (0.37) (0.53) (-0.45) Urban Lower Egypt -0.278*** -0.149** -0.292*'* -0.085 0.041 -0.619** (-5.13) (-2.43) (-3.49) (-1.14) (0.20) (-2.53) Urban Upper Egypt -0.332* -0.072 -0.584 ** -0.051 -0.315 -0.392 (-5.13) (-0.65) (-5.51) (-0.53) (-0.53) (-1.08) Rural Lower Egypt -0.4344*" -0.140*' -0.218**' -0.241"*'¢ 0.012 0.060 (-7.61) (-2.27) (-2.69) (-2.65) (0.03) (0.23) Rural Upper Egypt -0.454*** 0.075 -0.293t* -0.195 -0.175 -0.671* (-6.59) (0.84) (-2.80) (-1.34) (-0.48) (-1.72) Selection term (X) -0.227*'i -0.087 0.348**" -0.342*** -0.234 0.397 (-3.06) (-0.92) (3.32) (-2.82) (-0.91) (2.05) R2 0.471 0.439 0.290 0.457 0.502 0.305 Sample size 1,089 568 1,050 522 107 182 Significant at 10 percent. Significant at 5 percent ** Significant at 1 percent. Note: The dependent variable is the log of annual earnings. Standard errors are adjusted for the inclusion of the predicted selection term. t-ratios are in parentheses. a. Illiterate is the reference category. b. Greater Cairo is the reference category. Source: Author's calculations based on survey data. Table 3. Selected Measured and Estimated Variables for Public Sector Workers by Educational Level in Egypt, 1988 Government State-owned enterprises Variable Level one Level two Level three All Level one Level two Level three All Males Percentage of public sector workforce 18 6 16.0 13.1 47.7 14.6 6.5 4.5 25.6 Average tenure (years) 17 12 12 14 17 12 12 15 Average monthly salary (Egyptian pounds) 94 102 148 111 120 154 249 151 Ratio of private to public discounted lifetime earnings 1.51 1.21 1.41 1.38 1.28 1.11 1.25 1.23 1.93 1.22 1.41 1.41 1 42 1.07 1 17 1.24 Average displacement losses (Egyptian pounds) 4,978 14,251 17,160 11,43(0 7,625 21,430 29,957 15,056 Average losses in monthly salaries (months) 53 140 116 103 64 139 120 10( Number of observations 408 373 308 1,089 333 145 90 568 Females Percentage of public sector workforce 1.4 13.5 6.9 21.8 1.0 2.8 1.0 4.9 Averagc tenure (years) 13 9 8 9 12 10 13 l I Average monthly salary (Egyptian pounds) 63 85 111 92 81 129 193 133 Ratio of private to public discounted lifetime w earnings 1.17 0.54 1.14 0.77 1.20 0.64 2.03 1.05 11 a 1.40 0.57 1.17 0.79 1.51 0.68 2.33 1.07 Average displacement losses (Egyptian pounds) 7,878 21,519 25,163 21,810 11,817 32,702 15,234 24,658 Average losses in monthly salaries (months) 125 253 228 238 147 254 79 186 Number of observations 33 324 165 522 25 60 22 107 All Percentage of public sector workforce 20.0 29.6 20.0 69.5 15.6 9.3 5.6 30.5 Average tenure (years) 17 11 10 12 16 11 12 14 Average monthly salary (Egyptian pounds) 92 94 135 105 117 146 238 148 Ratio of private to public discounted lifetime earnings 1.48 0.90 1.32 1.19 1 27 0.97 1 40 1.20 Ti a 1 89 0.92 1.33 1.13 1 42 0.95 1 39 1.20 Average displacement losses (Egyptian pounds) 5,177 17,579 19,918 14,683 7,893 24,850 27,200 16,588 Average losses in monthly salaries (months) 56 187 147 140 67 170 114 112 Number of observations 441 697 473 1,611 358 205 112 675 Note: Level one refers to workers with less than a secondary school certificate. Level two refers to workers with a secondary or technical institute certificate. Level three refers to workers with a university or graduate certificate a. is the ratio of private to public discounted lifetime earnings limited to workers who are 35 or younger. Source: Author's calculations based on survey data. Assaad 131 two workers and six years for level-three workers. Third, queuing involves a total loss of private sector earnings (p) for females and the loss of half of poten- tial earnings for males (p = 0.5 for males and p = 0 for females). Fergany (1991) attributes the higher proportion of new entrants among unemployed females to the fact that males are more likely to engage in marginal or occasional economic activities while waiting for government employment. Fourth, nonwage benefits are proportional to wages. And fifth, an age cutoff of 35 is used to identify recent entrants to the public sector. To the extent possible, I assess the conse- quences of these assumptions on the estimates obtained. The ratio of the discounted stream of monetary earnings in the private and public sectors and the parameter l,, which is closely related to it, are shown in table 3 for each of the 12 cells and for all workers. The group with the highest ratio (and the highest il) consists of female SOE workers with level-three (univer- sity) education (rj, = 2.33), followed by male government workers with level-one (less than secondary) education (Ti, = 1.93). Among workers 35 and under, there are only 8 observations in the sample in the first group and 84 observations in the second group. The first group engages in costly queuing, but the second does not. All other groups have much smaller ratios than these two groups. Govern- ment wages and benefits are therefore just sufficient to attract these two catego- ries of workers to public sector employment. Given the relative imprecision of these estimates and the sparse number of observations in the first group, I use q = 2 as the baseline ratio of total compensation to monetary compensation in the public sector. Because the ratio of nonwage benefits is an important parameter in the subse- quent analysis, it is worth doing a sensitivity analysis to determine its robustness to the various assumptions made. First, I test the extent to which the rankings of the various groups and the estimate of rj change when the age cutoff is increased or decreased. At an age cutoff of 30, the same two groups of workers emerge as the lowest-lifetime-rent workers, with il, = 2.28 for male government workers with level-one education (38 observations) and il, = 2.52 for female SOE workers with level-three education (5 observations). At an age cutoff of 40, the ranking remains the same, with rl, = 1.75 for male government workers with level-one education (164 observations) and l = 2.18 for female SOE workers with level- three education (14 observations). Thus the estimate of rq ranges from 1.8 to 2.5, depending on the choice of age cutoff. Younger age cutoffs yield very sparse cells and therefore increasingly unreliable estimates. Older age cutoffs may in- clude workers who entered the public sector facing significantly different wage schedules. Second, I test for robustness to the assumptions relating to the length of queu- ing (t,) and the cost of queuing (p). Shorter queuing time and less costly queuing raise the lifetime rents of level-two and level-three workers but do not affect the rent of level-one workers, who are assumed not to queue. Shorter and less costly queuing may increase the rents of female level-three workers to the point where they are no longer the lowest-rent workers but does not displace male level-one 132 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 government workers from their position among the lowest-rent workers. Thus the robustness of the estimate needs to be tested only for longer and more costly queuing. To increase the cost of queuing to a maximum, I set the fraction of private sector earnings that can be earned while queuing (p) to zero and the length of the queuing period (I>) to 10 years for level-two workers and 9 years for level-three workers. Despite these changes, the two groups with the lowest lifetime rents (highest Tr,) remain the same. Increasing the queuing time to 9 and 10 years raises the Tl of female SOE workers with level-three education only marginally, to 2.56 at an age cutoff of 35. Sensitivity analysis on the discount rate shows that the marginal groups of workers remain the same for a range of discount rates from 0.03 to 0.07 and that il stays well within the range of 1.8 to 2.5. Finally, the assumption that the ratio of nonwage benefits is invariant across the two segments of the public sector is not important because a group of marginal workers is identified in each of the two subsectors and both give roughly similar estimates for r1. These sensitivity tests suggest that the identification of a marginal group of workers and the estimate of f are fairly robust to changes in the assumptions. I use il = 2.0 as the baseline estimate but also discuss results for a low estimate of = 1.8 and a high estimate of TI = 2.5. Displacement Losses Once the ratios of total compensation to monetary compensation (ri) are ob- tained, the estimation of worker-specific displacement losses is fairly straight- forward using equation A-7 in the appendix. Table 3 shows the average esti- mated displacement losses for the 12-cell classification and for all workers. The average losses of SOE workers are about £E16,600 (the equivalent of 112 months of salary), compared with £E14,680 (140 months of salary) for government workers. (The Egyptian pound, £E, was worth US$0.30 in 1988.) In general, displacement losses for female workers in both the government and SOE sectors are significantly higher than those of their male counterparts, with the exception of level-three female workers in SOEs. Displacement losses tend to be significantly higher for workers with second- ary and postsecondary education than for workers with lower levels of educa- tion but do not increase much between the secondary and university levels. Losses of level-two SOE workers are more than three times as high as those of level-one workers, whereas those of level-three workers are only 9 percent higher than those of level-two workers. Among women, level-two SOE workers have the high- est displacement losses, £E32,700 (254 months of salary). These patterns reflect the differential premiums placed on various levels of education in the private and public sectors. While public sector workers receive significant returns to secondary education, the private sector places little value on it, resulting in high displacement losses for workers at that level of education (see table 3). At the tertiary level, the returns to education continue to be higher in the public sector but are nonetheless significant in the private sector, especially for women (see Assaad 133 Assaad 1997 for a more extensive discussion of the returns to education in the private and public sectors in Egypt). The significant difference in displacement losses between workers at different levels of education has important implications for the design of severance pay programs. If the same package of benefits is offered to all workers to achieve a certain rate of exit, the likely outcome is that all level-one workers, who tend to have lower losses, will exit first, leading to a highly distorted occupational struc- ture. Some control can be achieved over the composition of the exiting workers by setting up separate programs for each level of education. This is the approach pursued below. The losses of SOE workers at each level of education are significantly higher than those of government workers, with the exception of level-three female workers. This reflects the sharper erosion in real wages experienced by govern- ment workers compared with SOE workers in recent years, making the compen- sation of government workers more comparable to what they can get in the private sector. Because of compositional differences, however, the losses of the average worker in the government and SOEs are comparable. Level-one workers, who have relatively low losses, make up more than half of the SOE workforce but only a quarter of government employment. Figure 2 plots the estimated displacement losses against tenure in the public sector for SOE workers at different levels of education using base case assump- tions. The solid line is a cubic spline that connects the median losses at each level of tenure. Because displacement losses are equal to the area between the public and private sector earning profiles up to retirement age, they must fall to zero as the worker approaches retirement, as shown on the figure. Moreover, because discounting reduces losses experienced far into the future, losses may first in- crease as workers gain seniority and then begin falling as the years of denied service decline. In all panels except for that of level-three SOE workers, females have higher losses, as indicated by the higher density of "Is" above the median line than below it. These profiles reveal that severance schemes that positively index payments on tenure, such as those that pay a given number of monthly wages per year of tenure, are fiscally costly because they pay the highest compensation to the work- ers with the lowest losses. Schemes that index on years of denied service do a better job of tracing these estimated loss profiles and therefore may be more cost-effective. Section III evaluates various severance pay schemes on the basis of how well they match compensation payments to worker losses. Sensitivity of the Loss Estimates Here I analyze the sensitivity of the loss estimates to the estimate of nonwage benefits. Varying rq over a range from 1.8 to 2.5 while keeping it constant across workers will change the magnitude of displacement losses, but will it change the rankings of various groups of workers within the 12-cell classification and affect the shape of the loss-tenure profiles shown in figure 2? As expected, the magni- 134 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Figure 2 Estimated Displacement Losses andJob Tenure by Educational Level for Workers in State-Owned Enteiprises in EgVpt, Base Case Displacement losses (Egyptian pounds) Level one 60,000 0 = male 1 female 30,000- o i0a 0~ii 0 10 20 30 40 Level two 60 o0o - 30,000 - 1 1 0 88 ° O I I 0 1 0 20 30 40 60 000 Level three _ o 30,000o 0 0 10 20 30 40 TenUre (yejr.i Note The solid line is a ctibic sphne that connects the median losses at each le-el of tenure Sou rce Author's calculations Assaad 135 tude of displacement losses varies significantly for the average SOE worker as rl goes from 1.8 to 2.5. Losses change from £E13,925 to £E23,285 as compared with £E16,588 in the base case (ri = 2). However, the ordinal ranking of losses in the 12-cell classification of workers by gender, education, and subsector changes very little. Furthermore, other than a change in scale, the shape of the loss- tenure profile shown in figure 2 for the base case changes very little as q is altered.4 Do the estimates of worker losses differ under the assumption that nonwage benefits are a constant amount per year for all workers? I use a methodology similar to that described in section II to calculate the constant amount. Rather than finding the maximum multiple of monetary earnings that equates lifetime compensation in the public and private sectors, I find the maximum additive shift that does so. The same two groups-female level-three SOE workers and male level-one government workers-show up as marginal. Under the other baseline assumptions, the size of lifetime nonwage benefits turns out to be £E13,370 for level-one male government workers and £E28,890 for level-three female SOE workers. On the basis of these estimates, I use an additive shift of £E28,900 on lifetime earnings to calculate an alternative set of displacement losses. This additive shift is prorated by the amount of time each worker has left in the public sector. Although the assumption that nonwage benefits are constant for all workers is somewhat unrealistic, it gives a sufficiently different pattern of nonwage ben- efits across workers compared with the base case of benefits proportional to earnings. Some of the consequences of this change in assumptions are obvious: the losses of low-wage workers increase and those of high-wage workers de- crease. In fact, the losses of male level-three SOE workers, the highest-paid cat- egory of worker, fall 47 percent and those of male level-one government work- ers increase 21 percent. Less expected is the fact that female losses change significantly less than male losses. Losses decline 9 percent for females com- pared with 25 percent for males. This is due to the fact that female workers in the sample are generally younger and therefore tend to have a longer period of constant nonwage benefits per year to look forward to in the public sector than males. Because of the compression of the loss profile across educational levels and the increasing gender gap, the ordinal rankings of losses change. The highest- loss group continues to be females with level-two education in the SOE sector. However, they are now followed in second place by female level-two workers in the government rather than male level-three SOE workers. With this way of esti- mating nonwage benefits, level-two workers have higher losses than level-three workers for both males and females. An examination of the loss-tenure profiles shows that with the exception of a change in scale and some minor variations in pattern, they are essentially similar to those obtained using the base case as- 4. The results for different values of tj are available from the author. 136 THE WORLD BANK ECONO0.11C REVIEW. VOL 13, NO I sumptions. Because most severance pay schemes tend to index on tenure, years of denied service, or age, the invariance of these profiles to the way nonwage benefits are estimated is what really matters. III. SIMULATION OF SEVERANCE PAY SCHEMES BASED ON ALTERNATIVE INDEXATION FORMULAS In the presence of heterogeneous workers, appropriate indexation of the sev- erance payment to observed differences in worker attributes should in theory reduce the cost of the program for a given desired rate of exit. A uniform com- pensation package will be fiscally costly because it must be set at a level that compensates the worker with the largest losses among those who exit. As a result, it would end up overcompensating most workers who accept the pack- age. However, compensation schemes that index on the wrong variables or that set the wrong parameters for the severance pay formulas could cost even more than a uniform payment scheme. For instance, a scheme that indexes on tenure rather than years of denied service costs more than a uniform payment scheme because the tenure scheme pays the most to workers who are close to retirement. Furthermore, the loss-tenure profiles indicate that losses include a fixed com- ponent as well as a variable (or indexed) component. Payment schemes that set compensations on a purely variable basis may perform worse than schemes that have both a fixed and a variable component. Finally, compensation plans that separate out workers who are close to retirement and give them the option of early retirement are likely to perform better than ones that apply the same in- dexation rule to all workers. In the subsequent analysis, the uniform payment scheme, which is the sim- plest scheme to implement, will serve as a benchmark for comparison of the various formulas. For another useful benchmark-the perfect indexation bench- mark-the payment is set at the estimated losses obtained in section II. The analysis consists of several simulations that optimize the parameters of alterna- tive severance pay formulas to achieve the desired rate of voluntary exits in each of three educational groups at the lowest possible cost. The average cost per retrenched worker and the total cost of the program under each of these alterna- tives are then compared with the two benchmarks. The effect of each program on the composition of the exiting and remaining workforces is also investigated. Although it is possible in practice to set different exit rates for each educational level or for that matter for various other categories of workers, for the purposes of this analysis the target exit rate is set at 30 percent for each of the three levels. I start with severance pay schemes that apply to all workers within each of the three educational categories and then consider schemes that offer an early retire- ment option for workers 50 years and older and a regular severance package for workers below that age. In all cases, the payment amount is calculated as if it were a lump-sum payment, but this does not preclude various payment methods, includ- ing annuities, combinations of annuities and lump sums, and pension payments. Assaad 137 A worker is assumed to exit if the compensation the worker receives under any of the programs equals or exceeds the estimated loss from displacement. For any given level of the fixed payment C,, the coefficient c, is set at the minimum level that achieves the desired exit rate, namely 30 percent of the workers at each educational level. The value of the fixed payment C, is optimized by select- ing the value that minimizes the average cost of severance per retrenched worker under the Ith formula. Because different programs are implemented at each level of education, different parameters are computed at each level for each severance pay formula.5 Schemes without Provisions for Early Retirement Here I discuss the uniform payment scheme and five severance pay schemes. First, under a uniform payment scheme, there is no indexation. The scheme (formula 1) is characterized by C = Cl. This is the simplest possible scheme and serves as the high-cost benchmark. The analysis considers separate programs at the three levels of education so that the payment is indexed only on education. Any program that results in a higher cost per retrenched worker is probably not worth considering. The uniform payment is set at a level that would just com- pensate workers whose losses are equal to the 30th percentile of the loss distri- bution at each level of education. Formula 2 has a fixed payment plus a given amount per year of denied ser- vice, so that C = C2 + c2 (60 - Age). In this scheme, the amount of compensation declines with tenure and thus captures some of the negatively sloped portion of the loss profiles shown in figure 2. Formula 3 has a fixed payment plus a given amount per year of tenure. That is, C = C3 + c3N, where N is years of tenure. Positive indexation by tenure is a com- mon feature of severance programs. This is the simplest such scheme. It may cap- ture the loss profile of younger workers fairly well, but it does a poor job of match- ing the declining losses with the tenure of older workers. It therefore overcompensates those workers. A more common variant of this formula is to index on wage and tenure by paying a certain number of monthly wages per year of tenure. An ex- ample of such a program is the Leather Corporation package in Sri Lanka dis- cussed in Fiszbein (1992). I tried this variant, but it produced results that were clearly inferior to the basic scheme from a cost point of view. Formula 4 has a fixed payment plus a given number of monthly wages. C = C4 + C4WG, where WG is the estimated monthly government wage. Linking com- pensation payments to monthly wages is quite common. This formula explores that linkage independent of the role of tenure or years of denied service. Here again, the most senior workers receive the highest compensation, even though they have the lowest losses. Formula 5 has a fixed payment plus a given number of monthly wages per year of denied service, so that C = C5 + C5WG(6O - Age). This scheme attempts to 5 The parameter values are available from the author. 138 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Figure 3 Average Compensation per Retrenched Worker and the Fixed Component of Severance Pay by Educational Level in Egypt, Base Case Average compensation per Level one worker (Egyptian pounds) 10 000 9,000 FFormLla 2 (index on vears of denied service) 8,000 Formula 3 (index on vears of tenure) 7,000 ' \._. _-- Formula 4 (index on monthly wages) ---A--- Formula 5 (index on monthly wagei 60ooo per years of denied ser iie *, ,>*. * * * ' - ' -+ * t. ------ Formula 6 (index on monthlv 5,000 x, - -wages y cars of tenure, and vears of A ~~~~~~~denied service) 4,000 * 3.000 0 1,000 2 000 3,000 4,000 5,000 Level two 35,000 .0 ...40. 00 20 100 20 23,000 20,000 15 000 o,o 0 4,doo 8,000 '12 000 1G 000 20,000 Level three 23 000 31,000 29 000 27 000i 25 000 9 23 000 21,000 \_ 17,000!, 15,000 0 4,000 8 000 12,000 16,000 20,000 Fixed compensation (Egyptian pounds) Mote ,See text and table 4 for more details on the severance pay formulas Source AAuthor's calculations Assaad 139 mimic the curvature in the loss profile by linking compensation positively to wages, which are an increasing function of tenure, and to years of denied ser- vice, which are a decreasing function of tenure. Formula 6 has a fixed payment plus a multiple of the monthly wage times years of tenure times years of denied service. That is, C = C6 + C6WGN(60 - Age). This is the most complicated of the severance pay formulas considered. A variant of this formula was used in the Bulmulla package in Sri Lanka (see Fiszbein 1992). As mentioned above, the simulation exercise consists of optimizing the pa- rameters of each severance pay formula to achieve the target exit rate through voluntary exits at the lowest cost. Figure 3 shows the average cost per retrenched worker and the fixed component of the severance payment (Cl) for the various severance pay formulas being considered. Each point on any given curve is the result of an optimization process that sets the multiple of the relevant index (cl) to the minimum required for achieving the exit target. The point at the extreme right side of each chart where all the lines meet is the point for which the entire compensation is fixed. This is therefore the payment level under formula 1, the uniform payment benchmark. The extreme left of each chart represents a purely variable compensation in which the fixed component is set to zero. The results summarized in figure 3 show that some commonly used severance pay formulas are always more costly than the fixed payment benchmark. These include formulas 3 and 4, which index on tenure or monthly wage. The results from a formula that indexes on both monthly wages and tenure are not shown but behave in very much the same way. Adding years of denied service to the index (formula 6) reduces the costs somewhat for level-one workers but per- forms worse than the uniform payment benchmark for level-two and level-three workers. Purely variable compensation payments, which are commonly used in severance programs, generally perform poorly as well. Under most formulas, compensation costs rise as the fixed payment approaches zero. As expected from examining the shape of the loss-tenure profile, the best- performing formulas are those that index either on years of denied service alone or on years of denied service and monthly wage (formulas 2 and 5). Formula 5 performs best for level-one and level-three workers, and formula 2 performs best for level-two workers, but formula 5 is a close second. The best combina- tion of fixed and variable payments appears to have a fixed component that makes up from one-third to half of the total payment. Table 4 shows the average cost per worker and the total cost of the program at the optimum (lowest-cost) point for the various severance pay formulas, the full indexation benchmark, and the fixed payment benchmark. Next to the full indexation benchmark (formula 0), the lowest-cost scheme is the one based on formula 5. This can achieve the 30 percent exit target at a cost of £E10,876 per worker or a total cost of £E3.98 billion to exit 370,000 workers. The cost per worker under this formula ranges from £E4,159 for level-one workers to more than £E17,000 for level-two and level-three workers. Indexation with formula 5 can thus achieve savings of about 13 percent over the uniform payment bench- Table 4. Cost of Compensation Programs under Alternative Severance Pay Formulas (Egyptian pounds) Cost per retrenched worker Total cost of program (millions) Formula number and description Level one Level two Level three All Level one Level two Level three All 0 Full indexation 2,735 12,301 12,966 7,555 500 1,310 853 2,663 1 Fixed payment 5,518 19,808 19,696 12,513 1,039 2,231 1,354 4,624 2 Fixed payment and multiple of years of denied service 4,889 16,463 19,137 11,068 917 1,832 1,280 4,029 3 Fixed payment and multiple of years of tenure 5,567 19,783 19,243 12,446 1,041 2,212 1,291 4,544 4 Fixed payment and multiple of monthly wages 5,569 19,761 19,114 12,416 1,048 2,210 1,282 4,540 5 Fixed payment and multiple of monthly wages times years of denied service 4,159 18,187 17,276 10,876 775 2,029 1,178 3,982 6 Fixed payment and multiple of monthly wages, years of tenure, and years of denied service 5,103 19,771 18,451 12,058 970 2,211 1,238 4,419 Number of exiters (thousands) 188 113 69 369 188 113 69 369 Note. The calculations are based on a 30 percent exit target. Level one refers to workers with less than a secondary school certificate. Level two refers to workers with a secondary or technical institute certificate. Level three refers to workers with a university or graduate degree The optimal parameters C, and c, for each alternative i are available from the author Source: Author's calculations based on survey data Assaad 141 mark, with the greatest savings being obtained for level-one workers. Formula 2 is a close second at £E11,068 per worker and £E4.03 billion overall. A slightly better performance (£E10,351 per worker and £E3.79 billion overall) can be achieved by combining formula 5 for level-one and level-three workers with formula 2 for level-two workers. It should be kept in mind, however, that even if the same formula is used for all three levels of education, the optimal param- eters would be significantly different. For instance, the optimal parameters for formula 5 are £E1,500 plus 2.25 monthly salaries per year of denied service for level-one workers, £E12,600 plus 1.83 monthly salaries per year of denied ser- vice for level-two workers, and £E5,400 plus 3.15 monthly salaries per year of denied service for level-three workers. As indicated in figure 3, formulas 3, 4, and 6 perform no better than a fixed payment in most cases. Moreover, all of the severance pay schemes fall well short of the savings that are possible with full indexation. The best-performing formula among these commonly used schemes (formula 5) exceeds the full in- dexation benchmark by nearly 45 percent. I conducted a sensitivity analysis to test the robustness of these results to different estimates of nonwage benefits. The level of compensation changes de- pending on the size of q and on whether nonwage benefits are proportional to wages or simply constant. However, the interesting issue is whether the relative performance of the various formulas changes. Varying il from 1.8 to 2.5 pro- duces charts virtually identical to those shown in figure 3, except for a change in scale.6 Assuming that nonwage benefits are constant rather than proportional to wages produces slightly different shapes, but the ranking of the performance of each of the five severance pay formulas remains the same. Thus, although the magnitude of the estimates of compensation costs varies depending on how nonwage benefits are estimated, the conclusions on the relative performance of each formula are robust to different estimation methods. Schemes with Early Retirement Provisions The early retirement schemes offer separate severance packages for workers age 50 and above and for those below age 50 to take advantage of the declining losses of workers approaching retirement. Under Egyptian social security regu- lations, workers who are 50 or above can still receive a retirement pension if they retire early. However, their pensions are reduced significantly compared with what they would receive if they retired at age 60. The amount of reduction is determined by a complicated formula that distinguishes between basic and variable components of the wage. The rules are described in some detail in Assaad (1996). Based on a detailed analysis of workers in the Delta Spinning and Weav- ing Company, the loss of benefits amounts to 66 percent of full benefits at age 50, 57 percent at age 53, and 40 percent at age 57 (Integrated Development Consultants 1994: table B.5). 6. The figures are not shown here but are available from the author. Table 5. Cost of Compensation Programs with an Early Retirement Option (Egyptian pounds unless otherwise noted) Exit rate Average cost per retrenched worker Total cost of program (millions) Retiremett scheme (percent) L.evel one' Level twob Level three' All Level one' Level twob l.evel threec All Target 70 percent for early retirement Workers age 50 and overd 72 2,594 6,014 12,336 6,205 273 117 212 601 Workers under age 50 22e 6,240 18,441 19,069 9,470 513 1,736 979 3,228 All 30 4,194 16,316 17,379 8,217 786 1,853 1,191 3,829 Target 80 percent for early retirement Workers age 50 and overd 82 2,918 6,526 12,939 6,846 349 143 257 749 Workers under age 50 21' 5,813 18,220 18,300 9,035 397 1,671 893 2,961 All 31 3,969 15,964 16,749 8,080 746 1,814 1,150 3,710 Target 90 percent for early retirement Workers age 50 and overd 91 3,473 6,854 13,291 7,485 468 156 282 906 Workers under age 50 19e 5,209 17,998 18,322 8,399 277 1,629 854 2,760 All 30 3,965 15,753 16,748 7,956 745 1,785 1,136 3,667 Target 100 percent for early retirement Workers age 50 and over"1 100 4,080 7,269 13,456 8,220 607 182 298 1,087 Workers under age 50 17e 4,450 17,790 17,775 7,657 170 1,535 799 2,503 All 30 4,156 15,428 16,368 7,958 777 1,717 1,097 3,590 Number of workers (thousands) Age 50 and over 149 25 22 196 149 25 22 196 Under age 50 472 344 199 1,016 472 344 199 1,016 All 621 369 221 1,212 621 369 221 1,212 Note. Level one refers to workers with iess than a secondary school certificate. Level two refers to workers with a secondary or technical institute certificate. Level three refers to workers with a university or graduate degree The optimal parameters C, and c1 for each alternative I are available from the author. For severance pay formulas, see the text and table 4. a. Indexation formula 4 was optimal for level-one workers under age 50. b Indexation formula 2 was optimal for level-two workers under age 50. c Indexation formula 5 was optimal for level-three workers under age 50. d. Indexation formula 7 (defined in the text) was used for workers age 50 and above e. Based on the All category for the three levels. Source: Author's calculations based on survey data Assaad 143 Assuming that workers 50 and older would voluntarily retire early if they were offered full retirement benefits, one approach would be to offer them a compensation package equal to the present value of the lost benefits they and their survivors would incur for as long as they receive pension benefits. Inte- grated Development Consultants ( 1994) pursues this approach but does not take into account the losses incurred due to lost benefits after retirement. However, for some workers, retirement, even with full benefits, may not be more desirable than continuing to work. Indeed, retirement at age 60 is mandatory rather than voluntary. Moreover, it may be very difficult to calculate the value of lost ben- efits. The calculation would require detailed actuarial information on the ex- pected life span of the retirees and their spouses to calculate the duration over which losses are incurred. Rather than rely on an approach that compensates workers for lost benefits, I use an opportunity-cost approach that implies matching the compensation to the individual-specific losses as estimated in section II. The main difference in the early retirement plan is that it uses a severance pay formula that pays work- ers 50 and above a lump-sum payment plus a multiple of their monthly wage until retirement according to the formula: C = C7 + c71, where I is the present value of the current annual earnings until retirement (formula 7).7 Both the lump sum (C7) and the multiple of the monthly wage (C7) are optimized to achieve a given exit target of people 50 and older at the lowest cost. Because workers 50 and older typically have lower losses, cost minimization might mean that the exit target for that group should be close to 100 percent. This may be impracti- cal for operational reasons; therefore, I examine alternative scenarios in which the early retirement exit target for that age group ranges from 70 to 100 percent. Workers under age 50 are offered a separate severance package based on one of the severance pay formulas. The parameters of that formula and the choice of the formula itself are optimized to achieve the exit of the balance of people necessary to reach the overall 30 percent exit target at the lowest possible cost. Workers younger than 50 are offered the plan that exits the balance needed to reach the 30 percent overall target at minimum cost. Table 5 summarizes the results for the various early retirement schemes under consideration. The plan with the lowest average cost per worker is the one that exits 90 percent of workers 50 and over, with the plans exiting 80 and 100 percent very close behind. Because it is difficult to fix the percentage of exiting workers exactly at 90 or 80 percent, these plans in fact exit 91 and 82 percent of workers age 50 and over. All the early retirement plans offer significant savings over the plans without early retirement. The plan that exits 90 percent of work- ers age 50 and over and 19 percent of workers under age 50 costs £E7,956 per 7. 1= J ()e dt = 12WG(T) e e where W is the current monthly wage in the public sector. 144 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I exiting worker, which represents a 36 percent saving over the uniform payment benchmark and a 27 percent saving over the best-performing plan without early retirement. In fact, it is only 5 percent above the full indexation benchmark (the first row in table 4). Now that the older workers are offered an early retirement plan, the optimal formula for level-one workers who are under age 50 changes from formula 5 to formula 4. Formulas 2 and 5 remain the optimal formulas for level-two and level-three workers, respectively. Because the compensation is based on individual-specific losses, including nonwage benefits, the multiple of the current wage paid to early retirees until the mandatory age of retirement may actually be larger than 1. In fact, under the 90 percent exit scenario for workers age 50 and over, the optimum multiple is 1.64 for level-one workers, 1.71 for level-two workers, and 1.62 for level-three work- ers. It may seem strange that these workers will be told not to report to work and will receive more in monetary compensation every month than they are currently receiving in wages. It should be kept in mind, however, that out of these payments they have to make both the employer and employee contributions to the social security fund if they wish to receive full benefits when they reach the age of 60. These contributions amount to approximately 40 percent of the monthly wage. Despite the fact that early retirees would be paid more than their current monthly wage, the early retirement plans cost less than the plans with no early retirement provisions because they allow for better matching between severance payments and worker rents. The simulations indicate that the cheapest workers to compensate for leaving the public sector are level-one workers who are age 50 and over. Severance costs for older level-two workers are also significantly lower than those for their younger counterparts. It may therefore be optimal to exit 100 percent of workers 50 and older with low and intermediate levels of education and limit the exit of older workers with university education, who require relatively high compensation. This may also be desirable from an opera- tional point of view because it may retain managerial talent in the public sector. As a general rule, it makes sense from a fiscal perspective to maximize the exit rate among workers with low levels of education until managerial and opera- tional considerations preclude any further reductions. Simulation Results on Workforce Composition The main compositional variable over which I have attempted to maintain some control is the proportion of exiters in each of three broad educational categories, which is kept at 30 percent. In the early retirement plans, I also fix the target exit rate for workers age 50 and older. In practice, it may be necessary to control the composition of the remaining labor force on several other dimen- sions, such as the proportion of managerial or production workers or the pro- portion of males and females. This can be done by offering each of these catego- ries a separate severance program, as is done here with the three educational categories. Assaad 145 Because exit rates are not set ex ante for specific age, gender, regional, or occupational categories, it is interesting to see what they turn out to be ex post under the various severance pay formulas that prove cost-effective. Table 6 shows these ex post exit rates as produced by the simulations for different categories of workers. The full indexation plan is not shown because it exits exactly the same people as the fixed payment plan. I also do not show the exit rate by educational category because it is set at the 30 percent target. In terms of age composition, most of the options without special provisions for early retirement exit workers age 50 and older at a rate of somewhere be- tween 88 and 94 percent, close to optimum rates found in the early retirement schemes. Workers under age 35 exit at very low rates, ranging from 9 to 18 percent. The highest exit rates for that group are obtained for the severance pay formula that indexes only on years of denied service (formula 2), thus paying these young workers relatively high compensation. Workers in the middle age group exit at rates close to the target rate. As expected, females exit at lower rates than males because they have higher losses. However, the difference between males and females is not very large. The Alexandria and Suez Canal region and to a lesser extent the Rural Lower Egypt region have disproportionately high exit rates. Conversely, Upper Egypt, a re- gion with poor private sector prospects, has disproportionately low exit rates. In terms of occupation, the most noteworthy result is the very high exit rate among workers in managerial occupations, who exit at a rate of close to 80 percent under most scenarios. This is probably because managers are more likely to be represented among the group over age 50 that exits at a high rate. The option that has the lowest exit rate for managers is formula 2, which also has the highest proportion of young people exiting. The disproportionate departure of managers may be welcome if the new owners of privatized SOEs prefer to hire their own managerial staff. However, if the retention of existing managers is deemed necessary, they could be offered a separate severance package that sets their target exit rates to lower levels ex ante. The other occupational group that exits disproportionately is service workers. These workers, who are mostly made up of janitors, guards, and messengers, are generally considered fairly unpro- ductive in the Egyptian context. Again, if that exit rate is deemed insufficient, they could be offered a more generous severance package that would induce more of them to exit. IV. CONCLUDING REMARKS This article had two major objectives: to estimate the losses that public sector workers would incur if they were displaced from their lobs and to simulate various voluntary severance schemes to determine how well they match compensation pay- ments to these estimated losses. It measured the displacement loss as the difference between what the worker can expect in terms of full compensation by staying in the public sector relative to the opportunity cost of working in the private sector. A Table 6. Simulated Exit Rates for Various Categories of Workers under Selected Severance Pay Formulas (percent) Early retirement target Fixed Severance formula, (percentage of workers age 50 and older) Category of worker payment 2 3 4 5 6 70 80 90 100 Age Below 35 9 18 9 9 11 8 12 12 11 10 Between 35 and 49 27 20 26 27 26 28 32 28 25 23 S0 and older 94 88 94 94 88 93 73 82 91 100 Gender Male 31 30 31 31 31 31 31 31 31 31 Female 27 32 27 27 28 28 29 27 25 23 Region of residence Greater Cairo 30 27 29 29 29 29 26 28 29 30 Alexandria and Suez Canal 43 47 43 43 42 44 45 44 43 41 Urban Lower Egypt 29 26 29 29 28 31 29 29 30 28 Rural Lower Egpyt 39 31 39 39 31 39 38 35 35 35 Urban Upper Egypt 27 31 26 27 27 26 34 32 29 27 Rural Upper Egypt 17 17 17 17 19 19 18 18 16 16 Occupation Professional/technical 23 28 23 23 25 23 27 25 24 23 Managertal 79 62 77 77 81 77 75 78 80 81 Production 27 28 27 27 27 28 27 27 27 26 Clerical 30 26 30 30 26 30 28 29 29 28 Services 37 40 34 37 37 31 31 32 33 37 a. Refer to table 4 and the text. Source: Author's calculations based on survey data. Assaad 147 good indicator of the size of displacement losses is the ratio of total compensation in the public and private sectors disaggregated by various worker characteristics. Another good indicator is the number of years of public sector employment that the worker will be denied by displacement. I also found that women, who tend to face strong barriers to entry into wage jobs in the private sector and thus have poorer earnings prospects there, have significantly higher losses than men. There- fore severance programs that do not distinguish along gender lines are likely to result in a feminization of the public sector labor force. The most challenging aspect of estimating displacement losses was to come up with a reliable estimate for the value to workers of the nonwage aspects of public sector jobs. The method pursued in doing this relied on several assump- tions, including the assumption that a marginal group of workers that receives no lifetime rents in the public sector can be identified by worker attributes ob- served in the data. Another assumption was that the ratio of nonwage benefits to monetary earnings is invariant to worker characteristics. These are fairly strong assumptions, and I attempted to carry out several sensitivity analyses to test the robustness of my results to their possible violation. The main result of the simulations was that the severance programs that pro- vide higher payments to long-tenure workers are likely to overpay many work- ers in order to achieve their desired exit targets because losses tend to decline as workers approach retirement. Schemes that base payment on years of denied service are likely to suffer less from overpayment because they offer lower com- pensation to these longer-tenure workers. Moreover, early retirement schemes that target older workers with separate plans are also likely to be cheaper than plans that apply to the entire workforce. I also showed that, in the Egyptian context, the losses of workers with secondary education and above are signifi- cantly higher than those of workers with lower levels of education. Thus if a balance is to be maintained in the educational composition of the remaining labor force, separate schemes, each with its own exit target, must be offered to workers at various educational levels. However, to the extent that operational requirements allow it, it is cheaper to retrench a higher proportion of unskilled workers. Moreover, most standard severance pay schemes are more expensive than a simple uniform payment to all workers that would achieve the same num- ber of exits. The exceptions are schemes based on years of denied service, either alone or in combination with the wage. Another important finding was that the optimal programs generally have compensation payments that combine a fixed component with a variable component that varies across workers. The optimum size of the fixed component is generally anywhere from one-third to half of the total payment. Programs with a purely variable payment, such as those that pay a given number of months of wages per year of tenure or per year of denied service, tend to be more costly than even the uniform payment benchmark. The plans that seem to approximate the estimated displacement losses most closely are early retirement plans that offer workers who are age 50 and older a multiple of their current monthly wage until retirement such that 80 to 90 per- 148 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I cent of these workers voluntarily exit the public sector. The balance of the nec- essary retrenchment is achieved by offering younger workers one of the sever- ance programs described above. In the base case, such a scheme turned out to cost only 5 percent more than the full indexation benchmark that pays workers the exact amount of their estimated losses. Besides producing estimates of the relative costs of various programs, the simulations also produced the composition of exiters and stayers in terms of various observable characteristics, such as age, gender, region of residence, and occupation. The most noteworthy result was that the various severance pro- grams tend to exit workers in managerial occupations in disproportionate num- bers, with nearly 80 percent of them exiting under some schemes. This results from the fact that managers are typically older and that older workers tend to have smaller losses. Several limitations of the analysis should be mentioned. First, by assuming that the returns to tenure in the public sector do not transfer to the private sector but that returns to general labor market experience do, I assumed that private employers value such experience. If that is not the case, reemployment wages will be lower than predicted, and losses will be underestimated. This would be more of an issue for workers with long tenure who would lose more of their specific human capital as they move to the private sector. The loss-tenure pro- files I estimated may therefore underestimate the losses of older workers. Second, some women and possibly older men who leave the public sector will probably prefer to stay home rather than join the private sector. In effect, this means that their reservation wage is below public sector compensation but higher than what they would get in the private sector. In that case, the opportunity cost of their labor should be the reservation wage, not the private sector wage, and the losses of these workers would be overestimated. Third, I assumed that the worker would time the exit in a fairly broad win- dow so as to minimize transitional unemployment. However, some exiters might reduce their labor supply after separation in the face of large reductions in the wage rate and in response to the income effect of a severance payment. As a result, they may not immediately take up a private sector job. If these temporary withdrawals from the labor force are voluntary, however, they do not really qualify as transitional unemployment. They simply mean that the reservation wage of these workers is temporarily higher than their private sector wage. Finally, the strong assumptions needed to calculate nonwage benefits are clearly a limitation. The sensitivity analysis indicated, however, that the main results on the shape of the loss profiles and the relative performance of different severance pay formulas appear to be quite robust to changes in both the magnitude of nonwage benefits and their structure across workers. If workers at different points in their careers place a significantly different value on the nonwage aspects of their jobs, then the loss profiles would change. Short of that, the profiles and the evaluation of the different severance pay schemes that depend on them are fairly robust to the violation of assumptions on the structure of nonwage benefits. Assaad 149 Despite these limitations, this article provided a fairly strong argument for looking at the structure of opportunity costs and wage profiles when designing severance programs. It showed that significant overpayment can be avoided by matching compensation payments to the expected losses of workers. It also pro- vided a method for estimating these losses from standard labor force surveys that are available in most countries. APPENDIX. THE MODEL USED TO ESTIMATE THE VALUE OF NONWAGE JOB ATTRIBUTES AND WORKER LOSSES For any experience level T and public sector tenure NG, the predicted log earnings of a government (or SOE) worker are given by: (A-i) lnE,G(T) = + &1GT - 2G + &3GNG + OG'zG where X, is a vector of observed characteristics, XiG is the inverse Mill's ratio from the multinomial logit selection model into the government, SOE, and pri- vate sectors, and G 1 aG = [&lG 1 &2G 5 &3G ] and 6G are the parameter estimates of the selectivity-corrected earnings equation in the government or SOES. (In equa- tion A-1 the subscript G refers to government, but the same equations would also apply to SOE workers.) Accounting for the fact that the returns to public sector tenure are not transferable to the private sector but that the worker be- gins accumulating tenure in the private sector after displacement, the same worker's predicted log earnings in the private sector are given by: (A-2) lnE1R(T) = ¾X, + &kRT - &2RT + 6C3R(T -1r) + 6R'zG where subscript R refers to the private sector, t is the time of displacement, and fR¾ acR = [a1R, &2R, &3R] and 0R are the equivalent parameter estimates from the private sector earnings equation. To estimate the value of nonwage benefits, I need an expression for lifetime rents, which are given by the difference in discounted earnings streams in the two career paths at entry. The net present values of the discounted estimated earnings over the life of the contract in the public and private sector paths, L,G and L,R, respectively, are given by: (A-3) LXG = Jf E,U exp(-pT)dT + I EG exp(-pT)dT (A-4) L,R = | E1R exp(-pT)dT where p is the discount rate, and %t, tr are the time in the queue and the time of retirement, respectively, both measured from the date of entry into the labor market. E,, represents the worker's earnings while waiting in the public sector 150 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I queue and is given by Et>(T) = PE,R(T), where 0 < p <1. EFG Is the total annual compensation in the public sector, which is assumed to be some multiple rj of public sector earnings as follows: E,G(T) = rE1G(T), where T1 > 1. The marginal group of workers, indexed by m, has zero lifetime rents, imply- ing that the present values of total lifetime compensation in the two career paths are equal: LmG = LmR. All other workers are assumed to have positive rents. Let rl, be the ratio of total compensation to monetary compensation that will equalize lifetime compensations in the two career paths for workers in group 1: (A-5) = L,R - J.~" E,U exp(-pT)dT E,G exp(-pT)dT Thus 'qr, the ratio that equalizes the compensation streams for the marginal group of workers, is given by: (A-6) ilm = max(T,q). Assuming that the ratio of total compensation to monetary compensation is independent of worker characteristics, rq = a1m for all public sector workers. The displacement losses R,G of a public sector worker with experience T are therefore given by: (A- 7) R,G = 11E1 E.Gexp(-pT)dT - ,Rexp(-pT)dT. I T The closed-form expression for the integrals as a function of the earnings equa- tion parameters is: | E,G exp(-pT)dT = 2G expL A )] XIq [P a I 3G] + 2&2GTr -D [P iG -3G] + 2&2Gc ~~ \i~2&2G ~2& 2G exp(-pT)dT = expl-&3]rI ex P - 1 'R t ~R exn &2 X K( [P -lR &3R] + 22RT 2 [P - o 1p - k3R + 262Rtjj Lhere oI2 R tR where FD is the cumulative normal distribution function. Assaad 151 Table A-1. Parameter Estimates for a Multinomial Logit Equation for Selection into Employment in the Government, State-Owned Enterprises, and the Private Sector, Egypt, 1988 Males Females State-owned State-owned Variable enterprises Private sector enterprises Private sector Constant 1.824"* 10.89** 4.347'* 11.330- (1.94) (12.87) (2.19) (6.05) Age -0.041 -0.350' -0.215' -0.454 - (-0.81) (-7.61) (-1.88) (-4.19) Age2 / 100 0.024 0.306" 0.253 0.500*- (0.40) (5.38) (1.62) (3.35) Educational attainmenta Read and write -0.078 -0.683 Fx* 0.455 -1.084 (-0.42) (-3.93) (0.51) (-1.40) Primary -0.451k -1.3904*' 0.847 -1.258 (-1.89) (-5.84) (1.00) (-1.49) Preparatory -0.115 -1.290 -0.230 -1.238 * (-0.43) (-4.91) (-0.29) (-1.79) General secondary -0.709*- -1.539*x 0.002 -1.331" (-2.05) (-4.62) (0.00) (-1.83) Vocational secondary -0.880*x* -2.678* -1.459 '3.900* - (-4.51) (-14.00) (-2.71) (-8.48) Technical institute -1.935X- -3.546k * x -2.697 * _4,739*X (-6.03) (-11.67) (-3.71) (-7 44) University or above -2.164**4 -4.027* -2.792<* -3 794x (-8.58) (-14.35) (-4.22) (-7.40) Bachelor of engineering 1.553 F* 1.791*" 2.308 0.446 (4.14) (4.24) (2.59) (0.38) Bachelor of science 1.724*** 1.075 (3.03) (1.68) Bachelor of commerce 1.222**X 1.533x 1.846-- 0.699 (4.02) (4.48) (3.45) (1 34) Region of residenceb Alexandria and Suez Canal 0.301 -0.219 -0.169 -0 527' (1.68) (-1.12) (-0.55) (-1.60) Urban Lower Egypt -0.619 -0.732" -1.012* -1.359 (-3.50) (-4.07) (-2.96) (-3.89) Urban Upper Egypt -1.934^-" -1.454<*' -2.531"* -1.870- (-8.17) (-7.34) (-3.34) (-3.71) Rural Lower Egypt -1.000`** -1.131 -1.382X' -0.917' (-5.75) (-6.54) (-2.56) (-1.98) Rural Upper Egypt -1.572*' -1.554'** -0.296 -0.046 (-7.19) (-7.60) (-0.44) (-0.07) Parents' characteristics Father farmer or agricultural worker -0.255*' -0.604**-' -0.668 0.284 (-1.87) (-4.40) (-1.32) (0.71) Father educated 0.209 0.083 0.573** 0.922'** (1.21) (0.46) (2.05) (3.06) Mother educated -0.645t' -0 134 -0.613 0.154 (-1.98) (-0.47) (-1.43) (0.42) (Table continues on the following page.) 152 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Table A-1 (continued) Males Females State-owned State-owned Variable enterprises Private sector enterprises Private sector Household-level variable Currently married 0.125 -0.331* 0.216 -0.487 (0.64) (-1.90) (0.66) (-1.59) Number of children fewer -0 474' -0.529 ' than 2 (-1.91) (-2.11) Number of children 3 to 6 -0 064 0.028 (-0 37) (0.15) Number of children 7 to 11 0.303* 0.321* (1.94) (2.00) Other private nonagricultural wage workers -0.171 0.443* (-0 52) (1.65) Other public wage workers -0.079 -0.471-' (-0.47) (-2.77) Log-likelihood -2,171.7 -501 8 Number of observations 2,707 811 Significant at 10 percent. Significant at 5 percent. * * Significant at 1 percent Note: The dependent variable is the worker's sector of employment: y = 0 for government, y = 1 for state-owned enterprises, and y = 2 for the private sector. The parameters of the government equation are normalized to zero. t-ratios are in parentheses. a. Illiterate is the reference category. b. Greater Cairo is the reference category. Source: Author's calculations based on survey data REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Alderman, Harold, Sudharshan Canagarajah, and Stephen Younger. 1996. "A Com- parison of Ghanaian Civil Servants' Earnings before and after Retrenchment." Jour- nal of African Economies 4(2):259-88. Assaad, Ragui. 1996. "Structural Adlustment and Labor Market Reform in Egypt." In Hans Hopfinger, ed., Privatization and Economic Liberalization in the Socialist Coun- tries of the Arab World. Stuttgart, Germany: Klett/Perthes. . 1997. "The Effects of Public Sector Hiring and Compensation Policies on the Egyptian Labor Market." The World Bank Economic Review 11(1):85-118. Diwan, Ishac. 1993a. "Public Sector Retrenchment and Efficient Severance Pay Schemes." Policy Research Department, World Bank, Washington, D.C. Processed. . 1993b. "Public Sector Retrenchment and Severance Pay: Nine Propositions." Policy Research Department, World Bank, Washington, D.C. Processed. Fergany, Nader. 1991. "Overview and General Features of Employment in the Domes- tic Economy." Final report. CAPMAS, Labour Information System Project, Cairo. Pro- cessed. Assaad 153 Flszbein, Ariel. 1992. "Labor Retrenchment and Redundancy Compensation in State- Owned Enterprises: The Case of Sri Lanka." South Asia Region, World Bank, Wash- ington, D.C. Processed. Integrated Development Consultants. 1994. "El Delta Spinning and Weaving Company: Organizational Structure and Human Resources." Final report to Bechtel/PID. Cairo. Processed. Kikerl, Sunita. 1996. "Privatization and Labor: What Happens to Workers When Gov- ernments Divest?" Private Sector Development Department, World Bank, Washing- ton, D.C. Processed. Lee, L. F. 1983. "Generalized Econometric Models with Selectivity." Econometrica 51:507-12. Levy, Anat, and Richard McLean. 1997. "Optimal and Suboptimal Retrenchment Schemes: An Analytical Framework." Policy Research Department, Rutgers Univer- sity, New Brunswick, N.J. Processed. Nunberg, Barbara. 1994. "Experience with Civil Service Pay and Employment Reform: An Overview." In David L. Lindauer and Barbara Nunberg, eds., Rehabilitating Gov- ernment: Pay and Employment Reform in Africa. A World Bank Regional and Sectoral Study. Washington, D.C.: World Bank. Robbins, Donald. 1997. "Public Sector Retrenchment and Efficient Severance Payment Schemes: A Case Study of Argentina." Policy Research Department, World Bank, Washington, D.C. Processed. Tansel, Aysit. 1997. "Workers Displaced due to Privatization in Turkey: Before vs. after Displacement." Policy Research Department, World Bank, Washington, D.C. Pro- cessed. Younger, Stephen. 1996. "Labor Market Consequences of Retrenchment for Civil Ser- vants in Ghana." In David Sahn, ed., Economic Reform and the Poor in Africa. Ox- ford: Clarendon Press. THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 155-83 The Algerian Retrenchment System: A Financial and Economic Evaluation Elizabeth Ruppert The government of Algeria has adopted a new retrenchment system to facilitate labor shedding in a public sector characterized by redundant workers and in an environment of already high unemployment. This article assesses the financial viability of the re- trenchment system and the inherent welfare costs and benefits associated with layoffs. A financial flows model tracks the Unemployment Insurance Fund's revenue and ex- penditure flows during the projected five-year adjustment period. It finds that even in the presence of massive retrenchment (21 percent of formal sector employment), the fund accumulates reserves equivalent to nearly 2 percent of gross domestic product. Because many displaced workers will end up in the informal sector, the resulting pro- ductivity gains or losses depend crucially on the initial level of productivity in the public sector. At the same time, retrenchment entails unambiguous benefits by reduc- ing subsidies to state-owned enterprises, thereby generating efficiency gains. Consider- ing these two effects together, the welfare model estimates that retrenching 13 percent of the formal sector will generate annual net welfare gains ranging from costs of $358 million to gains of $774 million. The Algerian economy is struggling to break away from the central planning model it adopted following independence from France in 1962. Algeria's vast oil reserves and external borrowing during the 1970s and early 1980s financed heavy investment to develop an industrial sector that relies on capital-intensive production technology. Over time, the large public sector gradually lost com- petitiveness. Economic reform was initially undertaken in the late 1980s and has been stop-and-go since then. This period was characterized by political turmoil, during which social and political pressures emerged to derail the economic re- form process. In 1991, renewed stabilization efforts were interrupted by politi- cal crisis resulting from the cancellation of legislative elections and were fol- lowed by backtracking on macroeconomic reforms. It was in this context that the Algerian authorities, facing balance of payments pressure engendered by depressed oil prices, undertook a broad program of stabilization and structural reform in 1994. Elizabeth Ruppert is with the Middle East and North Africa Regional office at the World Bank. This article is the result of a joint effort on the part of the Social and Economic Development Group of the Middle East and North Africa Regional office, and the Public Sector Retrenchment and Efficient Compensation Schemes project undertaken by the Development Research Group at the World Bank. The author gratefully acknowledges the valuable contributions and support of Martin Rama. ©i 1999 The International Bank for Reconstruction and Development/THE WORLD BANK 155 156 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I The transition to a market economy involves restructuring Algeria's public industrial sector by removing or modifying controls and regulations that distort price signals and by opening up to international competition. This is a difficult chore, given the state's dominance in industry; state-owned enterprises account for 75 percent of total formal sector industrial employment. Moreover, public enterprises were (and mostly still are) highly subsidized, faced soft budget con- straints with help from a compliant public banking sector, and served as a ve- hicle for the job creation objectives of the government. These firms therefore became bloated and unproductive but remained intact despite poor performance. Industrial production fell by half during the decade from 1986 to 1996, while industrial employment experienced no net change, indicating an increase in the number of redundant workers, defined here as underutilized. The extent of re- dundancy is unknown, but industrial public enterprises currently operate at a mere 35 percent capacity. In spite of potentially vast redundancies, however, the unemployment rate stands at nearly 30 percent. In the framework of the structural reform process started in 1994, the Alge- rian authorities adopted a national retrenchment system to facilitate layoffs while protecting displaced workers. Concurrently, they introduced hard budget con- straints in public enterprises and tight fiscal restraint in the government sector. The retrenchment system consists of a severance pay component and a monthly unemployment insurance benefit, a feature that is uncommon in developing coun- tries and was only recently introduced in the transition economies of Central and Eastern Europe (see World Bank 1995: ch. 13). All formal sector firms are subject to a mandatory payroll tax specifically for unemployment insurance and are eligible to undertake massive layoffs with access to the system's benefits. The system's stated objectives are to facilitate enterprise restructuring through labor shedding and to provide income support to laid-off workers, mitigating the im- pact of the transition. Although not specifically targeted for restructuring, pri- vate firms can also benefit from greater flexibility in hiring and firing decisions. This new retrenchment system replaced an earlier scheme consisting exclusively of generous severance packages paid in a lump sum to workers at the moment of separation. Only minimal labor reallocation occurred under the old system, which proved untenable for a variety of reasons, chief among them the required union approval of retrenchment and an incentive structure that supported the government's employment promotion policies. It appears that these impediments to labor shedding have been reduced, although not entirely removed, within the context of the new system and the broader structural reform program. This article assesses the financial viability of the new retrenchment system and the welfare implications of layoffs. The existing literature thoroughly ad- dresses the financial returns to retrenchment (see, for example, Haltiwanger and Singh in this issue), the effectiveness of downsizing efforts as reflected by rehir- ing rates, and the impact on the postseparation earnings of laid-off workers (Alderman, Canagarajah, and Younger 1995; Lindauer and Nunberg 1994; and Ruppert 157 Rama and Macisaac in this issue). The main contribution of this article is to provide an economic evaluation of the welfare impact of retrenchment through changes in production and taxation. Algeria's new retrenchment system represents a specific example of a downsizing mechanism, serving as a case study for financial and welfare analy- sis. The financial analysis relies on a simulation model of the Unemployment Insurance Fund (Caisse Nationale d'Assurance Ch6mage-CNAC) that tracks the fund's quarterly revenues and expenditures during the transitional period in which the public industrial sector is restructured through, among other things, massive layoffs. The financing model reproduces the evolution of the CNAC balances from 1995 to 2003 to assess whether the fund is financially sound. In order to analyze the system's impact on economic welfare, I propose a two- sector partial equilibrium model in which the formal sector contains surplus workers who are redundant. The alternative sector, also referred to as the infor- mal sector, consists of workers who are compensated at a wage equivalent to their marginal productivity. The wage gap between the formal and informal sectors creates incentives for job search and unemployment and is specified herein using the Harris-Todaro approach (see Harris and Todaro 1970). Given a con- stant wage gap, eliminating formal sector redundancies through layoffs may result in unemployment at the individual level, but in the aggregate, there is a reallocation of labor toward the informal sector and no change in the equilib- rium unemployment rate. Although certain key variables are unobservable, such that calibration requires simplifying assumptions, the welfare model provides a framework for measuring the changes in aggregate production and the efficiency gains from eliminating subsidies; taken together, the model generates a value of the net welfare cost of layoffs. Section I describes the retrenchment system and its benefit and financing struc- ture and discusses the associated incentives. Section II presents the financial flows model and assesses the system's viability across a range of potential layoff sce- narios. Section III develops the welfare model of economic costs and benefits and measures the net change in welfare associated with projected layoffs follow- ing the switch to the new retrenchment system. It also tests the model's robust- ness through sensitivity analysis. Section IV discusses the policy implications of the models' estimates. 1. DESCRIPTION OF THE SYSTEM The Algerian retrenchment system is a hybrid that resembles periodic sever- ance payments spread over time. It shares its basic contributory structure and benefit design with many Central and Eastern European countries and members of the Organisation for Economic Co-operation and Development. However, the additional financing provided by firms in the form of an initiation fee is unusual. The appendix contains a detailed description of the system. 158 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I Eligibility The retrenchment system was designed for workers and firms in the formal sector, specifically public enterprises, government administration, and a rela- tively small number of private firms. In 1995, 2.8 million workers were em- ployed in the formal sector, comprising 1.2 million public enterprise employees, 1.2 million civil servants, and 400,000 private sector workers. All formal sector firms are required to pay the unemployment insurance payroll tax. Firms also must be up-to-date in their payroll tax payments and must complete each re- trenched worker's application for admittance to benefits and submit it to the unemployment insurance authority. Workers laid off for economic reasons who have been affiliated with social security for at least three years and who receive no alternative earnings are eligible to collect benefits. Benefit Structure The new retrenchment system consists of a severance pay component and an unemployment insurance benefit component. At the time of separation, firms are required to pay retrenched workers a severance package equivalent to one month's salary per year of tenure up to three years, for a maximum of three months of salary. In addition, workers receive a monthly indemnity paid by the CNAC at a level proportional to each worker's salary; a reference wage is calcu- lated as the average of the monthly and the minimum wages and is subject to a replacement rate that declines over the period of eligibility from 100 to 50 per- cent. Benefit levels are subject to a minimum and a maximum, equal to three- fourths and three times the minimum wage, respectively, spanning a fairly nar- row range. The system administers indemnity payments for a period proportional to work history, that is, two months per year of tenure in excess of three years. All beneficiaries are guaranteed benefits for at least one year, up to a maximum of three years. Financing Structure The unemployment insurance component is financed by two sources of rev- enue: a 4 percent payroll tax on all formal sector employees, shared by the em- ployer and employee (2.5 and 1.5 percent, respectively), and an initiation fee paid by firms to the CNAC for each retrenched worker. By charging an initiation fee proportional to the displaced worker's salary and job tenure, the costs asso- ciated with retrenching are made explicit because firms incur a direct cost for each worker laid off that is factored into their decision to retrench. This effect is similar to the experience rating observed in various unemployment insurance systems in the United States. The Algerian system does not perfectly internalize the cost of layoff into the employer's decision. For example, in the event that a worker quickly finds another job, the old employer is still required to pay the initiation fee even though the CNAC pays no benefits to the worker concerned. The sizable initiation fee requirement serves to boost fund reserves by covering Ruppert 159 more than half of the total stream of monthly indemnities to be paid to the beneficiary and is instrumental in covering the initially high benefit outlays of a system facing large layoffs associated with enterprise restructuring. Although the CNAC is a public institution, it is fiscally independent of the central government-an unusual feature even for the transition economies. Fi- nanced entirely by payroll tax contributions and initiation fees, it does not ad- minister or finance any active labor market policies such as job training or place- ment schemes, nor does it have recourse to government funding in periods of reserve shortfalls. Instead, it must either amend its contribution and benefit pa- rameters to restore financial viability or borrow from commercial lenders. Given the size of Algeria's highly subsidized public sector, however, the CNAC (through both payroll taxes and initiation fees) is ultimately financed by taxes on the formal private sector and by government oil revenues. Comparison with the Old System The new system of unemployment insurance replaced an earlier retrenchment scheme that consisted of a lump-sum severance payment proportional to salary and work tenure (equivalent to one month of salary for each year of experience), paid by the enterprise at the moment of separation. A comparison of layoff costs under the two systems demonstrates that the marginal statutory costs incurred by firms under the new system are close to those of the old regime (see table 1). Consider an average retrenched worker with 10 years of experience. Under the old scheme, firms were obliged to pay 10 months of salary immediately to the worker, whereas under the new scheme, firms must pay 8.6 months of salary (3 months severance plus 5.6 months initiation fee), of which only 5 months of Table 1. Comparison of the Financing of the Old and New Retrenchment Systems in Algeria Indicator Old system, before 1994 New system, after 1994 Costs paid by firms Severance (to workers) One month of gross salary Three months of gross salary per year of tenure Initiation fee (to CNAC) None 80 percent of monthly gross salary per year of tenure in excess of three years: two months of salary paid on separation and the balance paid within one year Maximum (severance 15 months of gross salary 15 months of gross salary plus initiation fee) (equivalent to 15 years (equivalent to 18 years of of tenure) tenure) Payroll tax None 2.5 percent of taxable wage bill Costs paid by workers Payroll tax None 1.5 percent of taxable salary Source: Ministry of Labor and Social Protection. 160 THE WORLD BANK ECONOM[flC REVIEW, VOL 13, NO 1 salary are due within the first 3 months following separation. Although it is marginally cheaper to retrench under the new system, the principal gains come from spreading charges over time, thereby easing liquidity constraints. Comparison of the old and new systems presupposes that firms actually laid off workers under the old regime; this was not the case. The labor shedding witnessed since implementation of the new system reflects a reorientation of the Algerian government's priorities, manifested by the government's withdrawal from employment and production decisions. Specifically, public enterprises are no longer mandated to create jobs, in contrast with earlier employment promo- tion policies that were facilitated by indirect subsidies. The fundamental change in the government's role is most readily apparent in the tightening of previously soft budget constraints. Under the old system, the maximum severance payment was equivalent to 15 months of salary; firms found it cheaper in the short run to keep redundant workers on the payroll. The government's reduced subsidy of the industrial sector appears to be the real force driving retrenchment activity. The shift to hard budget constraints (evidenced by emerging salary arrears and interenterprise debt), together with greater autonomy in employment and pro- duction decisions, induced firms to rationalize production and factor costs by reallocating labor. Another indication of the government's new priorities is the diminished role of unions. The required approval of unions under the old system impeded lay- offs. The government eliminated this legal provision and achieved union sup- port of the retrenchment legislation through political maneuvering and collec- tive bargaining. For example, the imposition of the payroll tax was accommodated by the government's agreement to take over a transfer program previously fi- nanced by a tax on firms. Although firms have greater autonomy in their em- ployment decisions under the new system, the influence of unions remains sig- nificant, reflected by the similarity of firms' costs under the two schemes (for example, the same maximum equivalent to 15 months of salary). Redistribution and Incentives The CNAC, although nominally separate from treasury support, is in effect subsidized by civil servants through the employee portion of the payroll tax on the public administration wage bill (and at a more fundamental level by the private sector taxation and government oil revenues cited above). Public ser- vants make up almost half of the eligible labor force, but they are not de facto subject to layoff at present, indicating an implicit subsidy of those workers who are retrenched. Redistribution results from the benefit structure as well; because indemnities are capped, the flat rate contribution constitutes an in- come redistribution from higher wage earners to those actually laid off. And despite the apparent link between benefits and contributions, benefits are com- pressed in favor of lower-wage earners (because they are calculated as a weighted average using the minimum wage). This eliminates any regressive income re- distribution from low-salaried unskilled workers to highly paid skilled em- Ruppert 161 ployees and at the same time provides a minimum level of protection to the affected population. The new retrenchment system effectively changes firms' incentive structure through lower costs and greater autonomy, and because retrenchment is not voluntary, firms avoid efficiency losses associated with adverse selection by re- taining their most productive workers. There is some risk of moral hazard, how- ever; introducing hard budget constraints may give rise to labor hoarding, as witnessed in the transition economies of Central and Eastern Europe, especially for firms in serious financial straits. The benefit structure does not on the whole distort beneficlaries' incentives. Although monthly income support discourages job search, the declining replace- ment rate means that benefits diminish over the course of eligibility, falling by half. The actual level of compensation is a function of the reference salary; al- though beneficiaries receive near-full replacement of the reference salary (an average of 90 percent) for at least 6 months and up to 18 months, the effective replacement rate averages 60 percent of gross salary for the same period. The long duration of benefits, averaging about 20 months and up to a maxi- mum of 3 years, could discourage job search by promoting attachment to unem- ployment, thereby swelling the ranks of the long-term unemployed. An increase in unemployment would be particularly undesirable in Algeria, where there is already a tendency toward entrenched long-term unemployment. For example, the average unemployment spell for workers over age 30 is longer than that for younger workers. The risk of exacerbating unemployment over the long run may have negative dynamic implications for the labor market by shifting it to a higher rate of equilibrium unemployment. Burda (1993) presents a dynamic theo- retical model of equilibrium unemployment and vacancies, and Ruppert (1996) derives the model's application to Algeria. Job search incentives are supported by the system's lack of enforcement. Be- cause the benefit structure and eligibility are conditioned on need, beneficiaries are prohibited from earning alternative revenue. In reality, however, weak insti- tutions and monitoring are likely to result in little actual enforcement, especially outside the formal sector; potential earnings from informal activity, therefore, provide the standard incentives. Early Results In its first two years of operation (1994-96), the new system functioned rela- tively smoothly. Already 42,000 formal sector workers in both state-owned en- terprises and private firms were retrenched under its auspices, representing about 1.5 percent of formal sector workers. The unemployment insurance system was introduced rapidly by using the administrative capacity of the existing social security system. Administrative costs are relatively low as a result; by end-1996, only 124 personnel operated the system in 48 administrative regions. Among the total workers laid off and approved for benefit eligibility, about 65 percent actually receive benefits. After only two years, nearly $20 million had been col- 162 THE WORLD BANK ECONOMIC REVIEW, VOL 13. NO 1 lected in initiation fees, and the monthly revenue inflows from payroll taxes were approximately $11 million.' Because monthly benefit outlays averaged about $3.3 million, the system is more than amply financed to meet its current obligations. Il. ASSESSMENT OF FINANCIAL VIABILITY This section considers the system's financial viability over the medium run. The Financial Flows Model On the basis of work by Worden (1992), I constructed a financial flows model to evaluate the system's financial viability over the short and medium terms. The model monitors the revenues and expenditures of the CNAC by tracking flows into unemployment, flows out of unemployment, benefit outlays, payroll contri- butions, and the average level of initiation fees and their payment over time. The model reproduces the evolution of fund balances quarterly, from 1995 through 2003. The CNAC balance in period t is defined to be equal to the positive finan- cial inflows resulting from the payroll tax, initiation fees, and CNAC balance carried over from the previous period, t - 1, minus the outflows to benefit pay- ments, administrative costs, and social security taxes for beneficiaries. The CNAC balance in period t (Z,) is defined very generally as the sum of revenues minus expenditures plus government transfers in period t plus interest income on the previous period's fund balance, that is, (1) Z, = Rt- Et + Gt + (1 + r,) Zt,I where R, denotes fund revenues, E, is fund expenditures, G, is a net government transfer (positive or negative), and r, is the real interest rate earned on the fund balance carried over from the previous period, Z,1. Furthermore, revenues can be decomposed as follows: (2) Rt=Tr+ F, where T, represents total payroll taxes collected each quarter and Fr is total initiation fees paid into the CNAC. Payroll contributions consist of the firm's portion of the tax, Tf, and the worker's portion of the tax, Tc,, which are applied to the worker's net wage; tq is the ratio of average net wage (net of social benefits and regional adjustment premia on the government-regulated base salary) to gross wage, and w, is the average monthly wage, such that rlw, represents the average taxable monthly wage. Formal sector employment is denoted by Lt, and y, is a tax collection efficiency parameter (between 0 and 1) to reflect imperfect revenue collection. Because the wage w, is defined on a monthly basis, the total 1. Conversions from local currency are made at an exchange rate of 55 dinars per 1 U.S. dollar; all figures are in constant 1995 prices Ruppert 163 tax is multiplied by 3 to obtain quarterly taxes. Total payroll contributions in quarter t are therefore: Total initiation fees, F, reflect the requirement that for each new entrant to benefit eligibility, the initiation fee must be paid in an initial lump sum equiva- lent to two months of salary up front in quarter t, with the balance paid over the next four quarters. The average initiation fee per worker, ft, is calculated as 80 percent of the average gross monthly wage for each year of average work tenure in excess of three years. Let ct represent the separation rate in period t and Ot denote the share of layoffs actually admitted to eligibility. Total initiation fees are defined in equation 4. The first term in equation 4 represents lump-sum payments, and the second term denotes the sum of the initiation fee balances left over from the previous four quarters. That is, (4) F = 2w,O,a,L, t+L 4 l(ft_, -2 ,) 0,Ct 6AZL,- Total expenditures, denoted Et, consist of benefits paid out to system benefi- ciaries, B, the cost of administering the system, A, and charges for social secu- rity taxation, SS, that is, (5) EEt=Bt+A,+SS,. Benefits are paid to those admitted to the system subject to the eligibility criteria and at a benefit level determined in relation to the reference wage. The quarterly outflow of benefits is defined in equation 6, where 4t-, represents the share of beneficiaries entering in period t - l who remain eligible for benefits in period t, and P,-, denotes the average benefit paid to each worker. The product is summed over 12 quarters (l = 1,. . .,12), implying that beneficiaries are not eli- gible for benefits beyond 36 months, and eligible workers separated in period t do not receive benefits until the following period t + 1. Specifically, (6) Bt = 3 1=- The average benefit level paid out, P ,-, is calculated as a weighted average of (a) those beneficiaries receiving the minimum benefit, 3/4 Mt (the share of total beneficiaries is equal to Xt-,), (b) those beneficiaries receiving the maximum ben- efit, 3m,2 (share equal to v, ), and (c) the remaining beneficiaries whose salaries qualify them for benefits according to a declining replacement rate, Pti, of the reference wage, wR . The shares k,, and v_ are based on an assumed distribu- tion of wages. The average benefit level is therefore: (7) P_ z =X 4 4mt-, + vt, 3mt-, + (1- kt- RVt-)Pt-'Wt-' 164 THE WXORLD BANK ECONOMIC REVIEW, VOI 13 NO 1 Recall that the reference wage wR is defined as the mean of the minimum wage and the average wage. Administrative costs are simply A,, and SS, is the total social security contribution made by the CNAC on account of beneficiaries, such that: (8) SS, =132(X 4 ot%6Q ct1L,) where a is the social security tax rate applied to the monthly minimum wage for those remaining eligible for benefits in period t (denoted by the expression inside parentheses). Again, by summing over the range i = 1,. . .,12, I implicitly assume that social security tax payments for new entrants begin from the moment that laid-off workers receive benefits. Finally, the level of employment evolves over time. Employment in the formal sector at the beginning of period t consists of those remaining after retrench- ment of oT,1L_1l and separations to early retirement (pt-l L,1 in period t - 1, as well as newly hired workers joining the formal sector, J,1. That is, (9) L, = (1 - ,,, - pt-i) L_11 + J, 1 Calibration of the Financial Flows Model In order to calibrate the model, I need data on the variables defined above, where available, and assumptions on likely parameter values otherwise. The financial flows model relies on information regarding average wages and the pattern of layoffs observed during the first two years of the system's operation. The lack of time-series data on actual CNAC balances limits the model's accu- racy; however, using observed data points on cumulative fund revenues and expenditures as benchmarks, I rely on the efficiency parameters e and Y to cap- ture, respectively, the unemployment insurance administration's inefficiencies in processing applications and imperfect tax compliance. Assumptions on the model's parameters are summarized in table 2. Because there is little or no en- forcement of compliance with eligibility criteria once a worker receives benefits, I assume that outflows of benefit recipients from eligibility occur only after ben- efits expire. Based on the assurances of the Algerian authorities, there is no transfer of funds to the central government budget, nor any explicit government subsidy of the system's charges (that is, G = 0). The early results described in section I indicate an accumulation of reserves; in practice, funds are starting to be invested in interest-bearing instruments. I set the real interest rate, r,, equal to 0. Finally, in an effort to assess financial viabil- ity under conservative and perhaps pessimistic assumptions about the Algerian labor market, I assume that formal sector employment declines by the rate of retrenchment, (S, and separations to early retirement, (p, and that no new hiring occurs (that is, J, = 0). Ruppert 165 Table 2. Calibration of the Financial Flows Model Variable Valuea Average gross monthly wage, w, (dollars) 239 Monthly minimum wage, m, (dollars) 82 Formal sector employment, Lo (millions of workers) 2.8 Firm's payroll tax, tf (percent) 2.5 Worker's payroll tax, 'w (percent) 1.5 Ratio of net wage to gross wage, -q 0.75 Share of layoffs admitted to benefit eligibility, 0, ol 0.65 6, in 1998 0.95 Tax collection efftciency parameter, y, 0, Average initiation fee per worker based on an average tenure of 10 years, f, (months of salary) 5.6 Average reference wage, wR (dollars)b 160 Average benefit in each quarter of duration (dollars) First quarter (100 percent replacement of wR) 160 Second quarter (80 percent replacement of w1l) 128 Third quarter (60 percent replacement of w'( 96 Fourth quarter (50 percent replacement of WR) 80 Monthly administrative cost, A, (dollars) 41,000 Share of beneficiaries still eligible for benefits at the end of the period, 9, Share of total beneficlaries receiving the minimum benefit, A, d Share of total beneficiaries receiving the maximum benefit, v, d Social security tax rate paid by CNAC on behalf of beneficiaries, a (percent) 15 Net government transfer to the CNAC, G, 0 Real interest rate (on the unemployment insurance fund balance), r, 0 New hires to formal sector employment, J, (millions of workers)' 0 Separations to early retirement, 'p, (percentage of L,)f 0.2 a. Dollar amounts are in constant 1995 prices. b. wRt = l/, (Wf + m). c. t 1is determined by assuming that benefit duration is normally distributed around an aserage of 20 months There is no evidence on the actual distribution of benefit duration; however, the results are not sensitive to this assumption d. ?,, and v, are calculated according to the wage profile in each period, assumed to follow a log- normal distnbution around mean w,. e The budget laws of 1996 and 1997 stipulate zero net recruitment for the civil service f. Based on retrenchment activity to date indicating that, on average, 5 percent of workers in restructured firms were admitted to early retirement Source. Ministry of Labor and Social Protection, Unemployment Insurance Authority, National Statistical Office, author's estimates. Alternative Scenarios In order to assess the financial viabtlity of the unemployment insurance sys- tem, I consider various scenarios based on observed characteristics regarding layoffs and system parameters. Two different layoff rates are presented here: a base case in scenario A and a high case in scenario B. The base case imitates the layoff rate already observed, such that when industrial restructuring is com- plete, 13 percent of total formal sector employment will have been retrenched, amounting to 365,000 of the 2.8 million workers initially employed in the for- 166 THE W'ORLD BANK ECONOMIC REVIEW, VOL 13, NO I mal sector. In the interest of evaluating the limits of the fund's solvency, I as- sume that layoffs occur in a relatively short time frame in which separation rates, c7, peak during 1997-98 and layoffs are completed by the end of year 2000. This time frame is consistent with the Algerian authorities' current plan for comprehensive industrial restructuring and privatization. Under these as- sumptions, the resulting fund balances are positive and growing over time (see table 3); revenue outpaces expenditure, resulting in fund balances that exceed 1 percent of gross domestic product (GDP) by 1997 and rise to nearly 2.5 percent of GDP in the next decade. Would the system continue to be viable in the event of very high layoffs? Scenario B, summarized in table 3, illustrates the effect on fund balances in the case where total layoffs amount to 21 percent of formal sector employment, or 588,000 workers (parameter value selection is discussed in section III). The pe- riod of high layoffs is assumed to begin in 1997, with 330,000 workers laid off over a two-year period, that is, by end-1998. Although net annual cash inflows are near zero in 1998-2000, funds previously accumulated are sufficient to cover this shortfall, as reflected by the overall balance, which consistently exceeds 1 percent of GDP. The results of scenarios A and B in table 3 suggest that the system is amply financed and solvent over the crucial period of industrial restructuring, and, in fact, the system is projected to accumulate significant reserves over time. The Table 3. Algerian Unemployment Insurance Fund Accounts, Alternative Scenarios, 1996-2003 (percentage of gross domestic product) Scenario 1996 1997 1998 1999 2000 2001 2002 2003 Scenario A: 13 percent layoffs Revenues 0 53 0.79 0.84 0.63 0.54 0.44 0.38 0.36 Expenditures 0.12 0.38 0 69 0 44 0.26 0.14 0.02 0.00 Net inflows 0.41 0.41 0.15 0 19 0 28 0.31 0.36 0.36 Balance 0.90 1.27 1 36 1.49 1.70 1.93 2.19 2 45 Scenario B: 21 percent layoffs Revenues 0.53 0.81 0.99 0.84 0.67 0 46 0 35 0 33 Expenditures 0.12 0.42 0.90 0.89 0.65 0.32 0.05 0.00 Net inflows 0.41 0.39 0.09 -0.05 0.03 0.14 0.30 0.33 Balance 0.90 1.25 1.28 1.17 1.15 1.24 1.47 1.73 Scenario C: 13 percent layoffs and no initiation fee Net inflows 0.34 0.13 -0.19 0.02 0.17 0.26 0.36 0.36 Balance 0.80 0.89 0.66 0.65 0.79 1.02 1.33 1.62 Scenario D: 13 percent layoffs and no payroll tax after 1996 Net inflows 0.41 -0.10 -0.35 -0.27 -0.14 -0 09 -0 02 0.00 Balance 0.90 0.76 0.37 0.08 -0.06 -0.15 -0.17 -0.16 Note- Results for 1997 onward are simulations. Source. Author's calculations Ruppert 167 positive balance of the CNAC is partially driven by the high initiation fee, which covers more than half of the total stream of monthly indemnities to be paid to the beneficiary. Insolvent firms unable to meet their retrenchment cost obliga- tions are assumed to be covered by the government. Suppose that the system were financed uniquely by employer and employee payroll tax contributions, as depicted in scenario C in table 3; the resulting balances would still be positive, although considerably lower (on the order of 57 percent lower than the base case scenario A by 1999). The fund balance is quite sensitive to the payroll taxation rate. For instance, lowering the tax on firms from 2.5 to 1.5 percent starting in 1997 would slow the accumulation of fund reserves such that by year 2000 the balance would suffer a 25 percent decline (equivalent to 0.4 percent of GDP). The system would not have adequate resources to finance benefits if there were no payroll contri- bution at all. In light of the reserves accumulated by end-1996, however, elimi- nating the payroll tax starting in 1997 would not dissipate fund reserves until 2000 (see scenario D in table 3), and the system would thus be viable during the period of major layoffs. The financial flows model presented here permits an assessment of the system's financial solvency across a range of layoff scenarios. It concludes that Algeria's CNAC is adequately financed even under high layoffs and is likely to accumulate significant reserves. The fund's financial viability was, in fact, a principal objec- tive of the system design. Given the macroeconomic constraints attendant on Algeria's ongoing adjustment program (requiring the elimination of large fiscal deficits), it was crucial to avoid a budget implication. III. EVALUATION OF WELFARE COSTS The system should be evaluated not only from a financial point of view but also from an economic point of view; this is difficult given that the initial condi- tion of the labor market reflects a distorted equilibrium. The Algerian labor market is characterized by high and increasing unemployment; the unemploy- ment rate rose steadily from 17 percent in 1986 to 28 percent in 1995, illustrat- ing the economy's inability to absorb new entrants. The predominantly public formal sector is subject to extensive labor market regulation. The total burden of social taxes levied on the wage bill (shared by workers and firms) amounts to 34 percent, up from 32 percent under the old regime. Social taxes cover pen- sions, social security, accident insurance, unemployment insurance, early retire- ment, and social works. In addition, wages are distorted and downwardly rigid due to a salary grid system not subject to market rules. Public sector wages depend on a government-defined base salary that is augmented by a series of premia for merit, hardship compensation, regional considerations, and various entitlements, most of which are determined through collective bargaining. The labor market is characterized not only by excess supply, as evidenced by unemployment, but also by redundancies, primarily in the overstaffed public 168 THE WORLD BANK ECONOM.1IC REVIEW, VOL 13, NO 1 sector. Before implementation of the new retrenchment program, public firms adhered to the government's employment promotion policies, and firms were rewarded by explicit and implicit subsidies that kept them afloat even though they were not competitive. State-owned firms are generally inefficient and un- likely to produce on the production possibilities frontier. Unfortunately, little information is available on the marginal productivity of public industrial firms, except that average industrial productivity fell by half during 1986-96. I can conclude only that their marginal productivity lies somewhere between zero and the level that a similar private firm facing the same prices and wages would achieve. The extent of redundancy is integral to this analysis; it is therefore crucial to be clear about its measurement. Redundancy can be defined from the firm's perspective (also referred to as the private perspective), in which wages exceed the marginal productivity of labor in the firm, or from the social perspective, in which productivity in the firm is lower than productivity elsewhere. Under cer- tain conditions, workers may be redundant from both perspectives-for example, if wages in the public and alternative sectors are equal but public wages exceed marginal labor productivity in the public sector and private labor earnings are equal to marginal labor productivity in the alternative sector. In this example, the public sector worker who is redundant from the firm's perspective is also redundant from the social perspective. These conditions do not hold in the Alge- rian case, however, because distorted high public wages exceed average labor earnings in the alternative sector. Consider instead a public sector worker with zero productivity; within the realm of possibility in Algeria, this low-case sce- nario depicts a worker who is redundant from both perspectives. Under either definition, redundancies indicate a distorted labor allocation, suggesting that policy adjustments should be considered within the context of a second-best approach. In the presence of multiple distortions, removing only one, namely the sub- sidy, will not return the market to the efficient labor allocation resulting from fully flexible wages, nor will it move public sector firms onto their production frontiers in response to a single new market incentive. This economic evaluation therefore takes a second-best approach; that is, as long as distortions remain, retrenchment represents only a second-best policy. The analysis aims to isolate and quantify the effects of reducing private redundancies. In this section, I de- velop a model to estimate the relative welfare costs and gains associated with retrenchment in that environment. The Welfare Model The assessment of welfare costs is separated into two main effects: changes in production and taxation. Following implementation of the new system concur- rent with an end to the government's subsidy of public enterprises, the model predicts that firms retrench workers. At the level of the individual, a displaced worker either enters unemployment, finds another formal sector job, or relo- Ruppert 169 cates to the alternative sector. The model assumes that those laid-off workers who relocate to the alternative sector are compensated below the formal sector wage. By construction, the wage gap between the formal and alternative sectors is constant and depends on the level of unemployment; in the aggregate, there- fore, the decline in formal sector employment is equal to the increase in alterna- tive sector employment, with no change in the long-term unemployment rate. Because this is an equilibrium analysis, any transitional fluctuations in unem- ployment are not considered. At the new steady-state equilibrium, the total change in production associated with layoffs is measured by the gap in actual marginal productivity between the formal and alternative sectors. This impact on produc- tion, whose sign depends on the ex ante level of retrenched workers' productiv- ity, is offset or compounded by efficiency gains from eliminating the subsidy (the second main effect). Because the government's expenses are reduced, the diminished need for fiscal resources leads to lower taxation. The net welfare cost of retrenchment under the new system is therefore measured as the sum of changes in output and distortions from taxation. In order to analyze the retrenchment system's impact on economic welfare, I posit a two-sector model with homogeneous workers. The formal sector em- ploys L workers who are compensated at an average wage w. Government sub- sidies and regulation support the wage w above its market-clearing level. Some portion of these workers is assumed to be redundant from the firm's perspective, that is, their wage exceeds their marginal productivity. The alternative sector, also referred to as the informal sector, consists of workers who produce the same good using less capital-intensive technology. Workers in this sector are compensated by labor earnings wa equivalent to their marginal productivity and below the formal sector wage. The formal sector wage is assumed to be con- stant, due to existing regulations and a government-defined salary grid. The model uses a Harris-Todaro specification of relative wages in the two sectors, such that the higher formal sector wage induces workers to search for formal jobs until the expected returns to job search are just equal to informal sector labor earnings; this earnings gap determines the unemployment rate. The Harris-Todaro setup suffers from well-known limitations, chiefly that the prob- ability of earning the formal sector wage is equal to 1 minus the unemployment rate. This restriction implies that all formal jobs turn over every period such that everyone in the labor force has a uniform probability of being hired. Harris and Todaro effectively ignore other factors that affect workers' true likelihood of finding a formal sector job, as well as the fact that individual expectations tend to deviate from actual probabilities. The model could be extended to correct for these shortcomings (as addressed below in the calibration). The model assumes that there are three potential and mutually exclusive states of activity for labor force participants: employment in the formal sector at wage w, work in the informal sector at alternative earnings wa, or unemployment. In reality, there is likely to be some overlap among these states of activity; for example, some public sector workers may undertake informal activity for addi- 170 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I tional earnings. By assuming no overlap, the analysis may overestimate the net production gains for displaced moonlighters if their reallocation to the informal sector does not entail an increase in informal activity. The total labor force, denoted LF, is assumed to be constant and is defined as follows: (10) LF=L+I+ U where L is formal sector employment, I represents the informal sector, and U is total unemployment. Furthermore, the unemployment rate, u, is defined as the share of unemployed workers in the total labor force: (11)LF - L -I LF The Harris-Todaro hypothesis implies that workers equalize the returns to job search in the formal sector and potential labor earnings in the informal sec- tor, wa. Assuming that the probability of finding a formal sector job is equal to 1 minus the unemployment rate, (12) w (1 -U) = Wa. Workers who remain unemployed have no earnings. The model relies on the central assumption that the marginal productivity of labor in the informal sector is independent of the employment level in that sector, such that wa is constant. Because the formal sector wage w is constant, equation 12 implies that the un- employment rate u is constant as well. Whereas this result appears somewhat rigid, the evaluation ignores transitional unemployment because it is an equilib- rium analysis. Using observable data on formal sector wages and unemploy- ment, equation 12 provides a measure for wa. Solving for u and substituting into equation 11 yields the employment level: (13) L+I=LF LF where the right-hand side is constant, implying that the sum of L and I is also constant. This implication provides the basis for evaluating the effect on pro- duction of a change in formal sector employment resulting from enterprise re- structuring. In the aggregate, any retrenchment in the formal sector constitutes an equivalent expansion of the informal sector. In order to measure the effect of such a change, I first need to determine the difference in productivity between the two sectors. Productivity in the informal sector is given by the Harris-Todaro hypothesis, the observable formal sector wage, and existing unemployment. However, the actual labor productivity in state-owned enterprises is unknown. In order to cover a range of plausible outcomes, I consider two extreme assumptions: re- dundant workers in the public sector have zero productivity, and public firms Ruppert 171 Figure 1 Retrenchment and Changes in Production with No Subsidy Remaining (s, = 0) Wage, w w \4 Informal sector wage, wA 6 F Y'(LO) B cx E F Informal sector wage, w, G / G H y'(L) + s + h L, Lo Formal sector employment L Note See text for explanation y is production, S is the subsidy per worker, and h is the severance cost per retrenched worker produce on their production frontier. The reality is likely to fall somewhere between these lower and upper bounds. Under the assumption of zero productivity (in which case workers are redun- dant from both the private and social perspectives), the productivity of a worker retrenched from the public sector and reallocated to the informal sector will increase from zero to wa. The total change in output resulting from retrench- ment is therefore equal to the number of workers laid off between period 0 and period 1 times the new level of productivity, namely wa. In figure 1, total layoffs are depicted by Lo - L, on the horizontal axis (zero subscripts denote initial values under the old retrenchment system), and productivity increases are given by wva on the vertical axis; the total increase in output is thus equivalent to G + H. At the other extreme, suppose that redundant workers have positive produc- tivity on the efficiency frontier. In the formal sector, firms produce output y with labor L and fixed costs C and face decreasing returns to labor. The cost of labor is the gross wage w (including payroll taxes).2 Under the old severance pay system, Algerian firms faced distorting labor regulations in the form of sever- ance charges for retrenching workers, a rigid salary structure, and high wages 2 The gross wage w is equal to Wne, (1 + 0), where r, denotes the payroll taxation rate, including social security contributions. 172 THE WORLD BANK ECONOMIC REVIEW, VOL 13 NO 1 supported by government subsidies to maintain employment at some level Lo. The profit maximization problem was: (14) Max y(L) - wL + so L - ho (Lo- L) - C L where so represents the subsidy per worker and ho represents the severance cost per retrenched worker. The resulting first-order condition is: (15) y'(L) = w - so - ho. Formal sector workers Lo have a marginal productivity (w - so - ho) under the assumed upper bound of productivity but receive wage w. The initial equilib- rium of Algeria's formal sector labor market (before the change to the new re- trenchment scheme) is described by point Lo, w in figure 1. When so = ho = 0, however, the labor demand curve shifts to the left, resulting in a lower employ- ment level at the fixed formal wage it, (L1 in figure 1, where L1 is less than Lo). Excess workers, Lo - LI, are therefore redundant from the firm's perspective because their wage exceeds their marginal productivity. Whether they are re- dundant from the social perspective depends on the relative magnitude of earn- ings in the informal sector, wa. Introducing the unemployment insurance system changes the cost structure of firms in the formal sector. The severance cost per retrenched worker declines to hb, which captures both the severance payment and the initiation fee. I as- sume that there is a simultaneous reduction in the subsidy to sl, providing the impetus to retrench workers who are redundant from the firm's perspective (evi- denced by newly hard budget constraints and actual retrenchment under the new system). Firms also face the additional cost of the payroll tax for unemploy- ment insurance, Tul. The new profit maximization problem becomes: (16) Max y(L) - w(l + tui)L + s L - h1 (Lo - L) - C L with first-order condition: (17) y'(L) = w(I + ru0) - s - h1. If all subsidies are eliminated such that s, = 0, redundant workers Lo - L, are no longer supported by subsidies to formal sector firms and are retrenched, and the new equilibrium in the formal sector is described by point L1, w in figure 1. The magnitude of the labor reallocation depends on the costs faced by firms under the new retrenchment system. The resulting net change in total produc- tion is equal to the change in the marginal products of all workers retrenched, which depends on the magnitude of the ex ante marginal productivity with re- spect to alternative earnings. Given that all workers Lo - LI are redundant from the firm's perspective, production will increase if workers are also redundant from the social perspective, and production will decline otherwise. Ruppert 173 Under the assumed upper bound of productivity-that is, marginal produc- tivity is equal to y'(L)-the net change in production will depend on whether alternative informal sector earnings are less than or greater than y'(LO). In figure 1, for wa less than y'(LO), retrenched workers move from productivity y'(LO) down to wa, and the fall in production is equal to A + B + C + E + F. By contrast, for alternative earnings greater than y'(LO) (denoted WA in figure 1), the employ- ment shift to the alternative sector causes production to change by D - A; as the productivity of workers laid off at the margin rises from y'(LO) to WA, the gains in production (denoted by area D) are offset by the losses in production for workers subsequently retrenched whose higher marginal productivity actually falls to WA (denoted by area A). Thus far I have addressed only the changes in production associated with retrenchment. However, introducing the unemployment insurance system has another key effect, namely the associated decline in subsidy that offsets produc- tivity losses. If, for example, firms retrench all privately redundant workers Lo - Ll, then the subsidy is eliminated completely, easing pressure on the government's fiscal resources. Because taxation entails deadweight losses, reducing taxation gives rise to efficiency gains. The net welfare cost of retrenchment in Algeria will depend on the labor market's initial conditions in terms of the relative marginal productivity of workers in the formal and informal sectors and the marginal cost of public funds. In order to measure the changes in output, I need information on the key param- eters that determine the productivity of workers in the formal sector. With esti- mated values of the subsidy, s, the retrenchment cost per worker, h, and the unemployment rate, u, I use the first-order conditions (equations 15 and 17) to measure the wedge in productivity per worker between the two sectors. To- gether with information about the level of privately redundant workers and the gains from reduced government subsidy, I estimate the net welfare cost associ- ated with layoffs. To derive the initial level of productivity (w - so - ho), I need to know the extent of private redundancies. Suppose that all subsidies are eliminated (s, = 0) and public sector firms shed exactly all workers who are redundant from the firm's perspective; then equation 17 can be solved for hb because Cul and w are known. From the comparison of firms' retrenchment costs under the old and new systems (in table 1), I can solve for ho in terms of hl. Suppose, instead, that not all privately redundant workers are laid off and some degree of subsidy remains (s, > 0); this seems plausible in the case of Alge- ria. In figure 2, layoffs equivalent to Lo - LI are less than total initial redundant workers Lo - L2. For redundant workers whose productivity is zero (the lower bound of the potential range), partially eliminating redundancies leads to pro- duction gains denoted by area K that are smaller than if all redundant workers Lo - L2 were retrenched. For worker productivity at the upper bound, marginal productivity rises from y'(LO) to y'(Ll) under partial layoffs but still falls short of 174 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 Figure 2 Retrenchment and Changes in Production with Some Remaining Sitbsidy (s, > 0) Wage, ua IV informal sector wage, to '2 1iFormal sector employment. L Note See text for explanation y Is productuon, s is the subsidy per worker, h is the sev erance cost per retrenched worker, and t,, is the firm's payroll tax for unemployment insurance wage w.A The resulting production losses denoted by areas I and J are smaller than they would be if all redundancies were eliminated. The only missing element needed to solve for the initial subsidy so is the pro- duction function, whose first-order condition describes the demand for labor. I assume that 6 is elasticity of labor demand. Solving for L under the old and new retrenchment systems (that is, for Lo and LI) and combining the two yield an expression for the rate of total layoffs a as a function of the marginal labor productivities under the two systems, (18) 1 - 6 = °-s°-h1 Now that I can derive the marginal productivity of the formal sector from equation 18, a measure of productivity in the alternative sector is necessary to estimate potential production losses. This follows directly from equation 12, where w, is a function of the unemployment rate. Finally, the welfare costs associated with lost production derived above are offset by efficiency gains from lower subsidies; the marginal subsidy so falls to s, (s, > 0) under the new retrenchment system. The government's lower burden 3. Note that only the case w4 < y'(L,) is depicted Ruppert 175 with respect to public enterprises implies a lower equilibrium level of tax rev- enue. Because the marginal cost of public funds is positive, this saving translates into real economic gains. (The welfare model ignores the distribution of costs.) To quantify these gains, I need information on the marginal cost of funds. Calibration of the Welfare Model The key data necessary to calibrate the model are summarized in table 4. Introducing the unemployment insurance system raises the effective magnitude of the payroll tax from 32 to 34 percent; because payroll taxes are levied on net salary, which averages 75 percent of gross wages (described by rj in section II), the resulting increase in the tax rate on gross wages is 0.015. The relative mag- nitudes of bo and hb are derived from table 1. I rely on Hamermesh (1993) for the value of labor demand elasticity 6. Hamermesh surveys the literature on measures of aggregate constant-output labor demand elasticity for homogeneous labor and concludes that for industrial countries, it is likely to fall in the range [0.15, 0.75]. (By convention, labor demand elasticity is measured negatively, that is, in terms of -6, such that 6 > 0.) Recent estimates of labor demand elastic- ity in Morocco and Argentina also fall within the range (0.36 for state-owned enterprises in Morocco and between 0.30 and 0.75 in Argentina), suggesting that it is not unreasonable to apply the Hamermesh values to Algeria (see Guasch and others 1997 and World Bank 1997). In order to avoid underestimating the production losses and overestimating the efficiency gains of retrenchment in Algeria, I assume that 6 takes a value equivalent to the upper bound 0.75. The model's results are not very sensitive to this assumption (as discussed below). The model additionally requires a measure for the cost of public funds in order to calculate the efficiency gains from lower subsidies. Most available evi- dence refers to industrial countries: Browning (1987) finds that the marginal welfare cost of a tax on labor ranges from 32 to 47 percent in the United States; Hansson and Stuart (1985) estimate a range of 69 to 129 percent of revenue raised by a tax increase in Sweden. Developing countries are likely to encounter relatively higher costs due to less efficient tax administration and the use of less efficient tax instruments; Ahmad and Stern (1987) estimate welfare costs associ- ated with a range of tax instruments in India at 77 to 85 percent. To calibrate Table 4. Calibration of the Welfare Model Variable Value Average gross monthly wage, w (dollars)a 239 Net payroll tax for unemployment insurance, t,1 (percent) 1.5 Ratio of severance cost per retrenched worker under the new and old retrenchment systems, hblh, 0.86 Constant-output elasticity of demand for labor, 8 0.75 Marginal cost of funds (percent) 75 a. In constant 1995 prices. Source: Author's estimates. Table 5. Annual Net Welfare Gains from Retrenchment, 10 Scenarios Variable 1 2 3 4 S 6 7 8 9 10 Layoffs (percent) 13 13 21 13 13 13 13 21 13 13 Redundancies (percent) 13 21 21 50 65 13 21 21 50 65 Ratio of informal to formal sector wages, w,Iw 72 72 72 72 72 50 50 50 50 5( Initial subsidy per worker, soIw (percent) 4.7 8.6 8.6 22.3 29 2 4.7 8.6 8.6 22.3 29.2 Final subsidy per worker, s,lw (percent) 0 4.2 0 18.9 26.2 0 4.2 0 18.9 26.2 Ratio of initial marginal productivity to gross wage, (w - S- ho)lw (percent) 93.6 89 7 89.7 75.9 69.0 93.6 89.7 89.7 75 9 69.0 Change in production (millions of dollars) Maximum productivity -258.7 -216.7 -385.1 -68.1 2.1 -488.4 -446.3 -756 1 -297.7 -223.3 Zero productivity 751.5 751.5 1,213.9 751 5 751.1 521.9 521.9 843.0 521.9 521.9 Efficiency gains (millions of 36.6 88 3 108.4 252.0 322.4 36.6 88 3 108.4 252 0 322.4 dollars) Net welfare gains (millions of dollars) Maximum productivity -222.2 -128.4 -276.8 183.9 324.5 -451.8 -358.0 -647 7 -45.7 99.0 (-0 5) (-0.3) (-0.7) (0.4) (0.8) (-1.1) (-0 9) (-1.6) (-0.1) (0.2) Zero productivity 788.1 839 8 1,322.3 1,003.5 1,073.9 558.4 610.2 951 4 773 8 844.2 (1.9) (2.0) (3.2) (2.4) (2.6) (1.4) (1.5) (2 3) (1.9) (2.0) Note- Annual welfare gains assume that retrenchment is complete. The share of GDP is in parentheses. Source: Author's calculations Rup pert 177 the welfare model, I assume that the marginal cost of funds in Algeria is equal to 75 percent, such that each dollar reduction in total subsidy translates into a gain of 75 cents. The welfare cost of retrenchment depends on actual layoffs and the magni- tude of redundancies. Information on redundancies is generally unavailable, however, and imprecise at best, compounded by the fact that subsidies are largely implicit. I consider four values for redundancies as a share of formal sector em- ployment: 13, 21, 50, and 65 percent. The value 13 percent reflects firms' ob- served labor-shedding behavior to date as an indication of the degree of redun- dant workers. The extent of redundant workers may be understated, however, based on the assertion that some degree of subsidy is likely to persist, enabling some redundant workers to remain employed in the formal sector. The value 21 percent is based on data from Egypt; El Khawaga (1993) finds public enterprise redundancies on the order of 21 percent using a 1986 Egyptian census. The value 50 percent reflects the fact that nonoil industrial value added declined 50 percent in real terms over the past decade, concurrent with no net change in employment. And finally, the value 65 percent is based on estimates of idle pro- duction capacity suggested by average capacity utilization rates of 35 percent in 1996. Whereas all of these values define redundancy from the firm's perspective, the very large magnitudes suggest that public industrial output is far from the production possibilities frontier, implying very low or zero marginal productiv- ity for the workers in question. In order to cover a range of plausible scenarios, the analysis considers the 10 combinations of redundancy, layoffs, and unemployment delineated in table 5 and generates an upper and lower bound of net welfare gains associated with each combination. Because the model relies on the Harris-Todaro hypothesis for measuring informal sector earnings, the unemployment rate is central to esti- mates of production losses. Official statistics from the Algerian Ministry of La- bor and Social Protection estimate unemployment at 28 percent of the total labor force, translating into an alternative wage equal to 72 percent of w and a formal sector hiring rate equivalent to 72 percent (from equation 12). Scenarios 1 through 5 reflect this value. The actual hiring rate (the probability of finding a job) is likely to be much lower, however, given the lack of job turnover. I there- fore consider a lower hiring rate of 50 percent in scenarios 6 through 10; infor- mal sector earnings are thus assumed to equal half of the formal sector wage, averaging close to the minimum wage. Results I now have all the elements necessary to estimate the welfare effect of re- trenchment under the new system. Table 5 summarizes the results, disaggregat- ing the net welfare gains into the change in production and the gains in effi- ciency. Under the 10 scenarios, changes in annual production measured at the upper bound of marginal productivity range from losses of $756 million (sce- nario 8) to gains of $2 million (scenario 5). The wide disparity is driven by the 178 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 gap between the alternative wage and the initial marginal productlvity y'(L.) across the various scenarios. In other words, production losses are dominated by the rectangle J depicted in figure 2 (not drawn to scale). At the lower bound of zero marginal productivity, production gains range from $522 million to $1.2 billion annually by the end of the enterprise restructuring period (estimated at five years in section II). Lower marginal productivity of public sector workers actually leads to higher welfare gains, as workers are reallocated to the informal sector. Note that the increase in production from zero productivity to average informal sector productivity is invariant with respect to redundancies but rather depends on the number of layoffs. The efficiency gains from reducing taxation are more modest but nevertheless significant, ranging from $37 million to $322 million per year. Which scenario best describes the Algerian situation? I propose that scenarios 7 and 9, with a lower informal wage, layoffs of 13 percent, and private redun- dancies of 21 and 50 percent, respectively, most closely resemble the reality of the Algerian labor market. Retrenchment on the order of 13 percent leads to annual welfare effects ranging from costs of $358 million (0.9 percent of GDP) to gains of $610 million (1.5 percent of GDP) under scenario 7. At the upper bound of marginal productivity, the drop in annual production that results from shift- ing employment to the informal sector, depicted by areas I and J in figure 2, amounts to $446 million; under zero productivity, by contrast, annual produc- tion increases by an estimated $522 million, depicted by area K in figure 2. The associated efficiency gains resulting from a cut in the marginal subsidy are esti- mated at $88 million. Under the higher redundancies posited in scenario 9, re- trenching 13 percent of the formal sector yields annual net welfare gains ranging from -$46 million to $774 million. The net welfare gains resulting from these two cases can be easily compared with other scenarios; indeed, the model can be extended to predict welfare costs under any plausible retrenchment scenario. The estimates derived herein are strik- ing because they illustrate that large-scale retrenchment may significantly in- crease welfare-by up to 2 percent of GDP. At the same time, however, the wide range of estimated welfare gains suggests that without more data on unobserv- able variables such as redundancies and informal sector wages, the model can- not generate a conclusive estimation of the economic cost of retrenchment. Because the range of net welfare gains depends on the magnitude of private redundancies, I consider the sensitivity of results to this parameter and find it to be modest and fairly uniform across the 10 scenarios. A 1 percentage point increase in the degree of redundancies results in only small changes in welfare costs that average zero at the upper productivity bound and are negligible at the lower bound (except when all redundant workers are laid off). Efficiency gains are on average $7 million higher, due to the greater initial distortion reflected by higher redundancies. The results of this analysis are moderately sensitive to parameter assumptions about the elasticity of labor demand, the level of unemployment, and the mar- Ruppert 179 ginal cost of funds. The degree of sensitivity to these parameters, illustrated in table 6, is assessed relative to scenario 7. I consider, for example, a labor de- mand curve that is less elastic by 1 percentage point, that is, 3 equal to 0.74. The resulting net welfare gains are 0.7 percent higher than in scenario 7 at the upper bound of productivity, with an even smaller increase in welfare gains at the lower bound of productivity (0.2 percent). Less responsive labor demand im- plies smaller production losses and greater efficiency gains by cutting a higher initial subsidy. For unemployment, the effect is more significant. An unemploy- ment rate of 27 percent results in a higher alternative wage, raising net welfare gains by 2.9 percent (at the upper bound). Alternatively, a small decline in the marginal cost of funds from 75 to 74 percent lowers the net gains slightly (by 0.2-0.3 percent), because reducing the subsidy engenders smaller efficiency gains; the difference translates into additional costs of $1 million annually. IV. POLICY IMPLICATIONS AND CONCLUSIONS Does the Algerian retrenchment system effectively and efficiently achieve its objectives of facilitating layoffs and easing the transition costs by providing in- come support to affected workers? An affirmative answer to this question would suggest that the system could be imitated in other developing countries. Whereas the above analysis concludes that the CNAC is financially sound, the scope for its application elsewhere depends critically on institutional capacity. In light of its complicated benefit design and intensive administrative and monitoring require- ments, the Algerian system as is appears ill-suited for less-developed countries, although simplifying modifications could expand its applicability. This article addresses the more fundamental issues of financial and economic costs by evaluating the Algerian retrenchment system in terms of its financial viability and the welfare costs associated with layoffs in a labor market charac- terized by high unemployment and redundancies. Because retrenchment is some- what cheaper under a system that allows greater flexibility in labor decisions, firms are able to shed redundant workers, thus raising the average level of labor Table 6. Sensitivity Analysis Percentage change in welfare gains Change in parameter assumption Maximum productivity Zero productivity Constant-output elasticity of demand for labor, o, declines from 0.75 to 0.74 0.7 0.2 Unemployment rate declines from 28 to 27 percent 2.9 1.7 Marginal cost of funds declines from 75 to 74 percent -0.3 -0.2 Note: Results of the sensitivity analysis are relative to scenario 7, in which layoffs are 13 percent, redundancies are 21 percent, and the informal sector wage is half the formal sector wage (see table 5). Source: Author's calculations. 180 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I productivity and reducing the wage bill. The financial flows model presented in section II assesses the system's financial viability and projects a buildup of CNAC reserves on the order of 2 percent of GDP, implying that the system does not risk insolvency. These results have certain policy implications. The fact that the system is not only financially viable but also likely to accumulate considerable reserves indi- cates potential overfunding and raises questions regarding the use of funds. This is especially important in an inflationary environment; annual inflation around 20 percent during 1995-96 dissipated the real value of accumulated reserves. There is certainly scope for improving the financial efficiency of the new system, specifically by lowering the initiation fee or the payroll tax rate or by raising benefit levels. Although a cut in the payroll tax would reduce the burden on firms and workers, tax rates could be raised to meet unexpected financing needs, according to a general solvency provision in the legislation. Because Algeria faces a critical period of adjustment, this option would reduce labor costs in the im- mediate run, postponing higher taxes to a period in which firms would be better able to respond. Although the financial evaluation suggests that the unemployment insurance system is adequately financed, the economy may incur costs in the form of lost output under mass layoffs. The model presented in section III calculates the welfare cost of retrenching formal sector workers. Assuming that the aggregate reduction in formal sector employment due to retrenchment is equal to the ag- gregate increase in informal employment, the welfare model estimates total pro- duction losses (in the case of redundancy from the firm's perspective) or gains (for socially redundant workers) associated with layoffs that are offset or com- pounded by efficiency gains from cutting subsidies. These efficiency gains are generated by the fact that the government's marginal cost of funds is positive; reduced subsidy of public enterprises consequently diminishes the government's need for fiscal revenue. The net welfare cost of retrenchment depends on the initial number of redundancies and their ex ante marginal productivity, the ex- tent of layoffs, and informal sector earnings. The model estimates welfare costs for the two most likely outcomes, described in scenarios 7 and 9 in section III. Retrenching 13 percent of the formal sector is estimated to generate annual net welfare gains ranging from costs of $358 million to gains of $774 million. Whereas this range of potential gains appears rather broad for deriving precise conclu- sions, the model provides a method of evaluation that will be more useful when the necessary data are available. There are several directions in which the welfare model could be extended to address related issues. For example, it could be refined by introducing labor supply effects on informal sector earnings following the influx of laid-off work- ers. Alternatively, the dynamic potential of the private (including informal) sec- tor could be incorporated into the analysis, resulting in larger estimates of wel- fare gains. Whereas the welfare model takes a partial equilibrium approach, a general equilibrium analysis could account for the additional costs associated Ruppert 181 with transitory shocks to aggregate demand in the wake of massive retrench- ment. Furthermore, large layoffs could have dynamic implications for the CNAC's financial viability, if firms are unable to make the necessary initiation fee pay- ments, as well as for sectoral and economywide output as a whole. If, for ex- ample, restructuring results in widespread firm closures, additional constraints on growth could emerge with respect to aggregate demand, the availability of inputs to production, and fiscal pressures large enough to affect the macro- economy. There is also scope for assessing the short-term effects of retrench- ment on production losses because reallocation to informal activity is not in- stantaneous; this aspect cannot be addressed using equilibrium analysis. I raise these suggested extensions as areas for future research. Although redundancies involve costs to efficiency, eliminating them completely may be undesirable: there is a risk that layoffs resulting from cutting the subsidy to public firms will worsen welfare, consistent with the theory of the second best with respect to removing a single distortion in the context of many. These con- flicting potential outcomes highlight the complexity of public sector downsizing and the inherent risks to the economy as a whole. It is not clear from table 6 exactly what would be an optimal retrenchment strategy, although the welfare model is a tool for answering this question. The optimal level of layoffs there- fore cannot be determined without detailed information on the number and ac- tual level of marginal productivity of redundant workers. The analysis presented here is useful because it provides a framework for measuring financial and eco- nomic costs under various scenarios and assesses the cost impact of changing the system's parameters. In light of Algeria's various labor market distortions, the new retrenchment system appears to meet the objective of facilitating labor shed- ding, but the magnitude of the net economic gains remains somewhat uncertain. APPENDIX. THE ALGERIAN RETRENCHMENT SYSTEM Elhgibthlty criteria for the worker Laid off for economic reasons. Receives no other earnings, except for unpaid leave. Ineligible for early or regular retirement. Affiliated with social security system for at least three years. Paid unemployment insurance contributions (1.5 percent of taxable wage) for at least six months prior to layoff. Employer paid the initiation fee to the unemployment insurance fund. Registered "job seeker" with employment office for at least three months. Does not refuse job offer or retraining. Eligibility criteria for the employer Firm must be up-to-date in payment of social security contributions and unem- ployment insurance contributions (employer portion of unemployment insur- ance tax = 2.5 percent of taxable wage bill). 182 THE WORLD BANK ECONOMIC REVIEW. VOL 13, NO I Firms with more than nine employees must present an employment reduction or restructuring plan (volet social) to workers' participation committees and labor unions for agreement, negotiation, and mediation, with recourse to arbitration. For each employee laid off, the firm pays a severance package equal to three months of salary, notifies the unemployment insurance fund, and pays the initiation fee. Initiation fee is equal to 80 percent of the gross monthly salary for each year of work experience in excess of three years, up to a maximum of 12 months of salary (equivalent to 18 years of tenure): * Initial payment equivalent to two months of salary must be paid up front. * Balance paid over one year. Benefit structure Severance pay equal to one month of salary per year of service up to three years, paid in lump sum by the employer directly to the worker. Continued eligibility for health insurance, maternity benefits, and family ben- efits. Duration period equal to two months per year of unemployment insurance contributions paid to the last employer: 12-month minimum, 36-month maximum. Duration period divided into four equal subperiods. Monthly reference wage equal to half the sum of the gross monthly wage and the minimum wage. Benefit level equal to graduated replacement rate of the monthly reference wage: * 100 percent during the first subperiod * 80 percent during the second subperiod * 60 percent during the third subperiod * 50 percent during the fourth subperiod. Benefit level minimum is three-quarters of the minimum wage; the maximum is three times the minimum wage. Benefits are subject to the employee portion of social security contributions. After exhaustion of benefits, continued eligibility for health insurance and fam- ily benefits for one year. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Ahmad, Ehtisham, and Nicholas Stern. 1987. "Alternative Sources of Government Rev- enue: Illustrations from India, 1979-80." In David Newbery and Nicholas Stern, Ruppert 183 eds., The Theory of Taxation for Developing Countries. New York: Oxford Univer- sity Press. Alderman, Harold, Sudarshan Canagarajah, and Stephen Younger. 1995. "A Compari- son of Ghanaian Civil Servants' Earnings before and after Retrenchment." Journal of African Economies 4(2):259-88. Browning, Edgar K. 1987. "On the Marginal Welfare Cost of Taxation." American Economic Review 77(1):11-23. Burda, Michael C. 1993. "Unemployment, Labor Markets, and Structural Change in Eastern Europe." Economic Policy: A European Forum 8(April):101-37. El Khawaga, Leila. 1993. "Underemployment in the Public Sector: The Case of Egypt." Proceedings of the Expert Conference on Unemployment in the ESCWA Countries [in Arabic]. Amman, Jordan. Guasch, J. Luis, Alejandra Cox Edwards, Indermit Gil, and Carola Pessino. 1997. "Es- timating the Benefits of Labor Reform in Argentina." Latin America 1 Department, World Bank, Washington, D.C. Processed. Hamermesh, Daniel S. 1993. Labor Demand. Princeton, N.J.: Princeton University Press. Hansson, Ingemar, and Charles Stuart. 1985. "Tax Revenue and the Marginal Cost of Public Funds in Sweden." Journal of Public Economics 27(August):331-53. Harris, John, and Michael Todaro. 1970. "Migration, Unemployment, and Develop- ment: A Two-Sector Analysis." American Economic Review 60(1):126-42. Lindauer, David L., and Barbara Nunberg, eds. 1994. Rebabilitating Government: Pay and Employment Reform in Africa. A World Bank Regional and Sectoral Study. Wash- ington, D.C.: World Bank. Ruppert, Elizabeth. 1996. "Unemployment Insurance in Algeria: Implications for a La- bor Market in Transition." Policy Research Working Paper 1659. Policy Research Department, World Bank, Washington, D.C. Processed. Worden, Kathleen. 1992. "For UIM: An Unemployment Insurance Trust Fund Model." Country Department 1, Europe and Central Asia Region, World Bank, Washington, D.C. Processed. World Bank. 1995. World Development Report 1995: Workers in an Integrating World. New York: Oxford University Press. . 1997. "Kingdom of Morocco: Growth and Labor Markets: An Agenda for Job Creation." Report 16598-MOR. Human Development Sector, Middle East and North Africa Regional office, World Bank, Washington, D.C. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 185-209 Labor Earnings in One-Company Towns: Theory and Evidence from Kazakhstan Martin Rama and Kinnon Scott One-company towns, characterized by the presence of a large employer in a local labor market, are a frequent legacy of state-led development strategies. How will downsizing or closing unprofitable state-owned enterprises affect these towns? This article develops a simple model combining monopsony power in the labor market with a Keynesian closure of the product market and uses it to interpret the findings of previous studies. The article evaluates the impact of the company's employment level on the town's labor earnings in Kazakhstan, where one-company towns are still preva- lent. The evaluation is based on data from the 1996 Living Standards Measurement Survey. The results show that labor earnings in the town decrease roughly 1.5 percent when the share of its population working for the company decreases 1 percent. The results are robust to changes in the definition of labor earnings and to the inclusion of a variety of other community characteristics in the analysis. These results and the theo- retical model are combined to evaluate the welfare impact of company downsizing and, consequently, to derive the optimal extent of labor retrenchment. Many developing countries, and especially transition economies, have one- company towns whose distinctive feature is the presence of a large employer in a local labor market. In spite of their name, one-company towns are not neces- sarily urban agglomerations. In some transition economies, the label applies to the rural area surrounding a large agricultural producer, such as a state-owned farm. More often, however, it applies to mining towns and to the communities developed next to large manufacturing plants, such as steel mills or armament factories. In all cases, the company accounts for a substantial share of the jobs in the town, and even those who do not work for the company depend on it to make a living. If the company were to cease its operations, the one-company town could easily become a ghost town. One-company towns also exist, or at least existed, in industrial countries. The coal mining company towns in the U.S. region of Appalachia at the turn of the century were in some respects similar to the one-company towns in tran- sition economies and developing countries nowadays. In coal mining company towns, a vertically integrated enterprise provided a variety of affiliated services Martin Rama and Kinnon Scott are with the Development Economics Research Group at the World Bank. This article is part of the research project Public Sector Retrenchment and Efficient Compensation Schemes, supported by the Research Committee of the World Bank through grant RPO 679-51. C) 1999 The International Bank for Reconstruction and Development/ THE WORLD BANK 185 186 THE WORLD BANK ECONOMIC REVIEW. VOL 13, NO I to its workers, from housing to stores to medical services. However, in some of the coal mining company towns no positive externality stemmed from the com- pany because the only jobs in town were company jobs. In this respect, the one- company towns in developing countries and transition economies may resemble more the independent towns considered by Fishback (1992), where housing and stores do not belong to the main employer. Central planning, state ownership, and directed credit have made the one- company town setting more prevalent in developing countries than it was in industrial countries by fostering the settlement of large plants in relatively iso- lated communities. Large plants were supposed to exploit economies of scale, reduce import dependence, or satisfy sentiments of national pride. Remote loca- tions were sometimes justified by the availability of a key natural resource, but they were also chosen to promote regional development, to assert the nation's sovereignty, or to hide plants from the outside world. At present, most of those plants are inefficient and overstaffed. The demise of state-led development will be incomplete, it seems, until these plants are dramatically downsized, truly priva- tized, or closed altogether. Decisions affecting state-owned enterprises in one-company towns should not be based solely on considerations of profitability. Keeping these companies in operation certainly entails a cost for the rest of society because their deficits translate into higher taxes or lower social expenditures. But shutting them down, or significantly downsizing their operations, entails a cost for the populations that surround them and depend on them to sustain their economic activity. The optimal policy decision involves a tradeoff between these two costs. Unfortu- nately, relatively little is known about the magnitude of the cost to the sur- rounding populations. As in other areas of public policy, externalities are more difficult to quantify than deficits. In the absence of information about the im- pact of the company's size on the town's earnings, decisions regarding labor retrenchment could be misguided. This article analyzes the one-company-town problem from both a theoretical and an empirical point of view. The theoretical part develops a simple model of the one-company-town setting, combining aspects of the monopsony model of the labor market and the Keynesian model of the product market. The monop- sony aspect of the model is warranted because the company may not behave as a wage taker in the town's labor market. After all, the one-company-town set- ting provides a textbook example of market power by employers. The Keynesian aspect of the model captures the fact that many of the other, smaller businesses in town cater to the company and its employees. This simple model is used as a framework to interpret the findings of previous studies, to derive the specifica- tions for the empirical analysis, to evaluate the welfare impact of downsizing, and to discuss the corresponding policy implications. The empirical part uses data from Kazakhstan, a country with several fea- tures that make it relevant for our study. First, it is to some extent representative of the transition economies in Central Asia and Eastern Europe. Its output per Rama and Scott 187 capita is close to the median for this group of countries, and other indicators of economic development occupy a roughly intermediate position as well. Second, the one-company-town setting is still prevalent in Kazakhstan, and, because of the vast surface of the country, one-company towns tend to be quite isolated from one another. Third, the data on individual earnings and community char- acteristics that are needed for the empirical analysis have been collected and processed recently. A previous use of these data in the context of a poverty assessment indicates that they are of reasonable quality (World Bank 1998). Section I discusses the one-company-town problem in an intuitive way. Sec- tion II formalizes this discussion under the form of a simple analytical model. In section III, this model serves as the background for a review of the literature. Section IV provides a brief description of Kazakhstan and its labor market. Sec- tion V addresses data issues, including the construction of the variables measur- ing labor earnings and company size. Section VI regresses labor earnings on individual characteristics and the size of the largest employer in town. Section VII shows that the results are not affected when other community characteris- tics, apart from the size of the largest employer, are taken into account. Section VIII combines the empirical results and the theoretical model to assess the wel- fare impact of company downsizing and to derive the policy implications. Sec- tion IX concludes. 1. THE PROBLEM The textbook partial equilibrium diagram of the labor market can be used to discuss the one-company-town problem in an intuitive way. This diagram repre- sents the determinants of employment and labor earnings in an integrated, rela- tively frictionless labor market. Consider the labor market of a town, under the assumption that neither local firms nor individuals can move easily in or out of it. Suppose that no local firm is able to affect labor market outcomes on its own. Suppose also that a large number of individuals can either work or perform other activities, such as studying or rearing children. In this simple setting, it is possible to analyze how the settlement of a large firm in the town affects its labor market equilibrium. The results shed light on what would happen if an existing large firm were to cease or downsize its activities. The upward-sloping line in figure 1 represents labor supply. It indicates how many individuals (L on the horizontal axis) would prefer to work rather than perform other activities for any given wage rate (W on the vertical axis). The downward-sloping line marked "labor productivity without company" repre- sents labor demand by local firms. It indicates the marginal productivity of la- bor in these firms or, conversely, the number of workers these firms would be willing to employ at any given wage rate. This line is downward-sloping under the assumption of a decreasing marginal productivity of labor; it would be flat if labor productivity were constant. In the absence of a large company, the labor market equilibrium would lie at the intersection of the supply and demand lines. 188 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I The equilibrium wage in the town would be W°, and the equilibrium employ- ment level would be LO. When a large company settles in the town, at least some workers can be em- ployed more productively, as shown by the upward shift of the marginal pro- ductivity line in figure 1 (the new line would be flat if labor productivity at the company were constant). Assume first that the company behaves in the same way as local firms. Under this assumption, the new labor market equilibrium would obtain for an employment level L1 and a wage level Wi. However, it could be in the interest of the company to pay less than W'. At a lower wage level, more individuals would prefer to perform other activities rather than work for a wage. Because some of those who work for a wage are employed by local firms, employment in the large company would fall. The company's output would fall, but its labor costs would fall, too. It is in the interest of the company to pay less than W' as long as the resulting fall in output is smaller than the corre- sponding fall in labor costs. Let W2 be the wage level that maximizes the profits of the company. This level is necessarily higher than WO; otherwise, the company would be unable to compete for workers with local firms. It follows that the equilibrium level of employment is higher than it would be in the absence of the company (L2> LO). Employment by local firms, however, may be either higher or lower than before, depending on the impact of increased employment and wages on local consump- tion. There are many goods and services that cannot be "imported" from other Figure 1 Employment and Wages in a One-Company Town Wage (W Labor Labor productivity productovlty without with company company Labor L " L) L2 Li Employment (L) Rama and Scott 189 towns, so their prices would increase with consumption, and local firms would find it profitable to hire more workers after the company settles in the town. The broken line in figure 1 represents the labor demand of local firms in the new equilibrium (as the model in the next section shows, this line is not necessarily downward-sloping). For a wage level W2 employment by local firms would be L2T. This new employment level is larger than LI when local goods and services account for a large share of the town's consumption. It is also larger than LO when the productivity of labor is constant. Whether the company behaves as a wage taker or as a wage maker depends, among other things, on its objectives. In state-owned enterprises, those objec- tives are not always clearly specified. And even when they are, managers do not necessarily face the appropriate incentives to fulfill them. However, figure 1 shows that the presence of a large company raises labor earnings in town as long as labor market equilibrium lies on the town's labor supply line. This result holds regardless of whether the company exploits its ability to push wages down. II. A SIMPLE MODEL To interpret the findings of previous studies as well as to implement the em- pirical study of the case of Kazakhstan, we organize the preceding discussion under the form of an analytical model. This model does not thoroughly describe the one-company-town setting. Instead, it highlights how the size of the com- pany can affect employment and earnings in the town. With this objective in mind, we keep our hypotheses regarding technology and preferences simple. The introduction of a Keynesian closure in an otherwise standard monopsony model of the labor market distinguishes this model from the previous literature (see Boal and Ransom 1997). Assume that local firms use labor only and that their productivity is constant regardless of their volume of output. An example of a local firm with these characteristics would be a barbershop. Haircuts involve little more than the barber's time, and the average time per haircut does not vary much with the size of the shop. If units are chosen appropriately, each employee produces one unit of output over the relevant time period. Assume that labor productivity is con- stant in the company as well, at level K. Parameter K can be seen as a measure of the capital stock or the size of the company. These assumptions on technology can be written as follows: (1) YT = LT (2) Yc = KLc where Y is output, T refers to local firms, and C refers to the company. Under these assumptions, the demand lines in figure 1 would be flat. Total employment in the town is the sum of employment in local firms and employment in the company. But for the town's labor market to be in equillb- 190 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 rium, total employment must equal the number of individuals willing to work at the prevailing wage rate. Those individuals who have a better alternative prefer not to work, unless the wage rate increases enough to match the value of that alternative. This value represents their reservation wage, R. As the wage rate goes up, individuals with increasingly higher reservation wages are drawn into the labor force, and total employment expands: (3) L=LT+Lc (4) W=R (5) R = R(L) with dLRR £ Parameter £ can be interpreted as a measure of the elasticity of the town's labor supply. The larger is £, the more responsive is labor supply to changes in the wage level. Increased labor mobility between towns should therefore raise the value of £. Finally, it is assumed that the output of local firms is consumed in town, whereas the output of the company is sold out of town. The unit price of the company's output is supposed to be given. The unit price of local output, in turn, is determined by supply and demand. On the supply side, competition between local firms implies that this price has to be equal to the wage level W (in the example above, the price of the haircut cannot be distinguished from the earnings of the barber). On the demand side, it is assumed that a fraction c( of total income in the town is spent on locally produced goods. (This assumption corresponds to the case where the households' utility function is of the Cobb- Douglas type, exhibits constant returns to scale, and includes the consumption of local goods among its arguments.) Total income is the sum of the value added by local firms and the wage bill of the company. Equilibrium in the market for local goods is attained when the supply of local goods measured in money terms WYT equals demand DT: (6) WYT = DT (7) DT = (x(WYT+ WLC). Parameter cx can be interpreted, in Keynesian terms, as a marginal propensity to consume. Employment by local firms depends on the employment and pay decisions of the company because those decisions affect income and, therefore, the demand for local output.' It follows from equations 1, 3, 6, and 7 that 1. The simple specification chosen for DT implies that there is no equlibrium in the absence of a company, except as a limit case when K -X 0. A full analysis of the case where no large company operates in town would require the introduction of a second group of local firms, producing goods that can be sold out of town. This would complicate matters without shedding additional light on the consequences of downsizing. Rama and Scott 191 (8) L= Lc. 1-cL This relationship between L and Lc is reminiscent of the Keynesian multiplier. For every job created by the company, more than one job is created overall. The size of this multiplier depends on the marginal propensity to consume local goods a. The number of jobs created by the company, in turn, depends on whether it behaves as a wage taker or as a wage maker. The company's profits X are a function of its labor productivity, its labor costs, and its employment level: (9) Tr = (K - W)Lc. If the company treats the wage level as given, it expands its employment level as long as K is higher than W. But for more individuals to be drawn into the labor force, the wage level W has to increase. Eventually, the marginal profit from hiring an extra worker falls to zero, and the labor market reaches a new equilib- rium. In this equilibrium the wage level W' verifies (10) W' = K. The company could take advantage of its size to influence labor costs, how- ever. This is because the number of workers Lc it attracts varies with the wage level W it pays. The relationship between Lc and W depends on the elasticity of labor supply by individuals, as well as on the endogenous response of employ- ment by local firms to changes in employment and pay in the company. In ana- lytical terms, equations 4, 5, and 8 imply (1) W=R 'Lc- Equation 11 can be interpreted as the labor supply curve faced by the company. The company's labor supply curve is flatter than the town's labor supply curve because a higher wage level leads to a higher labor demand by local firms, so that Lc increases by less than L. However, under the chosen assumptions, the elasticity of the company's labor supply curve is the same as that of the town's labor supply curve, namely £. The optimal wage level from the point of view of the company, W2, can be obtained by replacing the labor supply curve faced by the company in the profit function represented by equation 9. The first-order condition for this problem is (10') W2 K 1 + (VEF) 192 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I As long as the local labor supply curve is not infinitely elastic, this wage level is lower than W1. Because fewer individuals are drawn into the labor force, total employment is lower when the company behaves as a wage maker. III. PREVIOUS STUDIES Previous studies have evaluated the empirical relevance of some of the key assumptions in the model. For instance, in the model the town's labor supply is not infinitely elastic, implying that workers do not move in and out of town in large numbers in response to regional differences in labor earnings. The litera- ture on local labor markets has implicitly dealt with this assumption, by assess- ing whether the labor earnings of otherwise similar workers vary across regions in a country. The model also assumes that the company is large enough to enjoy a significant monopsony power and possibly take advantage of it. Several stud- les have focused on labor markets with only a few employers and tried to deter- mine whether wages in those markets are below their competitive levels. An- other key assumption is that employment and pay in the company have an impact on labor earnings in the town. Some research has implicitly dealt with this as- sumption by documenting everyday life in one-company-town settings.2 Even in a well-integrated economy, with good infrastructure and transporta- tion, efficient housing markets, and a highly mobile labor force, regional dis- parities in labor earnings can be significant. In the United States, wage inequal- ity has increased considerably over the past quarter of a century, but it has done so at markedly different paces depending on the region. Using data from popu- lation surveys covering almost two decades, Topel (1994) estimates the determi- nants of the relative wages of low-skilled and high-skilled males at the regional level. His results suggest that distinctly local factors affect relative wages. The extent of labor markets thus appears to be limited by geography. In a previous study in the same spirit, Topel (1986) also shows that wages are responsive to employment shocks at the state level. The specification he uses can be inter- preted as a version of the labor supply function in equations 4 and 5, with con- trols added for education, experience, marital status, and other characteristics of workers, all measured at the state level. Topel finds that wages are indeed responsive to local employment shocks, particularly in the case of older and less-educated workers. This finding suggests that local labor supply is upward- sloping, as is assumed in the theoretical model. A related issue is whether large employers enjoy some monopsony power in local labor markets. This issue has been addressed for workers with specific labor skills in well-delimited geographic areas. The market for hospital nurses is 2. There is also a vast literature on the consequences of company downsizing on the earnings of the workers dismissed by the company Whereas these consequences tend to be sizable, workers in other firms are not necessarily affected by the downsizing process. Actually, most analyses in this literature are carried out under the assumption that downsizing entails no externalities. As a result, this literature does not shed much light on the one-company-town problem Rama and Scott 193 a case in point. Because there are only a few hospitals in each city, monopsony power by hospitals could be considerable. Frequent complaints about the "chronic shortage of nurses" indicate that hospitals would be willing to hire more nurses at the prevailing wage rate, but not to raise their wage rate to attract them. Equilibrium vacancies of this sort can be expected when firms pay wages below the relevant marginal product. Sullivan (1989) assesses the monopsony power hypothesis by estimating the elasticity of the supply curve of nurses faced by individual hospitals. The speci- fication used by Sullivan is similar to that of the labor supply function faced by the company, as described by equation 11, with wages and employment mea- sured at the hospital level. Sullivan's model includes controls for employment, wages, number of patient-days, and average length of stay in other hospitals. The regression analysis yields values of the elasticity £ ranging from 1.3 in the same year to 3.8 over a three-year period. From these results, Sullivan concludes that hospitals enjoy a significant monopsony power in the market for nurses. Surprisingly, however, the estimated values of E are similar for metropolitan and nonmetropolitan areas. Because there are many more hospitals in metro- politan areas, the labor supply curve they face should be much flatter (that is, £ should be higher). Therefore, Sullivan's results may capture something different from monopsony power. They would be consistent, for instance, with efficiency wages. If the probability of being caught shirking fell with the size of the hospi- tal, large hospitals would have to pay higher wages than small hospitals in order to elicit the same level of effort from their nurses. But in that case, the equation estimated by Sullivan could not be interpreted as a quasi labor supply curve. Monopsony power by employers was also likely in the coal mining company towns of Appalachia at the turn of the century. Miners in some of these towns lived in company housing and were paid in scrip that could be used only in the company store. Some of the companies controlled even the police. If these com- panies could not set wages below the corresponding marginal product of labor, then it is difficult to claim that companies elsewhere can do it. Boal (1995) con- tends that they could not. His regression analysis, based on data from 30-odd coal mining counties in West Virginia over the period 1897-1932, explains county-level wages as a function of county-level employment in the same year and in the previous year. His specification is similar to that of the local labor supply curve in equations 4 and 5. The population of the county and county dummies are included among the independent variables in the regression. Con- trols for average wages and employment in the other mining counties are also added, depending on the monopsony hypothesis used. Boal finds a large coeffi- cient for employment in the same year, but the coefficient is almost as large, and has the opposite sign, for employment in the previous year. He concludes that employer monopsony power was limited in turn-of-the-century coal mining, except in the short run. These findings have led to a reinterpretation of the Appalachian coal mining company town. Fishback (1997) argues that company housing and company 194 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 stores were mutually advantageous arrangements for workers and employers, rather than mechanisms set up by employers to exploit their workers. But it is worth mentioning that Appalachian company towns were not far apart and were linked by rail lines built to haul the coal. Findings concerning Appalachian com- pany towns may therefore be of limited relevance for evaluating the impact of downsizing on more isolated one-company towns. Detailed case studies are another source of information about one-company towns. The history of the town of Pullman (currently a suburb of Chicago) by Buder (1967) confirms how difficult it is to disentangle paternalism from exploi- tation, or mutually advantageous arrangements from monopsony power. George Pullman, who revolutionized railroad travel with his development of the sleep- ing car, also had a serious interest in the labor problems of the time. In 1880, he launched the construction of a model town to house his workers next to his new factory. His intention was to apply principles of business efficiency to meet the needs of his workers and establish a more peaceful system of labor relations. The bitter strike of 1894 suggests that workers thought otherwise. Another case study implicitly dealing with monopsony power concerns gold ming in southern Africa. The account of this history by James (1992) makes it clear that the Southern African Chamber of Mines was not as enlightened as George Pullman. This chamber was formed in 1889 to organize labor supply and mitigate competition for workers between mines. Shortly after it was formed, it established centralized labor-recruiting agencies that fixed the African work- ers' food supply, length of shift, working day, and wage rate. Individual mines could set their own wage scales, but their average wages could not be higher than those of other mines. This system ensured that wages for miners remained constant, in real terms, between 1911 and 1969. Although James's account does not prove that gold mines paid wages below the corresponding marginal pro- ductivity of labor, it suggests that they did. IV. THE CASE OF KAZAKHSTAN From an empirical point of view, it would be ideal to have detailed data not just on one but on several one-company towns. Variation in the size of the com- pany relative to the size of the town could then be used as a natural experiment. Conceptually, the experiment would involve a comparison of the labor earnings of workers who are alike in all respects, except that some live in towns where large companies operate whereas others live in towns with smaller companies. If the labor earnings of these workers differed in a systematic way across towns, the difference could be attributed to the externality created by the company. The regression analysis in section VI emulates this natural experiment. Kazakhstan has dozens of one-company towns. Labor mobility in and out of them is quite low, both for geographic and for institutional reasons. Kazakhstan is the ninth largest country on earth, with a population of less than 17 million people scattered across 2.7 million square kilometers. Labor mobility is con- Rama and Scott 195 strained, among other reasons, by the system of unemployment benefits. Unem- ployed workers need to hold a propiska (a local passport) in order to draw these benefits, which may dissuade them from searching for jobs outside their region of residence. Mobltity is also constrained by a poorly developed housing market outside Almaty. The provision of social services by firms may further discourage mobility, although substantial progress has been made in transferring these ser- vices to local authorities. Kazakhstan has some features that make it rather atypical among transition economies. For example, the sharp ethnic divide between the Russian north and the Kazakh south has not so far led to conflicts such as those experienced by some of the other former Soviet republics. We take into account this divide when we assess the robustness of the empirical results. Another atypical feature is that during the Soviet period Kazakhstan received the highest amount of sub- sidies under the old system. This massive transfer of resources is probably one of the reasons underlying the large number of one-company towns. In other respects Kazakhstan is not an exceptional country. Its level of devel- opment is close to the median for transition economies. With an output per capita of roughly $3,000 measured at purchasing power parity, it ranks number 14 among 23 former socialist countries in Eastern Europe and Central Asia. The private sector's share in the economy is not particularly low. A gradual process of privatization and enterprise restructuring has occurred since 1991, and the assets of many nonagricultural firms were transferred to managers' and work- ers' collectives in 1993. Since then, privatization has been extended to more than 90 percent of farms and 80 percent of farmland. Finally, few institutional constraints exist on the employment and pay deci- sions of firms in Kazakhstan (Klugman and Scott 1997). Unionization is high, but it is widely believed that trade unions are not a significant source of labor market rigidity. The minimum wage has been drastically eroded by price liberal- ization and high inflation, to the point where it lacks any significance, except as a scalar for wages in the public sphere (but even the lowest-paid public sector worker earns three times the minimum wage). Also, employers obtained the right to set wages in 1987 and to lay off workers in 1990. Workers are entitled to "adequate" notice before dismissal and to severance pay amounting to three months of wages. However, newly privatized firms are under no direct or indi- rect obligation to maintain employment levels. In light of these weak institu- tional constraints, the labor market equilibrium concepts used in the theoretical part of the article should be appropriate for Kazakhstan. V. THE DATA Another factor that makes Kazakhstan an interesting case is the availability of good data. An integrated multipurpose household survey was conducted in 1996. This survey had the same features as the Living Standard Measurement Surveys (LSMS) implemented with the assistance of the World Bank in many de- 196 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO 1 veloping countries and transition economies since 1985 (see Grosh and Glewwe 1995 for a description of these surveys). The LSMS for Kazakhstan covered a nationally representative sample of more than 7,200 individuals in about 2,000 households. A probability sample was used, instead of the old quota sample of the existing Family Budget Survey, which was typical of the former Soviet Union. Because the data were collected during a single month, the survey can be seen as a pure cross-section of households, with no time dimension involved. The sur- vey gathered information on individual characteristics and earnings, as well as on household consumption. In addition, its community questionnaire reported the distance between each community and the nearest large farm or industrial enterprise, as well as the number of individuals from the community who worked for that farm or enterprise. Measuring labor earnings in a transition economy like Kazakhstan is not an easy task. Official wages do not provide a complete picture of official labor compensation. Official wages may underestimate total earnings because they do not include nonmonetary benefits. During the Soviet period, these benefits tended to be an important component of total labor earnings because they served to attract and retain workers when the firms' autonomy to set wages was limited. Typically, nonmonetary benefits took the form of social services, such as health and child care, or access to sanatoriums and recreational places (Klugman and Scott 1997). A portion of these benefits has been retained. Alternatively, official wages may overestimate total earnings, due to substantial arrears in their pay- ment. Many firms facing a decline in demand for their output have refrained from adjusting employment downward, relying instead on reduced hours and "administrative" leave. To compensate for the shortfall in earnings, workers increasingly combine attachment to their formal employers with growing in- volvement in informal activities (Murthi 1998). In light of these measurement problems, in the empirical analysis we use three different measures of labor earnings: wages officially earned, wages actually re- ceived, and per capita consumption. The first two measures are defined at the individual level, whereas the third one is calculated at the household level. Because of nonmonetary benefits, wage arrears, and informal activities, consumption per capita could in principle provide a better picture of actual earnings from labor than the other two measures. However, consumption per capita is also affected by the size, composition, and asset ownership of households, so that it is not strictly comparable across individuals. These shortcomings imply that none of the three earnings measures is entirely reliable on its own. But econometric results that are consistent across the three measures should carry some credibility. Table 1 reports the means and standard deviations of the three earnings mea- sures and the individual characteristics that are usually considered among the determinants of labor earnings. Figures for consumption per capita are con- structed using the same methodology as in the poverty assessment of Kazakhstan in World Bank (1998). Table 1 shows that the average per capita consumption is similar to the average wage, although some household members do not earn Rama and Scott 197 Table 1. Sample Statistics from the Kazakhstan Living Standards Survey, 1996 Standard Number of Variable Mean deviation observations Age (years) 37.162 9.707 2,278 Gender (male = 1) 0.522 0.497 2,278 Schooling (years) 12.122 2.332 2,278 Wages officially earned 5,077.3 4,593.8 2,278 Wages actually received 5,791.0 6,492.4 1,311 Per capita consumption 5,063.9 3,595.5 2,354 Company's population share 0.055 0.096 98 Note: Monetary variables are measured in tenges per month, at 1996 prices. Source' Authors' calculations based on survey data. any income. This similarity confirms that households have other earnings that are not necessarily legal or reported in addition to their wage earnings. How- ever, the importance of these other earnings should not be overstated because labor force participation rates are high in Kazakhstan. Moreover, ethnic Rus- sian families, who account for roughly 40 percent of the population, typically have one child. As a result, the number of household members not earning any income is lower than elsewhere. Other measurement issues concern the size of the company. Unfortunately, the number of individuals who work for the nearest large employer is not avail- able for 34 of the 132 communities. Table 2 displays the medians and standard deviations of the available data, expressed as percentages of the community's population. When there is no large employer in or close to the community, a zero is reported. Unfortunately, the questionnaire did not foresee the possibility Table 2. The Company's Population Share, Kazakhstan, 1996 (distance at which the nearest large employer is located) Nature of In the Within a 10- Between 10 and More than 50 the community community kilometer range 50 kilometers kilometers Town Median 0.013 0.108 0.009 0.015 Standard deviation 0.037 0.147 0.016 0.008 Number of observations 25 11 4 2 Other urban Median 0.035 0.043 0.013 0.000 Standard deviation 0.093 0.042 0.018 0.000 Number of observations 8 4 2 1 Rural Median 0.178 0.005 0.000 0.000 Standard deviation 0.145 0.029 0.031 0.048 Number of observations 12 7 10 12 Note: The company's share of population is defined as the fraction of total population in the enumerating district working for the largest local employer Source: Authors' calculations based on survey data. 198 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I of more than one large employer per community. The data are disaggregated according to the three types of community considered by the LSMS and the dis- tance from the community to the company. We use this disaggregation to im- pute the company's share of population for the communities with missing data. Consider, for instance, the case of a town where the nearest large company is within a 10-kilometer range but where data on its employment level are missing. Our imputation procedure assumes that the share of the town's population work- ing for the company is equal to 0.108. Econometric analyses using the imputed data are likely to yield more significant estimates than those obtained using the reported data only. If the results turn out to be consistent regardless of whether imputed data are used, these estimates should carry some credibility. VI. COMPANY SIZE AND EARNINGS The impact of the size of the company on the town's labor earnings is as- sessed by estimating a quadratic version of the quasi labor supply curve in equa- tion 11, as follows: (11') log W = P30 + PJX + + DnX" + P3 +1Lc + P3÷2L 2. The X variables capture earnings determinants such as education and experi- ence. These variables and the earnings variable Ware measured at the individual level. The share Lc of the population employed by the company, however, is measured at the community level. Consequently, equation 11' can be interpreted as an earnings function augmented so as to include one community characteris- tic among its explanatory variables. The idea of introducing community characteristics among the determinants of individual earnings or individual consumption is not new. For instance, Borjas (1995) shows that individual wages in the United States are affected not only by individual educational attainment but also by the average educational attain- ment of the neighborhoods in which the individuals grew up. Community char- acteristics are also included among the determinants of consumption per capita in developing countries. Thus, Narayan and Pritchett (1997) show that village social capital (measured as the membership of a variety of groups, such as burial societies) has an impact on individual consumption in Tanzania. Similarly, Ravallion and Wodon (1997) find that location dummy variables are important in explaining per capita consumption in Bangladesh. To our knowledge, the size of the largest employer in town has not been considered among the relevant community characteristics. The quadratic specification chosen for equation 11' allows the elasticity of the labor supply curve faced by the company to vary with its size. More specifically, (12) dlogLW dLo 1 1 dlogW- dlogW Lc - ~+1Lc +2l3.+2L Rama and Scott 199 It follows that the company has no monopsony power (that is, the labor supply curve it faces is infinitely elastic) when it employs a negligible share of the community's population. How this elasticity varies as Lc increases depends on the estimated values of coefficients Pin+i and i3n+2- If both coefficients are equal to zero, the labor supply curve faced by the company is infinitely elastic regardless of the company's size. Table 3 presents the results obtained when estimating equation 11' using each of the three measures of labor earnings, both with and without imputed data on the size of the company. The regressions using consumption per capita as a measure of labor earnings also include several household characteristics among their explanatory variables. These characteristics are the household's size, its dependency ratio, and the number of rooms it occupies in its house or apartment (shared housing is still common in Kazakhstan). The number of rooms occupied can be seen as an indication of household wealth in a country where capital and housing markets are not well developed yet. The coefficients on individual characteristics should be smaller in the regressions where con- sumption per capita is the explained variable because earnings are pooled in a household. The absolute size of the estimated coefficients J3n+1 and N+2 iS larger when earnings are measured through wages, either officially earned or actually re- ceived, than when they are measured through per capita consumption. The sta- tistical significance of coefficients 13n+l and n+2 is higher as well. The last row in table 3 reports the implicit effect of a change in the company's size on the town's labor earnings, evaluated at the sample mean (that is, for Lc = 0.055). The aver- age of this effect across all six specifications is roughly 1.5, implying that labor earnings in town would decrease around 1.5 percent if the share of the town's population employed by the company were to decline 1 percent. The impact appears to be higher (around 2.2) when labor earnings are measured through wages actually received and to be lower (around 0.6) when they are measured through per capita consumption. These results suggest that large employers enjoy a significant monopsony power in the one-company towns of Kazakhstan. The elasticity of the labor supply curve faced by large employers can be assessed using equation 12. At the sample mean, this elasticity ranges from 6.7 to 15.5 in the specifications where labor earnings are measured through wages (either officially earned or actually paid). If the estimates of Sullivan (1989) and Boal (1995) are to be taken literally, the labor supply curve faced by large employers in the one-company towns of Kazakhstan would be more elastic than the supply of nurses faced by U.S. hospi- tals but less elastic than the labor supply curve faced by Appalachian coal min- ing companies. Only the results from the specifications where labor earnings are measured through per capita consumption are close to those obtained by Boal for Appalachian coal mining towns. Figure 2 shows the levels of labor earnings W and of labor productivity K associated with different sizes of the company, under the assumption that large Table 3. Determinants of Individual Earnings in Kazakhstan, 1996 Earnings measure Wages officially earned Wages actually received Per capita consumption Variable a b a b a b Company's population share (Lv) 1.6313 2.7475- 2.5764 " 3.9988' 0.8991 0.5978' (3.520) (6.788) (3.424) (6.202) (2.189) (1.666) Company's population share -4.1558 -6.5079 -8.2478' "" -11.7943 ' < -1.6377 -1.0214 squared (L ()2 (-3.247) (-5.599) (-3.487) (-5.464) (-1.444) (-0.993) Years of schooling 0 0746'-" 0.0837 0.0922" 0.0812'-** 0.0172"' 0.0234 (9.383) (13.134) (7.546) (8.228) (2.399) (4.117) Years of cxperience 0 0302 0.0251 * 0 0229 0.0275 -0.0040 -0.0006 (5.141) (5 229) (2.338) (3.458) (-0.724) (-0.140) Ycars of experienice squared -0.0006 -0.0004' -0.0005 -0.0007' 0.0001 -0.00(( (-4.167) (-3.852) (-2.20(5) (-3.429) (0.412) (-0.035) t- Gender (male = 1) 0.4447'' 0.4304< 0.3682" 0 3664 -'-' 0.0409 0.0389 (12.706) (15.144) (6 836) (8.368) (1.311) (1.542) Independent term 6.8119 6.6700': 6.7247- 6.7875-<' 11.0838 11.0189 (58 302) (71.694) (36.335) (45.506) (95.753) (116.632) Household characteristics No No No No Yes Yes Adjusted R2 0.146 0.166 0.112 0.123 0.185 0.201 F test 44.51 75.09 19.51 31.13 39.64 63.67 Number of observatioiis 1,534 2,239 880 1,294 1,533 2,238 Downsizing impact (dlog W/dL,) at the mean 1.17 2.03 1.67 2.70 0.72 0.48 Significanit at 10 percent. Significant at 5 percent. Significant at 1 percent. Note: The dependent variable is the log of earnings, log W For each earnings variable, the regressions in column a arc based on the conipany's actual share of popularion, whereas the regressions in column b also include imputed shares of population. All regressions were estimated by ordinary least squares t-values are reported in parentheses. Source: Authors' calculations. Rama and Scott 201 employers behave as wage makers. This figure was constructed based on equa- tions 11', 12, and 10', using the values of the coefficients 0,+i and I3n+2 reported in the third column in table 3. The figure illustrates the tradeoff between the Keynesian and the monopsony effects of an increase in the size of the company. As the company expands its employment, it gains power over the local labor market, as shown by the widening gap between the two curves in figure 2. There- fore, the monopsony effect becomes more important in relative terms. However, the Keynesian effect dominates over the relevant range of company sizes, as shown by the upward slope of the lower curve. Based on the coefficients in the third column in table 3, wages in town increase with Lc as long as the company employs less than 15.6 percent of the town's population. Because the largest employer employs less than this in more than 90 percent of the districts in our Figure 2. Wages in the Town and the Size of the Company Deviation from WI (log) 0 30 IMPlicit prdciiy K 0 25 if the company i s a wage maker 0 20 w 0 15 / / ~~~~~Pre icted wage inm the town .... ...... ......... ... /.,........7 0 10 / 0 05 - Sample mean -0 05 - I 000 002 004 006 008 010 012 014 Source Authors' calculations Company size (L) 202 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I sample, company downsizing can be expected to reduce the town's labor earn- ings. This conclusion remains valid when estimates from other columns in table 3 are used instead. We also tried several modifications to the specification in equation 11' to check the robustness of our results. One of the modifications was to replace years of education by a series of dummies for education degrees (primary, sec- ondary, vocational, technical college). The results did not change much. An- other modification was to include dummy variables for the distance between the main employer and the town, as defined in table 2. These dummies, either alone or interacted with the company's size, reduced the overall fit of the regressions without yielding any consistently different pattern. Finally, we tried to classify individuals depending on whether they worked for the company. Because this information was not provided by the LSMS survey, this attempt was based on the sector of activity of the main employer in a district and that of the individuals surveyed in that district. Unfortunately, too many individuals were difficult to classify, so we did not pursue this line of analysis. VII. TOWN CHARACTERISTICS AND EARNINGS The results in section VI could be objectionable on the grounds that associa- tion is not causation. Labor earnings in communities where large companies operate could be higher not because of a positive externality from employment in the company but rather because of other differences between these and other communities. For instance, the average educational attainment in communities with large companies could be higher, or the social capital could be more devel- oped, than in communities with smaller companies. To the extent that downsizing in the company does not affect these other community characteristics, it would be inappropriate to predict the impact of downsizing on labor earnings based on the estimated values of coefficients ,,+1 and ,+2. The results in section VI could therefore be biased due to the omission of relevant community characteristics. The one-company towns of Kazakhstan are admittedly different from other communities. Table 4 reports the correlation coefficients between the share of the town's population working for the largest employer and a variety of other community characteristics. These coefficients are calculated with and without using imputed data on the company's size. They suggest that one-company towns tend to have a larger share of their employment in manufacturing, to be more Russian (as opposed to Kazakh), and to be more Orthodox (as opposed to Mus- lim). Table 4 also reports the correlation coefficients between Lc and the X variables in equation 11'. These correlations are much weaker than the correla- tions with community characteristics. The gender composition and average edu- cational level of one-company towns are roughly the same as those of other communities. If anything, one-company towns could have a slightly younger population. Rama and Scott 2 03 Table 4. Companies, Towns, and Individuals in Kazakhstan, 1996 (correlation coefficients with the company's share of population) Variable Actual share Imputed share Community characteristics Population in the community -0.121 -0.079 Kazakh share of population -0.117 -0.152' Russian share of population 0 125 0.145 ' Muslim share of population -0.136 -0.157- Orthodox share of population 0.166 0.185-- Manufacturing share of working-age population 0.204'" 0.205- Agriculture share of working-age population 0.092 0.008 Commerce share of working-age population -0.131 -0.075 Unemployment share of working-age population -0.058 -0.051 Illiteracy share of adult population -0 027 0.034 Number of observations 98 132 Individual characteristics Age (years) -0.055 -0.038 Gender (percentage male) 0.005 -0.005 Schooling (years) -0.038 -0.015 Number of observations 1,534 2,239 Significant at 10 percent. - Significant at 5 percent. Note: Values are correlation coefficients between the share of the town's population that is working for the largest employer and a variety of other community and individual characteristics. Source: Authors' calculations. The main results remain roughly unchanged when these other community characteristics are included among the right-hand-side variables in equation 11', as shown in table 5. (Strictly speaking, not all the community characteristics listed in table 4 are included in the regressions, due to multicollinearity prob- lems. For instance, the correlation coefficient between the Russian and Kazakh shares of population is close to -1.) Only when labor earnings are measured through officially earned wages do the coefficients P,n+l and tN+2 become statisti- cally insignificant. But the coefficients obtained when measuring labor earnings through per capita consumption, which are statistically insignificant in table 3, are now significant. The average effect of a change in the company's size on the town's labor earnings, evaluated at the sample mean, is 1.25. This is not very different from the average effect in table 3. Of course, there could be other com- munity characteristics, not captured by the 1996 LSMS, that underlie the ob- served association between company size and labor earnings. But the commu- nity characteristics considered in table 5 cover a wide variety of aspects, from ethnicity and religion to employment structure and literacy. It is thus likely that coefficients ,+ and 8n+2 appropriately capture the kind of externality consid- ered in the theoretical model. Table 5. Individual Earnings and Town Characteristics in Kazakhstan, 1996 Wages officially earned Wages actually received Per capita consumption Variable a b a b a b Company's population share (Lv) 0.6977 0.6728 1.6343** 2.0310*** 2.6327* 2.3566*** (1.454) (1.531) (1.966) (2.631) (5.958) (5.605) Company's population share -0.0198 -0.1238 -4.1381* -5.0016'* -7.1830"** -6.6121** squared (LC)2 (-0.015) (-0.103) (-1.654) (-2.095) (-6.059) (-5.749) Russian share of population 0.3196*** 0.4035***-< 0.4017'* 0.3751*' 0.8762*` 0.7619"^'* (4.290) (6.481) (3.162) (3.377) (11.995) (11.997) Manufacturing share of working-age 0.0331 0.1645* -0.1862 -0.3500" -0.4177"'" -0.3422"** population (0.284) (1.691) (-1.023) (-2.256) (-3.883) (-3.676) Agriculture share of working-age -0.7753"*4 -0.6425 * * -0.8408" -0.8349*'* 0.0325 0.0408 population (-11.419) (-11.502) (-7.252) (-8.455) (0.514) (0.754) Commerce share of working-age -1.2052**" -0.8435 *' -1.2281"" -0.7633" -1.2283 -1.5709-* population (-5.030) (-3.818) (-2.493) (-1.649) (-5.510) (-7.381) Unemployment share of working-age 0.2276 -0.0665 0.6663'" 0.2257 -0.0797 -0.3550** population (1.035) (-0.371) (1.831) (0.735) (-0.393) (-2.068) Illiteracy share of adult population 0.1236 0.0495 1.2915 1.1790 -1 .5198** -1.5981 * * (0.179) (0.076) (0.875) (0.830) (-2.379) (-2 576) Individual characteristics Yes Yes Yes Yes Yes Yes Household characteristics No No No No Yes Yes Adjusted R2 0.303 0.324 0.207 0.209 0.290 0.292 F test 54.24 80.16 19.31 25.22 40.91 55.30 Number of observations 1,468 1,981 843 1,100 1,467 1,980 Downsizing impact (dlog WIdL,) at the mean 0.69 0.66 1.18 1.48 1.84 1.63 Significant at 10 percent. t' Significant at 5 percent. . Significant at 1 percent. Note: The dependent variable is the log of earnings, log W. For each earnings variable, the regressions in column a are based on the actual population share of the company, whereas the regressions in column b also include imputed population shares. All regressions were estimated by ordinary least squares. t-values are reported in parentheses. Source: Authors' calculations. Rama and Scott 205 VIII. POLICY IMPLICATIONS The estimated externality from employment in the company to labor earnings in the town is an important input with which to assess the socially optimal ex- tent of downsizing, but this assessment has to be based on welfare comparisons rather than on earnings comparisons. The fact that the estimated externality is positive and statistically significant suggests that the theoretical model discussed in section II is empirically relevant. This model can therefore be used to calculate the welfare effect of downsizing in the company and to derive the appropriate policy implications. We illustrate this approach by discussing the optimal sub- sidy to an inefficient plant operating in a one-company town. A more general analysis should also consider other policy instruments, such as subsidies to local firms and income transfers to individuals who do not have jobs. Therefore, the rule we propose to determine the subsidy rate will be optimal only to the extent that no other policy instruments are available. We leave a more general analysis of the problem to the interested reader. A benevolent government should care not only about the well-being V of persons whose lives are directly affected by economic activity in the one- company town but also about the well-being F of those who pay taxes to sup- port that activity. The first group includes those who work for the company, those who work for local firms, those who live in the town but do not perform economic activities, and the owners of the company. In analytical terms, (13) V = WL + fR(x)dx + (K - W)Lc. L The first term on the right-hand side of equation 13 represents total labor earn- ings in the town. If welfare is proportional to income (that is, if there is no risk aversion), this term measures the well-being of the town's workers. The second term measures the well-being of those who live in the town but perform noneco- nomic activities. In this term, x is an integration variable. The third term mea- sures the profits made by the owners of the company, thus reflecting their own well-being. This third term is included under the assumption that the company behaves as a wage maker. Supporting the company entails a double burden for the rest of the society. A first, direct cost is represented by the amount of resources that are transferred to the company and thus cannot be used for consumption or investment purposes. A second, more subtle burden results from the use of distortionary taxes to raise those resources. Because these taxes may affect private sector decisions regard- ing labor force participation, on-the-lob effort, and savings, among others, trans- ferring one tenge to the company costs more than one tenge to the rest of the society. In analytical terms, the function F can be written as (14) F = (1 + k)SLc 206 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I where S is the subsidy per worker and parameter X captures the marginal tax burden. Parameter X is larger the more inefficient the tax system. It could be quite large in transition economies and developing countries. The problem faced by a benevolent government is to choose the level of the subsidy S that maximizes total well-being, V - F. Let K* be the "true" labor productivity of the company in the absence of transfers from the government. The subsidy S is thus equivalent to (15) S=K-K*=sK where s represents the subsidy rate. Equation 15 and the model described in section II are the constraints of the government's maximization problem. Its solution is easier to interpret if we solve separately for the effects of S on V and on F. The impact of the subsidy on the well-being of those who are directly affected by economic activity in the town can be evaluated by totally differentiating equa- tion 13, using Leibniz's rule, and replacing equations 8, 10', 11, and 15. This yields (16) dS 1-Ic 1 +(te Lc The impact of the subsidy is bigger the larger the marginal propensity ou to con- sume local goods. This is because those who work for the company spend on local goods, so that the subsidy creates jobs both in and out of the company. But the allocation of the newly created jobs between the company and local firms is inefficient because of the wedge between wages and labor productivity in the company. The impact of the subsidy is therefore higher the higher the elasticity £ of the town's labor supply. The impact of the subsidy on the well-being of those who are asked to sup- port the company, in turn, can be assessed by differentiating equation 14 and, again, replacing equations 8, 10', 11, and 15. The following result obtains: (17) -d= (1 + ±) (1 + se)Lc. dS The cost of the subsidy to the rest of society is higher the larger the elasticity e of labor supply. This is because more individuals are drawn into the labor force as the wage paid by the company increases. Employment in the company thus ex- pands more substantially, and the total bill for the taxpayer ends up being higher. Not surprisingly, the cost to the rest of society is also higher the larger is the inefficiency X of the tax system. The optimal subsidy rate can be obtained by equating dVIdS with dFldS and solving for s. That rate verifies Rama and Scott 207 (18) 1-o 1+(1)j) Equation 18 makes it clear that in our simple model of the one-company town, the rationale for subsidizing the company comes from the Keynesian closure of the product market. If all goods and services could be imported from other towns, a would be equal to zero and s would become negative. The larger is the Keynesian effect, the higher is the optimal subsidy rate. It is also clear that the optimal subsidy rate decreases as the inefficiency X of the tax system increases. Finally, the elasticity E of the town's labor supply has opposite effects on the optimal subsidy rate. On the one hand, a high value of this elasticity makes the subsidy very expensive because employment in the company increases substantially. But on the other hand, a higher £ reduces the monopsony power of the company, thus making the labor market of the town more efficient. The net effect of a higher elasticity of labor supply is therefore ambiguous. IX. CONCLUSION The results in this article suggest that a town's labor earnings increase when a large company settles in, and they can be expected to decline if this company downsizes or closes. This positive association between company size and labor earnings could hold even in cases where the company takes advantage of its monopsony power to set wages below the marginal productivity of labor. Labor earnings in the town could be even higher if the company behaved as a wage taker. But even if it behaved as a wage maker, there should be a positive exter- nality of the company on the town's labor earnings provided that the local labor supply is not infinitely elastic and that some of the goods and services consumed by the local population are locally supplied. These two conditions are likely to be verified in many developing countries and transition economies. In the case of Kazakhstan, this article estimates the externality at around 1.5 percentage points for every 1 percent of the local population employed by the company. But the externality could be larger in the short run. This 1.5 percent figure results from a comparison between communities harboring companies of different sizes, so it captures differences in labor earnings between equilibrium situations. Most of those companies, if not all of them, were presumably in operation when the transition to a market economy started, almost a decade ago. In spite of limited labor mobility, there must have been some labor reallo- cation across these communities in response to differences in earnings. The ex- tent of labor reallocation ought to be much smaller over shorter periods of time, so that a downsizing process affecting 1 percent of the local population would probably reduce local labor earnings more than 1.5 percent. Whereas the estimates in this article highlight a potential side effect of downsizing operations in one-company towns, their implications for public policy 208 THE WORLD BANK ECONOMIC REVIEW, VOL 13, NO I are not straightforward. Even if the 1.5 percent figure could be taken literally (which it cannot), policy analysis should not be based on this earnings effect only but rather on the overall welfare impact of company downsizing. Assessing that impact requires a theoretical model such as the one we have proposed here. We interpret the positive earnings externality from the company as evidence that our theoretical model may be empirically relevant. But fully exploiting that model to identify the optimal government policy toward one-company towns is clearly beyond the scope of this article. Our discussion of the optimal subsidy to the company should be interpreted as an illustration only. A more general dis- cussion should consider other policy instruments, such as subsidies to local firms or income transfers for those who do not work. A cautionary note is warranted. The theoretical part of the article is based on assumptions that are not too restrictive, so that its implications may have gen- eral validity. The empirical part of the article, however, concerns a specific coun- try. Other transition economies and developing countries may differ from Kazakhstan in the extent of their labor mobility across communities or in the degree to which the product markets of those communities are connected. Con- sequently, the effect of the company's employment on the town's labor earnings may be quite different from the effect estimated in the case of Kazakhstan. The effect estimated here should be seen as indicating a plausible order of magni- tude. Careful studies of actual downsizing operations, in Kazakhstan and else- where, would be needed to assess its relevance. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Boal, William M. 1995. "Testing for Employer Monopsony in Turn-of-the-Century Coal Mining." Rand Journal of Economics 26(3):519-36. Boal, William M., and Michael R. Ransom. 1997. "Monopsony in the Labor Market." Journal of Economic Literature 35(1, March):86-112. Bor]as, GeorgeJ. 1995. "Ethnicity, Neighborhoods, and Human-Capital Externalities." American Economic Review 85(3):365-90. Buder, Stanley. 1967. Pullman: An Experiment in Industrial Order and Community Planning, 1880-1930. New York: Oxford University Press. Fishback, Price V. 1992. "The Economics of Company Housing: Historical Perspectives from the Coal Fields." Journal of Law, Economics, and Organlzation 8(2):346-65. . 1997. "Operations of 'Unfettered' Labor Markets: Exit and Voice in American Labor Markets at the Turn of the Century." NBER Working Paper Series on Historical Factors in Long-Run Growth 105. National Bureau of Economic Research, Cam- bridge, Mass. Grosh, Margaret E., and Paul Glewwe. 1995. A Guide to Living Standards Measure- ment Study Surveys and Their Data Sets. LSMS Working Paper 120. Washington, D.C.: World Bank. Rama and Scott 209 James, Wilmot G. 1992. Our Precious Metal: African Labour in South Africa's Gold Industry, 1970-1990. Bloomington: Indiana University Press. Klugman, Jeni, and Kinnon Scott. 1997. "Measuring Labor Market Activities in Kazakhstan: The Implications of Using Official Statistics." In Jane Falkingham, ed., Household Welfare in Central Asia. New York: St. Martin's Press. Murthi, Mamta. 1998. "Kazakhstan: Labor Market Study." Europe and Central Asia Human Development Sector Unit, World Bank, Washington, D.C. Processed. Narayan, Deepa, and Lant Pritchett. 1997. "Cents and Sociability: Household Income and Social Capital in Rural Tanzania." Policy Research Working Paper 1796. Policy Research Department, World Bank, Washington, D.C. Processed. Ravallion, Martin, and Quentin Wodon. 1997. "Poor Areas or Only Poor People?" Policy Research Working Paper 1798. Policy Research Department, World Bank, Washington, D.C. Processed. Sullivan, Daniel. 1989. "Monopsony Power in the Market for Nurses." Joufrnal of Law and Economics 32(2, October):S135-78. Topel, Robert. 1986. "Local Labor Markets." Journal of Political Economy 94(3):S1 11- 44. . 1994. "Regional Labor Markets and the Determinants of Wage Inequality." American Economic Review: Papers and Proceedings 84(5):16-22. World Bank. 1998. "Kazakhstan: Living Standards during the Transition." Report KZ- SR-19398. Europe and Central Asia Human Development Sector Unit, World Bank, Washington, D.C. Processed. August 23-27, 1999 Buenos Aires, Argentina The XII World Congress of the IEA will be organised by the Asociaci6n Argentina de Economia Politica.The five day programme includes plenary sessions, invited sessions and presentation of contributed papers. Invited lectures will cover two topics:"Inequahty in the WorldToday", with a programme arranged by Professor Richard Freeman (Harvard, USA), and "The Current State of Macroeconomics", arranged by Professor Jacques Dreze (CORE, Belgium) Callfor Papers Submissions of papers related to any economic area are welcomed The Programme Committee will screen the contributed papers Authors will be notified byApril 15.Three volumes of proceedings on macroeconomics, inequality and Latin American issues will be published Those wishing to participate with a paper should send three copies of the paper by February 14, 1999, to Prof David de la Croix, IRES - Universite Catholique de Louvain, Place Montesquieu 3, B-1348 Louvain - la - Neuve, Belgium,Tel + 3210473453 Fax. + 3210473945 e-mail delacroix@ires.ucl.ac.be Programme Committee Chairman David de la Croix (Uriversit6 Catholique de Louvain), BinaAgarwall (University of Delhi), Kenneth Arrow (Stanford University), Anthony Atkinson (Oxford University), Costas Azariadis (University of California), Parkash Chander (Indian Statistical Institute),Vittorio Corbo (Catholic University of Chile), Karel Dyba (Ministry of Economics, Czech Repubhlic),V'ctor El'as (University of Tucumdn),Jean-Paul Fitoussi (OFCE - Paris),Jean-Michel Grandmont (CNRS), Daniel Hamermesh (University of Texas),Yujiro Hayami (Aoyama Gakuin University), Seppo Honkapohja (University of Helsinki),Valeri Makarov (Russian Academy of Sciences), Mustapha Nabli (World Bank), Fernando Navajas (University of La Plata), Dan O 'Flaherty (Columbia University), Luigi Pasinetti (Universlta Cattolica del Sacro Cuore), Suzanne Scotchmer (University of California),Amartya K Sen (Harvard University), Hans Werner Sinn (University of Munich), Rehman Sohban (Centre for Policy Dialogue-Dhaka), Erich Streissler (Universitat Wien), Kotaro Suzumura (Hitotsubashi University),Jean-Plerre Urbain (University of Maastricht),Alan Woodland (University of Sidney) and Stefano Zamagni (Universita deglh Studi di Bologna) Organising Committee Chairman Enrique A Bour (University of Buenos Aires, FIEL) e-mall enrique@fiel org ar, Vice Chairman Rolf R Mantel (University of San Andres) e- mail rrmantel@infotrek com ar,Executive Secretary Victor A Beker (University of Belgrano, Univ Buenos Aires) e-mail beker@ub edu ar ,9 I- I Introducing a publication with the most vital commodity of all: thorough, reliable information. Subscribe now to Global Commodity Markets, the only publication that gives you the complete picture on world commodities. With the accuracy and objectivity of the World Bank. This new World Bank quarterly replaces -and significantly expands upon- its predecessor, Commodity Markets and the Developing Countries. You'll find price forecasts for 46 primary com- modities and detailed market reviews for 27 major ones. The detailed reviews, featured on generous two-page spreads, are packed with useful information, from end-of-year stocks to major *_ = producers to futures prices. The World Bank also provides objec- tive market analysis. A full-service subscription includes not only the quarterly publication but also electronic Monthly Updates and electronic access to the quarterly. The Updates offer the most recent commodity prices and a brief review of market developments. A full-service subscription is just $645 per calendar year. You may also subscribe to the quarterly alone for $495. Subscriptions in 1999 are for a calendar year and will be prorated on a quarterly basis. If you need dependable information on world commodities, subscribe now to Global Commodity Markets. It's your best vantage point for a global view. II] * Yes, sign me up for Global Commodity Markets. Payment Orders from individuals must be accompanied by pay- ment or credit card information Credit cards accepted io Full-service subscnption, including the quarterly publication, electronic Monthly Updates, only for orders addressed to the World Bank Check wih and electronic access to the quarterly, at $645 a year P your local distributor about acceptable credt cards i i Quarterly publication (pnnt only) at $495 a year* Subscnptions are fora calendaryear and will be prorated on a quarterly basis 0D Check no mi the amount of$_i is enclosed When ordenng directly from the World Bank, make Name _ check payable in U S funds only drawn on a U S bank to The World Bank Please send your check with your order Title __l_I 0 Charge $ _ _ to my Organization E American Express O Mastercard 0 Visa Address Card no City State Zip | Expiration date _ / _ _ Phone Fax | i. Name as it appears on the card (required for all credit card charges) jE-mad Il Signature _ Nate: We must have a fax number or e-mail address to notify you of Monthly Updates 0 Bill me Instiutional customers IN THE U S ONLY Please provide both if possible Please include purchase order Mall order to: The World Bank, PO Box 7247-7956, Philadelphia, PA 19170-7956, USA Fax order to. 201-476-2197 Orderonline: httplAvwwworldbankorglhtml/extpb/publicabonshtml Questions? E-mail us at booksowordbank org AGCER -- - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - - - - Distributors of World Bank Group Publications Prices and credit terms vary from Chinese Corporation for Promotion FINLAND IRAN country to country Consult your of Humanities Akateeminen Kirlakauppa Ketab Sara Co Publishers local distributor before placing an 52, You Fang Hu Tong, PO Box 128 Khaled Eslamboli Ave, 6th Street order Xuan Nei DaJie FIN-00101 Helsinki Delafrooz Alley No 8 Beijing Tel (358 0)121 4418 PO Box 15745-733 ARGENTINA Tel (8610)66072494 Fax (3580)121-4435 Tehran 15117 World Publications SA Fax (86 10) 660 72 494 E-mail akatilaus@stockmann fi Tel (98 21) 8717819, 8716104 Av Cordoba 1877 COLOMBIA URL www akateeminen com Fax (98 21) 8712479 1120 Ciudad de Buenos Aires Infoenlace Ltda FRANCE E-mail ketab-sara@neda net ir Tel (54 11) 4815-8156 Carrera 6 No 51-21 Editions Eska, DBJ/Offilib Kowkab Publishers Fax (54 11) 4815-8156 Apartado Aereo 34270 48, rue Gay Lussac PO Box 19575-511 E-mail wpbooks@infovia corm ar Santafe de Bogota, D C 75005 Paris Tehran AUSTRALIA, FIJI, PAPUA NEW Tel (571)285-2798 Tel (33-1) 43-29-21-32 Tel (98 21) 258-3723 GUINEA, SOLOMON ISLANDS, Fax (571) 285-2798 Fax (33-1) 43-29-91-67 Fax (98 21) 258-3723 VANUATU,iAND SAMOA COTE D'IVOIRE E-mail offilib@otfilib fr IRELAND D A Informabon Services Center d'Edition et de Diffusion URL www offilib fr Government Supplies Agency 648 Whitehorse Road Africaines (CEDA) GERMANY Oifig an tSolathair Mitcham 3132, Victoria 04 B P 541 UNO-Verlag 4-5 Harcourt Road Tel (61) 3 9210 7777 Abidjan 04 Poppelsdorfer Allee 55 Dublin 2 Fax (61) 3 9210 7788 Tel (225) 24 6510, 24 6511 53115 Bonn Tel (353 1) 661-3111 E-mail service@dadirect com au Fax (225) 25 0567 Tel (49 228) 949020 Fax (353 1) 475-2670 URL www dadirect com au CYPRUS Fax (49 228) 217492 ISRAEL AUSTRIA Center for Applied Research URL www uno-verlag de Yozmot Literature Ltd Gerold and Co Cyprus College E-mail unoverlag@aol com PO Box 56055 Weihburggasse 26 6, Diogenes Street, Engomi GHANA 3 Yohanan Hasandlar Street A-1011 Wien PO Box 2006 Epp Books Services Tel Aviv 61560 Tel (43 1) 512-47-31-0 Nicosia PO Box 44 Tel (9723) 5285-397 Fax (431)512-47-31-29 Tel (3572)59-0730 TUC Fax (9723)5285-397 URL www gerold co/at online Fax (357 2) 66-2051 Accra R O Y International BANGLADESH CZECH REPUBLIC Tel 223 21 778843 PO Box 13056 Micro Industries Development USIS, NIS Prodejna Fax 223 21 779099 Tel Aviv 61130 Assistance Society (MIDAS) Havelkova 22 GREECE Tel (972 3) 649 9469 House 5, Road 16 130 00 Prague 3 Papasotiriou S A Fax (972 3) 648 6039 Dhanmondi R/Area Tel (4202)2423 1486 35, Stournara Str E-mail royi@netvision net Dhaka 1209 Fax (4202)24231114 106 82 Athens URL www royintcol Fax (8802)811188 URL wwws(cz/ 301) 364-8254 Palestinian Authority/Middle East Fax (880 2) 811188 DENMARK Fax (30 1) 364-8254 Index Information Services BELGIUM SamfundsLitteratur HAITI PO B 19502 Jerusalem Jean De Lannoy Rosenoerns Alle 11 Culture Diffusion Tel (972 2) 6271219 Av du Roi 202 DK-1970 Frederiksberg C 5, Rue Capois Fax (972 2) 6271634 1060 Brussels Tel (4535)351942 C P 257 ITALY Tel (32 2) 538-5169 Fax (45 35) 357822 Port-au-Prince LibreT aLYcosaCommissionaria Fax (32 2) 538-0841 URL www sl cbs dk Tel (509) 23 9260 Sanson a SPA BRAZIL ECUADOR Fax (509) 23 4858 Via Duca Di Calabria, 1/1 Publicaches Tecnicas Internacionais Libri Mundi HONG KONG, CHINA; MACAO Casella Postale 552 Ltda Libreria Internacional Asia 2000 Ltd 50125 Firenze Rua Peixoto Gomide, 209 PO Box 17-01-3029 Sales & Circulation Department Tel (39 55) 645-415 01409 Sao Paulo, SP Juan Leon Mera 851 302 Seabird House Fax (39 55) 641-257 Tel (5511) 259-6644 Quito 22-28Wyndham Street, Central E-mail licosa@ftbcc it Fax (55 11) 258-6990 Tel (593 2) 521-606, Hong Kong, China URL www ftbcc it/licosa E-mail postmaster@pti uol br (5932)544-185 Tel (852) 2530-1409 JAMAICA URL www uol br Fax (593 2) 504-209 Fax (852) 2526-1107 laJ Randle PubI shers Ud CANADA E-mail librimul@librimundi com ec E-mail sales@asia2000 com hk 206 Old Hope Road, Kingston 6 Renouf Publishing Co Ltd E-mail librimu2@librimundi com ec URL www asia2000 com hk Tel 876-927-2085 369 Canotek Road CODEU HUNGARY Fax 876-977-0243 Ottawa, Ontario KlJ 9J3 Ruiz de Castilla 763, Edif Expocolor Euro Info Service E-mail irpl@colis com Tel (613) 745-2665 Primer piso, Of #2 Margitszgeti Europa Haz JAPAN Fax (613) 745-7660 Quito H-1138 Budapest Eastern Book Service E-mail Tel/Fax (593 2) 507-383, 253-091 Tel (36 1)350 80 24, 350 80 25 3-13 Hongo 3-chome, Bunkyo-ku order dept@renoufbooks cam E-mail codeu@impsat net eo Fax (36 1) 350 90 32 Tokyo 113 URL www renoufbooks com EGYPT, ARAB REPUBLIC OF E-mail euroinfo@mail matav hu Tel (81 3) 3818-0861 CHINA Al Ahram Distribution Agency INDIA Fax (81 3) 3818-0864 China Financial & Economic Al Galaa Street Allied Publishers Ltd E-mail orders@svt-ebs co lP Publishing House Cairo 751 Mount Road URL 8, Da Fo Si Dong Jie Tel (20 2) 578-6083 Madras - 600 002 www bekkoame or lp/-svt-ebs Beli(1g Fax (20 2) 578-6833 Tel (91 44) 852-3938 KENYA Fax (86 10) 6401-7365 The Middle East Observer Fax (91 44) 852-0649 Africa Book Service (E A ) Ltd 41, Sherif Street INDONESIA Quaran House, Mfangano Street China Book Import Centre Cairo Pt Indira Limited PO Box 45245 PO Box 2825 Tel (202)392-6919 Jalan Borobudur 20 Nairobi Beiing Fax (202)393-9732 PO Box 181 Tel (2542)223641 E-mail Jakarta 10320 Fax (254 2) 330 272 mafouda@meobserver com eg Tel (62 21) 390-4290 URL wwwmeobservercomeg Fax (62 21) 390-4289 Legacy Books Oasis Official SINGAPORE, TAIWAN, CHINA THAILAND Loita House PO Box 3627 MYANMAR; BRUNEI Central Books Distribution Mezzanine I Wellington Hemisphere Publication Services 306 Sdiom Road PO Box 68077 Tel (64 4) 4991551 41 Kallang Pudding Road #04-03 Bangkok 10500 Nairobi Fax (64 4) 499 1972 Golden Wheel Building Tel (66 2) 2336930-9 Tel (254) 2-330853, 221426 E-mail oasisiactrix gen nz Singapore 349316 Fax (662) 237-8321 Fax (254) 2-330854, 561654 URL www oasisbooks co nz/ Tel (65) 741-5166 TRINIDAD & TOBAGO E-mail Legacy@form-net com Fax (65) 742-9356 AND THE CARRIBBEAN KOREA, REPUBLIC OF NIGERIA E-mail info@hemisphere com sg Systematics Studies Ltd Dayang Books Trading Co University Press Limited SLOVENIA St Augustine Shopping Center International Division Three Crowns Building Jericho Gospodarski vestnik Publishing Eastern Main Road, St Augustine 783-20, Pangba Bon-Dong, Private Mail Bag 5095 Group Trinidad & Tobago, West Indies Socho-ku Tae an Dunajska cesta 5 Tel (868) 645-8466 Seoul Tel (23422)41-1356 1000 Ljubljana Fax (868) 645-8467 Tel (82 2) 536-9555 Fax (234 22) 41-2056 Tel (386 61) 133 83 47,132 12 30 E-mail tobe@trinidad net Fax (82 2) 536-0025 Fax (386 61) 133 80 30 UGANDA E-mail seamap@chollian net PA BISTAN E-mail repansekligvestniks Gustro Ltd Eulyoo Publishing Co , Ltd 65, Shahrah-e-Quaid-e-Azam SOUTH AFRICA, BOTSWANA PO Box 9997, Madhvani Building 46-1, Susong-Dong Lahore 54000 For single titles Plot 16/4 Jinja Rd Jongro-Gu Tel (92 42) 735 3601 Oxford University Press Southern Kampala Seoul Fax (92 42) 576 3714 Africa Tel (256 41) 251 467 Tel (82 2) 734-3515 Vasco Boulevard, Goodwood Fax (256 41) 251 468 Fax (822)732-9154 Oxford University Press PO Box 12119, Ni City 7463 E-mail gus@swiftuganda com LEBANON 5 Bangalore Town Cape Town UNITED KINGDOM Librairie du Liban Sharae Faisal Tel (27 21) 595 4400 MIcroinfo Ltd PO Box 11-9232 PO Box 13033 Fax (27 21) 595 4430 PO Box 3 Omega Park, Alton Beirut Karachi-75350 E-mail oxford@oup co za Hampshire GU34 2PG Tel (961 9) 217 944 Tel (92 21) 446307 For subscription orders England Fax (961 9) 217 434 Fax (92 21) 4547640 International Subscription Service Tel (44 1420) 86848 E-mail hsayegh@hlbraine-du- E-mail ouppak@TheOffice net PO Box 41095 Fax (44 1420) 89889 liban com lb Pak Book Corporation Craighall E-mail wbank@microinfo co uk URL www librairie-du-liban com lb Aziz Chambers 21, Queen's Road Johannesburg 2024 URL www microinfo co uk MALAYSIA Lahore Tel (27 11) 880-1448 The Stationery Office University ot Malaya Cooperative Tel (92 42) 636 3222, 636 0885 Fax (27 11) 880-624b 51 Nine Elms Lane Bookshop, Limited Fax (92 42) 636 2328 E-mail issis co za London SW8 5DR P0 Box-1127 E-mail pbc@brain net pk SPAIN Tel (44 171) 873-8372 Jalan Pantai Baru PERU Mundi-Prensa Libros, S A Fax (44 171) 873-8242 59700 Kuala Lumpur Editorial Desarrollo SA Castello 37 URL wwwthe-stationery- Tel (60 3) 756-5000 Apartado 3824 Ica 242 OF 106 28001 Madrid office co uk/ Fax (60 3) 755-4424 Lima 1 Tel (34 91) 4363700 VENEZUELA E-mail umkoop@tm net my Tel (51 14) 285380 Fax (34 91) 5 753998 Tecni-Ciencia Libros, S A MEXICO Fax (51 14) 286628 E-mail libreria@mundiprensa es Centro Cuidad Comercial Tamanaco INFOTEC PHILIPPINES URL www mundiprensa com/ Nivel C2, Caracas Av San Fernando No 37 InternaEonalBooksourceCenter Mundi-Prensa Barcelona Tel (58 2) 959 5547, 5035, 0016 Col Toriello Guerra Intentoac oksoc etr Conseil de Cent, 391 Fax (58 2) 959 5636 14050 Mexico, O F 1127-A Antipolo St, Barangay, 08009 Barcelona ZMI Tel (52 5) 624-2800 Venezuela Tel (34 3) 488-3492 University Bookshop University ot Fax (52 5) 624-2822 Makati City Fax (34 3) 487-7659 Zambia E-mail infotec@rtn net mx Tel (63 2) 896 6501, 6505, 6507 E-mail barcelona@mundiprensa es Great East Road Campus URL rtn net mx Fax (63 2) 896 1741 SRI LANKA, THE MALDIVES PO Box 32379 Mundi-Prensa Mexico S A de C V POLAND Lake House Bookshop Lusaka c/Rio Panuco, 141-Colonia International Publishing Service 100, Sir Chittampalam Gardiner Tel (260 1) 252 576 Cuauhtemoc Ul Piekna 31/37 Mawatha Fax (260 1) 253 952 06500 Mexico, D F 00-677 Warzawa Colombo 2 ZIMBABWE Tel (525) 533-5658 Tel (482)628-6089 Tel (94 1) 32105 Academic and Baobab Books (Pvt Fax (52 5) 514-6799 Fax (48 2) 621-7255 Fax (94 1) 432104Lt NEPAL E-mail books%ips@ikpatmcompl E-mail LHL@sri lanka net 4 Conald Road, Graniteside Everest Media International Services URL www ipscg waw pl/ips/export SWEDEN PO Box 567 (P ) Ltd PORTUGAL Wennergren-Williams AB Harare GPO Box 5443 L vraraPortugal P0 Box 1305 Tel 2634755035 Kathmandu ivpartad 28,RaDCrm S-171 25 Solna Fax 263 4781913 Tel (977 1) 416 026 Apartado 2681, Rua Do Carm Tel (46 8) 705-97-50 Fax (9771) 224 431 1200 Lisbon Fax (46 8) 27-00-71 Tel (1) 347-4982 E-mail mail@wwi se NETHERLANDS Fe 1 3748 DETHERLNDeSoom/nternatione Fax (1) 347-0264 SWITZERLAND PbLcaties b v - ROMANIA LLibrairie Payot Service Institutionnel PO Box 202, 7480 AE Haaksbergen Compani De Librarii Bucuresti S A C(tm)tes-de-Montbenon 30 Tel (31 53) 574-0004 Str Lipscani no 26, sector 3 1002 Lausanne Fax (31 53) 572-9296 Bucharest Tel (41 21) 341-3229 E-mail lindeboo@worldonline nl Tel (401)3139645 Fax (4121)341-3235 URL www worldonline nl/-lindeboo Fax (40 1) 312 4000 ADECO Van Diermen RUSSIAN FEDERATION EditionsTechniques NEW ZEALAND Isdatelstvo Ch de Lacuez 41 EBSCO NZ Ltd 9a, Kolpachny Pereulok CH1807 Blonay Private Mail Bag 99914 Ma ospchni Pereu83 k Tel (41 21) 943 2673 New Market Tel (7 095) 917 87 49 Fax (41 21) 943 3605 Auckland Fax (7 095) 917 92 59 Tel (64 9) 524-8119 ozimarin@glasnet ru Fax (64 9) 524-8067 Coming in the next issue of THE WORLD BANK ECONOMIC REVIEW May 1999 Volume 13, Nuimber 2 * Household Income and Child Schooling in Vietnam by Jere R. Behrman and James C. Knowles * Winners and Losers from the Privatization and Regulation of Utilities: Lessons from a General Equilibrium Model of Argentina by Omar Chisari, Antonio Estache, and Carlos Romero * Determinants of Commercial Bank Interest Margins and Profitability: Some International Evidence by Asli Demirgiiu-Kunt and Harry Huizinga * Genuine Savings Rates in Developing Countries by Kirk Hamilton and Michael Clemens * Benefit Incidence and the Timing of Program Capture by Peter Lanjouiw and Martin Ravallion * Tax Incidence in Madagascar: An Analysis Using Household Data by Stephen D. Younger, David E. Sahn, Steveni Haggblade, and Paul A. Dorosh Sv)m9bc D.C. 23039 Uo$oAo 7Ldjpm8 (202X 477-123 v(oEg Md mb.. bm E$BM tO-IM34124E