THE WORLD BANK ECONOMIC REVIEW Volume 6 January 1992 Number I \9000 A SYMPOSIUM ON INFLATION IN SOCIALIST ECONOMIES IN TRANSITION Inflation and the Transition to a Market Economy: An Overview Simon Commander Lessons from Experiences with High Inflation Rudiger Dornbusch Price-Wage Dynamics and Inflation in Socialist Economies: Empirical Models for Hungary and Poland Simon Commander and Fabrizio Coricelli Are High Interest Rates Effective for Stopping High Inflation? Some Skeptical Notes Guillermo A. Calvo Stagflationary Effects of Stabilization Programs in Reforming Socialist Countries: Enterprise-Side and Household-Side Factors Guillermo A. Calvo and Fabrizio Coricelli Voluntary Choices in Concerted Deals: The Menu Approach to Debt Reduction in Developing Countries Ishac Diwan and Kenneth Kletzer Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia Martin Ravallion Alleviating Transitory Food Crises in Sub-Saharan Africa: International Altruism and Trade Victor Lavy The Working Behavior of Young People in Rural Cote d'lvoire Rob Alessie, Paul Baker, Richard Blundell, Christopher Heady, and Costas Meghir Social Security and Private Transfers in Developing Countries: The Case of Peru Donald Cox and Emmanuel Jimenez The Willingness to Pay for Education for Daughters in Contrast to Sons: Evidence from Rural Peru Paul Gertler and Paul Glewwe THE WORLD BANK ECONOMIC REVIEW EDITOR Ravi Kanbur CONSULTING EDITOR Sandra Gain EDITORIAL BOARD Kaushik Basu, University of Delhi John Holsen Guillermo A. Calvo, International Monetary Fund Gregory K. Ingram Alberto Giovannini, Columbia University Ravi Kanbur Mark R. Rosenzweig, University of Pennsylvania Mieko Nishimizu Joseph Stiglitz, Stanford University Jacques van der Gaag Sweder van Wiinbergen 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 issued three times a year, with an occasional additional issue devoted to a special area of current research or policy interest. The Review 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 operational aspects of eco- nomics, rather than primarily theoretical and methodological issues. It is intended for readers familiar with economic theory and analysis but not necessarily proficient in advanced mathemati- cal or econometric techniques. Articles will illustrate how professional research can shed light on policy 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 members of the Bank's staff; 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 in articles published are those of the authors and do not neces- sarily represent the views and policies of the World Bank or of its executive directors or the countries they represent. When maps are used, the designations employed are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the World Bank or its affiliates concerning the legal status of any country, territory, city, or area, or concerning the delimitations of its boundaries or national affiliation. Comments or brief notes responding to Review articles are welcome and will be considered for publication 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. Subscriptions to The World Bank Economic Review are available without charge to readers with mailing addresses in developing countries and in socialist economies in transition. Written request is required every three years to renew such subscriptions. See the coupon at the back of this issue for subscription charges and additional information. © 1992 The International Bank for Reconstruction and Development / THE WORLD BANK All rights reserved Manufactured in the United States of America ISSN 0258-6770 Material in this journal is copyrighted. Requests for permission to reproduce or translate articles should be addressed to the Director, Publications Department, The World Bank, Washington, D.C. 20433, U.S.A. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the intended reproduction is for noncommercial purposes, with- out asking a fee. Permission is not required to make photocopies for classroom and similar immediate use. This journal is indexed regularly in Current Contents/Social & Behavioral Sciences, Index to International Statistics, Journal of Economic Literature, Public Affairs Information Service, and Social Sciences Citation Indexg. TH]E WORLD BANK ECONOMIC REVIEW Volume 6 January 1992 Number 1 SYMPOSIUM ON INFLATION IN SOCIALIST ECONOMIES IN TRANSITION Inflation and the Transition to a Market Economy: 3 An Overview Simon Commander Lessons from Experiences with High Inflation 13 Rudiger Dornbusch Price-Wage Dynamics and Inflation in Socialist Economies: 33 Empirical Models for Hungary and Poland Simon Commander and Fabrizio Coricelli Are High Interest Rates Effective for Stopping High Inflation? 55 Some Skeptical Notes Guillermo A. Calvo Stagflationary Effects of Stabilization Programs in Reforming 71 Socialist Countries: Enterprise-Side and Household-Side Factors Guillermo A. Calvo and Fabrizio Coricelli ARTICLES Voluntary Choices in Concerted Deals: 91 The Menu Approach to Debt Reduction in Developing Countries Ishac Diwan and Kenneth Kletzer Does Undernutrition Respond to Incomes and Prices? 109 Dominance Tests for Indonesia Martin Ravallion Alleviating Transitory Food Crises in Sub-Saharan Africa: 125 International Altruism and Trade Victor Lavy The Working Behavior of Young People in Rural Cote d'Ivoire 139 Rob Alessie, Paul Baker, Richard Blundell, Christopher Heady, and Costas Meghir Social Security and Private Transfers in Developing Countries: 155 The Case of Peru Donald Cox and Emmanuel Jimenez The Willingness to Pay for Education for Daughters in Contrast 171 to Sons: Evidence from Rural Peru Paul Gertler and Paul Glewwe A SYMPOSIUM ON INFLATION IN SOCIALIST ECONOMIES IN TRANSITION This symposium draws primarily on papers presented at the Conference on Manag- ing Inflation in Socialist Economies, sponsored by the Economic Development Institute of the World Bank, in Laxenburg, Austria, March 1990. The articles in the symposium were refereed in the usual way. The Editorial Board invited the orga- nizer of the conference, Simon Commander, to write the introduction to the symposium. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 3-12 Inflation and the Transition to a Market Economy: An Overview Simon Commander European socialist economies were commonly characterized by low or negligible open inflation, full employment, and stable relative prices and real incomes. Associated features were significant repressed or hidden inflation and disequilibria in goods mar- kets. More recently, as economic and political reform has advanced, attention has shifted to translating repressed into open inflation. Where such reforms have proceeded against a background of a monetary overhang, the implications of that overhang for demand-side policies have figured prominently. In all instances, however, the key un- derlying issue has been the transmission mechanism for inflation once the initial im- pulse associated with price liberalization has been imparted. Although it is widely recognized that major price liberalization has to be central to any stabilization, there is considerable debate concerning the optimal pace and the dynamic effects of price reforms. The arguments for liberalization turn on the importance of achieving a rational, relative price structure that can eliminate imbalances in the goods market while enabling allocative efficiency gains. Yet, as recent experience in Eastern Europe testifies, a variety of price liberalization schemes have involved significant jumps in the price level that translate into acceleration in the rate of increase of prices. In addition, develop- ments on the real side have been seriously adverse, with large declines in re- corded output. There is further evidence of an initially slow, but accelerating, increase in the rate of open unemployment. The articles in this symposium concentrate on the problem of inflation in socialist economies in transition. Given recent experience in several reforming socialist economies with high inflation, Dornbusch draws out the main compar- ative lessons from other high-inflationary experiences, particularly in Latin America. Stabilizing inflation requires fiscal correction and reforms to the tax system. There is commonly a role for incomes policy in achieving a more rapid and coordinated disinflation, but only as a complement to, rather than a sub- stitute for, budget balancing. Simon Commander is with the Economic Development Institute of the World Bank. He would like to thank Sara Calvo and Peter Knight for helpful comments on an earlier draft. ( 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 3 4 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Commander and Coricelli use data from Hungary and Poland to analyze the propagation mechanisms for inflation in a partially reformed economy. Devel- opments on the cost side are critical in relating exogenous, policy-determined adjustments to the price level to increases in the rate of inflation. Further, the absence of conventional equilibrating mechanisms and of market-based re- straints on prices and wages gives rise to underlying inflationary pressure, indi- cating, among other things, the importance of incomes policy in any stabilization. Calvo deals with the role of interest rate policy in a stabilization program and outlines the possible adverse consequences of applying high interest rates over protracted periods. If a stabilization program does not enjoy full credibility, the contraction that high rates are supposed to induce may not result. This is likely to have particular relevance for the reforming socialist economies, given segmen- tation in the financial system and the absence of mechanisms to direct household savings to the firm sector. This theme is further developed by Calvo and Coricelli, who invoke interest rates and credit policy as major explanatory factors behind both output decline and inflation persistence in the Polish stabilization. Technological and credit market shocks, rather than demand-side developments, are given prominence, which raises questions about the consequences of standard macroeconomic pol- icy prescriptions in a reforming socialist economy. This introduction emphasizes the initial conditions of the prereform socialist economies and the mechanisms by which inflationary pressures are carried over into the accelerated reform period. Recent experiences with price reform and stabilization are summarized, with attention focused on the phenomenon of persistent inflation. I. THE PREREFORM SOCIALIST ECONOMIES In a centrally planned economy decisions are largely based on considerations of quantity. The partially reformed economy accords a greater role to (con- strained) market-based rules, with progressive price liberalization and the emer- gence of dual pricing systems. Bulgaria, Czechoslovakia, and Romania in 1989- 90 correspond to the former system; Hungary and Poland in the 1980s to the partially reformed case. In both regimes unemployment was not tolerated, mon- etary policy was basically passive, and enterprises not only did not have to behave as profit maximizers but, to a certain degree, could be indifferent to financial performance. Inflationary tensions in these systems had different manifestations. It was widely held that endemic shortages provoked excess demand in both goods and factor markets due to fixed or sticky prices and wages and that this resulted in repressed inflation. Rationing occurred for both households and enterprises. The gap between notional and effective demand at official prices in goods mar- kets translated into excess liquid balances or excessive inventories (Nuti 1986). Commander S Labor shortages resulting from imposed full employment caused enterprises to hoard labor. Along with the institutional bargaining relationships and the rela- tively weak bargaining strength of managers, the earlier systems were marked by powerful wage pressure on costs. Enterprises, particularly in partially reformed economies, pursued explicit cost-plus pricing routines. The pace at which cost increases were passed through to consumers depended on how automatic product-specific markups were and hence, in part, on the degree of the planner's or government's aversion to price increases. In this setting, and with high levels of concentration in industry without administrative controls over price and wage changes, inflation (open or re- pressed) could accelerate rapidly. In short, no endogenous equilibrating mecha- nism existed in the system. The underlying inflationary mechanism could thus be traced back to a combination of factors. The low productive efficiency of enterprises and the regional structures of production and trade associated with the Council for Mutual Economic Assistance (CMEA) arrangement generated shortages and imbalances in goods markets and hence chronic sellers' markets. Further, the soft budget constraints of the enterprise sector and the banking system fueled increases in subsidy payments and provoked fiscal crises. With pervasive price controls, exogenous price shocks tended to widen the gap be- tween (lomestic and international prices, thus creating more distorted relative price structures and fiscal pressures. Table 1 presents data on inflation in the European socialist economies and in industrial and developing countries during 1971-91. It is evident that selective price liberalization during the 1980s was associated with rising open inflation. The response of both centrally planned and partially reformed economies to the external shocks of the 1980s was generally to depress investment, maintain broadly stable real wages, and raise subsidies to enterprises through the budget, negative interest rates, or preferential credit allocations. The deteriorating per- formance of the enterprise sector and associated growth in subsidies had to be covered by inflationary finance. Servicing of the large external debt, restricted access to foreign borrowing in the 1980s, and limited instruments for placing domestic debt increased the rate of money creation associated with a given noninterest deficit. High inflationary episodes have occurred in Poland and Yugoslavia and, more recently, in the Soviet Union. In all cases the causal link between inflation and budget deficits is strong when money financing is the only option. Moreover, as Dornbusch emphasizes, not only will there be profound nonlinearity in the link between inflation and the deficit, but the real fiscal deficit is not exogenous because inflation determines real revenues. Inflation not only erodes revenues (Tanzi-Olivera effect), but in reforming socialist economies widespread with- holding of taxes by enterprises-the bedrock of the tax system-further deepens the fiscal crisis. An accommodating monetary stance traces back to the dynamics of subsidy policy resulting from price controls, the gap between domestic and external prices, and the pervasive soft budgets of the enterprises. Piecemeal price liberalization has tended to be strongly inflationary when fiscal correction was Table 1. Inflation in Eastern Europe, Industrial Countries, and Developing Countries, 1971-90 (percent) Country or Average, country group 1971-80 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 Bulgaria 2.0 0.5 0.3 1.5 0.7 1.8 3.9 0.1 1.2 6.2 64.0 330.0a Czechoslovakia 1.1 0.9 4.7 1.1 0.9 1.3 0.4 0.1 0.2 1.4 10.0 52.3b Romania 0.9 2.0 16.9 5.2 1.1 0.4 2.0 0.4 3.0 0.6 4.7 80.0a Soviet Union 0.3 1.0 3.0 1.0 -1.0 0.7 2.0 1.3 0.6 2.0 5.3 100. a0d CN Hungary 4.5 4.6 6.9 7.3 8.7 7.0 5.3 8.2 16.3 17.0 28.9 26.6b Poland 4.6 24.4 101.5 23.0 15.7 14.4 18.0 25.2 60.2 245.3 249.3 34.2b Yugoslavia 20.0 39.8 31.5 40.2 54.7 72.4 90.0 221.0 294.0 1,340.0 130.0 49.4a Industrial countries 8.6 10.1 7.6 5.2 4.7 4.1 2.4 3.0 3.4 4.5 5.0 2.0a Developingcountries 20.6 28.6 30.1 39.9 35.1 35.5 27.0 35.4 57.3 75.9 102.7 12.4c a. First 6 months. b. January-July 1991. c. January-April 1991. d. IMF estimate. Source: World Bank data. Commander 7 insufficient to offset subsidies generated by the gap between controlled and market prices. In Poland and Yugoslavia high inflation was preceded by flight from domestic money that would have increased the income velocity of domestic Ml and eroded the inflation tax base, thus raising equilibrium inflation (Coricelli and Rocha 1991). The accelerated dollarization characteristic of these cases reflected basic uncertainty regarding the direction of economic policy and the apparent lack of rules in determining price changes by the government. Anticipation of further price changes resulted in upward adjustments to market prices, thus widening the gap between administered and market prices, adversely acting on the fiscal deficit, and resulting in overshooting. In Yugoslavia currency depreciation, triggered by closing access to external credits to finance current account deficits and de facto wage indexation, proved a powerful inflationary channel. The inflationary spiral was triggered, and in part driven, by nominal exchange rate and wage increases consistent with real exchange rate and real wage targets and with the inflation tax generated by the foreign exchange target. However, although the central government accounts show no major fiscal imbalances, inclusion of the quasi-fiscal deficit yields a more conventional picture. Enterprise claims on the banking system expanded significantly through interest rate subsidies (Mates 1990; Rocha 1991). Real exchange rate shocks further resulted in enterprises being unable to service their foreign debts. Consequently, socialization of external liabilities and credit sub- sidies fueled the expansion of the quasi-fiscal deficit and was a major factor behind monetary expansion. De facto wage indexing raised the responsiveness of inflation to price shocks. A norrnal cost output pricing rule and a real wage target provoking a ratchet effect likely account for raising the inflation rate. An accommodating monetary policy would strengthen this result. Indeed, for Poland it appears that even in the high inflation period of 1989, wage stickiness, price controls, and an effectively indexed exchange rate served to maintain inflation inertia. Further, with indexa- tion a real depreciation would immediately feed through pricing and the budget into accelerated inflation. Piecemeal economic reforms that weakened centralized controls on both prices and wages tended to expose the latent inflationary pressures associated with the ownership and control structure of the economy. The weak competitive environment associated with high levels of concentration and market power and the underlying tendency for wage drift derived from either an explicit worker management system-as in Yugoslavia-or de facto worker bargaining power- as in Poland-are important explanatory factors. The coexistence of market and controlled prices with the associated emergence of nonsynchronization in price determination further imparted inertia. To the extent that the government ad- justment rule for controlled prices was targeted on excess demand, the likely outcomne through expectations was to destabilize the system (Commander and Coricelli 1990). 8 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 II. STABILIZATION SINCE I990 Since January 1990 most of the socialist economies in transition have at- tempted to put in place a stabilization program, a central component of which has been generalized price revisions leading to large upward shifts in the price level. Such programs have also explicitly aimed at a large contraction in sub- sidies and consequent fiscal correction alongside tight monetary policy and, in some instances, establishment of an independent central bank to signal the repudiation of the earlier accommodating monetary policy. In Czechoslovakia and Poland the exchange rate has been fixed to provide a nominal anchor to the stabilization. An additional nominal wage anchor has also been employed, with wage cuts enforced ex ante to avert a price-wage spiral. In Bulgaria the exchange rate has been allowed to float, with wages and money providing the main anchors. In Romania money has been used as the main anchor, with wages providing a secondary nominal anchor. A dual ex- change rate system was established in 1990, with convergence to a unified rate in 1991. Current account or internal convertibility has been established-except in Bulgaria and Romania-with varying degrees of trade opening. The stabilizations have aimed at very significant contractions in net domestic credit. In all instances output has fallen sharply. The collapse of the Soviet market and of CMEA trading relations appears to have accounted for some of the decline in output, but this is only a partial explanatory factor with a different incidence at both country and sectoral levels. A major decline in output preceded the dismantling of the CMEA and the oil price shock of 1990 in almost all cases. Wages have tended to fall sharply at the start of the stabilization, followed by some recovery. Unemployment has risen from a negligible base to a range of 6- 12 percent and is projected to accelerate more rapidly. In general, output has declined more sharply than unemployment has risen, resulting in reduced labor productivity. Relatively little privatization has been achieved as yet, and there has been minor formal bankruptcy of firms. The structure of the economies has remained broadly intact, but enterprises have in principle been forced to adapt their pricing and other behavior in response to the exogenous shift in rules. Persistent Inflation Large jumps in the price level, generally followed by periodic further revisions to remaining administered goods' prices, have generally eliminated any mone- tary overhang and shortages. In Poland a comprehensive trade opening resulted in the fairly immediate import of a new relative price structure (Lipton and Sachs 1990). But in Hungary and Poland price level effects have been followed by inflationary persistence. In the Polish case, 18 months after the start of the stabilization, the price level was nearly five times higher, output and real wages were roughly 40 percent lower, unemployment was at around 10 percent, but inflation was still running at 3-4 percent a month. The persistence of inflation can be related to the combined role of expecta- tions and credibility. Anticipation of further price jumps for incompletely liber- Commander 9 alized goods, such as energy and transportation, could be expected to feed back into current pricing behavior. Similarly, the structural stability of the system, given little privatization, and lags in effectively importing competition could continue to promote inflationary pressures characteristic of the partially re- formed economies. This could be expected to show up in the pricing behavior of enterprises and in the evolution of the markup, as well as in the behavior of wages. Strong wage pressure and slower than warranted employment adjust- ments would point to a lack of underlying credibility in the program as workers test the resolve of managers and government to enforce a hard budget constraint on firms. It can be reasonably assumed that any tradeoff between wage and price changes and unemployment is only partially or perversely perceived by agents. Calvo and Coricelli argue that the sources of persistent output contraction and inflation can be traced to the enterprise sector and rigidities in the financial system. Two initial conditions are particularly important. First, the financial system is segmented; there are different circuits for enterprises and households and no ready mechanism for channeling household savings to the enterprise sector. Second, there is a stock of interenterprise debt, reflecting one facet of the soft budget constraint. The principal impulse is through a supply-side shock, which raises input prices very significantly. A monetary crunch through tight credit ceilings and high interest rates has a fairly immediate impact on firms by raising the cost of working capital. This disqualifies firms from producing at full capacity and causes output contraction, labor redundancies, and the expansion of the interfirm credit market (see also Hrncir and Klacek 1991; Tardos 1990). The ex ante stock of interenterprise debt acts as a recessionary transmission mechanism, aside from complicating the discriminating role of a monetary squeeze across "good" and "bad" firms (Calvo and Frenkel 1991). Technological shocks through input price rises and credit market shocks that impart a liquidity squeeze may in part explain the depth and persistence of output reduction but does less well in explaining the persistence of inflation. On the cost side, it appears that the average markup over costs (net of subsidies) declined during 1990 but that this decline was concentrated in the last quarter (Schaffer 1991). Because wages fell significantly in the earlier part of the year, it seems likely that an undervalued fixed exchange rate allowed greater space for inflationary pricing and corresponding appreciation of the exchange rate. Price rather than wage shocks have dominated. There have been significant lags in the adjustment of domestic prices to international prices alongside nontrivial do- mestic relative price realignments (Blanchard and Layard 1991). Wage Pressure and Incomes Policy Available evidence indicates that autonomous wage-push or price-push through the raising of profit margins have not been major explanatory factors behind stubbornly high inflation rates in several of these economies during the most recent period. There is evidence, however, of reemerging wage-push pres- sures, particularly in Poland and Romania. The potential for wage drift is of concern especially because of uncertainty 10 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I over ownership and future profitability, given new sets of relative prices. To the extent that tight money and relatively high interest rates are associated with lower output, in turn yielding lower employment, probabilities of unemploy- ment and loss of access to labor or product market rents will feed back into current decisions by workers. This can be expected to have dear repercussions for wage claims, especially given the weight of labor-dominated firms in these economies. In such firms worker control can be treated as tantamount to a powerful trade union presence, where wages and employment are subject to a joint maximization. Workers will not be constrained directly by the labor de- mand curve but by the firm's profit level. Wages will tend to equate average rather than marginal product but will also depend on the degree of government aversion to bail-outs and hence the soft budget constraint (Commander, Cori- celli, and Staehr 1991). The institutional core and the absence of true performance-related discipline could be expected to fuel wage demands, particularly as workers seek to test the resolve of any reforming government to adhere to its announced policy goals of fiscal balance and tight money. A reforming government starts with low cred- ibility; it will need to shock workers and managers into behaving like profit maximizers and may thus welcome selective bankruptcies or partial closures to demonstrate commitment. Associated losses in output and employment would likewise be accepted. The institutional legacy and the tendency for wage drift emanating from decentralization and uncertainty provide powerful arguments for formal methods of wage regulation. There is no doubt that wage regulation will imply efficiency costs and impede the appropriate adjustment of relative wages, but the macroeconomic costs of abandoning wage controls in an environment in which the rules of the game are not fully perceived will also not be trivial. Conse- quently, a range of tax-based incomes policies have been applied. In Poland tax penalties were levied on above-norm wage payments, in which the object of regulation in 1990 was the firm's wage bill. Similarly, in Bulgaria, Czechoslovakia, Hungary, and Romania tax-based incomes policies have been implemented (Commander and Staehr 1991). An explicit incomes policy seems essential for attaining macroeconomic targets in the absence of a more complete introduction and comprehension of standard market-based equilibrating mecha- nisms. More difficult is the question of how long such controls should be main- tained and the impact of an emerging private sector on the relative wage struc- ture, labor allocation, and aggregate employment (Blanchard 1991). Institutional arrangements are crucial in determining the efficiency-both short- and longer-run-with which inflationary wage claims are contained. In this realm, some clear cross-country differences have emerged, with different, implicit longer-run models of wage bargaining as their underpinnings. In Poland there appears to be the expectation that a centrally imposed incomes policy would give way to a competitive labor market endogenously generating wage behavior consistent with a low-inflation regime. In Bulgaria and Czechoslo- vakia, by contrast, more emphasis has been given to creating a tripartite struc- Commander 1 1 ture of wage bargaining, which encompasses trade unions, managers, and gov- ernment. In Czechoslovakia the outcome of the 1991 bargain was to agree upon a level of real wage reduction broadly consistent with the macroeconomic tar- gets while also restricting the range of uncertainty over real wage movements given major price shocks. The latter will be an important feature when informa- tion regarding the size and timing of shocks is incomplete, which will guard in part against high nominal wage claims fueled by anticipation of future price rises. Tlhe basic model resembles more the corporatist structure of the Nordic countries and Austria. As elsewhere, the ability to hold to agreed norms would depend on the ability to sustain a basic consensus among the partners, the willingness of individual unions to relinquish bargaining powers to a central union body, and the extent of unanticipated price shocks. III. CONCLUDING REMARKS Controlling inflation remains at the heart of the reforms in the erstwhile socialist economies. The problem is more than usually difficult given the need for price liberalization and a new relative price structure. The microeconomic features of the economies and the legacy of the past-as reflected in ownership arrangements, wage setting, organization of the financial system, and allocative mechanisms-risk nullifying or distorting the effects of conventional macro- economic policies. Credit policy appears to be an important case in hand. At the same time conventional associations, such as a Phillips curve, remain largely absent. In a context of uncertainty over ownership and employment status, unemployment cannot necessarily be expected to exercise restraint on wage demands; instead it may be associated with accelerated wage claims, subject only to the restraint placed by incomes policy. Further, rapid growth in unemployment and associated output losses have led to correspondingly sharp rises in claims on the budget by firms and the unem- ployed, thus raising doubts about commitment to fiscal correction. Supply shocks have been significant in all reforming economies and have been associ- ated with a decline in labor productivity. Sharp adjustments to relative prices, as through a major increase in input prices, and higher nominal interest rates could lower output and promote a decline in labor productivity. The scale of decline would depend on the willingness of enterprises to shed labor, this willingness itself being a function of the degree of competitive pressure faced by the enter- prises. Large undervaluation of the exchange rate at the outset of the program would provide an initial cushion and impart some measure of price rigidity, particularly in sectors-such as capital goods-in which competition was restricted. As the experience of Poland and other countries testifies, prices cannot instan- taneously jump to equilibrium values. Even with trade opening, domestic prices have appeared to adjust rather slowly to international prices, while staggered price adjustments of sensitive goods-such as housing and transportation- require further periodic shifts in the price level. The combination of lags in 12 THiE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I relative price adjustments and powerful oil price and subsequent CMEA shocks appear to be important factors in accounting for the persistence of inflation. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Blanchard, 0. J. 1991. "Notes on the Speed of Transition, Unemployment, and Growth in Poland.' Massachusetts Institute of Technology, Department of Economics, Cam- bridge, Mass. Processed. Blanchard, 0. J., and Richard Layard. 1991. "Post-Stabilization Inflation in Poland." Massachusetts Institute of Technology, Department of Economics, Cambridge, Mass. Processed. Calvo, Guillermo A., and J. A. Frenkel. 1991. "From Centrally Planned to Market Economy: The Road from CPE to PCPE." IMF Staff Papers 38 (2, June): 268-99. Commander, Simon, and Fabrizio Coricelli. 1990. "The Macroeconomics of Price Re- form in Socialist Countries: A Dynamic Framework." PRE Working Paper 555. World Bank, Economic Development Institute, Washington, D.C. Processed. Commander, Simon, Fabrizio Coricelli, and Karsten Staehr. 1991. "Wages and Unem- ployment in the Transition to a Market Economy." In Georg Winckler, ed., Central and Eastern Europe: Roads to Growth. Washington, D.C.: International Monetary Fund. Commander, Simon, and Karsten Staehr. 1991. "The Determination of Wages in Social- ist Economies: Some Microfoundations." PRE Working Paper 713. World Bank, Eco- nomic Development Institute, Washington, D.C. Processed. Coricelli, Fabrizio, and Roberto Rocha. 1991. "Stabilization Programs in Eastern Europe: A Comparative Analysis of the Polish and Yugoslav Programs of 1990." PRE Working Paper 732. World Bank, Country Economics Department, Washington, D.C. Processed. Hmcir, Miroslav, and Jan Klacek. 1991. "Stabilisation Policies and Currency Convert- ibility in Czechoslovakia." European Economy Special Edition No. 2:17-39. Lipton, David, and Jeffrey Sachs. 1990 "Creating a Market Economy in Eastern Europe: The Case of Poland." Brookings Papers on Economic Activity 1: 75-147. Mates, Nevin. 1990. "Inflation in Yugoslavia: Specific Forms of Public Deficit Caused by Parafiscal Operations of the Central Bank." Economics Institute, Zagreb. Processed. Nuti, D. M. 1986. "Hidden and Repressed Inflation in Soviet-Type Economies: Defini- tions, Measurements and Stabilization." Contributions to Political Economy 5: 37-82. Rocha, Roberto. 1991. "Inflation and Stabilization in Socialist Countries: Some Lessons from the Yugoslav Experience." In Simon Commander, ed., Managing Inflation in Socialist Economies in Transition. EDI Seminar Series. Washington, D.C.: World Bank. Schaffer, Mark. 1991. "A Note on the Polish State-Owned Enterprise Sector in 1990." Centre for Economic Performance, London School of Economics. Processed. Tardos, Martin. 1990. "The Present State of the Transforming Hungarian Economy." Paper prepared for Vienna Institute for Comparative Economic Studies Workshop on East-West European Economic Integration. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 13-31 Lessons from Experiences with High Inflation Rudiger Dornbusch In economies where price control has been the rule, the most serious concern may be recognition of the inflation problem. Beyond the initial correction of subsidies there is the broader issue of the risk of a serious inflation. This article looks at the problem of high inflation in developing countries in Europe and Latin America and draws lessons from historical experience. It analyzes the dynamics of the interaction among deficit finance, institutional innovation in financial markets, dollarization, and the shortening of wage contracts in high-inflation situations. When stabilization is undertaken, there is neither immediate, spontaneous resumption of longer adjustment periods for wages and prices nor instant increase of real money demand to noninflationary levels. Incomes policy-freezing exchange rates, wages, and prices-is advocated as an effective supple- ment to the inevitable budget cut to make up for institutional inertia and facilitate the start of the stabilization process. In early 1991 Argentina, Brazil, Peru, Poland, and Yugoslavia were in the midst of extreme instability or at best in the early stages of stabilization. Another group of countries, including the Soviet Union, Romania, and Bulgaria, were on the verge of slipping into high or even extreme inflation. A third group had already run the course and stabilized, as did Bolivia and Israel, or had avoided the extreme experience and opted for stabilization early and decidedly, as Mex- ico did. The evidence from some 20 experiences with high inflation establishes that the similarities between the experiences of various countries become sharper and clearer and the differences less significant as the inflation rate rises. The particu- lar mechanism by which monetary expansion occurs may differ-say, deficits of state enterprises rather than of a particular ministry-but the general pattern that runs from deficits to an expansion of money and credit is broadly the same, as are tlhe dynamics of inflation. Of course the experience in the post-communist economies is special in that it starts from repressed inflation, but even that is not very different from experiences in Argentina or Brazil, where cycles of price controls and hyperinflation are now common (Cardoso 1991; Dornbusch, Sturzenegger, and Wolf 1990). The experience of Poland and Yugoslavia, and Rudiger Dornbusch is Ford International Professor of Economics at the Massachusetts Institute of Technology. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 13 14 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 the extraordinary problems already apparent in the Soviet Union, Bulgaria, and Romania, suggest that discussion of high inflation is timely (Commander and Coricelli 1992; Coricelli and Rocha 1990). I. LESSONS FROM HISTORY Should hyperinflation make policymakers opt for zero inflation at any price? Or is there room in between, with cost-benefit analysis and with the lessons from a rich inflation experience across time and space? Three questions are of special interest in the context of inflationary instability and stabilization. First, how does a country fall into hyperinflation? Second, what is necessary to stop high inflation and return to normal growth? And, third, will Eastern Europe and the Soviet Union soon resemble Latin America, with some success stories, a few countries on the verge of high or extreme inflation, and some countries experiencing hyperinflation? The statistics are, of course, open to question, but the size of the Soviet Union's budget deficits in the past few years indicate the problem ahead (table 1). The data show persistent and increasing deficits. Inflation is still negligible in official markets, but the stage is set for a dramatic inflation unless both the overhang and the deficit are addressed at the outset of any attempt to restructure the economy. The answers to these questions can to a large extent be discerned from histori- cal experiences of high inflation (Bruno and others 1988; Yeager 1981; Dorn- busch and Fischer 1986; Dornbusch, Sturzenegger, and Wolf 1990; Dornbusch and Wolf 1990). Some of the lessons from the past are described below. The Similarity of Inflationary Experiences across Countries It is a mistake to believe that the problems of a particular country are unique. But it is common that policymakers in Brazil, Peru, or the Soviet Union cannot accept that the experience of their country is not unique in the essential facts of inflation. Yet their situation is not substantially different from that of the 20 Table 1. The Soviet Budget Deficit, 1985-90 1985 1986 1987 1988 1989 1990' Percentage of GNP Revenues 47.3 45.8 43.6 41.7 41.0 42.8 Spending 49.7 52.0 52.0 51.0 49.5 50.6 Budget deficit 2.4 6.2 8.4 9.2 8.5 7.9 Percentage per year Retail inflation - 2.0 1.3 0.6 2.0 4.8 Money incomes - 3.6 3.9 9.2 13.1 14.5 -Not available. a. Plan and estimates. Source: IMF, World Bank, OECD, and EBRD (1990). Dornbusch 15 other countries where policymakers also thought that their unique situation set ordinary economics aside (Dornbusch and Edwards 1991). The Danger of Complacency Hyperinflation is not around the corner whenever there is a budget deficit. But inflation can easily become a habit-and from there an unstable process. Com- placency comes at a disastrous price: society falls apart as the middle class disappears and society is divided between those who know how to get ahead with inflation and those that fall behind (see Guttman and Meehan 1976 and Fergusson 1975 for dramatic descriptions of the German experience). Pauperi- zation of the middle class rapidly corrodes social institutions. Public administra- tion, the tax system, and all social relations become undermined by corruption and fraud. The middle class revolts against the state, and the poor revolt against property. Participatory Democracy and Instability Political change toward a more participatory democracy has not been the traditional vehicle for stability. Political change may carry with it the expecta- tion of an improvement in opportunities and living standards. An inflationary response to raised expectations occurred in Europe in the 1920s, in the Soviet Union in 1919-21, during the Allende period in Chile, and with the growth of Solidarity in Poland. Destructive inflation may accomplish many things that had been considered politically impossible. The destruction that takes place calls for particularly stern measures to rebuild confidence and stability. Because democratic institu- tions do not facilitate hard choices, democratic countries have almost invariably implemented special procedures to adopt and implement the hard measures necessary for stabilization. The arrangements differ from national unity govern- ments (Israel in 1985) to restricted special powers for the executive (Poincare in France in 1926) or special parliamentary committees charged to interact expe- ditiously with the executive (Germany in 1923). In the end the politically impos- sible gets done because the destruction brought about by uncontrolled inflation is so devastating that it forces cooperation. Fiscal Austerity Stabilization does not imply that zero inflation must be achieved at any cost. Some policies incorporate a moderate rate of inflation because there is a steeply rising cost of disinflation. Policies that do not set a limit on inflation or policies that merely repress it for a while do not encourage confidence and stability. Policymakers need to have a good grasp of the role of control of the budget and incomes policy in stabilization. Without fiscal austerity stabilization cannot last; without incomes policy it cannot start. 16 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 2. Recent European Experience with High Inflation, 1986-90 (percentage per year) Year Hungary Turkey Poland Yugoslavia Soviet Union 1986 5.2 34.6 17.7 89.8 2.0 1987 8.7 38.8 25.2 120.8 1.3 1988 15.6 75.4 60.0 194.1 0.6 1989 16.9 69.6 251.1 1,239.9 2.0 1990 28.3 60.3 585.8 583.1 4.8 Source: IMF, International Financial Statistics (various issues), and IMF, World Bank, OECD, and EBRD (1990). Structural Reform and External Support A separate set of issues concerns the transition from stabilization to growth. What kind of policies at home and abroad can help to decrease the risk of protracted stagnation? Structural reform and external support play a role in reassuring potential investors and thus moving the economy to growth (Dorn- busch 1991). II. VARIETIES OF INFLATIONARY EXPERIENCE Table 2 shows inflation rates for several European countries in the range from moderate to acute and extreme inflation. Table 3 shows average annual inflation rates in industrial and developing countries. High inflation is a problem of the developing countries of Europe and Latin America; it is not endemic to Africa, Asia, or industrial countries. Extreme inflation (hyperinflation) is rare. The generally accepted operational definition of hyperinflation proposed by Cagan (1956) sets the benchmark at an inflation rate of 50 percent a month (12,875 percent at an annual rate): The term hyperinflation must be properly defined. I define hyperinflation as beginning in the month the rise in process exceeds 50 percent and as ending in the month before the monthly rise in prices drops below that amount and stays below for at least a year. The definition does not rule out Table 3. Inflation around the World, 1970-89 (percentage per year) Developing countries Latin America Industrial Middle and the Year countries Africa Asia Europe East Caribbean 1970 5.6 5.4 6.5 - 3.0 12.4 1979 9.2 15.4 7.5 20.9 10.2 50.2 1988-89 4.7 23.3 11.8 126.0 17.0 208.0 - Not available. Source: IMF, International Financial Statistics (various issues). Dornbusch 1 7 a rise in prices at a rate below 50 percent per month for the intervening months, and many of these months have rates below that figure. Until recently there were few cases of hyperinflation in modern history. But now, with fresh cases emerging in Latin America and possibly in Eastern Eu- rope, the phenomenon is more pervasive. The distinction between hyperinflation and cases of lower and yet extreme inflation is somewhat arbitrary. Whether the inflation rate is really 50 percent a month or only 20 does not make too much difference because in either case inflation will be the dominant factor in the economy and will overshadow most other issues. Countries experiencing inflation rates of 10 or 15 percent a month for any length of time are moving toward hyperinflation. Table 4 shows the movement into high inflation in Argentina, Bolivia, Brazil, Mexico, Peru,-and Israel in the 1980s. Unlike the inflation experiences associated with war disloca- tion or civil war, high inflation in these six countries is rooted in domestic mismanagement and, to some extent, in external shocks. In each case the transi- tion to extreme high inflation started off with quite moderate rates, then sud- denly gathered speed and became extreme. Only Mexico cut the process short and stabilized before all the mechanisms of instability could gather force. Stabilization cannot afford to be weak. "Soft measures do not create hard currencies" the German authorities said in 1948 when a drastic monetary reform had to be administered (Dornbusch and Wolf 1990). The lack of thorough reform in Argentina, Brazil, and Peru shows up in the continuation for more than five years of off-and-on-again extreme inflation, which of course is accom- panied by a dramatic decline in economic activity and the standard of living. III. THE SOURCES AND DYNAMICS OF HIGH INFLATION This section discusses the interaction of financing requirements and the finan- cial structure by assuming full wage-price flexibility. It focuses on the role of contracts in the inflation process and the dynamics of the interaction among Table 4. Recent Experiences with High Inflation, 1981-89 (percentage per year, December to December) Year Argentina Bolivia Brazil Mexico Peru Israel 1981 105 29 106 28 75 117 1982 165 133 98 59 64 120 1983 344 269 142 102 111 146 1984 627 1,281 197 66 110 374 1985 672 11,748 227 58 163 305 1986 91 276 145 86 78 48 1987 132 15 230 132 86 20 1988 343 16 682 114 10,205 16 1989 3,079 15 1,287 20 3,390 20 1990 2,314 17 2,938 27 7,482 16 Source: IMF, International Financial Statistics (various issues). 18 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 deficit finance, institutional innovation in financial markets, dollarization, and shortening contracts. Explosive inflation arises from the disintegration or melt- ing of several institutions. A framework of the main determinants of inflation highlights the roles of budget finance, tax and financial institutions, and con- tracts in creating high inflation. The following analysis not only identifies the determinants of inflation but explains the mechanics of the very sharp accelera- tion that has been witnessed on several occasions. Deficit Finance There is considerable controversy in high-inflation countries about the exact, or even the approximate, size of budget deficits. Reliable public data, covering an extended period of time in a comparable fashion, are simply unavailable. Various series differ in their coverage of the public sector, in the distinction between budget and cash bases, and in the inclusion of certain expenditure items, especially with respect to the quasi-fiscal deficit of the central bank. Both the adjustment of velocity and the presence of alternative means of financing the deficit (foreign borrowing, use of reserves, and domestic debt finance) help explain the lack of a tight link between inflation and the deficit. Controversies arise about the reason for the budget deficits, their endogeneity as a result of inflation, and their amplification by financial adaptation. That the actual outburst of inflation is often triggered by a foreign exchange crisis does not alter the fact that high inflation is a fiscal phenomenon. The most common view asserts that high inflation is the result of budget deficits. If the government spends more than it receives in tax collection, the remainder is financed by creating money. That means more money-too much money-chasing too few goods with the predictable outcome of inflation. This view needs considerable refinement to be entirely correct. Three directions of correction are essential. First, there is some room for noninflationary deficit finance. Second, deficits can be financed by debt. Third, there is a channel of causation that runs from inflation to deficits, as well as the other way around. A model of these important qualifications is given by equation 1. The deficit can be financed with high-powered money, with domestic debt, or with foreign debt: (1) gY = M/P + B/P +B e/P where g is the deficit ratio, Y is real gross domestic product (GDP), M is the domestic base money, B and B* are domestic and foreign debt, e is the exchange rate (domestic currency over dollars), and P is the price level. (A dot over a variable denotes the rate of change.) It is clear that deficits can be financed by borrowing from abroad or at home, thus entirely avoiding an increase in the money stock, at least for the time being. Focusing on the situation in which the entire deficit is financed by money will show how much inflation can be gener- ated by such a system. Dornbusch 1 9 Inflationary finance. Financing deficits by money creation means that any money that is not demanded at the current level of prices must be forced on the public by inflation. In a growing economy some extra real money balances are demanded in order to finance the growing level of transactions. But, beyond that, the demand for nominal money expands only to the extent that inflation erodes the purchasing power of existing real balances. To restore their real balances (at least partially), the public has to add to nominal money holdings. Thus inflationary finance automatically creates a demand for the money issue that finances the deficit. Keynes (1923, p. 37), in his splendid description of the inflation tax, noted the scope for inflationary finance even in a country with the poorest economic and political conditions: A government can live for a long time, even the German government or the Russian government, by printing paper money. That is to say, it can by this means secure the command over real resources, resources just as real as those obtained by taxation. The method is condemned, but its efficacy, up to a point, must be admitted. . . . so long as the public use money at all, the government can continue to raise resources by inflation . . . a govern- ment can get resources by a continuous practice of inflation, even when this is foreseen by the public generally, unless the sums they seek to raise in this way are very grossly excessive.... What is raised by printing notes is just as much taken from the public as is a beer duty or an income tax. What a government spends the public pays for. There is no such thing as an uncovered deficit. But, as Keynes has noted, significant inflation reduces the amount of money people choose to hold, because they will substitute toward assets that are more inflation proof. Thus, just as high taxation erodes the tax base, high inflation leads to a reduction in real balances and hence to an increase in the rate of inflation necessary to finance a given deficit. Moreover there may be a maxi- mum amount of resources the government can extract. The long-run relation between the money-financed budget deficit and the rate of inflation is shown in equation 2 (for a derivation, see Dornbusch 1985): (2) ir = (Eg - y)/(1 - fg), 1 > fg where ir and y are the rate of inflation and the growth rate of real GDP. The term a represents the noninflationary level of velocity, and a is the responsiveness of velocity to the rate of inflation. This equation shows that because the deficit is financed by money creation, there is inflation. But it also shows that the infla- tionary impact of a given deficit can differ widely, depending on the financial structure and the growth rate of output. There are three key points of this relation. First, the inflation rate is lower the higher the growth rate of output is. When output grows strongly, so does the 20 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I demand for real money. Accordingly there is room for some extra money to be issued without introducing the risk of inflation. Second, inflation is higher the larger the budget deficit is. Moreover this relation is nonlinear. As the govern- ment tries to finance a larger deficit, the required rate of inflation increases steeply. Depending on the particular form of the money demand equation, there may even be a maximum deficit that can be financed by money. Going beyond that range implies hyperinflation. Third, the inflation rate depends on the veloc- ity parameters in equation 2. The higher is the level of noninflationary velocity (that is, because of dollarization) the higher the rate of inflation associated with any given deficit. A high degree of responsiveness of velocity to inflation also implies a larger rate of inflation. The increase in inflation brought about by a one percentage point increase in the deficit is higher, the higher are inflation and the budget deficit from which one starts. Inflationary finance thus exerts a very powerful impact on inflation if it is used in large doses or in an environment where a high level of velocity, and strong responsiveness of velocity to inflation, leaves little scope for an inflation tax. Likewise, dollarization or a drop in growth bring about large increases in the inflation rate, more so the higher the initial extent of deficit finance. The Olivera-Tanzi effect. One of the striking effects of inflation is the erosion of the real value of taxation. If there is any delay between accrual and payment of taxes, the inflation in the interim will mean that the real value of what is paid is lower the higher the rate of inflation. With moderate inflation it makes no difference that 1987 taxes are paid in 1988. But when inflation is high, this effect wreaks havoc with the real value of tax collection. Keynes, commenting on the impact of inflation on the budget noted this point, as did Bresciani- Turroni (1937). Tanzi (1978) and others have recognized this effect in the spe- cific context of Latin American inflation. The empirical importance of this effect is large whenever inflation is high and tax collection lags are long and when there is no provision for tax indexation. External shocks and inflation. Suppose, as is the case in Argentina, that the public sector has a large external debt and an external debt shock occurs. Specifically, assume that before the disturbance any existing external debt was rolled over with interest fully capitalized through automatic "new money" and that there is no domestic debt. Let d* be the flow of external debt service (measured as a percent of gross national product [GNP]), and thus g is the total deficit ratio that is financed by money creation. Thus, (3) g = a(T) + d* From equation 3 reduced access to automatic capitalization of interest pay- ments implies that external debt service leads to increased deficit finance by money creation. The country has to earn the resources for external finance or else finance the purchase of foreign exchange by creating money. First, the Dornbusch 21 government will issue more money to finance the purchase of foreign exchange for interest payments (assuming, of course, that there are no expenditure cuts or tax increases). Second, there will typically be a real depreciation in order to improve the external balance. The increase in inflation resulting from an external financing disruption is larger, the larger is the debt service shock and the real depreciation, but it also depends on the responsiveness of velocity to inflation and on the degree to which increased inflation erodes real tax collection. Each of these factors will increase the inflationary impact of the debt shock significantly. The "balance of payments school" would argue that external balance prob- lems and the resulting depreciation of the exchange rate are the primary causes of the deficit. By contrast, the "quantity theory school" would point to budget deficits and their financing by money creation as the reason for inflation. Passive money is the essential ingredient in reconciling the quantity school and the balance of payments doctrine. Not surprisingly, suspension of reparation pay- ments in Germany and of debt service in Bolivia in 1985 were essential steps in the stabilization of inflation. In Argentina involuntary external debt service after 1982 became an important source of inflation in exactly the manner the balance of payments school emphasizes. Deteriorating terms of trade further aggravated the external debt shock by forcing real depreciation and hence an increase in the real value (in terms of GDP or the tax base) of the existing external debt service. Endogenous financial innovation and liberalization. The interest that tradi- tional depository institutions can pay is typically controlled. There may be an outright limitation on interest rates, or else institutions may be required to hold reserves or government debt at controlled rates. These restrictions make institu- tions unable to compete in financial markets where nominal interest rates more nearly reflect the ongoing inflation. New, unregulated financial institutions that offer depositors higher interest rates spring up and thus draw customers away from traditional depository institutions. There is a fall in the ratio of conven- tional money to GDP. The government loses part of its inflation tax base, and hence equilibrium inflation increases. The government may aggravate matters when it responds to the increasing inflation by raising reserve requirements or forcing traditional banks to hold government debt. Governments often promote this process, most obviously under the guise of financial liberalization. Since inflation is a tax on money (or commercial bank non-interest-bearing reserves), financial liberalization means that the public can avoid the tax on money. Financial liberalization may take the form of interest- bearing deposits or formal dollarization, each of which reduces the demand for high-powered money; velocity rises and so does the inflation rate associated with the financing of a given deficit by money creation. Thus, from an inflation point of view, financial repression, not liberalization, is appropriate. Financial liberalization requires that extra tax revenue be available to avoid the inflation- ary impact of a reduction in the captive inflation tax base. Governments that 22 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I condone dollarization likewise promote inflation. Dollarization is captured in equation 2 by both the coefficients a and A. The shift from the domestic mone- tary base into dollars reduces the base for the inflation tax and hence must increase inflation. One is tempted to explain inflation experiences in some countries by dollariza- tion and new financial intermediaries. Thus countries with stronger dollariza- tion have higher inflation. A government that experiences some inflation and makes dollarization easier will experience even more inflation. However, dollar- ization is also a response to inflation. The financial adaptation to inflation intensifies the inflationary process. In response to inflation there is a flight from money into interest-bearing financial assets, to the extent that they exist at all, or into dollars. But there is also an institutional adaptation: financial institutions spring up that offer protection against inflation. The better the protection they offer, the more substantial the flight from money or the larger the increase in velocity. Timing. Inflation controls and managed exchange rates can slow down the buildup of inflation. The loss of resources, or forced saving, is an alternative mode of financing. However, deficits do imply money creation and inflation. Moreover the longer the delay the more dramatic the inflationary explosion. This is especially the case when a managed exchange rate and reserve losses have financed the deficit in a relatively noninflationary manner. When these mecha- nisms are no longer possible, there will be a sudden shift toward the inflation tax at the same time that a real depreciation is required. The Role of Contracts in the Inflation Process As inflation accelerates, contracts shorten, and that shortening of contracts is itself a factor that causes inflation to accelerate. Institutional wage-setting mech- anisms often rely on a fixed contract length, with wage adjustments occurring at specified intervals. The adjustments are based on the cumulated increase in prices since the last adjustment. For example, earners might receive full compen- sation for past actual price increases at regular intervals, say yearly. Now sup- pose there is a shift to six-month intervals. There are two interesting questions. The first concerns the dynamics of shifting to shorter contracts. What is the threshold for inflationary erosion of wages that causes the shift, and what makes it economywide rather than just for a particular firm? The other interesting question is what happens when the frequency of adjustment increases yet fur- ther. This point has been developed especially by Pazos (1972). It is of interest here because contract deterioration is one of the important characteristics of an accelerating inflation and because exchange depreciation often plays an impor- tant role in setting off the process. If nominal wages are adjusted only periodically, the real wage follows a saw- tooth pattern. On each adjustment date the nominal wage is increased by the cumulated inflation since the preceding adjustment. Until the next adjustment Dornbusch 23 date the real wage declines as the ongoing inflation erodes the purchasing power of the constant nominal payments. By the end of the adjustment interval the real wage has declined below its period average. The higher the rate of inflation, moreover, the lower the average real wage, given the interval of adjustment. In a system of full, but lagged, indexation, the real wage can be cut only by moving to a higher rate of inflation. Thus, once-and-for-all depreciation of the currency immediately raises the rate of inflation and erodes existing contracts. But wage indexation ensures that inflation must be pushed to an even higher rate so that there is always some group of wage earners whose wages are still lagging the increasing rates of price increases. The same principle applies to the removal of subsidies undertaken to correct the budget. Measures undertaken to correct competitiveness or the budget can be effective only if they achieve a cut in the real wage, but because of full indexation that cut can take place only if inflation is allowed to run at a higher rate. This mechanism often sets the stage for inflation explosions. Consider a country that requires adjustments in the budget and external com- petitiverness. Suppose that the government lacks the political force to suspend full indexation, so that the removal of subsidies or a real exchange depreciation will speed up the inflation rate. Workers in the middle of their contracts, for example, will find that their real wages fall below what they consider a mini- mum standard of living. They cannot borrow, even in perfect capital markets. Hence they will call for a shorter interval between wage adjustments in order to recover the real wage losses imposed by inflation. They will ask for an advance of what they think is due. If the economy does, in fact, shift from, say, six- month to three-month indexation intervals, the inflation rate will simply double (Simonsen 1986). But once the contract structure has moved to a three-month scheme, it is unlikely that the indexation structure will return spontaneously to a longer interval, even if shocks are favorable. And there is nothing to make the three-month interval more stable than the six-month interval that was just aban- doned. New shocks will shift the economy to even more frequent adjustments and hence to correspondingly higher rates of inflation. At this stage the ex- change rate becomes critical. The dramatic escalation of inflation, seemingly out of proportion to the dis- turbances, arises from the endogeneity of the adjustment interval. This is due not so much to the direct impact on inflation of corrective exchange rate or price policies. It occurs because increases in inflation, which may be minor but highly visible (such as a 10 percent devaluation over and above a purchasing power parity rule or a removal of bread subsidies), lead to an increase in the frequency of wage adjustments, which brings on a much higher inflation rate. The endo- geneity of adjustment intervals is the mechanism that connects small inflation disturbances with a shift from 50 to 100 percent inflation or beyond to hyper- inflation. As long as full indexation remains, even seemingly small corrections are a dramatic threat to the stability of the inflation rate and hence may not be worth undertaking. Incomes policy designed to avoid inflationary explosion must avoid accelerating the frequency of adjustments. 24 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 1 Dynamics The actual dynamics of the economy emerge from the interaction of the inflationary aspects of deficit finance and the contracting process. A stable equilibrium may not actually exist. When inflation rises significantly and perma- nently, institutions adapt. In doing so, they help to increase inflation. Under conditions of extreme inflation, institutions break down. There is a near- abandonment of domestic money, which means the government must continue to increase inflation to get any seigniorage. Contracts are set for a shorter duration and are more likely to be dollar-based. In the analysis of inflationary experiences, it is common to assume adaptive inflationary expectations (Cagan 1956). There appears to be a significant slug- gishness in the initial phases of high inflation as well as a subsequent accelera- tion, which suggests exactly such an expectations mechanism. Adaptive infla- tionary expectations are often the key model device to slowing the impact of money on inflation. An alternative and perhaps more accurate model focuses much more on the dynamics of deterioration in contracts, both in the goods and labor markets, and on the inflationary adaptation of financial institutions. Insti- tutional dynamics seem to offer a more suitable framework for studying high inflation. As economic institutions break down and time intervals for contracts and adaptation to inflation become shorter, the inflation process becomes explosive. The economic time horizon shrinks along with contracts and maturities of finan- cial assets until, when the economy converges to a spot market with dollar pricing, the budget or external balance deficit leads to hyperinflation. Hyper- inflation is inevitable because the inflation tax, with sufficient financial adapta- tion, can be almost totally evaded, and hence the budget deficit cannot be financed. The Olivera-Tanzi effect, the shortening of contracts, and financial adaptation all react in a perverse way (from the perspective of stabilization) in that they widen the deficit and accelerate explosively the inflation process. IV. STABILIZATION The preceding discussion helps to explain why stabilization is difficult and often takes more than one attempt to succeed. In the process of high inflation all institutions break down. When stabilization is undertaken, there is neither im- mediate, spontaneous resumption of longer adjustment periods for wages and prices nor an instant increase of real money demand to noninflationary levels. As a result more sizable adjustments in the budget are required, and more dramatic measures are necessary to create the confidence that stabilization will, in fact, last. Because the fiscal measures have to be particularly large they are also particularly difficult and hence often can not be sustained. When they fail, inflation returns instantly at exceptionally high levels because institutional iner- tia had not recovered. Dornbusch 25 Incomes policy-freezing exchange rates, wages, and prices-can be an effec- tive supplement to the inevitable budget cut. It makes up for institutional inertia and, to that extent, gives a government a better chance to start stabilization. But, as is clear from the experiences of Argentina, Brazil, and Peru, failure to correct the budget implies that high inflation will soon return. The decline in the ratio of MI1 to GDP is not typically fully reversed in the initial stabilization. As a result financing even a moderate deficit is much more inflationary than it was before the experience of extremely high inflation. This hysteresis effect of high inflation (similarly apparent in contracts, pricing, and tax collection) sharply reduces the chances of stopping inflation with anything short of a dramatic budget cut. The task of stabilizing inflation involves stopping inflation quickly and avoid- ing the resurgence of inflationary pressures. To end inflation by incomes policy is relatively easy, but to keep it down requires fiscal support. The chief mistake in stabilization policy is to rely too much on incomes policy-fixed exchange rates and wage and price freezes-and too little on fiscal austerity. Such programs quickly lead to repressed inflation and overvaluation, in which tight monetary policy is introduced to sustain the imbalances. Ultimately that does not work, and another inflationary explosion offers the starting point for yet another stabilization. Argentina offers a clear example of this process with its successive failed stabilization programs in the past five years. Budget Balancing Budget: deficits are the ultimate source of inflation. When external financing or the domestic capital market cannot finance deficits, then the deficits must be adjusted. Two questions immediately emerge. The first is how large a deficit is consistent with stability; the second is how to cut deficits down to the required size. Argentina, Brazil, and Peru failed to adjust fiscal deficits in the aftermath of their 1985 heterodox stabilization. Wage-price controls and fixed exchange rates quickly stopped inflation and raised the political popularity of the presi- dent. The resulting possibility for fiscal stabilization was, however, not used. Instead the deficits persisted and were financed by creating money. As a result, inflation continues. Quasi-fiscal deficits. The starting point for budget balancing is the need for a transparent accounting of the consolidated government. Because the issue is control of monetary emission, it is essential that the central bank's "quasi-fiscal" deficit be part of the accounting. An accounting framework is needed for the consolidated government sector, including not only the central government and the central bank, but also state enterprises and local government. Extreme inflation invariably reflects deficits financed by writing checks on the central bank, whether it be by provincial authorities, as in China; by state enterprises, as in Yugoslavia; or by a government bank, as in Brazil. The deficits may have as 26 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 a counterpart purchases of foreign exchange, payments of wages, deficits of the railroads, external debt service, election spending by a governor, or simply corruption. In any case a deficit leads to money creation. Quasi-fiscal deficits arise from loans by the central bank at subsidized rates, losses on foreign exchange operations in the form of guarantees, forward con- tracts, or simply purchases at a high rate (under multiple exchange rates) and sales at a low one. In Peru in 1986-87, for example, exchange losses accounted for 2.3 percent of GDP. But central bank losses also arise from credit operations. Subsidized credit is no different from any other subsidy; in fact credit subsidies have long ceased being investment subsidies and have become simply a produc- tion subsidy that finances wages when prices are not allowed to reflect costs. Revenues. The second point on the reform agenda is to achieve a productive tax system. The reform must raise revenue on a substantially larger scale and far more efficiently. Increasing the yield of the tax system is dictated by the need to eliminate deficits. Inflation stabilization makes an immediate contribution be- cause the inflationary erosion of revenues ceases. But that is only a small part, perhaps as much as 2 percent of GNP in revenues. The major effort must be in reconstructing the tax system, including stopping the corruption and evasion that now undermine the collection of taxes as well as introducing and demon- strating mechanisms to increase compliance. The complacent acceptance of per- vasive tax evasion is the most regressive aspect of the Latin American tax sys- tem, and it must be carefully watched as Eastern Europe moves to taxation as the chief source of government revenue. The revenue effort must concentrate both on collecting taxes and on eliminat- ing subsidies in public sector enterprises. Many countries now have pervasive systems to manage public sector prices, both to control inflation and to try to prevent a decline in real wages. The implied revenue losses are extraordinarily large and cannot be justified by any of the objectives. For example, in Peru controlled telephone rates have reduced the real price of the service to one-tenth the 1985 level. It is difficult to argue that telephone rates have an important incidence either on inflation or on welfare of the poor, but they do contribute to deficits. Governments should therefore eliminate totally all subsidies. The resulting revenue gains must be applied to eliminate inflation, which in itself raises the welfare of the poor since inflation is a highly regressive tax. Part of the revenues should also be used for targeted food and employment programs for the poorest groups. Beyond cutting all subsidies and raising revenue under the existing structure, governments should use the crisis to institute a more efficient tax system. The system should produce more revenue with fewer distortions, which means elim- inating the pervasive exemptions from direct taxes and raising the rates to higher levels. A comprehensive value added tax of 15 percent, with a 5 percent sur- charge for luxuries, might be the starting point for discussion. Dornbusch 27 Government spending. For many observers the right direction for adjustment is to cut government spending, not to raise taxes. Inefficiency in government is pervasive, and public sector employment in many countries is unjustifiably high. But there is no presumption that the fiscal problem can be solved correctly by massive firing of public sector employees and privatization. There is a need to restructure public sector spending, from consumption to investment and pro- ductive services. But as for the level of spending, it certainly is not excessive. More of the spending absolutely and relatively should fall on infrastructure, health, and social services for the poorer groups. The current composition of spending is not only unproductive but probably also regressive. Most of infrastructure investment could be done by the private sector. That is certainly the case, for example, for telephone services, but also for public trans- port and even the road system. Mexico is now exploring such options with very substantial success. But the fact remains that infrastructure spending should not be the priority in balancing the budget, certainly not at the expense of health and education. Incomes Policy Fiscal austerity is the essential aspect of stabilization, but incomes policy is an important, desirable component. Incomes policy is designed to bring about a rapid, coordinated end to inflation. In a hyperinflation, incomes policy amounts to fixing the exchange rate. Because price setting is geared to the movements of the dollar, the move to a fixed exchange rate is enough to break the inflation and the expectation of inflation. But when annual inflation is only 100 or 200 percent, incomes policy is both more essential and more complicated. Without incomes policy ending inflation by demand management alone would create an extraordinary depression. The current inflation will be a weighted average of cost increases that are equal to past inflation, which enters costs by explicit or implicit indexation, and the current rate of exchange depreciation, e, plus a cyclical component, which is denoted by "GAP": (4) T = 7T _1 + (1 -T) e + +GAP. Because of the inertia represented by the cost increases resulting from explicit or implicit indexation, current inflation cannot get away from past inflation unless the government breaks the process by incomes policy. Incomes policy means fixing the exchange rate and stopping wage inflation. The government will have to intervene in loan contracts to reduce real interest burdens that otherwise would result from the unanticipated decline in inflation, and interven- tion will be required in wage contracts. Because these contracts have periodic adjustments for inflationary erosion, a sudden ending of inflation requires inter- vention. Some wage contracts have to be rolled back, and others need to be adjusted upward. 28 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Exchange rate policy. Exchange rate policy assumes a strategic role in stabi- lization, as does pricing in the public sector. The starting point of a program is invariably a fixed exchange rate. But if inflation does not end completely, sooner or later adjustments in the exchange rate and public sector prices are needed. The decision to abandon the fixed rate is a difficult one because it signals the government's acceptance of inflation as something inevitable. As a result there is a temptation to postpone exchange rate adjustment until a significant overvalua- tion has developed. Overvaluation in turn creates an expectation of a devaluation, and very high real interest rates become necessary to stop speculation. High real interest rates in turn increase domestic debt service and worsen the budget. Ultimately the exchange rate adjustment does have to come, but often the overvaluation has gone so far that an outright exchange crisis and collapse are the end of the abortive attempt to practice a fixed exchange rate. The pragmatic answer is to move after two or three months to a crawling peg, depreciating the exchange rate at a pace that maintains external competitiveness. The risk of an overvalua- tion maintains short economic horizons and stands in the way of recovery. The right time for a crawling peg is very early because the government should try to preserve maximum competitiveness. Holding onto an exchange rate too long may yield an extra month of low inflation, but it also sacrifices competitiveness and therefore prejudices the return of growth. Indexation. A major stabilization decision regards indexation. The common view is that indexation is responsible for the inflation and that accordingly it should be abolished. Moreover governments should declare a zero inflation target rather than create mechanisms that make it easier to live with inflation. However, it is not necessarily true that without indexation there is inflation stability because inflationary shocks, such as public sector price increases and depreciation, are not fully absorbed into lower real wages. Without explicit indexation the government becomes the judge of what wage increases to grant. The wage becomes politicized, which means invariably larger rather than smaller wage increases, and wage increases come sooner rather than later. In fact in economies in which a government seeks to avoid explicit indexation, as in Brazil after 1985, inflation soon becomes more unstable and susceptible to a far more rapid escalation than had ever been experienced under indexation. Indexation is a mechanism that creates inertia and also preserves inertia. Reintroducing half-yearly indexation may therefore be a key step in establishing the expectation of low inflation. Once the wage is locked away, a very rapid resumption of inflation will not be expected. As a result horizons can lengthen far more effectively than under threshold provisions or in the absence of any kind of formal indexation. Monetary policy. Monetary policy does not play an independent role in stabi- lization; it is dictated by the budget and the exchange rate policy. Following Dornbusch 29 stabilization real interest rates are too high. One could argue that to resume growth the economy needs reliquification. There is very limited room for reliq- uification, but that is best done by monetizing reserve inflows rather than by deficit finance or domestic credit creation. The alternative is an overly firm commitment to a zero inflation target. The policymaker might be tempted to make monetary policy (and the exchange rate) do what fiscal policy has not achieved. The risk is a long period of extraordi- narily high real interest rates and possibly an exchange overvaluation. They might stop inflation, but they also will destroy the real economy. The best and only lasting way to bring about low real interest rates and to achieve moderate inflation is by a balanced budget (including state enterprises, except where they are financed in the capital market) and a very competitive real exchange rate. V. CONCLUDING REMARKS: PRIORITIES Without financial stability, economic reconstruction and growth will simply not occur. If inflation is high and variable, then it will be the most important issue. It will take up policymakers' precious time. It will tempt them into superfi- cial remedies, which help in the short run but set back economic activity because they create uncertainty. It will also lead the private sector to focus on protection against inflation and government's arbitrary interventions. The first priority then must be to reduce inflation. A lastingly balanced budget is required to achieve financial stability. In economies in which price control has been the rule, the recognition of the problem of inflation may be the most serious issue. There has been no experi- ence with inflation, and most attention focuses on the popular revolt against removing some subsidies. But beyond the initial correction of subsidies there is the broader issue of the risk of a serious inflation. Serious inflation can emerge either because there is an initial monetary overhang or because the subsidy correction does not go far enough. Deficits remain, and money creation starts to interact with corrective inflation. (The Soviet Union is an obvious case in point.) In countries with major fiscal problems it is not politically impossible to make the necessary adjustments; it is politically difficult, but the adjustments will ultimately be made. The only question is how large the loss in the standard of living has to be before it is done and how much time and political capital will be lost (Blanchard and others, 1991; Fischer and Gelb 1990, and Dornbusch 1990). REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Bresciani-Turroni, C. 1937. The Economics of Inflation. London: Allen & Unwin. 30 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Blanchard, Oliver, Rudiger Dornbusch, Richard Layard, and Lawrence Summers. 1991. Economic Reform in the East. Cambridge, Mass.: MIT Press. Bruno, Michael, Guido Di Tella, Rudiger Dornbusch, and Stanley Fischer, eds. 1988. Stopping High Inflation. Cambridge, Mass.: MIT Press. Cagan, Phillip. 1956. "The Monetary Dynamics of Hyperinflation." In Milton Fried- man, ed., Studies in the Quantity Theory of Money. University of Chicago Press. Cardoso, Eliana. 1991. "From Inertia to Megainflation: Brazil's Macroeconomic Poli- cies in the 1980s.' In Michael Bruno and Stanley Fischer, eds., Lessons of Economic Stabilization and Its Aftermath. Cambridge, Mass.: MIT Press. Commander, Simon, and Fabrizio Coricelli. 1992. "Price-Wage Dynamics and Inflation in Socialist Economies: Empirical Models for Hungary and Poland.' The World Bank Economic Review, this issue. Coricelli, Fabrizio, and Roberto Rocha. 1990. "Stabilization Programs in Eastern Eu- rope: A Comparative Analysis of the Polish and Yugoslav Programs of 1990." World Bank, Washington, D.C. Processed. Dornbusch, Rudiger. 1985. "Stopping Hyperinflation: Lessons from the German Experi- ence in the 1920s." In Rudiger Dornbusch, Stanley Fischer, and John Bossons, eds., Macroeconomics and Finance: Essays in Honor of Franco Modigliani. Cambridge, Mass.: MIT Press. . 1990. "Priorities of Economic Reform in Eastern Europe and the Soviet Union.' Massachusetts Institute of Technology, Department of Economics, Cambridge, Mass. Processed. 1. 991. "Policies to Move from Stabilization to Growth." Proceedings of the World Bank Annual Conference on Development Economics 1990. Washington, D.C.: World Bank. Dornbusch, Rudiger, and Sebastian Edwards. 1990. "The Macroeconomics of Populism in Latin America." Trimestre Economico 57 (January-March): 121-62. , eds. 1991. The Macroeconomics of Populism in Latin America. University of Chicago Press. Dornbusch, Rudiger, and Stanley Fischer. 1986. "Stopping Hyperinflations: Past and Present'" Weltwirtschaftliches Archiv 122 (1, April): 1-14. Dornbusch, Rudiger, Federico Sturzenegger, and Holger Wolf. 1990. "Extreme Inflation: Dynamics and Stabilization." Brookings Papers on Economic Activity 2: 1-64. Dornbusch, Rudiger, and Holger Wolf. 1990. "Monetary Overhang and Reforms in the 1940s." Massachusetts Institute of Technology, Department of Economics, Cam- bridge, Mass. Processed. Fergusson, A. 1975. When Money Dies. London: William Kimber. Fischer, Stanley, and Alan Gelb. 1990. "Issues in Socialist Economy Reform." Massa- chusetts Institute of Technology, Department of Economics, Cambridge, Mass. Processed. Guttman, William, and P. Meehan. 1976. The Great Inflation. London: Gordon and Cremonesi. International Monetary Fund. Various issues. International Financial Statistics. Wash- ington, D.C. Dornbusch 31 International Monetary Fund, World Bank, Organisation for Economic Co-operation and Development, and European Bank for Reconstruction and Development. 1990. The Economy of the Soviet Union. Washington D.C. Keynes, John M. 1923. A Tract on Monetary Reform. Reprinted by the Royal Economic Society, London, 1971. Pazos, Felipe. 1972. Chronic Inflation in Latin America. New York: Praeger. Simonsen, Mario. 1986. "Indexation: Current Theory and the Brazilian Experience." In Rudiger Dornbusch and Mario Simonsen, eds., Inflation, Debt and Indexation. Cam- bridge, Mass.: MIT Press. Tanzi, Vito. 1978. "Inflation, Real Tax Revenue, and the Case for Inflationary Finance: Theory with an Application to Argentina." IMF Staff Papers 25 (September): 417-51. Yeager, Leland. 1981. Experiences with Stopping Inflation. Washington D.C.: American Enterprise Institute. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 33-53 Price-Wage Dynamics and Inflation in Socialist Economies: Eimpirical Models for Hungary and Poland Simon Commander and Fabrizio Coricelli This article analyzes the determinants of open inflation in transitional socialist econ- omies, with reference to recent experience in Hungary and Poland. A simple inflation model is centered on the transmission process and on the short-run dynamics of infla- tion. Further incorporating a number of features specific to socialist economies and working with quarterly data, dynamic price and wage equations are estimated. The estimated equations allow satisfactory exploration of the role and weight of foreign prices and domestic factors in propagating inflation. Foreign prices matter, but devel- opments on the cost side are critical in relating exogenous, policy-driven adjustments to the price level to increases in the rate of inflation. The absence of conventional market- based, equilibrating mechanisms requires that nominal anchors, particularly wage re- straints, feature prominently in any stabilization program adopted by reforming social- ist economies. An apparently unique feature of classical centrally planned economies has been the absence of inflation. However, it is generally accepted that inflation was repressed in centrally planned economies, with the result that open inflation has become a hallmark of economies making the transition to a market-based sys- tem. By it:self the translation of repressed inflation into open inflation cannot explain the acceleration-and the persistently higher rates-of inflation that has been observed in reforming socialist economies. Moreover the rapidity of the acceleration threatens to dilute the desired signaling role for prices and jeopard- izes the political sustainability of reform. Although an extensive literature exists that deals with the measurement of repressed inflation (Nuti 1986; Portes 1977; Feltenstein 1989), analysis of open inflation is still largely anecdotal (Honohan 1989 for China and Rocha 1990 for Yugoslavia are among the few exceptions). Little attempt has been made to pin down empirically the consequences of the particular behavioral rules and institu- tional routines, particularly cost-plus pricing, that are generally agreed to be characteristic of socialist economies in transition. In this article we try to isolate Simon Commander is with the Economic Development Institute, and Fabrizio Coricelli is with the Country Economics Department, both at the World Bank. i 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 33 34 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 key empirical regularities of the inflation process in reforming socialist countries. Several features of the socialist economies distinguish them from market- based economies and indicate areas for specific analytical treatment. Such dis- parities include the administrative determination of prices, the wage-setting framework and the imposition of full employment, external trading prices and regimes, and the monopolistic structure of goods markets. Further, such econ- omies might be expected to lack any endogenous mechanism of equilibration, given full employment and an absence of a conventional relation between output and prices. Several of these features are incorporated in an inflation model, calling on Hungarian and Polish experience and data. To highlight the short-run dynamics and propagation mechanism, price and wage models are estimated using quarterly data that cover much of the recent reform period. The analysis pays particular attention to the relation between prices and wages and to the role of the exchange rate in determining domestic inflation. Section I discusses the main institutional features and developments for prices and wages in Hungary and Poland. A simple structural price and wage model suitable for estimation is summarized in section II; the full model is laid out in the appendix. The reduced-form price and wage equations are estimated based on available quarterly data, with the results reported in section III. Section IV presents concluding remarks. I. INSTITUTIONAL FEATURES AND PRICES AND WAGES IN HUNGARY AND POLAND There are several institutional features related to prices and wages that are similar in Hungary and Poland. However, the inflationary experiences and the effect of wage determination on inflation have differed in the two countries. Main Features of Price Reform in Hungary and Poland Three main features can be extracted from Hungarian and Polish experience with price reform and inflation, especially during the 1980s. First, there has been a sustained upward shift in the price level. In the Hungarian case inflation has consistently ratcheted upward, exceeding 30 percent a year in 1990 with likely expansion to more than 50 percent in 1991. In the 1980s in Poland inflation remained consistently higher and exhibited far greater variance than in Hungary. This culminated in a high inflation burst in 1989, followed by a radical stabilization, which, by late 1990, brought inflation down to below 5 percent a month. Second, both economies have moved away from strictly administratively de- termined prices. A series of price liberalization measures has reduced adminis- tered prices as a percentage of total prices. Table 1 provides information on the distribution of different price categories. However, neither free prices in Hun- gary nor contract prices in Poland have been wholly free from administrative interference. Established ratios of costs to profits in the material sector and Commander and Coricelli 35 Table 1. Consumer and Producer Prices by Pricing Category in Hungary and Poland, 1982-90 Country and price category 1982 1984 1985 1987 1988 1989 1990 Hungary Consumer prices Fixed 15 15 6 6 0 0 n.a. Flexible 29 28 38 32 20 17 n.a. Free 56 57 56 62 80 83 n.a. Producer prices Fixed 33a 30 32 28 25 23 n.a. Free 67a 70 68 72 75 77 n.a. Poland Consumer pricesb Administrative 35 47 47 45 45 45 10 Regulated 15 3 3 2 0 0 0 Contract 50 50 50 53 55 55 90 n.a. Not available. Note: Data were not available for 1983 and 1986. Flexible consumer prices include those subject to maximum limit; fixed producer prices also include all prices with upper limits. a. Data for 1980. b. Expressed as share of total sales value for products covered by price category. Since 1990 changes in contract prices no longer require notification. Source: Data provided by the Ministries of Finance of Hungary and Poland. other guidelines have constrained price growth. In both economies this has diluted the potential impact on relative prices that might have been expected to result from a shift in the locus and timing of price adjustments. Consequently, major relative price rearrangements did not occur in Hungary until after 1987; in Poland significant change took place in 1981-82, 1988-89, and, more radi- cally, with the 1990 stabilization. In the 1990 stabilization program in Poland price controls were almost totally abandoned. Third, price liberalization has been associated with a greater explicit linking of domestic prices to foreign prices. In Hungary between 1973 and 1978 con- sumer prices expanded at least six percentage points below the rates of countries of the Organisation for Economic Co-operation and Development. In 1979-80 the price level was adjusted sharply upward to eliminate the wedge between producer and consumer prices. Producer prices-principally for exportables- were then explicitly linked to international prices through the competitive pric- ing rule. I[n Poland the impact of foreign prices on administered prices remained relatively weak. The equalization settlements system equalized transaction prices for exporters and importers to domestic market prices through subsidies and taxes. Figures 1 and 2 show that imported inflation has become progressively more powerful. In Hungary there was a pattern of staggered adjustment of producer and consumer prices to foreign prices from the mid-1970s onward. Import and domestic consumer prices moved more closely together during the 1980s. After 1986 Hungarian inflation began to exceed that of its convertible trading part- 36 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Figure 1. Changes in Prices and the Official Exchange Rate in Hungary, 1981.2-90.1 Percent 20 15- 10 Prices Exchange rate4 5 t . '',/ 1 I0t 1~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~f.f 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 Year and quaner Source. Intermatona Monetary Fund (vaious yat). ners. In Poland import and domestic prices moved together after 1982. Imple- nenting a competitive exchange rate policy aimed at boosting exports and re- ducing the gap between official and parallel rates resulted in recurrent depreciation and a strong imported inflation effect. Reduction in the share of administered prices in total prices and the explicit association between domestic and foreign prices appear to have allowed some linkage, albeit very weak, of price changes to the real side of the economy. Inflationary Experiences under lReform In both Hungary and Poland in the 1980s consumer prices, wages, and unit labor costs moved together closely (figures 3 and 4). However, similar ground rules for reform generated radically divergent inflationary paths. Thus the re- spective inflationary experiences in the two countries must be analyzed in rela- tion to the conditions under which the accelerated reforms of the 1980s were enacted. Inflation in Hungary. An annual inflation model covering 1960-87 and relat- ing change in consumer prices to lagged inflation, import prices, unit labor costs, and an excess demand term provides a reasonably robust explanation of Commander and Coricelli 37 Figure 2. Changes in Prices and the Official Excbange Rate in Poland, 1982.2- 89.4 Percent 120 Exchange rate 100 . 80 . 60 40 20 Prices ~ A, 0 7.. -20~~~~~~~~~_ -20 ,,, I,.I,,I, I ,, I,, I,, I, 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 1982 1983 1984 1985 1986 1987 1988 1989 Year and quarter Source International Monetary Fund (various years). Hungarian inflation, particularly during the pre-1982 period.' There is no evi- dence of structural change in Hungary up to 1987. Inflation has had a marked inertial component, as would be expected in a system with a significant share of administered prices. There is strong covariation of cost factors and domestic prices. Changes in unit labor costs and excess demand appear to have been more powerful determinants than first-round import price effects of the inflation rate. Inflation in Poland. In Poland during 1953-82, when prices were fully con- trolled from the center, consumer prices broadly reflected movements in total costs. However, within this aggregate movement, large changes in relative prices took place. Prices of basic items-food, transport, rent, and energy-lagged consistently behind the overall price index. In addition monetary growth out- paced price growth in the 1970s, creating a significant imbalance, which was further associated with a large premium of the black market exchange rate over the official exchange rate. The price reform of 1982, with the attendant jump in administered prices, aimed to change both relative prices and reduce the monetary overhang. The 1. Results of two-stage least-squares estimation of the annual inflation model for Hungary are avail- able from the authors. 38 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Figure 3. Cbanges in Prices and Wages in Hungary, 1981.2-90.1 Percent 20 15 10 Wages Prices 5 1234123412341234123412341234123412341234 1981 1982 1983 1985 1984 1986 1987 1988 1989 1990 Year and quarter Source; Internaional Monetary Fund (various years). initial jump in the price level translated into an upward shift of the rate of inflation, which stabilized before 1987 at around 15-20 percent a year. Empiri- cal evidence, including tests for structural breaks, suggests that the period after 1982 represents a new regime in price behavior and not simply an increase in the rate of inflation.2 This result seems consistent with the qualitative shift from a system of fully controlled prices to a mixed system of controlled and, in princi- ple, market-based prices. The 1980s were marked by major external imbalances, persistent current account deficits, and pressure on the exchange rate. The black market exchange rate premium remained very substantial throughout the 1980s-one indicator of the presence of repressed inflation in the system. Before 1989, however, the frequent devaluations of the official exchange rate did not succeed in reducing the size of the premium. An important factor for explaining the jump into high inflation in 1989 in Poland was a sharp deterioration in the fiscal accounts. Since 1981 the fiscal deficit had generally been restrained to under 1 percent of gross national product (GNP), and in 1988 it was held to 0.22 percent. In the first half of 1989 revenue losses amounting to more than 7 percent of GNP pushed the fiscal deficit close to 2. Chow tests and recursive estimation results are available from the authors. Commander and Coricelli 39 Figure 4. Changes in Prices and Wages in Poland, 1982.2- 89.4 Percent 100 80 - Prices 60 _ 40- Wages , 1 20 -20 .'% l * l 12341234123412341234123412341234 1982 1983 1984 1985 1986 1987 1988 1989 Year and quarter Source. International Monetary Fund (various years). 8 percent of GNP. Continuous depreciation of the exchange rate, liberalization of food prices, major shocks to administered prices, and the introduction of an ex-post indexing scheme in July 1989 generated a wage-price spiral that tipped the economy into hyperinflation. Money could not anchor the system in the absence of exchange rate and wage stability. Further, with the lifting of price controls and the large increases in administered prices, neither could the price controls anchor the system adequately. Sharp upward adjustments to adminis- tered input prices in late 1989 were initially passed through to final prices by producers. The deceleration of inflation after February 1990 can be attributed not only to a major fiscal correction but to the reinstallation of anchors in the system (Coricelli and Rocha 1991). The exchange rate and wages have been the princi- pal anchors, but temporary restraint on adjustments to remaining administered prices has also beend applied. Measures to break the accommodating role of money have been implemented through the independence of the central bank, interest rate policy, and credit ceilings. Tight money, input price adjustments, and a fall in domestic effective demand, alongside a trade liberalization aimed at importing international prices and a new relative price structure, led producers in "exposed" sectors-particularly consumer goods-to lower markups. 40 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Markups for producer goods were sustained longer (Frydman and Wellisz 1991). These measures reduced monthly inflation to less than 5 percent by October 1990. However, output declined by around 28 percent in the industrial sector during the first eight months of 1990, and unemployment increased to 7.5 percent of the labor force. Wage Determination Wage determination is critical as there is no effective endogenous mechanism of wage restraint in a socialist economy. In Hungary centrally imposed wage norms and bargaining rules were largely sustained, and wage drift was re- stricted. For Poland wage bargaining can best be interpreted in a noncooperative setting (Commander and Staehr 1990). The outcome was stronger wage and cost-side pressure on prices. Polish inflation was consequently powerfully linked to changes in the two key flow variables: wages and the exchange rate (Lipton and Sachs 1990). II. THE MODEL The model summarized here describes a simple markup pricing system in which price movements, either through government controls or through the behavior of monopolistic firms, reflect cost movements (for detailed features, see the appendix). The pricing routine is related to the absence of any strong, market-based restraint on either the price or wage side. Consequently, both price and wage controls are intrinsic to the stability of the system. Prices and market power are restrained by permissible margins on the markup, and wages are restrained by reference to expected inflation and some association to productivity. Features of the Partially Reformed Socialist Economy The sustaining conditions for price growth can be traced both to the fiscal- monetary side and to the discrete effects of exchange rate adjustment. On the fiscal-monetary side, the rate of monetization will be directly acted on by the consolidated fiscal deficit, including the quasi-fiscal deficit. Sustained inflation can be attributed to the presence of fiscal deficits that cannot be financed through debt issue on a sustainable basis (Anand and van Wijnbergen 1987). In systems in which administered prices are significant, if price controls re- press inflation in the short run, the elimination of budgetary subsidies associated with lifting price controls would translate that repressed inflation into open inflation. Moreover, with regard to the exogenous shift factors affecting the inflation rate, specific weight has to be attached to the role of discrete adjust- ments to administered prices by the planner. Such adjustments can be viewed as a lagged response by the planner to macroeconomic imbalances, particularly excess demand in goods markets, and as an attempt to modify the structure of relative prices. However, excess demand would be determined not simply by Commander and Coricelli 41 unanticipated changes in foreign price, monetary, and fiscal impulses, but also by the effect of fixed-price regimes. Dynamic adjustment is also affected by the timing of price decisions. Adminis- trative rules generally impose long lags between adjustments of controlled prices. A mixed system of controlled and market prices leads to two sources of nonsynchronization in the price setting (Commander and Coricelli 1990). The first relates to the nonsynchronization between the change of controlled and market prices. The second arises, as in market economies, from the decentraliza- tion, and thus lack of coordination, of the market. Liberalization of prices for certain categories of goods is likely to increase the flexibility of the price system. However, liberalization introduces an element of nonsynchronization in price changes and thus can increase price system inertia (Taylor 1979; Pazos 1972). Both price staggering and wage indexation could be expected to facilitate the translation of shocks to the price level into a persistent increase in the rate of inflation. The persistence of inflation is strengthened to the extenlt that enterprises operate a constant markup rule, wages are inflexible, and the planner adjusts administered prices in lagged response to excess demand in goods markets. Reduced-Form Price and Wage Equations The reduced-form dynamic models for prices and wages are given in equa- tions 1 and 2: (1) DP, = UO + alDP- I + a2(Dpw, + De,) + x3[Dw, - D(yIL),] + U4(Yt - y*) + oa5 [(M/RS),1 - (M/RS)* i1] - (6(P -P)t-1 (2) Dw = SO + 31DwCt + 032[D(yIL)J] + 03(W - )t-1 + 04DeB, + 05[(RS/P)t - (RS/P);]. In equation 1 prices (p) are a function of lagged prices, import prices (e denotes the exchange rate), unit labor cost (w, y, and L denote wages, output, and labor input, respectively), capacity utilization, an excess demand variable (constructed as the deviation from a perceived normal ratio of monetary holdings in the household sector [M] to retail sales [RS]), and a correction term. In equation 2 wages are related to the centralized wage norm (w,,), productivity, a correction term embodying a real wage target (w-), a black market exchange rate premium variable (eB) for Poland, and a variable that attempts to capture the sensitivity of wage demands to the availability of goods in consumer markets (the deviation from trend of real retail sales for consumer goods). The specification of the correction terms implies convergence to equilibrium values or to certain target values (p - p in equation 1 and w - w in equation 2). However, because socialist economies are commonly thought to be disequi- librium regimes, two aspects have to be emphasized. First, an error correction implies the presence of feedbacks, which induce changes in the dependent vari- 42 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 able when the system deviates from its long-run point of attraction. The feed- backs are total costs or money for prices, and prices or productivity for wages. This does not imply any concept of equilibrium, such as market clearing, be- cause such systems lack endogenous equilibrating mechanisms operating through the labor market or capacity. However, underlying regularities or rules of price and wage behavior were fairly stable during the 1980s. Interestingly, for pricing, these rules were embedded in the institutional setting in which price changes explained by cost changes were permitted and justified. The error cor- rection approach implies that the dynamics are affected by a tendency toward these underlying rules. Monetary variables are not included in the correction term, despite the fact that domestic money and prices are cointegrated, because money is passive in such economies. III. SHORT-RUN PRICE AND WAGE DYNAMICS: RESULTS FROM QUARTERLY DATA The reduced-form equations are estimated using quarterly data from Hungary and Poland. The specified periods are in part given by data availability. To explore the price-wage interdependency and to avoid simultaneity bias, we em- ploy the three-stage least-squares procedure.3 Hungary: 1981.3-89.2 The results of simultaneous estimation of the price and wage equations using quarterly data for Hungary are reported in table 2. Consumer price changes are related to lagged prices, the exchange rate, contemporaneous and lagged unit labor cost summarized in a first-period moving average, a money term proxying demand-side features, and a correction term relating prices to unit labor costs. Wages are related to consumer prices, productivity in the material sector, and a correction term, specifying a real wage target. The variables are stationary and significant at the 5 percent level. For the price equation the overall fit is quite acceptable, there is parameter stability when estimated recursively, and there is an absence of serious autocorrelation. The equation is not sensitive to altering the sample period. The wage equation is more problematic, appearing underidentified and with its fit deteriorating seri- ously toward the end of the sample. Nevertheless both equations capture the short-run dynamics reasonably well. The estimations indicate some key dynamic features for price and wage changes during the 1980s. Although foreign prices are important in the trans- 3. Before estimating the dynamic models, we tested for stationarity with Dickey Fuller and Augmented Dickey Fuller tests. We recognize that these are weak tests, but conveniently the time series proved to be largely integrated of the first order, I (1), so that first differencing of the data was appropriate. Estimation was initially done with levels and then in dynamic form with truncation of the lag polynomials and some restrictions on coefficients. The correction terms represent the lagged residuals from discrete, truncated levels equations, tested for nonstationarity of the residuals. Finally, the equations were tested for parame- ter stability and for sensitivity to changes in sample period. Commander and Coricelli 43 Table 2. Estimation Results from Quarterly System Three-Stage Least Squares, Hungary 1981.3-89.2 Estimated Item value Dependent variable: consumer prices Consumer prices lagged one period 0.23 (2.70) Official exchange rate 0.23 (4.11) Unit labor cost' 0.65 (6.49) M2/retail sales 0.21 (2.22) Correction termb -0.14 (-1.87) R2= 0.80 se = 0.016 F= 17.73 DW = 1.87 Dependent variable: wages Consumer prices 0.86 (5.94) Industrial productivityc 0.19 (1.89) Dummy variable for 1984.4 -0.05 (-2.32) Dummy variable for 1988.1 -0.05 (-2.26) Correction termd -0.52 (-3.65) R2 = 0.65 se 0.022 F 9.91 DW= 1.82 Note: All variables except for the dummy variables are in first differences of logs; t-statistics are in parentheses. Cointegration tests and results are available from the authors. a. One-period moving average of first differences of the log of wages minus the log of productivity. b. Correction term is lcpr = cl + C2 IULC (ULC is unit labor cost) lagged one period. c. Industrial production/industrial employment. d. Correction term is IWAGE = Cl + c2 lcPi (cPi is consumer price index) lagged one period. Source: International Monetary Fund and authors' estimations. mission of inflation, cost-side factors have greater weight. Demand-side pres- sures matter, and it seems likely that one mechanism works through the impact on the relative prices of administered and market goods as a factor determining the scale of adjustment to administered prices. Strikingly, no standard associa- tion between output or capacity utilization and prices exists. This emphasizes the lack of an underlying mechanism of equilibration in the economy despite the introduction of greater market-based features. The presence of administered prices imparts inertia to the system, but the models, as presently specified, cannot explicitly capture the price-price dynamics associated with nonsynchro- nization in price adjustment over administered and market prices. 44 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 The price equation. A lagged price term is included on the right side. Argu- ment for inclusion of this term relates to the weight of administered prices, where such prices can be considered to have stronger-than-normal properties of exogeneity. The coefficient on a one-period lag at 0.23 points to the fairly high degree of inertia in the system. The impact elasticity with respect to the ex- change rate at 0.23 is greater than in the annual model, but this difference could be expected given the more active exchange rate policy and stronger explicit linkage of domestic to foreign prices during this period. The demand pressure variable is very significant. Little can be said about the interactive effects on administered and market prices of demand pressure, but the elasticity indicates the somewhat diluted translation into aggregate price increase. In the initial single equation estimation, we inserted an output gap variable. It could have been expected that with the progressive introduction of market- based rules into the Hungarian economy that a more conventional relation of output to prices would emerge. Recursive estimation and shortening of the sample at the end of the period did indeed result in the term's picking up significance toward the end quarters. However, closer examination of the resid- uals indicated that the significance was largely taken from the noise associated with the price shifts during and after 1988, and the term was not included in the final estimations. The unit labor cost variable, with a moving average elasticity of 0.65, substantiates the powerful link of price growth to costs, as would be expected from the model elaborated above. The correction term is the argument that prices are cointegrated with unit labor costs. The Dickey Fuller test exceeds the 5 percent critical value for the null hypothesis for this specification. The coefficient enters the equation with the right sign and is significant at the 5 percent level. The mean lag of adjustment is reasonably rapid at around four quarters. Recursive estimation and Chow tests prompt inclusion of a dummy for the first quarter of 1988. There is no serious autocorrelation, even when the lagged- price term is dropped. The bands for the recursive estimations widen slightly at the end of the period, pointing to larger unexplained variation at the end of the sample. One-step residuals confirm this feature while Chow tests show a lack of smoothness toward the end of the period, with a persistent spike at 1988.1. This is unsurprising given the timing of changes to administered prices and the tax system in this period. The dummy for 1988.1 in the single equation estimation was significant but did not change the size of the other coefficients or improve the overall fit or properties of the model. Consequently, in the final system estimation it was omitted. The wage equation. The wage equation appears underidentified, and there is some negative autocorrelation in the residuals. There is a very marked deteriora- tion in performance at the end of the sample. No wedge term could be ade- quately specified capturing any gap between producer and consumer wage, and there is no explicit treatment, due to lack of data availability, of the wage tax Commander and Coricelii 45 effect. Up to 1985 wage changes were clustered in the first quarter; thereafter the annual wage round gave way to more random adjustments. Wage adjust- ments within years appear to have been adopted as an institutional routine associated with the acceleration in consumer prices during the later part of the sample. For consumer and producer prices there is a similar clustering of changes in the first quarter. For productivity-measured over the socialist mate- rial sector-changes generally occur in second and fourth quarters. The spike in the fourth quarter can partly be explained by the structure of the wage round and the effect of the wage tax in associating wage and productivity paths. Stability tests picked out breaks at 1984.4 and 1988.1, for which dummies were inserted. In general, institutional particularities, changes to the wage system, and wage tax rules and their incomplete specification in the model explain the limitations of the reported equation (Commander and Staehr 1990). The 0.86 coefficient on the price term in the wage equation points to the strong association of wage to consumer price movements. There appear to be no effective unanticipated price changes or price surprises. Testing this further by inserting a price surprise variable yielded a weak positive coefficient that was very insignificant. The absence of any price surprise, despite the acceleration in inflation during this period, can best be explained with respect to a real wage target maintained by the planner. Wages appear to have been fixed in relation to announced prices. To the extent that centrally given wage norms have aimed to impose an inflation penalty, bargaining has reconverged the wage and price path. The p:roductivity term is significant, and its coefficient has roughly the same magnitude as that for market economies. The correction term is written as a real wage target, is very significant, and points to a rapid adjustment speed. It should however be interpreted with some caution as an observed association between wage and price variables during the sample period, rather than as a long run equilibrium relationship. It is evident that there is an underlying linkage between price, productivity, and wage changes, as suggested by our model. Conse- quently, the wage does not enter as a completely exogenous variable. Poland: 1983.4-90.1 The Polish price system has been subjected to a set of major institutional changes. The first change was the creation of a mixed system of free and con- trolled prices in 1982. There were further modifications in 1988-89 and a major change in January 1990. Using annual data, Chow tests and recursive estima- tion identify 1982 as a structural break. Our estimations consequently use quar- terly information for the period after 1982. Preliminary testing for a break in 1990.1 does not indicate a comparable break in the pricing regime. The estimation results for Poland are presented in table 3. Before estimating the system, price and wage equations were run discretely to test for parameter stability, data coherence, and goodness of fit. All variables are highly significant. The model, as shown by stability tests, survives through the period of high, 46 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 3. Estimation Results from Quarterly System Three-Stage Least Squares, Poland 1983.4-90.1 Estimated item value Dependent variable: consumer prices Consumer prices lagged one period 0.33 (4.20) Official exchange rate 0.33 (7.60) Unit labor costs 0.43 (5.90) M3/retail sales 0.34 (5.40) Correction termb -0.17 (-1.60) R2 = 0.98 se = 0.04 F = 220.02 DW = 1.88 Dependent variable: wages Consumer prices 0.68 (16.40) Industrial productivity 0.58 (3.10) Deviation from trend of real retail sales 0.20 (2.80) Dummy variable for 1988.1 0.20 (5.80) Dummy variable for 1989.1 0.24 (6.60) Correction termc -0.40 (-4.50) R2 = 0.93 se = 0.0S6 F = 60.86 DW = 2.12 Note: All variables except for the dummy variables are in first differences of logs; t-statistics are in parentheses. Cointegration tests and results are available from the authors. a. One-period moving average of first differences of the log of wages minus the log of productivity. b. Correction term is lCPI = c, + C2 IULC + c3 IOER (ULC is unit labor costs, and OER is official exchange rate) lagged one period. c. Correction term is IWAGE = C4 + CSICPI + c61PRO (cPi is consumer price index, and PRO is produc- tivity in the industrial sector) lagged one period. Source: International Monetary Fund and authors' estimations. almost hyper, inflation of 1989-90. This seems to suggest that administered price changes, staggered adjustment of other prices, wage indexation, and an effectively indexed exchange rate maintained, if not increased, inertia in the inflation process. For the price equation the fit is very acceptable, and the variables are stable. For the wage equation the fit is very satisfactory, but there is clear evidence (detected through recursive estimations and Chow tests) of a large spike at 1988.1 and 1989.1. Dummy variables are included for both periods. These Commander and Coricelli 47 dummies can be interpreted as representing some independent wage push, which occurred in a period of increasing social unrest. The price equation. Prices were regressed on a lagged price term, a cost variable, the official exchange rate, a lagged demand pressure variable, and a correction term. A lagged price term is included for the reasons already alluded to in the Hungarian model. In this specification the measure of money was M3, excluding foreign currency deposits, deflated by retail sales. Prices are cointe- grated with unit labor costs and the official exchange rate. A cost-plus specification is used, linking the short-term dynamics to an adjust- ment to an underlying normal markup pricing rule. This specification includes monetary variables only as determinants of short-run movements, and not of a long-run equilibrium relation. Empirical tests were run for the presence of a significant long-run relation between prices and money. A simple standard money dlemand approach indicated the presence of cointegration between prices and money. However, the inflation transmission mechanism is very poorly re- flected by a dynamic equation based on this long-run tendency. This strengthens the view that the concurrent movement of prices, wages, and monetary aggre- gates in the long run is more the result of monetary accommodation than of a causal relation running from money to prices. Inflation is characterized by a relatively high degree of inertia throughout the period. This is captured by the presence of a lagged inflation term in the regres- sion and by a relatively low coefficient of the correction term, implying a some- what slow reaction to deviations from a normal relation between prices and costs. This pattern accords well with the observation of a relatively high and stable inflation during 1984-87. Moreover the stability of the relation through- out 1983-90 suggests that this degree of inertia survived during the acceleration of inflation after 1987. Although the coordination problem typical of systems with many price setters did not affect Poland, the inertia in inflation may have arisen from the peculiar features of a mixed system of market and controlled prices, in addition to the high degree of de facto indexation of wages to prices (Commander and Coricelli 1990). Different rules and timing of price decisions in the different price categories likely gave rise to a substantial staggering of price decisions and a consequent inertia in inflation. The wage equation. In the wage equation, wages were related to productivity in the material sector, prices, a variable denoting deviation from trend of real retail sales, a correction term, and dummy variables for 1988.1 and 1989.1. The correction term reflects cointegration between wages, prices, and productivity. The wage equation helps to stress the role of wages in the transmission of inflationary impulses, especially during 1988-90. Wages have been de facto almost fully indexed to prices throughout the period. Besides the inflation term, the other variables entering the equation show an effect of productivity on 48 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I wages, an effect of availability of goods on wage demands, and some effect of changes in the premium of black market over official exchange rate on wage movements. The latter effect may indicate that workers are concerned with the purchasing power of wages, not only in terms of domestic goods but also of dollar goods. This may be explained by the presence of shops selling foreign goods for foreign currency or by the use of the premium as a proxy for prices in nonofficial markets. The relation between wages and productivity is somewhat puzzling. The rather high coefficient may be partly explained by the system of bonus payments and by a similar seasonal pattern of wages and productivity changes. The high (0.20) elasticity of wages with regard to the demand pressure variable (deviation from trend of retail sales) is also striking. Contrary to theories of wage illusion, which might have been relevant in the 1960s and 1970s (Charemza and Gro- nicki 1988), the perception of a greater availability of goods in the market may have exerted an upward pressure on wage demands. The correction term can be interpreted as a real wage target, adjusted for productivity changes and the resulting payment of bonuses. Moreover the speed of adjustment to deviations of wages from the target level is rather rapid. The coefficient of the correction term, together with a coefficient of nearly 0.7 on the price term, points to a high degree of rigidity of real wages, both in the changes and in the levels. IV. CONCLUDING REMARKS This article sets up a simple inflation model centered on the transmission process and on the short-run dynamics of inflation in partially reformed socialist economies. The model has features derived from market economies with few producers and sticky prices. It attempts to capture some of the specific attributes of socialist economies, including chronic excess demand in goods markets. The bulk of the empirical analysis concentrates on the period after 1982, when significant market-related reforms were implemented. Although socialist econ- omies have been characterized by major disequilibria, we believe that appro- priately specified and tested econometric models can account for these features, without explicit recourse to disequilibrium econometric procedures. The dynamic price and wage models are estimated simultaneously. For both Hungary and Poland the estimation results allow us to explore the role and weight of foreign prices and domestic factors in propagating inflation. Foreign prices are shown to matter, but developments on the cost side are critical in relating exogenous, policy-determined adjustments to the price level to increases in the rate of inflation. Wages are indexed to prices for most periods; however, in Poland there are complex bargaining games and a corresponding inability to make centralized wage norms hold. Polish planners relied increasingly on price adjustments to address the macroeconomic imbalances that emerged, but these only further destabilized the system and failed to address the underlying sources of these imbalances. In contrast the Hungarian experience points to some of the Commander and Coricelli 49 ways in which administered prices can be used to stabilize the system (Com- mander and Coricelli 1990). An important implication is that nominal anchors, particularly wage re- straints, should be included in stabilization programs in reforming socialist economies. The absence of conventional equilibrating mechanisms, and hence of effective market restraints on wages alongside the presence of monopolistic pricing behavior, can easily result in powerful inflationary pressures and ulti- mately higher-than-warranted output costs. APPENDIX. DETAILS OF THE PRICE AND WAGE MODELS The economy has two sectors: material and nonmaterial. Inflation is a weighted average of prices in the material sector (Pm) and prices in the non- material sector (pj): (A-1) p = (1 - C)Pm + api. Each sector has a controlled and noncontrolled price subsector, denoted by subscripts c and n, respectively: (A-2) Pm = 031Prc + f2Pmn (A-3) Pn = 01Pc + 02PsIn For the material sector exposed to competition from the hard currency zone, export prices (Pr) are based on world prices (Pm) and the exchange rate (e): (A-4) Px = p. + e. Import prices are formed by world prices, the exchange rate, and tariffs (T): (A-5) pm = aT(p. + e). In equation A-5 a denotes a terms-of-trade parameter. Total prices in the mate- rial sector trading outside the system of the Council for Mutual Economic Assistance would be a weighted average (k) positively associated with the share of controlled to total prices. Thus (A-6) pm = k(pW + e). The following markup pricing rule is assumed, where p0 denotes output prices in the nonmaterial sector and that part of the material sector isolated from the competition of hard-currency goods: (A-7) PO = (1 + H) p1I+L y where A is the rate of markup, pi is the price of inputs, I is inputs volume, w is wages, L is labor volume, and y is output. The markup rule has standard properties. Changes in output prices will thus depend on unit factor costs and on the markup. Foreign prices do not matter. To 50 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I the extent that the markup relates to some normal level of output (Coutts, Godley, and Nordhaus 1978), the lower the weighted average in equation A-6, the more likely it is that price change will be positively associated with the level of capacity utilization. Capacity utilization (cu) can be written in the usual way as deviation of output from trend (y' ): (A-8) cu = (y - y-). The greater the weight of controlled prices in total prices, the weaker the link between price changes and capacity levels. One would therefore expect a classi- cal planned economy to lack any conventional equilibrating mechanism. In addition this introduces a relative price effect so that the coefficients depend not only on components of costs, but also on the elasticity of the markup with respect to the overall domestic price index. In terms of rates of change, this yields the following linearized, preliminary price equation: (A-9) Pt= to + alPt-I + %2(Pw + e) + U3a(WLlY) + U3b(PJI/Y) + a4(y - Y*). Prices are a positive function of the markup, foreign prices (through a2 and a3), deviations from trend of capacity utilization, and lagged prices. Price setting in the nonmaterial sector would be a subset of equation A-9, obtained by omit- ting a2. Equation A-9 closely resembles a standard markup model applied extensively to nonsocialist economies. We now accommodate some stylized features of so- cialist economies: the widespread presence of goods shortages and high liquidity ratios. Purchasing power in consumer markets may generate undesirable effects in the view of the planner. The presumed concern with relative prices on the part of the planner suggests that demand pressure in goods markets can best be captured in a price equation by a variable [(MIRS) - (M/RS):] measuring deviation from a perceived normal ratio of monetary holdings (M) in the house- hold sector to retail sales (RS). This indicates the measure of excess purchasing power in consumer markets and hence the assumed demand-side pressure for relative price adjustment by the planner. The inclusion of both an output gap (y - y'F in equations A-8 and A-9) and an explicit purchasing power variable is an attempt to pin down the particular excess demand features of these economies when controlled and market prices coexist and when the planner monitors their relative price. To the extent that supply remains insensitive to price changes, one would expect a strongly damped equilibrating relation between shortage and inflation. We can now set up in dynamic form the following inflation equation, where D is a first difference operator and a6 is the correction term. (A-10) Dp, = ao + a1Dp,-1 + a2(Dp., + De,) + a3[Dw, - D(y/L),] + a4D(y, - y*) + a5L(M/RS)_.1 - (M/RS)* 1] - a6(P -P*)-1- Commander and Coricelli S1 Wages An excess demand for labor regime obtains in the socialist economy. No usual convex relationship between labor market pressure and wage growth exists. Restrictions on profit-maximizing behavior at the level of the enterprise imply, in combination with full employment, that the wage share would not necessarily be endogenously restrained by productivity and prices. Wages are treated largely as exogenous and structured by centrally determined norms. Productivity- related wage components-a feature of all partially reformed economies- introduce a crude association to average productivity, rather than marginal products. Distinguishing again between material and nonmaterial sectors, aggregate wage growth is a weighted average. In so far as total wages converge to productivity-corrected wage levels for the material sector, a faster rate of pro- ductivity increase in the material sector would, other things being equal, acceler- ate inflation. Wage contracts are of annual duration, with explicit commitment to real wage stability. Available evidence indicates in general adherence to a real wage target (Adam 1979). In principle a form of forward-looking or ex-ante indexation should be operating. Indeed increases related to the centrally imposed norm reflect expected or planned inflation. In this case (A-1l1) Dw, = ot, + o!2E,L1(Dp) where w, is the centrally imposed wage, E is the expectation operator, and p is price inflation. The real wage is a decreasing function of expectational errors, or inflation surprises: (A-12) Dw, = a, + oU2Dp - a3[Dp - E,-(Dp)]. An ex-post indexation scheme would imply (A-13) DWC = a, + C12Dpt-1. Any increase in the inflation rate reduces the real wage, while the indexation scheme introduces inertia in the inflationary process, slows down a reduction in the inflation rate, and makes the wage setting insensitive to any announcement effects (which in the ex-ante specification will immediately reduce the increase in wages). Combining these features with the postulated relation between wages and average productivity yields the following equation to be estimated: (A-14) Dw, = 0 + O31Dw, + fl2[Dqm + Dp - D(1 + T)] where qm is average productivity in the material sector and T is the tax rate on wage growth exceeding the centrally imposed wage, adjusted for productivity. Wage growth consequently derives from a centralized component related to expected inflation and drift attributable to productivity changes in the material sector, offset by tax increases. Equation A-14 would not adequately capture the persistent increase in real wages up to the 1970s and apparent episodes of wage push during the 1980s in 52 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Poland. Assuming bargaining and a concern of workers with a real wage target, the insertion of an error correction term in the equation appears appropriate: (A-15) Dw, = 00 + f1Dw, + l32[Dqg + Dp - D(1 + r)] + /33(w -W*)t where the wage target (w*) is equal to price inflation. As with the price equation, equation A-15 is modified by introducing features specific to a socialist economy. Two variables may affect the wage dynamics. The first is an actual or implicit price index, which incorporates unofficial markets. The premium of the black market exchange rate over the official exchange rate (eB) is used as a proxy for this unrecorded inflation. In economies characterized by chronic shortages, availability of goods may significantly affect wage demands. Thus the second variable specific to a socialist economy mea- sures the deviation from trend of inventories of consumer goods at the retail level. The second variable is denoted by (RS/p)t - (RS/p)*, where RS is retail sales. The general form for the dynamic wage equation now becomes (A-16) Dw, = A0 + I3lDwCt + 032[Dq., + Dp, - D(1 + r)] + 3(w - w*)t.1 + 034DeB, + 035[(RSIp), - (RSIp)*]. However, as we maintain that consumer goods market imbalances and currency substitution have been more powerful in the Polish case, we will drop these terms at the outset from the Hungarian estimations. REFERENCES The word "processed" describes informally produced works that may not be commonly available through library systems. Adam, Jan. 1980. Wage Control and Inflation in the Soviet Bloc Countries. New York: Praegen Anand, Ritu, and Sweder van Wijnbergen. 1987. "Inflation and the Financing of Gov- ernment Expenditure in Turkey: An Introductory Analysis." World Bank, Washington, D.C. Processed. Charemza, Wojciech, and Mirosaw Gronicki. 1988. Plans and Disequilibria in Centrally Planned Economies: Empirical Investigation for Poland. Amsterdam: North-Holland. Charemza, Wojciech, and R. E. Quandt. 1982. "Models and Estimation of Disequilib- rium for Centrally Planned Economies." Review of Economic Studies 49 (155) :109- 16. Commander, Simon, and Fabrizio Coricelli. 1990. "The Macroeconomics of Price Re- form in Socialist Countries: A Dynamic Framework." PRE Working Paper 555. World Bank, Economic Development Institute, Washington, D.C. Processed. Commander, Simon, and Karsten Staehr. 1990. "The Determination of Wages in Social- ist Economies: Some Microfoundations." WPSEDI 713. World Bank, Economic Devel- opment Institute, Washington, D.C. Processed. Commander and Coricelli 53 Coricelli, Fabrizio, and Roberto Rocha. 1991. "Stabilization Programs in Eastern Eu- rope: A Comparative Analysis of the Polish and Yugoslav Programs of 1990." In Vittorio Corbo, Fabrizio Coricelli, and Jan Bossak, eds., Reforming Central and Eastern European Economies: Initial Results and Challenges. Washington, D.C.: World "Bank. Coutts, Kenneth, Wynne Godley, and William Nordhaus. 1978. Industrial Pricing in the United Kingdom. Cambridge University Press. Feltenstein, Andrew. 1989. "Inflation, Price Controls and Fiscal Adjustment: A Survey of the Literature and Analytical Framework." University of Kansas Lawrence. Processed. Frydman, Roman, and Stanislaw Wellisz. 1991. "The Ownership-Control Structure and the Behavior of Polish Enterprises during the 1990 Reforms." In Vittorio Corbo, Fabrizio Coricelli, and Jan Bossak, eds., Reforming Central and Eastern European Economies: Initial Results and Challenges. Washington D.C.: World Bank. Honohan, Patrick. 1989. "Inflation in China.' World Bank, Washington, D.C. Processed. International Monetary Fund. Various years. International Financial Statistics. Washing- ton, D.C. Lipton, David, and Jeffrey Sachs. 1990. "Creating a Market Economy in Eastern Eu- rope: The Case of Poland.' Brookings Papers on Economic Activity. 1:75-147. Nuti, D. M. 1986. "Hidden and Repressed Inflation in Soviet-Type Economies: Defini- tions, Measurements and Stabilization.' In Contributions to Political Economy. Pub- lished for the Cambridge Political Economy Society. London: Academic Press. Pazos, Felipe. 1972. Chronic Inflation in Latin America. New York: Praeger. Portes, Richard. 1977. "The Control of Inflation: Lessons from East European Experi- ence. Economica 44 (174): 109-30. Rocha, Roberto. 1990. "Structural Adjustment and Inflation in Yugoslavia." World Bank, Washington, D.C. Processed. Taylor, John. 1979. "Staggered Wage Setting in a Macro Model.' American Economic Review Papers and Proceedings. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 5S-69 Are High Interest Rates Effective for Stopping High Inflation? Some Skeptical Notes Guillermo A. Calvo High interest rates are considered an effective toolfor stopping high inflation. The case for a policy of high interest rates is developed in terms of a conventional IS-LM model. However, among other things, the model ignores some central aspects of modern credit markets. In particular high interest rates may give rise to nonperforming bank loans, thus seriously jeopardizing the effectiveness of the policy. Examples are developed in which it would be optimal to aim for equilibriums of low, rather than high, interest rates. One of these examples hinges on the existence of nonindexed domestic debt. This article looks at the effectiveness and desirability of relatively high interest rates in stabilization programs. High interest rates refer to nominal interest rates that would result in unsustainable, high real interest rates-that is, real rates of interest that would cause a state of generalized bankruptcy if maintained for a few months or a year-if the program's inflation target were to be achieved. The article evolves through a series of examples, beginning with the IS-LM model described in section I. Despite its simplicity (and early vintage), the model is still important in discussions of policy. The conventional IS-LM approach, which gives strong support to a policy of high interest rates, implies that tight money is effective in reducing economic activity and, through the Phillips curve, is also effective in lowering inflation. The larger the interest rate, the sooner price stalbility will be achieved. Section II questions the relevance of the IS-LM result on the basis of two observations. First, we live in a world in which countries are closely linked from a financial point of view. Therefore high interest rates are likely to signal expec- tations of high devaluation or inflation. Second, high inflation rates make ex- pected or ex ante real interest rates look much smaller than if the stabilization plan's inflation targets were to be fulfilled. Consequently, banks may continue lending despite high nominal rates. With the passage of tirne, however, firms will find themselves in serious financial straits, which eventually may result in a state Guillermo A. Calvo is in the Research Department at the International Monetary Fund. The author wishes to thank Bijan Aghevli, Morris Goldstein, Tim Lane, and three anonymous referees for their useful comments. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 55 56 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 of generalized bankruptcy. This phenomenon may grow almost unnoticed until it is too late to react within the limits of the original program. Section III discusses "sustainability" issues by studying the long-run implica- tions of raising the interest rate on bank deposits. A high rate of interest on bank deposits provokes results that are similar to a cut in the money supply, but does not call for an overall credit rollback. Thus the high rate may look attractive in the short run. However, in contrast with the IS-LM approach, the longer-run perspective illustrated in section III shows that a 1 percent increase in the rate of interest on bank deposits translates immediately into a rate of inflation that is 1 percent higher. Section IV makes the case for particular instances in which interest rates or related policy instruments may be used to eliminate equilibrium indeterminacy and, thus, facilitate the success of a stabilization program. Equilibrium indeter- minacy may be a serious problem if there exists a sizable stock of nonindexed government debt and the government is perceived as behaving in a discretionary manner (that is, the government is unwilling or unable to commit itself to a predetermined set of policy rules). Debt indexation may be a way to avoid equilibrium multiplicity and coax the system toward the best equilibrium outcome. Section V highlights the special circumstances of previously centrally planned economies undertaking radical economic transformation programs. It describes conditions under which a policy of high interest rates may be useful and the factors in Eastern European countries that would limit the usefulness of such a policy. I. THE CONVENTIONAL MODEL Much of current macroeconomic intuition owes a great deal to the enor- mously successful IS-LM model, originally developed by Hicks (1937). The model can be expressed by two market-clearing conditions, corresponding to the money and goods markets. The following log-linear equations represent closed- economy versions of the IS-LM model: (1) m - p = a - b(i -m) + y (money market) (2) y = c - d(i - re) (goods market) where m, p, and y denote the (logs of) the money supply, price level, and output; while i and im stand for the nominal interest rate on loans and on money. The expected rate of inflation is denoted by 1re. Parameters a, b, c, and d are all positive. The left side of equation 1 is the real supply of money; the right side is the demand for real monetary balances. Since money yields a return, the model assumes that the demand for money is a negative function of the difference between its opportunity cost, i, and its return, im. Furthermore the demand for Calvo 57 Figure 1. Determination of Output and Interest Rate Interest rate ( LM 0- y* Output (y) Note: The LUMcurve shifts to LU'when the nominal interest rate on money increases. The IS curve shifts toIS 'when the expected rate of inflation increases. money is unit-elastic with respect to output (this assumption can be easily re- laxed). The left side of equation 2 is the demand-determined level of output; the right side is output demand, which is assumed to be a decreasing function of the expected real interest rate. The relations described in equations 1 and 2 are depicted in figure 1 for a given price level, inflationary expectations, and interest rate on money. Equa- tion 1 is satisfied along the locus of points labeled LM, while equation 2 holds along the line denoted IS. The equilibrium levels of the nominal interest rate and output are indicated by i* and y'-, respectively. What does figure 1 say about the use of interest rate policy for stabilization? The conventional wisdom relies on a Phillips curve relationship, according to which the rate of inflation, ir, is an increasing function of excess demand in the goods market and the expected rate of inflation. This is represented by (3) 7r 1 j (y, ayf) + 7rh where Yfe is full employment output and j is a positive parameter. From equa- tions 1, 2, and 3, a cut in the money supply shifts the LM curve to the left, raises the interest rate, and puts downward pressure on output, which lowers the rate of inflation. The same results could be obtained by raising the interest rate on money, which would generate excess money demand and shift the LM curve to the left. Several South American countries have used the interest rate on money for stabilization. Manipulation of the interest rate on money does not require call- 58 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 ing back credit lines, as would be the case if the money supply were cut. Because the money supply normally includes bank deposits and bank deposits are the counterpart of bank loans, lowering the money supply sets in motion a contrac- tion of bank credit, which might cause productive firms and financial institu- tions to collapse. A rise in the interest rate on money would increase the equilibrium interest rate, which means that some investment projects would probably be dropped. This contraction, however, is likely to be more orderly than the one provoked by the indiscriminate curtailment of bank credit that a fall in the money supply might cause. The interest rate on money is an instrument that achieves the goals of a cut in the money supply at less cost in the short run. In the medium or long run the costs may be reversed, but the short-run benefits of a policy based on the interest rate on money may prove irresistible when a stabilization program begins to founder. High (nominal) interest rates could result from high inflationary expectations. For example, a rise in the expected rate of inflation would shift the IS curve in figure 1 to the right, resulting in higher interest, but also higher output. As indicated in equation 3, inflation rises because expected inflation and output have increased. This underscores the relevance of lowering inflationary expectations. II. A POLICY OF HIGH INTEREST RATES: ELEMENTS FOR A CRITIQUE A policy of high interest rates and tight money may be an effective tool to aid a disinflation process. However, persistently high interest rates may indicate lack of confidence in the program and may lead to a state of generalized bankruptcy. Unfortunately, it may take a long while for it to become apparent that serious financial trouble exists. A recent example of a policy of high interest rates is the Argentinean stabiliza- tion program of July 1989. The nominal rate of interest hovered around 15 percent a month from August to November, while the exchange rate was fixed against the dollar and inflation came down to about 5 percent a month. The program collapsed toward the end of November 1989, and high inflation re- sumed during the next two months. The Expected Rate of Devaluation and Inflation The conventional model is a very short-run depiction of the economy, in which financial linkages with the rest of the world are not explicitly considered. Some countries with high inflation face serious international debt problems, and their public sectors are subject to stringent borrowing constraints. The private sector in these countries, however, may not face similar constraints. Because of capital flight and unofficial foreign exchange, portfolio shifts between foreign and domestic assets may completely bypass the need to use foreign credit lines. Calvo 59 Therefore, at the margin, the domestic economy may function as if it operated under perfect capital mobility. Under perfect capital mobility, the domestic interest rate, i, is equal to the international nominal interest rate (the interest rate in terms of foreign cur- rency), I, plus the expected rate of devaluation of the domestic currency, -e, as follows: (4) i = I + Ee + k where I and k are assumed to be exogenous. Given fixed exchange rates and the expected rate of devaluation, any attempt to affect the domestic interest rate through monetary policy will simply provoke compensatory capital flows. This is one of the main messages of the Mundell- Fleming extension of the conventional model (Frenkel and Razin 1987). A cut in the money supply will induce the public to sell their dollars for domestic interest- bearing assets. The domestic currency will then tend to appreciate, and the central bank will be forced to intervene to keep the parity. The central bank will buy the dollars with domestic currency, which tends to undo the initial money- tightening operation. Therefore an active money or interest policy will eventu- ally require giving up either fixed exchange rates or capital mobility. Under floating exchange rates, monetary policy can be stronger (Frenkel and Razin 1987). However, the interest rate still satisfies equation 4, so that for a given (exogenously determined) expected rate of devaluation, the domestic in- terest rate is unaffected by monetary policy. Thus a policy of tight money will not raise the domestic interest rate, but will result in output loss. Equation 4 shows that the domestic interest rate rises, point by point, with the expected rate of devaluation. Thus a high domestic interest rate implies a high expected rate of devaluation, which is a likely signal that the stabilization program is not credible. In Dornbusch's (1976) extension of the Mundell-Fleming model under flexible exchange rates, the expected rate of devaluation becomes an endogenous vari- able. He shows that a permanent cut in the money supply will temporarily appreciate the domestic currency with respect to its long-run level. Conse- quently, the cut in the money supply induces the expectation of future devalua- tions, which, by equation 4, tends to increase the nominal interest rate. How- ever, the increase is a temporary phenomenon; the greater the degree of price flexibility, the shorter the period of increase. Because price flexibility is likely to be high in countries undergoing anti-inflation programs, the temporary appre- ciation of the currency does not seem to be an important factor for justifying high interest rates. Lower-Than-Expected Inflation Rates If inflationary expectations could be manipulated by a skillful politician, infla- tion would immediately decline, and no further short-run monetary stringency 60 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I would be required. A smaller expected rate of inflation would shift the IS curve to the left, resulting in some output loss, but the nominal interest rate would fall. Success in reducing inflationary expectations results in low interest rates. Conse- quently, tight money and high interest rates are likely to be more necessary when the public is reluctant to lower its expectations of inflation. Consider a situation of tight money and downward-inflexible inflationary expectations in the short run. As a result of the tight money policy, ex post inflation turns out to be less than expected. The ex post real interest rate, r, is the interest rate that a borrower will have to pay in real terms. It is defined as (5) r = i- r. It may be that, because the inflation rate is lower than expected, the public is quick to adjust downward its expectations about inflation. The tight money policy, however, makes the ex post real interest rate larger than the ex ante one, which implies unexpected redistribution of wealth from borrowers to lenders. This unplanned redistribution may become critical if expectations about infla- tion are slow to adjust. If real interest rates remain high in subsequent months, for example, the redistribution may cause the productive sector to deteriorate. Investors may have undertaken prudent investment prcjects in the expectation that the real interest rate would be low (given that the expected rate of inflation is high). However, a fall in the inflation rate by more than expected for an extended period of time may drive all or most firms into bankruptcy. Thus, if the management of expectations relies on stringent short-term mone- tary policy, generalized bankruptcy may result. A state of generalized bank- ruptcy implies that banks would have to terminate most of their loans, revealing the banks' weak financial position. Firms and banks may unwittingly be led into a scheme to hide these facts from the public and the regulators. As signs of trouble develop, the authorities may opt to loosen monetary policy. To solve bankruptcy problems, however, the ex post real interest rate must be lowered to about its ex ante levels, which would mean a return to high inflation. The longer the time that elapses between the initiation of the stabilization process and the return to high inflation, the higher will be the level of inflation required. The high-inflation solution may be postponed by increasing the remuneration of bank reserves held against deposits to expand the money supply. This would be equivalent to increasing the nominal interest rate on money. The increase in the money supply would result in an increase in loanable funds channeled through the banking system and would help to keep inflation under control. This emergency policy cannot be maintained indefinitely, however. III. SUSTAINABILITY ISSUES A policy of high interest rates may give rise to a state of generalized bank- ruptcy, which may impair the stabilization program in the medium run. In the short-run model, however, explicit attention was not paid to private or public Calvo 61 budget constraints. Thus we were not able to articulate the medium-term diffi- culties in a completely rigorous fashion. The impact of the rate of interest on bank deposits has played an important role in recent Latin American experiences. Bobember and Makinen (1983) pre- sent a fascinating account of the destabilizing effects of indexed deposits in Hungary after World War II. Increased Demandfor Money For the sake of simplicity, bank money is assumed to be the only form of domestic money held by the private sector (the private sector holds no domestic cash). However, individuals hold foreign exchange,f. Hence, assuming purchas- ing power parity and a unitary international price level, the private sector's real financial wealth, w, is given by (6) w=x +f where x stands for real monetary balances; in the notation of section II, x exp(m - p). The rate of return of foreign exchange in domestic currency is the expected rate of devaluation and that of domestic money is the interest rate on money. T he demand for money is assumed to be a function, L, of its opportunity cost, and thus (7) x = L(Ee - ij) L' < O. In general, the demand for money ought to be a function of financial wealth, but this term is omitted for simplicity. If the interest rate on bank deposits, im' increases and the exchange rate is constant, according to equation 7 the demand for money will rise. The public will try to shift their financial portfolios out of foreign currency and into domes- tic deposits (equation 6). If the government does not intervene, the shift into domestic deposits will put downward pressure on the exchange rate, thus caus- ing the domestic currency to appreciate. Because of purchasing power parity, this portfolio shift tends to lower the price level. Consequently, an increase in the interest rate on money immediately helps to reduce the inflation rate. If the monetary authority does not react to the higher demand for domestic money, the exchange rate and price level fall to accommodate the higher money demand. From the fiscal point of view, the increase in the interest rate on money does not bring about any additional resources. Given the expected rate of deval- uation, the burden of servicing deposits goes up. Therefore, except for the very short-lived inflation relief, the increased interest rates may cause greater fiscal difficulties. The monetary authority may intervene so as not to allow the exchange rate to appreciate (or depreciate) following a rise in the interest rate on money. As a result the central bank will accumulate foreign exchange according to (8) L(e X-0 62 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 where f is the change in foreign exchange and x5 denotes the stock of real monetary balances before the policy of higher interest rates is put into effect. If the accumulation of reserves has positive side effects, the policy reaps some short-run benefits. However, if reserves earn no real interest, the net fiscal effect is that the initial capital inflow given by equation 8 adds nothing to revenue. If the real interest rate earned on reserves is r*, then taking equation 8 into account, the net cost of servicing bank deposits, CN, becomes (9) CN = L(ee - i.) (im - r*) + r*x6 where we assume initially im = O. That is, the net cost of servicing bank deposits equals the direct cost, im x, minus interest on reserves accumulated as a conse- quence of raising the interest rate on deposits from 0 to a positive number, which, recalling equation 8, is given by (x - xj)r*. Hence a necessary condition for the fiscal situation to improve (compared with setting im = 0) is that the interest rate on money be less than the interest rate on reserves (im < r*). This is a very stringent condition, given that in actuality the rate of interest earned on reserves is less than 1 percent a month. The general implication, therefore, is that although positive nominal deposit rates may be desirable from a fiscal point of view, their optimal level, when authorities are committed to zero devaluation, is unlikely to exceed 1 percent a month. Endogenously Determined Rate of Devaluation Zero devaluation is, of course, a very strong assumption. In practice policy- makers commit themselves to such a policy for only limited periods. When the rate of devaluation is an endogenous variable, a realistic assumption is that the government runs a real deficit, denoted by g (assumed constant for simplicity). Money is issued to cover the real deficit and any cost of servicing bank deposits. Therefore, (10) M= Eg + inM - r*E(x - x5) where a dot over a variable denotes its time derivative. M is the nominal stock of money, and E is the exchange rate (the foreign price of output is assumed to be unity). The first term on the right side of equation 10 is the nominal value of the government's nonbank deficit, the second term is the interest on deposits, and the third term is the nominal value of the yield on the initial accumulation of reserves at the central bank. Dividing both sides of equation 10 by the exchange rate, (11) * = g-(rt + i.)x + r*x6 where e is the actual rate of devaluation. It is assumed that the private sector immediately adjusts its expectations to actual values (that is, it operates under perfect foresight). Thus, (12) Ee = C. Calvo 63 Consequently, by equations 7, 11, and 12, and holding real money balances constant (* = 0), (13) 0 = g - (r* + E - im)L(E - im) + r*x-. Thus, for a given interest rate on money, the steady-state equilibrium condition in equation 13 determines the rates of devaluation that are consistent with the model. In equation 13 the only thing that matters is the difference between the actual rate of devaluation and the interest rate on money. Hence, a higher interest rate on money brings about an increase in all equilibrium rates of devaluation (and inflation) that is exactly equal to the initial rise in the interest rate on money. Thus, under perfect foresight, an increase in interest rates for deposits will result in higher steady-state inflation. If, in addition, the real nonbank govern- ment deficit is not affected by the policy on interest rates, then the steady-state equilibrium rate of inflation increases point for point with the deposit interest rate. (For similar implications in models in which bonds take the place of interest-bearing money, see Sargent and Wallace 1981 and Calvo 1985.) This result underscores the importance of looking beyond the short run in evaluating interest rate policies. Contrary to our discussion in the previous section, here the higher interest rate does not give rise to bankruptcies. The end result, however, is not very different: inflation goes up. IV. THE ROLE OF AN ACTIVE INTEREST RATE POLICY In this section we develop examples in which the economy displays equilib- rium multiplicity and where there is room for an active interest rate policy. Suppose the fiscal deficit is fully "monetized," as implied by equation 13. Even when the rate of interest on money is zero, equilibrium may be consistent with more than one rate of devaluation. For instance, suppose the demand for money in equation 7 satisfies the following: (14) L(E - im) = exp [-a(e - i)] which corresponds to the Cagan model of the demand for money. Equation 13 may exhibit two solutions for the equilibrium rate of devaluation, as depicted in figure 2 at points E1 and e2. The government could eliminate inflation indeterminacy by choosing the smallest equilibrium rate of devaluation (el in figure 2) and establishing a system of predetermined exchange rates. (See Bruno and Fischer 1985 on inflation indeterminacy and the "heterodox" program in Israel.) By controlling the rate of devaluation, the government can move the economy away from the high-interest and into the low-interest equilibrium. Although this is an example of an active interest rate policy, note that the optimal policy consists of pushing the system to a low-interest equilibrium. 64 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Figure 2. Steady State Equilibriums Steady-state govermnent financing gap g+rxa 0 g+tx -(r+E)L(E) g+rx5-rL(O) 0 E1 2 Rate of devaluation of the domestic currency (E) Equilibrium Indeterminacy and Discretionary Policy Equilibrium indeterminacy may be an even more serious problem if the gov- ernment is likely to resort to discretionary policy. This point is illustrated by a simple example that also demonstrates the role that debt indexation may play (Calvo 1988 and 1989a). It is assumed that no interest is paid on money and that, contrary to the previous example, to finance the deficit the government resorts to issuing indexed and nonindexed bonds. Indexed bonds are a promise to pay the original value plus a return based on the real interest rate in units of output in the next period (the analysis is conducted in the context of a two- period model). Nonindexed bonds are nominal bonds that promise to pay the original value plus a return based on the nominal interest rate in units of domes- tic money in the next period. Furthermore it is assumed that there is perfect capital mobility and that the international one-period real interest rate is zero. Therefore, in equilibrium the domestic real interest rate must equal zero. Likewise, if there is perfect foresight, in equilibrium the nominal domestic inteiest rate is equal to the rate of inflation. In equilibrium the implicit real interest rate on nonindexed bonds must also be equal to the international interest rate (which is zero by assumption). To introduce discretionary policy in a simple manner, it is assumed that the rate of inflation cannot be determined until the next period. Moreover for simplicity we assume that in the next period the policymaker can choose the inflation rate to be either zero or some positive (possibly large) number (*). Zero inflation has no social cost. If, however, a positive rate of inflation is imple- mented, the inflation cost is positive and, for simplicity, is assumed to be equiva- lent to enlarging the real government deficit. Calvo 65 Because this is a two-period world, total expenditure in the next period has to be completely financed by taxes. Taxes are distorting, and, again for simplicity, their social cost is assumed to be proportional to tax revenue. It is assumed that the present government is stuck with a given amount of debt, which can, in principle, be split between indexed and nonindexed bonds (the sum of indexed and nonindexed bonds, however, is given). Total next-period taxes (Tj+1) in real terms are given by (15a) T1+1 = g + B, + B,(1 + i), if r = O (15b) Tt+1 = g + Bi + Bn (1 + i)/(1 + *) + c, if r > O where Bi is the stock of indexed bonds in terms of output, Bn is the stock of nonindexed bonds in terms of output, and & is the social output cost associated with high inflation. Equations 15a and 15b incorporate the equilibrium condi- tion that the domestic real interest rate must equal zero, since the return on indexed bonds cannot be affected by the government in the next period. Given that the rate of inflation can take only two admissible values, the only two types of equilibriums are a low-inflation equilibrium and a high-inflation equilibrium. Under a low-inflation equilibrium, people at present expect infla- tion to be zero, and thus the interest rate is equal to zero. For this to be a perfect- foresight equilibrium, it is necessary for the government in the next period to have the incentives to pursue a low-inflation policy, that is, to set the inflation rate to ze:ro. Since the costs of each inflation policy are given by equations 15a and 15b, it follows that setting the rate of inflation equal to zero would be the optimal choice if equation 15b exceeds 15a when the interest rate is zero. Thus the condit:ion for low-inflation equilibrium is (16) a - B,*I(l + T) > °- In a high-inflation equilibrium, people expect the rate of inflation to be greater than zero (7r = T). The policymaker will have an incentive to validate expectations of high inflation if equation 15a is larger than 15b. Thus the condition for high-inflation equilibrium is (17) Bni - c > O. Conditions 16 and 17 are depicted in figure 3. If the stock of nominal bonds is equal to zero, then condition 17 can never hold, whereas condition 16 always does. Hence the low-inflation equilibrium is the unique equilibrium. However, as shown in figure 3, there exists a critical value of nominal bonds, after which a high-inflation equilibrium is possible. In the interval between B1 and B2 in figure 3, both low- and high-inflation equilibriums are possible, that is, the economy exhibits rnultiple equilibriums. After B2 in figure 3 only high-inflation equilib- rium is possible, that is, only high inflation is a credible outcome. Because in equilibrium the interest rate is equal to the rate of inflation, then, by equations 15a and 15b, the social cost of taxes and inflation is given by 66 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Figure 3. Equilibrium Determination and the Stock of Nomtinal Bonds Bnlf- C) B1 equilibriu 0 B Stock of nominal Low-inflation B2 bonds (Bn) equilibrium / C- 27 Bn Note: The stock of nominal bonds is nonindexed government debt. g + Bi + Bn, if low-inflation equilibrium holds, and g + Bi + Bn + co, if high- inflation equilibrium holds. Hence low-inflation equilibrium is always socially preferable to high-inflation equilibrium. The intuition for this result is that, because of perfect foresight, the output value of the government's obligations will be the same irrespective of inflation. By assumption, the other items com- prising government expenditure are independent of the rate of inflation. The only difference between a high- and low-inflation equilibrium is the social cost associated with high inflation, which proves that high-inflation equilibrium is inferior to low-inflation equilibrium. Debt Indexation and Low-Inflation Equilibrium From the above example, debt indexation can be helpful to ensure that the economy goes to the low-inflation and low-interest-rate equilibrium. The gov- ernment can achieve a lower stock of nominal bonds by swapping indexed for nonindexed debt until the stock of nominal bonds becomes smaller than the critical level (B1 in figure 3). (See Guidotti and Kumar 1991 on the use of indexed debt in the recent successful anti-inflationary programs in Chile and Israel.) The example's simplicity, however, hides the fact that inflation is not the only tool governments can employ to partially repudiate their debts. Domestic debt, for example, can be "taxed away" without resorting to higher inflation. Although indexation removes inflationary incentives, it may result in other Calvo 67 forms of debt repudiation that might be even more costly. In practice, therefore, the optimal policy will depend on specific country considerations, like the stabil- ity of private ownership legislation. Multiple Equilibriums and Expectations about Inflation It may be that, for considerations outside the model, debt indexation is not possible o:r desirable. If a low-inflation equilibrium prevails, there is nothing the policymaker could or should do. Much more interesting is the case in which the stock of nominal bonds lies in the range of multiple equilibriums. For if the public expects high inflation, interest rates will be high, and the government in the next period will have incentives to engineer high inflation. But if the public expects zero inflation, interest rates will be low, and low inflation will be the incentive-compatible outcome. In that context the policymaker could lead the economy ito the best equilibrium by persuading the public that low inflation will be the final outcome. For that purpose it would be enough if the government refused to sell bonds with an (implicit or explicit) nominal interest larger than zero. Even if the policymaker succeeds in moving the system toward low-inflation equilibriurm, there is another potential complication. If the rate of inflation is zero and the stock of bonds is near B2 in figure 3, then a slight increase in the stock of nominal bonds would imply that the system would be suddenly thrown into the high-inflation equilibrium. This indicates that the stabilization program may be very susceptible to the level of outstanding nonindexed government debt. More important, this kind of example shows that a lack of trouble spots, and even apparent success-for instance, a dramatic initial fall in the rate of inflation--are not necessarily reliable signs that the stabilization program is running on solid ground. V. RECENT STABILIZATION PROGRAMS IN EASTERN EUROPE The policy of high interest rates has received renewed attention in recent stabilization programs in Eastern Europe and has been tried during the first few months of the 1990 stabilization program in Poland. One rationale for adopting high interest rates is that prereform inflation in some of these economies (notably Poland) was closely associated with the existence of substantially nega- tive real interest rates. Therefore high interest rates were expected to send a strong signal about a change in the monetary regime. Such a policy may be useful particularly when, as in Poland, it is carried out for a limited period, and enterprises start the program with excessive inventories (see Calvo and Coricelli, forthcoming). Indeed, a policy of high interest rates could dominate credit rationing-even if credit flows turn out to be the same as under high interest rates-because credit rationing has already been tried in the 68 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I past, and thus its adoption could fail to convey the impression that a drastic policy change has taken place. However, credit markets in Eastern Europe are notoriously segmented and incomplete. The productive sector strongly depends on central bank credit. Even when interenterprise credit exists, it relies very much on the presence of the central bank as a lender of last resort and tends to dry up when the central bank withdraws its implicit support of problem borrowers. Therefore a policy of high interest rates could have a direct negative effect on output. Aside from weaken- ing political support for the program, which may be caused by excess capacity and unemployment, output contraction has a direct negative impact on fiscal revenue, weakening policy sustainability and, thus, the program's credibility (Calvo and Coricelli 1991). VI. CONCLUSION High interest rates may jeopardize the success of an inflation stabilization program, and, if the program does not enjoy full credibility, the contraction that high rates are supposed to generate may not materialize (enough) in the short run. If high interest rates fail to achieve short-run adjustment, the adjustment may come later but at a much greater expense, because it may entail a state of generalized bankruptcy or a demise of the stabilization program. The policymaker's dilemma is not an easy one. On the one hand, high interest rates send a clear signal that a new regime has been put in place. Banking subsidies are reduced or eliminated, and savers have a bigger incentive to make their funds available for productive investment. Furthermore, in the short run high interest rates may decrease aggregate demand, helping the stabilization effort. On the other hand, as shown here, high interest rates may cause the program to fail. Whenever credibility issues are stake, the institutional and historical charac- teristics of the country in question play a central role (see Calvo 1989b). It is relatively easy to point to policies-such as a policy of high interest rates-that have a good chance of failing. A more positive suggestion that comes out of the analysis is that interest rates should be set in such a way that they do not contradict the sustainability of the program. A less ambitious stabilization pro- gram should be implemented. A program should not plan on real interest rates exceeding 15-20 percent a year for an extended period of time, unless there are very good reasons to argue that those rates are consistent with the country's available investment projects. This quantitative constraint on real interest rates is likely to be hard to satisfy in highly indebted countries because of debt overhang and related problems. In such countries financial liberalization will be slower. The binding constraint is not an undersized banking system, but a lack of investment possibilities given the uncertainties created by a large outstanding debt. The country has a debt problem, and it is therefore unlikely that more debt would be the solution. Calvo 69 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Bobember, William A., and Gail E. Makinen. 1983. "The Hungarian Hyperinflation and Stabilization of 1945-46." Journal of Political Economy 91 (October): 801-24. Bruno, Michael, and Stanley Fischer. 1985. "Expectations and the High Inflation Trap." MIT, Cambridge, Mass. Processed. Calvo, Guillermo A. 1985. "Macroeconomic Implications of the Government Budget." Journal of Monetary Economics 15 (January): 95-112. . 1988. "Servicing the Public Debt: The Role of Expectations." American Eco- nomic Review 78 (September): 647-61. 1989a. "Controlling Inflation: The Problem of Non-Indexed Debt.' In Sebastian Edwards and Felipe Larrain, eds., Debt Adjustment and Recovery: Latin America's Prospect for Growth and Development. New York: Basil Blackwell. 1989b. "Incredible Reforms?" In Carlos Diaz-Alejandro and Guillermo Calvo, eds., Debt, Stabilization, and Development. New York: Basil Blackwell. Calvo, Guillermo A., and Fabrizio Coricelli. 1991. "Stagflationary Effects of Stabiliza- tion Programs in Reforming Socialist Countries: Enterprise-Side and Household-Side Factors." The World Bank Economic Review, this issue. . Forthcoming. "Stabilizing a Previously Centrally Planned Economy: Poland 1990?" Economic Policy. Dornbusch, Rudiger. 1976. "Expectations and Exchange Rate Dynamics," Journal of Political Economy 84 (6, December): 1161-76. Frenkel, Jacob A., and Assaf Razin. 1987. "The Mundell-Flemming Model a Quarter Century Later." IMF Staff Papers (December): 567-620. Guidotti, Pablo, and Manmohan Kumar. 1991. "Domestic Public Debt of Externally Indebtecl Countries.' IMF Occasional Paper 80. Washington, D.C.: International Mon- etary Fund. Hicks, John R. 1937. "Mr. Keynes and the Classics: A Suggested Interpretation." Econo- metrica 5: 147-59. Reprinted in W. Fellner and B. F. Haley, eds., Readings in the Theory of Income Distribution. London: George Allen and Unwin, 1961. Rodriguez, Carlos A. 1988. "Comment." In Michael Bruno, Guido Di Tella, Rudiger Dornbusch, and Stanley Fischer, eds., Inflation Stabilization. Cambridge, Mass.: MIT Press. Sargent, Thomas J., and Neil Wallace. 1981. "Some Unpleasant Monetarist Arithmetic?' Quarterly Review of the Federal Reserve Bank of Minneapolis (Fall): 1-17. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 71-90 Stag:flationary Effects of Stabilization Programs in Reforming Socialist Countries: Enterprise-Side and Household-Side Factors Guillermo A. Calvo and Fabrizio Coricelli Dismanttling subsidies could give rise to serious macroeconomic difficulties in the short run. This article explores a view based on the enterprise sector as a central source and main channel of the stagflation phenomenon, using as an example the stagflation that followed the 1990 stabilization program in Poland. The stagflation phenomenon is linked to features of the financial market that are somewhat peculiar to reforming socialist economies: the weak credit links between households and enterprises, and the existence of large interenterprise debt. The policy implications of the enterprise-side view include more explicit consideration of initial conditions in the credit market, implementation of privatization schemes, and the development of a domestic banking system. Several socialist countries are in the process of transforming their economies into a system that relies much more heavily on market mechanisms. A pervasive characteristic of the associated transformation programs is the elimination of price distortions through dismantling a complicated network of subsidies. This is inspired by the valid microeconomic principle that an undistorted price system contributes to efficient resource allocation in the long run. In the short run, however, the elimination of subsidies could give rise to serious and protracted macroeconomic difficulties. For example, eliminating subsidies could bring about a sharp increase in the price of some key factors of production, such as coal and gasoline. Thus, if firms have limited access to credit or if interest rates are high, the increase in production costs could lead to a credit crunch. This article pays special attention to the recent Polish stabilization plan. The Polish plan exhilbits a sizable initial increase in production costs, and its relative long- evity helps to distinguish a little better between blips and trends. (A more com- Guillermo A. Calvo is in the Research Department at the International Monetary Fund, and Fabrizio Coricelli is in the Country Economics Department at the World Bank. The authors would like to acknowledge the useful comments by Eduardo Borensztein, Luis de la Calle, Roman Frydman, Stanislaw Gomulka, Peter Hole, Timothy Lane, Massimo Russo, Andres Solimano, Ulrich Thumm, and three anonymous :referees and to thank Barbara Ossowicka for her efficient research assistance. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 71 72 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I plete evaluation of the Polish program is taken up in Calvo and Coricelli [1991].) An initial stagflationary period accompanied the Polish stabilization program. This article explores a view based on the enterprise sector as a central source and main channel of the stagflation phenomenon. It is suggested that economic transformation programs of the Polish type may contain a significant supply- side shock stemming from the combined effects of a steep rise in input prices (both domestic and foreign) and interest rates, together with an expansion of credit that fails to keep pace with input prices. The "material" intensity of production, namely the use of inputs and energy per unit of output, is extremely high in socialist economies (Gomulka and Rostowski 1988). This implies that an increase in input prices has a particularly large impact on the costs of enter- prises. As a result, firms in Poland curtailed their demand for other firms' products and, eventually, their demand for labor. Thus the view discussed here does not preclude demand by enterprises to play an important contractionary role. In this context the withdrawal of the central bank from its role as "lender of last resort" (seen by the authorities as a necessary step for the switching from a "soft" to a "hard" budget constraint) may have magnified the credit squeeze. Also, given the initial credit segmentation, which delinks the current prof- itability and liquidity conditions of enterprises from their efficiency, such an attempt at introducing a hard budget constraint may have adversely affected firms in a manner largely unrelated to their efficiency. In this regard the presence of a sizable stock of interfirm debt in the period preceding the stabilization program-probably with good firms in the position of creditors-may have played an important role in further spreading across firms the recessionary effects of the liquidity crunch. There are similarities between the enterprise-side view and the neo- structuralist approach applied to nonsocialist developing countries (see, among others, Cavallo 1977, Taylor 1980, and van Wijnbergen 1982 and 1983). As in the neo-structuralist approach, the enterprise-side view stresses the possible con- tractionary effects of tight credit in the presence of serious imperfections in the financial market. However, the view exposed here links the stagflation phenom- enon to features of the financial market that are somewhat peculiar to reforming socialist economies: the weak credit links between households and enterprises, and the existence of large interenterprise debt. The latter, in turn, is linked to chronic "soft budget constraints" and the above-mentioned financial separation of households and firms during the prereform period. In addition, as shown in the appendix, we emphasize the effects of stock liquidity and make the central points in terms of a model with perfect price flexibility. In contrast, the neo- structuralist school relies on flow effects and Keynesian price-wage rigidities. Section I identifies the main relevant stylized facts of the Polish experience. Section II sketches some of the features of the financial market in reforming socialist economies. Section III lays out the enterprise-sector view of these facts, Calvo and Coricelli 73 relying on the simple model contained in the appendix. Section IV contains some policy implications. I. THE POLISH STABILIZATION PROGRAM On January 1, 1990, the Polish government launched a far-reaching program of economic transformation and stabilization. On impact, the inflation rate accelerated sharply and industrial output fell (figures 1 and 2). The initial effects of the program can be characterized as stagflationary. Although official figures may overstate the fall in output (Osband 1991), both the increase in the inflation rate and the drop in production significantly overshot official predictions. This stagflationary outcome is of particular interest because the Polish program can be classified as an exchange-rate-based, heterodox program. As shown by Kiguel and Liviatan (1990), this type of program is usually accompanied by initial expansion and very seldom results in a deep recession, such as occurred in Poland. This comparison may not be fully warranted because the Polish pro- gram entailed a major structural transformation. However, although the struc- tural transformation may call for a fall in output, this would occur only if new prices are lower than new average costs. Otherwise output will not fall, and the Figure 1. Inflation in Poland, 1989-90 (Monthly percentage change in consumer prices) 80 70 60 50 40 30 20 10 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec 1989 1990 Source. Central Statistical Office (1990). 74 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Figure 2. Industrial Production in Poland, 1989-90 (January 1989 = 100) 100 - 90 80 70- 60 Jan Feb Mar Apr MayJun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr MayJun Jul Aug Sep Oct Nov Dec 1 1989 1 1990 l Source, Central Statistical Office (1990). transformation would only be reflected in much lower investment being chan- neled to traditional sectors. The Program's Main Features The program was based on orthodox measures of fiscal and monetary tighten- ing and on unorthodox measures establishing two nominal anchors: the ex- change rate and nominal wages. Because credit ceilings were also imposed, there were actually three nominal anchors. Stabilization measures. The program tightened fiscal policy to restore a bal- ance in the general government accounts for 1990 (in-1989 the deficit was 7.5 percent of gross domestic product [GDP]). The goal was to cut subsidies so as to reduce expenditures by 8 percent of GDP and eliminate income tax relief, yield- ing additional revenue equal to 4 percent of GDP. Ceilings on net domestic assets of the banking system were imposed, allowing for a nominal growth of net domestic assets of 22 percent in the first quarter of 1990 and about 50 percent for the year as a whole. Ceilings on the lending of each commercial bank were also imposed but not until September 1990. On January 1 the National Bank of Poland increased its interest rate for refinancing from 7 to 36 percent monthly. Authorities planned to adjust this rate at least monthly in order to maintain positive real terms higher than the return on deposits in foreign currency. On January 1 the market for foreign exchange for current transactions was unified at a rate of 9,500 ZI/US$, which reflected a 31.6 percent devaluation of Calvo and Coricelli 75 the official rate. This rate was frozen, and it was announced that an active interest rate policy would be used to defend the peg. The exchange rate was maintained at that level until May 1991. The permitted increase in the wage bill was set at 30 percent of the rate of inflation in January; 20 percent in February, March, and April; and 60 percent for the rest of the year, except in July, when it was set to unity. Enterprises were subject to tax penalties of 200-500 percent on any increase above these ceilings. Liberalization measures. In January most remaining price controls were re- moved. No more than 3-5 percent of retail prices were set by the state, com- pared with 50 percent in the latter part of 1989. Foreign exchange could be provided without restrictions for most current transactions, as opposed to the previous system of rationing. Also enterprises were required to surrender all foreign exchange receipts to the central bank. Authorities eliminated all quan- titative restrictions on imports from the convertible currency area and intro- duced a unified, low customs tariff (around 4 percent) for commercial and personal imports. Recession Accompanying the Stabilization Program The initial reaction of the economy to the stabilization program was stagfla- tionary. In January the inflation rate overshot the officially expected rate, with an actual rate of about 79 percent against an official projection of 35 percent, while the decline in industrial production was around 30 percent with respect to January 1989 (see figures 1 and 2). The inflation rate decelerated sharply to 24 percent in February and 4.7 percent in March. However, it oscillated around 4- 5 percent throughout the rest of 1990. A new round of increases of administered prices pushed inflation to double digit monthly rates in January 1991. Industrial output (which is measured as output sold by state enterprises plus a rough estimation of sales by the private sector), after the sharp drop in early 1990, remained practically flat during the rest of 1990, with some signs of mild recov- ery at the end of the year. For 1990 as a whole, industrial output was about 25 percent below its level in 1989. Why the Stagflationary Response? One possible explanation for stagflation in Poland is the sharp fall in real wages during 1990 (table 1). The argument relies on the assumption that aggre- gate demand in Poland depends strongly on real wages and that state enterprises do not lower prices when faced with excess supply (because they follow a fixed markup rule, for example) (see Blanchard and others 1991). There are several factors that make this simple Keynesian view inadequate. First, the household sector has a sizable stock of liquid savings that can be quickly mobilized to cover a transitory fall in income (table 2). By the end of 1989, for example, the stock of foreign currency deposits in household portfo- lios amounted to more than 45 percent of GDP and almost three times the stock 76 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 1. Real Wages in Poland, 1989 and 1990 Month Index of real wages Wages in and year (January 1989 = 100") U.S. dollarsb 1 989 January 100 20 March 146 38 June 125 26 September 132 27 December 138 82 1990 January 79 67 February 73 76 March 99 105 April 84 97 May 77 91 June 76 93 July 81 105 August 83 110 September 86 119 October 92 135 November 98 146 Note: Data are for average monthly wages and include profit bonuses. a. Deflated by cPI inflation. b. Average wage in zlotys converted into U.S. dollars at the free market exchange rate. Source: Authors' calculations based on data from Central Statistical Office (1990) and National Bank of Poland (1990). of domestic money, and in February 1990 it amounted to more than 60 percent of the M3 money supply. Second, although the fall in real wages is substantial (about 30 percent), the index number that measures this change does not take into account the widespread shortages in Poland in 1989. Therefore the effective real wage (or the real-real wage, as some Polish economists like to call it) in 1989 could have been much lower than the one implied by available statistics. For instance, if measured in terms of dollars-using the black market exchange rate as a (undoubtedly rough) proxy for shortages in the goods market-wages increased substantially during 1990 (see table 1). Third, other things being equal, a fall in workers' incomes should be associated with a rise in firms' profits by the same amount. In socialist economies workers are shareholders, either directly as recipients of profit bonuses or indirectly as beneficiaries of state surpluses. Thus one should expect a less pronounced reaction to a wage cut than if wage income were the only source of noninterest income of the household sector. The household-demand view overlooks the importance of the supply shock provoked by the January increase in administrative input prices. As shown in table 3, the magnitude of the increase in input prices was extraordinary. Further- more, at the beginning of 1990 credit to firms (plus domestic liquid assets held by firms) per unit of cost fell by about 60 percent compared with its average during 1989. Another indication of the liquidity squeeze of state enterprises is that wages at those firms stood below the program's ceilings for the first quarter of 1990. Table 2. Deposits of the Household and Enterprise Sectors in Poland, December 1989-August 1990 Household sector Enterprise sector Total Foreign currency Total Foreign currency Month deposits Foreign currency depositsa as a percentage deposits Foreign currency deposits- as a percentage and year (trillion zl.) Trillion zI. Billion US$ of total deposits (trillion zl.) Trillion zl. Billion US$ of total deposits 1989 December 8.6 36.5 4.9 80.1 7.8 15.5 2.3 66.5 1990 January 11.0 45.2 4.8 80.4 15.2 14.1 1.S 48.1 February 13.6 46.3 4.9 77.3 17.0 11.2 1.2 39.7 March 16.5 47.8 5.0 74.3 19.7 9.8 1.0 33.2 April 18.3 49.3 5.2 72.9 22.7 8.6 0.9 27.5 May 20.3 SO.4 5.3 71.3 27.6 7.6 0.8 21.6 June 22.7 50.1 5.3 68.8 31.4 6.7 0.7 17.6 July 25.2 51.0 5.4 66.9 37.9 6.2 0.6 14.1 August 27.6 52.4 5.5 65.5 44.3 5.9 0.6 11.8 a. Foreign currency deposits are valued at the free market exchange rate. Source: National Bank of Poland (1990). 78 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 3. Changes in Administrative Input Prices in Poland, 1986-90 October January 1986a 1987a 1988, 1989b 1990b Natural gas to industrial users 100.0 Crude petroleum and productsc 31.2 39.0 57.0 80.0 - Hard coal to industrial users 22.1 36.7 60.4 100.0 400.0 Electricity to industrial users 18.2 36.6 50.0 142.0 300.0 Overall producer price 17.8 26.6 59.8 43.9 109.8 - Not available. a. Percentage change over previous year. b. Percentage change over previous month. c. Heavy oil for 1986-88. Source: Central Statistical Office (1990). There are two key financial aspects of the January stabilization program. One relates to the initial conditions characterizing the enterprise sector at the outset of the program. In this regard, we stress the importance of the interfirm credit market, which had flourished in previous years to reach a stock value of about 45 percent of GDP at the end of 1989. The other key aspect relates to a funda- mental change in the role of the central bank. In particular, the January program sent a clear signal that the central bank was withdrawing from its role as lender of last resort. This is a fundamental institutional change, which may have strong consequences on the functioning of the interfirm credit market. Before this institutional change took place, interfirm credit received the implicit full backing of the central bank. If a firm ran into financial difficulties and could not service its debt to other firms, the central bank provided the necessary funds. In the new regime, by contrast, the central bank is expected to be much less accommodat- ing. As a result lenders now have much stronger incentives to learn about the financial health of potential borrowers. Information about creditworthiness is likely to have been scant at the begin- ning of the plan for two reasons. First, when there was backing from the central bank, information about borrowers' creditworthiness was much less valuable to the lender. Second, the new rules of the game were not well known, and, in particular, it was too early to know which firms would survive and which ones would have to be phased out. As a result private credit markets may have been especially slow to react to the credit squeeze, which may have magnified the negative impact of the credit crunch on output and employment and contributed significantly to the persistence of these adverse effects. II. FINANCIAL MARKETS IN PLANNED SOCIALIST ECONOMIES: AN OVERVIEW The financial system of classical centrally planned economies may be simply characterized as being composed of highly separated monetary circuits: house- hold money and enterprise money. Enterprise money is passive in the sense that monetary flows in the enterprise sector adjust automatically to planned real flows. The central planner sets targets on real flows and, accordingly, extends credit to enterprises for their needs, taking into account administered wages and Calvo and Coricelli 79 prices. In practice the banking system is monolithic, with the central bank directly administering all the monetary transactions of enterprises. Money tends to be active in the household sector in the sense that budget constraints precede household actions (consisting mostly of consumption decisions) in the manner assumed by standard microeconomic theory. In a reformed planned economy, enterprises gain some independence about their decisions. They have some degree of freedom in determining wages, prices, input demand, and investments. Enterprise money can still be passive if the central bank accommodates all the actions of the enterprises, but credit accom- modation is not a necessary feature of the system; it is, in principle, a policy decision of the central bank. Because bankruptcy is ruled out, however, attempts to curtail central-bank credit are quickly followed by a compensatory expansion of interfirm credit. The compensatory adjustment is relatively quick to material- ize because interfirm credit bears little risk. The central bank is de facto forced to operate as lender of last resort (Kornai 1980) and is unable to control the supply of liquidity. As a result, stabilization programs tend to fail. Moreover repeated attempts to lower inflation may solidify the interfirm credit market. Thus stabilization programs launched under these no-bankruptcy conditions may induce a higher inflation plateau in the future and reduce even further the ability to control the supply of liquidity. Reformed planned economies still tend to be characterized by strongly seg- mented credit markets and very weakly integrated monetary circuits of house- holds and enterprise sectors. However, unlike centrally planned economies, the development of a thick interfirm credit market allows state enterprises in re- formed planned economies to escape official credit ceilings. The ability of the firms to circumvent credit ceilings results in a loss of monetary control by the monetary authority. Interfirm credit plays an important role in the reformed planned economies of Hungary (since the 1970s), Yugoslavia, Poland, and China. In Yugoslavia the stock of interfirm credit was estimated at 43 percent of gross social product in 1987; in Poland at around 48 percent of GDP at the end of 1989. In China the phenomenon is particularly relevant, and in the package of policy measures implemented by the Chinese government in 1989, the solution to this problem was deemed as one of the main targets. It is estimated that, in 1990, arrears in payments of interfirm credit amounted to about 20 percent of the working capital of all state enterprises, and arrears in payments of working capital credits from the banking system were of analogous size. III. THE ENTERPRISE-SIDE VIEW OF STAGFLATION Despite its oversimplified structure, the enterprise-side explanation appears able to account for the type of stagflation that occurred in Poland after the stabilization program was implemented. The recent stagflation may be partly due to the substantial increase in input prices, which may have been magnified by the presence of segmented credit markets. In particular there seems to be a 80 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I severe segmentation between the households and firms. A model of the enterprise-side view of stagflation is presented in detail in the appendix. In a more realistic model some of the fall in output could actually reflect an efficient reallocation of resources. For example, before the stabilization pro- gram, firms may have overaccumulated inventories and taken advantage of the highly negative real interest rates. Thus the fall in output of firms producing inventory may actually lead the system to a socially better equilibrium. Conse- quently, the welfare implications of a credit crunch require further scrutiny (Calvo and Coricelli 1991). Model Assumptions Consider an economy of identical (Ramsey-type) individuals whose utility function depends on present and future consumption streams. Households are assumed to have perfect access to international capital markets, where they can freely borrow and lend at the international rate of interest. This is an extreme assumption; however, it corresponds to a situation not too different from the one prevailing in Poland. Polish households carry in their portfolios a sizable stock of foreign currency deposits. Therefore at any time they are free to lower those balances to buy, for example, durable goods. This is equivalent to borrow- ing in international markets because-like borrowing-it results in a fall of the net financial position of Polish households in relation to the rest of the world. Furthermore, if those funds are held in accounts yielding the international risk- free interest rate (which is approximately the case in Poland for dollar- denominated bank deposits), then the cost of those "newly borrowed" funds corresponds to the interest rate assumed in the model. Firms are assumed to produce a tradable consumption good by means of an imported intermediate input (for example, coal or oil) and a fixed amount of labor. Firms are price takers but wage makers; firms are controlled by workers' councils, which are assumed to maximize the present discounted value of wages (evaluated at the international interest rate). This is a natural objective of firms in the present context, given the previous assumption that workers have perfect access to international capital markets. The key assumption of the model, how- ever, is that firms face a constraint of needing liquidity in advance to finance intermediate inputs. This means that in order to finance a given flow of expendi- ture, firms must secure a stock of liquidity that must bear a given "technical" relationship to the flow of expenditure. Liquidity at a given firm is defined as the money it holds plus credit lines from the central bank and suppliers available to the firm. In this fashion, firms face a liquidity constraint that has to hold at each point in time. The Case When Firms Are Unable to Borrow In the simple case in which firms are unable to borrow from either the central bank or the private sector (that is, other firms and households), the liquidity of the firms is entirely given by the stock of money they hold. At each point in time the stock cannot be changed unless the firms sell inventories (which is ruled out Calvo and Coricelli 81 in this simple version of the model). Thus, if firms start from a position in which the liquidity constraint is binding, then a one-step rise in the price of the input may provoke a liquidity crunch and an immediate output loss. The output loss occurs irrespective of what happens with output prices. This observation shows that a liquidity crunch can develop even when output prices increase in the same (or larger) proportion as input prices. Firms will try to offset the liquidity crunch by drastically lowering real wages until a ]level of liquidity consistent with full-capacity utilization is restored. Consequently, output will fall on impact and then steadily climb back to its higher, long-run level. Profits-which are equal to the accumulation of real monetary balances held by firms-will display a dramatic increase after the credit crunch. Thus, in the midst of this recession driven by the credit crunch, profits of firms will be spectacular. A fall in real wages is consistent with the Polish experience in early 1990 (see table 1), and it helps to explain why wage ceilings were not binding at the beginning of the program (Frydman and Wellisz 1991). The high profitability of enterprises could easily lead to the wrong conclusion that there is no credit problem. However, high profitability actually reflects the attempt by firms to relieve the credit crunch. Consequently, it would be some- what unwarranted to interpret the unexpectedly high profitability of the social- ized sector in Poland during January-April 1990 as a clear indication that credit does not lie at the heart of the recession. The fall in output implies a decline in permanent household income, which is, however, smaller than the fall in transitory household income. Income is ex- pected to return to its full-employment level. Thus, since consumption is deter- mined by permanent income, this version of the model predicts an initial trade balance deficit, as the decline in production is, on impact, larger than the decline in consumption. However, in Poland following the stabilization program there was a marked improvement in the trade balance. In 1990 it is estimated that the trade balance showed a surplus of US$2.7 billion compared with US$0.2 billion in 1989. The model can easily be modified to account for a temporary improvement in the trade balance. For example, enterprises may hold sizable stocks of invento- ries at the start of the program-partly, perhaps, in anticipation of the forthcom- ing liquidity crunch. Thus, in response to the liquidity crunch, firms are likely to react by using up their own inventories and lowering their demand for new intermediate goods. Therefore the liquidity crunch could lead to a reduction of total absorption (consumption by households and absorption by firms) that exceeds the fall in output, yielding a surplus in the balance of trade. In this case the fall in output is reduced because inventories act as a buffer stock. The Case with Heterogeneous Firms The model can be further enriched by assuming that firms are heterogeneous and that some firms produce intermediate goods used as factors of production by other firms in the economy. In a reformed planned economy, firms have 82 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I incentives to lend to one another in the expectation that the monetary authority will bail out firms that develop financial problems. Therefore, if the reformed planned economy launches a stabilization plan similar to the one in Poland in 1990, the economy is likely to start the program with a positive-and possibly large-stock of interfirm debt. In such a framework, consider once again the effects of a rise in the price of primary imported inputs. In addition to the adjustments implied by previous versions of the model, it is now possible for a debtor firm to alleviate the credit squeeze by falling into arrears with other firms. The cost of falling into arrears depends on the effectiveness and credibility of bankruptcy regulations. If such regulations are believed to be less than fully effective, then firms are likely to find it attractive to fall into arrears with other firms. This, in and of itself, will increase the riskiness of the interfirm credit market and dry up sources of funds for solvent and efficient firms. Furthermore in the short run the buildup of arrears plays against firms that are net creditors at the start of the stabilization program. This is worrisome, because there are no a priori reasons to expect the more efficient firms to emerge unscathed from these financial squabbles. The Official Banking Sector The assumption of no official bank credit is, of course, unrealistic. However, the model could easily be extended to incorporate an official banking sector that takes deposits from households and lends to enterprises. This sector played a prominent role in Poland and was the main source of the easy money that prevailed until recently. By setting credit ceilings for the entire banking system, the Polish stabilization program limited the amount of credit that banks could offer to enterprises. In addition the refinancing rate (that is, the interest rate charged by the central bank to the other banks in the system) was raised consid- erably during the first few months of the program (table 4) and became highly positive in terms of foreign exchange. Given the imperfections of the banking sector-and, in particular, the above-mentioned segmentation between house- Table 4. Interest Rates in Poland, December 1989-December 1990 Refinancing Average prime rate Year Month rate on loans 1989 December 11.7 14.1 1990 January 36.0 46.5 1990 February 20.0 22.0 1990 March 10.0 10.5 1990 April 8.0 8.5 1990 May 5.5 6.3 1990 June 4.0 4.5 1990 July 2.8 2.9 1990 August 2.8 2.9 1990 September 2.8 2.9 1990 October 3.6 3.4 1990 November 4.6 4.2 1990 December 4.6 5.1 Source: National Bank of Poland (1990). Calvo and Coricelli 83 holds and enterprises-the central bank's increase of its interest rate induced banks to offer lending interest rates well above the refinancing rate. This interest rate policy seems to account for the fact that the credit ceiling was not binding in early 1990. The excess supply of credit may appear to contradict the argument of a credit crunch. However, the fact that the banks' interest rates were so closely geared to the refinaoncing rate of the central bank demonstrates that domestic credit mar- kets are highly imperfect. Therefore it is perhaps a good approximation to envision the central bank as a credit monopolist able to set the price of credit. Thus, by raising the refinancing rate, the central bank would elicit a lower demand for credit and, furthermore, could even ensure that any credit ceiling would not be binding (assuming that firms would not expect to be fully bailed out). Under these circumstances the central bank could generate a credit crunch either by drastically lowering credit ceilings or, alternatively, by sharply increas- ing the refinancing rate. Consequently, the view that credit ceilings were too tight is not essentially different from the view that interest rates were set too high. The latter view is equivalent to saying that the virtual credit ceiling (as opposed to the actual credit ceiling) was too tight. Credit tightness-measured in terms of bank credit-seems to have eased in the second half of 1990. However, output has responded much less to credit expansion than to the credit contraction of the first half of the year. This is a puzzling development for the simple model outlined above and in the appendix. One explanation may be that the objective of firms is to maximize current wages rather than permanent wages, as postulated in the model. Thus, the easing of credit conditions would lead firms to offer higher wages instead of using the proceeds to relieve the liquidity squeeze. This seems consistent with the sharp increase in real wages in the second half of 1990 (Calvo and Coricelli 1991). Another explanation is that a credit crunch signifies a fundamental shakeup of interfirm relationships, leading firms to fall into arrears with one another and to break all sorts of explicit and implicit contracts. In such an environment, there- fore, a later expansion of credit may not be able to mend the damage caused by the initial credit crunch, thus leading to a slower recovery. Furthermore some of the arrears with other firms will be repaid. Therefore easing up bank credit would lead to a repayment of debts to both workers and suppliers, rather than an expansion of output. A measure of credit conditions that encompasses inter- enterprise credit as well would be more significant in this context, as bank credit and interenterprise credit would be substitutes for each other. Some evidence of this ability to substitute for each other is given in figure 3, which shows that the expansion of bank credit in the third quarter was matched by a contraction of interenterprise credit, leaving total credit for enterprises roughly unchanged. IV. SOME IMPLICATIONS OF THE ENTERPRISE-SIDE VIEW A credit crunch is likely to happen if the stabilization program starts with a large devaluation of the currency and removal of subsidies on primary (tradable) 84 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Figure 3. Interfirm Credit and Bank Credit in Poland, 1990 (December 1989 = 100) 100 _ Bank credit for working capital 80- Inerfirm credt.................... 60 - 4 0 I - _I I I I I1 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Note: End-period stocks deflated by the consumer price index. Source: National Bank of Poland (1990). commodities. If excessive credit tightness occurs (that is, if the fall in output exceeds what can be accounted for by considerations of efficiency) there are, in principle, three types of solutions: appreciate the currency, phase out the input subsidies more gradually, or increase central bank credit to the state enterprise sector. Appreciating the currency is the most attractive solution, because it would also lower the cost of primary inputs-thus directly relieving the credit crunch-and could most easily be presented to the public as a technical adjust- ment of the stabilization program. The other two solutions may be taken as signals that the government is weak and ready to abandon the stabilization plan. If central bank credit to the state enterprise sector is increased, the liquidity of firms should undergo a stock jump. If the increase occurs only in the flow of credit to the enterprise sector, there is likely to be a suspicion that the govern- ment has gone back to its old inflationary practices. In contrast, the stock increase need happen only once and could possibly be disguised under a cur- rency reform. A stabilization program should consider initial conditions in the credit mar-i ket. In this regard the initial stock and distribution of interfirm debt is of Calvo and Coricelli 85 paramount importance. Special attention should be given to avoid transmitting the financial difficulties of inefficient enterprises to efficient enterprises. Demand-side solutions may be counterproductive. Forcing firms to pay higher wages exacerbates the credit crunch directly, and increasing government demand for goods and services leaves output unchanged and is likely to worsen the balance of payments. There were some indications of this phenomenon in Po- land at the end of 1990, when the easing up of fiscal policy was accompanied by both a worsening balance of payments and an accelerating inflation. Proposals for privatization are generally based on considerations of micro- economic efficiency. The enterprise-side approach suggests that, when foreign companies are involved, privatization can play a fundamental role even during stabilization programs. Foreign companies are likely to have better access to international capital markets, which would help to relieve tight domestic credit conditions. Clearly, besides the implementation of privatization schemes, the development of a domestic banking system has enormous relevance. This will indeed facilitate the transfer of funds among firms and between households and firms, especially when stabilization policies are credible (Calvo 1989). APPENDIX. A MODEL OF THE ENTERPRISE-SIDE VIEW OF STAGFLATION The central formal ingredients of the model discussed in section III are sketched here. For simplicity, we will focus on the case in which all goods are tradable and there are no stocks of inventories. (Nontradable goods and other extensiorLs are discussed in Calvo and Coricelli 1991.) Households Households are assumed to be homogeneous and to have a utility function: (A-1) u(c,)e-rldt, r > 0 where u is a strictly concave and twice-differentiable, instantaneous, utility function; c denotes consumption of the representative household; and r is its constant subjective rate of discount. The household's budget constraint is given by the fol[lowing standard expression (Calvo 1986): (A-2) j cte-rtdt = ao + | (wt + gt - it mt)e rtdt where a, w, g, and m denote real (that is, in terms of consumption) financial wealth, labor income, government lump-sum transfers, and holdings of mone- tary balances corresponding to the representative household, respectively; more- over, i and r denote, respectively, the domestic nominal interest rate and the real (constant) international interest rate. It is assumed that the international real 86 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I rate of interest, r, is equal to the subjective rate of discount in expression A-1. Financial wealth, a, satisfies (A-3) at = mt + bt where b is the stock of international bonds held by the representative household (for example, dollar-denominated deposits). The consumer is subject to a cash- in-advance constraint that takes the following form: (A-4) m, 2, act, a > 0. The representative household is assumed to maximize its utility subject to its budget constraint; the liquidity constraint given by expression A-4 and its initial financial wealth, a, with respect to the path of consumption; and real monetary balances. If the government credibly announces that the exchange rate will be kept constant forever, then, by the assumption of perfect capital mobility, the domestic nominal interest rate equals the international rate. Therefore the solu- tion to the above optimization problem calls for setting consumption at a con- stant level, and the liquidity constraint is binding. Firms Firms do not have access to the international capital market. They produce tradable consumption goods by means of a tradable primary input, n, and labor. For simplicity, labor employment at each firm is fixed and equal to one. Each firm is subject to a cash-in-advance constraint of the form: (A-5) yptnt ' z, -y > 0 where p denotes the price of the intermediate factor of production in terms of consumption, and z stands for real monetary balances held by the firm. Variable p is determined by the exogenously given international relative price of the primary good in terms of consumption ("small country" assumption), adjusted by local tariffs and taxes. Therefore output net of direct intermediate-input cost is given by (A-6) if(n) -pn =- 0(n; p) where T(n) denotes the production function, which satisfies standard regularity conditions, and k can be interpreted as the "net" production function. Net output is denoted by y; hence, (A-7) yt=- (t, Pt). We assume, without loss of generality, that the liquidity constraint is binding (that is, -ypn = z). In addition, we assume that the firm can only accumulate money balances. Therefore, assuming zero international inflation, the change in real monetary balances held by the firm is (A-8) Zt = O(z,/P; P) - Calvo and Coricelli 87 where co is labor income (or the wage) offered by the representative firm. More- over, to simplify, the relative price of the primary input in terms of consumption, p, is assumed constant over time. The firm is controlled by a workers' council and therefore follows policies intended to maximize its workers' utilities. Thus, because workers (households) are assumed to have perfect access to international capital markets, the workers' utilities are optimized by maximizing the present discounted value of the workers' income. More specifically, the representative firm's objective function is (A-9) w, ce-r,dt. Therefore the representative firm's problem is simply to maximize the present discounted value of its employees' incomes by choosing the path of the wage rate, subject to the flow-budget constraint (equation A-8) and initial real mone- tary balances, zo. This is a linear optimization problem that has a "bang-bang" type of solution (Shell 1967). Let z = z_(p) be such that it satisfies the following equation: (A-10) 0 n(z/p, p)/p = r. Hence z. is the level of real monetary balances such that the net marginal productivity of real monetary balances at the firm equals the international rate of interest. If z0 = z., then it is optimal to set z, = zo for all t (that is, z_O is the firm's optimal steady state). However, if z0 - z., then co will be set as large as possible (in absolute value) to reach z_ in the shortest period of time. If no bounds are put on wages and if initial real monetary balances are larger than the steady-state optimum, then the firm should give an initial bonus to employees equivalent to the difference between the initial and optimum levels. Otherwise, if initial real money balances are less than the optimum level, the firm should order workers to bring to the firm's vault a stock of cash equivalent to the difference. In such a liquidity crunch an initial transfer would be equivalent to a loan from workers to the firm yielding a rate of return larger than the international interest rate. The amount of the transfer is optimal, because, by equation A-10, the marginal return of liquidity at the firm equals the opportunity cost of loan- able funds. In this fashion, therefore, the firm operates as if it had perfect access to international capital markets. To capture credit segmentation in a simple manner, we will assume that the wage rate, co, cannot fall below a well-defined minimum wage, co > 0 (the minimum wage is assumed to be smaller than long-run optimal output, that is, X < X (z. /p, p)). Under this amendment the optimal policy for the firm remains the same as in the previous case (in which co is subject to no bounds) if initial monetary balances at the firm, zo, exceed the long-run optimum, z_. However, 88 THE WORLD DANK ECONOMIC REVIEW, VOL. 6, NO. I if the firm starts on the liquidity-crunch region (that is, zo < z.), then it is optimal to set the real wage, w, at its minimum permissible level, , until real monetary balances at the firm, zo, attain their optimal long-run level, z<. Thus, in symbols, along an optimal path we have (A-1i1) if z, < ZO, then co = w. When firms are strapped for funds, they "borrow" from their own workers. This is not an unusual phenomenon in socialist economies. Consider now the effect of a once-and-for-all devaluation of the currency, assuming that, beforehand, firms were at their long-run optimum liquidity, z,. A devaluation raises the domestic prices of inputs and outputs in the same proportion-hence, leaving p unchanged. Thus z. remains the same, but now zo < z., that is, enterprises are subject to a liquidity crunch. By the above analysis, output falls, and firms react by lowering the real wage until liquidity returns to z_. Raising the relative price of the primary good, p, always leads to lower consumption output, but its effect on long-run optimal liquidity, z., is ambiguous and will not be pursued here. During this process, output (and net output y) recovers monotonically to the original steady state. Consequently, the firm's optimal response satisfies all the conditions discussed in the first part of section III. Government To close the model, we assume that the government consumes nothing and returns to households its entire fiscal revenue (net of interest payments). There- fore, as in Calvo (1986), one can show that (A-12) c = r (fo + | yte-rtdt) where f is the country's overall stock of international bonds (which is the sum of the international bonds held by households and those held by government). Thus, recalling that y stands for output net of primary input costs, equation A-12 simply states that consumption equals "permanent" income. The latter, in turn, equals the return on total domestic assets: international bonds plus the present discounted value of net domestic output of consumption goods. In the liquidity-crunch experiment in which all nominal prices are raised in the same proportion, we showed that y will initially fall and then steadily recover to its original steady-state value. Thus, by equation A-12, consumption falls on impact, but it does so by an amount that is less than the fall in initial output, yo. Consequently, the trade balance initially deteriorates, as stated in the text. Bank Credit The model can easily be appended to allow for situations in which bank credit is available and official credit limits are binding. For example, let us denote total Calvo and Coricelli 89 credit and liquidity available to the representative firm by e. Thus (A-13) e, = z, + b, where b denotes bank credit to the representative firm. The cash-in-advance constraint (expression A-5) is now assumed to take the following form: (A-14) ypnC et. Expression A-14 should thus be more properly referred to as a "liquidity-in- advance" constraint or, alternatively, as a "cash-and-credit-in-advance" constraint. The interest rate on bank credit is denoted by rb. Hence, if expression A-14 is binding, equation A-8 takes the following form: (A-15) Zt = (ftlp; p) - rb(ft - zt) - Lgt, for Et 2 Zt. In the previous version of the model we assumed that total liquidity, e, was always equal to firm-owned liquidity, z. Thus, z = e, and the second term of the right side of equation A-15 drops out, yielding equation A-8, as expected. Other- wise total liquidity could exceed own liquidity, that is, e > z, but the firm would have to service the resulting bank debt, giving rise to extra cost equal to rb(2 - z). If bank credit is not binding, the firm is free to borrow as much as desired, and it is optimal for the firm to choose e so as to maximize net-flow revenue as given by the right side of equation A-15. The first-order condition corresponding to such an optimum is (assuming p = 1 for notational simplicity) (A-16) ae(f; i)/d = rb. From identity A-6 and equation A-16-and recalling that the liquidity-constraint (expression A-14) is assumed to be binding-raising the interest rate on bank loans, rb, will result in output contraction and will induce early repayment of bank loans by enterprises (which is in line with developments in Poland during the first stages of the stabilization program). Symmetrically, if the interest rate on bank loans, rb, is lowered, output ex- pands and so does the demand for bank credit by enterprises. For a sufficiently low rb the credit ceiling (denoted by b) will be binding. At that point the own- liquidity accumulation equation (A-15) becomes (A-17) Zt = OI(Zt + b)/p; p] - rbb -t. One can show that during periods of credit rationing, that is, when the credit ceiling, b, is binding, the economy displays the same qualitative behavior as in the first case discussed above, in which, implicitly, the credit ceiling, b, was set equal to zero. 90 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Blanchard, Olivier, Rudiger Dornbusch, M. King, Paul Krugman, Richard Layard, and Lawrence Summers. 1991. Reforms in Eastern Europe. Cambridge, Mass.: MIT Press. Calvo, Guillermo A. 1986. "Temporary Stabilization: Predetermined Exchange Rates." Journal of Political Economy 94 (December): 511-33. . 1989. "Incredible Reforms.' In Guillermo A. Calvo, R. Findlay, P. Kouri, and J. Braga de Macedo, eds., Debt, Stabilization, and Development. Cambridge, Mass.: Basil Blackwell. Calvo, Guillermo A., and Fabrizio Coricelli. 1991. "Stabilizing a Previously Centrally Planned Economy: Poland 1990." Economic Policy, forthcoming. Cavallo, Domingo. 1977. "Stagflation Effects of Monetarist Stabilization Policies." PhD thesis. Harvard University, Cambridge, Mass. Processed. Central Statistical Office. 1990. Monthly Statistical Bulletin. Warsaw. Frydman, Roman, and Stanislaw Wellisz. 1991. "The Ownership-Control Structure and the Behavior of Polish Enterprises during the 1990 Reforms: Macroeconomic Mea- sures and Microeconomic Response." In Vittorio Corbo, Fabrizio Coricelli, and Jan Bossak, eds., Reforming Central and Eastern European Economies: Initial Results and Challenges. A World Bank Symposium. Washington, D.C. Gomulka, Stanislaw, and Jacek Rostowski. 1988. "An International Comparison of Material Intensity." Journal of Comparative Economics 12 (December): 475-501. Kiguel, Miguel, and Nissan Liviatan. 1990. "The Business Cycle Associated with Exchange-Rate-Based Stabilizations." PRE Working Paper 513. World Bank, Country Economics Department. Washington, D.C. Processed. Kornai, Janos. 1980. Economics of Shortage. Amsterdam: North Holland. National Bank of Poland. 1990. Monthly Bulletin. Warsaw. Osband, Kent. 1991. Index Number Biases during Price Liberalization. IMF Working Paper WP/ 91/76. Washington, D.C.: International Monetary Fund. Shell, Karl. 1967. "Optimal Programs of Capital Accumulation for an Economy in Which There Is Exogenous Technical Change." Essays in the Theory of Optimal Economic Growth. Cambridge, Mass.: MIT Press. Taylor, Lance. 1980. "IS-LM in the Tropics: Diagrammatics of the New Structuralist Macro Critique." In S idney Weintraub and William R. Cline, eds., Economic Stabili- zation in Developing Countries. Washington, D.C.: Brookings Institution. van Wijnbergen, Sweder. 1982. "Stagflationary Effects of Monetary Stabilization Poli- cies: A Quantitative Analysis of South Korea." Journal of Development Economics 10: 133-69. . 1983. "Credit Policy, Inflation, and Growth in a Financially Repressed Econ- omy." Journal of Development Economics 13: 44-65. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 91-108 Voluntary Choices in Concerted Deals: The Menu Approach to Debt Reduction in Developing Countries Ishac Diwan and Kenneth Kletzer This article examines the mechanics and attributes of concerted debt reduction agree- ments that offer creditors a choice between exit and relending options. The menu approach sets prices for different choices that implement a decentralized equilibrium. When banks can commit to choose from the menu and are not allowed to free ride, a menu can be designed that assures that the price paid for debt repurchased is equal to the marginal value of the debt claims. This can be achieved by taxing the gains that accrue to nonexiters with a request for new money. The equilibrium amount of debt reduction rises when the new money request is increased. The importance of banks' heterogeneity for menus to dominate simple concerted buybacks and the case in which debt reduction can be financed by loans from international financial institutions are discussed. A debtor country would gain by repurchasing some of its outstanding debt if the price it pays is low enough. Similarly, there is some (high enough) price above which its creditors would gain by selling some of their debt claims. For a Pareto- improving debt reduction plan to exist, these two price ranges must overlap. When a Pareto-improving exchange is possible, finding a mechanism that imple- ments such trade is an important remaining issue. In the case of country debt, market buybacks and concerted debt reduction are unlikely to do the job. If debt is repurchased on the secondary market, then the price that must be paid is the equilibrium price of debt claims after debt reduction (Dooley 1988). This is because a market buyback by the debtor leads to a rise in the secondary market price. Since any lender can choose between selling and retaining debt claims, the opportunity cost for selling is the market price after the debt reduc- tion. When lender participation in a buyback is voluntary, both the price paid for repurchased claims and the market price of remaining debt rise, and all creditors realize a net benefit at the expense of the debtor. The debtor is gener- ally better off not participating in a market buyback if the funds to be used to Ishac Diwan is with the International Economics Department of the World Bank, and Kenneth Kletzer is with the Economic Growth Center of the Department of Economics at Yale University. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 91 92 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I repurchase debt are available to it for other purposes (such as for domestic investment or consumption). Concerted agreements can be designed so that creditors commit to a coordi- nated debt reduction plan that is Pareto improving. Ideally, each creditor bank would sell a specific share of its claims at some price between the pre-debt reduction price and the expected post-debt reduction price. In practice, it may be difficult to bar individual banks from free-riding by not selling the agreed share. The problem of free-riding is exacerbated by the presence of heterogeneity across banks. If creditors differ in their relative valuations of country debt, a concerted buyback that does not discriminate between banks, and that at the same time hurts no bank, must occur at the reservation price of the bank with the highest valuation. With unobservable heterogeneity between creditors, the ability to discriminate between creditors in a concerted agreement will be lim- ited. Offering a menu of choices could still achieve Pareto improvement by allowing creditors to self-select, with only those with low valuations selling out at a particular offer price. Recently the menu approach has been used to reduce the external debt over- hang of some highly indebted countries. An agreed upon menu is a contract, which may be partly implicit, establishing a future opportunity set for the lenders. The menu approach requires that lenders can commit to choose from among a restricted set of actions ex post. The possible means for enforcing a menu include creditor country legal and political institutions and agreements between creditor nations. By agreeing to the contract, lenders face the possibility that they will be penalized for reneging ex post by refusing to allocate their existing debt claims across a portfolio using the menu options. While private creditors restrict their options ex post, they may be able to raise their net worth in negotiations. That is, lenders can increase the value of their loan portfolios ex ante by agreeing to choose from a menu of restricted options ex post. I. THE MENU APPROACH The menu approach to debt reduction combines characteristics of voluntary market buybacks and concerted debt reduction agreements. The menu approach thus retains the advantages, but not the inconveniences, of the other two mecha- nisms. The options on the menu and their relative pricing are negotiated first; in a second round each creditor freely chooses a preferred option. Overall, the discrimination allowed by the menu leads to cheaper debt reduction (Diwan and Spiegle 1990). This article analyzes the mechanics of simple debt-reducing menus that in- clude exit and relending options. Several types of exit options can be offered to fit the regulatory and tax characteristics of different types of banks. But often some banks would require too high a price to exit irrespective of the exit vehicle. Therefore a nonexit option is included in the menu in order to achieve agree- ment on a low exit price. The creditors that do not choose to exit get a capital Diwan and Kletzer 93 gain following the reduction in debt outstanding (because the price of debt increases when the debt stock decreases). Unless this gain is taxed, the nonexit option would dominate the exit option, and no exit would occur freely. This problem can be mitigated by imposing a capital gains tax on those banks that choose the nonexit option. While explicit taxation of creditors' gains may be politically or legally infeas- ible, there may be alternative schemes that effectively tax these windfall gains to the nonexiting debt-holders. Such a "tax" can be set by requiring lenders to extend new loans in proportion to the debt they retain. A commitment to provide new loans in proportion to retained debt creates a wedge between the price paid to repurchase debt and the secondary price of outstanding debt after the stock of debt has been reduced. In recent menu agreements, the options have included some form of buybacks and of new loans. When the new debt claims trade at a discount, a portion of any new loan represents such a tax. That is, if a dollar is lent and the secondary price is 60 cents, then the new debt claim is worth 60 cents and the rest is a tax with the revenue accruing to the debtor. The menu approach does not require assignment of particular choices from the menu to each lender. By offering a set of prices for different options, the menu leads to a decentralized equilibrium level of debt reduction ex post. There is a unique level of equilibrium debt reduction that corresponds to a given set of options because, in equilibrium, some banks would have to be indifferent be- tween the two options. The equilibrium amount of debt reduction is affected by the terms of the menu. When the new money call is set higher, more banks will prefer the exit option. The ex-post debt price will then increase, making the new money option more desirable. For the menu approach to debt reduction to create gains over and above those available with simple concerted buybacks, the offered options must add to existing market trading opportunities. In general the use of the menu approach allows for more gains when the secondary market for debt is inefficient, either because of incompleteness, distortionary public policies, or large transaction costs. When the cost of debt reduction is (partly) financed by new international loans, the results depend on whether the new loans share in the same seniority class as the old loans. II. A DEBT VALUATION MODEL AND THE MENU APPROACH Sovereign borrowers choose to service external debt obligations because it is in their enlightened self-interest to do so. Under sovereign immunity, creditors do not have recourse to an international court to assure settlement of debts through a direct lien on the country's assets as they would in the case of an insolvent domestic client. The threat of the imposition of sanctions in the event of nonpayment or insufficient payment provides sovereigns with the incentive to make debt payments. In addition to denials of future official aid flows, the penalties for default include restrictions on future trading opportunities on inter- 94 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I national markets for the debtor. These include the disruption of commodity trade, suspension of trade preferences, and reduced. access to international fi- nancial markets. Because the repayments made by a sovereign debtor are linked to the expected present value of the social cost of sanctions, they need not be strictly proportional to the total outstanding contractual debt burden. More- over, as the probability of sanctions being exercised (or being used by creditors as a threat to extract higher net repayments) increases, debtor countries can reduce the effectiveness of sanctions, for example, by underinvesting or by shifting resources to the nontradable and to the import-competing sectors of the economy. A simple model summarizes the various arguments made by the recent debt literature on the effect of market buybacks and contrasts those with the effects of menu-driven debt reduction. The model-a system of equations-can be used to determine how much debt reduction can be achieved with a given amount of resources or how much resources would be required to reduce debt by a given amount. A Simple Model of Debt Valuation We assume that the present value of expected debt service payments is an increasing and concave function of the nominal debt burden. In other words, the value of debt is a smooth and concave function (f(D) in figure 1) of the amount of outstanding total debt claims, D. In this analysis we consider only the case in which the marginal price of debt is positive: the debtor is on the increasing part of its debt value curve. But if the debt value curve is first upward, then down- ward sloping (as in the lower half of figure 1), the arguments below are un- changed. The debt value curve includes the effect that the debt burden has on the investment undertaken by the debtor; if debt reduction raises investment, the increase in the expected present value of repayments made on outstanding debt is included in the debt value curve. If creditors sell a portion of the initial debt they hold, the secondary market price of debt will rise from po in equation 1 to p1 in equation 2: (1) po=f(DO)IDO (2) Pi =f(Do - a/p1))/[Do - (aP1)] where po is the initial debt price, Do is the initial stock of debt, Pi is the new secondary market price of the reduced debt, and a is the amount of funds used to purchase debt. Because the debt value function is concave, the new secondary market price of debt exceeds the initial debt price for all positive values of debt reduction. The ex-post value of creditor assets (f(D1)) is: (3) f(D1) = Pi (Do - a/pl) + a = p1D0. The ex-post value of creditor assets is greater than the original value of debt claims because the new debt price is greater than the initial debt price. The capital gain (-y) realized by the creditors retaining claims is Diwan and Kletzer 95 Figure 1. 7he Debt Value Function and Equilibrium New Money Debt value 450 Initial debt price po (f(D)) _ f i ( D~~~~~~~() ;Debt stock (D) New money (N) fL(D)=p, Do Debt f'(D)=p0 ~~~~~~stock (D) (4) y= (p1 -Po) (Do -a/p1) > 0. All creditors realize a capital gain when the price of debt increases, whether or not they have sold a portion of their claims. Automatic Equilibrating Mechanisms in Simple Menus When contracts that bind the actions of private creditors are feasible and credible, the conversion of the status quo debt claims to a package of assets that has expected present value at least as great can be negotiated. It is assumed that 96 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I enforcement of an agreement by private creditors with their home governments and international financial institutions is possible, but that creditors do not need to accept a contract that reduces the present value of their assets. Once an agreement is reached, lenders face penalties if they refuse to perform according to the contract (in the recent Mexico deal, for example, creditors that did not participate were effectively made junior). If enforcement of a contract is possible only for a subset of the creditors, then a debt reduction menu can still be negotiated. However, the nonparticipants free ride, thus realizing a net increase in the value of their portfolios. The participating creditors can increase the value of their assets despite providing pecuniary external benefits to the nonpartici- pants. For simplicity, it is assumed that the entire stock of outstanding claims is represented in the negotiations. Suppose that the creditor banks and the debtor country have agreed on a simple menu of options represented by (p,n), in which, for each dollar of claim they hold, banks can choose to either exit at price p or to reschedule the loan and relend n dollars in addition (where n is a percentage of the debt they retain). With no loss of generality, we assume that new and old loans have equivalent terms. Given a simple menu, can we tell how much debt reduction and new money will actually be achieved? For simplicity, the equilibrium analysis below is developed for the case in which all banks are similar. The more general case with heterogeneous banks is discussed in section IV and more fully in Diwan and Spiegle (1990). After the deal is completed, debt prices are expected to be higher. A bank that relends a portion of its debt will have its old (rescheduled) claim revalued and will realize a capital gain as indicated in equation 4. However, its new claim will be valued at the secondary market price, implying a capital loss (equal to 1 - p1 per unit of new debt). Thus the opportunity cost for holding a unit of debt back from repurchase at the price agreed in the menu depends on the increased secondary market price and the amount of new money loaned. The opportunity cost, C, of holding a unit of debt back from repurchase at the increased second- ary market price is given by the following: (5) C = p1 (1 + n)-n. The opportunity cost of holding a unit of debt is equal to the exit price when the ex-post debt price satisfies the no-arbitrage condition in equation 6: (6) p1 = (p + n) / (1 + n). When the ex-post secondary market price of debt is expected to exceed the right side of equation 6, the new money option is preferred to the exit option. Banks would rather hold onto their debt than sell at an exit price of p because the capital gain on old claims is greater than the capital loss on agreed (by the menu) new lending. Thus less debt will be sold and more new money offered, resulting in less debt reduction. This leads to an increase in debt stock and thus to a reduction in the ex-post price. Because creditors are price takers when they optimize ex post and because the expected present value of the debt function is Diwan and Kletzer 97 strictly concave, the solution to portfolio value maximization by creditors is unique. Equation 6 must hold in equilibrium. When "sufficient" funds are available for debt reduction, competition among banks will ensure a unique equilibrium. The menu approach does not require that the ex-post choice of each creditor from the menu be negotiated and spe- cified or even that aggregate constraints be placed on the total amounts of debt reduction and new money. Prices, that is, the offer of the initial debt price for debt bought back, and any ratio of new loans to remaining debt claims required to obtain that price are sufficient. For a given amount of resources for debt reduction, we can calculate by how much debt stocks will be reduced. Given the initial debt stock, debt value function, and menu (exit price and relending ratio agreed on), equations 6 to 9 can be solved for new debt stock, amount of debt reduction, and new lending. (7) D1=Do-B+N (8) Pi = f(Di) / Di, i= 0,1 (9) n = N/(D1-N) where Di denotes debt stocks (subscript i = 0 indicates the level of the variables before the debt agreement and subscript i = 1 indicates the level after the deal is completed), B denotes the amount of reduction of debt stocks, and N denotes new lending. Lenders choose between the menu's two options (selling or relend- ing) in a manner that maximizes the value of their assets subject to the terms of the menu. What is remarkable is that any menu will produce an equilibrium in which all the banks, whether they exit or relend, retain a payoff exactly equal to the offered exit price. Also, when the menu approach is used to implement a debt reduction, increases in debt prices following the debt reduction are in the interest of the debtor since it receives a compensating transfer from the nonexiting creditors. This is in contrast with the pure market buyback case in which the opposite holds (Bulow and Rogoff 1988). Thus any menu effectively allows the debtor tco capture a share of the entire efficiency gains generated by debt reduc- tion. The actual gains that accrue to the creditors depend on the level of the agreed exit price. The closer the exit price is to the ex-ante (initial debt) price, the less the banks gain. But the closer the exit price is to the ex-post (increased secondary market) price, the larger is their gain. There are two limiting cases of interest: when the exit price is exactly set to the ex-post price (as happens with pure market buybacks) all efficiency gains go to the creditors and when it is set to the initial debt price, in which case the debtor captures all the efficiency gain. III. CHARACTERISTICS OF THE MENU APPROACH TO DEBT REDUCTION This section looks at comparative statics associated with changes in the terms of the menu exit and relending options and discusses the costs associated with the debt deal. 98 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Comparative Statics With the menu approach the equilibrium amount of debt reduction varies depending on the agreed levels of the exit price and the relending ratio. From equations 7 to 9: (10) f(D1) / D1 = (p + n) /(1+ n). Differentiating equation 10 with respect to the exit price, p, and the relending ratio, n, and rearranging yields: (11) dD1 /dpo = -1 / ID1 (1 + n) [p1 -f (D1)]} (12) dD, /dn = - [D1 /(1 + n)] {(1 - p1) / [p1 - f' (D1)]}. Equations 11 and 12 are both negative when the debtor is on the increasing part of its debt value curve. Thus, when the relending ratio is held constant and the exit price is increased, more debt reduction is achieved in equilibrium. This is because as the initial exit price rises, exit becomes more desirable. More debt reduction is then needed in order to increase the secondary market price sufficiently so that the relending option becomes as valuable as the exit option. Holding the initial debt price constant and increasing the relending ratio increases the equilibrium amount of debt reduction achieved. As the relending ratio is increased, the exit option becomes more desirable than the relending option. But, in equilibrium, both options must be equally desirable. As a result a larger debt reduction will be achieved in order to raise further the ex-post debt price and increase the attractiveness of the relending option. The Cost of the Deal In the simple model, given knowledge of the debt value function, the new relending ratio and exit price can be determined so the deal costs exactly the amount of funds available to repurchase debt, ca. If there is uncertainty about the debt value function, then any particular menu will lead to a random amount of debt reduction (and consequent demand for buybacks). Given an exit price, there need not be a windfall gain to creditors, but the amount of resources required to "back the deal" will be uncertain. If the amount of relending turns out to be too large, only a portion of the available resources will be used, but creditors will remain at the payoff level implied by the exit price. However, if the amount of relending turns out to be too low, the available resources will not be sufficient to buy back all the tendered debt. If this occurs, a distribution mecha- nism will be needed to allocate the scarce buyback resources. For example, buybacks can be distributed on a pro-rata basis. If all banks are similar, they all end up with a similar mix of cash and new loans, per unit of old debt, and this payoff would be lower than that implied by the exit price. Because of this possibility creditors will initially insist on a higher exit price to protect them- selves when the debt value function is uncertain. Since relending may not end up Diwan and Kletzer 99 as desirable as exiting, the exit price must be large so as to produce a portfolio of a given value. An underfunded deal is not in the interest of the debtor country. An ade- quately funded menu will allow the debtor to discriminate more fully between different types of creditor banks. As discussed below, banks are likely to be heterogeneous with respect to their valuation of country debt. Since equilibrium deals allow banks to self-select from the menu the options that they value most while ad hoc distribution mechanisms do not, equilibrium deals with adequate funding allow the debtor to reap additional gains (ex ante) over deals in which options are distributed in some ad-hoc way. IV. THE MENU APPROACH WITH HETEROGENEOUS BANKS To simplify the equilibrium analysis of menu choices, it was assumed that all banks are similar. Removal of the assumption of homogeneous banks does not affect the equilibrium and comparative statics results in any important qualita- tive way. If banks value country debt differently, the banks may be ranked by their valuation. A simple menu divides the banks into two groups: low valua- tors, which value the exit option more than the new money option, and high valuators, which place greater value on the relending option. Separating these two groups is a bank that is indifferent between the two options. For this marginal bank, a no-arbitrage condition applies (equation 6) such that the exit price given by the menu is equal to the value of the relending option. The comparative statics are then similar to those in the homogeneous case. In par- ticular, when the new money call increases, the exit option becomes more valu- able to all banks, reducing the ex-post value of debt to all banks. As a result, the marginal bank shifts toward banks with higher valuation, and the equilibrium amount of debt reduction increases (see Diwan and Spiegle 1990 for details). Because it values the two options equally, the marginal bank retains no sur- plus from the deal. Banks on both sides retain a surplus that increases with the distance from the marginal bank. Thus, when banks are not homogeneous, a single "new money" tax is not sufficient to ensure that the "surplus" of every lender is the same. But the menu dominates simple concerted debt reduction deals because exit would not occur in a concerted agreement at any price below the valuation of the bank with the highest valuation. More gains can possibly be achieved for the debtor by extending the menu and achieving finer discrimina- tion. For example, several new money and exit options could be offered that differ in the timing of the payments made by the banks. Menus would not be needed if banks were homogeneous. A buyback price could be agreed on for which all banks would agree to sell a certain proportion of their portfolio at this price. A pure concerted approach, rather than an approach that combines concerted with market features, would be sufficient. Banks' heterogeneity is the principal reason why the menu approach to debt reduction dominates a simple concerted approach. 100 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I If the secondary market for sovereign debt is efficient, the marginal value of a debt claim should be the same for all participating banks since the market intermediates away banks' heterogeneity. Banks would then be effectively homo- geneous. Heterogeneity in valuation occurs only if the market is not fully effi- cient in intermediating the valuations of different banks. This can be the case if asset markets are incomplete. If lenders are unable to completely diversify against the risk of country debt using all available assets, then the change in the surplus caused by an overall debt reduction can vary across lenders even though the value of a marginal claim is the same. Reduction of the debt outstanding changes the distribution of returns to holding a part of the country's debt. The expected present value of the stock of debt held by each lender will not change by the same percentage, in general. This means that a single capital gains tax rate, imposed using the new money option, is insufficient to tax away all the potential gains to heterogeneous lenders. Heterogeneity may also occur because there are distortionary public policies in place (such as subsidized deposit insurance) or there are some other types of transaction costs. Distortionary public policy and transactions costs create a wedge between the price at which debt is sold and the net payout that accrues to the seller. As a result, debt is not transacted in the secondary market when differences in valuation are smaller than this wedge. It can be shown that this wedge increases as the quality of the debt claims deteriorates (see Demirguc- Kunt and Diwan 1990). If the secondary market for debt is inefficient, banks can differ for a number of reasons in spite of the possibility of trading. For example, a commercial bank may have other business interests in the debtor country, thus creating an extra benefit to lending (Sachs 1989). Other reasons for heterogeneous banks include differences in their expectations (Williamson 1988), differences in their nation- ality and thus in their tax and regulatory environment (Bouchet and Hay 1989), alternative business opportunities, and bank size in the presence of fixed costs associated with recontracting and monitoring. These factors give rise to a sec- ondary market in the first place, but an inefficient market will not eliminate the effects of heterogeneity. Heterogeneity across lenders also implies that the choices the banks make from a menu will depend on their own characteristics. This issue is explored in depth in Demirguc-Kunt and Diwan (1990), using the experience of the Brazilian 1988 debt deal. They show that 80 percent of bank choices between exit and relending options can be explained by measures of financial strength, exposure, nationality, and long-term interest in the country. V. THE MENU APPROACH WITH DEBT REDUCTION FINANCED BY INTERNATIONAL FINANCIAL INSTITUTIONS If funds are supplied by international financial institutions seeking to benefit the debtor, then a package that effectively taxes the capital gains of lenders will reduce the net transfer to the commercial creditors. Commercial lenders can Diwan and Kletzer 101 agree to choose between selling debt back to the debtor and holding claims with the obligation to provide new loans. A menu can be designed to ensure that lenders are indifferent between the two options. The status quo payoff can be determined if the relationship between the present value of debt and the face value of debt is known. In general, the new loans forthcoming under the relending option can be used by the debtor country either for domestic absorption or for further debt re- purchases. If they are used for buybacks, the funds available for buybacks are increased above the contribution of the international financial institutions. The debt is then reduced further, and the relending ratio should be higher since the secondary market price rises further. After the debt reduction program, the remaining debt is given by: (13) D, = Do-[(F + N- A)/p] + N where F is the amount of funds supplied by the international financial institu- tions, N is the total of new loans made by the remaining creditors, ,l is the part of the new financing (N + F) kept by the debtor country to finance domestic absorption, and p is the price paid in the buyback. It is assumed that the net resource transfer to the country, u, is a monotone increasing function of the funds supplied by the international financial institutions (N + F) and the new loans made by creditors. The amount of new funds used to repurchase debt is equal to the funds supplied by the international financial institutions plus new loans by creditors minus the funds used to finance domestic absorption (N + F - A). In equilibrium the following no-arbitrage condition must now hold: (14) N +f(DO) = (F + N-A) +±f(D1). The left side of equation 14 is the value of the creditors' assets before the buyback. The right side expresses the value of their assets after the operation has been completed. In addition to the value of remaining debt, the creditors receive a net amount (F + N - 4). When the funds supplied by international financial institutions, F, exceed the net resource transfer to the country, ,a, the private lenders gain. In this case, if required, the private lenders are willing to pay a positive price for debt reduction. The amount of new money necessary for creditors to retain no net increase in the value of their portfolios is positive. There is a positive equilibrium value of new loans by creditors, given the level of funds supplied by international financial institutions, such that equation 14 holds whenever a part of the funds supplied is spent on buybacks. The new secondary market price, which is a function of the amount of funds supplied by international financial institutions and the net resource transfer to the country, is higher than the initial debt price whenever the equilibrium level of new loans made by creditors exceeds zero. This follows from the concavity of the debt value function. When clebt reduction creates no change in the value of creditors' assets, the 102 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I implicit tax rate imposed on remaining debt is the difference between the new secondary market price of debt and the initial debt price. The capital gains realized by remaining debt holders are taxed 100 percent. The capital gain due to debt reduction is -y = [Do - (F + N - /i)/p0] (p1 - po), and the total tax revenue, which accrues to the debtor, is X = N(1 - p,). The equilibrium level of new lending by creditors is obtained when the capital gain is equal to the tax revenue. This is equivalent to solving equation 14. The amount of new money required per unit of old debt held back from repurchase is (15) n = N I (Dl -N) = (p - po) / (1 - pl) which is equivalent to equation 6. Since the buyback occurs at the ex-ante price, the exiting creditors also break even. The effect of the funds supplied by international financial institutions on the equilibrium level of new loans by creditors and the new equilibrium level of debt stocks can be calculated using equations 13 and 14: (16) dD, /dF = -(1 - ' )(1 -p)/ [(1 -po)f' + (po -f' )jt'] and (17) dN/dF = (po-f' )(1-A' ) /[(1-po)f' + (po-f' )' ] wheref' = f' (D1) and A' = a' (F,N). When the marginal net resource transfer to the country is less than one, a rise in the funds supplied by international financial institutions increases both the equilibrium level of new loans by credi- tors and the new equilibrium level of debt stocks until the equilibrium marginal value of debt equals the initial debt price (f' (D) = p). Further debt reduction beyond this point through increases in financing pro- vided by international financial institutions decreases the size of the net resource transfer. As long as the marginal value of debt is below the buyback price (initial debt price), remaining debtholders realize an increasing capital gain from further repurchases. Therefore, given the constraint that creditors are at least as well off with the program as the status quo, the compensating concession in new money that can be demanded rises with debt reduction. However, when debt is reduced far enough so that the marginal value of debt is greater than the initial debt price (f' (D) > po), the implicit tax needed to assure that creditors are indifferent between selling and holding debt declines with further reductions in the debt. If the proportion of available funds used to repurchase debt can change, then an increase in pi(F + N) for every value of (F + N) acts as a decrease in F. Thus, until debt is reduced to the point that the marginal value of secondary debt equals the buyback price, the debtor gains both net resource transfers and debt reduction from further buybacks. If the objectives of the donor place weight only on the debtor's benefits (excluding any other objectives, such as the cost of funds), then the optimal amount of remaining debt for the donor is less than or equal to the solution to setting the marginal value of debt equal to the initial debt price (f' (D) = p). Figure 1 depicts the debt value curve and initial value of debt in the top panel. The relationship between new loans made by creditors and the ex-post equilib- Diwan and Kletzer 103 Figure 2. Equilibrium New Money and Gift Financing New money (N) Gift financing (F) f'(D)=p,, F-Y rium level of debt is portrayed in the lower panel for solutions to equations 13 and 14. The maximum amount of new money contributed by creditors is at- tained when the marginal value of debt equals the initial debt price. Figure 2 illustrates the relationship between the equilibrium level of new loans by credi- tors and the amount of gift financing by international financial institutions. Equilibrium lending rises as financing increases until the marginal value of debt equals the initial debt price and then declines toward zero. The debt stock is zero when the gift is large enough to buy back the entire initial debt stock at the initial debt price. When debt is repurchased with a loan, rather than a gift, from official sources, seniority considerations also matter. When all debt, regardless of the creditor, has equal priority in repayment, the results developed above occur. The only difference is that when part of the initial transfer is a loan, less new money is forthcoming than in the pure gift case because the new loan reduces the ex- post debt price. Consequently, less debt reduction can be achieved for a given transfer. However, when the initial transfer takes the form of senior debt, it impairs the value of existing debt claims since junior debt is now serviced only after the senior loan is repaid. In that case, despite the gains that are due to debt reduction, the value of private creditors' assets is reduced (see the appendix for a treatment of financing by international financial institutions as junior and senior debt). It is probable that loans from international financial institutions are senior to 104 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I commercial loans, but that they also create efficiency gains due to conditionality (that is, they lead to an upward shift in the debt value curve). The net effect of those loans on the payoffs of commercial creditors can then be positive. The menu approach can help ensure that those net benefits are taxed so that the debtor country can retain a larger share of the efficiency gains due to conditionality. VI. CONCLUDING REMARKS When sovereign debt trades at a discount on secondary markets, a market buyback leads to an increase in the secondary market price. The wealth of private creditors increases because part of the funds used in the buyback is a transfer payment to them. When banks are homogeneous, the inclusion of exit and relending options in a menu from which private creditors freely choose can eliminate the wealth transfers due to the debt reduction. It is sufficient to set the buyback price equal to the ex-ante status quo price. Any new money call will do the job. In equilibrium creditors will provide enough new money so that the ex- post price leaves them indifferent to the exit option. An increase in the new money call reduces the cost of the menu and the extent of debt reduction achieved. Commitment by the banks to a menu including this new money option assures that debt repurchased is bought back at its marginal price instead of at its average price. The menu approach does not require assignments of particular choices from the menu to each lender. Rather, it implements debt reduction through a price system, allowing different creditors to select different portfolios in equilibrium from a common set of options. With heterogeneous banks, some transfer of resources will occur when participation in the debt reduction plan is voluntary, and the buyback price will generally need to be above the pre-buyback price. APPENDIX. THE MENU APPROACH WITH LOANS FROM THE INTERNATIONAL FINANCIAL INSTITUTIONS Debt may be repurchased with a loan from official sources interested in pro- viding debt reduction and new liquidity. In the first case considered, all debt, regardless of the creditor, has equal priority in repayment. That is, a government or international financial institution does not possess legal seniority privileges in relation to private creditors; this institutional assumption is ad hoc. The second case considered explores the effect that seniority privileges would have if the debt of international financial institutions was senior to private debt. For sim- plicity, only the special case in which the entire new money tax is used for domestic absorption is analyzed. A loan can be made by the third party in several ways, one of which is equivalent to providing private new money. The loan amount can be provided in Diwan and Kletzer 105 exchange for a debt obligation of equal face value. Therefore, the expected present value of the acquired debt is less than the loan amount, and the agency is part donor. Another extreme is that the nominal debt obligation exceeds the loan amount by the discount on debt, so that the expected present value of the new debt obligation is equal to the amount paid (the loan amount). The Case of Equal Priority in Payment In this case if the expected present value of the new debt obligation is equal to the loan amount, the additional debt bought at the secondary market discount cannot reduce debt unless private creditors are forced to accept losses. This is because the reduction in private debt is made up one for one with new debt. Any new money tax imposed on private creditors results in an ex-ante appropriation of their wealth. If the expected present value of the new debt obligation is less than the new loan amount, the new money loan can yield both debt reduction and more private loans. If the new loans are entirely used to buy back debt, then private creditors are indifferent between the status quo and equilibrium debt reduction under the menu if equations A-1 and A-2 hold: (A-1) f(DO) + N = p1 [D1 -L] + L + N (A-2) D1 = Do + N + L-(L + N)lp and where p1 = f(D1)/D1, where D denotes debt stock, N denotes new lending, p denotes the price of debt, L denotes the loan, and subscripts 0 and 1 indicate initial and after the deal, respectively. The right side of equation A-1 is the value of the assets held by all private creditors ex post. It also expresses that the ex- ante price of debt equals the ex-post price minus the tax imposed through new loan commitments. When the buyback price is set equal to the initial debt price, equation A-1 (for concave debt value functions) can be solved to obtain the positive equilibrium level of new lending. Creditors will be indifferent between selling debt claims at the ex-ante price and holding them with the commitment to provide a proportionate share of the equilibrium new money. The effect of an increase in the loan amount on the extent of debt reduction and net resource transfer to the debtor can be calculated from equations A-1 and A-2: (A-3) dN/dL = [po (1 -p1) / (1 - po)A] -1 (A-4) dD1/dL =-(1 - p1) /A < 0 where A is given by (A-5) A = (1 - L/D1)f' (D1) + (L/D1)pl. Comparison with the gift case in section V reveals that when part of the initial transfer is a loan, less new money is forthcoming than in the pure gift case 106 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I because the new loans reduce the ex-post debt price. Consequently, less debt reduction can be achieved for a given transfer. The Case of Senior Loans by the International Financial Institutions In this case the new loan provided by an international financial institution impairs the value of existing debt claims. This is because after a senior interna- tional financial institution loan is made, the value of private debt equals the present value of debt claims only when they are serviced after the senior loan is repaid. If the senior loan is used to repurchase debt and no new money is required, then the buyback price, p1, is given by: (A-6) p1 = [f(D + F - B) -f(L)] /(D - B) where F is the amount of new funds provided by the international financial institution and B is the amount of debt repurchased. The amount of the loan used for buybacks is the buyback price times the amount of debt repurchased. If all of the new funds are used for buybacks, then the following equilibrium condition holds: (A-7) L/B = f(D + F-B)-f(L)] / (D-B) from which the equilibrium buyback price can be derived. Since the debt value function is concave, this has a solution for F < B, if 0 < F < D. The top part of figure A-1 depicts an equilibrium price when a buyback is financed by a senior loan from an international financial institution. The amount of the loan used to repurchase B units of debt is shown on the right side of the figure as p1B. For the values of the ex-post price and the amount of debt repurchased shown, the demand for buybacks does not use up all of the loan. When all of the loan is used to repurchase debt, the equilibrium values of debt repurchased and the ex-post price are larger, so that p1B equals the loan amount. The bottom part of figure A-1 shows the effect of dilution: po is the price of private debt before the senior loan is made, and p1 is the ex-post equilibrium price. Comparison of the ex-post equilibrium price paid in a buyback to the ex-ante price (before the senior loan is made) shows that the value of private creditors' assets is reduced, and the present value of all expected debt repayments falls. Letting F/B equal the ex-post price, p, equation A-7 implies the solution for the ex-post price: (A-8) p = [f(D + F - L/p) -f(L)] / (D - F/p). As the ex-post price times the debt stock approaches the loan amount (all private debt is repurchased), the right side of equation A-8 remains less than unity, in contrast to the case of equal seniority. Therefore a solution to equation A-8 exists for an ex-post price and loan amount such that all the private debt is repurchased and the ex-post debt burden is smaller than the initial debt stock. Diwan and Kletzer 107 Figure A-1. Debt Reduction Financed with a Senior Loan '.450 Debt value . (f (D)) , + ~~~~~~~~f (D) ,.' ~~~~~~~~~~~~P L DO+L-B Do+L Debt stock (D) Debt value 450 (f (D)) L DO+L-B Do Debt stock (D) This reduction in the expected present value of debt repayments is financed by the dilution of the private creditors' debt claims caused by the senior loan. The difference, f(D) - f(L), is the net transfer from private creditors to the debtor; this is a transfer of future rather than current resources. To further illustrate the dilution effect of an increase in senior debt, suppose that private creditors are offered the ex-ante secondary market price, p0, in the buyback (say for political reasons). From equation A-8, if the repurchase price is set to po = f(D)/Do on the right side, then the solution for the ex-post secondary market price of private debt (appearing on the left side) is less than po. Remain- 108 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO . 1 ing private creditors suffer a capital loss and prefer to sell back debt. An equilib- rium buyback price such that creditors are indifferent at the margin between selling and retaining debt is lower than the initial price. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Bulow, Jeremy, and Kenneth Rogoff. 1988. "The Buyback Boondoggle." Brookings Papers on Economic Activity 2. Bouchet, Michel, and Jonathan Hay. 1989. "The Rise of the Market-Based Menu Ap- proach and Its Limitations." In Ishrat Husain and Ishac Diwan, eds., Dealing with the Debt Crisis. A World Bank Symposium. Washington, D.C.: World Bank. Demirguc-Kunt, Asli, and Ishac Diwan. 1990. "The Menu Approach to Developing Country External Debt: An Analysis of Commercial Banks' Choice Behavior." PRE Working Paper 530. World Bank, International Economics Department, Washington, D.C. Processed. Diwan, Ishac, and Mark Spiegle. 1990. "Menu-Driven Debt Reduction Schemes with Heterogeneous Creditors." World Bank, International Economics Department, Wash- ington, D.C. Processed. Dooley, Michael. 1988. "Buybacks and the Market Valuation of External Debt.' In Jacob Frenkel and others, eds., Analytical Issues in Debt. Washington, D.C.: Interna- tional Monetary Fund. Sachs, Jeffrey. 1989. "Efficient Debt Reductions?' In Ishrat Husain and Ishac Diwan, eds., Dealing with the Debt Crisis. A World Bank Symposium. Washington, D.C.: World Bank. Williamson, John. 1988. Voluntary Approaches to Debt Relief. Paper 25. Washington, D.C.: Institute for International Economics. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 109-124 Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia Martin Ravallion Recent evidence suggests that food energy intakes of the poor respond less to income than was once thought. However, it is not intake alone that is of concern, but under- nutrition. Two facts confound assessments of how undernutrition responds to incomes and prices: individual nutrient requirements vary in a generally unobserved way, and intakes are observed with error. By modeling observed intake distributions economet- rically, straightforward stochastic dominance tests can permit robust qualitative infer- ences about such responses. An application to Indonesia in the mid-1 980s indicates that regional distributions of food energy intake are influenced by average income levels, intraregional inequalities, and local prices of staple food grains-all of which have unambiguous effects on undernutrition. The results suggest that any adverse effects on inequality of a growth process would need to be large to outweigh the desirable effect on undernutrition. In addition higher food staple prices still have adverse effects on undernutrition after allowingfor their likely positive effects on rural incomes. The attainment of adequate nutritional levels is an important criterion in eval- uating the success of development policies. However, such evaluations have often been hampered by both conceptual and technical problems associated with how undernutrition is measured. Most important among these is that the nutri- tional requirements for good health vary across individuals and over time in generally unknown ways, and nutrition intakes are typically measured with error. In light of these problems, what can we say about how the instruments of development policies influence the extent of undernutrition? Development policies have often emphasized the role of economic instru- ments, particularly incomes, in reducing undernutrition. However, recent household-level studies raise doubts about the effectiveness of these instruments Martin Ravallion is in the Welfare and Human Resources Division of the Population and Human Resources Department at the World Bank. This article is a product of a research project supported by the World Bank's Research Committee. The author is grateful to Monika Huppi and Alok Bhargava for their help with the data and estimation and to Indonesia's Central Bureau of Statistics for providing the data tapes of the 1984 and 1987 National Socioeconomic Surveys. Helpful comments on the paper were made by Shubham Chaudhuri; the Review's referees; seminar participants at the Department of Economics, Cornell University; and participants at the 1991 Eastern Economics Association meeting. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 109 110 THEWORLDBANKECONOMICREVIEW,VOL. 6,NO. 1 in improving nutritional levels. Mounting empirical evidence suggests that the income elasticity of household food energy intake in developing countries is lower than had been thought a decade ago (Behrman 1988; Behrman and others 1988; Bouis and Haddad 1988; Alderman forthcoming). There are reasons to suspect that the methodologies used in earlier studies may have overestimated that elasticity (Behrman and Deolalikar 1987; Bouis and Haddad 1988). The downward revision that seems to be warranted is far from negligible. Whereas 10 years ago an income elasticity of energy intake for the poor of about 1.0 would probably not have been seriously questioned, elasticities a good deal less than 0.5 would be considered more in keeping with recent estimates (Behrman and Deolalikar 1987; Bouis and Haddad 1988; Ravallion 1990; Strauss and Thomas 1989; Bhargava 1991). This substantial downward revision of estimates of how nutrition intake re- sponds to income has some potentially profound implications for development policies. The fight against hunger has been one of the strongest motivations for development, and raising incomes of the poor (through both the growth process and policies to reduce inequalities of income) has long been seen as the main weapon in that fight. The recent evidence has led some observers to suggest that this weapon may be quite blunt or, indeed, virtually useless. Does this aspect of our approach to development policy need a major revision? This article takes a further look at the question of whether aggregate under- nutrition responds to incomes and prices. The point of departure is the realiza- tion that nutrient intakes alone are not of interest but, rather, the adequacy of those intakes relative to needs (Ravallion 1990). Although this is obvious at a conceptual level, it is difficult to measure. The methodology proposed here is potentially far more robust to the inevitable errors and unknowns in measuring attainment relative to needs than are previous econometric studies. Beginning with an econometric model of how nutrient intake distributions vary across regions or sectors of the economy, theoretical results on stochastic dominance are used to infer the effects of changes in incomes and prices on undernutrition. The usefulness of stochastic dominance theory for ranking dis- tributions in terms of some objective function has been known for more than 20 years (Hadar and Russell 1969; Rothschild and Stiglitz 1970), although the relevance to poverty and nutrition analysis has come to be appreciated only quite recently (Atkinson 1987; Foster and Shorrocks 1988; Kakwani 1989). In assessing impacts on undernutrition, the dominance approach, outlined in sec- tion I, has the advantages over past methods that it uses all of the information available on the distribution of nutrient intakes, and it places far fewer restric- tions on the unknown distribution of individual nutrient requirements. It can also allow inferences that are more robust to intake measurement errors and to arbitrary choices about the specific measure of undernutrition. This study combines the dominance approach with an econometric model of intake distributions, thus permitting dominance tests of the comparative static effects on undernutrition of changes in the explanatory variables of the intake Ravallion 111 model. In section II the methodology is used to explore Indonesia's progress in reducing undernutrition during the mid-1980s, especially the regional dimen- sions of that progress and the role played by changes in incomes and prices. In section III other questions concerning the distribution of the benefits of growth are explored: How responsive is undernutrition to changes in income inequality? Will growth that is associated with an increase in inequality be able to reduce undernutrition? If income redistribution entails a "growth tradeoff' (due to, for example, adverse effects on aggregate savings), how severe would that tradeoff have to be to neutralize the effect of greater equity on undernutri- tion? Section IV summarizes the conclusions. I. APPROACHES TO ASSESSING IMPACTS ON UNDERNUTRITION The biggest single problem in past econometric studies of nutrition is that one rarely knows what nutrients the surveyed individuals require to maintain normal physiological functions without symptoms of deficiency. Requirements may vary widely among individuals because of differences in metabolic rates at rest and activity levels. Two sources of variability can be distinguished: interpersonal variability (genotypic variations around the assessed requirements of some refer- ence person) and intertemporal variability for a given person, which has been interpreted as the outcome of physiological regulatory mechanisms influencing energy utilization in the human body (as in, for example, Sukhatme 1978 and Srinivasan 1981). This latter source of variability has led Sukhatme and others to advocate a cutoff point in measuring undernutrition that is well below stipulated require- ments for a given person. Although the existence of intrapersonal variability may lead one to use caloric cutoff points that differ from the stipulated require- ments, whether the cutoff point should be set lower or higher will depend on the relative importance attached to "type 1 errors" (incorrectly classifying the person as undernourished) and "type 2 errors" (incorrectly classifying the person as adequately nourished) (Dasgupta and Ray 1990; Kakwani 1989). Advocacy of a lower cutoff point can be interpreted as a judgement that type 1 errors matter more than type 2 errors. This is clearly a questionable judgement (Dandekar 1981; Osmani 1987; Dasgupta and Ray 1990). For this discussion the source of the variability is not specified, so that the distribution of individual-specific requirements at the survey date could reflect either interpersonal or intertem- poral variability. Uncertainty about nutritional requirements is not the only problem. Nutrient intakes, although more readily observed than requirements, are typically mea- sured with error, such as those resulting from imperfect recall of food consump- tion. In effect this will also mean that the individual requirements specified are incorrect and should ideally be adjusted for the intake measurement error. This measurement problem is another reason for treating requirements as a random variable. 112 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I By making explicit assumptions about the interpersonal distribution of nutri- ent requirements, measures of undernutrition can be readily constructed from household or individual survey data, and, using an appropriate econometric model of intake determination, income and price effects can then be measured (Ravallion 1990). But the assumptions made about requirements are often arbi- trary, and they may influence the conclusions drawn from such an exercise. Fortunately, the main interest is often in the qualitative effects of changes in incomes and prices on undernutrition: is some policy combination, interpreted through a set of price and income changes, moving undernutrition in the right direction? In this case there is an alternative approach utilizing stochastic domi- nance tests, which place far fewer restrictions on the unknown distribution of individual requirements. With this approach each person may have a different requirement, and the distribution need not have any recognizable form. Although this approach allows a far more general class of possible distribu- tions of requirements, it still imposes two potentially important assumptions about the distribution. First, as is commonly assumed following Sukhatme (1961), intakes and requirements are taken to be independently distributed. One would suspect that the influence of common factors such as age, weight, and activity levels would lead intakes and requirements to be positively correlated. Nonetheless there is some evidence suggesting that there is little or no correla- tion between intakes and requirements (Sukhatme 1961), and Kakwani (1989) has further found that estimates of the proportion of the population of India deemed to be undernourished are quite insensitive to the assumption one makes about the correlation between intakes and requirements. Second, it is assumed that the distribution of requirements does not change over time or, in the specific cases here, is independent of incomes and prices. Whether that is true will depend on how the changes in incomes and prices occur. If intakes improve as a result of greater effort leading to higher incomes, for example, then undernutri- tion need not improve. The reasonableness of these assumptions should be kept in mind when interpreting the empirical results below. Under these assumptions one can invoke well-known theoretical results on stochastic dominance to at least partially order the intake distributions in terms of any well-behaved measures of undernutrition. To illustrate, suppose the in- take distribution shifts from state A to state B, as a result of (say) changes in the income distribution; state B, for example, may be reached after economic growth in state A. If intakes increase for all individuals in the move from A to B and there is no change in any individual's requirement, then undernutrition must fall. But that is an unnecessarily demanding test for a reduction of undernutrition. A more useful test can be constructed as follows. Let F,(z) denote the proportion of the population that fails to reach a given intake level z in state t. In the vocabulary of poverty measurement, F,(z) is the headcount index of undernutri- tion when a single requirement cutoff point is set at z. As z is allowed to vary over its entire range, F,(z) traces out the cumulative distribution function of Ravallion 113 intakes in state t. If FA(z) is below FB(z) at all intake levels (or, more precisely, FA(z) is nowhere above FB(z) and at least somewhere below), then the proportion of the population that is undernourished will be lower in A than B. This is called a first-orcler dominance test for comparing undernutrition in two states. First-order dominance can be an extremely useful test for determining whether there is more poverty in one state than another for any unknown, but fixed, poverty line (Atkinson 1987; Foster and Shorrocks 1988). Furthermore the underlying theoretical result that supports this test can be readily generalized to accommodate any fixed distribution of poverty lines or (as in this application) nutrient requirements (Kakwani 1989). This is the case because our best esti- mate of the proportion of the population that is undernourished for any distri- bution of requirements is the expected value of F,(z), where the expectation is taken over that distribution of requirements. If two intake distributions have the same requirement distribution and first-order dominance holds, then the ex- pected value of FA(z) must exceed FB(z) when both expectations are evaluated over the distribution of requirements. Undernutrition is higher in state A than in state B. First-order dominance of one intake distribution over another also implies an unambiguous ranking of the two distributions in terms of a broader class of undernutrition indicators than the simple headcount index. Let the level of undernutrition of a person with intake x and requirement z be u(x,z), which is positive for x < z, but zero otherwise. For the headcount index, u(x,z) = 1 for x < z, but zero otherwise. Thus the headcount measure is insensitive to differ- ences in the severity of undernutrition. A more general assumption is that the function u is strictly decreasing in intakes and increasing in requirements for all x < z and vanishes at x = z. Aggregate undernutrition is then the expected value of u(x,z) over the relevant intake and requirement distributions. For example, if u(x,z) = (1 - x/z)- for x < z and any positive parameter a, then we have the measure of undernutrition proposed by Kakwani (1989). (This generalizes the Foster, Greer, and Thorbecke 1984 class of poverty measures so as to allow the poverty line to be a random variable.) It can be readily demonstrated that first- order dominance then implies an unambiguous ranking of two distributions in terms of aggregate undernutrition. This result holds for any functional form for u(x,z) with the aforementioned properties and any fixed distribution of requirements. If the first-order dominance test is inconclusive (with cumulative frequency distributions intersecting at one or more interior points), then some requirement distributions and some reasonable measures of undernutrition will rank the two intake distributions inconsistently compared with others. Here an analogous second-order dominance test may prove useful. In particular suppose one con- siders only the undernutrition measures that are strictly decreasing in intakes (thus excluding the headcount index). It can be shown that if the area under the cumulative distribution function of intakes in state A is less than that in state B at all intake levels, then aggregate undernutrition is lower in A than B for any 114 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 requirement distribution. Kakwani (1989) proves this for the aforementioned class of undernutrition measures based on the Foster-Greer-Thorbecke func- tional form. Slightly modifying the argument in Atkinson (1987), one can show that the claim also holds for a broad class of other functional forms for u(x,z). If this second-order test fails, one can invoke higher-order dominance tests, requir- ing the further restriction of the class of admissible measures of undernutrition, or one can restrict the set of admissible requirement distributions. In applying the dominance approach, the distribution of intakes at each of two dates can simply be compared, as in Ravallion and Huppi (1991). This is useful in assessing overall progress in alleviating poverty or undernutrition. However, any observed change is likely to be the result of changes in many factors, and a simple comparison of distributions reveals nothing about the relative importance of different factors. A more illuminating approach, which is followed here, first models the intake distribution as an explicit function of the variables of interest and then uses this model to conduct dominance tests. (For another, somewhat different, example of this approach, see Ravallion and van de Walle 1991.) This methodology does not yield estimates of intake elasticities at the house- hold or individual level, such as those widely discussed in the recent literature. This may be viewed as a limitation of the methodology, but, as argued else- where, this limitation is not serious, since the individual intake elasticity can be a poor guide to the effects of policy on undernutrition (Ravallion 1990). If quan- titative results are desired, a better measure is provided by the elasticities of the cumulative intake distribution F(x) with respect to the variables of interest. These are derived naturally in the proposed methodology. II. AN APPLICATION TO INDONESIA The data tapes of the National Socioeconomic Surveys (SUSENAS), conducted by Indonesia's Central Bureau of Statistics, estimate household energy intakes by applying caloric unit values to the reported consumption of about 170 food and beverage items during the previous week. Most of the recent literature focuses on caloric intake as the sole determinant of nutritional well-being, and that approach will be followed here. The 1984 and 1987 SUSENAS consumption surveys contain stratified random samples of 50,000 households in each year. Population estimates were derived using the household-specific inverse sampling rates given on the SUSENAS data tapes. Because respondents were asked about expenditures on foods not prepared at home but not about the quantities consumed of such foods, the survey underesti- mates energy intakes. Food prepared outside the home is an important source of calories in Indonesia, particularly for certain socioeconomic groups such as those living alone. Although undernutrition will probably be overestimated, it is less clear how this would affect distributional comparisons over time and across regions. Average energy intake from foods prepared away from home is esti- Ravallion 115 mated to be 128 calories per person per day in 1987 and 112 calories per person per day in 1984 (Biro Pusat Statistik 1989a). Assuming that calories obtained from this source increase with income, this omission could cause the income response of true intakes to be underestimated. Although it is reasonable to assume that "eating out" is a normal good, however, it is also a heterogeneous good, and casual observation suggests that food from street stalls is popular among both poor and not-so-poor in Indonesia. Furthermore, while expendi- tures on prepared foods increase with income (Biro Pusat Statistik 1989b), it is less likely that intakes from this source will do the same, as the unit values will undoubtedly also increase. The exclusion of calories consumed outside the home is not likely to bias the elasticity estimates below. The SUSENAS indicates a marked improvement in the distribution of energy intakes in Indonesia during the 1980s. Ravallion and Huppi (1991) construct the cumulative frequency distributions of intakes for 1984 and 1987 from the SUSENAS consumption surveys. Their results indicate that the 1987 intake distri- bution lies below that for 1984 up to a high intake level. First-order dominance therefore holds over a very wide range of potential cutoff points. Furthermore the second-order dominance condition holds over the whole range of intake cutoff points. A wide range of measures of undernutrition will thus indicate an improvement between 1984 and 1987, irrespective of the distribution of re- quirements, assuming that the latter did not change. This conclusion is unlikely to be affected by excluding foods prepared away from home. Do these aggregate results mask significant regional variation? After dividing the 27 provinces of Indonesia into urban and rural areas, which produced 52 regions (Jakarta being solely urban, and East Timor having only rural observa- tions in the sample), the intake distributions were constructed for each region, and the dominance tests applied. An unambiguous improvement was indicated for 29 regions, while an unambiguous worsening was indicated for only three (the rural areas of Riau, Bengkulu, and Maluku). The first-order test was am- biguous for the remaining 20 regions. For each of these the 1987 distribution crossed the 1984 just once from below, that is, there was an improvement at the lower end of the distribution. The average crossover point was at approximately 1,950 calories per person per day (probably around 2,100 with food eaten away from home). The second-order test resolved this ambiguity for 12 regions, all but one of which (urban Central Sulawesi) showed an unambiguous improve- ment. The regions for which the second-order test was ambiguous were urban and rural Aceh, rural West and South Kalimantan, rural South-East Sulawesi, and the urban areas of Riau, Jambi, and North Sulawesi. Although first-order or second-order dominance tests thus indicate a reduc- tion in undernutrition for about 80 percent of the regions, the quantitative extent of this improvement varies considerably across regions. Table 1 gives the mean and standard deviation of the percentage change in the cumulative distri- bution of intakes between 1984 and 1987 for selected intake levels. Both the average percentage decline and its standard deviation tend to be higher at the 116 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 1. Average Percentage Change in the Proportion of the Population below Selected Calorie Intake Levels for 52 Regions of Indonesia, 1984-87 Calories per person per day 1,200 1,400 1,600 1,800 2,000 2,200 Mean percentage change -46.8 -25.5 -12.9 -9.3 -5.2 -2.0 Standard deviation 38.7 31.4 24.7 18.5 14.2 12.6 Coefficient of variation 0.8 1.2 1.9 2.0 2.7 6.4 Note: Calorie intake excludes energy from food not prepared at home. Source: Author's calculations based on regional data files constructed from the household-level data tapes of Indonesia's National Socioeconomic Surveys for 1984 and 1987. lower end of the distribution, although the coefficients of variation increase with the intake level. The large standard deviations signal the magnitude of differ- ences in nutritional improvements. To what extent can differences in income growth explain this regional diver- sity in the rate of nutritional improvement? Table 2 reports the elasticity of the proportion of the population below various intake levels with respect to income and to total food expenditure (obtained by regressing percentage changes against percentage changes between 1984 and 1987; see Ravallion 1991 for further information). An inverse relationship is evident at all points for both income and expenditure, with the (absolute) elasticity decreasing sharply as the intake level increases. Absolute elasticities are higher for food expenditure than for income at all points. (These elasticities are not comparable with the intake Table 2. Income and Expenditure Elasticities of the Proportion of the Population below Selected Energy Intake Levels by Region of Indonesia, 1984-87 Calories Elasticity with respect to per person Food per day Income expenditure 1,200 -1.190 -1.480 (0.259) (0.168) 1,400 -0.633 -0.852 (0.185) (0.128) 1,600 -0.435 -0.537 (0.127) (0.094) 1,800 -0.355 -0.424 (0.092) (0.067) 2,000 -0.228 -0.265 (0.070) (0.054) 2,200 -0.155 -0.166 (0.061) (0.050) Note: Elasticities are based on OLS estimation with data as percentage changes over the period. Standard errors are in parentheses; n=52. Calorie intake excludes energy from food not prepared at home. Source: Author's calculations based on regional data files constructed from the household-level data tapes of Indonesia's National Socioeconomic Surveys for 1984 and 1987. Ravallion 117 elasticities often quoted in the literature discussed above; they are the elasticities of the proportion of the population below each intake cutoff point rather than the elasticities of intake at that point.) These results suggest that, first, regions with higher rates of income growth tended also to have greater rates of improvement in the intake distribution. Second, the proportionate shift induced by a given income growth rate tended to be greater at lower intake levels. And third, there is considerable regional varia- tion around these trends. There are, for example, many regions in which the intake distribution noticeably worsened, despite at least modest growth in mean incomes. This last observation could reflect any number of factors, including higher prices for staple foods, worsening intraregional income inequalities, or deterio- rating related health services. To further explore the determinants of the shifts in the intake distributions, the following model was estimated across regions and for each of six values of the intake cutoff: (1) L87(x) = ZiY(x) + 3(x)L84(x) + Ei(x), i = 1, . . ., 52 where Lit(x) log[Fi(x)/(1 - Fi(x)] t = 1984, 1987 x is the value of the intake cutoff, and Z denotes a vector comprising mean individual income in the region, mean income squared, the Gini index of in- equality in consumption within the region, the average price of rice, and an urban-rural dummy variable equal to 1 if the region is rural. A logit transform is applied to the intake frequencies in equation 1 to avoid truncation in the distri- bution of the model's dependent variable. The model's random error terms ei are assumed to include region-specific random effects, entering additively, and the lagged dependent variable is treated as endogenous. The Bhargava and Sargan (1983) maximum likelihood estimation method for dynamic random effects models is appropriate for estimating equation 1 and has been used to obtain the results reported in table 3, below. The mean income and Gini index were estimated from the household-level data tapes. Income data in the SUSENAS include imputed values of own produc- tion. An allowance was made for urban-rural price differences (following Rav- allion and Huppi 1991) but not for regional price differences, apart from rice. (There is no satisfactory price index for this purpose for Indonesia. One candi- date, a deflator based on the raw data for the consumer price index, was tested as an additional independent variable, but proved insignificant.) Only Gini indexes for total consumption expenditure were included because this probably indicates inequality better than a Gini index based on income. The urban-rural dummy variable may pick up some of the measurement error in the dependent variable associated with the greater importance of energy obtained from pre- pared food[s in urban areas. Although dynamic effects are allowed in the model, there is little sign of such 118 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Table 3. Estimation Results from Models of the Determinants of Energy Intake Distributions across Regions of Indonesia, 1987 Calories per person per day 1,200 1,400 1,600 1,800 2,000 2,200 Intercept 46.23 20.64 21.65 20.21 -18.68 -39.52 (0.92) (2.94) (0.68) (0.12) (2.85) (4.1) Rural dummy 0.62 O.S9 0.89 1.24 1.66 1.96 variable (0.27) (0.38) (0.12) (0.14) (0.56) (0.35) Logmean income -10.07 -4.28 -4.54 -4.29 3.61 8.12 (0.062) (0.07) (0.03) (0.02) (0.04) (0.17) Log income 0.42 0.15 0.156 0.13 -0.28 -0.52 squared (0.02) (0.02) (0.004) (0.003) (0.03) (0.02) Logriceprice 1.38 0.68 1.03 1.14 1.60 1.59 (0.14) (0.22) (0.11) (0.03) (0.17) (0.86) Giniindex 7.22 6.ss 5.12 5.68 7.42 9.29 (3.51) (2.40) (1.33) (1.23) (3.47) (4.32) Lagged dependent 0.31 0.26 0.07 -0.21 -0.52 -0.79 variable (0.27) (0.17) (0.18) (0.16) (0.58) (0.43) Note: These are dynamic random effects models, using the Bhargava-Sargan maximum likelihood estimator. Standard errors are in parentheses; n = 52. The dependent variable is the logit transformation of the ordinate of the energy intake frequency distribution in 1987. Calorie intake excludes energy from food not prepared at home. Source: Author's calculations based on regional data files constructed from the household-level data tapes of Indonesia's National Socioeconomic Surveys for 1984 and 1987. effects in the results, and so these will be ignored in the following discussion. The -y parameters are estimated quite precisely (table 3). Several alternative specifications were also estimated. These (alternately) deleted the lagged depen- dent variable and the squared income term, treated income as endogenous, and used mean consumption and mean food expenditures as alternatives to income. Similar results were obtained with and without the alternative specifications; details are available from the author. In addition rejection of the null hypothesis that ,8(x) = 1 implies rejection of the (nested) alternative specification using proportionate changes as the dependent variable. The results in table 3 indicate that, for all intake levels considered, the intake distribution is strictly decreasing in mean income (the turning point of the quad- ratic function of log income is outside the range of the data in all cases) and strictly increasing in both the mean price of rice and the Gini index. Thus first- order dominance is satisfied, implying that undernutrition falls with higher average incomes, lower rice prices, and lower inequality for any (fixed) distribu- tion of requirements. III. FURTHER IMPLICATIONS OF THE EMPIRICAL RESULTS Interpretation of the quantitative results is easier if one calculates the implied elasticities of the intake distribution F(x) with respect to each of the variables of interest. Table 4 gives these elasticities of F(x) at the mean of each variable for each value of x. The income elasticity (this time controlling for the other vari- Ravallion 119 ables in the above model) is again found to decrease sharply as the intake cutoff point rises. The same is true of the elasticities with respect to the Gini index and the price of rice. Table 5 gives simulated effects on the intake frequency distributions of 10 percent changes in mean income, the Gini index, and price of rice, alternately, all evaluated at variable means. The new intake distribution resulting from any change dZ is estimated by its linear approximation F87[1 + (1 - F87)dZy]. Both income growth and greater equity alleviate undernutrition, and they have roughly equal proportionate impacts; around the middle of the distribution, a 10 percent increase in the mean income is found to have about the same impact on the intake distribution as a 10 percent decrease in the Gini index of inequality. This suggests a further question: what is the effect on the intake distribution if income growth is accompanied by increased inequality? The correlation coeffi- cient betwveen the Gini index and mean incomes across the 52 regions for 1987 is significant and positive at 0.47, although for 1984 the correlation coefficient is just 0.24, and the correlation coefficient between proportionate changes in the two variables between 1984 and 1987 is a weak, although positive, 0.23. There is at best a slight hint of a Kuznets relationship in these data. One should be wary of giving this correlation a causal interpretation, but it is still of interest to ask: if income growth occurs with an increase in inequality consistent with the above correlation, would undernutrition still fall? An ordinary least squares regression of the Gini index against the log of mean income in 1987 gives (t-ratios in parentheses): (2) Gini index =-0. 223 + 0. 0481og(mean income) (1.679) (3.737) with R2 = 0.22. (Using a logit transformation of the Gini index to avoid truncation of the dependent variable yields similar results; see Ravallion 1991.) Table 4. Estimated Elasticities of Cumulative Frequencies of Energy Intake at Selected Intake Levels, Indonesia 1987 Calories Elasticity with respect to Critical per person Mean Gini Rice income per day income index price sharea 1,200 -1.31 1.86 1.30 0.99 1,400 -0.95 1.44 0.55 0.57 1,600 -0.84 0.90 0.66 0.78 1,800 -0.77 0.74 0.55 0.71 2,000 -0.71 0.66 0.52 0.74 2,200 -0.56 0.55 0.34 0.61 Note: Calorie intakes exclude energy from food not prepared at home. a. Maximum revenue from rice production as a proportion of total income, which is consistent with a positive total elasticity with respect to rice price (that is, allowing mean income to vary). Source: Author's calculations based on regional data files constructed from the household-level data tapes of Indonesia's National Socioeconomic Surveys for 1984 and 1987. Table 5. Actual and Simulated Cumulative Frequency Distributions of Energy Intakes for Indonesia, 1987 (percentage of individuals with intakes below each cutoff) Cumulative frequency With 10 percent With 10 percent change in increase in mean Calories Mean income Gini index Rice price income and per person an increase of 0. 005 per day Actual Increase Decrease Increase Decrease Increase Decrease in the Gini index 1,200 6.02 5.23 6.81 7.14 4.90 6.80 5.24 5.43 1,400 19.38 17.53 21.23 22.18 16.58 20.44 18.32 18.02 1,600 35.98 32.94 39.02 39.21 32.75 38.35 33.61 33.51 1,800 52.22 48.21 56.23 56.10 48.34 ss.07 49.37 48.90 2,000 67.27 62.49 72.05 71.74 62.80 70.79 63.75 63.27 2,200 78.s5 74.14 82.96 82.83 74.27 81.23 75.87 74.89 Note: Calorie intake excludes energy from food not prepared at home. Source: Author's calculations based on regional data files constructed from the household-level data tapes of Indonesia's National Socioeconomic Surveys for 1984 and 1987. Ravallion 121 The implied elasticity at the mean of the Gini index is 0.18; a 10 percent increase in the mean income would only yield about a 2 percent increase in the Gini index at the mean. At this rate of change in inequality, growth would still have a positive effect on undernutrition and almost as strong as the previous partial elasticities had implied. The last column of table 5 gives a revised esti- mate of the effect of a 10 percent increase in income, incorporating the associ- ated change in inequality. If the growth process induces a Kuznets effect, it would have to be very much more dramatic in its impact on inequality than the above statistical correlation indicates before it would reverse the conclusion that income growth alleviates undernutrition. It is sometimes also argued that policy interventions aimed at alleviating undernutrition by redistributing incomes in favor of the poor will have an ad- verse effect on aggregate income, so that the effect on undernutrition would be mitigated. The above results suggest that the percentage decrease in the mean income would need to be of about the same magnitude as the percentage de- crease in the Gini index before the redistribution was neutral in its effect on undernutrition. Such a tradeoff may not be plausible. The empirical results also suggest that undernutrition will be quite responsive to changes in the price of rice, although the elasticity of the intake distribution function falls even more sharply as one moves to higher intake levels (see table 4). Of course this is a partial elasticity, holding both mean income and the Gini index constant. If the share of income obtained from rice production exceeds some critical level, then the income effect of an increase in the price of rice will outweigh the direct adverse effect on undernutrition. (To a first-order approx- imation, the elasticity of income with respect to the price of rice is simply the share of income from rice production.) The results in table 2 can be used to evaluate this critical income share, assuming that (for lack of any plausible alternative assumption) the price change is distributionally neutral. These in- come shares are given in the final column of table 4. The critical values are considerably higher than the average income shares from rice production in rural areas of Indonesia. For example, using 1987 data for rural central Java, the average share of income obtained from food grain production (no further disaggregation is possible) is about 20 percent, and this varies little with income. It is evident, then, that an increase in the price of rice will have an adverse effect on both urban and rural undernutrition, allowing for typical income responses. All these comparative static results hold without qualification for any inter- personal distribution of requirements, provided that this distribution is not also a function of incomes and prices. If requirement distributions shift, then caveats are called for, as noted in the previous section, and the above results may be either strengthened or weakened. Take, for example, an increase in income that arises from an increase in demand for agricultural labor (due to, say, a green revolution allowing multiple cropping). This type of work tends to be strenuous so that the requirement distribution may well shift to the right, reflecting an increase in the mean nutrition requirement. Effects on undernutrition are no 122 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I longer obvious. However, income growth associated with labor-intensive indus- trialization would probably have the opposite effect on the distribution of nutri- tion requirements; even unskilled modern sector jobs tend to have lower energy requirements than those of the traditional agricultural sector (James and Scho- field 1988), thus strengthening the comparative static results on the effects of income growth on undernutrition. IV. CONCLUSIONS Although recent household-level consumption studies have revealed much about the micro-level determinants of nutrient intakes, empirical assessments of undernutrition and its response to changes in incomes, prices, or other variables of interest are still confounded by the fact that, while nutrient intakes can be estimated with tolerable precision from such surveys, nutrient requirements are unobserved. Some studies of the impacts of income or other changes on nutri- tional status have simply ignored requirements, while others have taken a cutoff point that is assumed to be constant for all individuals or constant for all individuals of the same type, with a modest number of types identified. Neither approach is really a convincing foundation for assessing impacts on undernutri- tion, when (as can be reasonably expected) nutrient requirements for good health can vary widely across individuals and in ways that the researcher can have little hope of knowing. The potential hazards for inference are plain enough. This article has used an alternative empirical approach that exploits the po- tential for establishing partial orderings of intake distributions in terms of mea- sures of undernutrition for any unknown but fixed distribution of requirements in a population. Every individual may have a different requirement, which may change over time, although the overall distribution is assumed to be static and independent of the intake distribution. The theoretical results needed to assess comparative static effects on a broad class of measures of undernutrition are then straightforward applications of well-known results from the theory of stochastic dominance. In implementing the approach here, an attempt has been made to better understand the regional dimensions and economic determinants of the fall in aggregate undernutrition in Indonesia during the 1980s, as found by Ravallion and Huppi (1991). Cumulative distributions of household intakes per person for each of 52 regions of Indonesia have been constructed for 1984 and 1987. Ordinates of these distributions are then modeled as functions of regionally specific variables of interest, focusing particularly on average incomes and rice prices and on the inequality of consumption within regions. Dominance tests have then been applied to assess qualitative effects on undernutrition. A reduction in undernutrition between 1984 and 1987 is indicated for about 80 percent of the regions by either first-order dominance tests (more than half of the regions) or second-order tests. In modeling the intake distributions across Ravallion 123 regions, the empirical results strongly indicate that aggregate undernutrition in Indonesia responds to changes in incomes and prices. These results are consis- tent with the view that growth in average household incomes under conditions of nonincreasing inequality and reasonably stable prices for food staples has reduced undernutrition in Indonesia. This conclusion is not inconsistent with recent claims that intake responses of individuals to income changes are quan- titatively small; the induced shifts on intake distributions can be far from small. The results also throw light on the relative importance of the identified factors influencing undernutrition. Both growth in average incomes and reductions in inequality will reduce aggregate undernutrition. Even if inequality increases along with income, an implausibly large contemporaneous increase in inequality would probably be required to outweigh the desirable impact of income growth. In addition adverse effects on national income of transfers from rich to the poor would need to be roughly proportional to the associated changes in the Gini index to wipe out the desirable effect on undernutrition of greater equity. Higher rice prices will have an adverse effect on undernutrition, and this will persist under plausible rural income effects associated with the price change. Clearly, the combined nutritional impact of positive (negative) income growth, a de- crease (increase) in inequality, and a decrease (increase) in food staple prices can be substantial. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Alderman, Harold. Forthcoming. "New Research on Poverty and Malnutrition: What Are the Implications for Research and Policy." In Michael Lipton and Jacques van der Gaag, eds., Including the Poor ? Atkinson, Anthony B. 1987. "On the Measurement of Poverty." Econometrica 55 (4): 749-64. Bhargava, Alok. 1991. "Identification and Panel Data Models with Endogenous Re- gressors." Review of Economic Studies 58 (1): 129-40. Bhargava, Alok, and J. D. Sargan. 1983. "Estimating Dynamic Random Effects Models from Panel Data Covering Short Time Periods." Econometrica 51 (6): 1635-60. Behrman, Jere R. 1988. "Nutrition and Incomes: Tightly Wedded or Loosely Meshed?" PEW/Cornell Lecture Series on Food and Nutrition Policy. Cornell University, Ithaca, New York. Processed. Behrman, Jere R., and Anil B. Deolalikar. 1987. "Will Developing Country Nutrition Improve with Income?" Journal of Political Economy 95 (3): 492-507. Behrman, Jere R., Anil B. Deolalikar, and Barbara L. Wolfe. 1988. "Nutrients: Impacts and Determinants." The World Bank Economic Review 2: 299-320. Biro Pusat Statistik. 1989a. Konsumsi Kalori dan Protein Penduduk Indonesia Per Pro- pinsi 1987. Jakarta: Biro Pusat Statistik. . 1989b. Pengeluaran Untuk Konsumsi Penduduk Indonesia 1987. Jakarta: Biro Pusat Statistik. 124 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Bouis, Howarth, and Lawrence J. Haddad. 1988. "Comparing Calorie-Income Elas- ticities Using Calories Derived from Reported Food Purchases and a Twenty-Four Hour Recall of Food Intakes: An Application Using Philippine Data." Discussion Paper 88. Warwick University, Development Economics Research Centre. Processed. Dandekar, V. M. 1981. "On Measurement of Poverty." Economic and Political Weekly 15: 1241-50. Dasgupta, Partha, and Debraj Ray. 1990. "Adapting to Undernutrition: The Clinical Evidence and Its Implications." In Jean Dreze and Amartya Sen, eds., The Political Economy of Hunger. Oxford: Oxford University Press. Foster, James, J. Greer, and Erik Thorbecke. 1984. "A Class of Decomposable Poverty Measures." Econometrica 52 (3): 761-66. Foster, James, and A.F. Shorrocks. 1988. "Poverty Orderings." Econometrica 56 (1): 173-77. Hadar, Josef, and William R. Russell. 1969. "Rules for Ordering Uncertain Prospects." American Economic Review 59 (1): 25-34. James, W. P. T., and E. C. Schofield. 1988. "Human Energy Requirements: A Manual for Planners and Nutritionists." Food and Agricultural Organization of the United Nations, Rome. Processed. Kakwani, Nanak. 1989. "On Measuring Undernutrition." Oxford Economic Papers 41 (3): 528-52. Osmani, S. R. 1987. "Controversies in Nutrition and Their Implications for the Eco- nomics of Food." WIDER Working Paper 16. World Institute for Development Eco- nomics Research, Helsinki. Processed. Ravallion, Martin. 1990. "Income Effects on Undernutrition". Economic Development and Cultural Change 38 (3): 489-516. U = -. 1991. Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia. LSMS Working Paper 82. Washington, D.C.: World Bank. Ravallion, Martin, and Monika Huppi. 1991. "Measuring Changes in Poverty: A Meth- odological Case Study of Indonesia during an Adjustment Period." World Bank Eco- nomic Review 5 (1): 57-82. Ravallion, Martin, and Dominique van de Walle. 1991. "The Impact on Poverty of Food Pricing Reforms: A Welfare Analysis for Indonesia."Journal of Policy Modeling 13 (2): 281-99. Rothschild, Michael, and Joseph E. Stiglitz. 1970. "Increasing Risk: 1. A Definition." Journal of Economic Theory 2 (3): 225-43. Srinivasan, T. N. 1981. "Malnutrition: Some Measurement and Policy Issues." Journal of Development Economics 8 (1): 3-19. Strauss, John, and Duncan Thomas. 1989. "The Shape of the Expenditure-Calorie Curve." Rand Corporation, Santa Monica, California. Processed. Sukhatme, P. V. 1961. "The World's Hunger and Future Needs in Food Supplies." Jour- nal of the Royal Statistical Society (Series A) 124: 463-525. . 1978. "Assessment of Adequacy of Diets at Different Income Levels." Economic and Political Weekly 13: 1373-84. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 125-138 Alleviating Transitory Food Crises in Sub-Saharan Africa: International Altruism and Trade Victor Lavy The donor community has responded to unexpected or transitory drops in domestic food production in many countries in Sub-Saharan Africa. An empirical framework estimates and analyzes the correlation between this food aid and domestic production. Both emergency food aid and commercial imports are used to offset the effects of negative output shocks in Sub-Saharan African countries-the major recipients of global emergencyfood aid. On average every one-ton drop in cereal production is offset by the delivery of 0.8 tons of cereal and dairy products from abroad (over four years). Most food aid arrives within a year of a shock, and correlation of shocks over time along with the differences between crop years and calendar years may explain the link between aid and production. The economic and political considerations also determine the global response to the emergency food needs of countries in Africa. The interna- tional response is not contingent on the form of government or the level of political and human rights violations. Poorer countries and those with well-established non- emergency food aid programs receive larger amounts of emergency aid when needed. Food aid has long helped low-income countries to sustain domestic food sup- plies and cope with transitory food insecurity, and it has also been used to accelerate agricultural development and increase food production. But food aid has been heavily criticized in the past decade on several grounds. First, it has been charged that the international response to food crises is slow, meager, and inefficient. A related accusation suggests that countries granting emergency food aid discriminate on the basis of the political and economic orientation of the recipient country. Second, food aid has been thought to discourage domestic food production, thus leading to long-term dependence on donors. Finally, it has Victor Lavy is with the Department of Economics at the Hebrew University of Jerusalem. When this article was written, he was a consultant to the Population and Human Resources Department of the World Bank. The article was written under the auspices of the Technical Department of the World Bank's Africa Regional Office as a background paper to the study on Food Aid in Sub-Saharan Africa. The author would like to thank Shlomo Reutlinger for his assistance and useful insight and comments on the paper, the three referees of this journal for their detailed comments and suggestions, and the World Bank Research Committee for its financial support. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 125 126 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I been said that, by alleviating shortages, food aid enables countries to postpone or even cancel politically costly economic reforms (Maxwell and Singer 1979; Stevens 1979; Singer and others 1987; Hopkins 1984; Wallerstein 1980; Bhag- wati 1986; Srinivasan 1989; Maxwell 1989; Lavy 1990). This article investigates empirically the validity of some of these criticisms in the granting of emergency food aid to Sub-Saharan Africa during 1979-87. The response of the donor community to unexpected or transitory drops in domestic food production in 26 countries is examined. Food emergencies are charac- terized here as resulting from unexpected declines in output. However, famines or emergencies are often due to a combination of a breakdown in entitlements and a decline in output rather than just a drop in actual food supply. This study compares the role of food aid and commercial food imports in offsetting these production shocks and covering the shortfall in domestic food supplies. Several hypotheses that postulate the factors determining the response of donors to the needs of different countries are tested. Section I presents the stylized facts of food insecurity in Africa and documents the relative importance of emergency and total food aid by country. Section II suggests an empirical framework for estimating and analyzing the correlation between food aid and domestic production; the results are presented in section III. Section IV analyzes the economic and political determinants of the global response to the emergency needs of countries in Africa, and section V concludes. I. TRANSITORY FOOD INSECURITY: THE STYLIZED FACTS Unexpected temporary reductions in food production can have dramatic ef- fects on food consumption. Several mechanisms are available to stabilize con- sumption, including the commercial importation of food supplies and the grant- ing of emergency food aid. Both options have been used in Sub-Saharan Africa, where domestic food supplies have been unreliable for the past decade. Table 1 summarizes the coefficients of variation (net of trend) of cereal pro- duction during 1970-87. These coefficients are the absolute value of the average annual deviation from the (detrended) mean of production during the sample period, relative to that mean. The countries of the Sahel (Burkina Faso, Ethio- pia, Mali, Mauritania, Niger, Somalia, and Sudan) had the largest annual fluc- tuations, followed by Botswana, Madagascar, and Tanzania. The coefficients reported are averages; the variability in the latter years was much higher than average, because of the 1983-85 drought. Output for even the most stable food producers in Africa fluctuates more sharply than in most other developing coun- tries (World Bank 1988). Cereal production in Sub-Saharan Africa was not only highly variable but also declined per capita during the sample period. Table 2 gives regional annual growth rates for grain production and population from 1970 to 1982. East and West Africa had the highest rates of population growth and the lowest rates of cereal production growth, which resulted in a decline in per capita cereal pro- Lavy 127 Table 1. Statistics on Food Production, Aid, and Imports, 1970-87 Average Average Variability of enmergency totalfood Average commercial food production food aid as a aid as a food imports as (coefficient percentage of percentage of a percentage of Country of variation) totalfood aid food production food production Angola 21.3 50 9 84 Burundi* 21.8 9 10 4 Benin* 22.6 18 2 19 Burkina Faso* 39.2 25 3 8 Botswana 77.6 13 195 808 Central African Republic 15.9 10 2 17 Chad 20.3 59 8 10 Cameroon* 13.8 21 1 18 Comoros* 20.0 5 19 124 Congo* 35.8 0 10 862 Cape Verde 99.8 77 1,450 219 Ethiopia 34.6 59 9 8 The Gambia 27.7 34 10 57 Ghana 23.9 18 7 30 Guinea-Bissau 56.3 50 13 24 C6te d'Ivoire* 23.7 20 1 26 Kenya 32.4 23 2 11 Lesotho* 26.4 6 20 82 Liberia* 22.8 0 0 38 Madagascar 26.4 27 1 10 Mauritius 105.7 2 805 9,530 Mauritania 43.4 54 111 525 Mali 36.8 31 6 13 Malawi 33.5 20 1 2 Mozambique 13.7 68 9 60 Rwanda 21.1 23 3 6 Senegal 29.9 45 10 65 Sierra Leone 12.2 S 2 18 Somalia 42.6 81 29 71 Sudan 48.8 60 6 20 Tanzania 52.8 26 3 8 Togo* 18.4 8 2 19 Uganda 33.9 47 1 3 Zaire 25.7 46 1 35 Zambia 25.1 67 1 23 Zimbabwe 43.8 87 1 6 * These countries are not included in the empirical analysis in section II. Source: Author's calculations based on FAO data. duction of 2.2 and 0.8 percent a year, respectively. Positive trends were reported in all other regions, with cereal production up 1.8 percent both in East Asia and the Pacific Rim and in the industrial economies. Africa's negative growth rates during this period were somewhat offset by an increase in food aid and commercial food imports, which augmented total do- 128 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 2. Growth of Cereal Production and Population by Region, 1970-82 (average annual percentage change) Cereal Cereal production Region production Population per capita World 2.3 1.8 0.5 Industrial economies 2.5 0.7 1.8 East Africa 0.8 3.0 -2.2 West Africa 1.9 2.7 -0.8 East Asia and Pacific 3.5 1.7 1.8 South Asia 2.7 2.4 0.3 Latin America 3.2 2.4 0.8 Source: World Bank (1986). mestic production by an average of 12 and 15 percent in East and West Africa, respectively. As seen in table 1, Botswana, Cape Verde, Mauritania, Mauritius, and Somalia depended heavily on food aid, and Botswana, Cape Verde, Com- oros, Congo, Mauritania, and Mauritius depended heavily on commercial food imports. Cross-section correlations indicate that countries with a high coeffi- cient of variation of production receive more aid (as a percentage of total domes- tic production). The simple correlation coefficient between the variation of production and the ratio of food aid to domestic production is 0.45. This result is even stronger when the ratio of total food aid is replaced by emergency food aid. It appears that low domestic output triggers donor aid-mostly emergency help. Countries that are receiving food aid, and therefore are experiencing food shortages, also tend to import part of the food deficit. The simple correlation coefficient between the share of commercial food imports and the share of total food aid in production is 0.25, and, when emergency food aid is substituted for total food aid, the correlation is 0.31. II. THE EMPIRICAL FRAMEWORK This section describes the relationship between transitory changes in domestic production and emergency food aid or commercial food imports. Since negative shocks to domestic food production are assumed to be exogenous and are the primary trigger of international emergency food aid, which is assumed to be endogenous, the potential simultaneity between food aid and food production is not problematic. The same applies for transitory output shocks and their effect on commercial food imports. Here, however, negative as well as positive shocks can lead to a change in imports. As a result trade can serve as a symmetric stabilizer of food consumption, whereas emergency aid is used only as a buffer during food shortages. The trade mechanism works mainly through changes in imports from some average level, since food exports are negligible for most African countries. Some exceptions, such as Botswana, Kenya, Malawi, and Zimbabwe, are food exporters on either a continuing or at least sporadic basis. Also some food aid is in fact being given by way of "triangular transaction;' for Lavy 129 example, the United Kingdom gives food aid under the International Food Convention through purchases of maize from Zimbabwe that it sends to Zambia. More iFormally, let Yt denote the transitory component of food production in year t, Et denote emergency food aid, and M, denote emergency food imports. The last is the deviation of commercial food imports from trend and excludes emergency food aid. Response functions for aid and commercial imports in year k have the form k E ¢Yt + Ut, if yy7 t=O and y is a threshold below which an output shock will trigger an inflow of food aid or commercial food imports. This threshold is probably different in the two response functions, may not be equal to zero, and may vary from country to country. The emergency aid response model can be treated as a limited depen- dent variable model, with emergency aid taking only nonnegative values. Emer- gency food aid in any year may be positive even if the transitory output shock in the same year does not cross the threshold, because the level of emergency food aid in a given period includes a lagged response to previous shocks. An alternative view of the response model is a switching regime model. One regime is the case of large negative output shocks triggering current (and future) food aid. The other is the case of small negative shocks or positive shocks that do not lead to aid flows. Several difficulties arise in estimating such a model. First, since the threshold level is unknown, the switching model is with un- known regimes. Second, the transitory food production shocks are unobserved and must be estimated. Third, the form of the emergency food aid and emer- gency food import response functions may be jointly dependent, so that the coefficients should be jointly estimated. Fourth, the lagged response of foreign aid and imports to output shocks may be very important, implying the need to allow for an unrestricted lag structure. To obtain a measure of the output shocks, an autoregressive model with a time trend for food production is estimated separately for each country. The residuals from these regressions are then used in the two response functions. In the emergency food aid response function, a spline is used so that only the negative shocks are needed to explain the variance of emergency aid, thus con- straining the coefficients on positive output shocks to be zero. This means that 130 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I the threshold is constrained to zero. This approach is preferred over one that is based on estimating a switching model with an unknown threshold, because it allows greater freedom in exploring both the lag structure and the possibility that the coefficients vary across countries. Some flexibility in the response can be achieved by introducing nonlinearity in the effect of output shocks on emergency food aid. It might be expected that the response for small negative shocks is minimal or even nil, but increases with the severity of the negative shock. Adopting this approach of making the response a quadratic function of the size of the shock allows for different responses for different countries, as long as the countries' shocks are of different size. The equations for emergency food aid and emergency food import equations can be estimated jointly, using generalized least squares techniques to allow for an unrestricted error structure. As long as the set of explanatory variables is not identical in the two equations, efficiency is gained through joint estimation, which exploits the cross-equation correlation. If the aid response equation in- cludes only the negative output shocks while the import response equation includes both the negative and positive shocks, this condition is met. The crucial variable needed to estimate the model is emergency food aid. This variable is available for 1979-87 for all the Sub-Saharan countries. Given this short time series, a separate estimation for each country is not possible, and therefore data were pooled. As noted above, differences in the responses across countries may exist. Since the response functions do not include constants, models with individual effects (different intercepts) are inappropriate. An alter- native approach is to allow for varying slopes by country, with response being a function of characteristics of the recipient country. The response functions are therefore augmented with an interaction variable between the output shocks and country characteristics as detailed below. III. INTERNATIONAL ALTRUISM: THE EXTENT AND SPEED OF THE RESPONSE In the aid data for this study (unpublished data supplied by the World Food Programme), food aid is divided into three categories: emergency aid, program aid, and project aid. Each of these categories is subdivided into cereal and noncereal aid. All figures are in tons of grain or grain equivalent. Since data on Africa often suffer from large measurement errors, price and output data must be interpreted with care. To minimize the use of unreliable data, the sample was limited to 1979-87, for which careful attention was given to the quality of the figures. Because the only extended food output and input data available for most of Sub-Saharan Africa are for total cereal production, these are used to estimate random shocks to domestic production and imports. Production and imports are not an adequate estimate of consumption, however, due to the importance of changes in stocks in agrarian societies. The problem, of course, is that the available stock data usually refer only to public stocks, whereas most stocks are actually held by farmers or consumers. Therefore, the stabilization of consump- Lavy 131 Table 3. Estimation Results of Response Functionsfor Cereal and Noncereal Emergency Food Aid Cereal Noncereal With constant Without With constant Without term constant term term constant term Constant -783.0 n.a. 532.4 n.a. (0.0) (1.2) Positive output shock Contemporaneous 0.027 0.026 -0.001 0.001 (0.6) (0.6) (0.1) (0.1) Lagged one year -0.017 -0.017 0.001 0.001 (0.4) (0.4) (0.1) (0.2) Lagged two years -0.126 -0.126 -0.011 -0.011 (1.9) (1.9) (1.4) (1.4) Lagged three years 0.048 0.048 0.003 0.004 (1.5) (1.5) (0.7) (0.9) Negative output shock Contemporaneous -0.073 -0.073 -0.008 -0.009 (1.5) (1.5) (1.3) (1.5) Lagged one year -0.276 -0.276 -0.026 -0.027 (5.9) (6.0) (4.4) (4.6) Lagged two years -0.037 -0.037 -0.003 -0.003 (0.8) (0.8) (0.5) (0.6) Lagged three years -0.124 -0.124 -0.013 -0.014 (2.7) (2.8) (2.2) (2.4) R2 0.387 0.471 0.236 0.362 Sample size 126 126 126 126 F-statistic 10.875 15.025 5.826 9.924 n.a. Not applicable. Note: t-statistics are in parentheses. Source: Author's calculations, based on unpublished data from the World Food Programme. tion from current supplies is modeled here. And because several countries re- ceived no emergency food aid or received aid for only a short period, the sample includes only those 26 countries, listed in table 1 without asterisks, that received some emergency food aid for an extended period. Observations that lacked values for emergency aid were discarded, leaving 126 usable observations. Response functions for cereal and noncereal emergency food aid were first estimated using the contemporaneous and three lagged values of the output shocks, where both negative and positive shocks were included, as well as a constant term (table 3). As suggested earlier, positive shocks to output should have no effect on emergency aid, and indeed the coefficients on the positive shocks were not statistically significant. The evidence suggests therefore that the addition of contemporaneous and lagged positive output shocks, regardless of the lag length, do not contribute to the explanation of variations in current emergency aid flows. The constant term was also insignificant, meaning that there is no autonomous level of emergency aid. Table 4 presents the generalized least-squares estimates of the emergency food aid response functions to negative output shocks for cereals and noncereals. Experiments with various lag patterns and length show that all emergency aid is 132 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 4. Response of Emergency Food Aid to Negative Output Shocks Estimated equation With one lag With two lags With three lags With Without With Without With Without constant constant constant constant constant constant term term term term term term Cereals Constant 15,223.0 n.a. 5,671.8 n.a. 2,107.4 n.a. (1.6) (O.S) (0.2) Negative output shock Contemporaneous -0.031 -0.046 -0.037 -0.043 -O.OOS -0.007 (1.0) (1.5) (1.2) (1.5) (0.2) (0.2) Lagged one year -0.275 -0.292 -0.253 -0.257 -0.278 -0.280 (7.1) (7.8) (6.6) (6.9) (7.1) (7.3) Lagged two years n.a. n.a. -0.102 0.109 -0.071 -0.073 (2.9) (3.3) (1.9) (2.0) Lagged three years n.a. n.a. n.a. n.a. -0.091 -0.092 (2.2) (2.3) R 2 0.315 0.40 0.355 0.44 0.374 0.46 Sample size 126 126 126 126 126 126 F statistic 19.7 27.78 Noncereals Constant 2,873.3 n.a. 2,052.0 n.a. 1,656.3 n.a. (2.5) (1.7) (1.4) Negative output shock Contemporaneous -0.006 -0.009 -0.007 0.009 -0.003 -0.005 (1.7) (2.5) (1.9) (2.5) (0.8) (1.1) Lagged one year -0.023 0.027 0.029 -0.023 -0.024 -0.026 (5.0) (5.8) (4.6) (4.9) (5.0) (5.4) Laggedtwoyears n.a. n.a. -0.009 -0.011 -0.005 -0.007 (2.5) (2.8) (1.2) (1.5) Lagged three years n.a. n.a. n.a. n.a. -0.010 -0.091 (2.0) (2.2) R2 0.204 0.312 0.228 0.350 0.245 0.367 Sample size 126 126 126 126 126 126 F-statistic n.a. n.a. n.a. n.a. 11.131 19.283 n.a. Not applicable. Note: t-statistics are in parentheses. Source: Author's calculations, based on unpublished data from the World Food Programme. received within four years of the initial shock, so that the estimates using three lagged values of output shocks are discussed here. For cereals the contem- poraneous partial correlation is negative, but not significantly different from zero. The sum of the coefficients on the contemporaneous and lagged shocks is -0.45, suggesting that every one-ton reduction in domestic grain production is compensated by almost half a ton of grain in the form of emergency food aid. Most of this-more than 60 percent-arrives the year after the negative shock. This lag in response may be the result of sluggish reaction by the donor commu- nity, but, since the output shocks may be correlated, they could also induce a lagged effect on emergency food aid. For example, if the output shock is caused by a severe drought that forces farmers to temporarily leave their farms, the following year's crop could also be affected. Lavy 133 The timing of the response for emergency noncereal aid is similar to that for cereals: the contemporaneous response is small, with more noncereal aid arriv- ing in the years after the negative shock. The total response of noncereal aid is smaller, however, with only 0.13 of a (grain-equivalent) ton of noncereals being received in aid in the three years following a negative shock. For some countries for some years there are several missing values of emer- gency food aid. They are probably zero, but they may be unreported. In order not to throw the country out of the sample, if each missing value is replaced by a zero value, the sample size increases, and the accumulated aid response declines (to 0.35 for cereals and 0.04 for noncereals). For these estimates of emergency aid response, recall that the threshold that triggers a response has been constrained to equal zero. This implies that every reduction in output beyond that expected according to the long-term trend is assumed to result in emergency aid. In reality, however, emergency aid is pro- vided only when the reduction in output falls beyond a threshold that is less than zero. If ithis threshold was allowed empirically to be different from zero, the response would be larger than has been estimated here. Indeed higher responses were estlimated when negative threshold levels were chosen, but the choice of any threshold is arbitrary. An alternative to using an arbitrary negative threshold is to allow for nonlinearity in the effect of the negative shocks. When the squared value of the output shock was included as a regressor, its coefficient was negative and significantly different from zero, which implies that the aid re- sponse is more vigorous as the negative shocks are more severe, thus leading to a larger total response. With respect to the sluggishness of the aid mechanism, the crop year in which output is recorded and the calendar year in which aid is recorded do not corre- spond exactly; some of the aid recorded as arriving in the year following the shock actually arrived contemporaneously. So emergency food aid insulates do- mestic food grain consumption from random shocks to domestic production to a larger degree than the results here imply. But aid that arrives in the years follow- ing a shock may cause problems. If the output shock has a short life, the late emergency aid may actually arrive during a period of normal food production and may depress food prices. This effect on prices will be mitigated, however, if the late emergency aid serves to replenish stocks. The estimates here suggest that this "late" (more than a year after the shock) flow is small relative to both total aid and total production, and thus cannot be very harmful. Random shocks to domestic production may also induce a stabilizing re- sponse from commercial imports as domestic food prices increase relative to food prices elsewhere. This mechanism reduces food shortages in a free trade regime. Because many African governments do not, in fact, allow free trade and control imports and food prices, changes in relative prices may be very limited. The import response may nonetheless occur if governments are willing and able to buy on the world market. Table 5 summarizes the parameters of the import response to production shocks. The dependent variable is the deviation- 134 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 5. Response of Cereal Imports to Current and Lagged Production Shocks All imports Commercial imports With constant Without With constant Without term constant term term constant term Constant -19,705.3 n.a. -22,300.4 n.a. (2.5) (2.2) Positive output shock Contemporaneous 0.001 -0.014 -0.036 -0.052 (0.0) (0.4) (0.8) (1.2) Lagged one year -0.043 -0.051 -0.011 -0.022 (1.1) (1.2) (0.2) (0.4) Lagged two years -0.080 0.078 -0.095 -0.095 (1.5) (1.S) (1.4) (1.4) Lagged three years 0.0S0 0.043 0.027 0.019 (1.9) (1.6) (0.9) (0.6) Negative output shock Contemporaneous -0.106 -0.091 -0.048 -0.034 (2.7) (2.3) (0.8) (0.6) Lagged one year -0.170 -0.158 -0.288 -0.274 (5.0) (4.6) (6.0) (5.6) Lagged two years -0.054 0.047 -0.063 -0.057 (1-6) (1-3) (1-3) (1.2) Lagged three years 0.014 0.026 -0.045 -0.032 (0.4) (0.8) (1.0) (0.7) R2 0.352 0.330 0.424 0.408 Sample size 144 144 126 126 n.a. Not applicable. Note: t-statistics are in parentheses. Source: Author's calculations, based on unpublished data from the World Food Programme. positive or negative-from the trend of imports derived from estimating the autoregression of imports for each country. Since the dependent variable is a deviation from trend, one would expect the constant term of the regression to be zero, but, for estimating both commercial imports and all imports, the estimated constant term was significantly different from zero. Results with and without the constant are reported in table 5. The evidence that trade serves as a symmetric buffer indicates that the trade buffer is stronger when there are shortages (through imports) than when there are surpluses. Coefficients on positive shocks are generally negative, as expected, but are not statistically significant. In the case of negative production shocks, shocks lagged more than one period do not have significant coefficients, and commercial im- ports do not respond contemporaneously to a negative shock. The response of commercial imports in the period following the negative shock is, however, strong and significant. Again this apparent lag may result from the difference between the calendar year and the crop year. IV. ECONOMIC AND POLITICAL DETERMINANTS OF DONOR GENEROSITY The global response to the need for emergency aid varies, leading to the hypothesis that the total flow of emergency aid from all donor countries and Lavy 135 international organizations may be influenced by political and economic consid- erations. To identify the factors that determine the global response to food needs in Sub-Saharan Africa, the aid response equation was reestimated with the addition of interaction terms between current negative output shocks and other political and economic variables that may reasonably be thought to affect the responsiveness of donor countries. Several hypotheses are tested here. The first is that poorer countries receive more generous responses. The poverty level of a country is measured by per capita income and food consumption. The second hypothesis is that donors may be more responsive to the needs of countries that already receive large amounts of food aid. This hypothesis is based on the assumption that established food aid flows have organized channels of transmission, making emergency aid more effective and less costly. A related factor is the cost of transportation, which in Africa is mostly a function of the land component of transportation. Inland countries may receive less aid because transportation costs are higher for them. The third hypothesis is that the donors' response may depend on the political orientation of the recipients. The socialist countries in the sample include An- gola, Ethiopia, Guinea, Mozambique, and Tanzania, and, in testing this hy- pothesis, this group was enlarged to include the mixed socialist countries of Madagascar, Mali, Rwanda, Somalia, Sudan, and Zambia. The fourth hypoth- esis is that the political rights and civil and economic liberties in the recipient country may influence the donor community. Emergency aid may also depend on the form of government. To test these hypotheses, we used Gastil's (1984) ranking of African countries according to political rights, civil liberties, and state of freedom. The first two indexes rank countries from one (best) to seven (worst), while the third index characterizes countries as not free, partially free, or fully free. Most countries in Africa fall between five and seven on the first two indexes. The exceptions appearing in the data set used here are Botswana, Ghana, Mauritius, Senegal, and Zimbabwe. The countries ruled by a military regime are Central African Republic, Chad, Ethiopia, Ghana, and Mauritania. Given the extensive multicollinearity between all these variables, their impact on aid response was estimated one at a time. When the effect on the response was suspected to originate from correlation between the interaction term and a third variable, that variable was added as a control in the regression. Table 6 summarizes the qualitative results of including the interaction terms in the aid equation for cereals and noncereals. It indicates the sign of the effect on aid and whether it is different from zero at the 5 percent significance level. The signs of the coefficients are the same for the two types of food aid, which is not surpris- ing given the high correlation between the two. Almost all of the coefficients are statistically different from zero. For economic determinants the hypothesis that the response is better for poorer countries is accepted for both cereal and noncereal aid. Replacing the gross national product per capita as a measure of poverty with other poverty measures, such as per capita food production or per capita daily cereal con- 136 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 6. Factors Affecting the Response of Aid and Trade Emergency Commercialfood imports Factor Cereal Noncereal (negative shocks) GNP per capita _ Nonemergency food aid + * + * Access to sea + * + * + * Socialist + * + * + * Semisocialist + + + Military government + + * + * Political rights + * + * + * Civil liberties + * + * + * Status of freedom + + * + * Note: + means a positive correlation; - is negative. * denotes cases in which the interaction term was significantly different from zero (at a = 0.05). sumption, did not change this result in any significant way. The aid response varied greatly with the average level of nonemergency food aid received (as a share of domestic production); countries with a high share of nonemergency food aid received more aid in crisis situations than did countries with low shares of nonemergency aid. If low-income countries tend to receive more non- emergency food aid, however, the correlation between emergency aid and in- come per capita could be erroneous given the high collinearity between the two components of food aid. Nevertheless the nonemergency aid effect in the response equation remained unchanged, even when income per capita was controlled for in the estimation. This robust result can be interpreted as follows: the response to emergency needs is positively correlated with the ease (low cost) of delivering food aid, and the cost of transportation and delivery tends to be lower for countries that receive high levels of nonemergency aid. This conclusion is further supported by the significant positive correlation between good geographical location (access to the sea) and the response to output shocks. Although these results suggest that the emergency needs of inland countries are not met with the same generosity as is extended to those with seaports, it is possible that the emergency needs of inland countries are met by means other than food aid. The results for the political variables are quite surprising. Countries classified as socialist, with military governments, and with a low score in terms of protect- ing political, economic, and civil liberties tend to receive more aid in emergency or crisis situations. This tendency is probably a result of the high correlation between some of the other determinants of the aid response and the political and human rights factors. Indeed socialist countries and countries with restricted political and economic freedoms are at the lower end of the income distribution in Africa. They receive more aid because they are poorer, and the international response during crises is not contingent on the form of government or the level of political and human rights violations. Regular flows of food aid, however, are highly negatively correlated with socialist or military governments and with poor protection of civil, economic, and political freedoms. These results, how- Lavy 137 ever, likely suffer from sample selectivity biases because some countries with severe food crises did not receive any emergency aid and therefore are not included in the sample (for example, Burkina Faso and Congo). Table 6 shows the results from reestimating the import response function using only the negative output shocks (the positive shocks were not significant) and their interaction with all the political variables. Surprisingly, this estimation is almost identical to the aid equation. Although the market-oriented economies were expected to be more likely to import food for emergency needs, the results here suggest that emergency imports are higher for the socialist countries. This somewhat unexpected result could reflect the fact that the very poor socialist countries finance emergency commercial imports with general financial aid or with balance of payments relief aid. V. CONCLUSIONS This article has provided empirical evidence that throws doubt on the validity of some often heard criticisms of food aid. Both emergency food aid and com- mercial imports are used to offset the effects of negative output shocks in Sub- Saharan Africa. Within one year of an unexpected one-ton drop in cereal pro- duction, about 60 percent of this shortfall is replaced by increased food aid and commercial imports. Food aid continues to flow in response to the shock for another two years, although at a much reduced rate. Several studies have interpreted the negative simple correlation between food aid and domestic food production as an indication that food aid depresses domestic production. The results here suggest that this correlation may be due to the contemporaneous response of emergency food aid flows to output shocks. Most food aid arrives within a year of a shock, and correlation of shocks over time along with the differences between crop years and calendar years may explain the link between aid and production without resorting to the discourage- ment hypotheses. The dynamics of domestic food production cause and trigger the flow of emergency food aid, so these flows should be treated as endogenous. Emergency aid should be netted out of total food aid in an exercise that evalu- ates the efficiency of food aid. This analysis of African countries-the major recipients of global emergency food aid--suggests that on average a shock to agricultural output in the form of a one-ton drop in cereal production would lead to a flow of 0.64 tons within a year. Extending the period to four years increases the compensation to 80 per- cent. In other words, every one-ton drop in cereal production is offset by the delivery of 0.8 tons of cereal and dairy products from abroad. There is, how- ever, a lag in this response over four years, although most of the aid is received in one to two years. This estimate is biased downward, however, since the thresh- old that was assumed to trigger emergency flows was only zero. Contrary to criticism, the pattern of emergency aid flows in Sub-Saharan Africa does not provide evidence of discrimination by donors on the basis of 138 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I political factors. The international response to food crises in Sub-Saharan Africa is not contingent on the form of government or the level of political and human rights violations. On the contrary, countries classified as socialist with military governments and with a low score in terms of protecting political, economic, and civil liberties tend to receive more aid in emergencies or crises. Poorer countries and those with well-established nonemergency food aid programs re- ceive larger amounts of emergency aid when needed. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Bhagwati, Jagdish. 1986. "Food Aid, Agricultural Production, and Welfare." In S. Guhen and M. R. Shroff, eds., Essays on Economic Progress and Welfare, in honor of I. G. Patel. New Delhi: Oxford University Press. Gastil, Raymond D. 1984. Freedom in the World: Political Rights and Civil Liberties, 1983-84. Westport: Greenwood Press. Hopkins, Raymond. 1984. "The Evolution of Food Aid." Food Policy 9 (4): 345-63. Lavy, Victor. 1990. "Does Food Aid Depress Food Production: The Disincentive Di- lemma in the African Context.' PRE Working Paper 460, World Bank, Population and Human Resources Department, Washington, D.C. Processed. Maxwell, Simon. 1989. "The Disincentives of Food Aid: A Pragmatic Approach." Insti- tute of Development Studies, University of Sussex. Processed. Maxwell, Simon, and Hans W. Singer. 1979. "Food Aid to Developing Countries: A Survey," World Development 7 (3): 225-47. Singer, Hans, John Wood, and Tony Jennings. 1987. Food Aid: The Challenge and the Opportunity. Oxford: Clarendon Press. Srinivasan, T. N. 1989. "Food Aid: A Cause of Development Failure or an Instrument for Success?" World Bank Economic Review 3 (1): 39-65. Stevens, Christopher. 1979. Food Aid and the Developing World: Four African Case Studies. New York: St. Martin's Press. Wallerstein, Mitchel B. 1980. Foodfor War, Foodfor Peace. Cambridge: MIT Press. World Bank. 1986. Poverty and Hunger: Issues and Options for Food Security in Devel- oping Countries. Washington, D.C. .1988. The Challenge of Hunger in Africa: A Call to Action. Washington, D.C. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1: 139-154 The Working Behavior of Young People in Rural Cote d'Ivoire Rob Alessie, Paul Baker, Richard Blundell, Christopher Heady, and Costas Meghir One of the major features of structural adjustment is an attempt to reallocate labor- and hence output-through changes in relative prices. This article assesses how price changes affect the working patterns of young people in rural C6te dI'voire. The analysis is based on a model of the labor supply of rural households and on the construction of composite price indexes. The data come from the C6te dIvoire Living Standards Survey for 1985 and 1986. The panel aspect of the data allows the work choice made in one year to depend on the individual's choice in the previous year. Results indicate that the price of agricultural output generally is a positive incentive on the decision to partici- pate in the labor force. However, this result depends heavily on the employment and education of the individual in the previous period. Those not already working are less likely to respond to favorable movements in the prices of cash crops by entering the work force. This article reports on research on the determinants of working patterns among young people in rural C6te d'Ivoire. Because young people's decisions about working are closely related to decisions about schooling, the article also focuses on the relationship between work and school. The motivation for this study arises from an interest in the effects of structural adjustment programs on work incentives. Structural adjustment is expected to increase output by altering the incentives facing producers. The effect of improved incentives could, in princi- ple, be measured by directly observing changes in output. The time lag between the change in incentives and the change in output can be substantial, however, especially when tree crops are involved, so that data over a long period may be Rob Alessie is with the Department of Economics, University of Tilburg, Netherlands; Paul Baker is with the Institute for Fiscal Studies, London; Richard Blundell and Costas Meghir are with the Depart- ment of Economics, University College London and the Institute for Fiscal Studies; and Christopher Heady is with the School of Social Sciences, University of Bath. This article is the final report of a project financed by the World Bank on labor market transitions and structural adjustment in C6te d&lvoire. The work was undertaken while Rob Alessie was visiting the University College London from Tilburg. The authors are grateful to the World Bank for financial support on this project and to Francois Bourguignon, Paul Glewwe, Valerie Kozel, John Newman, Jacques van der Gaag, and three anonymous referees for their helpful comments. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 139 140 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I necessary to implement this approach. Changes in incentives will also affect input use, and even for tree crops these changes will be more immediate and therefore more readily observable in data from a short period. This article concentrates on labor input. A microeconomic approach that examines the labor supply decisions of indi- viduals and households has several advantages over an aggregative approach that simply focuses on overall labor participation rates. First, the micro- economic approach provides more units of observation and therefore permits the separate identification of the effects of a greater number of determinants of labor supply. Second, this approach utilizes heterogeneity in the population rather than aggregating across groups, so that empirical results are richer. Third, aggregate labor participation rates may mask the dynamics of labor supply. Small net changes in participation between two years may result from large numbers of people joining and leaving the labor force. Understanding these gross flows into and out of the labor force is essential to predicting the effects of policy changes on the overall participation rate. This article concentrates on young people for two reasons. First, the young undergo more transitions in labor market status than do older cohorts, which suggests that there is a considerable amount of choice about whether or not young people work. This is not to say that the individual exercises the choice. Decisions about work and school may be made jointly with other members of the household or even just by the head of the household. Second, it is possible that the decision to work may be accompanied by a decision to leave school or at least to reduce time in school. In this case the potential long-term cost of reduced education must be considered in addition to the short-run gain of increased output. This long-term cost is likely to be more serious if it is difficult to reenter school. This possibility should be explored in future research. The factors influencing labor force participation are expected to be different for urban and rural young people, so it is appropriate to examine each sector separately. The rural sector was chosen for this analysis because it is larger and is the sector in which structural adjustment is expected to have the greatest impact on labor supply and output. The data used in this study come from the Cote d'Ivoire Living Standards Survey (CILSS) for 1985 and 1986. C6te d'lvoire is particularly suited for this analysis because it has been carrying out a World Bank-sponsored structural adjustment program, which started in 1981 (Deaton and Benjamin 1987; Glewwe and de Tray 1988). For the CILSS half of the households surveyed in 1985 were reinterviewed in 1986, thus providing an opportunity to observe how individuals have adjusted their labor supply in response to changes in the eco- nomic environment. These observations of individual transitions permit an anal- ysis of dynamic behavior that cannot be identified in cross-section data and enable an assessment of the influence of past choices on current behavior. This study is among the first to make use of the panel aspect of the CILSS. It therefore Alessie and others 141 has the added interest of testing whether these data are sufficiently rich and accurate to allow the estimation of behavioral responses to economic change. Section I outlines the theoretical basis for the labor supply models that are estimatecl. Section II describes the data, including a detailed explanation of how price indexes for individual households are constructed. Section III presents the estimation results, and section IV summarizes the findings and indicates direc- tions for further research. I. THE THEORETICAL MODEL OF THE LABOR SUPPLY OF RURAL HOUSEHOLDS Most of the rural households sampled in the CILSS are engaged in agricultural production, and their labor supply consists almost entirely of work on their own farm. It is therefore appropriate to analyze labor supply decisions in the context of agricultural household models as described in Singh, Squire, and Strauss (1986). These models represent behavior as the result of collective household decisions about production, labor supply, and consumption. The Separable Model In general, decisions about household production and labor supply are inter- dependent and must be modeled simultaneously. However, under the assump- tion of complete competitive factor and product markets, the decisions have a recursive structure. Production decisions are made on the basis of maximizing the profit of the farm unit, which produces income for the household. House- hold labor supply and consumption are then chosen to maximize collective household utility, given consumer prices, wages, and other income. This model of labor supply can be expressed as (1) L, = F(cp,w,I,z) where L, is a measure of labor supply, cp is an index of consumer prices, w is the market wage rate, I is profit and other nonlabor income, and z is a vector of household characteristics. In this "separable" model, agricultural producer prices will affect household labor supply only indirectly through nonlabor income. Increased output prices will raise profits and, assuming that leisure is a normal good, may reduce household labor supply. Because of the recursive structure of the model, the household may simultaneously decide to increase output and total labor input in response to increased prices: the difference between total labor input and house- hold labor supply is covered by labor market transactions. Consider the incentive effects of structural adjustment in the separable model. The suggestion that increased agricultural prices will reduce household labor supply relies on the implicit assumption that wages are constant. In fact one would expect the increased demand for labor that follows from increased agri- cultural prices to push wages up. These increased wages will have both income 142 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 and substitution effects. The substitution effect will tend to increase household labor supply as leisure becomes more expensive. The direction of the income effect will depend on whether the household is a net buyer or net seller of labor. In either case the income effect will be proportional to the extent of net labor purchases (or sales) and so will be small for the typical Ivorian farm household, which buys or sells relatively small quantities of labor. Thus, in contrast to the situation of nonfarm households, the substitution effect of wage increases can be expected to dominate the income effect and provide the main route through which increased agricultural prices might stimulate labor supply. Because the wage rate is an exogenous variable in equation 1, the effect of output price changes on labor supply cannot be captured by simply estimating that relationship. The analysis requires the addition of a relationship between product prices and the wage rate: (2) w = G(pp,x) where pp is an index of product prices, and x is a vector of other factors affecting wages. The producer price index in equation 2 will be composed of the prices of all crops produced using labor from the local market in order to capture the role of prices in determining the demand for labor. Since many crops have long periods of production, however, labor demand will depend in part on expected future prices for the product. Ideally, therefore, the price index should also incorporate these expectations. Equations 1 and 2 could be estimated separately and then combined to pro- vide an estimate of the effect of product prices on labor supply. Alternatively the equations could be combined in a reduced form that could be estimated directly: (3) Ls = H(cp,pp,I,z,x). Although the "structural" estimation of equations 1 and 2 would reveal more detailed information about household behavior and labor market response, the estimation of equation 3 will provide the direct answer to the main question concerning the response of labor supply to structural adjustment. Also equation 3 can be estimated without data on market wage rates, which is a considerable advantage because there is little good quality data on local rural wage rates in Cote d'Ivoire. Finally, as will be shown below, the distinction between equations 1 and 3 becomes blurred as soon as the assumption of perfect product and labor markets is relaxed. More General Models There are two main ways in which the labor market might violate the assump- tion of completeness. First, the labor market may be very thin-or even nonexistent-so that households may have difficulty in either buying or selling labor. The CILSS data suggest that labor market activity in C6te d'Ivoire is rather limited. Landless laborers frequently obtain access to land through sharecrop- Alessie and others 143 ping rather than through wage labor. Second, hired labor may not be a perfect substitute for household labor because of the need for greater supervision. It is very difficult to obtain empirical evidence about the substitutability between household and hired labor, but the possibility of their being less than perfect substitutes has theoretical plausibility. Both oiF these violations imply that household labor supply is not a function of the market wage, but of a shadow wage reflecting the marginal value of extra household labor. This shadow wage will be influenced by the market wage, if there is a labor market, but will depend also on factors that influence the supply and demand for labor within the household. On the demand side, output prices and the amount of available land would be positively related to the shadow wage. On the supply side, the number of household members who are prepared to work would be negatively related to the shadow wage. The number available for work is endogenous, however, so that the nonseparable equivalent to equa- tion 1 is (4) Ls =f(cp,I,W,pp,LAND,z) where LAND is the quantity of land available to the household. Two of the variables in equation 4 have a somewhat different interpretation here. First, farm profit is no longer well defined because of the difficulty of valuing household labor. It is possible to define a shadow profit, based on the shadow wage, and its value will depend on the variables already included in equation 4. There is therefore no need to include a separate term for profits, so that I simply denotes nonfarm income. Second, the producer price index pp should now represent the current and expected future prices of goods produced by this particular household, because those prices determine the demand for labor by the household. This suggests that each household should have its own producer price index, with weights reflecting the relative importance of the different crops in its output mix. Once again, if there is a labor market, it is possible to substitute for w, using equation 2 to obtain the reduced form: (5) L, = h(cp,I,pp,LAND,z,x). Strictly speaking there should be two producer price indexes, one specific to the household and the other representing labor demand in the whole local labor market. In the empirical work the household-specific index is as described in section II, whereas differences in local labor markets will be captured by re- gional dummies. Also it is possible that the area of land farmed is also a choice variable of the household. In that case a full analysis of labor force participation would also require modeling land area decisions. The introduction of producer prices into equation 4 is the most interesting consequence of using the nonseparable model. The absence of producer prices from equation 1 constituted the major difference between it and the reduced- form equation 3, which reflects the fact that producer prices have no direct effect 144 THE WORLD BANK ECONOMIC REVIEW. VOL. 6, NO. 1 on labor supply in the separable model. In the more general model, producer prices have a direct effect on household labor supply, in addition to any effect through the market wage rate. Modeling Individual Labor Supply The models presented so far describe the determinants of the household deci- sion about overall labor supply. This article, however, concentrates on the labor supply of only part of the household: young people. It is therefore necessary to modify the earlier analysis to focus on the labor supply of a subset of the family. The main consideration is the effect of interdependent behavior of the household members. In particular the working status of other family members will affect the labor force participation of young people for three reasons. First, the work- ing status of other family members will affect the total income of the household. Second, for a given household size, the number of members at work will affect the need for young people to perform household duties and so stay out of the labor force. Third, in the nonseparable model, the number of household mem- bers working on the farm will affect the marginal productivity of additional workers. When all household members have the same skill, larger numbers of house- hold workers would be expected to reduce the marginal product of the young person. If the young people are less skilled and require supervision, however, the availability of supervisors might increase their marginal product. These compli- cations require the addition of the labor force participation of other household members to equation 5. This is clearly an endogenous variable, and the estima- tion procedures below account for this. II. THE DATA AND THE HOUSEHOLD-SPECIFIC OUTPUT PRICE INDEXES The data used for this study are from the Cote d'Ivoire Living Standards Survey (CILSS), for which two years of data (1985 and 1986) are currently available. The survey, conducted annually, covers some 1,600 African house- holds a year in communities selected to be nationally representative. The survey design uses a two-stage sampling procedure, with 100 primary sampling units selected from a list of cities, villages, and rural areas sorted by region and type of location. Within each unit 16 households are interviewed. In half of the units all households are surveyed again the following year, while for the remaining half a new sample of households is drawn. In the following year these new households are reinterviewed, and the existing panel households are discarded. The survey thus provides a rolling panel of 800 households. In practice some of these households will move between the sampling years or may refuse to participate in the second interview, thus reducing the panel size. In a preliminary matching of households for 1985 and 1986, 715 households were reinterviewed, providing information on 1,930 individuals. The CILSS consists of three questionnaires: a household questionnaire, a com- Alessie and others 145 munity questionnaire, and a price questionnaire. The household questionnaire, which forms the main part of the survey, requests detailed information on house- hold income, expenditures, education, health, labor force participation, and other household and individual characteristics. With regard to changes in indi- viduals' work activity, the survey asks for current employment status, kind of employment (self-employed or a wage laborer), occupation and industry, and hours and weeks at work. Agricultural households are asked for details about land holdings, type of agriculture, crops grown, and livestock raised. The survey also distinguishes between production for own consumption and for market, so that the value of home production can be estimated. The vil:lage questionnaire is completed in all rural clusters to collect informa- tion on ccmmunity characteristics common to all households within a cluster. In particular the questionnaire requests information on local economic activities, economic infrastructure, and social services. The price questionnaire, which is completed in all clusters, collects information on a limited number of basic food items and a few nonfood items. A Picture of Work Patterns and Transitions Before considering the labor market activity of the chosen subsample, the general pattern of labor market activity of all panel individuals living in rural areas should be considered. Table 1 shows, for both years and by age group, the proportion of individuals within the panel who reported having worked in the preceding 12 months.1 Roughly 95 percent of individuals between the ages of 24 and 60 report some labor market activity. There is an increase in overall re- ported labor market activity in 1986, which is concentrated in age groups with the lowest percentage of workers, namely individuals under 12 or over 60. A somewhat curious feature is the slight decrease in work activity for the 12-to-18 age group. Table 1. Labor Force Participation of Rural Individuals, by Age Group, 1985 and 1986 Age group Percentage who worked in Number of (in 1985) 1985 1986 observations 7-11 22.5 41.3 417 12-18 71.9 69.8 391 19-23 86.2 90.0 130 24-30 94.2 96.7 154 31-40 96.8 96.7 217 41-50 97.4 96.1 230 51-60 95.1 94.5 182 Over 60 66.5 73.2 209 Total 71.4 76.0 1,930 Source: World Bank data. 1. Many panel individuals reported age changes between interviews of more than one year. In selecting the sample, only those individuals for which the absolute value of the difference in age from one interview to the next is less than four years are retained. Some 200 individuals were discarded due to this selection rule. 146 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 2. Work Status Transitions, by Age Group, 1985 and 1986 Number of people who worked Age group Neither 198S5 1986 Both (in 1985) year only only years Total 7-11 202 43 121 51 417 (48.4) (10.3) (29.0) (12.2) (100.0) 12-18 53 65 57 216 391 (13.6) (16.6) (14.6) (55.2) (100.0) 19-23 9 4 9 108 130 (6.9) (3.1) (6.9) (83.1) (100.0) 24-30 3 2 6 143 154 (1.9) (1.3) (3.9) (92.9) (100.0) 31-40 4 4 3 206 217 (1.8) (1.8) (1.4) (94.9) (100.0) 41-50 3 6 3 218 230 (1.3) (2.6) (1.3) (94.8) (100.0) 51-60 5 5 4 168 182 (2.7) (2.7) (2.2) (92.3) (100.0) Over 60 45 11 25 128 209 (21.5) (5.3) (12.0) (61.2) (100.0) Total 324 140 228 1,238 1,930 (16.8) (7.3) (11.8) (64.1) (100.0) Note: Percentages of age groups are in parentheses. Source: World Bank data. To examine movements of individuals between the states of working and not working, table 2 shows the number and proportion of individuals according to their working status in each year. Between the ages of 19 and 60 the number of individuals leaving or entering the work force is very small. The vast majority of changes in working status are in fact concentrated among individuals aged 18 and under in 1985. The subsample to be used in the empirical work below is comprised of 391 panel individuals aged 12 to 18 in 1985 and living in rural areas. Of this group 122 individuals, or 31.2 percent, report movements be- tween working states across the two years of data, although labor force partici- pation changes by only 2.1 percentage points. The small net change clearly masks large gross flows. Young people's decisions on work activity may be closely related to decisions on schooling. The school system in C6te d'Ivoire is modeled on the French system, inherited under colonial rule. Six years of elementary education lead to the Certificat d'Etudes Primaires (CEPE), which is awarded on the basis of a nationwide examination. The CEPE is a prerequisite for entrance to secondary school, but, since secondary school places are scarce, the score required to gain placement often exceeds that required to obtain the CEPE certificate. Four years of lower secondary education lead to the Brevet d'Etudes du Premier Cycle, which, if successfully completed, allows the student to enter three years of upper secondary education leading to the Baccalaureate. Alternatively, students with a CEPE certificate or who successfully complete some or all of their lower second- ary education can enter various training programs. In the Ivorian education Alessie and others 147 Table 3. Work Status and School Attendance, 1985 and 1986 Percentage attending Percentage working Age school in in Number of (in 1985) 1985 1986 1985 1986 observations 12 68.4 64.5 46.1 48.7 76 13 57.7 53.9 64.1 59.0 78 14 40.0 36.5 73.1 59.6 52 15 29.1 25.5 87.2 83.6 55 16 29.3 21.9 82.9 80.5 41 17 12.8 10.3 84.6 92.5 39 18 12.0 10.0 86.0 88.0 50 Total 40.2 36.3 71.9 69.8 391 Source: World Bank data. system many students at all levels repeat grades, thus the number of years for which students are enrolled will on average exceed the minimum numbers out- lined here. Table 3 shows the percentage of individuals who attend school and/or work at any time in the 12 months before the survey, according to their age in 1985. Work activity reported is concentrated among those individuals aged 15 to 18 in 1985. Overall, school attendance falls steadily between the ages of 12 and 16, with a distinct drop for those 17 or over in 1985. As is to be expected, school attendance falls for all age groups between the two survey years. Clearly, deci- sions are being made regarding not only work activity but also school attendance. The examination of work and school participation rates in cross-section does not shed light on the movement of individuals between the states of working and not working and on the relationship of those movements to schooling. Table 4 shows a cross-tabulation of the number of individuals in the sample according to their work and education status in 1985 and 1986. A total of 231 individuals, nearly 60 percent of the sample, did not attend school in either year. In terms of work status transitions, 57 individuals entered the work force in 1986, having not worked in 1985. Correspondingly, 65 individuals left the work force in 1986. Of those who changed their working status, most (75) did so while Table 4. 'Work and School Status of Rural Youth Ages 12 to 18, 1985 and 1986 Work Did not work Both years 1985 only 1986 only either year Total School Both years 20 47 28 44 139 1985 only 6 0 8 4 18 1986 only 1 2 0 0 3 Did not attend either yea:r 189 16 21 5 231 Total 216 65 57 53 391 Source: World Bank data. 148 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I attending school in both years. Of those who attended school in 1985, only 8 left education to enter the work force in 1986. Clearly, participation in work does not necessarily preclude school attendance. What has not been considered, however, is the intensity of work or schooling undertaken by individuals within the various states of work and education. The survey does contain information on the intensity of work activity, as captured by both the number of weeks worked during the year and the number of hours worked each week. Work intensity is an important issue that should be investi- gated in the future, but here it is just noted that, as expected, the number of hours worked each week by those not attending school is significantly higher than for those who work while attending school. A Household-Specific Measure of Production Price Changes One of the major features of structural adjustment is an attempt to reallocate the mix of labor-and hence output-through changes in relative prices. It is important, therefore, to identify the effect of price changes on an individual's work activity. One difficulty in estimating the response of working behavior to price incentives is the limited amount of variability observed in the two years of price data. The effect of an output price change on work incentives for a particu- lar household, however, will depend on the importance of that particular good in the household's overall output. This means that even if all households face the same price changes for each crop, the changes in incentives are likely to be greatest for those households that specialize in the production of crops whose prices have risen most. A great advantage of the panel is that household-specific indexes of price changes can be calculated based on the first year's production shares, thereby avoiding the possible endogeneity of current period weights. Therefore, this measure of incentive changes may vary considerably among households, at least to the extent that the composition of agricultural production differs across households and changes in prices differ across outputs. An important part of the data work underlying this study is the construction of these composite price indexes. The household survey collects information on the quantities sold and the prices received for 24 crops. For 11 of these crops it was not possible to con- struct crop price indexes, either because the value of the crop in total production was rather small or because insufficient price information was available due to the fact that the price of the crop was generally reported in noncomparable and imprecisely defined units. The 13 crops used to construct the price indexes are cocoa, coffee, oil palm, plantains, fruit trees, cola nut, cotton, peanut, cassava, yam, maize, rice, and vegetables. Average regional prices were calculated for each of five regions: north (north of Kassou Lake), southwest 1 (between Buyo Reservoir and the Guinea and Liberia borders), southwest 2 (central southern area to the southwest of Kassou Lak'e), southeast 1 (north of Abengourou), and southeast 2 (Abidjan hinterland Alessie and others 149 to the south of Abengourou). In calculating average prices for each crop, obser- vations for which the quantities sold were reported in nonmetric units were excluded. Also excluded were some outlier observations, where the selection of outliers was based on a 95 percent range of data. Seasonal variation in prices could not be corrected for because the survey did not ask when during the year crops were harvested and sold. Having calculated average regional prices, composite divisia price indexes for each household were derived using the following formula: (6) P86h = Zw85(f np86i- fnp85i) where wjl4S is the value share of commodity i in total 1985 gross cash crop production for household h, and P86i (P85i) is the average regional price of good i in 1986 (1985). Although the value shares are household specific, the price changes vary only with region. Two separate price indexes are calculated for each household. For the gross production index (PRODIND) the value of production used in household weight- ing is defined as the amount of the harvest sold, plus the value of replacement capital such as seeds, the value of any harvest given away, and the value of home consumption. For the cash crop index (CASHIND) the value of production is the net value of cash crops sold, where cash crops include coffee, cocoa, cotton, and cola nut. It was necessary to construct the CASHIND index because the observed wide dispersion in subsistence crop prices within regions makes the PRODIND index somewhat suspect. This dispersion is due, in part, to the small number of observations for some subsistence crops. Cash crop prices are generally con- trolled and hence more uniform, yielding a more reliable cash crop price index. Some apparent cash crops are excluded from the CASHIND index. For rubber, for example, no household in the sample had positive recorded production. Some households produce fruit products sold for cash, but these are usually recorded in noncomparable units. In calculating both indexes, only the value of primary crop production was included because of insufficient price information on trans- formed production and on animal products. Table 5. Regional Mean Price Change Indexes, 1985-86 Region Southwest Southwest Southeast Southeast Index Total North 1 2 1 2 Gross production index 0.29 -0.17 0.23 0.45 0.76 0.27 (PRODIND) (0.42) (0.27) (0.19) (0.27) (0.39) (0.30) Cash crop production index 0.03 0.01 0.04 0.03 0.03 0.03 (CASHIND) (0.02) (0.01) (0.02) (0.02) (0.03) (0.01) Note: Standard deviations are given in parentheses. Source: Authors' calculations based on World Bank data. 150 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table S reports the regional mean values of the price change indexes. The prices of cash crops clearly do not change by as much as the prices for gross production. The difference between the two indexes for the northern region and those for the other regions is a consequence of the fact that the procurement price for cotton, which is the main crop within the region, was not raised by the government in 1986. The price of cocoa and coffee, which are more important in the other regions, increased by 14 and 5 percent, respectively. These figures are confirmed in Berthelemy and Bourguignon (1989). III. EMPIRICAL RESULTS: INCENTIVES AND AGRICULTURAL WORKING BEHAVIOR Empirical studies of individual participation in education and work are usu- ally restricted to cross-sectional analysis because of the scarcity of panel data in developing countries. This study is therefore in the almost unique position of being able to exploit the repeated observations on each individual, available across pairs of years in the C6te d'Ivoire survey. This will enable the investiga- tion of transitions in work behavior over time as well as the assessment of the usefulness of panel data over cross-section data in such investigations. Since many decisions are based on horizons longer than two years, the models attempt to exploit the richness of the cross-section data to capture long-term influences, leaving the time series aspect to provide information on short-run transitions. Perhaps the most natural statistical model for an analysis of working behavior over time is a discrete transition model, in which current period work status is related directly to a vector of individual characteristics, zi, conditional on last period's work status. If there were only two initial states-described by whether an index S9 equals unity if the individual works, or zero if the individual does not work-then the probability of working in the current period (Si = 1) would be given by (7) Pr[S! = 1] = F(zi O)Pr[Si? = 11 + F(z, -y)Pr[Si = 0] where F(.) describes each discrete conditional probability, ,3 is a vector of param- eters of interest conditional on working in the initial period, and 'y is a vector of parameters of interest conditional on not working in the initial period. If 3 and oy are equal, there would be no need to condition on initial state. In this case, apart from time varying variables, Pr[S? = 1] = Pr[SI = 1], and cross- section data alone could be used to estimate the underlying parameters. If the initial state matters, however, repeated observations over time on each individ- ual are needed to avoid bias in the estimation of ,B and -y. Referring to table 4, the individual's work status in the initial period takes one of three easily recognizable states. First, 281 young people worked in the first period. Second, of those who worked in 1985, 73 individuals also attended school in the first period. Comparing these two groups will enable us to assess whether being involved in some education in the initial period influences the Alessie and others 151 determinants of continuing to work in the next period. Finally, 84 individuals report participation only in education in the first period. Although the purpose of conditioning on past behavior is to identify the probability of working in the second period conditional on the work status of that individual in the recent past (Heckman 1981; Nakamura and Nakamura 1985), the observance of different parameter vectors f and -y is consistent with either state dependence or unobserved heterogeneity of individuals. The experi- ence of working in the initial period, for example, may somehow change peo- ple's constraints or incentives so that they are more likely to work in the next period. Or, different individuals may simply have different probabilities of working due to unobserved characteristics. Because of the short panel available, this analysis cannot distinguish between these two explanations of behavior. The results in this section should be interpreted as coming from a reduced- form transition model, since neither the direct effects of current period schooling on work status nor the effects of current period decisions on work and schooling by other household members are estimated. Because not only the region but also the cluster to which each individual belongs can be identified, the initial empiri- cal specification included cluster-specific dummies. These dummies could be grouped into five broad regional dummies, and indeed further grouping is possi- ble even in a reasonably tightly parameterized setting. Furthermore all models were initially estimated on samples separated by gender and religion, but, in the results reported here, the gender and religion effects are represented by dummies. The model estimated corresponds to equation 5, which contains several vari- ables whose measurement posed a complex and important problem. The house- hold price change indexes are perhaps the best illustration of this, and their definition was described fully in section II. Initial period product shares were used to weight regional price changes so that this index can be treated as exog- enous in the conditional probability model. Given other factors in the model, a strong positive effect of this variable on the probability of work in the second period might be expected if incentives are having a strong effect on work activ- ities. However, given the reduced-form nature of the labor supply equation noted above, the cash crop price variable will also capture the income effects of a change in output prices. As noted in section I, one objective of this study is to assess the relative importance of such effects. To capture a long-run or life-cycle measure of other income and wealth, consumption expenditure is used as an explanatory variable. This is likely to be a current period endogenous variable, so consumption from the initial survey is used. In addition this is restricted to cover food expenditures only, because they not only make up the largest share of a household's budget but also appear to be the best measured consumption item. These expenditures include the value of home production. The income and other household decision variables from the first period will be important in the estimation only so far as they capture longer- run effects not captured by the work status variable of the initial period. 152 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 The influence of local conditions on behavior can be captured through the extensive information on local infrastructure variables and schooling availabil- ity. Other noneconomic influences can be measured through factors such as religion, age structure, and general demographic variables. Two primary mea- sures of work status in the panel survey relate to working behavior in the past week or during the past year. The results were derived using the latter definition alone; these results were not only quantitatively similar to those using work status in the past week, but were also generally more precisely determined. This is encouraging, since it suggests the results are not unduly affected by seasonal work patterns. The results also use a similar measure for the past 12 months for schooling behavior. A binary probit model describing the conditional probability of each individ- ual working in 1986 was estimated by maximum likelihood for each of the three initial states, and the results are presented in table 6. In all the results a reason- ably parsimonious parameterization of each model is specified. Indeed a general Table 6. Work Status in Period 2 Conditional on Work and School Status in Period 1 Status in period 1 Work Work and school School only Constant -2.58 -2.96 1.71 (-2.66) (-1.41) (1.05) CASHIND 15.97 38.18 -14.46 (2.76) (3.33) (-1.49) Age 0.35 0.18 -0.02 (5.94) (1.29) (-0.17) Number of household workers -0.09 -0.17 -0.19 (-1.77) (-2.07) (-2.27) Food expenditures -0.45 -1.03 0.92 (-1.77) (-1.66) (1.40) Muslim 0.43 1.36 (1.13) (2.19) Male -0.53 (-1.16) Regional dummies Southwest 1 -2.00 n.a. -0.13 (-3.38) n.a. (-0.14) Southwest 2 -2.35 n.a. -0.94 (-4.56) n.a. (-1.09) Southeast 1 -1.41 0.04 -0.65 (-2.48) (0.07) (-0.72) Southeast2 -1.26 1.22 -1.88 (-2.11) (2.38) (-2.01) Number of observations 281 73 84 n.a. Not applicable. Note: The dependent variable is binary and equals 1 if the individual worked in period 2; it equals 0 otherwise. The t statistics are in parentheses. Source: Authors' calculations based on World Bank data. Alessie and others 1S3 feature of the results was the dominating importance of the cash crop price effects measured by CASHIND over the general index PRODIND, so only CASHIND results are given. The inferiority of PRODIND as an explanatory variable proba- bly reflects problems in measuring the prices of noncash crops. A full set of results, programs used, and detailed data transformations are available from the authors. Starting with individuals who worked in the initial period, the individual's age and the number of other household workers in the initial period were found to be of some importance. A variable measuring land size was also included in preliminary models, but this was never found to play a role once regional effects were allowed for. Food expenditures in the initial period were added to capture any longer-run income effects not reflected in the discrete state conditioning variable. Its negative coefficient confirms that income affects work status. It was thought that the individual's religion could be important, but its effect was small and insignificant. Location variables could play an important role in shaping both preferences and constraints. The regional dummies are clearly seen to be important, although some further grouping looks possible. Finally, several infra- structure variables and a gender variable turned out to be of little significance for this group. The issue of state dependence can be examined by comparing the behavior of the subset of 73 individuals from this group who had not only worked in the first period but had also been in school. The results for this subgroup are reported in the middle column of table 6. A similar overall pattern is observed, but an even stronger cash crop price effect appears. In this smaller sample both southwest regional dummies were grouped with the base region (north). Finally, the results for the group of 84 individuals who were wholly engaged in education in the first period contrast distinctly with those for the first two groups. Most notably, the cash crop price effect is negative (although insignifi- cant). Those who are not already working are less likely to respond to favorable movements in the prices of cash crops by entering the work force. In fact the probability of working declines with increases in the price index, which indicates that the income effect of price movements dominates for this group. IV. CONCLUSIONS This article has been concerned with modeling the participation in work of young people in rural Cote d'Ivoire. Since many young people combine work and school, the situation is more complex than a simple choice between the two. The econometric analysis has attempted to identify the factors that determine the choice to work. The panel aspect of the data allowed the choice made in one year to depend on the individual's choice in the previous year. The analysis of the decision to participate in the labor force shows that, among other factors, the price of agricultural output generally has a positive incentive effect, but this result depends heavily on the employment and educa- 154 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 tion state in which the individual is placed in the first period. This is of interest not only because it suggests that, for those already working, the substitution effect of the change in wages outweighs the income effect, but also because it shows that the construction of household-specific price indexes can help to explain labor supply behavior with panel data sets of this type. Using food consumption to replace income as a measure of welfare was more disappointing. In contrast to most studies of labor supply, this measure does not appear to exert a strong influence. This could be a result of the inaccurate measurement of food consumption. Although the food consumption data were better than other available consumption measures, brief analysis of the data showed both an unexpectedly large inequality in food consumption per capita and implausibly large year-to-year variation in food consumption per capita for many of the panel households. Further research to find a more accurate measure of consumption is warranted. The results of this study suggest that the CILSS data can be used to analyze labor supply behavior and that their panel aspect is of considerable value. The analysis also suggests several extensions. A preliminary examination of hours of work and their relation to schooling suggests that the modeling and estimation of working hours could be very important in estimating the effect of structural adjustment on overall labor supply. Moreover the analysis presented here and the proposed analysis of working hours should be extended to other age groups, and the decision to pursue education should also be explicitly addressed. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Berthelemy, Jean Claude, and Francois Bourguignon. 1989. "Growth and Crisis in C6te d'Ivoire'" D6partement et Laboratoire d'Economie Th6orique et Appliquee, Paris. Processed. Deaton, Angus, and Dwayne Benjamin. 1987. "Household Surveys and Policy Reform: Cocoa and Coffee in the C6te d'Ivoire." Research Program in Development Studies Discussion Paper 134, Princeton University. Processed. Glewwe, Paul, and Dennis de Tray. 1988. The Poor during Adjustment: A Case Study of C6te d'Ivoire. LSMS Working Paper 47. Washington, D.C.: World Bank. Heckman, James J. 1981. "Heterogeneity and State Dependence." In Sherwin Rosen, ed., Studies in Labor Markets. University of Chicago Press. Nakamura, Alice, and Masao Nakamura. 1985. "Dynamic Models of the Labor Force Behavior of Married Women Which Can Be Estimated Using Limited Amounts of Past Information." Journal of Econometrics 27(3):273-98. Singh, Inderjit, Lyn Squire, and John A. Strauss, eds. 1986. Agricultural Household Models: Extensions, Applications, and Policy. Baltimore, Md.: Johns Hopkins Uni- versity Press. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 155-169 Social Security and Private Transfers in Developing Countries: The Case of Peru Donald Cox and Emmanuel Jimenez Do social security systems "crowd out" private transfers from younger to older genera- tions? This question has generated much theoretical discussion, but little empirical work exists to confirm or refute this crowding-out hypothesis. We investigate the connection between social security and private transfers in Peru, using the Peruvian Living Standards Survey, and find that private transfers from young to old would have been nearly 20 percent higher without social security benefits. This indicates that the Peruvian social security system is less effective at delivering benefits to the elderly than a simple assignment of government expenditures would suggest. Social security's dis- placement of private transfers, while significant, is less than that predicted by models with widespread altruistic transfers. How effective are social security retirement benefits at raising the incomes of older households in developing countries? On the surface it seems that a "pay as you go" social security system, which taxes working households and distributes the proceeds to those who have retired, should benefit older households at the expense of younger ones. But this simple assessment could be inaccurate because of the effect of public transfers on informal private transfers among families. Consider the hypothetical case of an elderly couple supported by their chil- dren. If a social security program is created that taxes the children and channels the funds to the elderly household, the children may reduce their private trans- fers because their social security contributions accomplish what they once did informally. Such a response would weaken the impact of social security on income distribution. Indeed some U.S. researchers, notably Becker (1974) and Donald Cox is on the faculty of the Department of Economics at Boston College and a consultant to the World Bank. Emmanuel Jimenez is with the Population and Human Resources Department of the World Bank. The work on this article was supported in part by RPO 674-49, granted to the Public Economics Division by the World Bank's Research Support Budget. The authors wish to thank the government of Peru and the Bank's Welfare and Human Resources Division for providing access to data and the three anonymous referees and participants at a World Bank seminar for comments on an earlier draft. Jorge Castillo, Kalpana Mehra, and Reza Firuzabadi were invaluable in setting up workable data files. Fiona Mackintosh provided editorial assistance. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 155 156 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Barro (1974), argue that private transfers could render social security programs completely ineffective. Are such considerations important in developing countries that are consider- ing reforms in their social security system? Private transfers are a major compo- nent of household income and expenditure in virtually all of the developing countries surveyed by Rempel and Lobdell (1978) and Cox and Jimenez (1990), and much of the transfer income flows from children to parents. For example, more than a quarter of the private transfers in a sample of Kenyan households were given to parents from their children (Knowles and Anker 1981). Butz and Stan (1982) found significant transfers from young to old in Malaysia, as did Ravallion and Dearden (1988) in Java. In fact some researchers argue that, in developing countries, security for one's old age is the most important motive for having children, so that the prevalence of transfers from young to old should not be surprising (Leibenstein 1975). (For alternative views on this controversial hypothesis, see Vlassoff and Vlassoff 1980 and Nugent 1985.) While the potential linkage between social security and private transfers has obvious significance for policy, very little empirical work has been done on this issue. Part of the problem is the lack of adequate data; data sets containing both private transfer and social security information are scarce. Many countries for which information about private transfers exists have no social security pro- grams or devote only a negligible fraction of their budgets to public transfers of income. Data are available for some industrialized countries, such as the United States, but using these data to make inferences about developing countries would likely be misleading, given fundamental differences in the size of the social security programs and the patterns of intrafamily transfers. In fact, in the United States, private transfers from young to old are negligible (Cox and Raines 1985), which would make it difficult to gauge any connection between social security and private transfers. Furthermore comparisons among countries with data for private transfers are not feasible because of disparities both in survey definitions of private transfers and in sampling techniques. Some surveys, for example, count transfers in kind, whereas others do not, and some surveys focus on low-income subsamples, whereas others use representative cross- sectional samples. A new data set, the Peruvian Living Standards Survey (PLSS), is used here to assess the link between social security and private transfers. Peru is particularly well suited as a case study of the effects of social security on private transfers. The country is an ideal balance between the extremes of very poor nations, like Kenya, which have significant private transfers but little public transfer spend- ing, and the United States, which has a large social security system but negligible private transfers from young to old. In Peru a significant fraction of private transfers flows from young to old, and the social security program has expendi- tures equal to more than 3 percent of the gross domestic product (Suarez- Berenguela 1987). Furthermore the PLSS contains the labor-market and demo- Cox and Jimenez 157 graphic information that must be controlled for in order to isolate social secu- rity's impact on private transfers. The sections that follow review theories about private transfers in more detail and describe the Peruvian social security system. The empirical sections provide an overview of private transfers in Peru and measure the extent to which social security supplants private transfers. I. THEORIES ABOUT PRIVATE TRANSFERS The response of private transfers to social security depends on motives for giving. Economic models generally predict that if motives are purely altruistic and private transfers are widespread, social security will completely crowd out private transfers. If transfers are at least partly motivated by self-interest, com- plete crovvding out may not occur, even if altruistic motives are also present. Modern analysis of the connection between social security and private trans- fers began with the seminal work of Barro (1974), which uses the altruistic framework for private transfers that was advanced by Becker (1974). In this model the utility of the child, Uk, depends on own consumption, Ck, and the utility of the parent, Up. The parent's utility, in turn, depends on own consump- tion, Cp,,so that (1) Uk = Uk[Ck, Up(Cp)3. Both child and parent are subject to budget constraints such that the child's consumption must equal own income, Ik, net of any transfers, T, to the parent. The parent's consumption must equal own income, Ip, plus transfers received from the child. The maximization of equation 1, assuming an interior solution, implies that transfers will be given to achieve the optimal consumption of each family mem- ber. Each member's consumption depends only on aggregate family income, Ik + IP, and not on the distribution of its components. A social security program that forces a transfer from child to parent but leaves aggregate family income unchanged will have no effect on either family member's consumption. The child will reduce private transfers by the exact amount of the forced public transfer to maintain own consumption and that of the parent's at their previous levels. In summary the theory of altruistically motivated transfers predicts that transfers through social security merely supplant (that is, completely crowd out) private ones, leaving individual consumption and well-being unchanged. There are several reasons why complete crowding out may not occur. Al- though it is technically possible for crowding out to be incomplete in a purely altruistic model if there are corner solutions, most models achieve incomplete crowding out because of the introduction of self-interest as a motive for giving. As an illustration, consider a model based on pure self-interest, in which neither parent nor child cares about the other. Suppose it is difficult to use capital 158 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 markets to borrow or save, so that parents lend to their children, who repay when they reach middle age.1 Changes in social security would influence self- interested lending behavior only to the extent that the current income and life- time wealth of generation members are affected. Now suppose social security taxes and benefits increase in a way that leaves the lifetime wealth of the younger generation unchanged but increases the lifetime wealth of the older one. The current income of the young would be depressed, but current desired consump- tion would remain unchanged. If the young are liquidity-constrained, they would borrow more from the older generation, which would imply higher re- payments to the older generation later on. The enhanced lifetime wealth of members of the older generation would reinforce this effect, because it would put them in a better position to lend by lessening any credit constraints they would face. The expansion of social security would thus result in higher future transfers from young to old, contrary to the prediction of the altruistic model. This example is somewhat contrived, as it is designed only to illustrate the possibility that self-interest can generate a different outcome than does altruism. It is also possible to construct examples in which the qualitative predictions for self-interested motives are similar to those of altruistic models. For example, a common theme in the development literature is that private transfers from young to old might be repayments for past assistance or insurance premiums against income shortfalls (Rempel and Lobdell 1978). If an expansion in social security reduces the lifetime wealth of younger generations, such transfers from young to old in the form of loan repayments can decline because current con- sumption and borrowing will have fallen. The exact crowding out of private transfers by public ones, however, which is the primary implication of the altru- ism model, does not generally occur if transfers are motivated by self-interest. An alternative specification contains a mixture of altruism and self-interest. Both Lucas and Stark (1985) and Kotlikoff and Spivak (1981) posit that a combination of self-interest and altruism governs private transfer behavior. Self- interest prompts households to enter an agreement to, say, lend or insure, but altruism creates the mutual trust necessary to circumvent the moral hazard inherent in such agreements. These models are attractive because they explain why households motivated by self-interest are induced to repay loans. This eclectic specification of transfer motives will not result in complete crowding out except in special cases. Even if altruism of the Barro-Becker type is combined with self-interest, only one transfer motive can determine the compar- ative statics results in a given instance, depending on which motive is operative at the margin (Cox 1987). For example, suppose the transfer of a household's 1. There is evidence that private transfers respond to capital market imperfections (Cox 1990). If transfers are used to smooth consumption, transfers and earnings should move inversely over the life cycle. Elsewhere (Cox and Jimenez 1990) it is shown that the U-shaped age pattern for the probability of transfer receipt is almost the mirror image of the age-earnings profile, which supports the idea that private transfers respond to capital market imperfections. Cox andJimenez 159 last bit of income is motivated by self-interest and purchases some service from the recipient. In this instance marginal changes in pretransfer incomes induced by modifications in social security will not produce the Barro-Becker crowding- out results even though the donor has altruistic feelings, because altruistic mo- tives are nlot operative at the margin. The donor is made happier if the recipient receives a windfall income gain but is not willing to make a transfer to raise the recipient's utility. Furthermore the Barro-Becker model employs an important assumption about the bargaining framework between donor and recipient-the donor dominates. Alternative frameworks, such as Nash bargaining, do not result in complete crowding out (McElroy and Horney 1981; Cox 1987; Ko- tlikoff, Razin, and Rosenthal 1990). A final motive for private transfers is one that is often neglected by economists but emphasized by other social scientists: the existence of social norms, such as guilt feelings, pressure family members into helping other members in need (see Becker 1988). In many societies such feelings can extend to include distant relatives or village members. Although precise models about how these norms evolve are not available, they nonetheless are likely to be important in explain- ing why even individuals motivated by self-interest transfer resources to one another. But social norms need not be modeled similarly to pure altruism. For example, social norms can be interpreted so that the perception (by others in the community) of giving, rather than its outcome, determines the donor's utility. In this case complete crowding out need not occur. II. THE SOCIAL SECURITY SYSTEM IN PERU Peru's social security system (the Spanish abbreviation is IPss), which began in the early 1930s as a risk-sharing plan for blue-collar workers, covered about 40 percent of the economically active population by 1985.2 Participation is manda- tory for all employees of public or private firms and cooperatives; self-employed workers have an option to participate in the system. Because of the focus on wages and salaries, coverage is concentrated in the high-income urban and formal sectors of the economy: white-collar workers, blue-collar workers in large firms, and the military. The main sources of funds are payroll contributions by insured persons and their employers. Since 1987 covered workers have been expected to contribute 5 percent of their salaries to IPss, whereas employers contribute 14 percent of wages and salaries paid, plus 1 to 12 percent for work injury compensation, depending on the risk of the occupation (U.S. Social Security Administration 1987). Self-employed workers contribute 15 percent of their monthly income, but the income basis for their contributions is subject to maximum and mini- mum levels. The government contributes to the system primarily as an employer. 2. This section draws heavily on Mesa-Lago's (1985) and Suarez-Berenguela's (1988) description of the IPSS. 160 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Peru's social security system provides pension and health care benefits, with one-half of the employee and employer tax revenues earmarked for pensions and the other half funding medical and maternity expenses. Medical and maternity benefits comprise a little under two-thirds of IPss's total expenditures, with pension funds used to replenish the depleted health budget. In 1983 the system as a whole ran a deficit equal to 6.7 percent of total receipts (International Labour Organization 1987). III. DESCRIPTION OF THE DATA The data set used in this study is the Peruvian Living Standards Survey (PLSS), conducted by the World Bank in conjunction with the Peruvian Instituto Nacio- nal de Estadistica. The PLSS contains socioeconomic information for a sample of 5,109 households surveyed between June 1985 and July 1986. The household is the unit of observation for the analysis below. This limits the analysis to interhousehold transfers, but this might not be a strong limita- tion. Many intrahousehold allocation models, such as Rosenzweig (1986), as- sume that households reallocate resources to maximize a unified welfare func- tion. Also these models predict that households will at least partly offset redistribution effects, as found here in the analysis of interhousehold transfers. Households missing data on private transfers, age, education or gender of household head, parental schooling, illness, household size, or consumption were deleted, as were households with no indicator of urban or rural residence. These deletions reduced the sample size to 4,184. Furthermore, because social security is primarily an urban phenomenon, with urban households accounting for 85 percent of social security coverage, rural households were excluded, reducing the sample to 2,241. Survey respondents were asked the following: "Has any member of your household sent money or goods-regularly or irregularly-to persons who are not members of your household during the past three months?" Respondents were also asked to report the relationship between the recipients and the house- hold head (for example, son, parent) and the value, in intis, of cash and transfers in kind given in the preceding three months. The same questions were asked for interhousehold transfers received. Of the 2,241 households in the sample, 760 (or 33.9 percent) reported giving a private transfer to another household, and 723 (32.3 percent) reported receiv- ing a private transfer from another household. Two hundred eighty-two (12.6 percent) reported both giving and receiving a transfer, while 1,040 (46.4 per- cent) neither gave nor received. Because some households both gave and received transfers, this analysis fo- cuses on net transfers received (transfers received minus transfers given) and net transfers given (transfers given minus transfers received). A household is defined as a net transfer recipient if gross transfers received exceed gross transfers given. A household is a net transfer donor if gross transfers given exceed gross transfers Cox and Jimenez 161 received. The sample contained 574 net transfer recipients (25.6 percent of the sample), 613 net transfer donors (27.4 percent), and 1,054 households for which net transfers were zero (47 percent). Because 282 households both gave and received a transfer, some givers are included in the "net transfer recipients" category, and some recipients are included in the "net transfer donors" category. For simplicity, throughout the rest of the article net transfer recipients will be referred to as "recipients" and net transfer donors as "donors" or "givers. The average of net transfers received for the entire sample is 77.70 intis monthly--4 percent of average total monthly income. To put this figure in perspective, this is roughly two and a half times the average social security pension income. The average net transfer given is 71.05 intis. Curiously, there seems to be no tendency to exaggerate transfers given or to underreport transfers received. If anything, any reporting bias appears to have gone in the other direction. This finding runs counter to the evidence reported in Cox and Raines (1985), in which reported transfers given often exceeded transfers received for a cross-section of the U.S. population. The average net transfer receipt among Peruvian recipients is 303.36 intis-22 percent of the average income of recip- ients before private transfers. The average net transfer given among donors is 259.76 intis-11.8 percent of average donor income. Private transfers are thus nontrivial in both number and magnitude. Survey respondents were asked to report the main sources of transfers re- ceived and the destinations of transfers given. A summary of the sources of transfers received is shown in table 1. Most of the transfers occur between parents and children. After these two categories, the most significant is that of "other relatives, who are the source of 29 percent of the transfers received, although less than a fifth of total intis received. Very few transfers come from grandchildren or spouses, but nonrelatives account for a significant minority of transfers, Because this analysis focuses on interhousehold transfers, interspousal transfers are not considered unless the spouses are living in separate households. The category that is the focus of the empirical work below, transfers from children to parents, is the largest source of transfers, accounting for 32.5 percent of all private transfer income. Table 1. Sources of Private Transfer Income in Peru, 1985-86 Average Percentage of Number of Percentage transfer total intis Source transfers of transfers (intis) received Parents 191 27.6 248.3 26.1 Other relatives 204 29.4 176.5 19.8 Children 182 26.3 324.9 32.5 Grandchildren 9 1.3 101.3 0.5 Spouse 26 3.8 477.3 6.8 Nonrelatives 81 11.7 320.4 14.3 Note: Because households can receive from more than one source, the number of transfers reported here exceeds the number of households receiving net transfers. Source: Calculations from the Peruvian Living Standards Survey, 1985-86. 162 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 IV. WHAT DETERMINES TRANSFER BEHAVIOR? This section investigates empirically the question posed at the outset: Does social security crowd out private transfers to parents from children? There are two issues to explore. First, what is the connection between social security and the occurrence of a transfer? Second, when a transfer occurs, how does social security affect the amount? Probit analysis is used to analyze the first question. The analysis focuses on the sample of 1,121 households whose head is age 45 or over. Indexing house- holds by h and adding a normally distributed stochastic component, the latent variable that determines the transfer decision is expressed as (2) th = ao + alh + bXh Eh, and Th > 0 iff th > 0, Th = 0 otherwise. When the latent variable th crosses the threshold 0, transfers, Th become posi- tive. Otherwise, they are zero. The vector 'h is a set of three variables having to do with pre-private-transfer income: a dummy variable indicating whether the household receives social security income; the amount of social security income; and the amount of non-social-security income, which includes earnings, finan- cial income, rental income, and income from various other sources, such as gambling. The vector Xh contains education and demographic variables that may affect the incidence of transfers, including dummies for whether anyone in the house- hold has been ill during the past four weeks or unemployed during the past 12 months. To measure additional household resources, a dummy indicating whether the household head is a homeowner is included. The vector also con- tains dummies for educational level and a quadratic in age. Previous studies of transfer behavior indicate that the gender of the head of household is an impor- tant determinant of transfer behavior. Evidence from developing countries (for example, Lucas and Stark 1985 for Botswana and Kaufmann and Lindauer 1986 for El Salvador) indicate a positive relationship between transfers and female status, and a similar pattern has been found for the United States (Cox 1987). A dummy variable indicating whether the head of household is female is therefore included. Finally, Xh contains dummy variables for marital status, the number of children under age 30 living outside the household, whether there are no children under 30 living outside the household, and household size. Heckman's (1979) generalized tobit analysis is used to examine the effect of social security on transfer amounts. The estimating equation for transfer amounts is given by: (3) Th = cO + cIh + dXh + E('77 I Th > °) where ih is a random error component whose expectation is conditional on the Cox and Jimenez 163 observation of a positive transfer. The composition of the Ih and Xh vectors here is similar to that used in the probit equation, except that age enters linearly.3 Table 2 contains the probit and generalized tobit estimates for transfers re- ceived. The dummy variable for whether the household receives income from social security is negative, large, and statistically significant at the 0.01 level. At sample means, having income from social security reduces the probability of transfer receipt by 8 percentage points. This is the main finding of the article, and it forms the basis of our assessment of the impact of social security on transfers from young to old. First, however, we discuss the other findings of interest from table 2 as well as the determinants of giving. Neither the level of social security income nor the level of non-social-security income (labeled simply "income" in table 2) is statistically significant in deter- mining transfer receipt, and their coefficients are very small. At sample means, a 1 ,000-inti increase in income reduces the probability of receiving a transfer by only half a percentage point. The most important income measure affecting the receipt of a transfer is the dummy for the receipt of social security income. One possible explanation for the pronounced nonlinear effect of social secu- rity is that the dummy for the receipt of benefits captures the effects of poten- tially weaker family ties among those covered by ipss. This issue was explored further by searching the PLSS data for family-tie indicators to add to the list of regressors already used. Four variables that could potentially serve as additional measures of the strength of family ties were a dummy indicating whether the main respondent was born in a rural area, the number of times the respondent migrated, the number of years the respondent had lived in the area, and occupa- tion. These variables were added to the transfer probit and found not to exert a statistically significant effect on transfer incidence. The rural-birth dummy was positive and on the margin of statistical significance, but including these vari- ables had a negligible effect on the coefficient of the social security dummy. The rmost plausible explanation for the pronounced discontinuity in social security's effect stems from the inclusion of health benefits in social security payments. Because people collecting social security benefits also enjoy better health coverage and because private transfers respond to health risks, a sharply discontinuous effect on the incidence of private transfers is not all that surprising. The pirobability of receiving a transfer rises with age up to age 81 and then declines. At the sample mean (age 57), being one year older raises the proba- bility of transfer receipt by 2.4 percentage points. The probit results indicate that transfers are targeted toward homeowners, the unemployed, and house- 3. The probit equation used to generate the inverse Mills ratio in the generalized tobit contains the same vector of explanatory variables as does the probit in equation 2, plus additional terms. Income, age, marital status, female status, illness, and unemployment are entered interactively. The extra variables are used to identify the generalized tobit. Results from the expanded probit equation are available on request. 164 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 2. Probit and Generalized Tobit Estimates of Transfers Received by Parents from Children in Peru, 1985-86 Probit Asymptotic Generalized to bit Variable Coefficient t-value Coefficient t-value Income variables Income -0.290 x 10-4 -1.14 0.029 1.16 Social security recipient -0.565 -3.26 129.695 0.65 Social security income 0.148 x 10-3 0.70 -0.295 -1.29 Education dummy variables- Initial -3.377 -0.13 n.a. n.a. Primary 0.271 1.60 15.850 0.11 Secondary 0.259 1.32 443.218 2.53 Technical -0.172 -0.43 780.340 1.58 Postsecondary -0.131 -0.30 36.904 0.08 University 0.058 0.22 427.192 1.60 Household cbaracteristics Age 0.204 3.46 6.605 0.78 Age squared -0.001 -2.72 n.a. n.a. Unmarried 0.057 0.29 -195.191 -1.05 Female 0.454 3.49 355.273 2.46 II] 0.091 0.86 -251.910 -2.33 Unemployed 0.393 3.50 -35.258 -0.30 Homeowner 0.273 2.34 -68.908 -0.58 Number of youngb -0.097 -0.51 -95.663 -0.50 Young in schoolc 0.146 0.76 100.045 0.54 No young -0.272 -1.86 149.073 0.96 Household size -0.008 -0.42 45.155 2.15 Constant -9.029 -4.83 -436.435 -0.49 Inverse Mill's ratio n.a. n.a. 22.545 0.10 Recipients 175 175 Observations 1,121 175 R-squared n.a. 0.22 Dependent variable mean 0.16 330.29 Note: In the probit analysis, the dependent variable is a transfer receipt (transfer receipt = 1 if the transfer is received from a child, 0 if otherwise). In the generalized tobit analysis, the dependent variable is the net transfer amount received. The sample consists of households whose heads are age 45 or over. n.a. Not applicable. a. The reference category is no education. b. Number of children under age 30 living outside the household. c. Number of children who are in school living outside the household. Source: Calculations from the Peruvian Living Standards Survey, 1985-86. holds headed by females. Education is not significantly related to transfer receipt. For the PLSS, households were asked for information on children under the age of 30. Having no children under 30 reduces the probability of transfer receipt, but the other child-related coefficients are not significantly different from zero. More than half of the transfer recipients have no young children outside the household, so they must be receiving transfers from children over 30. Cox and Jimenez 165 The generalized tobit estimates are also given in table 2. Transfer amounts are higher for those with secondary, technical, and university education (the refer- ence category is "no education") and lower for those who are ill. Amounts are also higher for female-headed households and larger households. None of the income measures in the generalized tobit is statistically significant. Estimated equations for transfers given by younger to older generations are presented in table 3. The sample is restricted to the 1,875 households having at least one living parent residing outside the household. The probability of giving a transfer rises with earnings, but at a diminishing rate. At the sample mean a 1,000-inti increase in earnings raises the probability of giving a transfer by 2.2 percentage points. The probability of giving also increases with other income. Education has a negligible effect on the probability of giving. The probability of giving a transfer rises with age up to age 38, then falls. Female-headed house- holds and the unemployed are less likely to give, but the ill are more likely to do so. The education levels of parents and of the spouse's parents are included as indicators of the pretransfer resources of members of the older generation. These indicators show a mixed pattern for transfers given. For example, the probability of giving is inversely related to the mother's education but positively related to the spouse's mother's education. Generalized tobit estimation yields few significant variables influencing trans- fer amounts given. Transfers given increase with donor earnings, but, as in the probit equation for giving, education has a negligible effect on donations. The unmarried give more, and amounts given increase with the in-laws' education. V. THE EFFECTS OF SOCIAL SECURITY ON TRANSFERS FROM YOUNG TO OLD Using the estimates in tables 2 and 3, one can determine how private transfers would change if social security pension benefits were taken away. Such a deter- minatior[ can be made by comparing predicted transfers with and without social security. Predicted transfers are the product of the predicted transfer probability and the predicted amount. Because probit analysis has been used, the predicted transfer probability is given by the normal cumulative distribution function evaluated at dO + alh + LXA, where ao, a, and b are estimated coefficients and 'h and Xh are means of the vectors of explanatory variables. Predicted transfer amounts are the products of the estimated coefficient vector and the vector of explanatory variables evaluated at sample means: (4) Predicted transfers = F(ao + di, + bX,) x (co + cIh + dXb) where F is the normal cumulative distribution function. The sample-selection term and its estimated coefficient are included in predicted values for transfer amounts. (For a discussion of prediction in tobit models, see Maddalla 1984, chap. 6.) Predicted transfers without social security are given by equation 4 with social-security-related elements of the vector Ib set to zero. For transfers received 166 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Table 3. Probit and Generalized Tobit Estimates of Transfers Given by Children to Parents in Peru, 1985-86 Probit Asymptotic Generalized tobit Variable Coefficient t-value Coefficient t-value Income variables Earnings 0.128 x 10-3 3.38 0.022 5.56 Earnings squared -0.677 x 10-8 -2.33 n.a. n.a. Other income 0.381 x 10-3 1.99 -0.052 -1.06 Education dummies- Initial 0.820 1.45 -66.177 -0.51 Primary -0.116 0.44 -24.783 -0.36 Secondary 0.275 1.01 -31.490 -0.45 Technical 0.338 1.12 -43.154 -0.57 Postsecondary 0.296 0.94 -48.061 -0.63 University 0.289 0.99 -18.952 -0.26 Household characteristics Age 0.203 2.39 0.202 0.21 Age squared -0.004 -2.16 n.a. n.a. Age cubed 0.234 x 10-4 1.77 n.a. n.a. Unmarried 0.242 1.16 219.491 4.78 Female -0.367 -2.57 38.067 1.07 Ill 0.179 2.41 -21.979 -1.31 Unemployed -0.136 -1.28 21.888 0.87 Homeowner -(.105 -1.36 15.444 0.95 Household size -0.048 -2.52 2.026 0.47 Number of living parentsb 0.092 -2.22 -2.703 -0.30 Education of parents Father 0.016 1.21 -3.984 -1.39 Mother -0.023 -1.47 2.344 0.69 Spouse's father -0.022 -1.77 6.164 2.22 Spouse's mother 0.029 1.99 5.735 1.79 Constant -4.361 -3.48 157.314 1.38 Inverse Mill's ratio n.a. n.a. -90.513 -2.00 Donors 305 305 Observations 1,875 305 R-squared n.a. 0.30 Dependent variable mean 0.16 106.51 Note: In the probit analysis, the dependent variable is a transfer given (transfer given = 1 if the transfer is given to a parent, 0 if otherwise). In the generalized tobit analysis, the dependent variable is the net transfer amount given. The sample consists of households whose heads or spouses have at least one parent living. n.a. Not applicable. a. The reference category is no education. b. Includes head's and spouse's parents. Source: Calculations from the Peruvian Living Standards Survey, 1985-86. by old from young, the actual mean is 51.56 intis, the predicted mean is 52.13 intis, and the predicted mean without social security income is 59.20 intis. Removing social security causes a 7.07-inti increase in predicted private trans- fers received by older generations from younger ones. Thus, without social security, transfers per household from young to old in the sample would rise Cox and Jimenez 167 from 51.56 intis to 58.63 intis. The boost in transfers amounts to 12.1 percent (7.07 / 58.63) of pre-social-security transfers. Because donors cannot be matched with recipients in the PLSS, the 12.1 per- cent figure does not take into account how the removal of social security would affect donor behavior. Taking away social security increases the disposable earn- ings of potential donors, and, as indicated in table 3, giving is positively related to earnings. If the social security tax is fully shifted to workers, they would pay both the employee and employer portions of the assessment, or a tax rate of 19 percent. Removing the tax would raise disposable earnings by 23 percent (0.19/(1-.0.19). The estimates in table 3 imply that the increase in disposable earnings from removing the tax would generate a predicted increase in private giving of 2.92 intis. Adding together the increase in transfers that would be received with the increase in transfers that would be given as a result of the tax- reduction effect yields an increase in total transfers of 9.99 intis. Without social security, then, total transfers per household would be 61.55 intis. The estimated reduction in transfers due to social security would be 16.2 percent (9.99/61.55). Another way to express this result is that without social security, private transfers from young to old would have been 19.4 percent larger (9.99/51.56). These simulation results indicate that social security has a significant impact on private transfers, but its effect is considerably less than the Barro-Becker prediction of complete crowding out. The estimates in tables 2 and 3 can be interpreted in the light of alternative transfer motives. Some findings in table 2 are inconsistent with the strict Barro-Becker altruism motive for private trans- fers. The altruism model predicts a large reduction in private transfers, with increases in the pretransfer income of recipients. Instead, two of the income measures in the generalized tobit equation for transfer receipts are positively related to transfer amounts, although the coefficient on the amount of social security income is negative and large, which is consistent with the altruism hypothesis (although none of the coefficients is precisely estimated). One in- triguing explanation is that donors respond more strongly to changes in the recipent's income if the changes are beyond the recipient's control. Presumably recipient:s have more control over their earnings than they have over public transfer receipts. Although the probit analysis yields a negative coefficient on the receipt of social security income, which suggests altruism, probit results are a less discrimi- nating test of transfer motives than are tobit results. The reason is that a nega- tive coefficient on social security in the probit equation is consistent with exchange-motivated transfers as well, because a parent who expects to receive social security payments has less incentive to enter into exchange relationships with children, such as intergenerational lending or insurance. The same argu- ment applies to the positive coefficient on unemployment status in the probit for transfer receipt. The positive relation between transfers and unemployment may be due to either altruistic or exchange motives. 168 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 One problem with making inferences about transfer motives from the equa- tions that estimate transfer amounts, however, is the lack of information about potential donor income. For example, a possible reason why the coefficient on recipient income is positive is that the characteristics of potential donors are omitted from the equation. Positive correlation between recipient and donor incomes imparts an upward bias to the income coefficients in table 2. Although far from conclusive, some of the evidence is consistent with inter- family exchange. For example, females are much more likely to receive transfers in Peru, which is consistent with evidence from other countries (Lucas and Stark 1985 for Botswana; Kaufmann and Lindauer 1986 for El Salvador; Cox 1987 for the United States). Part of the reason for the gender difference in transfers received may be attributable to the possibility that women are more involved in the interfamily exchange of services in kind. "Altruistic" explanations, such as women being compensated for deficiencies in earning potential, are less convinc- ing, because the estimates control for income. Part of the gender difference could also be caused by gender differences in life expectancy, resulting in a targeting of private transfers to widows. VI. CONCLUSION Like many other countries, Peru established a social security system in part to ensure the well-being of the elderly. But who really benefits from the system? Assignment of public benefits does not occur in a vacuum-the private sector responds by withdrawing some of its support. The extent of the decline in private transfers is an empirical question addressed in this article. Peruvian households whose heads receive social security pension income are less likely to receive private transfers from their children. Without social security, the amount of private transfers from young to old would have been almost 20 percent larger. Thus, an evaluation of policy effects should take private behavior into account. REFERENCES Barro, Robert J. 1974. "Are Government Bonds Net Wealth?" Journal of Political Econ- omy 82 (6, November/December): 1095-17. Becker, Gary S. 1974. "A Theory of Social Interactions." Journal of Political Economy 82 (6, November/December): 1063-93. .1988. "Family Economics and Macro Behavior." American Economic Review 78 (1, March): 1-13. Butz, William P., and Peter J. E. Stan. 1982. "Interhousehold Transfers and Household Structure in Malaysia.' Population and Development Review 8 (Supplement): 92-115. Cox, Donald. 1987. "Motives for Private Income Transfers?' Journal of Political Econ- omy 9S (3, June): 508-46. . 1990. "Intergenerational Transfers and Liquidity Constraints." Quarterly Jour- nal of Economics 105 (1, February): 187-217. Cox and Jimenez 169 Cox, Donald, and Emmanuel Jimenez. 1990. "Social Objectives through Private Trans- fers: A Review." The World Bank Research Observer 5 (2): 205-18. Cox, Donald, and Fredric Raines. 1985. "Interfamily Transfers and Income Redistribu- tion'" In Martin David and Timothy Smeeding, eds., Horizontal Equity, Uncertainty, and Measures of Well-Being. University of Chicago Press. Heckman, James J. 1979. "Selection Bias as Specification Error." Econometrica 47 (Jan- uary): 153-61. International Labour Organization. 1987. The Cost of Social Security, 1981-83. Geneva. Kaufmann, Daniel, and David L. Lindauer. 1986. "A Model of Income Transfers for the Urban Poor." Journal of Development Economics 22 (2, July/August): 337-50. Knowles, James C., and Richard Anker. 1981. "An Analysis of Income Transfers in a Developing Country." Journal of Development Economics 8 (2, April): 205- 26. Kotlikoff, Laurence J., Assaf Razin, and Robert W. Rosenthal. 1990. "A Strategic Altru- ism Model in Which Ricardian Equivalence Does Not Hold." Economic Journal 100 (403, December): 1261-68. Kotlikoff, Laurence J., and Avia Spivak. 1981. "The Family as an Incomplete Annuities Market.' Journal of Political Economy 89 (2, April): 372-91. Leibenstein, Harvey. 1975. "The Economic Theory of Fertility Decline?' Quarterly Jour- nal of Economics 89 (1, February): 1-31. Lucas, Robert E. B., and Oded Stark. 1985. "Motivations to Remit: Evidence from Botswana.' Journal of Political Economy 93 (5, October): 901-18. Maddalla, G. S. 1984. Limited-Dependent and Qualitative Variable in Econometrics. Cambridge University Press. McElroy, Marjorie B., and Mary Jean Horney. 1981. "Nash-bargained Household Gen- eralization of the Theory of Demand." International Economic Review 22 (2, June): 333-49. Mesa-Lago, Carmelo. 1985. El Desarrollo de la Seguridad Social en America Latina. Santiago: Comision Economica para America Latina y el Caribe. Nugent, Jeffrey B. 1985. "The Old-Age Motive for Fertility." Population and Develop- ment Review 11 (1, March): 75-97. Ravallion, Martin, and Lorraine Dearden. 1988. "Social Security in a 'Moral Economy': An Empirical Analysis for Java.' Review of Economics and Statistics 70 (1, February): 36-44. Rempel, Henry, and Richard A. Lobdell. 1978. "The Role of Urban-to-Rural Remit- tances in Rural Development." The Journal of Development Studies 14 (3, April): 324-41. Rosenweig, Mark R. 1986. "Program Intervention, Intrahousehold Distribution, and the Welfare of Individuals: Modeling Household Behavior." World Development 14 (2, March): 233-43. Suarez-Berenguela, Ruben M. 1987. Financing the Health Care Sector in Peru. LSMS Working Paper 31. Washington, D.C.: World Bank. U.S. Social Security Administration. 1987. Social Security Programs throughout the World. Washington, D.C.: Government Printing Office. Vlassoff, Michael, and Carol Vlassoff. 1980. "Old Age Security and the Utility of Children in Rural India." Population Studies 34 (3, November): 487-99. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 171-188 The Willingness to Pay for Education for Daughters in Contrast to Sons: Evidence from Rural Peru Paul Gertler and Paul Glewwe In most of the developing world the education of women lags behind that of men. This could come about from a lack of parental desire for educated daughters or from a perception by the parents that there is a lower net return to education for girls. The relation between gender and education in rural Peru is explored using data from the 198S-86 Peru Living Standards Survey. A model of educational choice is developed. The estimated demand functions are used to assess the impact of user fees on demand and revenues. The empirical evidence indicates that parents are more willing to pay for reduced travel time to secondary school for boys than for girls. However, parents are willing to pay increasedfeesfor girls'schooling sufficient to generate teachers' salaries. Education is almost universally recognized as an avenue for raising living stan- dards. Thus the factors that influence educational decisions have an important impact on individual and social welfare. In developing countries one factor that plays a consistent role in educational decisions is gender: females usually attain lower levels of education and have lower school enrollment rates than do males. In almost all developing countries the government is the main provider of educa- tion, but in recent years funds for education have been constrained as many developing countries have experienced economic stagnation. In times of tighter educational opportunities the educational position of women, relative to men, may suffer. In many developing countries the government provides educational oppor- tunities at no direct cost to households, or at very little cost relative to the expenses it incurs. Present financial burdens on governments have led some to suggest that households should pay substantial fees to provide funds for the government to expand both the quantity and the quality of schooling (Birdsall 1983; Thobani 1983; World Bank 1986; Jimenez 1987). Whether such a policy would discriminate against poorer households has been investigated in Gertler and Glewwe (1989), who find that such households are willing to pay sizable Paul Gertler is at the RAND Corporation, Santa Monica, California, and Paul Glewwe is in the Population and Human Resources Department at the World Bank. ) 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 171 172 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Table 1. Primary and Secondary School Enrollment Rates for Boys and Girls in Various Income and Country Groups and in Peru, 1986 (percentage of age group) Primary Secondary Group Boys Girls Boys Girls Income group Low 113 92 42 27 Lower-middle 108 100 57 50 Upper-middle 107 101 71 67 High 103 102 91 93 Country group Sub-Saharan Africa 73 58 20 12 East Asia 131 117 50 39 South Asia 98 69 41 22 Latin America and the Caribbean 110 108 54 56 Developed economies 103 102 92 94 Peru 125 120 68 61 Note: All rates are gross rates and thus may be greater than 100 because some of the children enrolled may be older than the age group normally associated with a given level of schooling. Source: World Bank (1989). fees to send their children to school. Yet there is little rigorous investigation of the effect of school fees on the educational attainment of girls relative to that of boys. This question is examined here using data from rural Peru. We find that although parents are less willing to pay for the schooling of girls relative to that of boys, they are still willing to pay a fee that would be adequate to finance teachers' salaries. 1. Low LEVELS OF EDUCATION AMONG WOMEN-THEORY AND EVIDENCE Primary and secondary school enrollment rates for girls and boys in various income and country groups and in Peru are given in table 1. The gap between girls' and boys' enrollment rates is largest for the poorest countries; it effectively disappears for high-income countries, at least at the primary and secondary levels. The gap is most pronounced in Sub-Saharan Africa and South Asia, less pronounced in East Asia, and almost nonexistent in Latin America and the Caribbean and in developed economies. However, even in Latin America and the Caribbean, progress in school enrollment for girls is uneven across countries. There are two main aspects of a decision by parents to send their children to school. The first is the consumption aspect, which is that parents may prefer educated children regardless of the financial benefits of education. The second is the investment aspect, which proposes that education may be valued because it brings financial (or other) returns to parents. Thus differences in school enroll- ment for girls and boys must occur because of differences in the direct value parents place on educated children or because the economy and society may be Gertler and Glewwe 173 such that, from the parents' perspective, the net return to educating boys is higher than that to educating girls. Of course, both factors may be operating. Differences in the net returns to education can be divided into the cost of, and the gross returns to, education and into the returns to parents and the returns to children. The net return to education is the gross return minus the cost (both in terms of discounted value). If the net gains are lower for girls, either their costs are higher or their gross financial gains are lower relative to boys, or both. Furthermore parents may not benefit equally from financially successful sons and daughters. If social customs place more responsibility on sons than on daughters to support their parents, the parents may have a stronger incentive to educate their sons, even if the net returns to education are the same for boys and girls. II. GENDER AND EDUCATION IN RURAL PERU The relation between gender and education in rural Peru is explored using the Peru Living Standards Survey. This survey was carried out jointly by Peru's Instituto Nacional de Estadistica and the World Bank (see Grootaert and Arriagada 1986). The Extent of the Gender Gap Tables 2 and 3 present data on the differences in education by gender in urban and rural areas in Peru. The data on the adult population in table 2 indicate that in recent years the gap between women and men in the level of education Table 2. Mean Years of Schooling Attained and the Gender Gap in Education in Peru, 1985-86 Mean years of education Gendergap Age All Male Female Yearsa Percentageb Adult population 20-29 - 8.6 7.4 1.2 16.2 30-39 - 7.7 5.8 1.9 32.8 40-49 - 5.6 3.7 1.9 51.4 50-59 - 4.9 2.9 2.0 69.0 60+ - 4.0 2.4 1.6 66.7 School-age population in urban areas 6-10 1.3 1.4 1.3 0.1 7.7 11-15 5.2 5.2 5.0 0.2 4.0 16-20 8.5 8.6 8.5 0.1 1.2 21-25 9.9 10.1 9.6 0.5 5.2 School-age population in rural areas 6-10 0.8 0.8 0.8 0.0 0.0 11-15 3.5 3.6 3.3 0.3 9.1 16-20 5.5 5.7 5.1 0.6 11.8 21-25 5.8 6.1 5.5 0.6 10.9 - Not available. a. Difference between male and female mean years of schooling. b. The difference as a percentage of female mean years of schooling. Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. 174 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Table 3. Boys and Girls in the Primary and Secondary Grades in Urban and Rural Peru, 1985-86 (percentage of students) Urban Rural Grade Boys Girls Boys Girls 1 55.8 44.2 57.1 42.9 2 60.4 39.6 59.7 40.3 3 63.4 36.6 57.7 42.3 4 68.4 31.6 62.9 37.1 5 58.4 41.6 66.7 33.3 6 60.7 39.3 67.0 33.0 7 61.6 38.4 72.7 27.3 8 64.0 36.0 68.2 31.8 9 60.2 39.8 69.1 30.9 10 68.0 32.0 60.3 39.7 Primary (1-5) 60.6 39.4 59.9 40.1 Secondary (6-10) 62.5 37.5 67.7 32.3 Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. attained has narrowed. For the population age 40 and above, the mean years of schooling of males is more than 50 percent higher than that of females, but it is only 33 and 16 percent higher for males ages 30-39 and 20-29, respectively. For the school-age population in urban and rural areas, the data indicate that males and females have attained lower average years of schooling in rural areas. The gender gap is most obvious at the secondary level in rural areas. Data on the percentage of boys and girls in grades 1 to 10 in urban and rural areas are presented in table 3. In both urban and rural areas in all grades more than half the students are boys. Girls make up the smallest percentage of stu- dents in rural areas in the secondary grades. Cultural Attitudes and the Cost of Educating Girls Do parents in rural areas of Peru think that education is less important for girls than for boys? Some observers have found evidence to support this hypoth- esis. According to Fernandez (1986, p. 3), in rural areas of Peru "families maintain preferential attitudes toward boys and discriminatory ones toward girls:' More than 90 percent of rural women in Peru do attend school at some point in their lives but, at the secondary level, boys appear to go farther (table 2). This is consistent with the notion that girls need only a basic education to perform their social roles, whereas boys require higher levels of education. According to Vargas (1987, p. 3), "Many rural and urban families of the lower and middle classes prefer to send their sons to school and give their daughters a bare minimum of education and then put them in charge of domestic tasks." One possible reason that boys go farther in school than girls is that the cost of sending girls to school may be higher. The direct cost of school fees is likely to be the same, but a more important cost may be hours of work lost to the household due to a child's school attendance. Fernandez (1986, p. 5) finds that "in a study Gertler and Glewwe 175 of peasants and seamstresses, 42 percent of 684 peasant women interviewed declared that they had left school because of their responsibilities on the farm .and around the house." Data from the 1985-86 Peru Living Standards Survey for hours spent on schoolwork and on other work for children ages 10-14 and 15-18 are given in table 4. The data are for two areas of rural Peru, for students and nonstudents, and for girls and boys. Girls ages 15-18 in rural Costa and ages 10-18 in rural Sierra and Selva spend substantially more hours on work that is not related to school than do boys. More important, the loss in hours worked for a girl who attends school is higher than the loss for a boy who attends school. Differences in the Benefits of Education for Males and Females Given this cost disparity by gender, could there also be a benefit disparity? Stein (1972) presents anecdotal evidence that Peruvian households consider males to be "worth more" than females. The issue of whether the economic environment makes girls "less valuable" than boys is explored in this section by using studies based on data from the 1985-86 Peru Living Standards Survey. Wages and education. One way in which educating boys could be "worth more" than educating girls is that the wages of men may be higher than those of women. More specifically, the increase in wages brought about by increases in schooling may be higher for boys than for girls. Stelcner, Arriagada, and Moock (1987) and King (1989) have examined the returns to schooling for men and women, respectively, in both urban and rural Peru. They estimate the percentage Table 4. Time Spent on Schoolwork and Other Work by Boys and Girls in Rural Peru, 1985-86 (hours per week) Other work Schoolwork Outside the house Housework Total Area Boys Girls Boys Girls Boys Girls Boys Girls Rural Costa Students Age 10-14 23.6 22.8 7.7 4.7 9.3 11.8 17.0 16.5 Age 15-18 20.7 17.8 12.7 12.0 7.8 10.2 20.5 22.2 Nonstudents Age 10-14 0.0 0.0 21.3 11.8 6.8 15.8 28.1 27.6 Age 15-18 0.0 0.0 26.8 12.3 5.0 26.5 31.8 38.8 Rural Sierra and Selva Students Age 10-14 24.4 25.0 8.7 8.8 9.7 12.8 18.4 21.6 Age 15-18 24.6 24.6 13.1 9.3 8.1 15.7 21.2 25.0 Nonstudents Age 10-14 0.0 0.0 25.1 25.7 8.7 17.9 33.8 43.6 Age 15-18 0.0 0.0 36.3 27.7 6.9 22.9 43.2 50.6 Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. 176 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. t increase in wages due to an additional year of schooling. To the extent that women's wages are lower than those of men for reasons other than schooling, similar percentage increases for both sexes amount to a lower increase in wages to women, relative to men, for an additional year of schooling. Two results stand out from the research. First, at the primary level the returns to education to men and women are quite similar in both urban and rural areas, and all estimates are statistically different from zero. Second, although the re- turns to secondary education are always statistically different from zero for men, this is not the case for women. More strikingly, the point estimate of the return to secondary education for women in rural areas is very low (and insignificant) compared with that for men. These results suggest three things. First, there is a similar financial incentive to send both girls and boys to primary school in both urban and rural Peru, so one would not expect discrimination against girls at the primary level in either area due to different (gross) returns to education. Second, although the point esti- mates are not precise for females, the rates of return to secondary school in urban Peru do not appear to vary by gender, so there is little incentive to discriminate against girls in urban areas. Third, there is an incentive to discrimi- nate against girls and in favor of boys at the secondary level in rural areas, since rates of return to girls are not significantly different from zero while they are so for boys. Thus, even if rural parents had no previous bias against educating girls, the increase in wages for males and the lack of increase in wages for females as a result of increased schooling may persuade them to send boys, but not girls, to secondary school. Returns to education in nonagricultural self-employment. Many men and women are not wage earners, but are self-employed, either in agriculture or nonagricultural enterprises. Moock, Musgrove, and Stelcner (1989) estimated the rates of return to schooling on the earnings of self-employed individuals working in nonagricultural family enterprises. Separate estimates were made for all enterprises, those run only by females, and those run at least in part by males, for Lima, other urban areas, and rural areas. In urban areas estimated rates of return to education are significantly different from zero for most levels of education. However, the returns to secondary education in female-only firms in urban areas are not significantly different from zero. In rural areas there are no significant returns to anyone at any level of education. Thus there does not appear to be any financial gain to secondary education for self-employed women in either urban or rural areas. Returns to education in agriculture. Another possibility is that education may be a profitable investment if it increases agricultural productivity. Jacoby (1989) takes up this issue in a study that estimates the returns to education on gross farm revenues in rural Sierra. Jacoby's results do not distinguish between males and females, but in general indicate that education has a substantial effect on Gertler and Glewwe 177 agricultural productivity only at the postsecondary level. Thus it is not surpris- ing that rural women are less likely to have attended secondary school than rural men, who at least have significant returns to education for wage work. III. THE VALUE OF SECONDARY EDUCATION FOR MALES AND FEMALES IN RURAL PERU From the data in section II, two points became clear about differences in education for boys and girls in rural Peru. The first is that girls appear less likely to attend secondary school than do boys. The second is that the returns to secondary schooling are almost nil for women, while for men secondary school- ing has significant returns in the wage labor market. Thus secondary education may not be a rational investment for girls, but it does seem to be a rational investment for boys. Consequently, girls may be less likely to be enrolled than boys. However, education also has a consumption aspect, and the fact that many girls are enrolled in secondary school indicates that, for many families, secondary schooling for girls is valued. (It is also possible that it makes them more productive in work at home.) A model of the demand for schooling is used to calculate the willingness of parents to pay for improvements in their children's schooling. The model is estimated to get a general picture of the differential value placed on boys' and girls' schooling and to examine what will happen if user fees are raised in rural Peru. The Model Because education is both a consumption good and an investment good, it is valued for its own sake and because it provides financial returns. The fact that some children are not attending school indicates that, in their parents' eyes, the advantages of sending them to school are outweighed by the disadvantages. The advantages are the benefits that parents receive from well-educated children, both in direct material returns (for example, support in old age) and in the satisfaction of having educated children. The main disadvantages are the associ- ated costs, which may be a heavier burden on poorer families. Parents plan for the total investment in a child's education based on assump- tions about future costs and benefits. Because schooling occurs in annual incre- ments, parents must decide each year whether to enroll their child in another year of school. As time passes and more information about expected costs and benefits is revealed, parents update their expectations and revise their plans. Although it might be optimal to model the complete dynamic process, the limitations of a single cross-section of data require a focus on the decision to enroll for another year of schooling. The contemporaneous enrollment decision is chosen over the educational achievement outcome because we are interested in price effects and because the latter requires data on all past prices and incomes 178 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I when choices about education were made. However, the enrollment decision model is consistent with a fully dynamic process. Ultimately, a family enrolls a child for another year of school if the expected future benefits outweigh the associated costs. A model of the demand for schooling is formalized by specifying the utility obtained from each schooling option (Gertler and Glewwe 1989). Every house- hold is assumed to have a utility function that depends on the human capital of its children and the consumption of goods and services. Investment in another year of schooling raises a child's human capital at the cost of reduced consump- tion of other goods and services. In any year the choice is discrete; the options are to not send the child to school, to send the child to the nearest school, or to send the child away from home to a distant school. The expected utility conditional on sending a child to school is given by (1) Ui = U[(Ho + Si), Ci] + i, i = 1,f- where Ho is the current stock of the child's human capital, Si is the increment to that human capital from another year of education at school i, Ci is the house- hold's consumption of goods and services after incurring the direct and indirect costs of sending the child to school i, and ei is a random taste shifter. Subscripts I and f refer to local and faraway (that is, the child lives away from home) schools, respectively. If the parents decide not to send the child to school, the utility is: (2) U0 = U(HO, C0) + Eo where C0 is the level of household consumption possible without sending the child to school, and Eo is a random taste shifter. The decision to send a child to school depends on the quality of education received (Behrman and Birdsall 1983, 1985), which is determined by school and teacher characteristics, such as the availability of textbooks and levels of teacher training. The quality of education can also include household characteristics such as parents' education. It is assumed that the higher the quality of education, the greater the increase in human capital from another year of schooling and thus the higher the utility from the schooling option, other things being equal. One would expect that the only reason for a child to go to a faraway school would be to obtain a higher quality of education than is available at the local school. The local and faraway school options have budget constraints of the form (3) C, + P,*= Co = Y, i = lf where P7 and P7 are the prices of sending a child to local and faraway schools for one year, respectively, and Y is the family's yearly disposable income. The price of schooling involves both direct and indirect costs. The former include expenses such as school fees and outlays for textbooks. The crucial indirect cost is the opportunity cost of a child's time. In developing countries children often con- tribute substantially to family income by working, including housework. Table 4 Gertler and Glewwe 179 demonstrates that there are large differences in hours worked between students and nonstudents in rural Peru. Direct and indirect costs are not necessarily the same for all households. Even if fees and textbook costs are the same for all households, direct transportation costs may vary. If the school is far away, outlays on transportation may be required, and, if the school attended is so far that the child must live away from home, lodging costs are also incurred. Indirect costs may also vary because of differences in time lost from work due to school attendance and variations in wages earned by children. The variation in direct and indirect costs allows estimation of the effect of raising user fees on school enrollments. Given the three schooling options and their associated utilities, the house- hold's unconditional utility maximization problem is (4) U* = max(Uo, Ul, Uf) where U* is maximum utility, and UO, Ul, and Uf are the conditional utility functions specified in equations 1 and 2, given the constraints in equation 3. The stochastic terms in the utility function imply that the solution to the utility maximization problem gives the probability that each alternative is chosen. In a discrete choice framework these probabilities can be interpreted as demand functions. These demand functions are used to solve for the unconditional indi- rect utility and expenditure functions and to assess the welfare impact of policy changes in terms of compensating variations. Solving the utility maximization problem in equation 4 yields a system of demand functions reflecting the probabilities associated with each alternative. The functional form of each demand function depends on that of the conditional utility function and on the distribution of the stochastic variables. A parsi- monious form of the conditional utility function that does not impose a constant marginal rate of substitution (and thus allows estimates of willingness to pay to vary by household income level) is the semiquadratic, which is linear in human capital and quadratic in consumption. The conditional utility function for the two schooling options is (5) Ui = Co(HO + Si) + al Ci + a2CZ?+ ei, i = l,f Each Ei has zero mean and finite variance and is uncorrelated across individuals. Consumption net of expenditures for schooling can be derived from equation 3 to yield: (6) Ci = Y-Pi-= Y-Pi-wT,, i=l,f where w is the opportunity cost of the child's time; Ti is the child's lost hours of work due to school attendance, which varies over the local and faraway school options; and Pi is the direct price (money cost) of sending the child to school i. Substituting equation 6 into equation S yields (7) Ui = ao(Ho + Si) + aX1(Y - Pi - wT1) + a2(Y -Pt- wT,)2 +e1, i=I,f 180 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I as the utility the household derives from sending the child to either a local or faraway school. The utility from keeping the child at home is (8) UO = uOHO + U1Y + a2Y2 + Eo. The identification of the parameters in equations 7 and 8 requires that the values of human capital and consumption differ across the alternatives. Varia- tion in prices for the two schooling options identifies ct and Q!2 in equation 8. Because the decision rule involves comparing utility levels across alternatives, the conditional utility functions can be normalized relative to one of the alterna- tives without loss of generality. The utility from the no-schooling alternative is normalized to zero by subtracting equation 8 from equation 7 for both the local and faraway options. In this case equation 7 becomes (9) Ui-Uo = oSi -a1 (Pi + wTi) - a2 [2(P1 + wTi) Y - (Pi + wTi)2] + ei -Eo. The current stock of human capital is differenced out of the model. Income is differenced out of the consumption term, but it is not differenced out of the consumption-squared term. The linear consumption term represents just price, whereas the consumption-squared term includes both a price-income interaction term and a squared price term. If a linear utility function had been used, the squared terms in equations 7 and 8 would not be present, and the marginal rate of substitution across schooling alternatives would be constant. Income would then difference out of the decision rule, and consequently it would not influence the alternative chosen. In this specification the quadratic consumption term implicitly allows price effects to vary by income. Not all schools are the same; better schools will provide "more" human capi- tal per school year and thus be more attractive to parents. Quality cannot be observed directly, but one can observe various school and demographic charac- teristics thought to contribute to school quality and estimate a household human capital production function. However, the marginal utility of increments to human capital may also vary by demographic characteristics. Pollak and Wach- ter (1975) argue that the separate effects of demographic variables, such as parents' education, in the production of human capital and in the marginal utility function cannot be identified. Therefore a reduced-form model of the utility is specified from human capital accumulated by attending school i: (10) loSi = YX± + i where Xi is a vector of school quality and demographic characteristics. The coefficients in equation 10 differ for the local and faraway school alternatives. The random disturbance term, 7i, captures unmeasured portions of the quality function and may be correlated across alternatives. School quality variables are components of the human capital production function, while demographic characteristics may enter both in the production function and as direct determinants of the marginal utility received from incre- Gertler and Glewwe 181 ments to human capital. Elements in the vector Xi that are specific to households or individuals will not vary across schools and thus do not take the i subscript. Those which are school-specific will vary by schools, hence the need for the i subscript on X. Substituting equation 10 into the conditional utility function 7 yields (1) Ui = Vi + ci + q i=l,f where V = aoHo + -yXi + a1(Y - P, - wHi) + C12(Y - P, - wH,)2, and (12) UO = VO + Eo where VO = elY + U2Y2 for the no-schooling option. V simply represents the nonstochastic component of the utility obtained from each option. The Demand Functions and Welfare The demand functions for schooling are assumed to take a nested multinomial logit form (McFadden 1981) to allow for correlation across subgroups of alter- natives arid, therefore, nonconstant cross-price elasticities. Specifically, the joint distribution of the error terms is assumed to follow a type B extreme value distribution. The error terms of the schooling alternatives (in equation 11) may be correlated with each other, but not with those in the no-school alternative. Thus the probability of not going to school is (13) Tro = exp (VO) exp (VO) + [exp (VI/a) + exp (Vfl/)]a Intuitively, as the nonstochastic component (1VO) of the utility from the no-school option rises, the probability of not attending school increases. Similarly, as the nonstochastic component of the utility from either of the two schooling options (VI and V>) rises, the probability of not attending school declines. The demand for schooling alternative i is [exp (VI/o) + exp (Vf/uf)]I where a is one minus the correlation of the error terms in the conditional utility functions of the local and faraway schooling options introduced by the qi's, and the Vi's are given by equations 10 and 11. The intuition here is that once the probability of not going to school has been determined, the choice between the two schooling options is made. If the nonstochastic component of the utility from the local schooling option rises relative to the faraway schooling option, the probability of going to the local school increases through the increase in the fraction term in equation 14. An analogous situation holds if the nonstochastic component of the utility from the faraway schooling option increases. The estimated demand functions in equations 13 and 14 can be used to assess the impact of user fees on demand and revenue and, consequently, to assess the willingness to pay for closer or higher-quality schools. Willingness to pay is 182 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I calculated as a compensating variation-the amount of income needed to make the family just as well off after a price or quality change as before the change. In the case of a nested multinomial logit (Small and Rosen 1981), the compensat- ing variation, denoted by CV, for an increase in travel time is (15) CV = (1/X) fIn [exp (VO,) + [exp (VI/a) + exp (VfIou)]V] - In [exp (VO) + [exp (Vs/a) + exp (V;/a)]af1 where Vi and Vi' are evaluated at the original and new travel times, respectively, and X is the marginal utility of income. Intuitively, the change in the nonstochas- tic component of utility is given by the difference between the two main terms in the curled brackets. The marginal utility of income is then used to obtain the money value (compensating variation) of this change. IV. ESTIMATION OF THE MODEL The above model of the demand for schooling was implemented using data from the Peru Living Standards Survey. The data set and construction of the variables are the same as in Gertler and Glewwe (1989). The Variables "Consumption" is the value of per capita consumption associated with each option. It equals family income under the no-schooling option, and family in- come minus the direct and indirect costs of schooling under the other options. Consumption is observed under the current choice. Thus if the child is in school, family income is calculated as the sum of observed consumption and the price of the schooling option chosen. The indirect cost of the schooling option is the product of the local wage rate for children and the difference in hours worked by students and nonstudents, as given in table 4. If a child goes to a faraway school, it is assumed that all the child's labor is lost. The direct (monetary) cost of schooling was calculated from regional means of expenditures on tuition and books, plus mean transportation costs calculated for three different categories of travel time. Estimates were also done using predicted hours worked by students and nonstudents in order to account for sample selectivity that might be present in the figures given in table 4. Evidence for selectivity was weak, and the results were essentially unchanged (Gertler and Glewwe 1989). The age of the child is included as a variable because older children may be less inclined to continue schooling at a given grade level. "Year in school" is the number of school years completed by the child. It is included to see if some grades of secondary school are perceived to be more valuable than others. A dummy variable is included that takes the value of 1 if the child is female and 0 if the child is male. The number of years of completed schooling by the child's father and mother are also included. Parental education measures preferences for education, possi- Gertler and Glewwe 183 ble ability of children, and assistance given by parents to children in school- work. Another family-characteristic variable that is included is the number of siblings between the ages of 13 and 17. "Travel time" is interpreted as a school quality variable. Longer travel times to local schools reduce the hours of time spent by children in school and thus reduce the quality of education they receive. The other school quality variable is a lack of teachers at the local primary school, which is assumed to be highly correlated with a lack of teachers in local secondary schools. Unfortunately, there are no other variables that indicate quality of education or school quality. The Estimates Estimates of the model of educational choice for secondary school in rural Peru are given in table 5. The consumption and consumption-squared coeffi- cients indicate that the demand for schooling rises with expenditure levels but is nonlinear. The estimate of the correlation across the error terms is between 0 and 1, which indicates that the nested multinomial logit model is correctly specified. The other parameter estimates were allowed to vary by choice of schooling but in some cases have similar magnitudes. Age has a significantly Table 5. Estimation Results for the Nested Multinomial Logit Educational Choice Model Independent variable Coefficient t-statistic Consumptiona 0.5936 2.0621 Consumpton squaredb -0.0250 -1.6554 Correlation term (sigma) 0.6865 3.3285 Local schooling option Constant 3.5229 2.3672 Age -0.5115 -3.0834 Year in school 0.6121 2.3225 Female -0.9542 -2.5174 Father's education 0.0375 0.6045 Mother's education 0.1595 1.7885 Siblings age 13-17 0.0141 0.0829 Travel time -0.4270 -5.4625 Teacher shortage -1.3419 -2.7956 Faraway schooling option Constant 1.6774 1.0167 Age -0.4561 -2.6447 Year in school 1.0458 3.5927 Female -0.9472 -2.2289 Father's education -0.0006 -0.0088 Mother's education 0.0259 0.2729 Siblings age 13-17 -1.0731 -4.6619 Number of observations 718 - Log likelihood 569.65 a. Variable divided by 100 for estimation. b. Variable divided by 10,000 for estimation. Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. 184 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 negative effect on the demand for schooling in both options, indicating that conditional on finishing primary school, the demand for schooling is stronger for younger children. There is clearly a lower demand by parents for both types of schooling for female children. The father's education has little effect on the demand for either type of schooling. Finally, "year in school" has a strong positive effect on the demand for both types of schooling, which suggests that the final years of secondary school are more valuable than the initial years, all things being equal. The other parameter estimates varied by schooling options. The mother's education has a sizable positive effect on the demand for the local schooling option but has almost no effect on that for the faraway schooling option. The presence of other siblings of the same age, which is a proxy for other demands on scarce household resources, has no effect on the demand for the local school- ing option but has a strongly negative effect on the faraway option. Finally, the two variables indicating quality of education-travel time (which implies less classroom time) and a dummy variable (which indicates lack of teachers in the local primary school)-are significantly negative. The parameter estimates in table 5 are used in table 6 to predict the proba- bility of choosing the three schooling options by gender and by household expenditure level per capita. All other variables are evaluated at their mean levels. At all expenditure levels the probability of males being in local schools is higher than that for females, and the same is true for probabilities of being in faraway schools. For example, at the level of 5,000 intis per capita per month, the probability of a girl attending a local school is about 0. 19, whereas for a boy it is 0.28, and the probabilities for attending a faraway school are 0.13 and 0.19 for girls and boys, respectively. For both girls and boys the probability of going Table 6. Probability of Choosing Various Schooling Options for Males and Females by Household Expenditure Level Household expenditure Males Females level (intis per Far- Far- capita per Local away No Local away No month) school school school school school school 1,000 0.280 0.166 0.554 0.185 0.110 0.705 2,000 0.281 0.173 0.547 0.186 0.115 0.699 3,000 0.281 0.179 0.540 0.187 0.120 0.693 4,000 0.281 0.186 0.533 0.188 0.125 0.687 5,000 0.281 0.193 0.526 0.189 0.131 0.681 6,000 0.281 0.200 0.519 0.190 0.136 0.674 7,000 0.281 0.207 0.511 0.191 0.142 0.668 8,000 0.281 0.215 0.504 0.191 0.147 0.661 9,000 0.281 0.223 0.497 0.192 0.153 0.655 10,000 0.280 0.230 0.489 0.193 0.159 0.648 Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. Gertler and Glewwe 185 Table 7. Arc Price Elasticities by Gender Quartiles of household income level Range of increase Next Next All in schoolfees (intis) Lowest lowest highest Highest households Average household income 7,236 13,154 21,002 48,679 20,483 Males Local school 0-300 -0.19 -0.18 -0.16 -0.09 -0.18 300-600 -0.35 -0.33 -0.29 -0.16 -0.32 600-900 -0.53 -0.49 -0.44 -0.24 -0.47 900-1,200 -0.73 -0.67 -0.59 -0.33 -0.64 Faraway school 0-300 -0.22 -0.21 -0.19 -0.11 -0.18 300-600 -0.40 -0.37 -0.33 -0.19 -0.33 600-900 -0.59 -0.55 -0.49 -0.27 -0.49 900-1,200 -0.80 -0.74 -0.66 -0.36 -0.66 Females Local school 0-300 -0.22 -0.20 -0.18 -0.11 -0.19 300-600 -0.39 -0.36 -0.32 -0.19 -0.32 600-900 -0.57 -0.53 -0.47 -0.27 -0.48 900-1,200 -0.78 -0.72 -0.64 -0.36 -0.65 Faraway school 0-300 -0.24 -0.22 -0.20 -0.12 -0.20 300-600 -0.43 -0.39 -0.35 -0.20 -0.36 600-900 -0.63 -0.58 -0.51 -0.29 -0.52 900-1,200 -0.85 -0.78 -0.69 -0.38 -0.70 Note: Arc elasticities for a given price range are defined as the percentage change in demand (in this case the probability of attending a certain school) divided by the percentage change in price. The two endpoints of the price range are used to calculate the percentage change in the price. Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. to school rises with expenditure levels. For girls this is true of both local schools and faraway schools; for boys the probability of attending a local school is essentially unchanged, whereas that of attending a faraway school rises substan- tially. Although households that are better-off are more likely to send both girls and boys to school, the gap between girls and boys is found among all house- holds and does not disappear for the better-off rural households in Peru. The continued preference for sending boys to school also appears in the estimates of arc price elasticities of the demand for schooling given in table 7. For both boys and girls, arc price elasticities are less than unity in absolute value, indicating that the demand for schooling is relatively insensitive to price changes. The elasticities are lower for wealthier households but increase at higher levels of proposed increases in school fees, for both girls and boys. Price elasticities for girls are always higher in terms of absolute value, which indicates that the demand for schooling for girls is more sensitive to price increases than for boys. This suggests that increases in user fees may reduce schooling more for girls than for boys. But Gertler and Glewwe (1989) argue that price elasticities alone do not indicate whether parents are willing to pay to improve school 186 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 1 Table 8. Annual School Fee Parents Would be Willing to Pay to Reduce Travel Time by Two Hours in Rural Peru (June 1985 intis) Household expenditures per capita (intis per month) Male Female 1,000 440 374 2,000 486 414 3,000 542 462 4,000 612 523 5,000 702 602 6,000 823 702 7,000 994 855 8,000 1,253 1,080 9,000 1,691 1,462 10,000 2,600 2,252 Source: Authors' calculations from the unpublished 1985-86 Peru Living Standards Survey. quality or to construct new schools closer to their communities. Similarly, they do not necessarily indicate that parents are less willing to pay for improvements in schooling for girls than they are for boys. Estimates of the willingness to pay for improvements in schooling are given in table 8, which shows how much parents are willing to pay to reduce travel time to the nearest secondary school by two hours. If the government were to build a new school that reduced travel by this amount, the full cost of supporting teacher salaries would require an annual school fee of 400 intis (Gertler and Glewwe 1989). With one borderline exception, parents of both boys and girls are willing to pay this cost for closer schools. Table 8 shows that parents are more willing to pay for reduced travel time for boys than for girls, and that this holds for both poor and better-off households. Yet the differences here are not astounding; parents are still quite willing to pay to improve schooling for girls even though there is evidence that the utility of secondary education for women is less than that for men in rural Peru. Averag- ing across household expenditure levels, parents are willing to pay only about 17 percent more for boys than for girls. This indicates that parents value secondary education for girls even though economic returns to it in the labor market are doubtful. Several interpretations of these results, not necessarily mutually exclusive, are possible. First, secondary education for girls may be useful in their work around the home. Second, some parents may expect their daughters to migrate to urban areas, where there are substantial returns to secondary education for female wage earners. Third, to the extent that returns to higher education are signifi- cant for women (Moock, Musgrove, and Stelcner 1989; Jacoby 1989), some parents may expect their daughters to continue their education beyond second- ary school and thus are willing to pay for secondary education. Fourth, educated daughters may simply be directly valued by many parents. Gertler and Glewwe 187 V. CONCLUSION For developing countries whose budgetary resources are limited, raising reve- nue for education by raising school fees is a policy option. There is much anecdotal evidence that girls are not treated favorably, and it appears that the returns to secondary education for girls in rural Peru are not very strong. Exam- ination of the empirical evidence indicates that parents in rural Peru place more value on sending boys to secondary school than on sending girls. However, there is adequate demand for girls' schooling so that raising user fees to generate teachers' salaries (for a nearer school) will not result in a welfare loss that will cause parents to remove their daughters from school. This is true even among relatively poor rural households. Whether this is true in other countries remains to be seen; rigorous empirical work along the lines done here is needed. It is possible that larger gender differ- ences in the willingness to pay for schooling may lead to a marked decrease in school attendance among girls. The obvious policy remedy is to set lower school fees for girls relative to boys. This type of price discrimination will allow for more flexibility in raising revenue to expand educational opportunities while at the same time maintaining educational opportunities for girls. Such a policy has recently been adopted in parts of Bangladesh and Nigeria and should be taken seriously in any developing country where providing educational opportunities to girls is a high priority. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Behrman, Jere, and Nancy Birdsall. 1983. "The Quality of Schooling." American Eco- nomic Review 73 (5): 928-46. -. 1985. "The Quality of Schooling: Reply." American Economic Review 75 (5): 1202-05. Birdsall, Nancy. 1983. "Strategies for Analyzing Effects of User Charges in the Social Sectors." Discussion Paper 1983-9. World Bank, Country Policy Department, Wash- ington, D.C. Processed. Fernandez, Hernan. 1986. "Women's Educational Situation in Peru." Instituto Nacional de Investigacion y Desarrollo de la Educaci6n, Lima, Peru. Processed. Gertler, Paul, and Paul Glewwe. 1989. The Willingness to Pay for Education in Develop- ing Countries: Evidence from Rural Peru. LSMS Working Paper 54. Washington, D.C.: World Bank. Grootaert, Christiaan, and Ana-Maria Arriagada. 1986. "The Peru Living Standards Survey: An Annotated Questionnaire." World Bank, Population and Human Re- sources Department, Washington, D.C. Processed. Jacoby, Hanan. 1989. "The Returns to Education in the Agriculture of the Peruvian Sierra." World Bank, Washington, D.C. Processed. 188 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. I Jimenez, Emmanuel. 1987. Pricing Policies in the Social Sectors. Baltimore, Md.: Johns Hopkins University Press. King, Elizabeth. 1989. Does Education Pay in the Labor Market? The Labor Force Participation, Occupation, and Earnings of Peruvian Women. LSMS Working Paper 67. Washington, D.C.: World Bank. McFadden, Daniel. 1981. "Econometric Models of Probabilistic Choice.' In C. Manski and D. McFadden, eds., Structural Analysis of Discrete Data with Econometric Appli- cations. Cambridge, Mass.: MIT Press. Moock, Peter, Philip Musgrove, and Morton Stelcner. 1989. "Education and Earnings in Peru's Informal Nonfarm Family Enterprises.' PRE Working Paper 236. World Bank, Population and Human Resources Department, Washington, D.C. Processed. Pollak, R., and M. Wachter. 1975. "The Relevance of the Household Production Func- tion and Its Implications for the Allocation of Time." Journal of Political Economy 83:255-99. Small, Kenneth A., and Harvey S. Rosen. 1981. "Applied Welfare Economics with Discrete Choice Models.' Econometrica 49 (1): 105-30. Stein, William. 1972. "Mothers and Sons in the Andes: Developmental Implications." State University of New York at Buffalo, Department of Anthropology. Processed. Stelcner, Morton, Ana-Maria Arriagada, and Peter Moock. 1987. Wage Determinants and School Attainment among Men in Peru. LSMS Working Paper 38. Washington, D.C.: World Bank. Thobani, Mateen. 1983. Charging User Fees for Social Services: The Case of Education in Malawi. World Bank Staff Working Paper 572. Washington, D.C. Vargas, Virginia. 1987. "Reflecting on Women's Education in Peru.' Centro Flora Tri- stan, Lima, Peru. Processed. World Bank. 1986. Financing Education in Developing Countries: An Exploration of Policy Options. Washington, D.C. .1989. World Development Report 1989. New York: Oxford University Press. Free to Subscribers in Developing Countries To subscribe to The World Bank Economic Review (three issues a year) or The World Bank Research Observer (two issues a year), complete and return this coupon to World Bank Publications. The World Bank Economic Review is directed at an international readership among professional economists and social scientists. It publishes original research that emphasizes empirical applications and testing. The World Bank Research Observer is for anyone with a professional interest in economic development. Its overviews of key issues in development econom- ics are intended for policymakers, practitioners in the field, students, and economists seeking surveys of work outside their own specialties. Subscribers to either journal also receive the Proceedings of the World Bank Annual Conference on Development Economics. . READERS ,N D, g-i- -g-g gE PING SN: C...n....id' oplim,entarys sbsciptosobothi .o.as.r...iall t.raeswt OrEnclosed ankPubismycations, pa n U.S. .....o..ollar toWord. .. .OBiS T Strgee m lNWl R TI 0k C Te Word Ban Eco reve Washington, D 20433 USA U he World an ReSerc Ob,sere Coupond includingathr method ofnpayment,mandbmeilrto: .Woraldi Banraio Puaiteon O Bill me (institutionalvidcustomersionly). Icrdit puchard oumber o itp.nme ....................................~ ~ ~ ~ ~ ~ ~~~~~~................ In lude purhase oderoith p. o . number:... Plese..r...t..ot.sdes.f.....uscipto.couo. Subscrption oupon o ::. .. . . . Please print clearly. Name __________________________________ Title ___________________________ Organization name _____________________ Organization address_____________________ Country _______________ _ Postal code_ Please check only one O 01 International or regional organization, such as UN agency or develop- ment bank (includes field offices) O 02 Central bank; finance or planning ministry O 03 Other government agency or ministry (including embassies) o 04 Nongovernmental organization or association O 5s Research institution a 06 University faculty O 07 Library (university or public) O 08 Bookstore Oa 0 Commercial bank or other financial institution (including savings and loan associations and security dealers) o i0 Business- goodsn a 11X Business-services fe whery o 12 Student O 13 News media O 14 Other______ Check up to six categories Eo R0 General economics: theory, history, systems O Ri Economic growth: development, planning, fluctuations O R2 Economic statistics O R3 Monetary and fiscal theory and institutions O R4 International economics O R5 Administration, business finance, marketing, accounting a R6 Industrial organization, technological change, industry studies O R7 Agriculture, natural resources, rural development OI R8 Labor, population and human resources O R9 Welfare programs, consumer economics, urban and regional economics Plea...sinessmpletebothsides.o..th s rt co.on The Transformation of Economies in Centml and Easten Europe Issues, Progress, and Prospects Alan H. Gelb and Cheryl W. Gray Reviews the history of Central and Eastern Europe's economic systems and analyzes the transition toward a private market economy. 82 pages ! ISBN 0-8213-2870-5 / US$6.95 / Order Stock #11870 Managing Inflation in Socialist Economies in Transition Simon Commander, editor Addresses significant changes occurring over the last year, including the reform process, economic stabilization, and approaches to the establishment of free markets--including issues involved in privatization. 280 pages / ISBN 0-8213-1778-4 / US$15.95 I Order Stock #11778 Reforming Central and Eastern European Economies Iniil Results and Challenge Vittorio CofbO, Fabrizio Coricelli, and Jan Bossak, editors Analyzes adjustment lending, speed of adjustment and sequencing policies, the design of stabilization programs and their effects on the economy, and adjustment in foreign financial trade. 304 pages / ISBN 0-8213-1893-4 / US$24.95 / Order Stock 11893 The most useful, comnprehensive, and detailed single set of statistics available on the debt of developing countries. -Robert H. Cassen Director, Intemational Development Centre, University of Oxford World Debt Tables 199 1-92 * Features complete and up-to-date data on debtfor more than 100 developing countries * Includes the latest 1990 data and projections through 2000 * Periodic supplements are sent throughout the year as new data become available When you need current and reliable data on developing country debt, you will not find a better source than World Debt Tables 1991-92. This new edition gives you instant access to data gathered through the World Bank's exclusive Debtor Reporting System. You also get an analysis of the world debt situation-from top World Bank economists-that puts the data in perspective. World Debt Tables 1991-92 Volume 1. Analysis and Summary Tables Volume 2. Country Tables Complete two-volume set and supplements ISBN 0-8213-1984-1 / US$125.00 Order Stock #11984 Available separately Volume 1. Analysis and Summary Tables 193 pages / ISBN 0-8213-1971-X /US$16.95 / Order Stock #11971 To order Wor-ld Debt Tables 1991-92, list title, stock number, and price on the coupon that follows. -~~~ < Title Stock Number Price Subtotal US$ Shipping and handling US$ 3.50 Airmail surcharge outside USA (US$6.00 per copy) US$ Total US$ Check your method of payment. Orders from individuals must be accompanied by payment or credit card information. Ask local distributors about forms of payment acceptable to them. W Enclosed is my check payable to World Bank Publications in U.S. dollars. F Charge my F VISA [ MasterCard [I American Express These credit cards are accepted for orders addressed to Philadelphia. (Check with your local distributor about acceptance of credit cards in your country.) Credit card account number Expiration date Signature Institutional customers only: D Send invoice and reference purchase order no. Please print clearly Name Title Firm Address__ City_ State Postal Code_ _ Country Telephone Customers ir the United States: Detach this form and mail to World Bank Publications, P.O. Box 7247-8619, Philadelphia, PA 19170-8619, U.S.A. Customers outside the United States: Contact World Bank Publications, 1818 H Street, NW, Washington, DC 20433, U.S.A. for a list of World Bank publications distributors. Write to the local distributor in your area for information on prices and terms of payment. Coupons received in Washington from countries with authorized distributors will be returned to the customer. Customers in countries with no authorized distributor, return your coupon to World Bank Publications, P.O. Box 7247-8619, Philadelphia, PA 19170-8619, U.S.A. Coming in the next issue of THE WORLD BANK ECONOMIC REVIEW May 1992 Volume 6, Number 2 Articles on ... * The Economics of Farm Fragmentation: Evidence from Ghana and Rwanda by Benoit Blarel, Peter Hazell, Frank Place, and John Quiggin * The Short- and Long-Run Effects of Fiscal Policy by Edward F. Buffie * The Dynamics of Optimal Gradual Stabilizations by Alex Cukierman and Nissan Liviatan * Transportation Policy and Panterritorial Pricing in Africa by Mlark Gersovitz * The Business Cycle Associated with Stabilizations Based on Exchange Rates by Miguel A. Kiguel and Nissan Liviatan * Project Evaluation and Uncertainty in Practice in World Bank Projects by Gerhard Pohl and Dubravko Mihaljek * Household Saving in Developing Countries by Klaus Schmidt-Hebbel, Steven B. Webb, and Giancarlo Corsetti * The Impact of Pricing Policies toward Cocoa and Coffee Producers in C6te d'Ivoire by Pravin K. Trivedi and Takamasa Akiyama * Researching the Link between Trade and Productivity by James R. Tybout