THE WORLD BANK ECONOMIC REVIEW Volume 6 September 1992 Number 3 Measuring the Independence of Central Banks and Its Effect on Policy Outcomes Alex Cukierman, Steven B. Webb, and Bilin Neyapti On the Transmission of World Agricultural Prices Yair Mundlak and Donald F. Larson How Small Enterprises in Ghana Have Responded to Adjustment William F. Steel and Leila M. Webster The Dynamics of Optimal Gradual Stabilizations Alex Cukierman and Nissan Liviatan Uniform Commercial Policy, Illegal Trade, and the Real Exchange Rate: A Theoretical Analysis Stephen A. O'Connell Maize and the Free Trade Agreement between Mexico and the United States Santiago Levy and Sweder van Wijnbergen The Optimal Currency Composition of External Debt: Theory and Applications to Mexico and Brazil Stijn Claessens Household Saving in Developing Countries: First Cross-Country Evidence Klaus Schmidt-Hebbel, Steven B. Webb, and Giancarlo Corsetti Cumulative Indexes of Articles and Authors for Volume 6 THE WORLD BANK ECONOMIC REVIEW EDITOR Ravi Kanbur CONSULTING EDITOR Sandra Gain EDITORIAL BOARD Kaushik Basu, University of Delhi John Holsen Guillermo 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 Wijnbergen The World Bank Economic Review is a professional journal for the dissemination of World Bank-sponsored research that informs policy analyses and choices. It is directed to an interna- tional readership among economists and social scientists in government, business, and interna- tional agencies, as well as in universities and development research institutions. The Review emphasizes policy relevance and operational aspects of economics, rather than primarily theoreti- cal and methodological issues. It is intended for readers familiar with economic theory and analysis but not necessarily proficient in advanced mathematical or econometric techniques. Arti- cles 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 mav 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. I 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. The WVorld Bank Economic Review is published three times a year (January, May, and Septem- ber) by the World Bank. The annual subscription fee is US$25 for individuals and $45 for institutions. Single copies may be purchased at $10.95. Subscription orders should be sent to: World Bank Publications, Box 7247-7956, Philadelphia, PA 19170-7956 U.S.A. Subscriptions to The W'orld 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. t 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. The World Bank encourages dissemination of its work and will normally give permission promptly and, when the intended reproduction is for noncom- mercial purposes, without asking a fee. Permission to make photocopies is granted through the Copyright Clearance Center, 27 Congress Street, Salem, MA 01970 U.S.A. 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 Index®. It is available in microform through University Microfilms, Inc., 300 North Zeeb Road, Ann Arbor, Michigan 48106, U.S.A. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 353-398 THE WORLD BANK ECONOMIC REVIEW Volume 6 September 1992 Number 3 Measuring the Independence of Central Banks and Its 353 Effect on Policy Outcomes Alex Cukierman, Steven B. Webb, and Bilin Neyapti On the Transmission of World Agricultural Prices 399 Yair Mundlak and Donald F. Larson How Small Enterprises in Ghana Have Responded 423 to Adjustment William F. Steel and Leila M. Webster The Dynamics of Optimal Gradual Stabilizations 439 Alex Cukierman and Nissan Liviatan Uniform Commercial Policy, Illegal Trade, and the Real 459 Exchange Rate: A Theoretical Analysis Stephen A. O'Connell Maize and the Free Trade Agreement between Mexico 481 and the United States Santiago Levy and Sweder van Wijnbergen The Optimal Currency Composition of External Debt: 503 Theory and Applications to Mexico and Brazil Stijn Claessens Household Saving in Developing Countries: 529 First Cross-Country Evidence Klaus Schmidt-Hebbel, Steven B. Webb, and Giancarlo Corsetti Cumulative Index of Articles and Authors for Volume 6 549 THE WORLD BANK ECONOMIC REVIE W. VOl.. 6. NO. 3: 353-395 Measuring the Independence of Central Banks and Its Effect on Policy Outcomes Alex Cukierman, Steven B. Webb, and Bilin Neyapti Making the central bank an agency with the mandate and reputation for maintaining price stability is a means by which a government can choose the strength of its commit- ment to price stability. This article develops four measures of central bank indepen- dence and explores their relation with inflation outcomes. An aggregate legal index is developedforfour decades in 72 countries. Three indicators of actual independence are developed: the rate of turnover of central bank governors, an index based on a ques- tionnaire answered by specialists in 23 countries, and an aggregation of the legal index and the rate of turnover. Legal independence is inversely related to inflation in industrial, but not in develop- ing, countries. In developing countries the actual frequency of change of the chief executive officer of the bank is a better proxy for central bank independence. An inflation-based index of overall central bank independence contributes significantly to explaining cross-country variations in the rate of inflation. "Willpower is trying hard not to do something that you really want to do," said Frog. "You mean like trying not to eat all these cookies," asked Toad. "Right," said Frog. He put the cookies in a box. "There, now we will not eat any more cookies." "But we can open the box," said Toad. "That is true," said Frog. He tied some string around the box. He got a ladder and put the box up on a high shelf. "There, now we will not eat any more cookies." "But we can climb the ladder. . . ." (Lobel 1972) Institutions cannot absolutely prevent an undesirable outcome, nor ensure a desirable one, but the way that they allocate decisionmaking authority within the public sector makes some policy outcomes more probable and others less likely. An important example of this principle concerns the balance of authority between the central bank and the executive and legislative branches of govern- ment. Economists and practitioners in the area of monetary policy generally Alex Cukierman is with the Department of Economics, Tel Aviv University; Steven B. Webb is with the Country Economics Department at the World Bank; and Bilin Neyapti is with the Country Economics Department at the World Bank and the Economics Department, University of Maryland. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 353 354 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 believe that the degree of independence of the central bank from other parts of government affects the rates of expansion of money and credit and, through them, important macroeconomic variables, such as inflation and the size of the budget deficit. Ultimately, the central bank's authority and scope of action depends on the government. But governments often pass laws and follow customs that grant their central banks authority and autonomy to pursue price stability, even when it conflicts with other government objectives. Making the central bank an agency with the mandate and reputation for maintaining price stability benefits the economy and the government itself in various ways. Central bank indepen- dence is one of the means by which a government can choose the strength of its commitment to price stability (Cukierman 1992a, chap. 23, and 1992b; Lohmann 1992). A vast literature discusses the costs of inflation (see Fischer [1986] for a survey); the central bank's pursuit of price stability can help reduce these costs. Price stability is also necessary, although far from sufficient, for developing a local capital market where both government and businesses can borrow more conveniently and cheaply in the long run. International financing, such as for the countries recovering from hyperinflation in the 1 920s, has often been conditional on the central bank's mandate and authority to pursue the stability of prices and exchange rates. Pursuing price stability necessarily competes at least some of the time with other tasks that central banks can and often do perform-such as managing the government's financial transactions, financing the government's deficits with money issue, financing development projects, and bailing out insolvent busi- nesses, including banks and publicly owned enterprises. Although most govern- ments recognize the long-run benefit of price stability, other goals often loom larger in the short run. Assuring price stability, therefore, usually requires ensur- ing that the central bank is not forced to perform these functions, at least not when they would cause inflation. Sometimes the government or the treasury takes direct responsibility for limiting the demands on a subservient central bank. Even in these cases, but especially in the more typical case where the government has strong tendencies to focus on issues other than price stability, central bank independence and an explicit mandate to pursue price stability are generally regarded as important institutional devices for ensuring price stability. This belief has eluded comprehensive verification in the past because of the difficulties in measuring the autonomy of central banks independently from Many individuals helped assemble the data for this study. Particular thanks are due to Walter Wasser- fallen, Herman-Josef Dudler, John Flemming, D. Hiss, Wolfgang Schill, Marc-Olivier Strauss-Kahn, Ann Johannessen, and those who answered the questionnaire. The authors also received useful comments from James Alt, Tomas Balino, Edgardo Barandiaran, Paul Beckerman, Mario Blejer, Michael Bruno, Gerald Caprio, Max Corden, Andrew Crockett, Jose de Gregorio, Patrick Downes, Robert Effros, Valeriano Garcia, Sergio Pereiro Leite, Alfredo Leone, Klaus Richel, Lawrence Summers, V. Sundarara- jan, Mark Swinburne, Richard Webb, Eduardo Wiesner, and participants at the November 1991 NBER Conference on Political Economics. Cukierman, Webb, and Neyapti 355 observation of the inflation outcomes. Actual, as opposed to formal, central bank independence depends not only on the law, but also on many other less- structured factors, such as informal arrangements between the bank and other parts of government, the quality of the bank's research department, and the personality of key individuals in the bank and the (rest of the) government. Because of the difficulty in quantifying such features in an impartial manner, previous studies developed indexes of central bank independence based mostly on legal independence-and only for the industrial countries at that. The multi- country studies that attempt to rank independence for a cross-section of coun- tries include Bade and Parkin (1980), Skanland (1984), Parkin (1987), Alesina (1988), Masciandaro and Tabellini (1988), Bodart (1990), Swinburne and Castello-Branco (1991), and Grilli, Masciandaro, and Tabellini (1991), plus Leone's (1991) study of limits on lending in several industrial and developing countries. There is also a large literature of single- or multicountry case studies, which includes Mittra (1977), Schokker (1980), Eizenga (1983), Kearney (1984), Dotsey (1986), Epstein and Schor (1986), Bordes and Strauss-Kahn (1987), Bordo and Redish (1987), Eizenga (1987), Keenan and Mayes (1987), Suzuki (1987), Goodhart (1988), Holtfrerich (1988), Willett (1988), Fazio (1991), Cargill and Hutchison (1990), Mayer (1990), Meltzer (1991), Maxfield (1992), and Volcker, Mancera, and Godeaux (1991). Indicators based only on the law have two problems. First, the laws are incomplete in that they cannot specify explicitly the limits of authority between the central bank and the political authorities under all contingencies. These voids are filled by tradition at best and by power politics at worst. Second, even when the law is quite explicit, actual practice may deviate from it. This article develops unified and broadly based measures of central bank independence, uses them to rank central banks by their degree of independence, and explores the relation between their independence and inflation outcomes. The study goes beyond previous work in three dimensions. First, the set of countries is wider, including up to 72 countries (21 industrial countries and 51 developing countries). The wider sample makes it possible to examine whether there are systematic differences in central bank independence between industrial and developing countries. Second, the coverage in time goes back to the 1950s, if the bank existed then. Third, the study uses a wider range of information on central bank independence. The spirit of the law and its application in practice are generally more important than the letter of the law. In addition to coding characteristics of the central bank law, the study looks at the actual frequency of turnover of central bank governors and at the questionnaire responses from specialists on monetary policy in a subsample of 23 countries. The questionnaire was designed to identify divergences between the central bank's charter and actual practice. We use several different indicators of independence because, in addition to the noise that they contain, each indicator captures a somewhat different aspect of independence. Examining the relation of inflation to alternate indicators for independence 356 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 reveals that legal independence is an important determinant of inflation in the industrial countries. In developing countries, by contrast, governors' turnover is strongly and positively associated with inflation. This finding suggests that there are larger divergences between actual practice and the law in developing than in industrial countries (see, for example, Bodart 1990 and Leone 1991). I. MEASURES OF THE LEGAL INDEPENDENCE OF CENTRAL BANKS Legal independence is, of course, an essential component of actual indepen- dence, but it is also of interest for several other reasons. First, it indicates what is the degree of independence that legislators meant to confer on the central bank. Second, practically all existing attempts at systematically characterizing central bank independence rely solely on legal aspects of independence (Bade and Parkin 1980; Banaian, Laney, and Willett 1983; Skanland 1984; Parkin 1987; Alesina 1988; Masciandaro and Tabellini 1988; Grilli, Masciandaro, and Tabellini 1991). Establishing comparability with previous studies requires an index of legal independence for our sample of countries. The laws of central banks differ in their focus, scope, and degree of detail. Many provisions in central bank charters have no direct bearing on the issue of central bank independence. Ranking central bank charters by their degree of legal independence is therefore difficult and inevitably requires subjective judgment. Coding Legal Independence Our coding of the legal central bank independence followed two principles. First, we coded only a few narrow but relatively precise legal characteristics. Second, we used only the written information from the charters. Additional information on how the law is applied was deliberately left out, since it is reflected by separate indexes 'that are discussed in section II. These principles make it possible to rank central banks by their degree of independence in various legal dimensions with relatively few subjective judgments and to focus on con- crete details of the law rather than on a broader but more impressionistic view of it. The legal characteristics of the central bank as stated in its charter are grouped into four clusters of issues: * The appointment, dismissal, and term of office of the chief executive officer of the bank-usually the governor * The policy formulation cluster, which concerns the resolution of conflicts between the executive branch and the central bank over monetary policy and the participation of the central bank in the budget process * The objectives of the central bank * Limitations on the ability of the central bank to lend to the public sector; such restrictions limit the volume, maturity, interest rates, and conditions Cukierman, Webb, and Neyapti 357 for direct advances and securitized lending from the central bank to the public sector. The clusters were built up from 16 different legal variables, each coded on a scale of 0 (lowest level of independence) to 1 (highest level of independence). The detailed classification and codings appear in table 1. The codes are set so that a higher number indicates what we expected would lead to a stronger mandate and greater autonomy for the central bank to pursue price stability. In coding various central banks by the degree of independence within each group of characteristics, the following criteria were used. Central banks in which the legal term of office of the chief executive officer (CEO) is longer and in which the executive branch has little legal authority in appointing or dismissing the governor are classified as more independent in the CEO dimension. By the same logic, central banks with wider authority to formulate monetary policy and to resist the executive branch in cases of conflict are classified as more independent in the policy formulation dimension. For the objective of the central bank, there are six possible ratings, according to the prominence given to price stability compared with other stated objectives that might conflict with price stability. For instance, when the charter specifies price stability as the main or only goal, the bank is classified as being more independent in this dimension than a central bank with objectives in addition to, but not inconsistent with, price stability, such as stable banking. These banks are, in turn, classified as more independent than banks whose objectives include things like full employment, which might conflict with price stability. The objec- tives variable is designed to capture the legal mandate of the bank to single- mindedly pursue the objective of price stability. Only the central banks in the Federal Republic of Germany and the Philippines have unequivocal legal man- dates for price stability. The objectives variable does not, therefore, reflect the general level of independence from government, in contrast to the CEO and policy formulation variables. In Rogoff's (1985) terminology, the objectives variable measures the strength of the "conservative bias" of the bank's charter. Similarly, we classify a central bank with tighter limits on its lending to the public sector as more independent to pursue the objective of price stability. These limitations encompass a number of more detailed variables, such as sepa- rate limitations on advances and securitized lending and restrictions on maturi- ties and on interest rates. The stricter the limitation, the higher the independence coding given to the bank in that dimension. The comparability of various types of limitations is complicated because the limitations are specified in different ways in different countries. In a few countries limitations on lending are spe- cified in absolute cash amounts and in others as a percentage of central bank liabilities. The limitation is formulated in most cases as a percentage of govern- ment's revenues from taxes but in a minority of cases as a percentage of govern- ment's expenditures. The "bite" of these limitations obviously depends on the magnitudes of the reference variables. Other things being equal, however, abso- 3s8 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 1. Variables for Legal Central Bank Independence Variable Numerical number Description of variable Weight coding 1 Chief executive officer (CEO) 0.20 a. Term of office Over 8 years 1.00 6 to 8 years 0.75 5 years 0.50 4 years 0.25 Under 4 years or at the discretion of appointer 0.00 b. Who appoints CEO? Board of central bank 1.00 A council of the central bank board, executive branch, and legislative branch 0.75 Legislature 0.50 Executive collectively (e.g. council of ministers) 0.25 One or two members of the executive branch 0.00 c. Dismissal No provision for dismissal 1.00 Only for reasons not related to policy 0.83 At the discretion of central bank board 0.67 At legislature's discretion 0.50 Unconditional dismissal possible by legislature 0.33 At executive's discretion 0.17 Unconditional dismissal possible by executive 0.00 d. May CEO hold other offices in government? No 1.00 Only with permission of the executive branch 0.50 No rule against CEO holding another office 0.00 2 Policy formulation 0.15 a. Who formulates monetary policy? Bank alone 1.00 Bank participates, but has little influence 0.67 Bank only advises government 0.33 Bank has no say 0.00 b. Who has final word in resolution of conflict?a The bank, on issues clearly defined in the law as its objectives 1.00 Government, on policy issues not clearly defined as the bank's goals or in case of conflict within the bank 0.80 A council of the central bank, executive branch, and legislative branch 0.60 The legislature, on policy issues 0.40 The executive branch on policy issues, subject to due process and possible protest by the bank 0.20 The executive branch has unconditional priority 0.00 c. Role in the government's budgetary process Central bank active 1.00 Central bank has no influence 0.00 3 Objectives 0.15 Price stability is the major or only objective in the charter, and the central bank has the final word in case of conflict with other government objectives 1.00 Price stability is the only objective 0.80 Price stability is one goal, with other compatible objectives, such as a stable banking system 0.60 Price stability is one goal, with potentially conflict- ing objectives, such as full employment 0.40 Cukierman, Webb, and Neyapti 359 Table 1. (continued) Variable Numerical number Description of variable Weight coding No objectives stated in the bank charter 0.20 Stated objectives do not include price stability 0.00 4 Limitations on lending to the government a. Advances (limitation on nonsecuritized lending) 0.15 No advances permitted 1.00 Advances permitted, but with strict limits (e.g., up to 15 percent of government revenue) 0.67 Advances permitted, and the limits are loose (e.g., over 15 percent of government revenue) 0.33 No legal limits on lending 0.00 b. Securitized lending 0.10 Not permitted 1.00 Permitted, but with strict limits (e.g., up to 15 percent of government revenue) 0.67 Permitted, and the limits are loose (e.g., over 15 percent of government revenue) 0.33 No legal limits on lending 0.00 c. Terms of lending (maturity, interest, amount) 0.10 Controlled by the bank 1.00 Specified by the bank charter 0.67 Agreed between the central bank and executive 0.33 Decided by the executive branch alone 0.00 d. Potential borrowers from the bank 0.05 Only the central government 1.00 All levels of government (state as well as central) 0.67 Those mentioned above and public enterprises 0.33 Public and private sector 0.00 e. Limits on central bank lending defined in 0.025 Currency amounts 1.00 Shares of central bank demand liabilities or capital 0.67 Shares of government revenue 0.33 Shares of government expenditures 0.00 f. Maturity of loans 0.025 Within 6 months 1.00 Within 1 year 0.67 More than 1 year 0.33 No mention of maturity in the law 0.00 g. Interest rates on loans must be 0.025 Above minimum rates 1.00 At market rates 0.75 Below maximum rates 0.50 Interest rate is not mentioned 0.25 No interest on government borrowing from the central bank 0.00 h. Central bank prohibited from buying or selling government securities in the primary market? 0.025 Yes 1.00 No 0.00 Note: The ranking under each criteria indicates the degree of independence of central banks-the higher the code, the more independent the central bank. a. Often the law does not contain a separate provision on the resolution of conflict. In those cases, the variable was coded on the basis of the impression from reading the law in its entirety. If the law gives the impression that the government formulates policy guidelines that the bank simply follows, then the ranking is low. Source: Various central bank laws, Aufricht (1961, 1967); Bank for International Settlements (1963); Effros (1982); and the IMF'S computerized files on central bank laws. 360 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 lute cash limits are more binding than limits in terms of central bank liabilities, which, in turn, are more binding than limits in terms of government's revenues. The most accommodative limits are those which are specified in terms of gov- ernment's expenditures. These considerations were embodied in a "type-of- limit" variable and also influenced the classification of the variables for limita- tions on lending via advances and for limitations on lending via securities. Table 1 shows the details of the several variables for limitations on lending. Limitations on lending are also classified as stricter the nearer are the rates paid by government to market rates and the shorter are the maturities of the loans from the central bank to the public sector. They are also stricter the narrower the circle of institutions that is allowed to borrow from the central bank and the smaller the discretion of the executive branch to decide to whom and how much the central bank will lend. In addition, central bank laws that prohibit the central bank from buying government securities on the primary market are considered, all things being equal, stricter than laws that do not contain such a prohibition. The period considered covers the four decades from 1950 to 1989. It is divided into four subperiods: 1950-59, 1960-71, 1972-79, and 1980-89, which we refer to according to the decades in which they are centered. They correspond, respectively, to the gold-dollar standard period before most curren- cies had convertibility, the period of convertibility with the dollar, the period of the two oil shocks after the end of the Bretton Woods currency system, and the period of disinflation and the debt crisis. Legal variables were coded separately for each decade. Since central bank legislation changes relatively slowly, the codes are, in many cases, identical across subperiods. Nonetheless this pro- cedure captures important legislative changes for some countries. Only one code per decade was assigned for each country for each legal variable. Whenever a change occurred within a decade, the classification was done in line with the legislation that was in effect during at least half of that decade. When a central bank was founded within a decade, its legal variables were coded only if it existed for at least three years during the decade. The coded variables appear in Table A-1 following the concluding section. Aggregating the Legal Variables The individual components of legal independence are aggregated in two steps to yield a hierarchy of indexes. Later in the article we investigate the association of those indexes with other variables, such as other indicators of central bank independence and inflation. The basic data on the 16 legal variables described in table 1 were aggregated into eight legal variables as follows. The four variables concerning the appointment and term of office of the governor of the central bank were aggregated into a single variable labeled CEO, equal to the mean of the four components. The three variables under policy formulation were aggregated into a single variable by computing a weighted mean of the variables in that Cukierman, Webb, and Neyapti 361 group, with weights of 0.5 for the resolution of conflict, 0.25 for who formu- lates monetary policy, and 0.25 for active role of the central bank in formulating the government budget. The objectives variable was treated separately. The first four variables for limits on lending were treated separately; the last four vari- ables in the group were averaged with equal weights into a single variable. This aggregation procedure produces one summary legal variable for each of the first three groups in table 1, and five legal variables for the limitations on lending group. When an entry is not available (-) for one or more variables within a subgroup, only the variables with meaningful entries are aggregated. In such cases the weights of the missing variables are allocated proportionally to the remaining variables within the subgroup. When the legal variables appear at a high level of disaggregation (as in table 1), a missing observation on at least one variable precludes the use of that country or decade. Partial aggregation alleviates this problem by reducing the number of observations with entries that are not available. In addition, multicollinearity among the 16 legal variables reduces the precision of the estimated effect of each of them on inflation. Partial aggregation alleviates this problem, too. The eight legal variables from the first round of aggregation were aggregated further into a single index for each country and decade, using weights that we considered most plausible. The weights are indicated in table 1. When a compo- nent is missing, the weights of those remaining are expanded proportionately to sum to 1.0. For all the observations, the weights of the component variables summed to at least 0.7. The variable for the legal central bank independence aggregated in this way is similar to a variable aggregated with equal weights; the two are highly correlated. Table 2 ranks industrial and developing countries according to their aggregate variable for legal central bank independence for 1980-89 and also provides their average inflation rates in the 1980s. Countries classified by the World Bank as low- or middle-income, on the basis of 1985 incomes, are classified as develop- ing; the others are referred to as industrial. Observations for each country and decade are given in appendix table A-1. Austria, Germany, and Switzerland-all industrial countries-have the highest legal independence, while Morocco and Poland-both developing countries-have the lowest. Otherwise the two coun- try groups have very similar distributions of aggregate legal independence. The medians are virtually identical-0.33 and 0.34. II. INFORMAL INDICATORS OF ACTUAL INDEPENDENCE The legal status of a central bank is only one of several elements that deter- mine its actual independence. Many central bank laws are highly incomplete and leave a lot of room for interpretation. As a result, factors such as tradition or the personalities of the governor and other high officials of the bank at least par- tially shape the actual level of central bank independence. Even when the law is Table 2. Legal Central Bank Independence and Average Annual Inflation, 1980-89 Industrial economy Developing economy Legal Legal Legal central Average central Average central Average bank in- annual bank in- annual bank in- annual depen- rate of in- depen- rate of in- depen- rate of in- dence" flation', dence' flationb dence' flationb Economy (index) (percent) Economy (index) (percent) Economy (index) (percent) Germany, Fed. Rep. of 0.69 3 Greece 0.55 18 Botswana 0.33 10 Switzerland 0.64 3 Egypt 0.49 16 Zambia 0.33 25 Austria 0.61 4 Costa Rica 0.47 23 Ghana 0.31 37 Denmark 0.50 7 Chile 0.46 19 Romania 0.30 4 United States 0.48 5 Turkey 0.46 41 Bolivia 0.30 119 Canada 0.45 6 Nicaragua 0.45 128 Western Samoa 0.30 12 Ireland 0.44 9 Malta 0.44 3 China 0.29 8 Netherlands 0.42 3 Tanzania 0.44 27 Singapore 0.29 3 w Australia 0.36 8 Kenya 0.44 10 Korea, Republic of 0.27 8 iz Iceland 0.34 32 Philippines 0.43 13 Indonesia 0.27 9 Luxembourg 0.33 5 Zaire 0.43 45 Colombia 0.27 21 Sweden 0.29 8 Peru 0.43 108 Thailand 0.27 6 Finland 0.28 7 Honduras 0.43 7 South Africa 0.25 14 United Kingdom 0.27 7 Venezuela 0.43 19 Hungary 0.24 9 Italy 0.25 11 Bahamas, The 0.41 6 Uruguay 0.24 45 New Zealand 0.24 12 Portugal 0.41 16 Panama 0.22 3 France 0.24 7 Argentina 0.40 143 Pakistan 0.21 7 Spain 0.23 10 Ethiopia 0.40 4 Brazil 0.21 119 Japan 0.18 3 Lebanon 0.40 - Taiwan 0.21 5 Norway 0.17 8 Israel 0.39 72 Zimbabwe 0.20 12 Belgium 0.17 5 Barbados 0.38 7 Qatar 0.20 4 Uganda 0.38 72 Nepal 0.18 10 Nigeria 0.37 18 Yugoslavia 0.17 73 Malaysia 0.36 4 Morocco 0.14 7 Mexico 0.34 50 Poland 0.10 36 India 0.34 9 - Not available. a. The potential range of the index for legal central bank independence is from zero (minimal independence) to one (maximum independence). b. Inflation is computed in logs. Cukierman, Webb, and Neyapti 363 quite explicit, reality may be very different. For example, in Argentina the legal term of office of the governor is four years, but there is also a tradition that the governor of the central bank offers to resign whenever the government, or even the finance minister, changes. Argentine governors have invariably adhered to this tradition. As a consequence, the average actual term of office of the gover- nor was about one year from 1950 to 1989. Obviously the actual independence of the Argentine central bank is substantially lower than the legal indicators imply. It is hard to find systematic indicators of actual independence when it diverges from legal independence, and we do not pretend to resolve this mea- surement issue fully. Here we develop two indicators of actual, as opposed to legal, central bank independence from the actual frequency of change of the governor and from responses to a questionnaire sent to experts on each country. Turnover of Central Bank Governors This indicator is based on the presumption that, at least above some thresh- old, more rapid turnover of central bank governors indicates a lower level of independence. Indeed, more rapid turnover presumably creates dependence. If the political authorities frequently take the opportunity to choose a new gover- nor, they will at least have the opportunity to pick those who will do their will. Frequent turnover may reflect the firing of those who choose to challenge the government. A government would even have some incentive to appoint a gover- nor with a reputation for some independence, thereby gaining a temporary increase in the potential for stimulating output or collecting resources through seigniorage, and then use up his reputation, as happened with Arthur Burns at the Federal Reserve Bank of the United States in the 1970s. For high turnover rates, the tenure of the central bank governor is shorter than that of the executive branch. This makes the central bank governor suscep- tible to influence by the executive branch and discourages the governor from trying to implement longer-term policies, especially those that would extend beyond the election cycle. Because in most countries the electoral cycle is at least four years, it is likely that the threshold turnover, above which independence declines seriously, is somewhere between 0.2 and 0.25 changes a year ( for an average tenure of four to five years). One would expect that turnovers at the central bank that occur simultaneously with or shortly after changes in the government would indicate lower independence than turnovers that occur at other times. Further work will investigate this issue. If, however, a governor stays on for several years and perhaps outlasts several heads of government, thus presiding over price stability, the governor's reputa- tion can become strong enough to resist considerable pressure. The govern- ment's desire to preserve financial stability can deter it from challenging a well- established central bank governor. A low turnover does not necessarily imply a high level of central bank inde- pendence, however, because a relatively subservient governor may stay in office 364 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 3. Turnover Rates of the Central Bank Governor, 1950-89 (average number of changes a year) Economy 1950-89 1950-S9 1960-71 1972-79 1980-89 Industrial economy Iceland 0.03 - 0.09 0.00 0.00 Netherlands 0.05 0.00 0.08 0.00 0.10 Denmark 0.05 0.10 0.08 0.00 0.00 Luxembourg 0.08 0.10 0.08 0.13 0.00 Norway 0.08 0.10 0.08 0.00 0.10 Italy 0.08 0.00 0.08 0.25 0.00 United Kingdom 0.10 0.00 0.17 0.13 0.10 Canada 0.10 0.10 0.08 0.13 0.10 Germany, Fed. Rep. of 0.10 0.10 0.08 0.13 0.10 Australia 0.10 0.00 0.08 0.13 0.20 Finland 0.13 0.20 0.08 0.00 0.20 Switzerland 0.13 0.10 0.08 0.13 0.20 Belgium 0.13 0.10 0.08 0.13 0.20 United States 0.13 0.10 0.08 0.25 0.10 Ireland 0.15 0.10 0.17 0.13 0.20 France 0.15 0.00 0.17 0.25 0.20 Sweden 0.15 0.20 0.00 0.38 0.10 NewZealand 0.15 0.00 0.17 0.13 0.30 Austria 0.15 0.10 0.17 0.25 0.10 Japan 0.20 0.20 0.17 0.13 0.30 Spain 0.20 0.20 0.25 0.25 0.10 Developing economy Qatar 0.06 - - 0.14 0.00 South Africa 0.10 0.00 0.17 0.00 0.20 Barbados 0.11 - - 0.13 0.10 Taiwan 0.13 0.10 0.17 0.00 0.20 Philippines 0.13 0.00 0.25 0.00 0.20 Honduras 0.13 0.11 0.00 0.38 0.10 Tanzania 0.13 - 0.18 0.13 0.10 Malaysia 0.13 - 0.08 0.00 0.20 Israel 0.14 0.20 0.08 0.13 0.20 Zimbabwe 0.15 0.27 0.17 0.13 0.10 Mexico 0.15 0.10 0.08 0.13 0.30 Kenya 0.17 - 0.36 0.00 0.20 Greece 0.18 0.10 0.08 0.38 0.20 Hungary 0.18 0.38 0.17 0.13 0.10 Lebanon 0.19 - 0.24 0.25 0.10 a long time. This is probably true for countries with exceptionally low turnover rates, such as Iceland, Denmark, Norway, and the United Kingdom. Table 3 presents the average annual turnover rates in the sample countries for 1950-89 and for each decade within that period. These rates are presented separately for developing and industrial countries. Average turnover rates for 1950-89 range from a minimum of 0.034 (one change in 29 years) in Iceland to a maximum of 0.93 (average tenure of about 13 months) in Argentina. Turnover rates in developing countries extend into a range considerably above the highest rates in the industrial countries. The highest average turnover among the indus- Cukierman, Webb, and Neyapti 365 Table 3. (continued) Economy 1950-89 1950-59 1960-71 1972-79 1980-89 Developing economy (continued) Nigeria 0.19 - 0.17 0.25 0.10 Bahamas, The 0.19 - - 0.18 0.20 Morocco 0.20 - 0.25 0.00 0.20 Ethiopia 0.20 - 0.00 0.50 0.10 Colombia 0.20 0.20 0.25 0.13 0.20 Romania 0.20 0.40 0.08 0.13 0.20 Portugal 0.20 0.20 0.08 0.25 0.30 Thailand 0.20 0.40 0.08 0.25 0.10 Yugoslavia 0.23 0.30 0.17 0.25 0.20 Indonesia 0.23 0.20 0.33 0.13 0.20 Zaire 0.23 - 0.26 0.25 0.20 Nepal 0.24 0.27 0.33 0.25 0.10 Panama 0.24 - 0.56 0.00 0.20 Pakistan 0.25 0.10 0.33 0.25 0.30 Poland 0.28 0.20 0.25 0.13 0.50 Malta 0.28 - 0.27 0.38 0.20 Ghana 0.28 - 0.33 0.25 0.20 Venezuela 0.30 0.20 0.25 0.25 O.S0 Egypt 0.31 0.46 0.33 0.13 0.30 India 0.33 0.20 0.33 0.50 0.30 Peru 0.33 0.30 0.33 0.38 0.30 China 0.34 - - - 0.30 Uganda 0.34 - 0.36 0.50 0.20 Nicaragua 0.35 - 0.29 0.38 0.40 Singapore 0.37 - - 0.00 0.60 Zambia 0.38 - 0.38 0.25 0.50 Turkey 0.40 0.30 0.50 0.38 0.40 Botswana 0.41 - - 0.44 0,40 Korea, Republic of 0.43 0.31 0.67 0.13 0.50 Chile 0.45 0.20 0.33 0.50 0.80 Uruguay 0.48 - 1.03 0.38 0.30 Western Samoa 0.56 - - - 0.56 Costa Rica 0.58 0.20 0.83 0.88 0.40 Brazil 0.68 1.01 0.50 0.38 0.80 Argentina 0.93 0.71 1.08 0.88 1.00 - Not available. Note: Turnover rates were calculated if at least three years of data were available for the decade. Source: Correspondence with central banks. trial countries for 1950-89 is 0.2 (or an average tenure of five years) for Spain. More than half of the developing countries have turnover rates exceeding this maximum. Because average turnover rates for all industrial countries for 1950-89 are less than 0.2, these rates probably do not reveal much about the variations of independence within that group. But because turnover rates vary widely among the developing countries, however, they seem more likely to reveal variations in the independence of those governors. A governor's legal term of office does not seem to have much effect on the 366 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 4. Questionnarie Variables, Weights, and Numerical Coding Variable Numerical number Variable description Weight coding I Tenure of central bank CEO overlap with political authorities 0.10 Little overlap 1.0 Some overlap 0.5 Substantial overlap 0.0 2 Limitations on lending in practice 0.20 Tight 1.00 Moderately tight 0.66 Moderately loose 0.33 Loose or nonexistent 0.00 3 Resolution of conflict 0.10 Some clear cases of resolution in favor of bank 1.0 Resolution in favor of government in all cases 0.0 All other cases 0.5 4 Financial independence 0.10 a. Determination of the central bank's budget Mostly central bank 1.0 Mixture of bank and executive or legislative branches 0.5 Mostly executive or legislative branches 0.0 b. Determination of the salaries of high bank officials and the allocation of bank profits Mostly by bank or fixed by law 1.0 Mixture of bank and executive or legislative branches 0.5 Mostly executive or legislative branches 0.0 5 Intermediate policy targets 0.15 a. Quantitative monetary stock target Such targets exist; good adherence 1.00 Such targets exist; mixed adherence 0.66 Such targets exist; poor adherence 0.33 No stock targets 0.00 b. Formal or informal interest rate targets No 1 Yes 0 6 Actual priority given to price stability 0.15 First priority 1.00 First priority assigned to a fixed exchange rate 0.66 Price or exchange rate stability are among the bank's objectives, but not first priority 0.33 No mention of price or exchange rate objectives 0.00 7 Function as a development bank, granting credit at subsidy rates? 0.20 No 1.00 To some extent 0.66 Yes 0.33 The central bank heavily involved in granting subsi- dized credits 0.00 Cukierman, Webb, and Neyapti 367 actual turnover. To explore this issue, we regressed actual turnover rates in the four subperiods on the legal terms of office and on decade dummies to control for possible period-specific effects on turnover. The coefficient of the legal term- of-office variable was negative and statistically significant, but the adjusted R2 was low (0.07), thus indicating that actual turnover is affected by many other factors besides the legal term of office. Questionnaire on Central Bank Independence The other group of indicators of central bank independence is based on re- sponses to a questionnaire that was sent to a nonrandom sample of specialists on monetary policy in various central banks. Some questions involve the same issues that underlie the legal variables, but they focus on the practice rather than the law-for example, central bank objectives, their importance in practice, and the strictness of limitations on lending in practice. Some questions refer to additional issues, such as subsidized credits from the bank to the private sector, quantitative targets for the money stock, the determination of the bank budget, and the degree of actual tenure overlap between the governor and high officials in the executive branch. Although the judgments of those responding to the questionnaire are subjective and not entirely uniform, the responses help to identify divergence between actual and legal independence, particularly when the divergence is large. Answers to the questionnaire sufficed for coding most of the nine question- naire variables described in table 4 in 23 countries (The codings for each country are given in appendix table A-2). We coded only the parts of the questionnaires that could be translated into clear rankings and for which an adequate share of the questionnaires had responses. Since the questionnaire was worded in the present tense and since policymakers' thinking is dominated by the recent past, the responses are taken to refer to the 1980s. Variable 1 is designed to reflect the extent to which the terms of office of the governor and of the board of directors are likely to be independent from govern- ment. The more the turnover at the central bank coincides with turnover in the government, the less independent the bank is likely to be, and vice versa. Vari- able 2 reflects the actual limitations on lending in practice and is coded by applying criteria similar to those used to classify the legal limitations on lending described earlier. The lowest level of independence is assigned if there are no limitations on lending or if the government can adjust the limits very easily. Variable 3 reflects the extent to which conflicts between the government and the central bank are resolved in favor of the latter. Variable 4 captures two aspects of the financial independence of the bank: the determination of its budget and the setting of salaries of its top officials. It is calculated as a simple average of the variables 4a and 4b. The two parts of variable 5 reflect the relative importance of targets for the monetary stock or interest rates-these targets being precommitments agreed to by the bank and government. Money 368 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 stock targets would enhance the pursuit of price stability, because the bank could adhere to them in the face of pressure from the government. A target for the nominal interest rate, however, would typically work to limit the ability of the bank to respond to upsurges of inflation. Variable 5 is calculated as a simple average of the variables 5a and 5b. Variable 6 captures more directly the priority assigned to price stability, and variable 7 reflects the extent to which the central bank has the competing objective of providing subsidized credits to encourage development. Further details appear in chapter 19 of Cukierman (1992a). Aggregating the seven variables gives the results in table 5. The weights used in the aggregation are based on our priors and are shown in table 4; using equal weights gave an almost identical ranking. The indexes are reported only for countries in which the weights of the responses sum up to at least 0.7. Because it is based on subjective evaluations, the questionnaire-based index probably con- tains more noise than the index of legal central bank independence, but it also probably contains additional pertinent information about actual independence. The main limit of the questionnaire is the small number of countries with responses, but we hope to expand this in the future. Table 5. The Questionnaire-based Index of Central Bank Independence and Average Annual Inflation Questionnaire-based index of Average annual rate of Country central bank independence inflation, 1980-89 (percent) Germany, Fed. Rep. of 1.00 3 Costa Rica 0.81 23 Finland 0.78 7 Australia 0.76 8 Italy 0.73 11 Denmark 0.73 7 Bahamas, The 0.71 6 Luxembourg 0.66 5 France 0.65 7 United Kingdom 0.64 7 South Africa 0.64 14 Zaire 0.61 45 Lebanon 0.59 Ireland 0.57 9 Barbados 0.54 7 Uganda 0.53 72 Uruguay 0.49 45 Belgium 0.47 5 Turkey 0.44 41 Tanzania 0.38 27 Peru 0.22 108 Yugoslavia 0.17 73 Ethiopia 0.13 4 -Not available. a. Inflation is computed in logs. Cukierman, Webb, and Neyapti 369 Table 6. Rank Correlations between Indexes of Legal Central Bank Independence Country sample Correlation pair All Industrial Developing Legal index and rate of turnover 1950-89 0.000 0.018 0.011 1980-89 0.000 0.065 0.015 Legal and questionnaire-based indexesa 0.041 0.334* 0.056 Questionnaire-based index and rate of turnovera 0.068 0.050 0.031 Significant at 10 percent level. a. Covers 1980-89. The questionnaires clearly indicate that central banks in developing countries are less independent than those in industrial countries. Only two industrial countries-Ireland and Belgium-are below the median of 0.60, and only four developing countries-The Bahamas, Costa Rica, South Africa, and Zaire-are above it. This contrasts with the findings for legal independence, where the two country groups do not differ widely, but is similar to the finding for turnover. Relationships between Indexes of Independence Table 6 shows the rank correlations between indexes of central bank indepen- dence: the legal index, the turnover rate, and the questionnaire-based index. None of the various indexes of central bank independence are closely correlated. Only the correlation between the legal independence index and the questionnaire-based index of independence for the industrial countries is even marginally significant, which suggests that the law is a more important determi- nant of actual independence in the industrial countries. Since the correlation across these indexes is not high, they can be usefully combined to obtain a better measure of overall central bank independence, which is done toward the end of the article. III. INFLATION AND CENTRAL BANK INDEPENDENCE Do countries with more independent central banks have lower rates of infla- tion? The hypothesis that inflation should be negatively related to the legal and questionnaire variables has two bases. (Recall that for both variables a higher code in the range between 0 and 1 reflects a higher level of independence.) First, there is a presumption that central banks are more concerned about price stabil- ity than the political authorities (see Rogoff [1985], for instance). Because actual policy is normally the outcome of a compromise between the central bank and the executive branch, a more independent central bank will have a stronger impact on actual policy, and therefore average inflation will be lower (Alesina and Tabellini 1987; Cukierman 1992a, chap. 18). Second, the legal indepen- 370 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 dence variable is intentionally structured to reflect, among other things, the extent to which the central bank has an explicit mandate to pursue price stability at the expense of other objectives. For a given level of independence from the political authorities, a more focused legal mandate to pursue price stability is expected to result in a lower rate of inflation. The legal independence of the central bank is neither a necessary nor a suffi- cient condition for low inflation, although, other things being equal, less legal independence contributes to higher inflation. Some of the countries with the highest average rates of inflation, such as Argentina, Peru, and Nicaragua, have rankings of legal independence above the median. However, countries such as Belgium, Japan, Morocco, and Qatar, with very low rates of inflation, are ranked in the lowest quartile of legal central bank independence. To investigate systematically the relation between central bank independence and inflation, we regressed inflation on the various indexes of central bank independence. Because a higher numerical code assigned to the legal and ques- tionnaire variables indicates a higher level of independence, the hypothesis im- plies that the effect of each of these variables on inflation is negative. The effect of the turnover of central bank governors, at least above some threshold, is predicted to be positive. The inflation variable was transformed in order to reduce heteroskedasticity of the error and thus improve the efficiency of the estimate. Most countries had average inflation rates of 20 percent or less, but a few had three-digit inflation rates in some decades. Using the straight inflation rate would give undue weight to these outlier observations. So we transformed each year's inflation rate into inflation divided by one plus the inflation rate and then took the geometric average for the decade. This variable represents the annual real depreciation of a given amount of money; we call it D: (1) D = ir/(l + 7r) where ir is the inflation rate and D (hence, the transformed inflation rate) takes a value from 0 to 1.0. When inflation is 100 percent a year, D is 0.5. Table 7 presents regressions of the transformed inflation rate (D) on disaggre- gated indexes of legal central bank independence, along with the governor's turnover variable. Each observation pertains to one decade in one country. Not all countries have observations for all four decades, because some countries or central banks start after 1950. Inflation was counted only for years when the central bank existed, if it started late in the decade. The results show the impor- tance of the turnover rate for explaining variations of inflation in the whole sample and among the developing countries. None of the disaggregated legal variables has a significant coefficient (at the S percent level), and an F-test reveals that the variables as a group are not quite significant at the 10 percent level even within the group of industrial countries. To overcome the collinearity among the disaggregated legal variables, we ran the regressions with the aggregatc index of legal independence. Table 8 reports Cukierman, Webb, and Neyapti 371 Table 7. The Transformed Inflation Rate, Disaggregated Variables of Legal Central Bank Independence, and the Turnover Rate, 1950-89 All Industrial Developing Explanatory variable countries countries countries Intercept 0.09* 0.09*** 0.09* (2.47) (3.50) (1.71) CEO -0.00 0.02 0.01 (-0.10) (0.54) (0.18) Policy formulation 0.05 -0.02 0.09 (0.90) (-0.54) (1.08) Central bank -0.04 0.01 -0.08 objectives (-1.29) (0.42) (-1.61) Limitations on lending a. Advances -0.04 -0.02 -0.04 (-1.11) (-0.88) (-0.72) b. Securitized lending 0.03 -0.01 0.04 (0.73) (-0.13) (0.69) c. Terms of lending 0.06 0.01 0.08 (1.27) (0.42) (1.15) d. Potential borrowers 0.02 0.00 0.03 (1.15) (0.33) (0.90) e. Others -0.07 -0.06* -0.05 (-1.14) (-1.79) (-0.57) Rate of turnover of 0.30"** -0.07 0.30*** central bank governor (5.99) (-1.13) (4.47) D1ummy: 1950-59 -0.08*** -0.03** -0.10** (-2.92) (-2.05) (-2.14) Dummy: 1960-71 -0.09*** -0.02 -0.12*** (-4.04) (-1.48) (-3.57) Dummy: 1972-79 -0.02 0.03* * -0.03 (-1.10) (2.10) (-1.18) R2 0.29 0.29 0.27 F-statistic for 1.24 1.62 1.20 legal variablesa (0.28) (0.15) (0.30) Number of observations 177 60 117 Note: The dependent variable is the transformed inflation rate, D. The t-statistics are reported in parentheses under estimated coefficients. indicates significance at the 10 percent level, '* at the 5 percent level, and * * at the 1 percent level. a. The significance levels are in parentheses. the results. (We used the index based on our priors for weights. Using the index with equal weights produced similar results.) The key results come out when we split the sample into industrial and developing countries. For the industrial countries, the aggregate legal variable has a statistically significant coefficient with the predicted negative sign. Laws do make a differ- ence. The turnover rate, always low in any case for this subsample, has a negative sign, contrary to our prediction; the t-statistic indicates marginal statis- tical significance. The most anomalous case is Iceland, with the highest inflation rate and lowest turnover among industrial countries. Dropping Iceland from the industrial country subsample makes the estimated coefficient on the turnover rate slightly positive and totally insignificant. The coefficient on the aggregate index of legal central bank independence becomes more significant (the t-statis- 372 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 Table 8. The Transformed Inflation Rate, Aggregate Index of Legal Central Bank Independence, and the Turnover Rate, 1950-89 All countries with decom- All Industrial Developing posed turnover Explanatory variable countries countries countries variable Intercept 0.09*** 0.09*** 0.11' 0.10*** (3.55) (7.17) (2.51) (3.54) Legal central bank -0.02 -0.06** 0.01 -0.03 independence (-0.39) (-2.54) (0.11) (-0.45) (aggregate index) Rate of turnover 0.28*** -0.08* 0.28*** of central bank (6.64) (-1.81) (4.80) governor Decomposed turnovera High turnover range 0.27 ** (6.27) Low turnover range 0.20* (1.86) Dummy: 1950-59 -0.08*** -0.03*** -0.11### -0.086* (-3.31) (-2.94) (-2.62) (-3.33) Dummy: 1960-71 -0.09*** -0.02** -0.13*** -0.09*** (-4.45) (-2.11) (-4.14) (-4.47) Dummy: 1972-79 -0.02 0.03*** -0.04 -0.02 (-0.88) (2.90) (-1.28) (-0.86) PR2 0.26 0.34 0.23 0.25 Number of observations 214 79 135 214 Note: The dependent variable is the transformed inflation rate, D. The t-statistics are reported in parentheses under estimated coefficients. * indicates significance at the 10 percent level, at the 5 percent level, and * * * at the 1 percent level. a. The rate of turnover is in the high range if there are 0.25 or more turnovers per year; it is in the low range if there are fewer than 0.25. tic is -4.32), and the adjusted R2 increases to 0.61. Italy also has low turnover rates but high inflation, while Japan has high turnover compared with other industrial countries but low inflation. In Japan, the Ministry of Finance has an unusually strong anti-inflation attitude, as well as strong influence over the central bank. To explore further the composition of the legal variable, we made an index- the lending-limit index-from only the five components pertaining to limits on lending. These components drive the result for the industrial countries. The t-statistic for the lending-limit index is about the same as for the whole compo- site legal variable. The other components of the legal independence variable- CEO, policy formation, and objectives-do not make any significant contribu- tion to explaining inflation. The lending-limit index is not significant for the developing countries. For the developing countries, the turnover rate is highly significant with the predicted positive coefficient. But the aggregate legal variable remains insignifi- cant. This is not to deny that the legal charter has helped ensure the central Cukierman, Webb, and Neyapti 373 bank's independence and commitment to price stability in some developing countries. But statistical evidence for developing countries does not reveal that the central bank laws contribute to explaining the variation of inflation across periods and between countries. What is the relation between the results for the whole sample and the results for the two subsamples (industrial and developing countries)? In particular, why does the result of the importance of the legal variable for the industrial countries not show up in the whole sample, while the result of the importance of the turnover variable for the developing countries does show up for the entire sam- ple? Figure 1 illustrates these relations. The left panel in figure 1 shows the partial relation of the aggregate index of legal central bank independence with the transformed inflation rate (D). The downward-sloping line shows the significance of the aggregate index of legal central bank independence for industrial countries. The horizontal line shows that the index does not on average affect inflation (D) in developing countries. Combining the two samples masks the effect of legal central bank independence on inflation in industrial countries, because including the developing countries raises the average inflation rate (D) across the whole range of values for the aggregate index of legal central bank independence. The right panel shows the partial relation between the rate of turnover of the central bank governor and inflation (D). The upward-sloping line shows that the increasing rate of turnover leads to increased inflation in developing countries. The short and low horizontal line shows that the rate of turnover and inflation are both generally lower in the industrial countries. The overall regressions on the rate of turnover reveal a stronger effect of turnover on real depreciation of money in the entire sample than in the developing-country subsample, because Figure 1. Partial Relation of Inflation to the Legal Independence and to the Turnover Rate of Governors of Central Banks in Industrial and Developing Countries Inflation (D) Inflation (D) Developing * * * * * * Developing Legal independence Turnover rate 374 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 the overall regression also reflects differences between industrial and developing countries in the general levels of D and of turnover rates. To explore further the possibility that variations in the rate of turnover of the central bank governor at high rates of turnover might have a stronger effect on inflation than variations at low levels, the regression in the fourth column of table 8 decomposes the turnover variable into high and low turnover ranges. The cutoff of 0.25 turnovers a year or an average tenure of four years seems reasonable, in view of the typical length of electoral cycles. The coefficient on turnover for the high range is very significant, although it is significant for the low range at only the 10 percent level. This result is consistent with the view that turnover is more negatively associated with actual central bank independence in the high range for turnovers. Although the observations of the questionnaire results (and inflation) were available for a much smaller sample-only 22 countries and one period, the 1980s-than the sample for the turnover and aggregate legal central bank inde- pendence variables, we combined it into the regression reported in column 1 of table 8 and obtained the following results: (2) D = 0.27 + 0. 16 Legal-cBI + 0. 57 TOR - 0. 46 Question-cBI (0.62) (2.22);* (-2.97)- - adjusted R2 = 0.38, N = 22 where ** and * indicate significance at the 0.05 and 0.01 levels, respectively, Legal-cBI is the index of aggregate legal central bank independence, TOR is the rate of turnover of the central bank governor, and Question-CBI is the aggregate questionnaire-based index computed with the weights listed in table 4. The questionnaire variable has the predicted sign and is very significant statistically. The turnover variable also remains significant, which suggests that it reveals information about actual central bank independence that is not captured by the questionnaire. (The coefficient on the turnover variable may also reflect some simultaneity with the inflation variable, which is discussed below.) To test the robustness of our results, we tried several variants of the regres- sions in table 8. Using regular inflation rates instead of the transformed version (D) as the dependent variable yields qualitatively similar results. Using versions of the legal central bank independence and questionnaire variables that were aggregated using equal weights on the component variables did not substantially change the results. We also tried aggregating legal variables with weights derived from principal components analysis. For the industrial-country subsample, the first principal component gave results very similar to those with equal weights and with the weights we selected. The aggregations with weights from the second and third principal components did not have significant coefficients. For the overall sample and the developing-country subsample, the aggregation with weights from the first principal component was not statistically significant. The turnover rate also reflects the extent to which the government complies Cukierman, Webb, and Neyapti 375 with the law's specification of the governor's term of office. To address this issue more directly, we generated a variable (the compliance variable) equal to the ratio of the actual average term in office to the legal term of office in each country and decade. In most countries and decades, actual average terms in office (including reappointments) are shorter than the legal term. Our hypoth- esis is that the lower the actual tenure compared with the one in the law, the lower the actual independence of the central bank and the higher the inflation rate (D). When the compliance variable is entered instead of the turnover vari- able, it gives qualitatively similar results to those with the turnover variable- significant coefficients for the overall sample and the developing countries but not in the group of industrial countries. When the compliance and turnover variables are entered together on the right-hand side, however, the compliance variable loses all significance, while the coefficient on the turnover variable remains positive and significant. In other words, it seems to be turnover itself that affects the ability and will to control inflation, rather than the relation of actual to legally stipulated turnover. We also explored whether the low turnover would identify a subset of devel- oping countries where the degree of legal independence did contribute to ex- plaining inflation. An interaction term of the aggregate index of legal indepen- dence times a dummy for low turnover (0.25 or below) was not significant, however, for the whole sample or for the developing countries. Thus low turn- over does not seem to be sufficient to reveal or engender a systematic respect for the legal stipulations of independence. Previous work on inflation and central bank independence is based only on legal data and, in some cases, refers to a subset of the developed countries (Alesina 1988; Grilli, Masciandaro, and Tabellini 1991). These analyses find a significant negative relationship between some of their indexes of central bank independence and average inflation in some periods. Because the indexes of independence used here differ from those used in the previous studies, it is instructive to reexamine their samples with our indexes. We did this with the country subsamples from the earlier studies, and our variables give a better fit with the country subsamples from the previous studies than with the whole industrial-country subsample. The Grilli and others (1991) sample includes Greece and Portugal, which are in our developing-country subsample. (Further details appear in Cukierman [1992a], chap. 20.) The same degree of central bank independence may be associated with differ- ent rates of inflation when there are different economic shocks. It is also possible that the degree of independence affects the response of policy to shocks. A more dependent central bank, for example, may inflate at a higher rate in response to a slowdown in economic activity. The investigation of such interactions is be- yond the scope of this article, but it represents an intriguing avenue for future empirical research. In summary, legal independence is systematically and inversely related to 376 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 inflation in industrial, but not in developing, countries. In the latter group the actual frequency of change of the CEO of the bank is a better proxy for central bank independence. The divergence between the letter of the law and actual practice seems substantially higher in developing than in industrial countries. This may be due to a general norm of more adherence to the law in industrial countries. Two- Way Causality between Inflation and Central Bank Independence There may be two-way causality between inflation and the actual degree of central bank independence. As shown above, less independence contributes to higher inflation. However, high inflation is likely to result, at least after a while, in less independence. High inflation encourages processes that make it easier for the government to influence monetary policy even if the bank charter does not change. Most central bank laws are highly incomplete contracts that do not fully delimit the areas of responsibility of the bank and of the executive branch. In times of high inflation it is harder for the bank to closely control the money supply. In addition, high inflation is partly blamed on the bank, which tarnishes its public image and thus reduces its authority in relation to the treasury, even if this contradicts the charter. Argentina is a dramatic example. If inflation affects actual independence, we expect it also affects the rate of turnover of governors. Therefore, at least part of the positive relation found between inflation and turnover may reflect the effect of inflation on the independence of central banks and therefore on the turnover rate of their CEOS. The legal charters change infrequently and do not seem likely to be simultaneously determined with infla- tion outcomes. To examine the possibility that the results for the turnover variable in tables 7 and 8 result f;om a simultaneity bias, we reestimated the regressions using two- stage least squares and introduced an instrumental variable for the rate of turn- over of the central bank governor. The instruments used are transformed infla- tion (D) and the turnover variable in the previous decade, legal central bank independence, the legal term of office, and the decade dummies. Because periods are about a decade each, the previous observation is about 10 years away from the center of the current one-enough to be predetermined. The use of lagged values of D and turnover as instruments cuts the number of observations used in the equation by about a quarter, to 142. The coefficient of turnover remains positive, actually increases, and remains highly significant (its t-statistic is 6.98). Qualitatively similar results were also obtained with the developing-country subsample. To investigate further the causality between inflation and the turnover of central bank governors, we did a simple Granger causality test by estimating the bivariate autoregressive processes for inflation and turnover. The periods are the four approximate decades used throughout the article. The long periods seem appropriate for slow-moving processes such as the erosion or the buildup of central bank independence and its interaction with inflation. The estimated Cukierman, Webb, and Neyapti 377 processes for transformed inflation (D) and turnover (TOR) are given in equa- tions 3 and 4 with t-statistics in parentheses under the coefficients. (3) D = 0.02 + 0.79 D-1 + 0.17 TOR-1 (7.54)*-- (3.53)* (4) TOR = 0.09 + 0.54 D-1 + 0.43 TOR-1. (3.41)... (5.93)*-*' The coefficient of lagged turnover in the inflation equation is highly significant, as is the coefficient of lagged inflation in the turnover equation. A similar picture emerges when the straight inflation rate is substituted for D. The pattern also holds up for the developing-country subsample, but not for the industrial- country subsample, as the earlier results would lead one to expect. These results imply that there is a vicious circle between inflation and low levels of central bank independence. When sufficiently sustained, inflation erodes central bank independence. Then low independence contributes to higher inflation. Although central bank independence and price stability reinforce each other, the significant coefficients for the turnover variable reported in tables 7 and 8 do seem to reflect a true effect of central bank independence on inflation, and not just simultaneity. Central Bank Independence and the Variability of Inflation Variability of inflation imposes economic costs. Indeed, many of the costs of high inflation arise because it is usually more variable and uncertain when the average is high (Cukierman 1984, chap. 3-6). Furthermore, theory implies that inflation will be more variable when the central bank is less independent (Cukierman 1992a, chap. 18). Thus it is important to investigate whether our proxies for central bank independence actually affect the variability of inflation. As a measure of inflation variability, we take the standard deviation of D. The proxies for central bank independence explain inflation variability to about the same extent that they explain inflation levels (table 9). Turnover contributes significantly to explaining the variance of D in the overall sample and in the developing countries, but the variable for legal independence does not. For the industrial-country sample, the legal variable is significant at the 10 percent level. Again, with Iceland removed from the industrial subsample, the coefficient on turnover is much smaller and more insignificant, and the coeffi- cient for legal central bank independence is significant at the S percent level. The conceptual framework in chapter 18 of Cukierman (1992a) implies that the mean level of inflation and its standard deviation are both negatively related to the degree of independence of central banks across countries. As a conse- quence the mean level of inflation and its standard deviation are positively related to each other. There is empirical evidence that the variation in the inde- pendence of central banks explains some of the correlation between the average level of inflation rates and their variance (Cukierman 1992a, chap. 22). 378 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 9. The Standard Deviation of Transformed Inflation, Aggregate Index of Legal Central Bank Independence, and the Turnover Rate, 1950-89 All countries with All Industrial Developing decomposed Explanatory variable countries countries countries turnover variable Intercept 0.04* ** 0.04#** 0.05*** 0.04*** (4.16) (7.42) (2.88) (3.96) Legal central bank -0.00 -0.02* 0.02 -0.01 independence (aggregate (-0.20) (-1.77) (0.40) (-0.23) index) Rate of turnover of central 0.10* -0.02 0.09*** bank governor (6.23) (-1.28) (4.05) Decomposed turnovera High turnover range 0.10* (5.92) Low turnover range 0.08 * (1.95) Dummy: 1950-59 -0.01 0.00 -0.03 -0.01 (-1.46) (0.80) (-1.52) (-1.48) Dummy: 1960-71 -0.03*** -0.01**. -0.04*** -o.03*** (-3.87) (-3.14) (-3.22) (-3.88) Dummy: 1972-79 -0.01 -0.00 -0.01 -0.01 (-1.29) (-0.77) (-1.13) (-1.27) R2 0.19 0.18 0.15 0.19 Numberof observations 215 79 136 215 Note: The dependent variable is the standard deviation of transformed inflation (D). The t-statistics are reported in parentheses under estimated coefficients. indicates significance at the 10 percent level, at the 5 percent level, and t '* at the 1 percent level. a. The rate of turnover is in the high range if there are 0.25 or more turnovers per year; it is in the low range if there are fewer than 0.25. Central Bank Independence and the Growth of Credit to the Government Providing credit to the government would seem to be the most important channel through which the lack of central bank independence leads to inflation because the issue of how to finance its budget deficit is immediately relevant to the government. Providing credit to private and publicly owned business also contributes to inflation, however, as do problems with managing exchange rates. Political authorities are very concerned with these issues as they affect unemployment, bankruptcy rates, export incentives, and the domestic currency cost of imports. Although a systematic evaluation of these considerations lies beyond the scope of this article, we can at least examine the relation between central bank independence and its extension of credit to the government. Regressing the rate of growth of credit from the bank to the public sector on the two main indicators of central bank independence-the legal independence of the central bank and turnover variables-reveals a pattern similar to, but generally less strong than, what was found with the regressions for the trans- formed inflation rate (table 10). The aggregate index of legal variables was not significant at all, even in the industrial-country subsample. Turnover of the central bank governor contributes significantly to explaining credit growth to the public sector, although variation of turnover at low rates does not matter much. Within the subsamples of country groups, the coefficients are not statis- Cukierman, Webb, and Neyapti 379 tically significant. Yet they are consistent with the following two hypotheses. The first is that among industrial countries more legal independence limits credit expansion to the public sector, whereas the turnover rate is too low to matter much. The second is that among developing countries higher turnover reflects lower independence, which contributes to faster credit expansion, but the law does not matter much. That the results for the growth of central bank credit are weaker than the results for inflation suggests that issues other than deficit financing are more important than we had originally thought. An Overall Index of Inflation-Based Central Bank Independence Legal independence and turnover capture different dimensions of central bank independence, and each seems to be important for a different subset of coun- tries. This section combines the indicators with a weighting scheme in order to obtain an overall measure of central bank independence. Such weighting is perforce arbitrary, but we reduce the arbitrariness by setting the weights equal to the coefficients from the regressions in which they are used to explain the varia- tion in the transformed inflation variable (D). In this sense, the resulting index of overall independence is based on inflation. Different regressions for the industrial and developing countries generate the measures of overall central bank independence for members of each group. For Table 10. The Rate of Growth of Central Bank Credit to the Public Sector, 1950-89 All countries with All Industrial Developing decomposed Explanatory variable countries countries countries turnover variable Intercept 0.22*** 0.14<* 0.27"* 0.22*** (3.06) (2.15) (2.34) (2.71) Legal central bank -0.05 -0.13 0.20 -0.05 independence (aggregate (-0.33) (-1.21) (0.67) (-0.33) index) Rate of turnover of central 0.46 ** -0.06 0.27 bank governor (3.48) (-0.27) (1.52) Decomposed turnovera High turnover range 0.46 ** (3.38) Low turnover range 0.45 (1.46) Dummy: 1950-59 -0.19@** -0.09* -0.20* -0.19*** (-2.80) (-1.67) (-1.79) (-2.79) Dummy: 1960-71 -0.15*** -0.01 -0.22*** -0.15*** (-2.73) (-0.28) (-2.66) (-2.71) Dummy: 1972-79 -0.01 0.10** -0.06 -0.01 (-0.16) (2.12) (-0.83) (-0.16) R2 0.13 0.14 0.06 0.12 Number of observations 175 68 107 175 Note: The dependent variable is the rate of growth of central bank credit to the public sector. The t-statistics are reported in parentheses under estimated coefficients. * indicates significance at the 10 percent level, * * at the 5 percent level, and 3 ' * at the 1 percent level. a. The rate of turnover is in the high range if there are 0.25 or more turnovers per year; it is in the low range if there are fewer than 0.2S. 380 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 the industrial countries, the equation has only the aggregate index of legal independence, because the coefficient on turnover in tables 7 and 8 had a sign contrary to the prediction and because virtually all the observations from the industrial countries had turnover rates below the relevant threshold. For devel- oping countries, the equation has both turnover and the index of legal indepen- dence. The two equations have the same left-side variable, D, and so the pre- dicted values of D offer an inflation-based indicator or index of central bank independence that is comparable across the two subsamples. We estimated the equations across all periods. Table 11 shows the countries ranked by their Table 11. Ranking of Central Banks by an Overall Index of Independence during the 1 980s Rate of Transformed inflation Legal central bank turnover of rate (D) independence central bank Economy Fitted Actual (aggregate index) governor Germany, Fed. Rep. of 0.04 0.02 0.69 n.a. Switzerland 0.05 0.03 0.64 n.a. Austria 0.05 0.03 0.61 n.a. Denmark 0.05 0.05 0.50 n.a. United States 0.05 0.04 0.48 n. a. Canada 0.06 0.05 0.45 n.a. Ireland 0.06 0.07 0.44 n.a. Netherlands 0.06 0.02 0.42 n.a. Australia 0.06 0.07 0.36 n.a. Iceland 0.06 0.24 0.34 n.a. Luxembourg 0.06 0.04 0.33 n.a. Sweden 0.06 0.06 0.29 n.a. Finland 0.07 0.06 0.28 n.a. United Kingdom 0.07 0.05 0.27 n.a. Italy 0.07 0.08 0.25 n.a. New Zealand 0.07 0.09 0.24 n.a. France 0.07 0.06 0.24 n.a. Spain 0.07 0.08 0.23 n.a. Japan 0.07 0.02 0.18 n.a. Norway 0.07 0.07 0.17 n.a. Belgium 0.07 0.04 0.17 n.a. Qatar 0.11 0.03 0.20 0.00 Nepal 0.14 0.08 0.18 0.10 Zimbabwe 0.14 0.11 0.20 0.10 Hungary 0.14 0.07 0.24 0.10 Thailand 0.14 0.04 0.27 0.10 Nigeria 0.14 0.16 0.37 0.10 Barbados 0.14 0.05 0.38 0.10 Lebanon 0.14 - 0.40 0.10 Ethiopia 0.14 0.04 0.40 0.10 Honduras 0.14 0.05 0.43 0.10 Tanzania 0.14 0.21 0.44 0.10 Morocco 0.16 0.06 0.14 0.20 Yugoslavia 0.17 0.51 0.17 0.20 Taiwan 0.17 0.03 0.21 0.20 Cukierman, Webb, and Neyapti 381 indexes of central bank independence during the 1980s. The actual values of D in the 1980s are presented too, for comparison, along with the legal indepen- dence variable for all countries and the turnover variable for the developing countries. The overall index usually classifies extreme cases in the expected ranges. Thus Germany and the United States are classified near the top and Argentina, Brazil, and Venezuela toward the bottom. For 24 out of the 71 countries (with inflation data) in table 11, the distance between the actual and the predicted transformed inflation rate (D) is less than or equal to 0.03. Table 11. (continued) Transformed inflation Legal central bank Rate of rate (D) independence central bank Economy Fitted Actual (aggregate index) governor Panama 0.17 0.02 0.22 0.20 South Africa 0.17 0.12 0.25 0.20 Colombia 0.17 0.17 0.27 0.20 Indonesia 0.17 0.07 0.27 0.20 Romania 0.17 - 0.30 0.20 Ghana 0.17 0.28 0.31 0.20 Malaysia 0.17 0.03 0.36 0.20 Uganda 0.17 0.47 0.38 0.20 Israel 0.17 0.47 0.39 0.20 Bahamas, The 0.17 0.05 0.41 0.20 Zaire 0.17 0.34 0.43 0.20 Philippines 0.17 0.11 0.43 0.20 Kenya 0.17 0.09 0.44 0.20 Malta 0.17 0.02 0.44 0.20 Greece 0.17 0.14 0.55 0.20 Pakistan 0.19 0.06 0.21 0.30 Uruguay 0.19 0.33 0.24 0.30 China 0.19 0.07 0.29 0.30 India 0.19 0.07 0.34 0.30 Mexico 0.19 0.38 0.34 0.30 Portugal 0.20 0.14 0.41 0.30 Peru 0.20 0.64 0.43 0.30 Egypt 0.20 0.13 0.49 0.30 Botswana 0.22 0.09 0.33 0.40 Nicaragua 0.22 0.67 0.45 0.40 Turkey 0.22 0.28 0.46 0.40 Costa Rica 0.22 0.19 0.47 0.40 Poland 0.25 0.29 0.10 0.50 Korea, Republic of 0.25 0.05 0.27 0.50 Zambia 0.25 0.25 0.33 0.50 Venezuela 0.25 0.16 0.43 0.50 Western Samoa 0.27 0.05 0.30 0.56 Singapore 0.28 0.02 0.29 0.60 Brazil 0.33 0.68 0.21 0.80 Chile 0.33 0.15 0.46 0.80 Argentina 0.39 0.74 0.40 1.00 - Not available. n.a. Not applicable. 382 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 The coefficient of rank correlation between the actual and predicted real depreciation of money is 0.25 for the countries in table 11, which indicates that the combination of variables for turnover and legal central bank independence can predict a reasonable amount of the cross-country variation in inflation in the 1980s. Using only the legal variable (and country group dummies), the coeffi- cient of rank correlation is merely 0.19. The inclusion of turnover in the con- struction of an index of central bank independence improves the ability of this index to predict cross-country variations in inflation. Argentina and Brazil would have been ranked as 55th and 29th from the top in legal independence alone, instead of 71st and 69th in the overall index of independence. The large change of rankings reflects the effect of the turnover variable, which is very high for Argentina and Brazil. In most other cases, where including the turnover variable substantially improved the prediction of a country's inflation, low turn- over brought the prediction down closer to the actual. All the industrial countries are above the median of overall independence (D predicted 0.17), and most of the developing countries are below it. Among the countries with overall independence above the median-D estimated below the median-only two (Iceland and Tanzania) have actual D values above 0.17. This indicates that a reasonably high overall independence is highly likely to prevent high inflation. All of the countries with inflation above 50 percent (D 0.33) have less than median central bank independence. For 11 countries, how- ever, the actual D is less than 0.17, even though they have lower-than-median overall central bank independence (estimated D above 0.17). In other words, lower-than-median levels of overall central bank independence do not neces- sarily lead to high inflation. These findings are consistent with the view that below-median independence by itself does not necessarily result in high inflation. When there are adverse shocks, however, countries with independence levels within the low range are more likely to develop high and even exceptionally high rates of inflation. Austria, The Bahamas, Belgium, Luxembourg, Netherlands, and Panama have lower inflation in the 1980s than their central bank independence would indicate, because their monetary policy is dominated by a policy rule fixing their exchange rate to a relatively stable currency. Korea and Japan have lower infla- tion than their indicators of central bank independence predict, probably be- cause the governments, to which their central banks are subservient, have their own commitment to price stability. These examples demonstrate that high cen- tral bank independence is not necessary for price stability. IV. CONCLUSIONS The notion of central bank independence underlying this study is not uncondi- tional independence from government, but rather the independence to pursue the objective of price stability, even at the cost of other objectives that may be more important to the political authorities. Thus, our measures of independence Cukierman, Webb, and Neyapti 383 include indexes of institutional independence such as appointment procedures, as well as measures of the relative importance attached to price stability in the central bank law and in practice. Unavoidably, there were subjective or arbitrary decisions in coding, classify- ing, and weighing legal information and responses to questionnaires. Results of sensitivity analysis offer some reassurance that the main results are robust, but questions about various details may remain. This study aims to contribute to the systematization of future work by committing to a systematic and documented way of characterizing central bank independence. The study produces four different rankings of independence of central banks. The first is by legal independence, and the second is by governors' turnover rates. The third ranking utilizes responses of specialists to a questionnaire on central bank independence. The fourth ranking is based on an aggregation of the first two. Legal independence is an important and statistically significant determinant of price stability among industrial countries, but not among developing countries. Within the latter group, some countries have elaborately locked cookie boxes, to borrow a metaphor from the beginning of the article, but some of them either break or undo the lock when they are hungry. An important step in creating an independent central bank, therefore, must involve establishing respect for the central bank charter and management, even when they are not ideal. The rate of turnover of the governors contributes significantly to explaining inflation, and it is even more important in explaining variations in inflation across the overall sample of countries. An inflation-based index of overall central bank independence-combining legal and turnover information-contributes signifi- cantly to explaining cross-country variations in the rate of inflation. These results seem robust to possible biases due to reverse causality. The results imply that the discrepancies between actual and legal independence are wider on aver- age in developing countries than in the industrial countries. The turnover rate was not significant in explaining variations of inflation within the industrial group. Preliminary evidence from the 1980s suggests that when questionnaire vari- ables are used to explain variations in inflation, there is some additional infor- mation in the rate of governors' turnover but not in the legal variables. There seems to be a vicious circle between inflation and the lack of central bank independence, which deserves fuller investigation. Preliminary results here indicate a two-way (positive in both directions) causality between inflation and turnover of central bank governors, a proxy for lack of independence. Lower independence induces higher future inflation, which, in turn, reduces the subse- quent actual level of central bank independence, and so on. Success in control- ling inflation, however, seems to enhance the independence of central banks. Finally, central bank independence is only one of several institutional devices for ensuring price stability. Some of the structural and political determinants of central bank independence are discussed in Cukierman (1992a, chap. 23, and 1992b). Table A-1. Disaggregated and Aggregated Legal Central Bank Independence Variables, by Country and Decade Disaggregated legal variables Policy formulation variables CetaAgrae CEO variables Who bCentral Limitations on lending variables Aggregate Term of Who Dis- Other formu- Final Role in objectives Securitized Terms of Potential Type Maturity Interest Pri,nary central bank Economy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates nmarket independence and decade (la) (Ib) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4h) variable Argentina 1972-79 0.25 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.33 0.33 0.33 1.00 0.33 1.00 0.50 0.00 0.40 1980-89 0.25 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.33 0.33 0.33 1.00 0.33 1.00 0.50 0.00 0.40 Australia 1960-71 0.75 0.00 0.8.3 1.00 0.33 0.20 0.00 0.40 0.33 0.00 0.33 0.00 - 0.67 1.00 0.00 0.36 1972-79 0.75 0.00 0.83 1.00 0.33 0.20 0.00 0.40 0.33 0.00 0.33 0.00 - 0.67 1.00 0.00 0.36 1980-89 0.75 0.00 0.83 1.00 0.33 0.20 0.00 0.40 0.33 0.00 0.33 0.00 - 0.67 1.00 0.00 0.36 Austria 1950-59 0.50 0.00 0.83 1.00 0.67 0.60 0.00 0.60 1.00 0.67 0.33 1.00 1.00 1.00 1.00 0.00 0.65 1960-71 0.50 0.00 0.83 1.00 0.67 0.60 0.00 0.60 1.00 0.67 0.33 1.00 1.00 1.00 1.00 0.00 0.65 1972-79 0.50 0.00 0.83 1.00 1.00 0.60 0.00 0.60 1.00 0.67 0.33 0.33 0.33 1.00 1.00 0.00 0.61 1980-89 0.50 0.00 0.83 1.00 1.00 0.60 0.00 0.60 1.00 0.67 0.33 0.33 0.33 1.00 1.00 0.00 0.61 Bahamas, The 1972-79 0.50 0.00 0.83 0.50 0.33 - 0.00 0.60 0.33 0.33 0.33 1.00 0.33 1.00 0.25 0.00 0.41 1980-89 0.50 0.00 0.83 0.50 0.33 - 0.00 0.60 0.33 0.33 0.33 1.00 0.33 1.00 0.25 0.00 0.41 Barbados 1972-79 0.50 0.00 0.83 0.00 0.33 0.20 0.00 0.80 0.33 0.33 0.33 0.67 - 0.33 0.25 0.00 0.38 1980-89 0.50 0.00 0.83 0.00 0.33 0.20 0.00 0.80 0.33 0.33 0.33 0.67 - 0.33 0.25 0.00 0.38 Belgium 1950-59 0.50 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.33 0.33 - 1.00 0.50 0,00 0.15 1960-71 0.50 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.00 0.00 0.33 0.33 - 1.00 0.50 0.00 0.15 1972-79 0.50 0.00 0.00 0.50 0.00 0.20 0.00 0.00 0.00 0.00 0.33 0.33 - 1.00 0.50 0.00 0.17 1980-89 0.50 0.00 0.00 0.50 0.00 0.20 0.00 0.00 0.00 0.00 0.33 0.33 - 1.00 0.50 0.00 0.17 Bolivia 1950-59 0.00 0.00 0.83 1.00 0.67 0.20 0.00 0.60 0.33 0.00 0.00 0.00 0.00 1.00 0.25 0.00 0.30 1960-71 0.00 0.00 0.83 1.00 0.67 0.20 0.00 0.60 0.33 0.00 0.00 0.00 0.00 1.00 0.25 0.00 0.30 1972-79 0.00 0.00 0.83 1.00 0.67 0.20 0.00 0.60 0.33 0.00 0.00 0.00 0.00 1.00 0.25 0.00 0.30 1980-89 0.00 0.00 0.83 1.00 0.67 0.20 0.00 0.60 0.33 0.00 0.00 0.00 0.00 1.00 0.25 0.00 0.30 Botswana 1972-79 O.S0 0.00 0.83 O.S0 - 0.00 0.00 0.20 0.33 0.67 0.33 0.67 0.33 0.33 0.25 0.00 0.33 1980-89 O.S0 0.00 0.83 O.S0 - 0.00 0.00 0.20 0.33 0.67 0.33 0.67 0.33 0.33 0.25 0.00 0.33 Brazil 1960-71 0.00 O.S0 0.00 0.00 0.33 - 0.00 0.00 0.67 0.00 0.00 1.00 - 0.00 0.25 0.00 0.21 1972-79 0.00 0.50 0.00 0.00 0.33 - 0.00 0.00 0.67 0.00 0.00 1.00 - 0.00 0.25 0.00 0.21 1980-89 0.00 0.50 0.00 0.00 0.33 - 0.00 0.00 0.67 0.00 0.00 1.00 - 0.00 0.25 0.00 0.21 Canada 1950-59 0.75 0.75 0.83 1.00 0.33 0.20 0.00 0.20 0.33 0.67 0.44 0.67 0.33 0.67 0.75 0.00 0.45 1960-71 0.75 0.75 0.83 1.00 0.33 0.20 0.00 0.20 0.33 0.33 0.67 0.67 0.33 0.67 0.75 0.00 0.45 1972-79 0.75 0.75 0.83 1.00 0.33 0.20 0.00 0.20 0.33 0.33 0.67 0.67 0.33 0.67 0.75 0.00 0.45 1980-89 0.75 0.75 0.83 1.00 0.33 0.20 0.00 0.20 0.33 0.33 0.67 0.67 0.33 0.67 0.75 0.00 0.45 Chile 1950-59 0.00 1.00 1.00 0.50 0.67 0.00 0.00 0.20 0.00 0.00 0.67 0.00 - 0.00 0.25 0.00 0.26 1960-71 0.00 1.00 1.00 0.50 0.67 0.00 0.00 0.20 0.00 0.00 0.67 0.00 - 0.00 0.25 0.00 0.26 1972-79 0.50 0.00 0.83 0.50 0.67 0.20 0.00 0.80 0.33 0.33 0.67 1.00 - 0.00 0.25 0.00 0.46 1980-89 0.50 0.00 0.83 0.50 0.67 0.20 0.00 0.80 0.33 0.33 0.67 1.00 - 0.00 0.25 0.00 0.46 China 1950-59 0.50 - - 0.00 - 0.80 0.00 0.20 - 0.33 0.00 - - 0.00 0.25 1.00 0.29 1960-71 O.S0 - - 0.00 - 0.80 0.00 0.20 - 0.33 0.00 - - 0.00 0.25 1.00 0.29 1972-79 0.50 - - 0.00 - 0.80 0.00 0.20 - 0.33 0.00 - - 0.00 0.25 1.00 0.29 1980-89 0.50 - - 0.00 - 0.80 0.00 0.20 - 0.33 0.00 - - 0.00 0.25 1.00 0.29 Colombia 1960-71 0.00 0.75 0.83 0.00 0.00 0.20 0.00 0.00 0.67 0.00 0.33 0.00 0.67 0.67 0.25 0.00 0.27 1972-79 0.00 0.75 0.83 0.00 0.00 0.20 0.00 0.00 0.67 0.00 0.33 0.00 0.67 0.67 0.25 0.00 0.27 1980-89 0.00 0.75 0.83 0.00 0.00 0.20 0.00 0.00 0.67 0.00 0.33 0.00 0.67 0.67 0.25 0.00 0.27 Costa Rica 1960-71 - 1.00 0.67 1.00 - - 0.00 0.60 0.67 0.33 0.33 0.33 0.00 0.67 0.25 0.00 0.47 1972-79 - 1.00 0.67 1.00 - - 0.00 0.60 0.67 0.33 0.33 0.33 0.00 0.67 0.25 0.00 0.47 1980-89 - 1.00 0.67 1.00 - - 0.00 0.60 0.67 0.33 0.33 0.33 0.00 0.67 0.25 0.00 0.47 Denmark 1950-59 0.00 0.00 0.33 0.00 - 1.00 0.00 0.60 1.00 0.33 0.67 0.00 - 1.00 0.25 0.00 0.50 1960-71 0.00 0.00 0.33 0.00 - 1.00 0.00 0.60 1.00 0.33 0.67 0.00 - 1.00 0.25 0.00 0.50 1972-79 0.00 0.00 0.33 0.00 - 1.00 0.00 0.60 1.00 0.33 0.67 0.00 - 1.00 0.25 0.00 0.50 1980-89 0.00 0.00 0.33 0.00 - 1.00 0.00 0.60 1.00 0.33 0.67 0.00 - 1.00 0.25 0.00 0.50 (Table continues on the following page.) Table A-1. (continued) Disaggregated legal variables Policyformulation CEO variables vaiables Central Limitations on lending variables le.gal Term of Who Dis- Other forniu- Final Role in objectives Securitized 7erms of Potential Type Maturity Interest Primary central bank Economy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates market independence and decade (la) (lb) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4h) variable Egypt 1950-59 0.50 0.50 1.00 1.00 0.33 0.00 0.00 0,60 0.67 0.67 0.33 1.00 0.33 0.00 0.25 0.00 0.52 1960-71 0.50 0.50 1.00 1.00 0.33 0.00 0.00 0.60 0.67 0.67 0.33 1.00 0.33 0.00 0.25 0.00 0.52 1972-79 0.50 0.50 1.00 0.00 0.33 0.00 0.00 0.60 0.67 0.67 0.33 1.00 0.67 0.67 0.25 0.00 0.49 1980-89 0.50 O.S0 1.00 0.00 0.33 0.00 0.00 0.60 0.67 0.67 0.33 1.00 0.67 0.67 0.25 0.00 0.49 Ethiopia 1960-71 0.50 0.25 0.00 0.00 - - - 0.60 - - - - 0.00 0.25 0.00 - 1972-79 0.00 0.25 0.00 0.00 - 1.00 1.00 0.00 0.33 0.33 0.67 1.00 0.33 0.67 0.50 0.00 0.40 1980-89 0.00 0.25 0.00 0.00 - 1.00 1.00 0.00 0.33 0.33 0.67 1.00 0.33 0.67 0.50 0.00 0.40 Finland 1950-59 1.00 0.00 - 0.00 - 0.00 0.00 0.80 0.00 0.00 0.67 - - 0.00 0.25 0.00 0.28 1960-71 1.00 0.00 - 0.00 - 0.00 0.00 0.80 0.00 0.00 0.67 - - 0.00 0.25 0.00 0.28 1972-79 1.00 0.00 - 0.00 - 0.00 0.00 0.80 0.00 0.00 0.67 - - 0.00 0.25 0.00 0.28 1980-89 1.00 0.00 - 0.00 - 0.00 0.00 0.80 0.00 0.00 0.67 - - 0.00 0.25 0.00 0.28 France 1950-59 0.00 0.25 1.00 0.50 0.33 0.60 0.00 0.20 0.33 0.00 0.33 0.00 1.00 0.00 0.00 0.00 0.28 1960-71 0.00 0.25 1.00 0.50 0.33 0.60 0.00 0.20 0.33 0.00 0.33 1.00 1.00 1.00 0.00 0.00 0.36 1972-79 0.00 0.25 1.00 0.50 0.67 0.60 0.00 0.00 0.00 0.00 0.33 1.00 - - - 0.00 0.24 1980-89 0.00 0.25 1.00 0.50 0.67 0.60 0.00 0.00 0.00 0.00 0.33 1.00 - - - 0.00 0.24 Germany, Fed. Rep. of 1950-59 1.00 0.75 1.00 0.00 0.67 1.00 0.00 1.00 0.67 0.67 0.67 0.33 1.00 1.00 0.25 0.00 0.69 1960-71 1.00 0.75 1.00 0.00 0.67 1.00 0.00 1.00 0.67 0.67 0.67 0.33 1.00 1.00 0.25 0.00 0.69 1972-79 1.00 0.75 1.00 0.00 0.67 1.00 0.00 1.00 0.67 0.67 0.67 0.33 1.00 1.00 0.25 0.00 0.69 1980-89 1.00 0.75 1.00 0.00 0.67 1.00 0.00 1.00 0.67 0.67 0.67 0.33 1.00 1.00 0.25 0.00 0.69 Ghana 1950-59 0.50 0.00 0.83 0.50 0.33 0.00 0.00 0.60 0.67 0.67 0.33 1.00 0.33 0.67 0.75 0.00 0.49 1960-71 0.50 0.00 0.83 0.50 0.33 0.00 0.00 0.60 0.00 0.33 0.33 0.00 0.33 1.00 0.50 0.00 0.31 1972-79 0.50 0.00 0.83 0.50 0.33 0.00 0.00 0.60 0.00 0.33 0.33 0.00 0.33 1,00 0.50 0.00 0.31 1980-89 0.50 0.00 0.83 0.50 0.33 0.00 0.00 0.60 0.00 0.33 0.33 0.00 0.33 1.00 0.50 0.00 0.31 Greece 1950-59 0.25 0.75 0.67 0.50 0.33 0.60 0.00 0.80 0.67 0.67 0.33 0.33 1.00 0.67 0.75 0.00 0.56 1960-71 0.25 0.75 0.67 0.50 0.33 0.60 0.00 0.80 0.33 0.67 0.33 0.33 1.00 0.67 0.75 0.00 0.51 1972-79 0.25 0.75 0.67 0.50 0.33 0.60 0.00 0.80 0.67 0.67 0.33 0.00 1.00 1,00 0.75 0.00 0.55 1980-89 0.25 0.75 0.67 0.50 0.33 0.60 0.00 0.80 0.67 0.67 0.33 0.00 1.00 1.00 0.75 0.00 0,55 Honduras 1950-S9 0.75 0.00 0.83 1.00 0.33 1.00 0.00 0.00 - 0.33 0.67 0.33 0.33 0.67 0.25 0.00 0.43 1960-71 0.75 0.00 0.83 1.00 0.33 1.00 0.00 0.00 - 0.33 0.67 0.33 0.33 0.67 0.25 0.00 0.43 1972-79 0.75 0.00 0.83 1.00 0.33 1.00 0.00 0.00 - 0.33 0.67 0.33 0.33 0.67 0.25 0.00 0.43 1980-89 0.75 0.00 0.83 1.00 0.33 1.00 0.00 0.00 - 0.33 0.67 0.33 0.33 0.67 0.25 0.00 0.43 Hungary 1950-59 0.50 0.25 1.00 0.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 1.00 0.67 1.00 0.00 0.24 1960-71 0.50 0.25 1.00 0.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 1.00 0.67 1.00 0.00 0.24 1972-79 0.50 0.25 1.00 0.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 1.00 0.67 1.00 0.00 0.24 1980-89 0.50 0.25 1.00 0.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 1.00 0.67 1.00 0.00 0.24 Iceland 1950-59 1.00 - - - - - - - - - - - - _ _ _ 1960-71 1.00 0.75 0.83 0.50 0.33 0.20 0.00 0.40 0.00 0.00 0.33 1.00 - 0.33 0.25 0.00 0.34 1972-79 1.00 0.75 0.83 0.50 0.33 0.20 0.00 0.40 0.00 0.00 0.33 1.00 - 0.33 0.25 0.00 0.34 1980-89 1.00 0.75 0.83 0.50 0.33 0.20 0.00 0.40 0.00 0.00 0.33 1.00 - 0.33 0.25 0.00 0.34 India 1950-59 0.50 0.25 0.83 0.50 - 0.00 0.00 0.40 0.00 0.00 0.33 0.33 1.00 0.33 0.25 0.00 0.25 1960-71 0.50 0.25 (1.83 0.50 - 0.00 0.00 0.40 0.33 0.00 0.67 0.33 1.00 0.33 0.25 0.00 0.34 1972-79 0.50 0.25 0.83 0.50 - 0.00 0.00 0.40 0.33 0.00 0.67 0.33 1.00 0.33 0.25 0,00 0.34 1980-89 0.50 0.25 0.83 0.50 - 0.00 0.00 0.40 0.33 0.00 0.67 0.33 1.00 0.33 0.25 0.00 0.34 Iiidonesia 1950-59 0.50 0.25 0.00 0.50 0.67 0.00 0.00 0.00 0.33 0.33 0.00 1.00 0.33 0.00 0.25 0.00 0.24 1960-71 0.50 0.25 0.00 0.50 0.67 0.00 0.00 0.40 0.33 0.33 0.00 1.00 0.33 0.00 0.25 0.00 0.30 1972-79 0.50 0.25 0.00 0.50 0.67 0.20 0.00 0.40 0.00 0.00 0.33 1.00 0.33 0,33 0.25 0.00 0.27 1980-89 0.50 0.25 0.00 0.50 0.67 0.20 0.00 0.40 0.00 0.00 0.33 1.00 0.33 0.33 0.25 0.00 0.27 (Table continues on thefollowing page.) Table A-1. (continued) Disaggregated legal variables Policyformulation variables Central Aggregate CFO variables Who bank l.imitations on lending variables legal Term of Who Dis- Other formu- Final Role in ob)ectives Securitized Terms of Potential Type Maturity Interest Primary central bank Fconomy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates market independence and decade (la) (lb) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4b) variable Ireland 1950-59 0.75 0.50 0.83 1.00 - - 0.00 0.80 - 0.00 0.33 0.33 - 0.67 0.75 0.00 0.44 1960-71 0.75 0.50 0.83 1.00 - - 0.00 0.80 - 0.00 0.33 0.33 - 0.67 0.75 0.00 0.44 1972-79 0.75 0.50 0.83 1.00 - - 0.(0 0.80 - 0.00 0.33 0.33 - 0.67 0.75 0.00 0.44 1980-89 0.75 0.50 0.83 1.00 - - 0.00 0.80 - 0.00 0.33 0.33 - 0.67 0.75 0.00 0.44 Israel 1950-59 0.50 0.50 0.50 0.50 0.67 0.20 0.00 0.40 0.33 0.00 0.67 1.00 0.00 0.67 0.25 0.00 0.39 1960-71 0.50 0.50 0.50 0.50 0.67 0.20 0.00 0.40 0.33 0.00 0.67 1.00 0.00 0.67 0.25 0.00 0.39 °° 1972-79 0.50 0.50 0.50 0.50 0.67 0.20 0.00 0.40 0.33 0.00 0.67 1.00 0.00 0.67 0.25 0.00 0.39 1980-89 0.50 0.50 0.50 0.50 0.67 0.2(0 0.00 0.40 0.33 0.00 0.67 1.00 0.00 0.67 0.25 0.00 0.39 Italy 1950-59 0.00 0.75 0.67 1.00 - - 0.00 0.20 0.33 0.00 0.33 - 0.00 0.00 0.25 0.00 0.25 1960-71 0.00 0.75 0.67 1.00 - - 0.00 0.20 0.33 0.00 0.33 - 0.00 0.00 0.25 0.00 0.25 1972-79 0.00 0.75 0.67 1.00 - - 0.00 0.20 0.33 0.00 0.33 - 0.00 0.00 0.25 0.00 0.25 1980-89 0.00 0.75 0.67 1.00 - - 0.00 0.20 0.33 0.00 0.33 - 0.00 0.00 0.25 0.00 0.25 Japan 1950-59 0.50 0.25 0.83 0.50 0.67 0.00 0.00 0.00 0.00 0.00 0.33 - - 0.00 0.25 0.00 0.18 1960-71 0.50 0.25 0.83 0.50 0.67 0.00 0.00 0.00 0.00 0.00 0.33 - - 0.00 0.25 0.00 0.18 1972-79 0.50 0.25 0.83 0.50 0.67 0.00 0.00 0.00 0.00 0.00 0.33 - - 0.00 0.25 0.00 0.18 1980-89 0.50 0.25 0.83 0.50 0.67 0.00 0.00 0.00 0.00 0.00 0.33 - - 0.00 0.25 0.00 0.18 Kenya 1960-71 0.25 0.00 0.83 0.50 - 0.20 0.00 0.40 0.67 0.67 0.33 0.33 1.00 0.67 0.75 0.00 0.44 1972-79 0.25 0.00 0.83 0.50 - 0.20 0.00 0.40 0.67 0.67 0.33 0.33 1.00 0.67 0.75 o.00 0.44 1980-89 0.25 0.00 0.83 0.50 - 0.20 0.00 0.40 0.67 0.67 0.33 0.33 1.00 0.67 0.75 0.00 0.44 Korea, Republic of 1950-59 0.25 0.25 0.83 0.00 0.33 - 0.00 0.60 0.33 0.00 0.33 0.00 - 0.67 0.25 0.00 0.30 1960-71 0.25 0.25 0.83 0.00 0.33 - 0,00 0.60 0.33 0.00 0.33 0.00 - 0.67 0.25 0.00 0.30 1972-79 0.25 0.25 0.83 0.50 0.33 - 0.00 0.60 0.33 0.00 0.33 0.00 - 0.67 0.25 0.00 0.32 1980-89 0.25 0.25 0.83 0.50 0.33 - 0.00 0.60 0.00 0.00 0.33 0.00 - 0.67 0.25 0.00 0.27 Lebanon 1950-59 0.75 0.25 0.83 1.00 0.67 - 0.00 0.00 0.67 0.00 0.33 0.33 0.33 1.00 1.00 0.00 0.40 1960-71 0.75 0.25 0.83 1.00 0.67 - 0.00 0.00 0.67 0.00 0.33 0.33 0.33 1.00 1.00 0.00 0,40 1972-79 0.75 0.25 0.83 1.00 0.67 - 0.00 0.00 0.67 0.00 0.33 0.33 0.33 1.00 1.00 0.00 0.40 1980-89 0.75 0.25 0.83 1.00 0.67 - 0.00 0.00 0.67 0.00 0.33 0.33 0.33 1.00 1.00 0.00 0.40 Luxembourg 1980-89 0.75 0.25 0.83 0.00 0.33 - 0.00 0.60 0.00 0.00 0.33 1.00 - 1.00 0.25 0.00 0.33 Malaysia 1960-71 0.50 0.00 0.83 0.00 0.00 0.20 0.00 0.60 0.33 - 0.67 0.00 0.33 0.67 0.25 0.00 0.36 1972-79 O.S0 0.00 0.83 0.00 0.00 0.20 0.00 0.60 0.33 - 0.67 0.00 0.33 0.67 0.25 0.00 0.36 1980-89 0.50 0.00 0.83 0.00 0.00 0.20 0.00 0.60 0.33 - 0.67 0.00 0.33 0.67 0.25 0.00 0.36 Malta 1960-71 O.50 0.50 0.83 1.00 0.00 0.20 0.00 0.40 0.67 0.00 0.33 1.00 0.33 1.00 0.25 0.00 0.44 1972-79 0.50 0,50 0.83 1.00 0.00 0.20 0.00 0.40 0.67 0.00 0.33 1.00 0.33 1.00 0.25 0.00 0.44 1980-89 0.50 0.50 0.83 1.00 0.00 0.20 0.00 0.40 0.67 0.00 0.33 1.00 0.33 1.00 0.25 0.00 0.44 Mexico 1950-59 - 0.00 0.83 1.00 0.00 0.20 0.00 0.00 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.25 1960-71 - 1.00 0.83 1.00 0.67 0.20 0.00 0.00 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.34 1972-79 - 1.00 0.83 1.00 0.67 0.20 0.00 0.00 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.34 1980-89 - 1.00 0.83 1.00 0.67 0.20 0.00 0.00 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.34 Morocco 1960-71 - 0.25 0.00 0.00 - - 0.00 0.20 0.33 0.00 0.00 0.33 0.33 0.67 0.25 0.00 0.14 1972-79 - 0.25 0.00 0.00 - - 0.00 0.20 0.33 0.00 0.00 0.33 0.33 0.67 0.25 0.00 0.14 1980-89 - 0.25 0.00 0.00 - - 0.00 0.20 0.33 0.00 0.00 0.33 0.33 0.67 0.25 0.00 0.14 Nepal 1950-59 0.50 0.25 0.00 0.00 - 0.00 0.00 0.20 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.18 1960-71 0.50 0.25 0.00 0.00 - 0.00 0.00 0.20 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.18 1972-79 0.50 0.25 0.00 0.00 - 0.00 0.00 0.20 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.18 1980-89 O.S0 0,25 0.00 0.00 - 0.00 0.00 0.20 0.00 0.00 0.33 1.00 - 0.67 0.25 0.00 0.18 (Table continues on the following page.) Table A-1. (continued) Disaggregated legal variables Policyformulation variables Central Aggregate CEO variables Who bank Limitations on lending variables legal 'Ierm of Who Dis- Other formu- Final Role in objectives Securitized Terms of Potential Type Maturity Interest Primary central bank Economy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates market independence and decade (la) (Ib) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4h) variable Netherlands 1950-59 0.75 0.00 0.17 1.00 0.33 0.20 0.00 0.80 0.67 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.42 1960-71 0.75 0.00 0.17 1.00 0.33 0.20 0.00 0.80 0.67 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.42 1972-79 0.75 0.00 0.17 1.00 0.33 0.20 0.00 0.80 0.67 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.42 1980-89 0.75 0.00 0.17 1.00 0.33 0.20 0.00 0.80 0.67 0.00 0.00 1.00 1.00 0.00 0.00 0.00 0.42 New Zealand 1950-59 0.50 0.00 0.83 0.00 0.00 0.00 0.00 0.00 - - 0.00 0.33 0.33 1.00 0.75 0.00 0.18 1960-71 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.40 0.00 0.00 0.00 1.00 - 0.00 0.50 0.00 0.24 1972-79 0.50 0.00 0.83 1.00 0.00 0.00 0.0( 0.40 0.00 0.00 0.00 1.00 - 0.00 0.50 0.00 0.24 1980-89 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.40 0.00 0.00 0.00 1.00 - 0.00 0.50 0.00 0.24 Nicaragua 1960-71 0.00 0.00 0.83 1.00 1.00 - 0.00 0.00 1.00 0.33 0.67 0.00 0.33 0.67 0.50 0.00 0.45 1972-79 0.00 0.00 0.83 1.00 1.00 - 0.00 0.00 1.00 0.33 0.67 0.00 0.33 0.67 0.50 0.00 0.45 1980-89 0.00 0.00 0.83 1.00 1.00 - 0.00 0.00 1.00 0.33 0.67 0.00 0.33 0.67 0.50 0.00 0.45 Nigeria 1960-71 0.50 0.00 0.83 0.50 0.33 0.20 0.00 0.60 0.33 0.33 0.33 0.00 0.33 0.67 0.75 0.00 0.37 1972-79 0.50 0.00 0.83 0.50 0.33 0.20 0.00 0.60 0.33 0.33 0.33 0.00 0.33 0.67 0.75 0.00 0.37 1980-89 0.50 0.00 0.83 0.50 0.33 0.20 0.00 0.60 0.33 0.33 0.33 0.00 0.33 0.67 0.75 0.00 0.37 Norway 1950-59 0.75 0.00 0.33 1.00 0.00 0.20 0.00 0.00 - - 0.00 0.00 - 0.67 0.25 0.00 0.20 1960-71 0.75 0.00 0.33 1.00 0.00 0.20 0.00 0.00 0.00 0.00 0.00 0.00 - 0.67 0.25 0.00 0.15 1972-79 0.75 0.00 0.33 1.00 0.33 0.20 0.00 0.00 0.00 0.00 0.00 0.00 - 0.67 0.50 0.00 0.17 1980-89 0.75 0.00 0.33 1.00 0.33 0.20 0.00 0.00 0.00 0.00 0.00 0.00 - 0.67 0.50 0.00 0.17 Pakistan 1950-59 0.50 0.25 0.83 O.S0 0.00 0.00 0.00 0.40 0.00 0.00 0.00 0.3.3 - 0.67 0.25 0.00 0.21 1960-71 0.50 0.25 0.83 O.S0 0.00 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.67 0.25 0.00 0.21 1972-79 0.50 0.25 0.83 0.50 0.00 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.67 0.25 0.00 0.21 1980-89 0.50 0.25 0.83 0.50 0.00 0.00 0.00 0.40 0.(( 0.00 0.00 0.33 - 0.67 0.25 0.00 0.21 Panama 1950-59 - 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.00 0.25 0.00 0.24 1960-71 - 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.00 0.25 0.00 0.24 1972-79 - 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.00 - 0.00 0.25 0.00 0.22 1980-89 - 0.25 0.83 1.00 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.00 - 0.00 0.25 0.00 0.22 Peru 1960-71 0.00 1.00 0.83 1.00 0.67 0.20 0.00 0.40 0.00 0.67 0.67 0.33 0.33 1.00 0.25 0.00 0.43 1972-79 0.00 1.00 0.83 1.00 0.67 0.20 0.00 0.40 0.00 0.67 0.67 0.33 0.33 1.00 0.25 0.00 0.43 1980-89 0.00 1.00 0.83 1.00 0.67 0.20 0.00 0.40 0.00 0.67 0.67 0.33 0.33 1.00 0.25 0.00 0.43 Philippines 1950-59 0.75 0.00 0.83 0.50 0.67 0.20 0.00 1.00 0.33 0.00 0.33 0.67 0.33 0.33 0.25 0.00 0.43 1960-71 0.75 0.00 0.83 0.50 0.67 0.20 0.00 1.00 0.33 0.00 0.33 0.67 0.33 0.33 0.25 0.00 0.43 1972-79 0.75 0.00 0.83 0.50 0.67 0.20 0.00 1.00 0.33 0.00 0.33 0.67 0.33 0.33 0.25 0.00 0.43 1980-89 0.75 0.00 0.83 0.50 0.67 0.20 0.00 1.00 0.33 0.00 0.33 0.67 0.33 0.33 0.25 0.00 0.43 Poland 1950-S9 0.50 0.25 0.00 0.50 0.33 0.00 0.00 0.00 0.00 0.00 0.33 0.00 - 0.67 0.25 0.00 0.14 1960-71 0.50 0.25 0.00 0.50 0.33 0.00 0.00 0.00 0.00 0.00 0.33 0.00 - 0.67 0.25 0.00 0.14 1972-79 0.50 0.25 0.00 0.00 0.67 0.00 0.00 0.00 0.00 0.00 0.33 0.00 - 0.00 0.25 0.00 0.10 1980-89 0.50 0.25 0.00 0.00 0.67 0.00 0.00 0.00 0.00 0.00 0.33 0.00 - 0.00 0.25 0.00 0.10 Portugal 1972-79 0.50 0.25 1.00 1.00 0.33 0.40 0.00 0.60 0.33 0.33 0.67 0.67 0.00 0.67 0.25 0.00 0.48 1980-89 0.00 0.25 1.00 0.00 0.33 0.60 0.00 0.00 0.67 0.67 0.67 0.00 1.00 1.00 0.25 0.00 0.41 Qatar 1972-79 0.50 0.25 0.83 0.50 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.00 0.25 0.00 0.20 1980-89 0.50 0.25 0.83 0.50 0.33 0.00 0.00 0.40 0.00 0.00 0.00 0.33 - 0.00 0.25 0.00 0.20 Romania 1950-59 0.50 0.25 1.00 1.00 0.67 0.60 0.00 0.60 0.67 0.33 0.67 0.00 0.67 0.67 0.00 0.00 0.53 1960-71 0.50 0.25 1.00 1.00 0.67 0.60 0.00 0.60 0.67 0.33 0.67 0.00 0.67 0.67 0.00 0.00 0.53 1972-79 - 0.25 - 0.00 1.00 0.80 0.00 0.20 - - 0.67 0.00 - 0.00 0.25 0.00 0.30 1980-89 - 0.25 - 0.00 1.00 0.80 0.00 0.20 - - 0.67 0.00 - 0.00 0.25 0.00 0.30 Singapore 1972-79 0.25 0.00 0.83 0.00 - - 0.00 0.60 - - 0.33 0.00 - 1.00 0.25 0.00 0.29 1980-89 0.25 0.00 0.83 0.00 - - 0.00 0.60 - - 0.33 0.00 - 1.00 0.25 0.00 0.29 (Table continues on the following page.) Table A-1. (continued) Disaggregated legal variables Policy formulation variables Central Aggregate CEO variables Who bank Limitations on lending variables legal Termz of Who Dis- Other formu- Final Role in objectives Securitized Terms of Potential Type Maturity Interest Primary central bank Economy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates market independence and decade (la) (Ib) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4h) variable South Africa 1950-59 0.50 0.00 0.83 0.50 - - 0.00 0.20 0.00 0.00 0.33 1.00 - 1.00 0.25 0.00 0.25 1960-71 0.50 0.00 0.83 0.50 - - 0.00 0.20 0.00 0.00 0.33 1.00 - 1.00 0.25 0.00 0.25 1972-79 0.50 0.00 0.83 0.50 - - 0.0( 0.20 0.00 0.00 0.33 1.00 - 1.00 0.25 0.00 0.25 1980-89 0.50 0.00 0.83 0.50 - - 0.00 0.20 0.00 0.00 0.33 1.00 - 1.00 0.25 0.00 0.25 Spain 1950-59 0.00 0.25 0.00 0.50 0.33 0.00 0.00 0.00 0.33 0.00 0.33 0.00 0.00 0.00 0.00 0.00 0.13 1960-71 0.00 0.25 0.00 0.00 0.33 0.00 0.00 0.00 0.33 0.00 0.00 0.33 0.00 0.00 0.00 0.00 0.09 tQ 1972-79 0.00 0.25 0.D0 0.00 0.33 0.00 0.00 0.00 0.33 0.00 0.00 0.33 0,00 0.00 0.00 0.00 0.09 1980-89 0.25 0.00 0.00 1.00 0.33 0.00 0.00 0.60 0.33 0.00 0.00 0.33 0.00 0.00 0.00 0.00 0.23 Sweden 1950-59 0.00 1.00 - 0.50 - - 0.00 0.20 0.33 0.00 0.67 0.00 1.00 0.67 0.25 0.00 0.29 1960-71 0.00 1.00 - 0.50 - - 0.00 0.20 0.33 0.00 0.67 0.00 1.00 0.67 0.25 0.00 0.29 1972-79 0.00 1.00 - 0.50 - - 0.00 0.20 0.33 0.00 0.67 0.00 1.00 0.67 0.25 0.00 0.29 1980-89 0.00 1.00 - 0.50 - - 0.00 0.20 0.33 0.00 0.67 0.00 1.00 0.67 0.25 0.00 0.29 Switzerland 1950-59 0.75 0.25 - 1.00 - 1.00 0.00 0.00 0.67 0.33 0.67 1.00 - 1.00 0.25 0.00 0.53 1960-71 0.75 0.25 - 1.00 - 1.00 0.00 0.00 0.67 0.33 0.67 1.00 - 1.00 0.25 0.00 0.53 1972-79 0.75 0.25 - 1.00 - 1.0() 0.00 0.00 0.67 0.33 0.67 1.00 - 1.00 0.25 0.00 0.53 1980-89 0.75 0.25 - 1.00 - 1.00 0.00 0.00 1.00 - 1.00 1.00 - 1.00 0.25 0.00 0.64 Taiwan 1980-89 0.50 0.50 1.00 0.00 0.00 0.00 0.00 0.60 0.00 0.00 0.00 0.33 0.00 0.00 0.00 0.00 0.21 Tanzania 1960-71 0.25 0.00 0.83 0.50 0.67 0.20 0.00 0.40 0.67 0.33 0.33 1.00 0.33 0.67 0.75 0.00 0.44 1972-79 0.25 0.00 0.83 0.SO 0.67 0.20 0.00 0.40 0.67 0.33 0.33 1.00 0.33 0.67 0.75 0.00 0.44 1980-89 0.25 0.00 0.83 0.50 0.67 0.20 0.00 0.40 0.67 0.33 0.33 1.00 0.33 0.67 0.75 0.00 0.44 Thailand 1950-59 0.25 0.50 0.00 0.00 0.00 0.20 0.00 0.60 0.33 0.00 0.33 0.33 0.00 0.67 0.25 0.00 0.27 1960-71 0.25 0.50 0.00 0.00 0.00 0.20 0.00 0.60 0.33 0.00 0.33 0.33 0.00 0.67 0.25 0.00 0.27 1972-79 0.25 0.50 0.00 0.00 0.00 0.20 0.00 0.60 0.33 0.00 0.33 0.33 0.00 0.67 0.25 0.00 0.27 1980-89 0.25 0.50 0.00 0.00 0.00 0.20 0.00 0.60 0.33 0.00 0.33 0.33 0.00 0.67 0.25 0.00 0.27 Turkey 1950-59 0.50 0.75 1.00 1.00 0.67 0.20 0.00 0.40 0.33 0.00 0.33 0.00 0.67 1.00 0.25 0.00 0.39 1960-71 0.50 0.75 1.00 1.00 0.67 0.20 0.00 0.40 0.33 0.00 0.33 0.00 0.67 1.00 0.25 0.00 0.39 1972-79 0.50 0.75 0.83 0.00 0.33 0.80 0.00 0.60 0.33 0.67 0.33 0.33 0.00 0.67 0.25 0.00 0.46 1980-89 0.50 0.75 0.83 0.00 0.33 0.80 0.00 0.60 0.33 0.67 0.33 0.33 0.00 0.67 0.25 0.00 0.46 Uganda 1960-71 0.50 0.50 0.83 0.50 0.00 0.20 0.00 0.40 0.33 - 0.33 0.33 0.33 1.00 0.75 0.00 0.38 1972-79 0.50 0.50 0.83 0.50 0.00 0.20 0.00 0.40 0.33 - 0.33 0.33 0.33 1.00 0.75 0.00 0.38 1980-89 0.50 0.50 0.83 0.50 0.00 0.20 0.00 0.40 0.33 - 0.33 0.33 0.33 1.00 0.75 0.00 0.38 United Kingdom 1950-59 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.20 0.00 0.00 0.00 1.00 - 0.00 0.25 0.00 0.21 1960-71 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.20 0.67 0.67 0.00 1.00 1.00 1.00 0.75 0.00 0.43 1972-79 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.20 0.00 0.00 0.00 1.00 1.00 1.00 0.75 0.00 0.27 1980-89 0.50 0.00 0.83 1.00 0.00 0.00 0.00 0.20 0.00 0.00 0.00 1.00 1.00 1.00 0.75 0.00 0.27 United States 1950-59 0.25 0.50 0.00 1.00 - 0.20 0.00 0.40 1.00 0.33 0.33 1.00 - 1.00 0.25 0.00 0.48 1960-71 0.25 0.50 0.00 1.00 - 0.20 0.00 0.40 1.00 0.33 0.33 1.00 - 1.00 0.25 0.00 0.48 1972-79 0.25 0.50 0.00 1.00 - 0.20 0.00 0.40 1.00 0.33 0.33 1.00 - 1.00 0.25 0.00 0.48 1980-89 0.25 0.50 0.00 1.00 - 0.20 0.00 0.40 1.00 0.33 0.33 1.00 - 1.00 0.25 0.00 0.48 Uruguay 1950-59 0.25 0.00 - 0.50 0.67 0.00 0.00 0.20 - - 0.33 0.33 - 0.00 0.25 0.00 0.22 1960-71 0.25 0.00 - 0.50 0.67 0.00 0.00 0.20 - - 0.33 0.33 - 0.00 0.25 0.00 0.22 1972-79 0.25 0.25 0.83 0.00 0.67 0.20 0.00 0.40 0.00 0.00 0.67 0.00 - 0.00 0.25 0.00 0.24 1980-89 0.25 0.25 0.83 0.00 0.67 0.20 0.00 0.40 0.00 0.00 0.67 0.00 - 0.00 0.25 0.00 0.24 Venezuela 1950-59 0.50 0.00 - 0.00 0.00 - 0.00 0.20 - - 0.67 1.00 - 0.67 0.25 0.00 0.28 1960-71 0.50 0.00 - 0.00 0.00 - 0.00 0.20 - - 0.67 1.00 - 0.67 0.25 0.00 0.28 1972-79 0.50 0.00 0.83 0.50 0.33 1.00 0.00 0.40 0.67 0.00 0.67 0.00 0.33 0.00 0.50 0.00 0.43 1980-89 0.50 0.00 0.83 0.50 0.33 1.00 0.00 0.40 0.67 0.00 0.67 0.00 0.33 0.00 0.50 0.00 0.43 (Table continues on the following page.) Table A-1. (continued) Disaggregated legal variables Policy formulation variables Central Aggregate CEO variables Who bank limitations on lending variables legal Term of Who Dis- Other formu- Final Role in objectives Securitized Terms of Potential Type Maturity Interest Primary central bank Economy office appoints missal offices lates authority budget variable Advances lending lending borrowers of limit of loans rates market independence and decade (Ia) (Ib) (Ic) (Id) (2a) (2b) (2c) (3) (4a) (4b) (4c) (4d) (4e) (4f) (4g) (4h) variable Western Samoa 1980-89 0.25 0.25 0.83 0.00 0.00 0.20 0.00 0.40 0.33 0.33 0.33 0.00 0.33 1.00 0.25 0.00 0.30 Yugoslavia 1960-71 0.25 0.25 0.83 0.00 0.00 0.20 0.00 0.40 0.00 - 0.00 0.00 - 0.00 0.25 0.00 0.17 1972-79 0.25 0.25 0.83 0.00 0.00 0.20 0.00 0.40 0.00 - 0.00 0.00 - 0.00 0.25 0.00 0.17 1980-89 0.25 0.25 0.83 0.00 0.00 0.20 0.00 0.40 0.00 - 0.00 0.00 - 0.00 0.25 0.00 0.17 > Zaire 1972-79 0.50 0.00 1.00 0.00 0.67 0.20 0.00 0.60 - 0.00 0.33 0.33 0.33 0.67 0.75 0.00 0.35 1980-89 0.50 0.00 1.00 1.00 0.67 0.60 0.00 0.40 0.33 0.33 0.33 0.33 0.33 0.67 0.75 0.00 0.43 Zambia 1960-71 0.50 0.25 0.83 1.00 0.00 0.20 0.00 0.80 0.33 0.33 0.00 0.33 0.33 0.33 0.75 0.00 0.40 1972-79 0.50 0.25 0.83 1.00 0.00 0.20 0.00 0.80 0.33 0.33 0.00 0.33 0.33 0.33 0.75 0.00 0.40 1980-89 0.50 0.25 0.83 0.50 0.33 ().20 0.00 0.40 0.33 0.33 0.00 0.33 0.33 0.33 0.75 0.00 0.33 Zimbabwe 1950-59 0.75 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.33 0.33 0.00 0.67 0.33 0.33 0.25 0.00 0.23 1960-71 0.75 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.33 0.33 0.00 0.67 0.33 0.33 0.25 0.00 0.20 1972-79 0.75 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.33 0.33 0.00 0.67 0.33 0.33 0.25 0.00 0.20 1980-89 0.75 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.33 0.33 0.00 0.67 0.33 0.33 0.25 0.00 0.20 - Not available. Table A-2. Country Codes for the Questionnaire Variables and the Questionnaire-based Index of Central Bank Independence Financial independence Policy targets Function Limita- Resolu- Salaries Mone- Priority as devel- Questionnaire- Tenure tions on tion of and tary Interest of price opment based index of overlap lending conflict Budget profits stock rate stability bank central bank Country (1) (2) (3) (4a) (4b) (Sa) (Sb) (6) (7) independence Australia 1.00 0.67 0.00 1.00 0.50 0.33 1.00 1.00 1.00 0.76 Bahamas, The 0.50 0.67 0.50 1.00 1.00 0.00 1.00 0.67 1.00 0.71 Barbados 0.50 0.67 0.00 1.00 0.50 0.00 0.00 1.00 0.67 0.54 Belgium 1.00 0.00 0.00 1.00 1.00 0.00 1.00 0.67 - 0.47 Costa Rica 0.50 0.67 0.50 1.00 1.00 0.67 1.00 1.00 1.00 0.81 Denmark 1.00 1.00 0.50 0.50 1.00 0.00 0.00 0.67 1.00 0.73 Ethiopia 0.00 0.00 0.00 0.00 0.00 0.00 1.00 0.33 0.00 0.13 Finland 1.00 1.00 1.00 0.50 0.00 0.00 - 1.00 1.00 0.78 France 0.50 0.33 0.50 0.50 1.00 0.67 1.00 1.00 0.67 0.65 Ireland 0.50 0.33 0.00 0.50 0.00 0.00 1.00 1.00 1.00 0.57 Italy 1.00 0.00 0.50 1.00 1.00 0.67 1.00 1.00 1.00 0.73 Lebanon 0.50 0.33 0.50 1.00 1.00 0.00 1.00 0.33 1.00 0.59 Luxembourg 0.50 0.33 - 1.00 1.00 0.00 1.00 0.67 1.00 0.66 Peru 0.00 0.00 0.00 1.00 0.50 0.33 0.00 0.33 0.33 0.22 South Africa 0.50 0.33 0.50 1.00 0.00 0.67 1.00 - 1.00 0.64 Tanzania 0.50 0.33 0.50 1.00 0.00 0.33 1.00 0.00 0.33 0.38 Turkey 0.50 0.00 0.00 1.00 1.00 0.00 1.00 1.00 0.33 0.44 Uganda 0.50 0.33 0.50 1.00 1.00 0.33 - 1.00 0.33 0.53 Uruguay 0.50 0.33 0.50 1.00 0.50 0.00 - 0.33 1.00 0.49 United Kingdom 0.50 0.33 0.00 0.50 - 0.67 1.00 1.00 1.00 0.64 Germany, Fed. 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 Rep. of Yugoslavia - 0.00 0.00 0.50 0.00 0.33 1.00 - 0.00 0.17 Zaire 0.50 0.33 0.50 1.00 1.00 0.67 1.00 1.00 0.33 0.61 - Not available. Note: Numbers in parentheses in the column headings correspond to variable numbers in table 4. 396 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Alesina, Alberto. 1988. "Macroeconomics and Politics." NBER Macro-economics Annual 3: 13-52. Alesina, Alberto, and Guido Tabellini. 1987. "Rules and Discretion with Noncoordi- nated Monetary and Fiscal Policies." Economic Inquiry 25: 619-30. Aufricht, Hans. 1961. Central Banking Legislation. Washington, D.C.: IMF. .1967. Central Banking Legislation. Vol. II, Europe. Washington, D.C.: IMF. Bade, Robin, and Michael Parkin. 1980. "Central Bank Laws and Monetary Policy." University of Western Ontario, Department of Economics, London, Ontario. Processed. Banaian, King, Leroy Laney, and Thomas Willett. 1983. "Central Bank Independence: An International Comparison." Economic Review: Federal Reserve Bank of Dallas (March): 1-13. Bank for International Settlements. 1963. Eight European Central Banks. New York: Praeger under the auspices of the BIS. Bodart, Vincent. 1990. "Central Bank Independence and the Effectiveness of Monetary Policy: A Comparative Analysis." IMF, Central Banking Department, Washington, D.C. Processed. Bordes, Christian, and M. 0. Strauss-Kahn. 1987. "1977-86: Dix ans de Politique D'objectifs en France ou le 'Targeting' a la Francaise." SUERF (Societe Universitaire Europeene de Recherches Financiers) papers on Monetary Policy and Financial Sys- tems 4, Tilburg. Processed. Bordo, Michael, and Angela Redish. 1987. "Why Did the Bank of Canada Emerge in 1935?" Journal of Economic History 47 (June): 405-17. Cargill, Thomas, and Michael Hutchison. 1990. "Monetary Policy and Political Econ- omy: The Federal Reserve and the Bank of Japan." In Thomas Mayer, ed., The Political Economy of American Monetary Policy. New York: Cambridge University Press. Cukierman, Alex. 1984. Inflation, Stagflation, Relative Prices, and Imperfect Informa- tion. Cambridge: Cambridge University Press. . 1992a. Central Bank Strategy, Credibility, and Independence: Theory and Evi- dence. Cambridge, Mass.: MIT Press. 1992b. "Commitment through Delegation and Its Effect on Policy: The Case of Central Bank Independence." Paper presented at the Conference on Coalitions and Bargaining, Tel Aviv University, Department of Economics. Processed. Dotsey, Michael. 1986. "Japanese Monetary Policy: A Comparative Analysis." Eco- nomic Review: Federal Reserve Bank of Richmond 72 (November/December): 12- 24. Effros, Robert, ed. 1982. Emerging Financial Centers: Legal and Institutional Frame- work. Washington, D.C.: IMF. Eizenga, Weitze. 1983. "The Independence of the Federal Reserve System and of the Netherlands Bank." SUERF (Societe Universitaire Europeene de Recherches Financiers) Series 41A, Tilburg. Processed. Cukierman, Webb, and Neyapti 397 1987. "The Independence of the Deutsche Bundesbank and the Nederlandsche Bank with Regard to Monetary Policy: A Comparative Study." SUERF (Societe Univer- sitaire Europeene de Recherches Financiers) papers on Monetary Policy and Financial Systems 2, Tilburg. Processed. Epstein, Gerald, and Juliet Schor. 1986. "The Divorce of the Banca d'Italia and the Italian Treasury: A Case Study of Central Bank Independence." Harvard Institute of Economic Research, Discussion Paper Series 1269. Cambridge, Mass. Processed. Fazio, Antonio. 1991. "Role and Independence of Central Banks." In Patrick Downes and Reza Vaez-Zadeh, eds., The Evolving Role of Central Banks. Washington, D.C.: IMF. Fischer, Stanley. 1986. Indexing, Inflation, and Economic Policy. Cambridge, Mass.: MIT Press. Goodhart, Charles. 1988. The Evolution of Central Banks. Cambridge, Mass.: MIT Press. Grilli, Vittorio, Donato Masciandaro, and Guido Tabellini. 1991. "Political and Mone- tary Institutions and Public Financial Policies in the Industrial Countries." Economic Policy 13: 341-92. Holtfrerich, Carl-Ludwig. 1988. "Relations between Monetary Authorities and Govern- mental Institutions: The Case of Germany from the 19th Century to the Present." In Gianni Toniolo, ed., Central Bank Independence in Historical Perspective. Berlin: Walter de Gruyter. Kearney, Colm. 1984. "The British Anti-Inflation Strategy: Implementing Monetarism or Turning to Radcliffe?" SUERP Series 44A. Tilburg. Processed. Keenan, Peter, and David Mayes. 1987. "Monetary Targets and Monetary Management: The New Zealand Experience.' SUERF (Societe Universitaire Europeene de Recherches Financiers) Papers on Monetary Policy and Financial Systems 3, Tilburg. Processed. Leone, Alfredo. 1991. "Effectiveness and Implications of Limits on Central Bank Credit to the Government." In Patrick Downes and Reza Vaez-Zadeh, eds., The Evolving Role of Central Banks. Washington, D.C.: IMF. Lobel, Arnold. 1972. Frog and Toad Together. New York: Harper and Row. Lohmann, Susanne. 1992. "Optimal Commitment in Monetary Policy: Credibility versus Flexibility." American Economic Review. 82 (March): 273-86. Masciandaro, Donato, and Guido Tabellini. 1988. "Monetary Regimes and Fiscal Defi- cits: A Comparative Analysis." In H. S. Cheng, ed., Monetary Policy in Pacific Basin Countries. Dordrecht: Kluwer Academic Publishers. Maxfield, Sylvia. 1992. "Private Constituencies and the Power of State Monetary Au- thorities in Newly Industrializing Countries.' Yale University, Department of Political Science, New Haven, Conn. Processed. Mayer, Thomas, ed. 1990. The Political Economy of American Monetary Policy. New York: Cambridge University Press. Meltzer, Allan. 1991. "U.S. Policy in the Bretton Woods Era.' The 1991 Homer Jones Lecture. St. Louis, Mo. Mittra, Sid. 1977. Central Bank versus Treasury: An International Study. Lanham, Md.: University Press of America. Parkin, Michael. 1986. "Domestic Monetary Institutions and Deficits?' In J. M. Buchanan, C. K. Rowley, and R. D. Tollison, eds., Deficits. New York: Basil Blackwell. 398 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Rogoff, Kenneth. 1985. "The Optimal Degree of Commitment to an Intermediate Mon- etary Target'" Quarterly Journal of Economics 100 (November): 1169-90. Schokker, E. 1980. "The Central Bank and the State in the Light of European Integra- tion:' SUERF (Societe Universitaire Europeene de Recherches Financiers) Series 34A, Tilburg. Processed. Skanland, Hermod. 1984. The Central Bank and Political Authorities in Some Industrial Countries. Oslo: Norges Bank. Suzuki, Yoshio. 1987. The Japanese Financial System. Oxford: Clarendon. Swinburne, Mark, and Marta Castello-Branco. 1991. "Central Bank Independence and Central Bank Functions." In Patrick Downes and Reza Vaez-Zadeh, eds., The Evolv- ing Role of Central Banks. Washington, D.C.: IMF. Volcker, Paul, Miguel Mancera, and Jean Godeaux. 1991. Perspectives on the Role of a Central Bank. Proceedings of a conference held in Beijing, China, January 5-7, 1990. Beijing: Peoples' Bank of China. Willett, Thomas, ed. 1988. Political Business Cycles: The Political Economy of Money, Inflation, and Unemployment. Durham, N.C.: Duke University Press. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 399-422 On the Transmission of World Agricultural Prices Yair Mundlak and Donald F. Larson Two questions are asked about the relationship between domestic prices and world prices of agricultural commodities: are variations in world prices transmitted to domes- tic prices, and do these variations in world prices constitute an important component of variations in domestic prices? Domestic prices are regressed on world prices in various forms, taking into account the possible effects of exchange rates and inflation. The empirical analysis is based on data from the Food and Agriculture Organization of the United Nations for 58 countries for 1968-78 and for the countries of the European Community for 1961-85. The results show that most of the variations in world prices are transmitted and that they constitute the dominant component in the variations of domestic prices. Agricultural products are on the whole tradable, and every country trades in some agricultural products. In the absence of intervention it is expected that domestic prices of such products will vary with world prices. It is well known, however, that agriculture is subjected to considerable intervention, which cre- ates a gap between world prices and domestic prices and which generates cross- country variations in agricultural prices (see, for example, McCalla 1969; John- son 1973; Bale and Lutz 1981; Australia 1985; Anderson, Hayami, and Honma 1986; World Bank 1986). This is perhaps why it is sometimes claimed that world prices are irrelevant for the development of agriculture in countries that intervene in the pricing of their agricultural products; see, for instance, the implicit debate in Mellor and Ahmed (1988), Valdes and Siamwalla (1988), and Ahmed (1988). It is therefore natural to ask to what extent such an intervention reduces the influence of world prices on domestic prices. In this article we examine two major questions. First, what proportion of the variations in world prices is transmitted to domestic prices? Second, what pro- portion of the variations in domestic prices can be attributed to variations in world prices? Not independently, we also examine empirically the revealed rela- tionships between world prices and the degree of intervention. The outcome of this analysis is crucial for understanding the relationships between domestic and Yair Mundlak is with the University of Chicago and the International Food Policy Research Institute. Donald F. Larson is with the International Economics Department at the World Bank. The authors are grateful to Ronald C. Duncan and D. Gale Johnson for comments and suggestions on an earlier draft. © 1992 The International Bank for Reconstruction and Development / THE WORLD BANK 399 400 THIE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 world markets, and it has several ramifications. The facts and the ramifications are two distinct subjects, however, and it is important to deal with them sepa- rately. This article deals mainly with the empirical analysis. I. THE FRAMEWORK The framework draws on the law of one price, where the domestic price of commodity i in year t, Pit, is expressed as a product of the world price, P,-,, the nominal exchange rate, Et, and the tax policy Si, = (1 + fit), where T is the tax rate (or, if negative, the subsidy). Because countries engage in nontariff market interventions, including quantitative import restrictions, the tax policy term here includes the tariff equivalent of any restriction on domestic prices. The equation for the domestic price is (1) Pi, = P;',E,Si,_ This formulation ignores differences in product qualities and in transportation, storage, and marketing costs, as well as other domestic nontradable inputs. Also, the equation is based on the assumption that the exchange rate is neither under- nor overvalued so that the difference between domestic and foreign inflation rates is fully reflected in E. To allow for deviations from this assump- tion and for the effects that are not included in the equation, a disturbance term, denoted by U, is added to the equation. Equation 1 is rewritten with lowercase letters indicating logs: (2) Pi, = P*t + et + Sit + uit where u - IID(p,, a2) and E(eu) = E(su) = E(p*u) = 0. The mean of the disturbance, a, is not necessarily 0, for the reasons given above. The answer to both questions posed above is obtained, in principle, by com- puting the following regression: (3) Pit = ce + 0P* + yet + Ei,. Equation 2 can be expressed in terms of equation 3, subject to the restrictions H1: 0 = 1, and H2: _y = 1. The coefficient , is the elasticity of the domestic price with respect to the world price, to be referred to as the elasticity of transmission. The value of this elasticity is the answer to the first question. A value of 1 implies that the variations in world prices are fully transmitted to the domestic prices, whereas a value of 0 implies no transmission at all. There are several reasons why this elasticity would differ from unity. First, omitted variables, specifically tax-policy variables (s), are correlated with the world price. Second, there may be measurement errors in the world price. Such errors may reflect the fact that the world price used in a given study differs from the one pertinent for the particular country. Third, if the economy is closed, the world price is irrelevant. Of course, very few countries are completely closed, but many countries are Mundlak and Larson 401 partially closed by means of trade policies, and this may affect the value of the estimate. The contribution of world prices, measured in domestic currencies, to the variations in domestic prices is given by the value of R2 of the regression of equation 3. This answers the second question: a low value means that only a small proportion of the variations in domestic prices are accounted for by world prices and exchange rates. The marginal contribution of world prices condi- tional on the exchange rate is given by the square of the partial correlation coefficient between world and domestic prices. The foregoing discussion dealt with proportional changes, but it says nothing about differences in levels. This information is contained in the intercept, which need not equal 0. Under the restrictions H1 and H2, a = s + it, where s is the sample average, over commodities and time, of sit' Thus the intercept reflects the tax policy and the quantitative importance of the omitted variables from equation 2. More generally, when H1 or H2 are not maintained, or if the explan- atory variables are measured with error, the intercept will be affected. Il. THE DATA The regression was first computed for 58 countries for the period 1968-78, and the sample covered some 60 products. The number of products varied by country. Products not produced in a country were excluded from the analysis. The domestic prices are those given by the Food and Agriculture Organization of the United Nations (FAO) and described by the FAo as follows: Farm prices are in theory determined by farmgate or first point-of-sale transaction, when farmers participate in their capacity as sellers of their own products. Of course, data may not always refer to the same selling points, depending on the prevailing institutional setup in the countries. Also, different practices may prevail in regard to individual commodities (FAO 1987, p. 23). There is a common belief that FAO prices are subject to many problems. This may be the case, but we are unfamiliar with any study that indicates the sources of errors in the FAo data and their quantitative importance, or that substantiates this belief in any other way. One way to have a rough check on the data is to see, as we are doing here, to what extent they are correlated with world prices. If the empirical results showed weak relationships between the domestic and world prices, this could be explained in terms of measurement errors. This is not the case, however, and therefore we think that these data are indeed informative. To double-check, we repeat the analysis in section IV for the European Community, using prices of the U.S. Department of Agriculture reported in Herlihy and others (1989) for 1960-85. The world price is an export-unit value calculated in nominal U.S. dollars. It 402 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 is a ratio of the total world value of exports for each of the commodities divided by the total world exported quantities for the corresponding commodities. We use the framework developed in section I to examine and verify the pertinence of this variable. The exchange rates are annual averages as published by the Inter- national Monetary Fund. III. RESULTS USING FAO DATA In a cross-commodity comparison, the deviation from unitary transmission elasticity is surprisingly small. The time-series analysis for individual commodi- ties yields somewhat lower values, suggesting that policy does have some smoothing effect. In all cases, the relatively high values of R2 indicate that world prices constitute a major source of domestic price variations. Table 1. Estimated Transmission and Exchange Rate Elasticities, 1968-78 Nominal values Real values World Exchange World Exchange Country prices rates R2 prices rates R2 Argentina 0.98 0.97 0.94 0.97 0.42 0.71 Australia 0.92 0.63 0.82 0.93 1.04 0.80 Austria 0.98 1.08 0.87 0.98 1.11 0.86 Bangladesh 0.74 0.82 0.66 0.70 0.15 0.50 Belgium and Luxembourg 0.97 1.05 0.79 0.97 1.08 0.79 Brazil 0.87 1.25 0.83 0.86 -1.54 0.69 Burundi 0.87 2.30 0.77 0.88 1.87 0.76 Cameroon 0.89 0.47 0.74 0.88 0.71 0.72 Canada 1.00 -0.20 0.85 1.01 0.35 0.84 Chile 0.94 0.98 0.96 0.95 0.67 0.63 Colombia 0.94 0.85 0.76 0.93 0.60 0.69 Costa Rica 0.93 0.45 0.85 0.92 2.22 0.83 Cyprus 0.92 0.73 0.68 0.93 0.76 0.66 Denmark 1.02 0.81 0.83 1.02 0.86 0.82 Ecuador 1.02 0.07 0.72 0.99 0.53 0.67 Egypt 1.24 2.22 0.74 1.21 -0.04 0.71 El Salvador 0.90 n.a. 0.80 0.89 0.28 0.79 Finland 0.99 3.08 0.89 0.99 1.52 0.88 France 0.95 1.04 0.72 0.95 1.02 0.70 Germany, Fed. Rep. of 0.99 1.23 0.83 0.99 1.29 0.83 Greece 0.90 1.29 0.72 0.90 1.01 0.67 Guatemala 0.90 n.a. 0.82 0.91 1.00 0.80 India 0.77 -0.61 0.71 0.75 0.75 0.66 Ireland 1.02 0.98 0.86 1.03 1.28 0.84 Israel 1.01 0.84 0.78 1.00 0.83 0.67 Italy 0.92 0.68 0.74 0.92 1.47 0.69 Japan 0.89 -0.20 0.79 0.88 0.33 0.76 Kenya 1.07 0.80 0.83 1.08 1.00 0.82 Korea, Rep. of 0.91 1.32 0.74 0.90 -1.03 0.69 Malawi 0.90 -0.86 0.75 n.a. n.a. n.a. Mundlak and Larson 403 Initial Results Because the variables are in logs, their sample variations represent relative changes. Therefore, the variables have no units, and it is possible to pool the data over all commodities for all years. The estimates of equation 3 appear in the first three columns of table 1. The t ratios are all very high (double digits) and therefore are not reported here. The estimated transmission elasticity (from the world-prices column) varies between 0.74 and 1.24, with a median of 0.952. The values for 49 out of 57 countries fall in the range of 0.85 to 1.07. Thus the discrepancy from 1 is indeed very small. This indicates that the variations in world prices are almost fully transmitted to domestic prices. This is the answer to the first question. Turning to the second question, the values of R2 vary between 0.66 and 0.96, which Table 1. (continued) Nominal values Real values World Exchange World Exchange Country prices rates R2 prices rates R2 Malaysia 0.84 0.58 0.77 0.84 0.78 0.76 Mauritius 1.03 1.50 0.90 1.03 0.30 0.88 Mexico 1.02 0.56 0.79 1.02 1.04 0.75 Netherlands 0.98 1.09 0.76 0.98 1.09 0.76 New Zealand 1.03 0.21 0.76 1.04 0.90 0.74 Norway 0.98 1.23 0.89 0.97 1.11 0.88 Pakistan 0.82 0.44 0.76 0.79 --0.02 0.69 Panama 0.93 n.a. 0.77 0.95 0.84 0.75 Peru 0.86 0.98 0.77 0.85 0.58 0.65 Philippines 0.83 0.58 0.75 0.81 0.58 0.69 Portugal 0.96 1.02 0.79 0.97 1.15 0.76 South Africa 0.98 0.37 0.87 0.99 1.51 0.85 Spain 0.93 1.29 0.77 0.93 1.11 0.74 Sri Lanka 0.84 0.67 0.76 0.83 0.58 0.73 Sweden 0.95 2.35 0.82 0.95 1.87 0.82 Switzerland 1.01 0.68 0.74 1.01 0.68 0.73 Syria 0.98 4.79 0.76 0.97 0.86 0.72 Tanzania 0.97 1.41 0.81 0.98 1.16 0.79 Thailand 0.89 -0.86 0.80 0.89 0.75 0.78 Trinidad 1.01 1.04 0.66 1.01 1.05 0.63 Turkey 0.96 0.94 0.76 0.93 0.10 0.69 United Kingdom 0.96 0.75 0.89 0.96 1.37 0.88 United States 1.01 n.a. 0.82 1.02 0.00 0.81 Uruguay 0.81 0.98 0.94 0.79 0.97 0.72 Venezuela 0.94 5.13 0.71 0.92 -2.02 0.69 Yugoslavia 1.01 1.00 0.83 1.01 1.06 0.79 Zambia 0.89 1.56 0.86 0.89 1.56 0.84 Zimbabwe 0.97 1.69 0.88 0.96 1.55 0.87 n.a. Not applicable. Note: The values in the table are the estimated coefficients from a regression using data pooled over all commodities (the sample covered 60 products, which vary by country) and over all years (1968-78). Source: Authors' calculations, using data from FAO (various issues). 404 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 indicates that a high proportion of the variations in domestic prices are ac- counted for by the variations in world prices. There are no estimates of the exchange-rate elasticity for the United States or for Panama, Guatemala, and El Salvador, which used dollar-linked currencies. However, the values of the transmission elasticity for these countries are in line with those obtained for the other countries. The median of the exchange-rate elasticity, with these four countries ex- cluded, is 0.97, but the estimates vary greatly across countries. The variability in the exchange-rate elasticity reflects the problem of determining the appropriate measure of exchange rate for this analysis. In many countries this variable is volatile because of inflation and changes in exchange-rate regimes. When the exchange rate changes during the year, the rate that was applicable to a particu- lar commodity depends on the seasonality of that commodity and may differ from the variable used in the regression. A similar problem arises when there are multiple exchange rates, where whatever alternative is used represents a compromise. The effect of inflation on the results can be reduced by examining the identity in equation 2 in terms of the real exchange rate: (4) (Pit - PJ = (P2, - P2,) + (e, + pt- P) + Sit + ui, where Pt and Pt are the logs of domestic and world price deflators, respectively, and the terms in parentheses represent real values. The estimation of equation 3 is repeated, with the real values replacing the nominal values. We deflate the domestic and world prices by the domestic and U.S. consumer price index, respectively. The results appear in the last three columns of table 1. The results for the regressions in real and nominal values should be the same under H1 and H2. Indeed, the transmission elasticity is changed very slightly; its median value is 0.947. The median of the exchange-rate elasticity is 0.86, but for some countries the estimate differs significantly from the respective nominal-value regressions, and the cross-country variability still exists. Eliminating the effect of the exchange rate. Because we are interested largely in the transmission elasticity, it is desirable to eliminate the effect of the ex- change rate. We consider two options. First, we compute within-year regressions (that is, regressions with year dummies). Such regressions use the price differ- ences between commodities for each year, and those do not reflect the exchange rate. In this case, the regression equation takes the form (5) (Pit - P.) = c, + f3(p9t - P t) + eit where p, = ZApit/I, the time-price average over commodities, with I being the number of commodities. We use generic notations a, f, and e for the intercept, the coefficient of world prices, and the disturbance term, respectively, although their values are expected to vary from one equation to another. The prices in equation 5 are also real, but unlike those in equation 4 they are deflated by their Mundlak and Larson 405 own sample averages. In this case there is no difference between the nominal and real variables. If we let Pt be the consumer price index used to convert nominal to real values, then the real domestic price to be used in equation 5 is [(pi, - P,) - (P.-Pt)] = (pi - P.t), which is the nominal value. The same holds for real world prices. The results appear in the first column of table 2. The median value is 0.967, and on the whole the results are similar to those of the pooled regres- sion with exchange rates included. The second alternative is to express domestic prices in dollars: (6) (pi, - e) = °x + Opi + Eit. This approach was taken in an earlier version of this article, Mundlak and Larson (1990). The results of the estimation of equation 6 appear in the second column of table 2. They convey the same information as the previous regres- sions: the median value of the estimated transmission elasticity is 0.945. Estimation results with all countries pooled together. The world price is the export unit value and, as, such, it is not an average of domestic prices. After all, world trade constitutes only a small fraction of world production. To determine the extent to which the world price used here represents the domestic country price, the regression is estimated with all countries pooled together. In such an analysis the individual countries serve as repeated observations because they all face the same world price. The results are: 0.933 for R2, 0.941 for the transmis- sion elasticity, and 1.02 for the exchange-rate elasticity. The comparable esti- mates for real prices are 0.964, 0.943, and 0.980, respectively. It is thus con- cluded that the world prices used in the analysis are indeed representative of domestic prices. The Policy Bias The aforementioned regressions do not include a measure of the tax, s, as a variable because it is simply unobserved. This omission adds a component to the equation disturbance, and thus it reduces the R2, which measures the impor- tance of world prices in explaining the variations in domestic prices. More important, the omission may bias the transmission elasticity. It is often stated that countries pursue policies aimed at stabilizing domestic prices. Stabilization requires tax reductions when world prices are high and tax increases when world prices are low, which implies a negative correlation between world prices and taxes. Such a relationship is captured by (7) Si, = To + 7rp + v1, where v is the error of this equation and E(p*v) = 0. Combining equations 7 and 3, the regression equation for domestic prices is (8) Pi, = (a + To) + ( + T) P, + -e, + Lit where E = e + v. For convenience, we refer to ir as the policy elasticity. Under 406 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 2. Estimated Transmission Elasticities from Regressions Excluding the Exchange Rate, 1968-78 With domestic With commodity- Country With year dummies prices in dollars means' Argentina 0.990 0.966 1.000 Australia 0.933 0.930 0.944 Austria 0.984 0.979 1.007 Bangladesh 0.710 0.715 0.731 Belgium and Luxembourg 0.973 0.972 0.993 Brazil 0.853 0.902 0.871 Burundi 0.884 0.862 0.901 Cameroon 0.881 0.890 0.900 Canada 1.018 0.999 1.033 Chile 0.970 0.878 0.785 Colombia 0.944 0.922 0.972 Costa Rica 0.931 0.908 0.944 Cyprus 0.934 0.925 0.948 Denmark 1.025 1.037 1.049 Ecuador 1.012 0.987 1.036 Egypt 1.231 1.208 1.271 El Salvador 0.904 0.903 0.925 Finland 1.000 0.967 1.026 France 0.953 0.949 0.968 Germany, Fed. Rep. of 0.995 0.989 1.037 Greece 0.906 0.912 0.925 Guatemala 0.971 0.907 0.972 India 0.774 0.737 0.794 Ireland 1.030 1.022 1.045 Israel 1.012 0.972 0.989 Italy 0.940 0.909 0.957 Japan 0.888 0.942 0.914 Kenya 1.090 1.064 1.112 Korea, Rep. of 0.904 0.926 0.907 Malawi 0.923 0.888 0.950 equation 2, f = 1 and the policy contributes to a discrepancy from 1. If we attribute all of the discrepancy of the estimated elasticity from 1 to the policy, then a value of 0.95 for the transmission elasticity implies a value of -0.05 as an estimate for 7r, which is indeed very small. Equation 7 assumes a uniform policy for all commodities and all years. This assumption is too strong and should therefore be weakened by generalizing the equation. This can be done by allowing commodity-specific policy elasticity, denoted by 7ri, and a tax level, denoted by 7rt0: (7a) sit= toi + wi pi*t + Vit where E(p*v) = cov(iroi, pj't) = cov(ri, p1't) = 0 for all t. The effect of this extension can be evaluated through the computation of between-commodity and Mundlak and Larson 407 Table 2. (continued) With domestic With commodity- Country With year dummies prices in dollars meansa Malaysia 0.842 0.858 0.863 Mauritius 1.033 1.041 1.048 Mexico 1.040 0.985 1.058 Netherlands 0.989 0.985 1.016 New Zealand 1.051 1.029 1.068 Norway 0.976 0.977 1.006 Pakistan 0.804 0.744 0.829 Panama 0.970 0.937 0.971 Peru 0.852 0.868 0.902 Philippines 0.826 0.804 0.842 Portugal 0.970 0.959 0.982 South Africa 1.005 0.972 1.028 Spain 0.932 0.928 0.948 Sri Lanka 0.827 0.814 0.833 Sweden 0.955 0.930 0.986 Switzerland 1.018 1.039 1.043 Syria 0.977 0.978 1.002 Tanzania 0.989 0.977 1.012 Thailand 0.892 0.897 0.915 Trinidad 1.011 1.015 1.036 Turkey 0.943 0.952 0.961 United Kingdom 0.967 0.951 0.971 United States 0.958 1.005 0.958 Uruguay 0.800 0.796 0.809 Venezuela 0.933 0.910 0.963 Yugoslavia 1.020 1.011 1.041 Zambia 0.898 0.893 0.921 Zimbabwe 0.969 0.956 0.994 a. This column gives the values for the coefficient of world prices when using the between-commodity regression equation, in which the commodity price is an average over time and the world price is the only explanatory variable. Source: Authors' calculations, using data from FAO (various issues). within-commodity regressions. The appendix summarizes the formal relation- ships between the various estimators. Between-commodity regressions. Averaging the variables over time we obtain (9) p, = (a + 1ir0 + -ye) + (3 + iri) P, + A. where pi. = Yipit/ T, the commodity-price average over time, T is the number of years, and e = Xtet/ T. The exchange rate is thus subsumed into the intercept and the between-commodity regression has only the world price as an explana- tory variable. Its regression coefficient is (10) b(i) = >Zp1.(p* - P*) / (p* - ) 408 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Taking expectation using equation 9: (I1) Eb(i) = + A(i) and (12) a(i) = where Xi = (pZ _ p't )2 / y(p' - p.*)2 is the weight assigned to 7ri. Thus A(i) is a weighted average of the commodity-specific policy elasticity, iri. The values obtained for the coefficient of world prices, b(i), appear in the third column of table 2. The median value is 0.971, so that it differs very little from that of the pooled regression. This implies that the weighted average of the policy elas- ticities is quite small. Table 3. Estimated Transmission and Exchange Rate Elasticities from within-Commodity Regressions, 1968-78 Nominal values Real values World prices World Exchange World Exchange Country (U.S. dollars) prices rates prices rates Argentina 0.759 0.868 0.989 0.614 0.995 Australia 0.847 0.977 0.623 0.964 1.092 Austria 0.792 0.931 1.143 0.733 1.196 Bangladesh 0.630 1.094 0.604 0.911 0.938 Belgium and Luxembourg 0.828 0.921 1.145 0.750 1.154 Brazil 1.094 0.825 1.280 0.936 0.960 Burundi O.S79 0.58s 1.466 0.257 1.233 Cameroon 0.873 0.903 1.057 0.671 1.063 Canada 0.797 0.962 -0.152 0.894 0.595 Chile 0.600 0.938 0.981 -0.010 1.376 Colombia 0.648 0.665 1.555 0.319 1.242 Costa Rica 0.655 0.500 2.367 0.404 1.308 Cyprus 0.826 0.820 -0.131 0.479 -0.142 Denmark 0.948 1.007 1.010 0.944 1.074 Ecuador 0.719 0.833 1.225 0.548 1.136 Egypt 0.964 0.600 -1.314 0.317 -1.901 El Salvador 0.759 0.759 2.413 0.618 1.408 Finland 0.636 0.769 2.263 0.713 1.582 France 0.846 0.939 1.305 0.817 1.255 Germany, Fed. Rep. of 0.748 0.979 1.252 0.935 1.369 Greece 0.845 0.746 1.451 0.667 1.164 Guatemala 0.697 0.972 0.000 0.869 1.350 India 0.437 0.272 2.783 0.218 1.487 Ireland 0.806 0.971 0.892 0.951 0.947 Israel 0.767 1.025 0.831 0.643 1.130 Italy 0.688 0.620 1.349 0.448 1.158 Japan 1.144 1.205 0.905 1.107 1.072 Kenya 0.750 0.634 1.739 0.532 1.107 Korea, Rep. of 1.006 0.846 1.190 0.767 1.106 Malawi 0.488 0.829 -1.162 n.a. n.a. Mundlak and Larson 409 Within-commodity regressions. The basic underlying equation for the within-commodity regression is obtained by subtracting equation 9 from equa- tion 8, with roi and iri replacing ?rO and ir, respectively: (13) Pi, = (Pi + 7ri) (p*-pZ ) + y (et- e) + ,it-i '.. The within-commodity estimates were derived for the nominal and real versions with the exchange rate included and for the domestic variables in dollar prices. The results appear in table 3. The median of the transmission elasticities for the nominal prices, 0.937, is significantly higher than the corresponding values of 0.78 and 0.713 for the dollar prices and real prices, respectively. To examine the source of this difference in the estimates, we simplify the Table 3. (continued) Nominal values Real values World prices World Exchange World Exchange Country (U.S. dollars) prices rates prices rates Malaysia 0.837 1.010 1.291 1.008 1.430 Mauritius 0.989 0.822 1.503 0.868 1.052 Mexico 0.646 0.904 1.091 0.452 1.131 Netherlands 0.819 0.992 1.208 0.973 1.245 New Zealand 0.764 0.939 0.207 0.883 0.958 Norway 0.801 0.974 1.381 0.880 1.388 Pakistan 0.367 1.285 0.144 0.531 1.287 Panama 0.604 0.969 0.000 0.633 -1.527 Peru 0.782 0.931 1.032 0.516 1.071 Philippines 0.597 1.078 0.578 0.465 1.222 Portugal 0.800 0.820 1.351 0.663 1.165 South Africa 0.626 0.922 0.325 0.721 1.972 Spain 0.816 0.822 1.227 0.682 1.108 Sri Lanka 0.686 1.205 0.367 1.004 0.895 Sweden 0.579 0.775 1.898 0.493 1.588 Switzerland 1.054 1.099 0.844 1.209 0.945 Syria 0.872 0.843 1.633 0.764 1.219 Tanzania 0.765 0.635 1.745 0.516 1.094 Thailand 0.769 0.781 1.206 0.506 1.087 Trinidad 0.866 0.887 1.784 0.771 1.272 Turkey 0.904 1.046 0.966 0.810 1.094 United Kingdom 0.781 0.943 0.671 0.864 0.518 United States 0.817 0.955 0.000 0.860 0.000 Uruguay 0.730 0.893 0.953 0.594 1.033 Venezuela 0.599 0.669 2.186 0.202 1.758 Yugoslavia 0.851 0.549 1.855 0.657 1.089 Zambia 0.713 1.003 0.988 0.736 1.940 Zimbabwe 0.697 0.969 1.280 0.832 1.191 n.a. Not applicable. Source: Authors' calculations, using data from FAO (various issues). 410 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 exposition by ignoring the term with the exchange rate and writing the regres- sion coefficient for commodity i as if it were a simple regression:' (14) bi = ytP.t(P?t - P,:) / ( - p,.)2 and express the within-commodity estimator, w(i), as (15) w(i) = >p,it(p2* - p*-) / ZE(p4 -P*) = Zibicii where o, = t(P, - pdi)2 / t2(pit - pfl2. Thus the within-commodity estimator is a weighted average of the regression coefficients for commodity i. Taking expectation using equation 13: (16) Ew(i) = + Ei Ziri. The difference between equations 16 and 11 is in the way the commodity- specific slopes are taken into account. In equation 11 only one observation per commodity is used, and the exchange rate is eliminated from the equation. In equation 16 there are as many observations as years. If the difference of the transmission elasticities from 1 is considered as a weighted average of the commodity-specific policy elasticity, then the results suggest some variability in policy elasticity among commodities. We therefore examine this possibility more closely by presenting results for individual commodities and discussing possible sources of variations between the various estimates. A similar analysis can be conducted for an alternative specification that allows for systematic variations of policy over time. This is discussed in Mundlak and Larson (1990), where it is shown that at the median the time effect on the policy elasticity, derived for domestic dollar prices, is -0.04. It seems that the vari- ability over commodities is more important. To deal with the variability over time, a longer time series is needed. Specific Commodities Mundlak and Larson (1990) present transmission elasticities for wheat, cof- fee, and cocoa derived from equation 6, where the domestic prices are measured in-dollars. The results are reproduced here in table 4. Wheat is chosen because it is often stated that staple foods are more susceptible to intervention that insu- lates domestic markets from world prices. Coffee and cocoa are internationally traded under cartel arrangements, and as such they may show a larger gap in the variations of domestic and world prices. The median value of the transmission elasticity for wheat is approximately 0.65, with only 9 out of the 58 countries having a coefficient smaller than 0.5. 1. In practice b, is obtained from the following regression: P,, = a, + bip,, + c,e, + error. Equations 14 and 15 still apply when the prices are netted of the linear effect of e,. This adjustment does not affect the interpretation. Mundlak and Larson 411 The median value for coffee is 0.68; for cocoa it is above 0.84. The reason for concentrating on the estimates of equation 6 is that, with only 11 observations and a correlation between exchange rates and world prices, the results of the full equation 3 with an additional coefficient are less stable. This problem is over- come by pooling all countries together. The estimates of equation 3 for wheat with country-pooled data are 0.69, 1.03, and 0.969 for the transmission elastic- ity, exchange rate elasticity, and R2, respectively. Interestingly, the value of the transmission elasticity is very close to the median value of table 4. The policy elasticities for wheat, coffee, and cocoa are negative and larger in absolute value than those obtained for the pool of commodities. Still, for most countries, about 70 to 80 percent of the variations in world prices, depending on the commodity, are transmitted to domestic prices. Furthermore, the values of R2 are on the whole quite high, indicating that world prices are the main source of variations in domestic prices. The relationships between the within- and between-commodity estimates are illustrated in figure 1, where domestic dollar prices, measured in natural logs, are plotted against the log of world prices. Lines ac and bc represent regression lines with slopes f3 + 7ra and 03 + lrb fitted to observations on two commodities, where both slopes are less than 1. Ellipsoids mark the clusters of observations Figure 1. Relationship between Within-Commodity and Between-Commodity Estimates Log of domestic Between-commodities price in dollars (p) regression line commnodity a 'Witiin-commodity a regression line b Widtin-commodity b regression line C Observations on - comrmodity b Log of world price in dollars (p) 412 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 4. Estimated Transmission Elasticities for Wheat, Coffee, and Cocoa, 1 968-78 Wheat Coffee Cocoa Elastic- Elastic- Elastic- Country ity, b R2 ity, b R2 ity, b R2 Argentina 0.701 0.46 0.00 0.00 0.00 0.00 Australia 0.905 0.76 n.a. n.a. n.a. n.a. Austria 0.588 0.85 n.a. n.a. n.a. n.a. Bangladesh 0.655 0.37 n.a. n.a. n.a. n.a. Belgium and Luxembourg 0.626 0.81 n.a. n.a. n.a. n.a. Brazil 0.814 0.88 0.652 0.22 1.192 0.85 Burundi 0.518 0.31 0.680 0.99 n.a. n.a. Cameroon 0.152 0.68 0.530 0.88 0.615 0.86 Canada 0.954 0.66 n.a. n.a. n.a. n.a. Chile 0.836 0.29 n.a. n.a. n.a. n.a. Colombia 0.620 0.85 0.619 0.95 0.612 0.91 Costa Rica 0.554 0.93 0.940 0.99 1.075 0.94 Cyprus 0.477 0.86 n.a. n.a. n.a. n.a. Denmark 0.892 0.87 n.a. n.a. n.a. n.a. Ecuador 0.529 0.82 0.629 0.78 0.979 0.89 Egypt 0.562 0.86 n.a. n.a. n.a. n.a. El Salvador 0.621 0.78 1.054 0.93 0.927 0.69 Finland 0.413 0.83 n.a. n.a. n.a. n.a. France 0.582 0.85 n.a. n.a. n.a. n.a. Germany, Fed. Rep. of 0.646 0.87 n.a. n.a. n.a. n.a. Greece 0.715 0.83 n.a. n.a. n.a. n.a. Guatemala 0.687 0.83 0.862 0.96 0.975 0.91 India 0.405 0.85 0.142 0.39 n.a. n.a. Ireland 0.707 0.79 n.a. n.a. n.a. n.a. Israel 0.821 0.96 n.a. n.a. n.a. n.a. Italy 0.655 0.90 n.a. n.a. n.a. n.a. Japan 1.113 0.71 n.a. n.a. n.a. n.a. Kenya 0.780 0.81 1.006 0.99 n.a. n.a. Korea, Rep. of 0.903 0.81 n.a. n.a. n.a. n.a. Malawi 0.500 0.66 0.430 0.75 n.a. n.a. for the sample years for each commodity. The intercept of the price line of an individual commodity indicates a systematic difference between domestic and world prices or simply the level of distortion for the particular commodity. Thus we can have a slope of 1 and an intercept larger or smaller than 0, indicating a protection or tax, respectively. The lines are drawn with a common intercept so as to keep the same distortion rate. This is not essential, and other configura- tions are admissible. The commodity-specific regressions estimate the slopes of these lines, and the within-commodity regression provides estimates of a weighted average of these commodity slopes. However, the between-commodity regression line is a statis- tical fit to commodity averages, labeled p, and, unlike the within-estimator, the between-commodity estimator does not use the information represented in indi- Mundlak and Larson 413 Table 4. (continued) Wheat Coffee Cocoa Elastic- Elastic- Elastic- Country ity, b R2 ity, b R2 ity, b R2 Malaysia 1.008 0.97 0.837 0.76 0.844 0.94 Mauritius 0.692 0.82 n.a. n.a. n.a. n.a. Mexico 0.586 0.88 0.858 0.80 0.835 0.88 Netherlands 0.584 0.84 n.a. n.a. n.a. n.a. New Zealand 0.701 0.78 n.a. n.a. n.a. n.a. Norway 0.601 0.82 n.a. n.a. n.a. n.a. Pakistan 0.097 0.08 n.a. n.a. n.a. n.a. Panama 0.497 0.77 0.425 0.79 1.023 0.98 Peru 0.704 0.86 0.732 0.83 1.046 0.85 Philippines 0.609 0.85 1.018 0.87 0.754 0.92 Portugal 0.422 0.93 n.a. n.a. n.a. n.a. South Africa 0.454 0.93 n.a. n.a. n.a. n.a. Spain 0.546 0.90 n.a. n.a. n.a. n.a. Sri Lanka 0.586 0.64 0.809 0.71 1.085 0.82 Sweden 0.482 0.84 n.a. n.a. n.a. n.a. Switzerland 0.910 0.85 n.a. n.a. n.a. n.a. Syria 0.687 0.84 n.a. n.a. n.a. n.a. Tanzania 0.634 0.69 0.616 0.50 0.498 0.79 Thailand 0.995 0.84 0.461 0.86 n.a. n.a. Trinidad 0.729 0.81 0.604 0.75 0.702 0.97 Turkey 0.705 0.77 n.a. n.a. n.a. n.a. United Kingdom 0.708 0.77 n.a. n.a. n.a. n.a. United States 0.958 0.73 0,831 0.86 n.a. n.a. Uruguay 1.153 0.88 n.a. n.a. n.a. n.a. Venezuela 0.805 0.84 0.051 0.09 0.504 0.62 Yugoslavia 0.626 0.84 n.a. n.a. n.a. n.a. Zambia 1.187 0.96 0.715 0.52 n.a. n.a. Zimbabwe 0.624 0.84 0.434 0.75 n.a. n.a. n.a. Indicates not applicable. Source: Authors' calculations, using data from FAO (various issues). vidual commodity observations. Because there is a large between-commodity spread in the prices, the slope of the between-commodity regression differs from the slopes of the individual com- modities. An elasticity of 1 for the between-commodity estimate indicates that what is relatively expensive in the world market is also relatively expensive at home, or, more specifically, that the relative prices at home and abroad are, on average, the same. A slope smaller than 1 indicates that the more expensive the commodity is, the lower the tax rate or the larger the subsidy is. Decomposition of the Pooled Regression Results The pooled regression is a weighted average of the within- and between- commodity regressions (see the appendix), where the weights depend on the 414 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 variance components of the world prices. Table 5 presents a decomposition of the variations in world prices to the commodity and time components. The between-commodity variations dominate the within-commodity variations, and therefore estimates based on data pooled across time and commodities (first three columns of table 1) largely reflect the between-commodity variations. Thus the transmission elasticity of the pooled regression is close to the between- commodity value and close to 1 even though the commodity elasticities are smaller than 1. A slope smaller than 1 for an individual commodity in this framework is consistent with a stabilization policy implemented by changing tax rates. But is a deviation from 1 for an individual commodity an exclusive outcome of policy? The answer is probably no. There are two important effects that are likely to Table 5. Sum of Squares of World Prices, 1968-78 Within (4) Between (2) (3) Commodities (S) (6) (1) Commodities Time and time Commodities Time Country (Total) (i) (t) (it) (i) (t) Argentina 562.8 81.3 493.0 11.5 481.5 69.8 Australia 511.0 75.1 446.6 10.6 436.0 64.5 Austria 426.4 57.2 373.2 7.7 370.4 53.4 Bangladesh 401.2 62.9 349.3 11.0 338.3 51.9 Belgium and Luxembourg 414.7 58.9 358.6 9.2 357.6 56.8 Brazil 576.0 91.8 499.3 17.5 483.6 77.0 Burundi 385.4 46.1 346.8 7.5 339.3 38.6 Cameroon 475.8 63.6 418.7 13.0 413.0 57.9 Canada 370.0 53.1 323.6 6.6 316.9 46.4 Chile 386.1 68.8 325.1 9.8 318.7 61.1 Colombia 538.3 82.3 469.5 13.5 456.1 68.8 Costa Rica 442.4 59.2 390.4 11.5 384.1 52.4 Cyprus 363.2 56.0 310.7 7.8 309.8 53.0 Denmark 297.1 48.5 249.5 7.9 250.9 48.3 Ecuador 542.9 83.4 473.6 14.1 459.5 69.3 Egypt 369.1 75.3 304.4 10.6 293.8 64.7 El Salvador 457.3 64.7 401.3 12.5 393.2 56.5 Finland 273.2 39.6 235.3 6.2 235.0 38.4 France 459.2 69.2 399.2 9.9 388.7 60.1 Germany, Fed. Rep. of 429.8 55.6 375.0 10.3 371.9 56.7 Greece 507.6 78.5 439.7 10.9 427.6 68.0 Guatemala 437.6 58.9 386.1 11.2 379.2 51.9 India 519.8 84.0 449.0 15.1 436.6 70.9 Ireland 296.2 35.1 268.1 5.8 268.5 28.1 Israel 404.6 72.5 338.3 10.9 330.5 67.1 Italy 453.3 84.5 377.8 11.0 370.0 75.6 Japan 562.8 78.4 494.6 12.6 486.5 68.3 Kenya 430.2 70.3 471.1 12.9 459.8 59.2 Korea, Rep. of 482.4 63.0 425.1 7.8 424.7 57.7 Malawi 375.8 50.1 334.7 9.0 325.8 41.1 Mundlak and Larson 415 bias the estimates downward: tradability and "measurement error." Tradability. Although agricultural products are largely tradable, their domes- tic prices also reflect domestic inputs such as marketing, finance, storage, and transportation. (For an analysis of this subject, see Mundlak, Cavallo, and Domenech 1990.) To incorporate this extension, equation 2 is rewritten: (17) Pit = Ti(p,` + et + si, + uit) + (1 - )pd where pd is the natural log of the aggregate price of the domestic input, assumed to be the same for all products, and Ti is the share of the tradable component in the price of commodity i. Under equation 17, the slope of the individual com- modity price line is smaller than 1. The empirical transmission elasticity is now an estimate of ri + iri. As pd is omitted from the regression, this estimate is Table 5. (continued) Within (4) Between (2) (3) Commodities (5) (6) (1) Commodities Time and time Commodities Time Country (Total) (i) (t) (it) (i) (t) Malaysia 446.4 58.2 394.3 13.1 390.3 53.0 Mauritius 301.9 36.0 272.1 6.3 265.9 29.8 Mexico 522.9 92.6 440.5 14.8 433.2 82.8 Netherlands 322.3 56.1 269.1 7.5 268.7 53.6 New Zealand 400.1 51.5 355.3 6.7 348.6 44.8 Norway 274.5 38.8 236.1 6.9 234.6 39.6 Pakistan 411.6 75.2 346.2 11.6 337.2 65.5 Panama 320.5 40.2 286.7 8.4 280.5 34.0 Peru 556.7 90.0 483.0 16.9 462.6 74.3 Philippines 445.0 65.4 388.9 13.2 377.8 57.1 Portugal 516.6 69.9 453.9 12.2 448.0 63.2 South Africa 525.9 72.8 465.1 12.1 453.1 60.7 Spain 568.2 85.8 493.2 13.6 483.0 75.4 Sri Lanka 443.6 57.3 396.0 9.7 386.3 46.7 Sweden 365.1 53.2 310.5 10.7 315.1 56.4 Switzerland 351.6 46.5 310.8 6.4 301.3 41.0 Syria 397.4 71.8 335.9 9.7 325.6 61.7 Tanzania 535.0 78.2 469.3 14.5 457.3 65.8 Thailand 404.4 58.5 357.6 11.9 348.3 47.3 Trinidad 388.6 47.6 345.3 10.9 341.8 44.1 Turkey 455.6 78.6 387.2 10.2 377.0 68.4 United Kingdom 341.8 45.9 298.6 8.2 292.1 44.2 Uruguay 383.0 65.0 326.5 8.5 318.0 56.5 United States 528.0 81.5 457.8 13.2 445.8 70.5 Venezuela 420.8 64.3 367.0 11.3 355.0 53.9 Yugoslavia 488.1 76.7 420.6 9.2 411.4 67.5 Zambia 400.9 47.8 363.6 7.6 359.7 37.3 Zimbabwe 426.0 54.1 378.8 8.8 372.0 47.3 Source: Authors' calculations, using data from FAO (various issues). 416 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 biased. It is likely that the omitted variable, which is closely related to domestic inflation, is positively correlated with the exchange rate and that hence the bias is positive. Therefore, the deviation from 1 cannot be fully attributed to policy. We can carry this analysis a step further and rewrite equation 17 as (18) p,t - p = r,(p t + et + sit + uit) - Trptd. If we assume that the price of the domestic input can be approximated by the overall price level, then the dependent variable is real domestic price in the sense of equation 4. The difference is that in equation 18 the world price is nominal, whereas in equation 4 it is real, as well. If we ignore this difference, however, the estimates of equation 4 can be viewed as an approximation of the estimate of equation 18, with the last term omitted. But now the bias is negative, because the coefficient of ptd has a negative sign. Indeed, the estimates of the real regres- sions give somewhat smaller estimates for the transmission elasticities. The values of the within-commodity estimates are 0.937 for the nominal, 0.78 for the dollar prices, and 0.713 for the real. It is interesting to note that the effect of converting to dollars is similar to that of deflating by the overall price level. In terms of our discussion, the nominal estimate is biased upward, and the real is biased downward. Hence the difference from 1 obtained from the regression with real values can be viewed as an upper bound for an estimate of the sum of the share of the nontradable component and the policy elasticity, whereas the nominal regression provides a lower bound. Measurement error. There are two good reasons to think about measurement error of a conceptual rather than mechanical nature. First, the basic equation 2 is applicable at a time when a trade takes place. Trade is not carried out continu- ously, however. Between trades, the world price is changing without necessarily affecting domestic prices. Stored commodities maintain an intertemporal arbi- trage condition. As such, the spot prices respond to new information with respect to expected future world supply. Because arbitrage, either through trade or through storage, is costly, domestic prices, which are not backed by transac- tions, do not respond instantaneously to changes in world prices. Consequently, as illustrated effectively by Williams and Wright (1991), the dynamic paths of world prices and domestic prices within the year are likely to differ. When intrayear variations in the world price are summarized in equation 2 by a single figure, a discrepancy is built in between the domestic price and the pertinent world price. This is, of course, a short-term phenomenon, but it recurs with every new shock to the system. Because the prices are dated, this dynamic may matter and thus affect the results. Second, the problem of deciding on the right deflators to convert the world prices from nominal to real is similar to the question of what exchange rate to use, which was discussed above. The deflators and exchange rates may bias the estimates downward. The bias may be substantial and may lead to a rejection of Mundlak and Larson 417 the empirical validity of the law of one price. Also, in some countries (for example, Canada) the estimates of the exchange rate elasticities change very little over the sample period. Therefore the spread is not sufficient to get reliable estimates. IV. THE DATA SET FOR THE EUROPEAN COMMUNITY A potential problem of any empirical application is that the results emerging from the study reflect the idiosyncrasies of the way in which the data are col- lected, estimated, or reported rather than the underlying economic effects. This possibility carries a special weight in view of the doubt researchers express with respect to FAO data. Therefore we repeat the analysis on a separate data set for the European Community (EC), which has had an active agricultural policy as well as good data. The EC agricultural policy is well financed and sophisticated in its execution and reporting mechanisms. Because it is well financed, any wedge between domestic and international prices can be expected to be longer lived than in lower-income countries. The data are taken from Herlihy and others (1989), and cover 25 years of producer prices. The commodity coverage available from the Ec data is more limited than in the FAO data set, but it contains the major staple products. Table 6 presents some summary results. For the sample pooled across time and commodities, equation 3, the transmission elasticity varies be- tween 0.91 and 1.01, and the values of R2 vary between 0.84 and 0.92. The estimated exchange-rate elasticity, not reported in the table, varies between 0.79 and 1.06. The corresponding values for the pool of all the EC countries are 0.97, 0.97, and 1.03, respectively. The results for the between-commodities regres- sions are similar. The within-commodity estimates of the transmission elasticity (equation 5) are somewhat smaller, with a median value of 0.74, compared with a median value of 0.96 for the between-commodity regression. This pattern is similar to what we observed above for the first sample. It is also similar to the results reported in Mundlak and Larson (1990) for equation 6, with domestic prices measured in dollars. We use this similarity to report in table 7 the results in Mundlak and Larson (1990) for individual commodities, with dollar domestic prices. The results for the EC confirm the earlier results. Although the commodity coverage is different, the pooled elasticities for countries common to both sam- ples are remarkably close. The estimates for wheat vary between 0.54 and 0.91, with a median at 0.70 and a value of 0.77 for the pool of all the EC countries. Recall that the median for the country estimates in table 4 is 0.66, and the estimates for the pooled country data derived from equation 3 is 0.69. For some of the other commodities, the median elasticities are also somewhat lower than the aggregates. But for milk the estimated transmission elasticity is larger than 1 Table 6. Estimated Transmission Elasticities, European Community, 1960-85 Between Within Pooled Commodities Time Commodities Time Elasticity, R2 Elasticity, Elasticity, Elasticity, Elasticity, Country b b(i) R2 b(t) R2 w(i) R2 w(t) R2 Belgium and Luxembourg 0.98 0.84 1.00 1.00 0.76 0.94 0.74 0.99 0.99 1.00 Denmark 0.96 0.88 0.95 1.00 1.09 0.98 1.05 0.98 0.94 1.00 France 1.01 0.88 1.04 1.00 0.82 0.98 0.78 0.99 1.03 0.99 x Germany, Fed. Rep. of 0.96 0.85 0.97 0.99 0.54 0.88 0.47 0.99 0.96 0.99 Greece 1.00 0.92 1.02 1.00 0.68 0.99 0.65 0.99 1.02 1.00 Ireland 0.91 0.84 0.91 0.99 0.97 0.98 0.92 0.98 0.90 0.99 Italy 1.00 0.89 1.02 1.00 0.81 0.98 0.74 0.98 1.01 1.00 Netherlands 0.94 0.84 0.94 0.99 0.76 0.92 0.69 0.99 0.94 0.99 United Kingdom 0.95 0.90 0.96 0.99 0.90 0.95 0.86 0.98 0.96 0.99 All countries 0.97 0.97 1.02 1.00 0.89 0.98 0.85 0.99 1.00 1.00 Note: Exchange rate effects were included in estimates. Source: Authors' calculations, based on data from Herlihy and others (1989). Mundlak and Larson 419 for several countries, indicating a strong adjustment of domestic prices that was positively correlated with world prices. V. CONCLUSIONS By way of generalization, the deviation from unitary elasticity is, on the whole, surprisingly small. The deviation from unitary elasticity is in part the result of policy measures and in part the result of domestic inputs that are not necessarily synchronized with world agricultural prices. This does not imply that policies generated with respect to particular products are not important in affecting the prices of these products. They certainly affect the price levels, and, whenever a country taxes agriculture, the domestic prices will differ from world prices. Consequently, there are cross-country variations of prices. Such policies do not, however, prevent domestic prices from moving with world prices. Fur- thermore, world prices are the major contributor to variations in domestic prices. In this analysis it was assumed that the world price is independent of the disturbances in the price equation. On the face of it, this assumption might be too strong for the United States, and perhaps some other countries, when deal- ing with some specific commodities. If this assumption were violated, world price would be endogenous and the estimates would be subject to least-squares bias. However, this is not reflected in the results in any meaningful way. What, then, is the role for domestic supply and demand? They determine the traded quantities of the traded goods, and the prices of the traded goods affect to a large extent the prices of the specific factors in agriculture and thereby the supply of the nontraded goods. This is basically the mechanism of factor-price equalization. For instance, depressed world prices affect land prices, agricultural wage rates (through their effect on labor supply), and the price of quasi-fixed inputs. This spreads to all commodities. An important implication for thinking about the dynamics of world agricul- ture (Mundlak 1989) is that we can think of the world as a closed economy facing a downward-sloping demand function that serves as a constraint to pro- duction growth. The trend in world prices is determined by the relative growth in world supply and demand. In this century supply has outpaced demand, and as a result real world agricultural prices have declined. The essence of our analysis is that such a decline should have taken place in all countries, regardless of whether their supply actually increased in relation to demand. This implies that technical change and other permanent shocks that originate in one country but that are big enough to affect world prices eventually affect prices in all countries. Even though domestic policies affect prices, they cannot prevent the covariations of domestic prices with world prices in the long run, because price distortion is costly, and public resources, like private resources, are finite. Passive countries, which are shock takers, should implement the necessary structural adjustments called for by the shock-including the enhance- 420 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 7. Estimated Transmission Elasticities for Selected Commodities, European Community, 1960-85 Belgzum and Germany, Luxembourg Denmark France Fed. Rep. of Greece Elas- Elas- Elas- Elas- Elas- ticity, ticity, ticity, ticity, ticzty, Commodity b R2 b R2 b R2 b R2 b R2 Barley 0.714 0.873 0.972 0.940 0.691 0.865 0.684 0.858 0.632 0.873 Butter 0.522 0.822 1.101 0.930 0.607 0.915 0.736 0.907 0.716 0.916 Cattle 0.958 0.963 1.193 0.972 0.889 0.960 0.987 0.959 0.829 0.957 Cheese 0.921 0.947 1.287 0.976 0.844 0.972 1.022 0.989 0.773 0.920 Eggs 0.636 0.662 0.919 0.921 1.198 0.917 0.787 0.921 0.739 0.807 Maize n.a. n.a. n.a. n.a. 0.683 0.877 n.a. n.a. n.a. n.a. Milk 1.188 0.895 1.731 0.904 1.230 0.895 1.395 0.916 0.669 0.907 Oats 0.778 0.901 0.998 0.944 0.716 0.900 0.753 0.909 0.747 0.861 Pigs 0.722 0.938 0.819 0.956 0.571 0.910 0.754 0.950 n.a. n.a. Poultry 0.955 0.953 0.950 0.928 0.604 0.716 0.880 0.927 0.411 0.730 Potatoes 1.079 0.689 1.105 0.769 0.906 0.650 0.901 0.851 0.755 0.814 Rye 0.844 0.911 0.912 0.958 0.697 0.914 0.822 0.914 n.a. n.a. Sugar beets 0.741 0.712 0.885 0.825 0.798 0.777 0.718 0.766 0.645 0.649 Wheat 0.661 0.811 0.907 0.910 0.619 0.844 0.723 0.841 0.537 0.727 n.a. Not applicable. Source: Authors' calculations, based on data from Herlihy and others (1989). ment of technical change, if this is the source of the shock-rather than delay the process through taxation. This is certainly a very general statement, and it has to be properly interpreted when it comes to a particular policy; however, it is mentioned here in order to place possible implications of the analysis within a broader framework. APPENDIX. A SUMMARY OF THE FORMAL RELATIONS BETWEEN THE VARIOUS ESTIMATORS The analysis in the text differs somewhat from more familiar forms of panel data analysis. It is therefore useful to evaluate the results within a uniform framework. Let W, B(i), W(i), and W(it) be projection (symmetric and idempo- tent) matrixes that generate residuals. They can be defined in terms of their operation on an arbitrary vector, x, of order IT: Wx = (xi, - x), B(i)x = (xi - x..), B(t)x = (x1 - x), W(i)x = (xi - xi), W(t)x = (xi, - x), and W(it)x = (xit -X - X.t + X). The parentheses contain the typical elements of the vectors in question. The following identities can then be derived. (A-1) W = W(i) + B(i) (A-2) = W(t) + B(t) (A-3) = W(i) + W(t) - W(it) (A-4) = B(i) + B(t) + W(it). Mundlak and Larson 421 Table 7. (continued) Ireland Italy Netherlands United Kingdom All countries Elas- Elas- Elas- Elas- Elas- ticity, ticity, ticity, ticity, ticity, b R2 b R2 b R2 b R2 b R2 0.878 0.887 0.718 0.865 0.762 0.916 0.869 0.870 0.769 0.774 n.a. n.a. 0.654 0.843 0.880 0.804 1.072 0.883 0.786 0.733 1.105 0.949 0.873 0.966 0.911 0.928 1.023 0.943 0.974 0.849 n a. na. 0.837 0.343 1.040 0.976 1.116 0.900 0.980 0.671 0.785 0.883 0.548 0.793 0.762 0.874 0.634 0.612 0.779 0.633 n.a. n.a. 0.854 0.866 n.a. n.a. n.a. n.a. 0.769 0.851 1.551 0.974 1.393 0.903 1.312 0.910 1.173 0.972 1.294 0.818 0.845 0.868 0.773 0.878 0.719 0.924 0.819 0.898 0.794 0.801 0.767 0.898 0.642 0.908 0.733 0.933 0.735 0.892 0.718 0.860 0.748 0.669 0.534 0.840 0.937 0.946 0.913 0.727 0.770 0.680 1.018 0.718 0.950 0.733 1.016 0.757 0.923 0.719 0.961 0.641 n.a. n.a. 0.624 0.807 0.773 0.915 1.052 0.886 0.848 0.790 n.a. n.a. 0.774 0.703 0.790 0.744 0.795 0.562 0.818 0.785 0.698 0.811 0.637 0.827 0.674 0.833 0.898 0.885 0.768 0.688 If p and p° are the vectors of the two prices, then the regression coefficients of p or p can be presented in terms of a = p*Ap/p*Ap*. When A = W, B(i), B(t), the resulting estimators are b (pooled), b(i) (between commodity), and b(t) (between time), respectively. Also, when A W(i), W(t), and W(it), the coeffi- cients are referred to as within commodity (w[i]), within time (w[t]), and within time and commodity w( [it]), respectively. We can then decompose the pooled regression coefficient: (A-S) b = p -Wp/p*Wp* = Ow(i) + (1 - O)b(i) where 0 = p*W(i)p*lp*Wp* is the ratio of the within-commodity sum of squares and the total sums of squares, and the complement is a similar ratio for the between-commodity sum of squares. Table 5 presents a decomposition of the sum of squares of p- by sources. Because p* is the world price, the sums of squares should be the same for all countries. However, the set of commodities analyzed varies somewhat among countries, and therefore the numbers in the table differ accordingly. It is clear that the variations among commodities are greater by far than the variations over time; therefore, the pooled regression is closer to the between-commodity regression. A similar comparison can be made for the other estimators. Also, the results are easily generalized to multiple regressions, where the weights will be matrix, rather than scalar, weights (Mundlak 1978). 422 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Ahmed, Raisuddin. 1988. "Pricing Principles and Public Intervention in Domestic Mar- kets." In John Mellor and Raisuddin Ahmed, eds., Agricultural Price Policy for Devel- oping Countries. Baltimore, Md: The Johns Hopkins University Press. Anderson, Kym, Yujiro Hayami, and Masayoshi Honma. 1986. "Growth of Agricultural Protection." In Kym Anderson and others, eds., The Political Economy of Agricultural Protection: The Experience of East Asia. Sydney, Australia: Allen & Unwin. Australia, Bureau of Agricultural Economics. 1985. Agricultural Policies in the Eu- ropean Community: Their Origin, Nature, and Effects on Production and Trade. Policy Monograph 2. Canberra: Australian Government Publishing Service. Bale, Malcolm D., and Ernst Lutz. 1981. "Price Distortions in Agriculture and Their Effects: An International Comparison." American Journal of Agricultural Economics 63 (1): 8-22. FAO (Food and Agriculture Organization of the United Nations). Various issues. FAO Production Yearbook. Rome. Herlihy, Michael, Stephen Magiera, Richard Henry, and Kenneth Baily. 1989. Agri- cultural Statistics of the European Community, 1960-85. Statistical Bulletin 770. Washington, D.C.: U.S. Department of Agriculture. Johnson, D. Gale. 1973. World Agriculture in Disarray. New York: Macmillan. McCalla, Alex F. 1969. "Protectionism in International Agricultural Trade, 1850-1968." Agricultural History 43 (3, July): 329-44. Mellor, John W., and Raisuddin Ahmed, eds. 1988. Agricultural Price Policy for Devel- oping Countries. Baltimore, Md.: The John Hopkins University Press. Mundlak, Yair. 1978. "On the Pooling of Time-Series and Cross-Section Data." Econo- metrica 46 (1, January): 69-86. 1989. "Agricultural Growth and World Developments." In Alan Maunder and Alberto Valdes, eds., Agriculture and Governments in an Interdependent World. Pro- ceedings of the Twentieth International Conference of Agricultural Economists, Dartmouth, Aldershot, England. Mundlak, Yair, Domingo Cavallo, and Roberto Domenech. 1990. "Effects of Macro- economic Policies on Sectoral Prices." The World Bank Economic Review 4 (1): 55- 79. Mundlak, Yair, and Donald E Larson. 1990. "On the Relevance of World Agricultural Prices." wps 383. World Bank, International Economics Department, Washington, D.C. Processed. Valdes, Alberto, and Ammar Siamwalla. 1988. "Foreign Trade Regime, Exchange Rate Policy, and the Structure of Incentives." In John Mellor and Raisuddin Ahmed, eds., Agricultural Price Policy for Developing Countries. Baltimore, Md.: The Johns Hop- kins University Press. Williams, Jeffrey C., and Brian D. Wright. 1991. Storage and Commodity Markets. Cambridge, U.K.: Cambridge University Press. World Bank. 1986. World Development Report 1986. New York: Oxford University Press. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 423-438 How Small Enterprises in Ghana Have Responded to Adjustment William F. Steel and Leila M. Webster Monitoring of adjustment has focused on larger, often state-owned enterprises, many of which have been adversely affected by the more competitive environment. Little is known, however, about the impact of adjustment policies on small firms. Firm-level data from Ghana show that the adjustment process was well under way across sectors and within firms and that new investment was taking place. The adjustment process had strained most firms' operations-profits were squeezed between rising input costs and weak domestic demand and low-cost competing imports. Small-scale industries were forced to become more competitive to survive. Interviews with owners of small firms revealed considerable entrepreneurial initiative in changing product mix and seeking newly opened market niches. Sample entrepreneurs fell into two broad groups: dynamic, successful adapters with good prospects (found mostly among small-scale enterprises) and stagnant producers who had not adapted to the new competitive environment (found mostly among microenterprises). For potentially dynamic small firms, the most critical constraint was lack of access to finance for working capital and new investment. Many microentrepreneurs were seriously constrained by a lack of purchasing power among the lower-income population and by saturation of the sector. This article uses a survey of 82 manufacturing firms in Ghana to analyze how small, private enterprises responded to structural adjustment reforms and how their contribution to economic growth could be enhanced. Growth of private indigenous enterprises is important for sustained industrial development, given decreasing public ownership and uncertain prospects for foreign investment. But the absence of data on small firms, where most Ghanaian entrepreneurs are found, makes it difficult to assess their prospects in the changed environment. The survey reported here represents a rapid assessment technique for analyzing how different types of firms respond to adjustment policies and what constraints need to be addressed for them to realize their potential. William E Steel and Leila M. Webster are with the Industry and Energy Department of the World Bank. The data were gathered with the assistance of the Ghana National Board for Small-Scale Industries. The authors are grateful to E. K. Abaka, Frederick Gyebi Acquaye, and John Wayem for assistance and to Kwamena Adjaye, Surendra Agarwal, Benson Ateng, Nancy Barry, Stephanie Gerard, Don Mead, and two anonymous referees for their comments. ( 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 423 424 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Ghana introduced an Economic Recovery Program (ERP) in 1983 to redress some of the causes of its long economic decline. The ERP'S key elements had different implications for different types of industry. Supported by adjustment lending, import liberalization increased access to previously restricted inputs and spare parts (especially for small firms), but also broadened competition from imported products. Massive realignment of the highly overvalued ex- change rate created new opportunities for export and import substitution, but adversely affected import-dependent industries by sharply raising the price of imported inputs and the cost of financing them. In principle, price liberalization permitted firms to pass through higher costs, but in practice many could not because demand was restrained by stabilization measures and competition was high, especially in self-employment activities with low barriers to entry. The study focuses on small-scale enterprises (SSES) for two principal reasons. The first reason is to investigate the view that African countries lack the dynamic entrepreneurs needed to initiate and sustain the process of industrialization. In this sense, SSEs-defined for this study as having from 4 to 29 full-time workers-can be distinguished from microenterprises with fewer than 4 workers, which are more likely to be informal and oriented toward income for survival. In Ghana SSEs had been discouraged by excessive regulation, distorted incentives, and state domination of industry. Hence we hypothesized that liber- alization of markets would enable potential industrialists to emerge. The second reason for our focus on SSES is to assess the extent to which flexible organization and technology and relatively high labor intensity enable SSES to survive during adjustment. We hypothesized that SSES would be better able than large firms to respond quickly to the dramatic changes under Ghana's ERP and that smaller firms might absorb the surplus labor being shed by larger ones. A consensus emerging from the literature is that the fundamental issue is how to create a policy and business environment that enables SSES to contribute productively to industrial development, not whether SSES have a role to play (Liedholm 1990; Schmitz 1982). Hence the study examines constraints on firms' ability to respond to changing incentives and how these differ by firm size. In particular, we hypothesized that microenterprises-defined in the Ghanaian context as having fewer than 4 full-time workers (including self-employment)- would serve more as a safety valve for surplus labor and be less able to respond to the ERP, given their low barriers to entry and limited access to resources. The data were collected through interviews lasting one to two hours during three weeks in November 1989. The survey was conducted in two major urban centers, one medium-size town, and two rural towns to represent a cross-section of urban/rural and growing/stable settings. An effort was made to represent the range of activities in which SSEs are typically found, although the lack of suitable census data made it difficult to select a statistically representative sample. Sixty- six (80 percent) of the 82 firms surveyed fell into six subsectors that accounted for 81 percent of employment recorded in the industrial census (Ghana 1989) Steel and Webster 425 and 86 percent of total manufacturing employment in the population census (Ghana 1987): food processing, textiles and garments, wood products, soap and cosmetics, metal products, and building materials. The oldest firm started in 1950 and the most recent in 1989. Forty-eight firms (59 percent) started before 1984, and 33 (40 percent) began operations between 1984 and 1989. These data were supplemented by an additional survey of selected questions for large firms with 30 or more workers. The analysis focuses on differences across firm size categories, which gener- ally yielded the most interesting results. Selected firm characteristics such as subsector and age are also analyzed. Firm characteristics are to some extent interrelated: no firms established since 1986 were large, and microenterprises were slightly overrepresented in the food and metal subsectors, as were SSES in wood products. The small number of observations, however, precluded statis- tically disentangling the separate effects of different variables. The aim, rather, was to arrive at indicative answers to the research hypotheses. Preceding the analysis of survey findings, section I summarizes the literature on the role of SSES in industrial adjustment, and section II reviews the pre- adjustment context. Section III characterizes the entrepreneurs in the sample, and section IV examines changes in employment, output, and investment under the adjustment program. Sections V and VI analyze adaptive response and constraints on adjustment, and section VII presents the conclusions. 1. THE ROLE OF SMALL ENTERPRISES IN AFRICAN INDUSTRIAL ADJUSTMENT In the period after independence, many African countries attempted to leap directly to a modern industrial structure through public investment in large-scale industries. The state often took the lead for lack of a strong indigenous entrepre- neurial class and to avoid dependence on foreign investors. But inadequate attention to economic viability and market prospects resulted in substantial excess capacity, with many large firms unable to survive without heavy protec- tion or subsidies. Many enterprises were squeezed, first, by economic crisis and, subsequently, by adjustment policies that reduced protection, cut back subsidies, restrained demand, and changed relative prices. Given budgetary restraints and a policy shift away from direct ownership of productive enterprises, govern- ments have had to look increasingly to the private sector to take the lead in future industrialization. Both cross-sectional and time-series data suggest that the industrialization process normally involves initial rapid growth of production in small-scale en- terprises, some of which may expand into medium- and large-scale firms or survive in market niches even as large-scale industries gradually come to domi- nate the size distribution (Anderson 1982; Cortes and others 1987; Liedholm and Mead 1987; Little and others 1987; Nanjundan 1987; Staley and Morse 1965). In most African countries, the bulk of manufacturing employment is in self-employment and enterprises of fewer than 10 workers, while most output is 426 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 generally produced in large firms of more than 100 workers (Liedholm and Mead 1987). Thus, some observers argue that there is a "missing middle" in the size distribution of industry in African countries that must be filled for private initiative to sustain industrial development (Kilby 1988). We follow much of the literature in distinguishing SSES from household and cottage industries that are essentially self-employment with low barriers to entry. SSES may be defined as production units with hired labor that involve enough investment in capital or skills to constitute a barrier to entry but are small enough to be managed by one person. Few of the firms surveyed were part of an informal sector in the sense of being outside the legal framework. Although many of the smaller firms appeared "informal" in the organization of their production process, all but a few were registered as businesses and paid taxes at least to the local district. Many gradations of "formality" in different aspects of business operation were observed. Hence we have avoided the term "informal sector" and instead have used size categories that can be objectively measured. The number of workers is a measurable and therefore convenient indicator of size, although it must be recognized that dividing lines between groups are arbitrary. The following arguments have been used to justify the expansion of SSEs as a desirable strategy. * Successful industrialization must have an indigenous base, and expansion of the SSE sector would help develop entrepreneurial and managerial skills as a basis for efficient indigenous investment in and management of larger indus- tries (Bolton Committee 1971; Bruton 1990; Kilby 1988). * Because SSES tend to be relatively labor-intensive and to use low levels of technology, a strategy to expand the SSE sector is likely to be consistent with employment and income distribution objectives while allowing for sustained productivity increases through improvements in technology (Staley and Morse 1965; Steel and Takagi 1983). * Abolishing restrictive licensing, tax incentives for large firms, and direct allocation of credit, materials, and foreign exchange to large firms (espe- cially in the public sector) is likely to unleash investment by SSES in previ- ously untapped opportunities (Little and others 1987; UNIDO 1989). * SSES can respond flexibly under difficult and changing conditions because they do not depend heavily on infrastructure and because their typically low levels of technology allow product lines and inputs to be changed at rela- tively low cost (Morawetz 1974; Steel 1977). * Even when large-scale industries dominate, many SSES retain a competitive advantage by serving dispersed local markets, providing differentiated prod- ucts with low-scale economies for niche markets, or specializing as sub- contractors for larger firms (Anderson 1982; Anheier and Seibel 1987). Adjustment policies such as those pursued by Ghana can be hypothesized to open opportunities for SSES to expand. By introducing greater competitive forces Steel and Webster 427 and opening up export opportunities, adjustment policies are intended to yield a more efficient, dynamic structure of production by forcing inefficient firms to lower production costs or die out. A substantial negative impact could be ex- pected on inefficient large-scale firms that were created under heavy protection to import-intensive, import-substitution industries, which characterized much of Ghana's (and other African countries') industrial structure. However, this effect would be mitigated to the extent that firms could take advantage of increased availability of foreign exchange to utilize excess capacity. SSES could offset the negative effects of import liberalization to the extent that they could use local inputs to produce import substitutes or exports, thereby taking advantage of the incentives provided by a greatly devalued exchange rate. However, the ability of SSES to grow would also depend on the ability of the financial system to shift resources to them from industries that are declining and on how severely demand and credit were restrained under stabilization policies. 11. TRENDS BEFORE ADJUSTMENT As Ghana's economy declined in the late 1970s and early 1980s, foreign exchange scarcities curtailed production in its large import-substitution indus- tries, which depended heavily on imported inputs. Capacity utilization fell from 40 percent in 1978 to 21 percent in 1982. Although public policy prevented employment from falling in state-owned enterprises, large private firms recorded in the industrial statistics in Ghana reduced employment by a third during this period. Through innovative use of local raw and waste materials, some technically astute small entrepreneurs were able to fill gaps left by declining imports and large-scale output by producing import substitutes such as soap, metal products, and vehicle spare parts (Anheier and Seibel 1987; Dawson 1990). However, many SSES were constrained by lack of access to complementary imported in- puts, such as chemicals for tie-dye, screws for carpenters, ink for printers, perfumes for cosmetics producers, and even yarn for weavers of traditional kente cloth. There is no conclusive evidence as to whether opportunities for SSES outweighed constraints during the period of economic decline and tight controls. Stagnant income per capita in the 1960s stimulated growth in self- employment as people attempted to maintain family incomes. Women ac- counted for 63 percent of the total increase in nonagricultural employment from 1960 to 1970-half of them in manufacturing and 91 percent of them in enter- prises of fewer than 10 workers (Steel 1981). This process accelerated with steeply declining real incomes in the late 1970s and early 1980s. Female partici- pation in the formal labor force jumped from 64 percent in 1970 to 82 percent in 1984, and women accounted for the entire increase in manufacturing employ- ment during that period (Steel 1981; Ghana 1987). With rising inflation in the 1970s and 1980s, many modern sector workers turned to self-employment ac- 428 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 tivities to supplement their declining real wages, while others were forced to work on their own when they were laid off. As a residual source of employment and economic survival, the microenterprise sector is generally thought to have expanded in the face of a declining economy. III. ENTREPRENEURS DURING ADJUSTMENT The survey identified small enterprises as an important locus of investment under the adjustment program. Almost half of the firms with fewer than 30 workers were established after the 1983 reforms. In contrast, none of those with 30 or more workers was established after 1983 (including 31 respondents to a related survey of large-scale firms). Although it is not known what percentage of new firms would be "normal" for each size group, the difference in the propor- tion of new businesses between the small and large groups is striking. Many new owners were highly entrepreneurial in that they saw a profitable opportunity and took it. None followed their parents into business, as against 38 percent of pre-1984 owners. Their primary motivation was to apply their training (43 percent versus 13 percent of pre-1984 owners). These new entrepre- neurs were better educated than their predecessors, both formally and through significant past job experience. The number of new small firm owners who finished middle school was twice that of pre-1984 owners (60 percent versus 29 percent). The most successful SSE entrepreneurs moved into profitable niches producing specialized, nontraditional items, such as freezers, water coolers, and drums, to compete with increasingly expensive imports. Some undertook innovative pro- cessing of locally available materials, such as fuel briquettes from sawdust and knives from used band-saw blades and metal packing strips. Some left large companies to start their own business. Most started small and moved up, some more quickly than others. In contrast, new microentrepreneurs (with fewer than 4 workers) were more likely to have entered business through the apprenticeship route: 44 percent versus 25 percent of pre-1984 microentrepreneurs. Ghana's informal appren- ticeship system ensures that existing skills are maintained in trades such as garments, carpentry, vehicle repair, and metal-working; however, this also en- sures that the numbers of young adults with similar skills will increase, without necessarily imparting the entrepreneurial and innovative abilities needed to be- come more productive. For example, one seamstress compensated for declining demand by increasing her paying apprentices to 37. With negligible orders for garments, the girls learned to sew on paper bags. The seamstress solved her immediate need for income but trained her future competition. The apprentices' futures probably will depend more on the growth of income among Ghana's workers than on their abilities as seamstresses. Women owned 15 of the 82 firms surveyed; all but 2 of them had fewer than 10 workers and were found in food processing, textiles, and garments. These Steel and Webster 429 subsectors accounted for 78 percent of total female manufacturing employment in 1984 (Ghana 1987). Although female entrepreneurs generally were less edu- cated than their male counterparts in the sample, among the microenterprises women tended to be better educated than men (43 percent with 12 or more years of schooling as against 28 percent of men) and to have been in business much longer (an average of 17 years). One interpretation is that many women may view successful self-employment as a satisfactory, permanent source of income rather than as a step toward expansion. Another is that women have limited opportunities to expand their enterprises because of other substantial demands on their time. The problems women face in business in Ghana appear to arise mostly from their concentration in easy-entry, highly competitive activities and from gender-related differences in education and family responsibilities. IV. EMPLOYMENT, OUTPUT, AND INVESTMENT UNDER ADJUSTMENT This section analyzes how adjustment affected employment, production, and investment at the firm level and pays particular attention to differences between firms established before and after the ERP was launched. The evidence indicated that adjustment policies shifted the advantage away from larger firms and that entry and growth were particularly high among SSES. Changes in Employment Table 1 shows the average annual growth in employment by firm size during 1975-83 and 1983-89 for sample firms established by 1975. During 1975-83, incomes fell and import controls were tightened as foreign exchange became increasingly scarce. Nevertheless, employment grew at 7.6 percent a year in the medium- and large-scale firms in the sample, presumably because imports were channeled to these firms and because the government maintained strong pres- sure on both public and private sector firms not to lay off workers. Microen- Table 1. Employment Growth, by Age and Size of Firm, 1975-83 and 1983-89 (weighted average annual percentage growth) Growth since start Firms established Established Number offirms by 1975 Established during established by Firm size 1975-83 1983-89 by 1983 1984-89 1975 1983 1989 1-3 workers 0.0 16.0 9.7 7.6 4 17 33 4-9 workers -0.2 6.2 1.3 45.1 7 16 26 10-29 workers 0.5 3.1 8.1 19.1 9 9 16 30+ workers 7.6 -17.2 1.1 n.a. 5 7 7 n.a. Not applicable. Note: Size categories and weights are based on total employment in 1983 for firms established by then, to reveal the impact according to firm size at the beginning of adjustment. For firms established from 1984 to 1988, size categories are based on 1989 employment and weights on employment at start-up. Source: Survey data. 430 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 terprises and SSES, however, lacked government protection and suffered stagnant or declining employment as incomes and their access to inputs fell. The preceding data are for individual firms in the sample. In the aggregate, as some firms shut down, production and employment were falling in the large- scale (especially private) sector. Conversely, aggregate employment (although not necessarily production) in microenterprises and SSES is thought to have increased, as people sought alternative and supplementary sources of income. The picture reversed sharply under the ERP from 1983 to 1989. Employment fell rapidly in the medium- and large-scale sample firms, partly because many were able to shed excess labor and partly because they were squeezed between high costs of imported inputs and greater competition from liberalized imports. Employment rose in all the other size categories, in part reflecting expansion as smaller firms gained greater access to inputs and rural demand recovered. One striking feature of the recovery was that a much higher proportion of the firms surveyed reported an increase in employment (62 percent) than in produc- tion (39 percent; table 2). This was true for all size groups and most subsectors; employment growth lagged only in food products and soap and cosmetics. Increased labor absorption in part represents substitution in response to dras- tically eroded real wages and increased interest rates, making labor relatively cheap. A sharp increase in the number of microenterprises confirms that many workers turned to this sector for additional income (part-time workers ac- counted for 50 percent of microenterprise employment, as against an average of 13 percent for all firms surveyed). Table 2. Production and Employment Trends, by Size of Firm, for Firms Established by 1983 and during 1984-89 (percentage of respondents in each category) Number of workers Number of workers in firms established in firms established Large Allfirms by 1983 during 1984-89 firin Impact surveyeda 1-3 4-9 10-29 30+ 1-3 4-9 10-29 30+ survey Change in production Increase 39 29 31 33 43 34 70 43 n.a. 58 Decrease 43 65 44 67 43 46 20 14 n.a. 29 Change in employment Increase 62 47 56 78 71 56 80 71 n.a. 35 Decrease 16 18 25 11 29 6 0 29 n.a. 39 Number of firms 82 17 16 9 7 16 10 7 0 31 n.a. Not applicable. Note: Size categories are based on employment at the time of the survey in 1989. Compared with table 1, the microenterprise category includes some unsuccessful firms that declined, and the largest category includes some successful firms that grew. Firms reporting no change are not shown; so figures may not add to 100 percent. a. Excluding large-firm survey. Source: Survey data. Steel and Webster 431 Production and Competition Between 1983 and 1989, approximately the same proportion of firms in the total sample increased production (39 percent) as decreased (43 percent; table 2). A clear pattern emerges when firms are broken down by size and period of establishment. Firms established before the ERP had a relatively high propensity to decline under adjustment policies. In contrast, firms in all size groups estab- lished since 1983 were more likely to have increased their output than those already in existence. New firms with 4 to 29 workers (SSES) had especially high ratios of growing to contracting firms. This growth in the newer firms may be partly attributable to the initial spurt that characterizes many SSES in their first three years (Liedholm 1990), as well as to their choice of activities with high growth potential. Nevertheless, microenterprises with 1 to 3 workers were more likely to have decreased than increased output. Among firms with fewer than 30 employees, three-quarters reported that other SSES were the primary source of competition. Although 61 percent claimed that competition had increased since 1983, only 12 percent mentioned imports as a major source of that competition. The principal subsectors affected by import competition were metal products (21 percent of firms, mostly in agri- cultural machinery, which could be imported duty-free) and soap and cosmetics (29 percent of the subsector, especially cosmetics firms that did little more than repackage imported materials). Investment Contrary to some observers' perceptions, investment was taking place, at least among small firms. Nearly half of the microenterprises sampled were established since 1983 (high birth and death rates are expected in this group); while 38 and 44 percent of firms with 4 to 9 and 10 to 29 workers, respectively, entered since 1983. As import costs rose, many of these new firms sprang up to supply low- cost substitutes, particularly from local materials. Examples included pottery, locally mixed paints, and simple agricultural implements. In addition, about half of the firms established by 1983 had purchased some new equipment by 1989 (table 3). V. ADAPTATIONS Many firms adapted to new price incentives by altering their product lines- an important motivation for replacing outmoded equipment that had deterio- rated during the late 1970s and early 1980s. The product mix was altered during the ERP by 34 percent of all firms surveyed and generally by higher proportions for those establishecdby 1983 (table 3). As the construction industry picked up, for example, metal workers began producing metal gates and burglar alarm systems. Unable to compete with imports, a chalk producer shifted to producing starch for the textile industry. Changes in product mix were especially 432 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 3. Production and Input Changes under Adjustment, by Size of Firm, for Firms Established by 1983, 1983-89 (percentage of respondents in each category) Number of workers in firms Allfirms established by 1983 Change surveyeda 1-3 4-9 10-29 30+ Bought new equipment 49 43 56 56 57 Product mix changed 34b 40 27 63 45b Selling in different markets 15 12 19 0 43 Have exported (directly or indirectly) 9 6 8 7 14 Have considered exporting 38 28 46 44 43 Imported share of raw materials Actual percentage share 34.0b 19.1 19.1 62.8 52.6b Change since 1983 Greater 9b 11 11 25 8b Smaller 28b 22 11 13 83b Easier to get Imported inputs 73b 89 64 50 88b Domestic inputs 57b 64 44 17 67b Harder to get or too costly Imported inputs 12b 0 18 33 Ob Domestic inputs 20b 21 25 33 Ob Number of firms 82 17 16 9 7 a. Includes firms established after 1983 as well as before. b. Includes an additional 24 firms from a separate survey of large-scale enterprises. Source: Survey data. prevalent in the soap and cosmetics and the textiles and garments subsectors (83 and 50 percent of firms, respectively). Those manufacturing food, wood, and metal products had less need to change because their dependence on imported inputs was low (19, 10, and 18 percent, respectively) and their products were adapted to local tastes, making them less vulnerable to competition from stan- dardized imports. Along with changes in product mix, a number of firms shifted their marketing strategies. Large firms in particular had to seek out new domestic markets (43 percent; table 3). Devaluation raised the price of exports, and the response was especially high in the timber industry. New, small sawmills were buying logs from those with timber concessions and sawing them into lumber for larger exporting firms. Of the total sample, 38 percent were interested in exporting, even though only 9 percent had any direct or indirect experience with exports. SSE exports tended to be somewhat haphazard and on a small scale. For exam- ple, a tie-dye producer sold a relatively small volume of her fabric to a German buyer, and a producer of commercial freezers sold several units to traders from Nigeria and Mali. Use of Inputs The use of imported inputs was affected in two opposite ways by Ghana's adjustment policies. First, import liberalization made imports more available to Steel and Webster 433 small firms, previously excluded by the large-scale firms that were favored by the import licensing system. Second, devaluation raised the cost of imported inputs, favoring industries with high proportions of domestic inputs and value added. Among firms in the sample, the same proportion used imported inputs exclusively as used only domestic raw materials (22 percent). On the average, the proportion of imported raw materials used by firms with fewer than 10 workers (19 percent) was about a third that used by larger ones, making them less vulnerable to devaluation (table 3). While the majority of respondents of all sizes reported that imports had become easier to obtain, only for those with 10 to 29 workers did increased access outweigh the substitution effect: twice as many (25 percent) increased their import content as decreased (13 percent). None of the firms in wood, metal products, and textiles and garments reported any change in import content (partly because import shares were already rela- tively low). The overwhelming majority of firms with 30 or more workers reported substi- tution of domestic for imported inputs under the ERP (83 percent, in contrast to an average of only 16 percent of firms with fewer than 30 workers; table 3), although no significant change in average import content appears in recorded statistics for large-scale industry from 1984 to 1988. One reason for the rela- tively low substitution found among smaller firms was that relative prices may not have changed much in favor of domestic materials, most of which are tradable. A much larger share of SSES cited the high price of local raw materials as a problem than that of imported inputs (for which they previously had to pay black market prices). The price of agricultural raw materials had risen because food prices were liberalized. The price of cloth to the tie-dye and garment subsector remained high because extra tariff protection was provided to large textile mills, and costs in the greycloth factory were not significantly reduced. Since timber is exportable, its domestic price tended to rise pari passu with devaluation. Good scrap metal became increasingly scarce. Thus there may have been supply-side constraints to rapid expansion of some SSES that were based on local raw materials. VI. CONSTRAINTS ON ADJUSTMENT AND GROWTH Although some firms found market niches with good growth potential, many were not able to pass on fully the increased costs of raw materials and equipment to consumers in the form of higher prices because of weak demand and an inflow of competing imports. Most firms experienced a financial squeeze, in- cluding those on sound financial footing in the past. The survey revealed that if demand were eliminated as a problem, finance would be an even more severe constraint on growth. In sum, the incentive side of the adjustment process was working-less efficient firms were being squeezed-but the financial side was not functioning adequately to enable more efficient firms to grow. Respondents to the survey identified lack of access to credit (especially for 434 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 working capital), demand, and increased cost of inputs as their principal operat- ing problems, but the relative importance varied by size. The importance of credit increased with firm size, while demand problems were most evident among microenterprises. SSES were more concerned about the high price of local inputs, whereas larger firms placed more emphasis on imported input costs. SSES also were most eager to replace old equipment and hence were most affected by lack of access to term lending. The majority of SSES (58 to 64 percent of firms with 4 to 29 workers) saw their ability to respond to market incentives as constrained primarily by lack of access to institutional credit for working capital and for equipment (table 4). The other two main constraints-the high cost of local materials and the need to replace old equipment-indicated why credit was needed to facilitate their supply re- Table 4. Constraints by Size of Firm, 1983-89 (percentage of respondents in each category) Allfirms Number of workers Constraint surveyed 1-3 4-9 10-29 30+ Main constraint on sales Resources (could sell more) 52 44 58 64 43 Demand Can sell current production but no more 24 19 23 21 57 Can't sell current production 24 38 19 14 0 Credit harder to get 84 83 100 100 100 Bank loan Haveeverhadone 32 18 19 60 86 Have tried to get one in past 5 years 47 28 48 73 71 Competition is greater 61 53 64 33 71 Major competition from Imports 21' 13 8 7 50a Small firms 65a 75 72 72 37a Top four operating problemsb Credit For raw materials 57 55 54 63 71 To buy equipment 29 18 35 50 14 Demand Consumers lack of money 35 55 23 19 29 Too many other firms 17 21 12 25 0 Can't afford my product 11 15 12 6 0 Inputs Local material prices 22 30 23 69 14 Imported input prices 18 9 23 25 29 Can't get local materials 11 15 15 0 0 Equipment Needs replacing 23 12 35 31 14 Number of firms 82 33 26 16 7 a. Includes an additional 24 firms from a separate survey of large-scale enterprises. b. Firms could list up to four problems, so the percentages can add to more than 100 percent. Problems listed by less than 11 percent of firms are not shown. Source: Survey data. Steel and Webster 435 sponse. Many of these firms had orders for their products but insufficient work- ing capital to meet them, due in part to slim profit margins and delays in collecting from customers. Because responses were based on owners' percep- tions, internal management problems were likely to be underrepresented as constraints. Nevertheless, almost all the owners of firms with fewer than 10 workers expressed a desire to improve their business skills, particularly those relating to technology and accounting. Microenterprises were much more likely to be constrained by demand: 38 percent could not even sell their current production (table 4). They attributed weak demand primarily to people's lack of money-reflecting decreased real income per capita since the 1970s as well as restrained demand under the ERP. Increased competition from other small firms was also a factor: a growing number of competitors were slicing up a diminished pie. Thus the fortunes of the smallest firms depend most on growth of demand among low-income consumers (rural and urban), who are their principal markets. In contrast to firms with fewer than 30 workers, larger firms saw imports as their main competition, and none cited competition from other domestic firms among their top four problems. Although they did not feel as immediately constrained by demand as microenterprises, 57 percent thought that demand could not absorb further increases in their production (table 4). These results confirm the importance of import liberalization to force large firms to act competitively. Credit As could be expected under tight monetary policies that curtail government spending and restrain credit, almost all of the survey respondents agreed that credit had become harder to get under the ERP (table 4). The survey also indi- cated significantly less access to bank loans for firms with fewer than 10 workers (fewer than 20 percent had ever had a loan) than for those with 10 to 29 and more than 30 workers (60 and 86 percent, respectively). More than 70 percent of firms with 10 or more workers had applied for a loan since the ERP began, thus indicating a high demand for credit. Only 9 percent of firms said they would refuse a loan for working capital at 30 percent interest, and only 25 percent (mainly microenterprises) would decline credit for investment purposes. Microenterprises often demurred because they doubted that they could sell enough to repay their loans. Regulatory Environment Given the severity of finance and demand problems, the regulatory environ- ment was rarely mentioned among the top four problems, except for taxes (cited by 14 percent of all respondents and 30 percent of large-scale firms). Infrastruc- ture (especially electricity interruptions) and transportation costs were cited by S to 7 percent. Only 5 percent mentioned the business environment generally, and just 3 percent (large-scale firms only) listed regulations or licensing as a problem. 436 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Regulatory problems appeared to affect larger firms more than smaller ones, and exporters more than those producing for the domestic market. Location was the only significant regulatory concern of firms with fewer than 10 workers, many of whom had been affected by relocation and demolition efforts by urban authorities. Nevertheless, the business environment may be more of a concern for new investors than for existing producers who have already adapted to it. Uncer- tainty about the economy was seen as a restraint on new investment by 38 percent of respondents, especially those with 10 or more workers. More than a third of large firms also expressed some reservations about the government's attitude toward private investment. VII. CONCLUSIONS Structural adjustment policies have begun altering the structure of industrial production in Ghana. Changes in the exchange rate, trade policy, and price controls have had different effects on incentives and profits. Although both positive and negative effects were found within each size and subsector group, differences between group averages indicate that structural changes were occur- ring. Large firms were particularly affected by competition from liberalized imports, and the increased cost of imported inputs forced many to increase the content of local raw material. Small enterprises were especially responsive in adapting their product lines to changing market opportunities and in taking advantage of increased access to imported inputs. Many entrepreneurs sought new products, techniques, and markets, and others would do so if they had greater access to resources, especially finance. New small enterprises were gener- ally performing well. Nevertheless, the scarcity of credit and the absence of mechanisms to shift resources from declining firms to those with greater growth potential appeared to restrain investment to take advantage of new opportunities. SSES were quick to identify lack of credit for raw materials and equipment as major constraints to expansion, and they expressed strong interest in loans at current interest rates. Increased access to credit would greatly facilitate SSES' contributions to recovery and adjustment of the industrial sector. In contrast, microenterprises were much more likely to view weak demand as a major constraint-a consequence of the sharp fall in income per capita in the decade before the ERP and of restraints on demand under the ERP. Employment in microenterprises (including self-employment) mushroomed as formal sector employment and incomes diminished, but much of this growth was driven by excess supply of labor rather than by production-generating demand. Income and productivity may decline further without increased demand for the type of goods and services produced by this already overcrowded, low-wage sector. Policies that put more money into the hands of the low-income population (for example, through improving the terms of trade for farmers) are likely to have Steel and Webster 437 greater effect on microentrepreneurs' earnings than supply-side measures to as- sist individuals. As demand and credit problems are resolved, taxation and the business cli- mate may come increasingly into play as constraints on investment, at least for larger firms; smaller firms tend to be less concerned with regulations. Economic and political stability and a positive attitude toward private profit would help reduce uncertainty-an important consideration for new investors. The cost of doing business could be reduced through lower, more transparent taxation. Export procedures, as well as financing, need improvement for SSES to live up to their export potential. In addition to such measures, a long-run strategy should include complementary education and technical training to enhance the ability of Ghana's small entrepreneurs to contribute to industrial development. REFEREN CES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Anderson, Dennis. 1982. "Small Industry in Developing Countries: A Discussion of Issues." World Development 10 (11, November):913-48. Anheier, Helmut K., and Hans Dieter Seibel. 1987. Small-Scale Industries and Economic Development in Ghana: Business Behaviour and Strategies in Informal Sector Econ- omies. Cologne Development Studies 3. Saarbrucken, Germany: Breitenbach. Bolton Committee. 1971. Bolton Committee Report: Small Firms. Cmnd. 4811. Lon- don: Her Majesty's Stationery Office. Quoted in Ian M. D. Little and others, 1987. Small Manufacturing Enterprises. New York: Oxford University Press. Bruton, Henry. 1990. "Broad-Based Growth." Paper presented at a workshop on Re- search Priorities for Policy Reforms Supporting Broad-Based Growth and Democracy. USAID/APRE/SMIE. Washington, D.C. Processed. Cortes, Mariluz, Albert Berry, and Ashfaq Ishaq. 1987. Success in Small- and Medium- Scale Enterprises: The Evidence from Colombia. New York: Oxford University Press. Dawson, Jonathan. 1990. "The Wider Context: The Importance of the Macroenviron- ment for Small Enterprise Development." Small Enterprise Development 1 (3, September):39-46. Ghana, Republic of. 1987. 1984 Population Census of Ghana: Demographic and Eco- nomic Characteristics, Total Country. Accra: Ghana Statistical Service. .1989. Ghana National Industrial Census 1987. Accra: Ghana Statistical Service. Kilby, Peter. 1988. "Breaking the Entrepreneurial Bottleneck in Late-Developing Coun- tries: Is There a Useful Role for Government?" Journal of Development Planning 18:221-49. Liedholm, Carl. 1990. The Dynamics of Small-Scale Industry in Africa and the Role of Policy. GEMINI Working Paper 2. Washington, D.C.: USAID. Liedholm, Carl, and Don Mead. 1987. "Small-Scale Industries in Developing Countries: Empirical Evidence and Policy Implications." International Development Paper 9. Michigan State University, Department of Agricultural Economics, East Lansing. Processed. 438 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Little, Ian M. D., Dipak Mazumdar, and John M. Page, Jr. 1987. Small Manufacturing Enterprises: A Comparative Analysis of India and Other Economies. New York: Ox- ford University Press. Morawetz, David. 1974. "Employment Implications of Industrialization in Developing Countries: A Survey." Economic Journal 84 (335, September):491-542. Nanjundan, S. 1987. "Small and Medium Enterprises: Some Basic Development Issues." Industry and Development 20:1-50. Schmitz, Hubert. 1982. "Growth Constraints on Small-Scale Manufacturing in Develop- ing Countries: A Critical Review." World Development 10 (6, June):429-50. Staley, Eugene, and Richard Morse. 1965. Modern Small Industryfor Developing Coun- tries. New York: McGraw-Hill. Steel, William E 1977. Small-Scale Employment and Production in Developing Coun- tries: Evidence from Ghana. New York: Praeger. . 1981. "Female and Small-Scale Employment under Modernization in Ghana." Economic Development and Cultural Change 30 (1, October):153-67. Steel, William F., and Yasuoki Takagi. 1983. "Small Enterprise Development and the Employment-Output Trade-Off." Oxford Economic Papers 35:423-46. UNIDO. 1989. "Environment Conducive to Sustained Growth of Small- and Medium- Scale Enterprises.' Discussion Paper ID/WG.492/4 for the First Consultation on Small- and Medium-Scale Enterprises including Co-operatives (Bari, Italy), Vienna. Processed. THE WORLD BANK ECONOMIC REVIEW, VOL . 6, NO. 3: 439-458 The Dynamics of Optimal Gradual Stabilizations Alex Cukierman and Nissan Liviatan Inflation inertia may be quite tenacious because of the simultaneous interaction be- tween policy actions and inflationary expectations under imperfect credibility. This result is particularly relevant for understanding some of the failed efforts to stabilize inflation in South America. This article deals with the issue of inertia in the framework of imperfect information about the type of the policymaker and extends the existing models to an infinite horizon. Because policymakers do not have perfect control of inflation, a 'frivolous stabilizer" may deviate from the policies of a "serious stabilizer" without necessarily being unmasked immediately. When the difference in the ability of "strong" and "weak" policymakers to control inflation is large, unexpected inflation may be persistently negative for quite a while, thus causing reduced economic activity and giving the indication that credibility is low. If the policymaker persists with the stabilization, this pattern gradually disappears as his reputation rises. But before this final stage the serious policymaker has to compromise his inflation objective in view of adverse expectations about his type and pay the cost of imperfect credibility. In some countries, particularly in Latin America, inflation has been quite tena- cious in spite of recurring attempts at stabilization. Many of these attempts probably failed because the stabilization packages did not include a serious commitment to slow down the rate of growth of the money supply and to reduce the deficit. But even when such a commitment was in place, as in Chile and Argentina during the mid-1970s, inflation came down rather slowly and was accompanied by substantial and sustained reductions in the level of economic activity. It is obviously possible to claim that if the monetary brakes had been applied more strongly, stabilization would have been faster. It is not clear, however, that such a course of action would have yielded better overall results. The policy- maker may find it preferable to stabilize gradually because of credibility prob- lems, thus providing a rigorous foundation for "inflation inertia" (Kiguel and Liviatan 1988). Alex Cukierman is with the Department of Economics at Tel Aviv University, and Nissan Liviatan is with the Department of Economics at Hebrew University and the Country Economics Department at the World Bank. A previous version of the paper was presented at the September 1991 European meeting of the Econometric Society, in Cambridge, U.K. The authors would like to thank Ariel Rubinstein, John Driffill, and several anonymous referees. © 1992 The International Bank for Reconstruction and Development/ THE WORLD BANK 439 440 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 The behavior of the policymaker while in office depends on the public's expectations about inflation (Barro and Gordon 1983). The public's expecta- tions reflect, in turn, the possibility that the policymaker is not serious about achieving stabilization (not a "serious stabilizer"). When the public has little faith that the policymaker has really become more concerned about price stabil- ity, even a serious stabilizer does not necessarily reduce inflation quickly in order to avoid very substantial decreases in economic activity. Hence, when a serious stabilizer is in office, inflationary expectations as reflected in nominal contracts turn out to be too high ex post, which causes them to be revised and reduced. This reduction in inflationary expectations raises the policymaker's reputation as a serious stabilizer. But this learning process, although optimal, is gradual because actual inflation is affected not only by the deliberate decisions of the policymaker but also by unpredictable events over which the policymaker has no control. Because subsequent policy actions also depend on expectations, and because expectations are reduced slowly, even a serious stabilizer may be led to stabilize gradually. If in office for a sufficiently long period of time, the serious stabilizer will ultimately build up a good reputation and deliver a much lower rate of inflation. But the period of stabilization will be protracted, and during most of it unex- pected inflation will be negative, thus creating a persistent lull in economic activity. This lull occurs in spite of the fact that the serious stabilizer partially accommodates inflationary expectations in order to minimize the combined costs of low employment and high inflation. Much has been made of inflation inertia arising from long-term, overlapping wage contracts and backward-looking, formal or informal indexation arrange- ments. This article demonstrates that even without these sources of inertia, inflation may be quite tenacious because of the simultaneous interaction be- tween policy actions and inflationary expectations when there is imperfect cred- ibility. Policy actions respond to expectations, which in turn are updated opti- mally, but sluggishly, in light of actual inflation. Practically all previous models in which the public is uncertain about what type of policymaker is in office postulate a finite horizon for the policymaker (Backus and Driffill 1985a, 1985b; Barro 1986; Vickers 1986; Persson and van Wijnbergen 1987; Andersen and Risager 1987, 1988; Cukierman and Liviatan 1991a). In this article we characterize the path of inflation and other variables when the policymaker has an infinite horizon. This extension makes it possible to determine whether a serious stabilizer is eventually able to deliver the best performance. Because policymakers do not have perfect control over inflation, a "frivolous stabilizer" may deviate from the policies that would have been followed by a serious stabilizer without necessarily being revealed as a frivolous stabilizer immediately. Imperfect control of inflation makes it more costly for a serious stabilizer to build up credibility quickly. The first dynamic implication of our framework is that inflation comes down Cukierman and Liviatan 441 during a gradual stabilization process, in contrast to models with mixed strate- gies of the Backus and Driffill (1985a) or Barro (1986) type, in which inflation goes up during this process. The second is that as inflation comes down, so do inflationary expectations. In Barro and in Andersen and Risager, inflationary expectations are fixed over time. Moreover, because of their fixed horizon, these models imply that a serious stabilizer may not reap the benefits of better perfor- mance until the last few periods, or even the very last period, of the game. Hence, the fixed-horizon assumption of these models does not give a serious stabilizer who cares somewhat about economic activity much incentive to pur- sue stabilization. An infinite horizon thus seems necessary in order to under- stand situations in which it is optimal to embark on a stabilizing path even when the formation of reputation is lengthy and gradual. The analytical framework we develop provides a natural vehicle for the dis- cussion of the effect of foreign aid on the credibility of stabilization. Untied aid, by permitting a higher level of economic activity, reduces the incentive to inflate and increases the credibility of stabilization. Conditional foreign aid reinforces these tendencies and speeds up the process of stabilization even further. Section I presents a simple, infinite-horizon, Barro-Gordon (1983) type of framework, with uncertainty about the type of policymaker and with imperfect control of inflation. Section II introduces the public's (Bayesian) process of learning. A full characterization of the model's solution for the case in which a serious stabilizer opts for a gradual stabilization is developed in section III. Section IV presents the dynamic features of the equilibrium solution and uses it to interpret the stabilization in the United Kingdom under Margaret Thatcher and the stabilization in Chile. Section V discusses the effect of conditional and unconditional foreign aid on the credibility of stabilization. 1. THE MODEL AND ITS RATIONALE The model is designed to bring out the public's uncertainty concerning the likelihood that the policymaker is serious about stabilizing inflation and the effect of this uncertainty on policy. In order to capture the public's uncertainty, we assume that there are two types of policymakers: weak and strong. Both types of policymakers dislike inflation and desire to maintain employment above the natural level of employment. (The natural level of employment is the level at which expected and actual inflation rates are equal.) The desired level of em- ployment of the "weak" policymaker is higher than that of the "strong" policy- maker. More precisely, the combined cost of inflation and of being below the desired level of employment for each policymaker in period t is (1) A (N-N )2 + = W,s where z denotes cost; A is the relative preference for price stability versus em- ployment objectives and is a parameter common to both types of policymaker; 442 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 N,f is the desired level of employment of policymaker i; w denotes the weak policymaker and s the strong policymaker; N, is the actual level of employment; and 1rt is the actual rate of inflation. Neither policymaker has perfect control over the rate of inflation. In particu- lar, when a policymaker plans to generate inflation at rate 7r' , the actual rate of inflation is (2) =1Pt + E5 i = W'S where ri - is the actual inflation under policymaker i, and eit is a stochastic noise term whose variance is inversely related to the degree of control of the policy- maker over the rate of inflation. Imprecise control of inflation is the result of either imperfect control of the money supply (Cukierman 1992, chapter 9) or the policymaker's uncertainty about money demand. The public cannot deter- mine with certainty to what extent a change in monetary expansion occurs because of the policymaker's deliberate plans and to what extent it occurs be- cause of an error in the policymaker's forecast of money demand (Cukierman 1992, chapter 13; Canzoneri 1985). The change obviously may result from a combination of both possibilities. We assume that the stochastic noise term has zero expected value and is distributed uniformly. In particular, 1/2a,, -a_, < x as. The policymaker can affect employment by creating inflation that was unan- ticipated at the time nominal contracts were concluded. This situation leads to a conventional expectations-augmented Phillips relation that is summarized in equation 4: (4) Nt -Nn = a(r, - r-), a > O where N. denotes the natural level of employment, and -7re represents the rate of inflation expected at contracting time for the period of the contract. Equation 4 states that the deviation of employment from its natural level is positively related to unanticipated inflation. Substituting equation 4 into equation 1 and setting a = 1 for simplicity, (S) zt 5 z(dpt, jr"e) = A [di-- (7r-7re)]2 + 2t 2 ~~~~2 where (6) di-Ni- N Nn = w,s. Cukierman and Liviatan 443 Thus di is the (positive) divergence between the level of employment desired by a policymaker of type i and the natural level of employment. From the Barro-Gordon (1983) analysis, when the public is fully informed about the type of policymaker holding office and there is perfect control of the money supply (Ei being identically equal to zero), the equilibrium rate of infla- tion is (7) ,,i = Adi= w,s. This is the time-consistent or subgame-perfect equilibrium. It is obtained by letting the policymaker choose the level of inflation (7r) so as to minimize the costs in equation 5. The policymaker would do this by taking the rate of infla- tion expected at contracting time (1re) as given and then imposing rational expec- tation. Under perfect information, rational expectation amounts to the require- ment 7t = ir, (Cukierman 1992, chapter 3). Because the difference between the desired rate of employment and the natural rate of employment is positive, the equilibrium rate of inflation is positive. Moreover, because there is no uncertainty of any kind, wage setters fully antici- pate the subsequent action of the policymaker. As a consequence, employment is always at the natural level in spite of the fact that inflation is positive. Obvi- ously, the same level of employment could have been obtained with zero infla- tion, provided wage setters had believed the policymaker would choose zero inflation. But the wage setters have no reason to hold such a view, because zero inflation is not optimal for the policymaker after the wage contract has been made. This dynamic inconsistency of monetary policy induces a suboptimally high rate of inflation (Kydland and Prescott 1977). Because the difference between the desired level of employment and the natu- ral level of employment is greater for the weak policymaker than for the strong one, the equilibrium in equation 7 implies that the weak policymaker produces a rate of inflation that is higher than the rate of inflation produced by the strong policymaker. The intuition is that the weak policymaker is known to have a larger employment objective and is rightly expected to inflate at a higher rate. Hence, wage setters demand higher wage increases than in the case in which the strong policymaker is known to be in office. When stabilization programs are introduced, the public is usually uncertain about their outcome. We model this uncertainty by assuming that either the weak or the strong policymaker is in office forever but that the public is not sure which type is in office. As time passes, the public learns from the realizations of inflation which type is likely to be in office. But this process may be protracted: because both types of policymakers have imperfect control of inflation, realiza- tions of past inflation do not necessarily convey precise information about the type to the public. The assumption that one type of policymaker is in office forever is obviously not made for its realism. Policymakers do change, and the relative emphasis on employment versus price stability may change even within the same administration. The assumption is made to illustrate the potential difficulties that low credibility brings to a strong policymaker even in the favor- 444 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 able case in which a particular type of policymaker is in office forever and public uncertainty concerns only his or her type. The policymaker's plans about how to handle inflation are made by taking both the current and the future values of the costs z(-) into consideration. In particular, when in office, either type of policymaker makes plans for current and future inflation rates that will, given the information available at the time, minimize the expected present value of costs. This present value, as of the present period (denoted by 0), is given by C. (8) EPo 1tz(d,,7r,xt,e), 0 < A c 1, i = w,s where 0 is a discount factor that measures the policymaker's rate of time prefer- ence and Epo is an expected value conditioned on the information available to the policymaker when the policymaker picks the average rate of inflation of the initial period. The weak policymaker is more sensitive to the costs of low em- ployment in all periods because the difference between the desired and the natural level of employment is greater for the weak policymaker than for the strong one. The length of a period is determined by the length of nominal wage contracts. Within each period, nominal wage contracts are made on the basis of the ex- pected rate of inflation between the previous and the current period. Then the policymaker picks the planned rate of inflation for the period, taking those expectations (or nominal contracts) as given. This sequence shows that the government cannot precommit itself to a level of inflation. Actual inflation for the period is determined, through equation 2, by the policymaker's decision and by the realization of the uncontrollable inflationary shocks. The public observes the actual rate of inflation before it sets inflationary expectations and nominal contracts at the beginning of the next period. However, the public never ob- serves the two components of the actual rate of inflation-one component planned by the policymaker and the other uncontrollable-separately. It. STABILIZATION AND THE EVOLUTION OF REPUTATION Inflation may be quite high for a while because a weak policymaker has been in office. If a strong policymaker then settles permanently in office and an- nounces that a stabilization phase has commenced, the public will remain skepti- cal. This is because the earlier, weak policymaker would have had an incentive to make similar statements. We model the public's skepticism by assuming that its prior probability that the announced stabilization has been made by a strong policymaker is a number that is strictly bounded between zero and one. The smaller the prior probability, the lower the initial "reputation" of the policy- maker (Backus and Driffill 1985a, 1985b; Vickers 1986). Inflation uncertainty normally rises with the level of inflation (Engle 1982; Cukierman and Liviatan 445 chapter 18 of Cukierman 1992). We capture this feature in a simple manner by assuming that the weak policymaker, who always plans to inflate at a higher rate than does the strong policymaker, also does not control inflation as tightly (a > as). In spite of this difference in the precision of inflation control, the public may not be able to ascertain with certainty, even many periods after the policymaker has taken office, that a strong type is in office. The reason is that imprecise control of inflation by both policymakers prevents the public from clearly sep- arating one type from the other, even if, as is normally the case, they plan to produce different average rates of inflation. Figure 1 illustrates why this is so for arbitrary equilibrium strategies of the two types of policymakers. From equa- tions 2 and 3a, actual inflation when a weak policymaker is in office is between ?r- - a, and ir- + a,,,. Similarly, actual inflation when a strong policymaker is in office is between rp - a and 7rs + as. When actual inflation is in the common range, which is the range of the strong policymaker in figure 1, there is no way for the public to clearly identify that a strong policymaker is in office. But because the probability that a rate of inflation in the common range has been produced by a strong type is larger than the probability that it has been pro- duced by the weak type (as can be seen from figure 1), realizations of inflation in the common range raise the reputation of the policymaker in office. This intuitive argument is confirmed by Bayes' formula, which shows how reputation evolves when the realization of inflation is in the common range. Given the actual rate of inflation in the current period, Bayes' formula relates the public's subjective probability that the policymaker in office is strong to the probability that the policymaker was strong before the realization of current inflation. Equation 9 shows the updating formula when actual inflation falls in the common range. Figure 1. Strategies of Different Policymaker Types and the Corresponding Distributions of Inflation Probability density function Strong policymaker Weak policymaker ItP-a , 4 - a, 4 7 4 + a, + a inflation Note: The relative positions of the two distributions in the figure are meant to be illustrative. Other relative configurations, some of which are discussed in the text and illustrated in figure 2, are possible. 446 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 (9) ce,+, Pr [i= s I r,] Pr [Lr, I i = s] Pr[i = s] Pr[7r,IIi = s] Pr[i = s] + Pr[7r Ii = w] Pr[i =w 1 2a, cet 1 1 a = = 2a- a,t + 2a: (1-et a' + a,, (1 - a) where Pr is an abbreviation for probability. Because the ratio of the imprecision of inflation control of the strong policymaker to that of the weak one is smaller than one, the probability that the policymaker is strong is greater in the current period (after the realization of 7rt) than in the previous period for all t as long as inflation falls in the common range of figure 1. In other words, given that the public is still uncertain about the type in office, an additional realization of inflation in the common range always raises the reputation of the policymaker in office. (If the ratio of imprecision of inflation control is one, then realizations of inflation in the common range are not informative.) When the weak policymaker is in office, there is a positive probability that inflation will fall in the noncommon range. When inflation actually falls in this range, the policymaker is revealed as being weak with certainty, and the public's prior probability that the policymaker is strong jumps to zero. The public knows that such a realization of inflation could not have occurred had a strong policy- maker been in office. Thus, given the strategies postulated in figure 1 for the weak and strong policymaker types, the weak type will ultimately be revealed as weak, although that may take a long time. However, it will never be demon- strated to the public with full certainty that the strong type is in office when this is the case. But the reputation of the strong type will increase monotonically, reaching one asymptotically as time goes to infinity. III. EQUILIBRIUM STRATEGIES, POLICY CONVERGENCE, AND EXPECTATIONS UNDER GRADUALISM The policymakers' equilibrium choice of strategy determines the probability that the public will be able to identify the policymaker as weak or strong. When the divergence between the equilibrium strategies of the two types of policy- makers is small compared with the divergence in the precisions of their control over inflation, the probability that the public will be able to separate a strong policymaker from a weak policymaker may be zero. Conversely, the probability of the public's being able to make a distinction is usually positive, and may even be one, if the divergence between the equilibrium strategies is sufficiently large in relation to the difference in the precisions of inflation control. The case of a small divergence in equilibrium strategies leads to equilibria in Cukiermnan and Liviatan 447 which it is optimal for a strong policymaker to stabilize gradually. The case of a large divergence corresponds to situations in which it is optimal for a strong policymaker to take a chance with a "shock treatment." Which of the two methods of stabilization is optimal generally depends on the policymakers' repu- tations at stabilization, their rate of time preference, the precision of their con- trol over inflation, and other parameters. The main focus of this article is on equilibrium when the optimal method of stabilization is gradual. Characterization of Gradual Stabilizations This section fully characterizes the equilibrium strategies of the two types of policymakers, but under a (provisional) assumption. The assumption is that the range of possible inflation rates produced by the weak policymaker's strategy fully covers the range of inflation rates that could be produced by the strong policymaker's strategy. In this case, the dynamic optimization problem decom- poses into a series of one-period maximization problems. A convenient feature of the uniform distributions postulated for the noise terms (ei) is that all observa- tions on inflation are equally informative and are independent of the magnitude of inflation as long as they all fall within the common range. Hence, as long as the strategies (planned inflation rates) of the weak and strong policymakers are such that the range of possible rates produced by the strong policymaker is fully covered by the range of rates that could be produced by the weak policymaker, the probability that the type will be revealed is independent of the precise loca- tion of the planned inflation rates within this range. As a consequence, within this range, either policymaker can select a current strategy to maximize the value of current objectives without paying attention to future values of the objectives. Because equilibrium strategies depend on the process of forming expectations and this process depends in turn on what the public knows about these strate- gies, expectations and equilibrium strategies are determined simultaneously. The appendix shows that the equilibrium strategies are given by equation 10 or, alternatively, by equations 11 and 12, where B is a positive combination of parameters (whose precise form appears in the appendix). Expectations are given by equation 13, which is also derived in the appendix. (10) t= 1 = A (di + 7re) I w's (1 1) 7rs t = Ad, + (1 - a,) B(dw - d5) (12) 7p, = Ad, - atB(d - dJ) (13) 7re = ca,(Ad.) + (1 - cx)(Ad,). Equation 13 implies that inflationary expectations in period t are a weighted average of the discretionary rates of inflation that would have been chosen by each policymaker under perfect information. Because the difference between the 448 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 desired and natural level of employment is greater for the weak policymaker than for the strong policymaker (dw - d5 > 0), equations 11 and 12 imply that imperfect information causes both policymakers to have equilibrium strategies that converge toward each other compared with the equilibrium strategies they would have under perfect information. When the value of the reputation indica- tor is one-half (c°, = 1/2), the planned rate of inflation of the weak policymaker decreases and that of the strong policymaker increases by the same amount compared with their full-information counterparts. As the value of the reputa- tion indicator increases above one-half, the tendency of the weak policymaker to move toward the strong one increases, and the tendency of the latter to move toward its weak counterpart diminishes. The converse happens when the value of the reputation indicator decreases below one-half. Because the marginal costs of being away from the desired level of employ- ment are higher the further away actual employment is from its desired level (equation 1), both types of policymakers partially accommodate inflationary expectations. This is reflected by the positive, but smaller-than-one, coefficient [A/ (A + 1)] of the rate of inflation expected at contracting time (ire) in equation 10. Because the expected rate of inflation is affected by what the public knows about the equilibrium strategies of both policymakers, the expected rate is some- where between the rates of inflation that the strong and the weak policymakers plan to generate. Because the latter rate is larger, it follows that the weak policymaker's planned rate of inflation will be lower than it would have been under perfect information. The strong policymaker's planned rate of inflation is higher than it would have been under perfect information. The public's uncer- tainty pulls the policies of the two types toward each other. As reputation increases, expectations approach the equilibrium strategy of the strong policy- maker. Hence, the tendency of the strong policymaker's planned rate of inflation to converge toward that of the weak policymaker diminishes, and the tendency of the weak policymaker's planned rate of inflation to converge toward that of the strong policymaker increases. Alternatively, as reputation diminishes, the tendency of the strong policymaker to compromise on the full-information strat- egy increases. As the reputation indicator approaches zero, the strong policy- maker's planned rate of inflation approaches Ad, + B(d - d5) (equation 11). The difference between this strategy and the strong policymaker's strategy under perfect information tends toward B(d., - dj. Thus, with poor credibility, the actual policy of a strong policymaker may very well resemble that of a weak policymaker. To assure that the strategies given in equations 10, 11, and 12 are indeed equilibrium strategies, it is necessary to assure that neither the strong nor the weak policymaker wants to deviate from the range in which the rates of inflation that could have been produced by the weak policymaker fully cover the range of rates that could have been produced by the strong policymaker. A sufficient condition for the weak policymaker not to want to deviate from this range is (14) a, > 7rp- ptfor all t, Cukierman and Liviatan 449 which, using equations 6 and 10, is equivalent to (15) a, > A (N` - N). Condition 14 requires that the imprecision of inflation control by the strong policymaker (measured by aJ) is larger than the difference in the equilibrium strategies of the two types in equation 10. When this is the case, the weak policymaker has no incentive to change the planned inflation rate in a way that would eliminate the full coverage of the strong policymaker's distribution by the weak policymaker's distribution. The inflation rate the weak policymaker plans to generate is greater than the rate the strong policymaker plans to generate for all periods. The weak policy- maker obviously does not have an incentive to reduce the planned inflation rate, because doing so would not reduce the probability of revelation but would increase the weak policymaker's expected costs in period t. The weak policy- maker also does not have an incentive to increase the inflation rate it plans to generate above its value from equation 10, because doing so would raise current expected costs and may increase the probability of revelation.1 Equation 15 restates this condition in terms of the fundamental parameters of the model and clarifies that the condition is independent of the time index, t. The condition basically requires that the imprecision of inflation control by the strong policy- maker be large compared with the difference between the desired employment objectives of the two policymakers. Partial- versus Full-Revelation Equilibria Unlike the weak policymaker, who has an incentive, other things being equal, to reduce the probability of revelation, the strong policymaker has an incentive to increase it. If revealed as such, the strong policymaker will reap the benefits of a good reputation during the entire future. But to increase the probability of the public's being able to separate the strong policymaker from the weak one, the strong policymaker's planned rate of inflation must be lowered. Thus the range of inflation rates that could be produced only by the strong policymaker would be widened. If, given the weak policymaker's planned rate of inflation, the strong policy- maker chooses a planned rate of inflation denoted by *; in figure 2, the distribu- tion of both policymakers' planned inflation rates would fully overlap, thus making the probability of sharp separation zero. But, by reducing the planned rate of inflation to Tp, the strong policymaker can create a positive probability 1. The implicit assumption underlying this statement is that both policymakers believe that off- equilibrium observations of inflation do not induce updating in the reputation parameter, a, and in expectations. The game theory literature refers to assumptions regarding the beliefs of players about off- equilibrium situations as "conjectures" and to the type of conjecture used here as a "passive conjecture" (Rubinstein 1985). 450 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 2. Partial- and Full-Revelation Equilibria a. A Partial-Revelation Equilibrium II AI A B it4 it,, itp C Inflation b. Revelation with Probability of One 4 t, p Inflation of sharp separation. In the figure this probability is measured by the ratio of linear segments AB/BC. In the extreme case, the strong policymaker can reduce the planned rate of inflation enough to create a complete separation (with a probability of one) between the two types. Figure 2b illustrates such a situation. This corresponds to the notion of a separating equilibrium in signalling theory as illustrated, for example, by the work of Vickers (1986). When Is Gradualism the Optimal Strategyfor a Strong Policymaker? Returning to the case of strictly gradual stabilization, as in figure 1, in order for the strategy of the strong policymaker in equation 10 to be optimal, it must dominate all strategies that yield positive probabilities of sharp separation. This occurs when the current cost of deviating from the strategy in equation 10 to a strategy that opens a "window of separation" is larger than the expected present discounted value of the benefits of separation. The present cost arises because, by lowering current inflation, the strong policymaker deviates from the optimal strategy of balancing the costs of low employment and high inflation in the current period. Sharp separation yields future benefits, however, because once he or she is recognized as strong, the policymaker enjoys the benefits of higher employment and lower inflation levels associated with a perfect reputation (CZ = 1). The present costs of partial or full separation are more likely to be higher than its future benefits, the larger the difference between the policymakers' imprecision of inflation control (au, - aJ) is compared with the difference in their desired levels of employment (N - N7-) and the lower the discount factor (f) is. In addition, it is likely that no separation will be attempted when the initial reputation is relatively high. The precise condition underlying this statement is presented in part 1 of the appendix in Cukierman and Liviatan (1991b). The policymaker's rate of tirme preference affects the likelihood that the cur- rent costs of separation are larger than its future benefits. The less the strong Cukierman and Liviatan 451 policymaker cares about the future, the larger the importance attributed to the current costs of separation. The role played by the relative sizes of the difference between the imprecision of inflation control by the weak and the strong policy- makers (a, - aj) and the difference between the weak and strong policymakers' desired levels of employment (Nw - N7) can be understood intuitively as fol- lows. The larger the difference in imprecision of inflation control compared with the difference in the desired level of employment, the larger the divergence between the strong policymaker's optimal strategy for the current period and the strategy necessary to produce a positive probability of separation. Hence, when the difference in imprecision of inflation control is large in relation to the differ- ence in the desired level of employment, the current costs of separation are more likely to be prohibitively high. Finally, when reputation is high to start with, the marginal future benefits of full revelation are small and therefore not worth the current costs of revelation. IV. FEATURES OF OPTIMAL GRADUAL STABILIZATIONS AND APPLICATIONS Even under the relatively favorable conditions in which a strong policymaker gets into office at the beginning of the stabilization process and remains there forever, it may be optimal to stabilize gradually. Speaking somewhat loosely, gradual stabilizations are optimal when there is a lot of noise in the control of inflation and when the policymaker, even if strong, has a high rate of time preference. Under these circumstances the reputation of the strong policymaker as a stabilizer rises gradually (see equation 9). The speed at which it rises depends on the precision of inflation control of the strong policymaker (as) in relation to that of the weak policymaker (a>). This relative precision is conveniently mea- sured by the ratio a5/a,. The lower this ratio, the higher the relative precision of the stabilizing policymaker (that is, the strong one) and the faster the rate of growth of reputation. In the limit, when the ratio approaches 1, the process of reputation building may take forever. At the other extreme, when the ratio approaches 0, reputation rises extremely rapidly. Figure 3a illustrates two paths of reputation building starting from a common initial reputation ax. In both paths, the process of reputation building is gradual, but it is faster when the ratio is lower. From equations 11 and 12, when a firm reputation is finally established, the average rate of inflation stabilizes at the level irsF _ Ads, which may be a reasonably low level, provided the difference between the strong policymaker's desired level of employment and the natural level of employment is sufficiently small. But during the period of reputation buildup even a serious stabilizer (that is, the strong policymaker) partially compromises on its perfect information inflation by accommodating some of the public's suspicions. The extent of accommodation can be measured as the difference between the decisions of a serious stabilizer with and without perfect information. This measure of accom- 452 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 3. Reputation, Accomodation, and Unexpected Inflation during Stabilization a. Evolution of Reputation after the Inception of Stabilization, with High and Low Relative Precisions Reputation 1 indicator, a Low a, la, High a. /a. aXO Time (t) b. Paths of Planned Inflation and of Accommodation through the Stabilization Process Plnndinflation IF Accomodation Time (t) c. Average Value of Unexpected Inflation through the Stabilization Process Mean unexpected inflation, E, (7r, - n't) 0 Time (t) modation is 7rst - F = [A2/(1 + A)] (d - d5)(1 - ,), where 7r sF is the strong policymaker's planned inflation rate under full information. Thus, the larger the divergence in the employment targets of a weak and a strong policymaker and the lower the current level of reputation, the larger the extent of accommodation. As the reputation of the serious policymaker rises, the extent of accommodation diminishes monotonically and tends to disappear altogether after a sufficiently large number of periods. Concurrently, average planned inflation decreases monotonically and finally converges toward the full- information low inflation rate (7r,F in figure 3b). Figure 3b shows the paths of planned inflation and accommodation during the stabilization process. The speed at which inflation and accommodation diminish is determined by the rate of increase in reputation, which is, in turn, faster the larger the relative precision of inflation control by the strong policymaker. Equation 13 and the fact that the difference between the desired and the natural level of employment is greater for the weak than for the strong policy- maker imply that inflationary expectations also decrease monotonically during Cukierman and Liviatan 4S3 the stabilization process and finally converge toward the low inflation rate that obtains in full information. The speed at which inflationary expectations de- crease is again directly related to the relative precision of inflation control by the serious stabilizer. In spite of the fact that stabilization is gradual, the serious policymaker has to accept a prolonged period of employment below normal because the average rate of inflation is consistently lower than expected. From equations 2, 11, 12, and 13, 7r, - =re [A/(1 + A)] (d_ - d)(1 - at) + Es Because the expected value of es is zero, the average discrepancy between actual and expected inflation is negative on average. It is large initially and decreases monotonically toward zero as the reputation of the policymaker im- proves. Figure 3c depicts the path of the average discrepancy between actual and expected inflation. The tighter the relative control of the strong policymaker over inflation, the faster the rate at which unexpected inflation converges to zero. Hence, a serious stabilizer who has a tighter relative control of inflation stabilizes with a shorter and less severe reduction in economic activity. Con- versely, when the policymaker's relative control is known to be loose, the reces- sion induced by stabilization is longer and deeper. An additional implication of the analysis concerns the dependence of the policies of both policymaker types on the evolution of reputation. From equa- tions 11 and 12, it can be seen that, as reputation rises (as a,t goes up), the equilibrium strategies of both types gradually shift to lower rates of inflation. The intuition is that, as the reputation indicator goes up, expectations get nearer to the equilibrium strategy of the strong policymaker and further away from that of the weak policymaker. Other things being equal, this situation reduces infla- tionary expectations. The fact that the policies of both policymakers are pos- itively related to expectations tends to reduce the planned rates of inflation of both types. Essentially, the increase in reputation and the associated reduction in inflationary expectations pulls both types toward lower rates of inflation. The possibility that gradualism may be the optimal strategy sheds new light on several recent stabilization efforts in various countries. At the end of the 1970s and through the first half of the 1980s the British government under Margaret Thatcher implemented a gradual stabilization program. During this time Britain went through a prolonged period of unemployment accompanied by increasing real wages. Sargent (1986b, p. 150) criticized this gradualist ap- proach on the grounds that it "invites speculation about future reversals, or U-turns, in policy." Sargent is right in the sense that our strong policymaker could produce a probability of one of separating himself from the weak policy- maker by deflating at a sufficiently low rate. As demonstrated in the previous section, however, this course of action is not necessarily optimal. For appropri- ate configurations of parameters like highly imprecise control of inflation and a reasonably high initial reputation, the best stabilization strategy is gradualism. A somewhat similar argument can be made about the disinflation process in Chile in 1974-77, where the rate of monetary growth fell rather gradually (in 1977 annual monetary growth was still above 100 percent). Contrary to a claim by Harberger (1981, 1982), a gradual reduction in monetary growth may be 454 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 consistent with a monetary crunch. Applying the principles of our model, one could argue that, under imperfect credibility, even the gradual reduction in the rate of monetary growth set by the policymaker may systematically exceed the public's expectations of this reduction. This will result in a situation of tight money. Indeed Edwards and Edwards (1987) and Corbo and Solimano (1989) conclude that monetary policy in Chile in the early stage of stabilization was contractionary. Ireland during the 1980s also partially fits the pattern of gradualism. After 1982, both fiscal and monetary policies in Ireland became substantially more restrictive. There ensued a gradual reduction in inflation and a prolonged in- crease in the rate of unemployment (Dornbusch 1989). However, it is likely that part of the increase in unemployment, particularly during the second part of the 1980s, can be attributed to factors other than those modeled in this article (Blanchard and Summers 1986). Our model provides a framework for analyzing stabilization policies in set- tings of moderate or high inflation, in which the main motive for inflation is high economic activity. But it does not seem to be appropriate for stabilization of hyperinflation, as experienced by Germany or Austria after World War I. Hyper- inflation is a result of an unsustainable fiscal situation in which the policymaker inflates mostly or only to finance the government's budget; historically, in all major hyperinflations the revenue motive was paramount. In such cases the structure of nominal contracts becomes so condensed that the short-run tradeoff all but disappears (a in equation 4 tends to zero), and any differences in em- phasis on economic activity between policymakers become inconsequential. The credibility of stabilization relates, in this case, to the ability of the policy- maker to finance expenditures in a sustainable manner (including the use of an inflation tax). This means that the issue is the credibility of the government's solvency. In such cases, credibility can be established (as in Sargent 1986a) more swiftly for two reasons. First, any potential differences between policymakers with regard to employment are relatively unimportant because of a shrunken Phillips tradeoff. Second, balancing of the budget (usually with foreign assis- tance) sends a clear signal that the major motive for inflation has been elimi- nated. This differs from the setting of our model, which deals with high, but not hyperinflationary, conditions, in which the credibility issue is not related to solvency, but rather to the inflation-unemployment tradeoff. In the latter setting, the gradualist solution becomes more relevant; budget balance alone does not establish quick credibility, because financing governmental expenditures is not the main motive for inflation. V. THE EFFECT OF FOREIGN AID ON THE CREDIBILITY OF STABILIZATION The path of inflation can be affected by external intervention in the form of foreign aid. An increase in the country's resources as a result of foreign aid may reduce the incentive of any type of policymaker to use surprise inflation to Cukierman and Liviatan 455 increase employment and output. For example, foreign aid, by increasing the availability of raw materials and physical capital, may increase the natural levels of employment and thereby reduce the difference between the desired and the natural level of employment. Such a reduction reduces the equilibrium rate of inflation under both types of policymakers. Foreign aid may, however, be tied more directly to the performance of the stabilization program. For example the foreign aid for any given year may be made conditional on inflation being below some target level, say T . If -x* is below rs + a, in figure 1, it is clear that each policymaker will have an incentive to reduce the planned inflation rate, thus leading to a lower inflation path. The presumption is that both policymakers are better off accepting the aid and the associated conditions than rejecting this package. However, even if T is above 7rs + a. (but below ir- + a.), the strong policymaker still has an incentive to reduce the planned inflation rate, provided the following condition holds: with- out the constraint on the inflation rate, the weak policymaker would have inflated at a higher rate, and the public understands this fact. As a result, the public's expectation of inflation goes down compared with its expectation in the case of no conditionality. From equation 10, this decline in the expectation of inflation induces the strong policymaker to lower the planned inflation rate even further. The reason is that the public understands that tying the aid to perfor- mance would constrain the behavior of a weak policymaker. Foreign aid conditionality will, when it is binding, motivate the weak policy- maker more than the strong policymaker to reduce the planned rate of inflation. Consequently, the rate of inflation the weak policymaker plans to generate will move closer to the rate the strong policymaker plans to generate, and the mo- tivation of the strong policymaker to separate will therefore diminish. Thus foreign aid conditionality will tend to make the gradualist solution more likely. However, the level of inflation with foreign aid conditionality will be lower on average. In practice, the foregoing form of conditionality is problematic because the policymaker may be induced to produce an artificially low inflation rate by means of price controls or by similar methods (such as setting an unrealistically low exchange rate or artificially reducing public sector prices). It will therefore be advisable to formulate a more robust kind of conditionality, one that makes explicit reference not only to inflation but also to fundamentals such as the size and composition of public sector expenditures and revenues, and the stock of money or domestic credit. VI. CONCLUSIONS Imperfect control of inflation fundamentally alters the dynamics of inflation, reputation, expectations, and economic activity during stabilization (see also, Cripps 1991). When the difference in the ability to control inflation of strong and weak policymakers is large, unexpected inflation may be persistently nega- 4S6 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 tive for a while, causing reduced economic activity and giving the indication that credibility is low. But if the policymaker persists with stabilization, this pattern gradually disappears. Imperfect control also leads to a generalization of the concept of separation because it creates situations in which the equilibrium policies of different types of policymakers diverge without necessarily inducing clear-cut separation. Imperfect control of inflation is the result of factors that are related to the structure of the economy and of policymaking institutions. In particular, it is likely that the lower the degree of independence of the central bank, the lower the precision of inflation control (Cukierman 1992, chapter 18). When the fundamental cause of inflation inertia is imperfect information about the objectives of policymakers, there is no gradualism without pain. This contrasts with the role of gradualism when the basic reason for inertia is backward-looking nominal contracts. It is possible to devise, in such cases, patterns of gradual disinflation that eliminate the employment costs of stabilization. APPENDIX. DERIVATION OF EQUILIBRIUM STRATEGIES AND EXPECTATIONS UNDER GRADUALISM (EQUATIONS Io TO 13) To simultaneously solve for the equilibrium strategies and for expectations, we use the method of undetermined coefficients. In particular, we postulate that the equilibrium strategies of the two types of policymakers can be represented as the following two functions of di, i = w,s, and of expectations (A-1) irs = kdds+ ke.xi 7rw = rddw + re,7r where kd, ke, rd, and re are unknown coefficients to be determined. It can be shown that only the current expectation belongs in the solution and that, given linearity and decomposability, the solution is therefore unique. There also are history-dependent trigger strategies, but we rule them out because of their lim- itations and lack of descriptive realism. Discussions of the coordination problem and other problems of trigger strategies appear in Rogoff (1987, 1989) and Cukierman (1992, chapter 11). The public knows the decision rules in equation A-1 but is uncertain about the identity of the policymaker in office. Hence, inflationary expectations are given by (A-2) 7re C= !,1r- + (1 - a,)7C = a,[kddS + kete,] + (1 - a,)[rddW + reweJ]. Because the dynamic optimization problem in equation 8 reduces to a series of one-period problems, the equilibrium strategy of policymaker i at time t can be characterized by solving (A-3) min Ei,z(d', ipt + et, W'), i W,S where Eit denotes the information available to policymaker i when the policy- Cukierman and Liviatan 4S7 maker picks the planned rate of inflation for the period. The superscript i attached to the expected value denotes the fact that the information sets of the two policymakers differ. Equation 10 is obtained from the first-order condition for the problem in A-3 and by using equation S and the fact that, when the policymaker picks 7r;p, the policymaker takes ir as given. Equating the coeffi- cients of di and ire across equations A-1 and 10 yields (A-4) kd = rd = ke = r, = [A/(1 + A)]. Equation 13 follows by using A-4 in A-2. Equations 11 and 12 follow by letting B = [A2/(1 + A)]. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Andersen, T. M., and 0. Risager. 1987. "The Role of Credibility for the Effects of a Change in the Exchange Rate Policy." Seminar Paper 377. IIES, University of Stock- holm. Processed. . 1988. "Stabilization Policies, Credibility, and Interest Rate Determination in a Small Open Economy." European Economic Review 32: 669-79. Backus, David, and John Driffill. 1985a. "Inflation and Reputation." American Eco- nomic Review 75: 530-38. . 1985b. "Rational Expectations and Policy Credibility Following a Change in Regime." Review of Economic Studies 52: 211-21. Barro, R. J. 1986. "Reputation in a Model of Monetary Policy with Incomplete Informa- tion." Journal of Monetary Economics 17 (January): 3-20. Barro, R. J., and D. B. Gordon. 1983. "A Positive Theory of Monetary Policy in a Natural Rate Model." Journal of Political Economy 91 (August): 589-610. Blanchard, O., and L. Summers. 1986. "Hysteresis and the European Unemployment Problem." NBER Macroeconomics Annual 1:15-78. Canzoneri, M. 1985. "Monetary Policy Games and the Role of Private Information." American Economic Review 75: 1056-70. Corbo, Vittorio, and Andres Solimano. 1989. "Chile's Experience with Stabilization Revisited." The World Bank, Country Economics Department, Washington, D.C. Processed. Cripps, M. 1991. "Learning Rational Expectations in a Policy Game." Journal of Eco- nomic Dynamics and Control 15: 297-1S. Cukierman, Alex. 1992. Central Bank Strategy, Credibility, and Independence: Theory and Evidence. Cambridge, Mass.: MIT Press. Cukierman, Alex, and Nissan Liviatan. l991a. "Optimal Accommodation by Strong Policymakers under Incomplete Information." Journal of Monetary Economics 27: 99-127. . 1991b. "The Dynamics of Optimal Gradual Stabilizations." Working Paper 34-91. Tel Aviv University, The Foerder Institute for Economic Research, Department of Economics. Processed. 458 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Dornbusch, Rudiger. 1989. "Credibility, Debt, and Unemployment: Ireland's Failed Stabilization." Economic Policy 8 (April): 173-210. Edwards, Sebastian, and A. C. Edwards. 1987. Monetarism and Liberalization-The Chilean Experiment. Cambridge, Mass.: Ballinger. Engle, R. F. 1982. "Autoregressive Conditional Heteroscedasticity with Estimates of the Variance of U.K. Inflation." Econometrica 50 (July): 987-1007. Harberger, A. C. 1981. "Comment on Foxley." In W. R. Cline and S. Weintraub, eds., Economic Stabilization in Developing Countries. Washington, D.C.: Brookings Institution. .1982. "The Chilean Economy in the 1970's: Crisis, Stabilization, Liberalization, Reform." In K. Brunner and A. H. Meltzer, eds., Carnegie Rochester Conference Series on Public Policy, 17. Kiguel, M. A., and Nissan Liviatan. 1988. "Inflationary Rigidities and Stabilization Policies." wps 4. World Bank, Country Economics Department, Washington, D.C. Processed. Kydland, F. E., and E. C. Prescott. 1977. "Rules Rather than Discretion: The Inconsis- tency of Optimal Plans." Journal of Political Economy 85 (June) 473-591. Persson, Torsten, and Sweder van Wijnbergen. 1987. "Signalling, Wage Controls, and Monetary Disinflation Policy." World Bank, Washington, D.C. Processed. Rogoff, Kenneth. 1987. "Reputational Constraints on Monetary Policy." In K. Brunner and A. H. Meltzer, eds., Carnegie-Rochester Conference Series on Public Policy, 26, Spring, 141-82. 1. 989. "Reputation, Coordination and Monetary Policy." In Robert Barro, ed., Modern Business Cycle Theory. Cambridge, Mass.: Harvard University Press. Rubinstein, A. 1985. "Choice of Conjectures in a Bargaining Game with Incomplete Information." In A. E. Roth, ed., Game-Theoretic Models of Bargaining. Cambridge University Press. Sargent, T. J. 1986a. "The Ends of Four Big Inflations." In T. J. Sargent, ed., Rational Expectations and Inflation. New York: Harper and Row. .1986b. "Stopping Moderate Inflations: The Methods of Poincare and Thatcher." In T. J. Sargent, ed., Rational Expectations and Inflation. New York: Harper and Row. Vickers, J. 1986. "Signalling in a Model of Monetary Policy with Incomplete Informa- tion." Oxford Economic Papers 38: 443-55. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 459-479 Uniform Commercial Policy, Illegal Trade, and the Real Exchange Rate: A Theoretical Analysis Stephen A. O'Connell Countries on fixed exchange rates sometimes use uniform tariff cum subsidy (UTcs) schemes as a way of achieving a real depreciation without disturbing the nominal exchange rate. A potential drawback of this policy in relation to an across-the-board devaluation is that a UTCS scheme provides incentives for illegal trade. Using an opti- mizing model with currency convertibility and illegal trade, Ifind that welfare is lower under a uTcs scheme than under a corresponding across-the-board devaluation and that in some cases the real exchange rate actually appreciates in response to an increase in the uTcs rate. For countries on fixed exchange rates, achieving a real depreciation through macroeconomic policy alone may be highly contractionary if preexisting wage contracts or other rigidities prevent rapid reduction of domestic costs in relation to foreign costs. When competitiveness needs to be improved, but policy consid- erations prevent the use of the nominal exchange rate, a natural alternative is the uniform tariff cum subsidy (UTCS). By applying a constant ad valorem tariff to all imports and an equal subsidy to all exports, a UTCS achieves the same rapid adjustment in the domestic prices of traded goods that would be implied by a parity change. Laker (1981) studied the use of uniform trade taxes and closely related fiscal proxies for devaluation in France (1957-58), Israel (1955-62), India (1963- 66), and the Federal Republic of Germany (1968-69). UTCS schemes have been used more recently in a number of developing countries, including Ghana, Sudan, and Senegal. A recent case in point is provided by Cote d'Ivoire. C6te d'Ivoire is a member of the West African Monetary Union, a group of countries whose common currency (the CFA franc) is freely convertible into Stephen A. O'Connell is with the Department of Economics at Swarthmore College, Swarthmore, Pa. This article is a revision of World Bank PPR Working Paper 185 (see O'Connell 1989). The author wishes to acknowledge his indebtedness to the Macroeconomic Adjustment and Growth Division of the Country Economics Department at the World Bank for its hospitality; to the Council on Foreign Relations for its financial support; and to Bill English, Ravi Kanbur, Saul Lizondo, Dani Rodrik, Oren Sussman, and an anonymous referee for their helpful comments. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 459 460 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO 3 French francs by agreement with the French Treasury, which guarantees convert- ibility by extending overdraft privileges to the Union's Central Bank (Krumm 1985). Increases in domestic inflation starting in the mid-1970s, together with the recent nominal appreciation of the French franc and significant nominal depreciations in neighboring Ghana and Nigeria, have produced real apprecia- tions in C6te d'Ivoire and a number of other CFA countries. This has led several authors (Krumm 1987; Devarajan and de Melo 1987; Leiderman 1987) to suggest that an optimal macroeconomic policy package would include a deval- uation of the nominal exchange rate were it not for C6te d'lvoire's responsi- bilities to the CFA Zone, where the exchange rate against the French franc has been fixed since 1948. The government of C6te d'Ivoire implemented a UTCS scheme in mid-1986 with the precise objective of achieving a rapid increase in competitiveness. Initial experience with the UTCS in Cote d'Ivoire has not been encouraging. There is anecdotal evidence of widespread underinvoicing of imports, and in some cases the government has been substantially late in paying the export subsidy. The Ivoirien experience suggests a potentially important asymmetry between devaluations and UTCS schemes: UTCS schemes, unlike devaluations, provide incentives for illegal trade. This article analyzes the implications of this asymmetry for welfare and resource allocation within the framework of a com- monly used general equilibrium macroeconomic model. My findings support Laker's (1981) contention that illegal trade may significantly weaken the case for uniform trade taxes as proxies for devaluation. I. UTCS SCHEMES AND DUAL EXCHANGE RATES In the literature about trade theory, the close formal relationship between UTCS schemes and devaluations was pointed out by Meade (1951) and, later, by Bhagwati (1968) and Dixit and Norman (1980). Abstracting from issues of implementation, the key difference between the two policies is one of coverage. A recent treatment is given by Adams and Greenwood (1985), who derive a precise equivalence result between uniform trade taxes and exchange rate changes. They find that in a frictionless, competitive environment, a UTCS scheme covering trade in goods and services is fully equivalent to a devaluation of the commercial rate in a dual exchange rate system. This asymmetry of coverage has sometimes been regarded as a primary virtue of UTCS schemes. Keynes, for example, favored a UTCS as a way to address the overvaluation of sterling in the early 1930s. He believed a UTCS scheme would improve competitiveness without "the injury to the national credit and to our receipts from foreign loans fixed in terms of sterling which would ensue on devaluation" (quoted in Laker 1981, p. 118). For the typical developing coun- try, however, foreign assets and liabilities are denominated in foreign currency, so Keynes' argument does not apply. UTCS schemes therefore represent a particularly simple departure from unified O'Connell 461 exchange rates. As Adams and Greenwood (1985) show, this departure does not imply an asymmetry between the effects of UTCS schemes and across-the-board devaluations, provided that the private sector (correctly) views the policies in question as permanent. With flexible prices and perfect capital mobility, neither a permanent devaluation of the commercial rate nor a permanent across-the- board devaluation has real effects. Both policies produce once-and-for-all offset- ting adjustments in domestic prices and an immediate rebuilding of money balances through a balance of payments surplus. With perfect capital mobility and a fixed exchange rate, the rebuilding of money balances simply requires an exchange of foreign bonds for money between the private sector and the central bank (see Obstfeld 1981, 1986 and Adams and Greenwood 1985). Asymmetries emerge in this idealized environment only when the policies are expected to be temporary. Adams and Greenwood (1985) and Frenkel and Razin (1989) show that, although a temporary across-the-board devaluation has no real effects, a temporary UTCS or devaluation of the commercial rate lowers the real interest rate facing the private sector and is therefore equivalent to a subsidy on foreign lending. Actual experience with UTCS schemes suggests a further set of differences between UTCS schemes and across-the-board devaluations, differences arising from the close proximity of the former to traditional commercial policy (see Laker 1981). Unlike changes in the exchange rate, for example, UTCS schemes require an administrative apparatus for determining the values of traded goods. Moreover, the political and administrative aspects of policy implementation leave more room for rent seeking and discriminatory application in the UTCS case than in the devaluation case. Finally, as emphasized in this article, UTCS schemes drive a wedge between the domestic and foreign prices of traded goods; illegal trade therefore becomes a potentially important source of asymmetry between the two policies. Although I will be focusing on the differences between UTCS schemes and across-the-board devaluations, the basic insights of that analysis carry over, in modified form, to the "equivalent" dual exchange rate policy. Thus, although dual (or multiple) exchange rates do not introduce direct incentives for simple smuggling or customs fraud, these systems provide indirect incentives for illegal trade as part of more complicated attempts to arbitrage between different ex- change rates. A permanent devaluation of the commercial rate, for example, encourages the overinvoicing of exports and underinvoicing of imports, with the (purportedly unrelated) residual payment taking place at the more appreciated capital account rate. The exact equivalence between trade tax regimes and departures from unified nominal exchange rates is broken, however, because even without other distortions the microeconomics of illegal activity varies, depending on whether a given structure of relative prices is achieved through different exchange rates or trade taxes. Because arbitraging between dual ex- change rates requires the completion of at least two separate illegal transactions, it would appear that the dual exchange rate alternative would typically provide 462 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 smaller incentives for illegal trade than the otherwise equivalent UTCS. This remains an open question, however, because the relationship between the two alternatives depends on the details of evasion technologies and enforcement policy. II. AN EQUILIBRIuM FRAMEWORK BASED ON THE STANDARD MODEL OF A DEPENDENT ECONOMY I will examine the implications of illegal trade using versions of the standard model of a dependent economy. Although the model is readily adaptable to short-run analysis, I will assume perfectly flexible wages and prices and there- fore abstract from the short-run macroeconomic effects that would arise with nominal rigidities. I will also restrict the examination to one-period models, thereby dropping the distinctions between temporary and permanent policies and between across-the-board devaluations and devaluations of the commercial rate. And because the existence of illegal trade does not give rise to a black market in foreign currency as long as the capital account is open and free convertibility is maintained (with "no questions asked" about the source or destination of foreign exchange obtained by private individuals), I will exploit these features of the Cote d'lvoire case and will leave monetary and portfolio issues out altogether (see Pitt 1984, Macedo 1987, and Branson and Macedo 1989 for models in which tariffs and export taxes, combined with controls on foreign exchange or capital, give rise to a black market). I begin by reviewing the basic structure of the two-sector dependent economy model (Dornbusch 1980). The economy produces and consumes nontraded goods (N), importables (M), and exportables (X). The country is sufficiently small in world markets that the international prices of traded goods can be treated as exogenous. Given the import tariff and export tax (or subsidy) rates, the domestic prices of traded goods are also fixed, and importables and export- ables can be consolidated into a single composite "traded good." The associated price index, PT, is a linearly homogeneous function of the domestic prices of importables and exportables; in the case in which the import tariff and export subsidy are both equal to z ' °, PT is given by (1) PT =f(Px,PM) =f[E(l + z)P",E(l + z)P*] E(1 + z) where E is the nominal exchange rate, an asterisk denotes an international price, and world prices Px and P* are set equal to unity. The real exchange rate, e, is defined as the domestic relative price of nontraded to traded goods: e PNIPT = PN/e(l + z). I assume that labor is perfectly mobile in the period under analysis but that all other factors of production are specific to the sectors in which they are initially located. This is the Ricardo-Viner model of trade theory (see Dixit and Norman 1980 and, for a dependent economy application, van Wijnbergen 1986). The production functions qN(LN) and qT(LT) have the standard properties: q(0) = 0, O'Connell 463 q' > 0 and q" < 0, with the latter property (diminishing returns to labor) implied by the fixed supplies of sector-specific factors. For the moment, I also assume that q' (0) = X in order to ensure that even a small UTCS elicits a smuggling response; I generalize this to q' (0) > 0 in the discussion at the end of the section. Wages and prices are flexible, so that full employment prevails; perfect mobility of labor implies that a single economywide nominal wage pre- vails in equilibrium. Equilibrium can be described in a straightforward manner using revenue and expenditure functions (Dixit and Norman 1980). Because all activities are purely competitive, labor will be allocated to maximize domestic revenue from the two activities. Denoting the maximized value of revenues by R and assuming a fixed total supply of labor, L, R(PT,PN;L) maximizes PTqT + PNqN subject to the resource constraint LT + LN C L. Because R is homogeneous of degree 1 in all prices, deflating by the domestic price of traded goods yields real revenue, R/PT = r(l,e;L). The revenue function r has the property that its partial deriva- tives are the supply functions for traded and nontraded goods. Turning to the expenditure side, aggregate expenditure on traded and non- traded goods is simply PTCT + PNCN, where cj (j = T,N) is consumption of good j. The minimized value of expenditure for any social utility level U is denoted by Z(PT,PN;U). Because Z is homogeneous of degree 1 in all prices, the expenditure function can be written in terms of traded goods as Z I T = E(l,e;U); the partial derivatives of e are the compensated demand functions for traded and nontraded goods. e and r are related by the budget constraint e = r - T, where r is the real value of lump-sum taxes paid by the private sector. Government finances are equally straightforward. The government receives tariff revenue from imports, pays out a subsidy to exports, and collects lump- sum taxes (which may be negative) from the private sector. Because the subsidy rate on exports is the same as the tariff rate on imports, budget balance requires that lump-sum taxes equal the official trade balance in foreign currency, b0, multiplied by the UTCS rate: (2) T = ZEbo/PT = [z/(l + z)]b0. Equilibrium without Illegal Trade Without illegal trade, the official trade balance equals the overall trade bal- ance, which is zero in this one-period model. A UTCS scheme therefore has no net budgetary effect, so that T = 0. It would be straightforward to incorporate government spending on traded and nontraded goods in the amounts gT and gN; in this case, equation 2 would read r = g + zEPTb0/PT = g + IZ/(1 + z)]b0, where g = egN + gT- Equilibrium prevails when there is full employment (LT + LN L), the nontraded goods market clears (e[J1,e;U] = re[l,e;U]), and the private sector and government satisfy their budget constraints (e = r). Because full employ- ment is already subsumed in the revenue function, the latter two equations fully 464 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 1. Equilibrium without Illegal Trade Traded goods consumption, CT, and production, Slope = -e Nontraded goods consumption, cN, and production, q determine the equilibrium values of welfare and the real exchange rate (the equation for a zero trade balance is implied by these two). Figure 1 shows the familiar dependent economy diagram, with equilibrium at point 1. The key feature of the "legal trade" equilibrium is that neither the nominal exchange rate nor the UTCS rate affects the equilibrium value of welfare or the real exchange rate. This is apparent in the fact that E and z do not enter the equilibrium conditions e, = re and e = r. A nominal devaluation or a rise in the UTCS rate simply produces an equiproportional rise in wages and the prices of nontraded goods, with no change in the real exchange rate. Incorporating Illegal Trade Illegal trade complicates the analysis in two ways. First, because the activity uses real resources, the economy now has at least three sectors (production of traded goods, production of nontraded goods, and production of smuggling services). Second, although uniform taxes are designed to leave the domestic relative prices of traded goods unchanged, they may fail to do so with illegal trade. When this happens, importables and exportables can no longer be consol- idated into a single traded good. Section III focuses on the first of these complications, following Bhagwati and Hansen (1973) in assuming that legal and illegal trade take place independently. Under this assumption, smuggling by competitive firms does not affect the domestic prices of traded goods unless the marginal costs of smuggling are so low that legal trade is driven out completely in equilibrium. This means that when the two types of trade coexist, as is normally observed in practice (Pitt O'Connell 465 1981; Thursby, Jensen, and Thursby 1990), one can continue to consolidate importables and exportables into a single good. The resulting three-sector smug- gling model is simple and intuitively appealing. Although the smuggling model is appropriate for certain kinds of illegal trade (such as border smuggling), there are important categories of illegal trade (such as underinvoicing of imports or overinvoicing of exports) that use legal activity as a "cover" for illegal activity. In these cases, equilibrium is characterized by what Pitt (1981) called "price disparity": the domestic price of the imported (exported) good falls below (above) the full tariff-inclusive (subsidy-inclusive) price. Section IV, therefore, uses a model of invoicing fraud to illustrate the second channel-the domestic relative price of traded goods-through which illegal trade can affect welfare and the real exchange rate. III. SMUGGLING Smuggling firms use domestic resources to bring goods past the customs au- thorities. Four assumptions characterize the smuggling activity. First, smuggling services are produced by a combination of labor and factors of production specific to smuggling. Second, the cost of an additional unit of smuggling, for the smuggling industry as a whole, is an increasing function of the total amount being smuggled. Third, smuggling firms behave as price takers in the domestic market for traded goods. Fourth, the marginal cost of smuggling at the industry level rises rapidly enough that legal trade is not driven out completely in equilibrium. The first of these assumptions implies that smuggling firms must compete with domestic producers for labor, so that both traded and nontraded goods are implicitly used up in producing smuggling services. Smuggling costs are there- fore not characterized as denominated solely in either the good(s) being smug- gled (Bhagwati and Hansen 1973; Pitt 1981; Martin and Panagariya 1984) or nontraded goods (Sheikh 1974). Two kinds of market structure are consistent with the second and third as- sumptions. As in the Bhagwati-Hansen analysis, one can assume that although individual smuggling firms face constant costs per unit, the smuggling sector as a whole incurs increasing costs because of intra-industry, interfirm diseconomies of scale. The link between firm-level average costs and aggregate smuggling is the result of "congestion" on illegal trade routes; marginal cost pricing emerges as the result of price competition under conditions of constant returns at the firm level. The Bhagwati-Hansen analysis implies zero profits for the smuggling sector in equilibrium. An alternative structure, in which smuggling firms are profitable in equilibrium, is one in which there is a finite number of smuggling firms, and each faces an increasing marginal cost of smuggling. In this case, it must be assumed that the number of firms, while finite, is large enough that marginal cost pricing is a good approximation to their actual strategic behavior. Although 466 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 marginal cost pricing is the case usually analyzed in the smuggling literature, the appropriateness of this assumption clearly depends on the underlying strategic environment. If smuggling firms compete over prices as Cournot rivals, for example, the smuggling industry will approach marginal cost pricing as the number of firms increases (Thursby, Jensen, and Thursby 1990). Tirole (1988) discusses the conditions under which Cournot behavior can be derived as the equilibrium of a two-stage game in which competitive suppliers first choose quantities (in our case, first smuggle a certain amount into the country) and then compete over prices. The four assumptions make it possible to incorporate illegal trade by simply appending a third, "smuggling" sector (subscript S) to the standard analysis. To do this, let qs(Ls) be the total amount of smuggling, measured in terms of the composite traded good. The total amount of smuggling is the sum of import- ables and exportables smuggled into the country to avoid the tariff or receive the subsidy (an export subsidy provides an incentive for illegal importing of export- ables for reexport or sale on the domestic market). Each unit smuggled yields a net revenue equal to the difference between the domestic and the world price of the good. Because smuggling occurs under increasing marginal costs at the industry level and legal trade is not driven out in equilibrium, the marginal source of supply of the traded good is legal trade, and the domestic price of the traded good will simply be the world price inclusive of the uniform tariff or subsidy rate. Real net revenue from a unit of smuggling is therefore given by (PT- EP ) z PT (1+Z) As in the previous section, labor will be allocated under perfect competition so as to maximize net domestic revenue from the three activities. The revenue function therefore takes the form r[1,e,z/(1 + z);L]. On the expenditure side, because illegal trade leaves the price of domestic tradables unaffected, the expen- diture function is still given by e(1,e;U). The expenditure function and the revenue function are related through the private sector's budget constraint, e = r - r. The previous section showed that a UTCS scheme that is implemented with a zero trade balance will have no net effect on government revenues without illegal trade. With illegal trade, however, this is no longer true: a UTCS scheme will involve a net transfer of resources to smugglers that must be made up by higher lump-sum taxes. To see this, first note that the requirement of balanced overall trade in foreign currency terms implies (3) XO = MS + mO + XS where x and m denote the quantity of goods exported and imported, respec- tively, and the subscripts 0 and S denote official and smuggled quantities, respectively. Smuggled exports (x5) are included among imports (as well as being part of official exports) because these are reexports illegally brought in from O'Connell 467 other countries to take advantage of the subsidy. Equation 3 implies that the official trade surplus equals the total value of smuggling: (4) bo = x. - mO = ms + Xs = qs Because lump-sum taxes, as before, equal the official trade surplus multiplied by the UTCS rate, equation 4 implies that - = [z/(l + z)]qs. Lump-sum taxes are exactly equal to net revenues from illegal trade. Equilibrium Because qs = r3 ( ear/a[z/(l + z)], the real value of taxes can be written as r = [z/(l + z)]r3, and equilibrium can be characterized completely using the following two equations: (5) e(1,e;U) = rI1,e,( ± Z);l -( + r (6) ee(l;e,U) = reL1,e,(l +);L Equation 5 states that, as the labor market clears, the economy is on its overall budget constraint, and the government budget constraint is satisfied; equation 6 is the market-clearing condition for nontraded goods. Equations 5 and 6 jointly determine the real exchange rate and the social utility level as functions of the fixed total supply of labor and the UTCS rate (the third equilibrium condition, that the trade balance be zero, is implied by these two). As before, the nominal exchange rate does not appear in the equations. Any devaluation is immediately eroded by an equiproportional rise in wages and the price of nontraded goods, with no real effects. When the smuggling activity is prohibitively costly, so that r3 = q, e 0, the model reduces to the standard dependent economy model, in which the real exchange rate and the social utility level are determined independently of the UTCS rate. Without smuggling, there- fore, a (permanent) UTCS has no real effects. When smuggling is present, however, changes in the UTCS rate do have real effects. From equation 5, smuggling affects overall welfare by lowering social disposable income. It is not surprising that a UTCS scheme is unambiguously welfare-worsening, because a rise in the UTCS rate increases the incentives for smuggling. As proved in the appendix and illustrated in figure 2, with a positive UTCS rate (z > 0 in figure 2), any increase in the UTCS rate draws resources out of productive activities and lowers welfare. Moreover, starting from a VTCS rate of zero (z = 0 in figure 2), any increase in the UTCS rate that is not vanishingly small will lower welfare. This is not true for across-the-board devaluations (which leave welfare unchanged); thus there is a well-defined sense in which illegal trade tends to render uniform trade taxes inferior, or "second-best," to the policy they are intended to proxy. A related question is whether smuggling itself worsens welfare with uniform 468 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 2. Equilibrium with Smuggling (Real Appreciation Case) Traded goods consumption, CT, and production, qT E=r-[z/(l+z)lr3 _ z > Nontraded goods consumption, CN, and production, z& Note:- A very sarri increase in z starting at z - O does not affect welfarxe or the real exdihnge rate to a fulst approxirration. trade taxes: for a given UTCS rate, is welfare higher or lower with smuggling? The answer is again unambiguous: smuggling lowers welfare (as in Bhagwati and Hansen 1973, when legal trade is not fully displaced). The reason is that the UTCS scheme is itself not a distortion, so that the loss of productive resources as a result of smuggling occurs in an undistorted economy. If the focus had been, instead, on nonuniform commercial policies (such as import tariffs alone), the possibility of welfare gains might have emerged, because in this case the costs of illegal activity would have been offset to some degree by an amelioration of the policy distortion (Sheikh 1974; Pitt 1981; Deardorff and Stolper 1990). lThe appendix shows that starting from a positive UTCS rate, the real exchange rate appreciates (rises) or depreciates (falls) according to the following criterion: (7) = - as - = dz < PT3 < AT where A, is the income elasticity of demand for good j and V.3 is the cross elasticity of supply in sector j with respect to Iz /(1 + z)] (as with welfare, a vanishingly small UTCS rate that is introduced starting from z = O will leave the real exchange rate unchanged). A change in the UTCS rate therefore affects both supply and demand, as illustrated in figure 3. A rise in the UTCS rate drives up the economywide wage and draws labor into the smuggling activity; the effect on the relative supply of nontraded to traded goods, qNlqT, depends on the relative cross elasticities of supply of these goods with respect to a rise in z/(1 + z). The relative supply curve (RS in figure 3) shifts to the left (toward traded goods) if the supply O'Connell 469 Figure 3. Real Exchange Rate Determination with Smuggling Real exchange rate, e RS (z > O) ss \ RS (z 0) RdD (z 0 ) RD (z > O) Relative supply of nonuraded to traded goods, qv IqT, and relative dernand of nontraded to traded goods, CN ICT Vote: A very srnall increase in z starting at z - O does not affect the real exchange rate to a first approxirnation. response is higher in nontraded goods and to the right if the response is higher in traded goods (equivalently, relative supply shifts toward the sector with a smaller elasticity of supply with respect to its own product wage, measured at the original real exchange rate). Relative supply is unchanged if UN31V13= Ohn the demand side, the movement of labor into smuggling produces a fall in real disposable income. The effect of this on the relative demand curve (RD in figure 3) depends on relative income elasticities. The relative demand curve shifts to the right if ttN /1T < 1 because in this case a fall in income produces a shift in demand toward nontraded goods; the relative demand curve shifts to the left if nontraded goods have the higher income elasticity. With homothetic pref- erences (AN = PLT), there is no effect on relative demand. Equation 7 suggests that with illegal trade a UTCS scheme designed to achieve a real depreciation may have the opposite effect. Starting at point 1 in figure 3, where the UTCS rate is zero, a real appreciation occurs (point 2 in figure 3) if the income elasticity of demand for nontraded goods is relatively low or if the cross elasticity of supply of nontradables with respect to the UTCS rate is relatively high in nontraded goods compared with traded goods. InFcome Distribution Effects Resources devoted to smuggling end up engineering a transfer of income from taxpayers to smugglers that is mediated through the government budget. Assum- ing that taxes are lump-sum, the fact that these redistributions take place through the government budget has no effect on welfare or the real exchange rate. In practice, however, governments do not have access to nondistortionary 470 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 tax instruments. The reduction in public sector revenues therefore requires some combination of increases in other distortionary taxes and cuts in public sector expenditures. With respect to welfare, as long as the shadow value of increased government revenue is positive, the increased budgetary stringency associated with a UTCS scheme lowers welfare. Because these budgetary effects do not occur under a devaluation, this situation further undermines the relative efficacy of uniform taxes. Effects on the real exchange rate depend on where in the budget, and when, the required adjustment takes place. Suppose, for example, that trade taxes are the only tax instrument available to the government. The government budget constraint would then imply that either current government expenditure or (in a dynamic setting) future trade tax rates or expenditure must adjust to the fall in current trade tax receipts. If current government expenditure bears the burden, there will be upward or downward pressure on the real exchange rate, depending on whether the private sector's marginal propensity to consume non- traded goods is above or below that of the public sector. Two Remarks on UTCS Schemes and Illegal Trade The following two remarks elaborate on the comparison between the effects of UTCS schemes and devaluation. Remark 1: High uniform tax rates and immobile labor. Although this article has emphasized what appears to be an inexorable link between uniform trade taxes and illegal trade, in practice this link may only come into play for suffi- ciently high uniform tax rates. In particular, I have assumed that there are no fixed costs to initiating the smuggling activity and that the marginal product of labor in smuggling is infinite at LS = 0. These assumptions imply that any finite UTCS rate, regardless of how small, will move labor into smuggling. With fixed costs, however, the UTCS rate would have to reach some critical minimum level (z > 0) before there would be any smuggling response. The same would hold if the marginal product of labor were finite at Ls = 0. In either case, the second- best nature of UTCS schemes as proxies for devaluation would emerge only for a sufficiently large UTCS rate. There would also be no real effects from a UTCS scheme in the short run if labor were immobile. With immobile labor, nominal wages in all sectors would adjust in the short run to maintain real product wages at their original levels, and the price of nontraded goods would rise in direct proportion to the increase in traded goods. Real effects would begin to emerge only as labor began to move between sectors in response to real wage differentials. Remark 2: Temporarily sticky wages and prices. Although the asymmetry between UTCS schemes and devaluations is a fundamental feature of medium- to long-run equilibrium, the analysis is more complicated if wages and prices are temporarily sticky. In this case, changes in the exchange rate can alter the relative return to legal and illegal activities in the short run, provided that the O'Connell 471 costs of smuggling are not denominated completely in traded goods. For exam- ple, suppose that the price of nontraded goods is flexible, so that the market for nontraded goods always clears, but that unemployment arises from a sticky economywide nominal wage (that is, the economy is on the border of the Keyne- sian and classical unemployment regions in a disequilibrium framework). Now consider a rise in the UTCS rate or nominal exchange rate that achieves a given increase in the domestic price of traded goods, PT = E(1 + z), on impact. Because these policies lower the product wage in the traded goods sector, wiE(l + z), by the same amount, they lead to the same increase in the demand for labor there. The UTCS, however, has a greater effect on overall labor demand because it lowers the product wage in the smuggling sector, w/Ez, by a larger amount. The short-run employment response is therefore larger when the change in the prices of traded goods is achieved through a UTCS than when it is achieved through a devaluation. Both policies produce an increase in employment and income, together with expenditure switching toward nontraded goods at the initial price of nontraded goods. The price of nontraded goods will therefore rise to clear that market, leading to a further increase in the demand for labor as the product wage in the nontraded goods sector falls. The final increase in employ- ment and rise in the price of nontraded goods will be larger for the UTCS, however, given the stronger impact on employment in smuggling. The overall welfare comparison of the two alternatives in the short run is unclear: the UTCS creates more employment, but it has more smuggling and therefore a higher resource cost. Of course, the asymmetry emphasized earlier emerges over time as wages and prices adjust to market-clearing levels. IV. INVOICING FRAUD A channel for tariff avoidance that is important in a number of developing countries is the underinvoicing of imports. Bhagwati and Desai (1970), for example, report significant overinvoicing of exports in India in response to export incentives that were introduced along with import restrictions between 1963 and 1966 to simulate a devaluation (Laker 1981; see, also, Pitt 1981 and Bhagwati 1974). Although the use of official reference prices or specific tariffs would seem an easy solution to the problem of underinvoicing, implementation of realistic reference price systems may be very costly for nonhomogeneous imports. Cote d'Ivoire, for example, at the time the UTCS scheme was intro- duced, was already getting rid of specific tariffs because of their inefficiencies. In addition to these difficulties, solving the underinvoicing problem will increase the incentive for direct smuggling. Although invoicing fraud (including overinvoicing of exports) and smuggling can both emerge in response to tax-induced divergences between the interna- tional and domestic prices of traded goods, the analysis of the previous section does not carry over directly to the underinvoicing case. There are two key 472 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 4. Effective Tariff and Subsidy Rates u)itb Invoicing Fraud Effective import tariff rate, (Z) (;,, and effective export 45- subsidy rate, ax/ / Gc,,= c0Z= z for z < 7 UTCS rate, z Note, The diagram assumes identical penalty functions for import underinvoicing and export overinvoicing. See O'Connell (1989) for details. Penalties conditional on being caught are assumed to be a linear function of illegally appropriated revenues, plus an amount proportional to the volume of affected imports or exports. If the latter component, which acts as a fixed cost of fraudulent activity given the volutne of trade, is zero, then Z = 0. differences in the structure of costs. First, although it was reasonable to think of smuggling as using up real resources (for example, in using inefficient transport routes), underinvoicing simply involves producing a fraudulent record of a transaction. The private cost of this activity may include bribes or penalties, but it seems appropriate to a first approximation to assume that the activity absorbs no real resources. The supply- and demand-side reallocations that were empha- sized in the previous section will therefore not play a role here. The second difference is that it is less natural in the case of invoicing fraud to think of legal and illegal trade as separate activities. In the smuggling model, the amount of smuggling could be determined independently of the extent of legal trade, because smuggling costs were independent of the amount of legal trade. No such separation is possible for customs fraud, because individuals engaged in this activity must use (the appearance of) legal trade to cover their illegal activity. This means that, in contrast to the earlier analysis, illegal trade will affect the domestic prices of traded goods. In particular, because the price of the imported good is driven below the full tariff-inclusive price and the price of the exported good is driven above the subsidy-inclusive price, the domestic price of exports in terms of imports rises as a result of an increase in the UTCS rate. This relative- price effect is the key channel through which customs fraud affects the real equilibrium. Figures 4 and 5 bring out the main points of this analysis of invoicing fraud (a full analysis appears in O'Connell 1989). Figure 4 shows the effective import tariff and export subsidy rates, a. and a,, as functions of the UTCS rate. The effective import tariff is below the statutory tariff rate because importers under- invoice in order to lower their tariff bill, and in competition this cost saving is O'Connell 473 Figure 5. UTCS versus Improved Terms of Trade Production of nontraded goods, qv, and consumption of nontraded goods, CN Production Consumption possibflte pos bwtwe Production of Ł Consumption of export goods, qj / I importable goods, cm trade Original terms of trade -- ------~ ~~ ~ __-_-___-__-_-__- Improved terms 45' of trade Consumption of importable goods, Cm passed on to consumers in the form of a lower price. The effective export subsidy is above the statutory subsidy rate because exporters are inflating export values in order to collect a higher subsidy. Therefore (8) 0 c um(z) c z and ua(z) 2 z. Furthermore 0 1. Any scheme of uniform trade taxes will therefore have the same effect on the relative prices of domestic traded goods as does an export subsidy. As shown in figure 5, this inadvertent commercial policy distorts the alloca- tion of resources and lowers welfare. For simplicity, consider an economy in which only nontraded goods and importables are consumed and only nontraded goods and exportables produced. The upper half of the diagram shows feasible production and consumption patterns, given the economy's endowment of fac- tors of production and the exogenous terms of trade. The balanced-trade line in the southwest quadrant takes the form P*x - P*m, where x and m are actual exports and imports, respectively. Together with the 450 line in the southeast quadrant, the balanced-trade line translates any production pattern in the north- west quadrant into the corresponding consumption pattern in the northeast quadrant. 474 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 It is apparent from the diagram that the introduction of a finite UTCS appreci- ates the real exchange rate for imports (the domestic relative price of nontraded goods to imports) and moves the consumption equilibrium from point 1 to point 2, where welfare is lower. On the production side, the real exchange rate for exports depreciates, and production shifts in favor of exports, which now com- mand a higher relative price. The welfare loss occurs without any loss in real resources; welfare would be even lower in the presence of social costs of the type emphasized in the previous section. It is instructive to compare the UTCS equilibrium with what happens when the same improvement in the domestic terms of trade occurs as a result of an exogenous movement in the external terms of trade-PX / PM. The difference is that an improvement in the world terms of trade represents an increase in consumption possibilities for the economy. The balanced trade locus rotates counterclockwise in the southwest quadrant of the diagram, and the consump- tion possibilities locus expands to reflect the economy's increased command over imports. Welfare rises to point 3. As long as both goods are normal, the real exchange rate in terms of importables appreciates; what happens to the real exchange rate in terms of exportables depends on income and substitution ef- fects (Gavin 1988). If the income effect dominates, consumption of the non- traded good rises, and the real exchange rate in terms of exportables must depreciate. If the substitution effect dominates, consumption and production of nontraded goods fall, and the real exchange rate in terms of exportables appreciates. The fiscal effects of a UTCS scheme with invoicing fraud are qualitatively identical to those emphasized in the smuggling case. In particular, although designed to be revenue-neutral when trade is balanced, uniform trade taxes produce a fiscal shock that must be made up for by expenditure cuts or higher taxation. As before, the presumption is that this adds to the welfare burden of the policy. V. CONCLUSION This article has examined the similarities and differences between uniform trade taxes and exchange rate changes. Without illegal trade, a UTCS scheme is equivalent in all real respects to a devaluation of the commercial rate in a dual exchange rate system. It is not surprising that when the possibility of illegal trade is taken into account, the equivalence between uniform trade taxes and dual exchange rates is broken. More important is the fact that illegal trade undermines the attractive- ness of both of these alternatives as proxies for across-the-board devaluation. These weaknesses are particularly important in developing countries, because they seem most likely to emerge when governmental administrative capabilities are thinly stretched and the required tax-cum-subsidy rate is large. The analysis identified two channels through which illegal trade alters the O'Connell 475 operating characteristics of UTCS schemes. The first is a resource-cost effect: when illegal trade uses up domestic resources, an increase in trade taxes drives down real income by drawing resources into smuggling, thereby producing negative supply and demand shocks in both the tradables and the nontradables sectors. With respect to the real exchange rate, it is striking that using a UTCS to raise the domestic relative price of traded goods may backfire and end up actu- ally appreciating the real exchange rate in terms of importables. This outcome emerges if the nontradables sector has a sufficiently low income elasticity of demand or a relatively high cross elasticity of supply with respect to the UTCS rate. The second channel is through the domestic relative price of traded goods. In the presence of some types of illegal trade, such as invoicing fraud, a rise in the UTCS rate will raise the domestic price of exports in relation to imports, appre- ciating the real exchange rate for imports and depreciating the real exchange rate for exports. In this case, a trade tax package that was designed to be neutral between traded goods actually ends up introducing an inadvertent net subsidy to exports. Both the resource-cost effect and the relative-price effect produce efficiency losses for a small open economy and therefore lower welfare in relation to what would prevail under an across-the-board devaluation. These findings suggest that whatever the macroeconomic benefits of avoiding a change in the official exchange rate, illegal trade introduces a nonequivalence between UTCS schemes and across-the-board parity changes that is unambiguously unfavorable to the former. A number of extensions of the analysis, particularly on the macroeconomic front, would help complete this account of the operating characteristics of UTCS schemes. The first is to add investment to the model and investigate the relation- ship between investment response, the real exchange rate, and fiscal revenues under a UTCS when the government does not have lump-sum taxes. The key issue here is that, although the tariff component of a UTCS simultaneously satis- fies both the relative-price and the revenue objectives of the government, the export subsidy component brings out a conflict between these two objectives. The government may therefore have an incentive to renege on the export subsidy component of the package. If private investors anticipate this possibility, the investment response in the exported goods sector is likely to be considerably weaker than under the corresponding devaluation. The second extension would be to introduce explicitly some wage and price stickiness so as to reinstate the real balance and relative-price effects that are the traditional channels for short-run real effects of devaluations and UTCS schemes. I emphasized in the introduction that parity changes are often defended on the grounds that they directly lower the real wage in terms of traded goods, thereby helping to avoid "transitional unemployment" associated with more gradual approaches to real depreciation. On the negative side, this short-run gain must be weighed against the other short-run, primarily contractionary, effects of par- 476 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 ity changes or UTCS schemes that appear when prices are sticky. It would help to know how, if at all, these short-run tradeoffs are affected by the illegal trade associated with the UTCS option. Finally, although I have assumed perfect foresight and permanent changes in trade taxes or exchange rates, it is worth emphasizing that the balance of pay- ments effects of UTCS policies depend crucially on how these policies affect expectations regarding the financial exchange rate. Laker (1981, p. 137) found that "in all four countries [surveyed in his article], the fiscal proxies were subse- quently abandoned in favor of an explicit change in the exchange rate." This suggests that introducing a UTCS scheme is likely to raise subjective probabilities of devaluation of the financial rate (either alone or as part of an across-the- board devaluation). This in turn means an increase in the domestic nominal interest rate and a portfolio shift away from domestic currency and toward foreign bonds. If the interest elasticity of money demand is high enough, the overall balance of payments may well deteriorate as a result. Indeed, if there are limits to international borrowing by the central bank, it may be impossible (without other policy action) for the authorities to rule out an equilibrium in which implementation of the UTCS leads to self-fulfilling expectations of a bal- ance of payments crisis and devaluation of the financial rate. APPENDIX. WELFARE AND THE REAL EXCHANGE RATE IN THE SMUGGLING MODEL Totally differentiating equations 5 and 6 in the text, gives ( de] 1 ELe -E z(l + z)-lr33 [ LdU A Lree-Eee (1 + z) re3 re3 where subscripts e, U, and 3 denote partial derivatives with respect to e, U, and [z/(1 + z)], and where A [z/(1 + z)]re3eeU + eu(ree - eee) = reeelU - releeu- EUEee > 0. Thus, (A-2) dU( +z) 1 ([r23 - r,er33] + eeer33) (A-3) [,ud = (1 + z)2 [eUr3l - eUre3(l - eEeU where equation A-3 uses the fact that r3l + er3e + [z/(l + z)]r33 = 0, by homogeneity of degree 1 of r3. Consider the effect on welfare first. By convexity of the revenue function, the term in square brackets in equation A-2 is nonpositive; this implies that the entire expression is negative, except at z = 0, where it is zero (although an infinitesimal rise in z starting at z = 0 produces a first-order increase in smug- gling services, it does so without drawing more than an infinitesimal amount of O'Connell 477 labor from productive sectors and therefore without affecting overall income or utility to first order). UTCS schemes therefore unambiguously worsen welfare in this model: any finite increase in the UTCS rate draws resources out of productive activities and lowers welfare. Now consider deldz. Under the assumption that the marginal product of labor in smuggling goes to infinity as qs goes to zero, one can show that rl 3 = re3 = 0 for z = 0. An infinitesimal change in z starting at z = 0 therefore has no effect on e. For z > 0, however, rj3 is strictly less than zero. To derive equation 7, note that e = e1 + eeel, by homogeneity of degree 1 of e. This implies Eu = elu + eeE". Using this to substitute for the last term in equation A-3, gives (A-4) de = 1 1 (E,url3-ElUr,3) dZ (1 + Z)2 A eU1 Income elasticities of demand for the traded and nontraded goods are given by AT = e1uE/eue1 and IAN = EeUE/EUEe, respectively. Using these, equation A-4 implies > asre3 /E, >IN (A-5) de/dz = Oas_ = _ < r,3/E C< /1T The cross elasticities of supply of nontraded and traded goods with respect to [z/(1 + z)] are VN3 = ri3[z/(1 + z)] /r and PT3 = r13[z/(1 + z)]/r1. Because re = e, by market clearing and r, = e1 by a zero trade balance, the last expression in equation A-5 is VN3/ vT3. This yields equation 7 in the text. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Adams, Charles, and Jeremy Greenwood. 1985. "Dual Exchange Rate Systems and Capital Controls: An Investigation." Journal of International Economics 18 (1/2, February): 43-63. Bhagwati, Jagdish. 1968. "The Theory and Practice of Commercial Policy: Departures from Unified Exchange Rates." Special Papers in International Economics. 8. Prince- ton University, Department of Economics, International Finance Section, Princeton, N.J. , ed. 1974. Illegal Transactions in International Trade. Amsterdam: North- Holland. Bhagwati, Jagdish, and Padma Desai. 1970. India: Planning for Industrialization- Industrialization and Trade Policies since 1951. Oxford, U.K.: Oxford University Press for the Development Centre of the Organisation for Economic Co-operation and Development. Bhagwati, Jagdish, and Bent Hansen. 1973. "A Theoretical Analysis of Smuggling." Quarterly Journal of Economics 87(2, May): 172-87. 478 THE WORLD BANK ECONOMIC REVIEW. VOL. 6, NO. 3 Branson, William H., and Jorge Braga de Macedo. 1989. "Smugglers' Blues at the Central Bank: Lessons from Sudan." In Guillermo Calvo, Ronald Findlay, Pentti Kovri, and Jorge Braga de Macedo, eds., Debt, Stabilization, and Development. Ox- ford, U.K.: Basil Blackwell. Deardorff, Alan, and Wolfgang Stolper. 1990. "Effects of Smuggling under African Conditions: A Factual, Institutional, and Analytic Discussion." Weltwirtschaftliches Archiv 126(1): 116-41. Devarajan, Shantayanan, and Jaime de Melo. 1987. "Adjustment with a Fixed Exchange Rate: Cameroon, C6te d'Ivoire, and Senegal." The World Bank Economic Review 1 (3): 447-87. Dixit, A. K., and Victor Norman. 1980. Theory of International Trade. Cambridge: Cambridge University Press. Dornbusch, Rudiger. 1980. Open Economy Macroeconomics. New York: Basic Books. Frenkel, Jacob, and Assaf Razin. 1989. "Exchange-Rate Management Viewed as Tax Policies." European Economic Review 33(4, April): 761-81. Gavin, Michael. 1988. "Structural Adjustment to a Terms of Trade Disturbance: The Role of Relative Prices." Journal of International Economics 28(3 /4, May): 217-43. Krumm, Kathie. 1985. "Adjustment Policies in the Ivory Coast in the Framework of UMOA." CPD Discussion Paper 1985-6. World Bank, Country Policy Department, Washington, D.C. Processed. -. 1987. "Adjustment in the Franc Zone: Focus on the Real Exchange Rate." CPD Discussion Paper 1987-7. World Bank, Country Policy Department, Washington, D.C. Processed. Laker, John F. 1981. "Fiscal Proxies for Devaluation: A General Review." IMF Staff Papers 28 (1, March): 118-43. Leiderman, Leonardo. 1987. "Price and Wage Flexibility in the Adjustment Process of Franc Zone Countries: The Ivory Coast." University of Chicago, Department of Eco- nomics. Processed. Macedo, Jorge Braga de. 1987. "Currency Inconvertibility, Trade Taxes, and Smug- gling."Journal ofDevelopment Economics 27 (1/2, October): 109-25. Martin, Lawrence, and Arvind Panagariya. 1984. "Smuggling, Trade, and Price Dis- parity: A Crime-Theoretic Approach." Journal of International Economics 17 (3/4, November):201-17. Meade, J. E. 1951. The Balance of Payments: The Theory of International Economic Policy. Vol. 1. London: Oxford University Press. Obstfeld, Maurice. 1981. "Capital Mobility and Devaluation in an Optimizing Model with Rational Expectations." American Economic Review: Papers and Proceedings 71 (2, May): 217-21. . 1986. "Capital Controls, the Dual Exchange Rate, and Devaluation." Journal of International Economics 20 (1 /2, February): 1-20. O'Connell, Stephen A. 1989. "Uniform Trade Taxes, Devaluation, and the Real Ex- change Rate: A Theoretical Analysis." PPR Working Paper 185. World Bank, Country Economics Department, Washington, D.C. Processed. Pitt, Mark M. 1981. "Smuggling and Price Disparity." Journal of International Eco- nomics 11 (4, November): 447-5 8. O'Connell 479 1984. "Smuggling and the Black Market for Foreign Exchange." Journal of International Economics 16 (3 /4, May): 243-57. Sheikh, Munir A. 1974. "Smuggling, Production, and Welfare." Journal of International Economics 4 (4, November): 355-64. Thursby, Marie, Richard Jensen, and Jerry Thursby. 1990. "Smuggling, Camouflaging, and Market Structure." Purdue University, Department of Economics, West Lafayette, Ind. Processed. Tirole, Jean. 1988. The Theory of Industrial Organization. Cambridge, Mass.: MIT Press. van Wijnbergen, Sweder. 1986. "Macroeconomic Aspects of the Effectiveness of Foreign Aid: On the Two-Gap Model, Home Goods Disequilibrium and Real Exchange Rate Misalignment." Journal of International Economics 21 (2, August): 123-36. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 481-502 Maize and the Free Trade Agreement between Mexico and the United States Santiago Levy and Sweder van Wijnbergen Setting the price of maize in rural Mexico above the world price is inefficient and likely to have negative distributional effects because many subsistence producers, and all landless workers, are net buyers; in fact it screens out the relatively poor rather than the relatively rich. The policy objective, therefore, should be to move toward free trade. This would yield large gains in efficiency. The Free Trade Agreement provides an ideal opportunity to pursue this objective. It will provide freer entrance into the United States for other agricultural products as well as a broad range of manufactured products. Insuring secure and sustained access for labor-intensive agricultural and manufactured products can help ease the impact on the labor market of a transition away from subsistence maize cultivation. Maize is perhaps the single most important commodity in Mexico. In rural areas it is the main food consumed by farmers; in urban areas it is the main input into tortillas, a key component of urban workers' diets. Maize cultivation occupies between one-third and one-half of the country's arable land and employs one out of three rural workers. It is grown by a large number of small-scale producers on rain-fed land and by relatively fewer large-scale farmers on irrigated land. But because many small-scale producers, or subsistence farmers, have plots of very poor quality, maize is directly associated with rural poverty. In addition poverty in Mexico is to a large extent a rural phenomenon (Levy 1991). Governments in Mexico have announced maize self-sufficiency as a national goal. Governments also have expressed their commitment to poor maize pro- ducers by subsidizing production. The process of land reform in Mexico gave farmers some land, but, as the extensive margin was exhausted, the quality of the land distributed diminished. Of the 43 million hectares distributed between 1958 and 1976, 91 percent were hillside, mountainous terrains; 8.4 percent were rain-fed land; and only 0.4 percent were irrigated land (Salinas 1990). Raising the producer price of maize was one way to increase the value of the Santiago Levy is with the Economics Department at Boston University. When this article was written, Sweder van Wiinbergen was the lead economist for Mexico at the World Bank. The authors thank Hans Binswanger for useful discussions, Hartwig Schafer for valuable assistance, and the three anonymous referees for very helpful suggestions. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 481 482 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 asset distributed. Governments have also expressed their commitment to the poor by subsidizing maize consumption, although the system of support to consumers operates mostly in urban areas. Rural consumers obtain most of their maize at the producer price. Attempts are made to subsidize rural consumers through a network of Conasupo (the government's food marketing and distribu- tion agency) stores, where maize is sold at a discount. But these subsidies are small and do not systematically reach rural consumers. Therefore, rather than referring to the consumer and producer prices of maize, we instead refer to the urban and rural prices of maize. To accomplish its goals of maize self-sufficiency and support for poor maize producers, the government controls imports, and intervenes directly in market- ing and distribution through Conasupo. These policies thus raise distributional issues between urban and rural areas, and within rural areas, because not all rural producers grow maize and only a subset of maize producers are net sellers. But because significant land and labor resources are allocated to this crop, maize policies also have important effects on efficiency. The soon-to-be-negotiated Free Trade Agreement (FTA) between Mexico and the United States has placed maize at the forefront of policy debates in Mexico. Policymakers face a dilemma: continue present policies, or include maize in the FTA. This article analyzes maize pricing policies in Mexico and explicitly calcu- lates the costs of keeping maize outside the FTA. We argue that efficiency and distributional gains can be made by liberalizing maize. For a general analysis of agricultural pricing in developing countries, see Sah and Stiglitz (1987); for analysis of individual countries, see, for example, Braverman, Hammer, and Gron (1987) for Cyprus and Newbery (1987) for the Republic of Korea. I. A FRAMEWORK FOR ANALYSIS We opt for a partial equilibrium approach mostly because of the forbidding data requirements for a full-fledged general equilibrium model. This is in the spirit of much recent work on taxation in developing countries (Newbery and Stern 1987; Newbery 1987). We show that the tools used in this approach are in fact quite flexible and easily adaptable to reflect much country-specific detail. Sah and Stiglitz (1987) use similar tools, but go quite far toward a general equilibrium analysis. Their focus on the consequences of rural-urban interaction on growth is different from ours, however. Four groups are associated with maize in Mexico. First, small-scale producers consist of farmers with three hectares or less of rain-fed land, who derive their income partly from producing maize and partly from participating in the labor market; they consume a significant share of their own production. Second, landless rural workers derive their income from wages and at times compete with subsistence producers on the supply side of the rural labor market. Third, large-scale producers mostly own irrigated land, are net buyers of labor, and Levy and van Wijnbergen 483 derive their income from producing a wide variety of crops. Fourth, urban consumers comprise many groups with different income levels and expenditure patterns. To evaluate maize policies, we distinguish between income changes incurred by individual groups and aggregate welfare. Calculations of aggregate welfare tell policymakers whether a particular policy needs to be changed; calculations of real income loss or gain per group indicate which groups might need assistance. The Market for Maize Maize pricing policies in 1989 fixed the rural price, pr above the world price, pw, but set the urban price, pu, below the world price. In this respect Mexico's policies differ from those of many other developing countries, where both the urban and rural prices of the key staple are below the world price, or those of middle-income countries like Korea, where the rural price of rice ex- ceeds the urban price, but both exceed the world price (Newbery 1987). Figure 1 illustrates the 1989 interventions in the market for maize in Mexico. Panel a shows average and marginal cost curves for maize production on rain-fed land; panel b does the same for irrigated land. Average costs on rain-fed land are higher than on irrigated land. In addition, the supply curve of maize on irrigated land is more elastic, which reflects the fact that irrigated land gives greater crop choice to producers, thus resulting in higher cross-price elasticities of supply between maize and other crops. The rural maize market is depicted in panel c. The supply curve is obtained from the horizontal addition of the marginal cost curves of the two types of producers; the demand curve reflects own- consumption and other rural uses of maize (animal feed and seeds). At pr the rural sector is a net exporter, in the amount AB. Increasing pr above p,w gives rents to subsistence producers but gives larger inframarginal rents to large-scale producers on irrigated land. Panel d shows that at pu urban demand for maize is OC. This is met by imports from rural areas in the amount AB (purchased by Conasupo at the price pr ), together with imports from abroad in the amount DC (purchased by Conasupo at the price pw). Since urban consumers pay pu for the full amount OC, this pricing-cum-import control scheme requires a subsidy given by the shaded areas of panel d, or the sum of rectangles AIJB and EFGI. Clearly, maize producers are subsidized, rural maize consumers are taxed, and urban maize consumers are subsidized. Rectangle EFGI is the subsidy received by urban consumers; rectangle AIJB is the subsidy received by rural producers. The effects on imports and the fiscal balance of equating the rural and urban prices of maize to the world price are as follows. At p wproduction by both types of producers declines, together with land rents. Rural consumption increases, which reduces net maize exports to the urban sector from AB to QR. But at pw urban consumption falls from OC to OK. This partly reduces the increase in imports associated with lower domestic supply. In fact, it is conceivable that 484 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Figure 1. 7he Market for Maize in Mexico a. Production in rain-fed lands b. Production in irigated lands Marginal cost Marginal cost pm-f, PMu - - Average cost - ----- - -------- Average cost I I _ _ _ _ _ _ _ _ _ _ _ _ __L I 8.8 million tons 1.5 million tons c. The rural market d. The urban market Supply A pMr W A B ------- A B pm Q RZ H pm i ~~~~~~~~~~~~Demand Demand 6.3 4.0 O T D K C I 1_ _ _ _ _ _ _ _ _ _ _ _ _ __.6l 10.3 million tons 6.5 million tons imports could fall (if, say, rural maize demand is inelastic and urban consump- tion represents a large amount of total consumption). Eliminating subsidies generates fiscal savings measured by the shaded area of panel d. Real Income Effects: Direct Price Effects Consider a simple set-up where there are only three tradable products: maize, vegetables (a proxy for all other agricultural products), and manufactures (deno- ted by subscripts m, v, and q, respectively). Increasing the number of goods leaves the argument intact as long as their prices are not affected by changes in the price of maize, as is the case for tradable goods. Introducing nontradables complicates the argument because we need to incorporate the effect of changing maize prices on the price of the nontradable. But in Mexico maize accounts for less than 1 percent of gross national product, so even large changes in maize prices are unlikely to have an important effect on the price of nontradables. Levy and van Wijnbergen 485 To sharpen the argument, we also assume that rain-fed lands can produce only maize, while irrigated lands can also produce vegetables, and that all rain-fed lands are owned by subsistence producers and all irrigated lands are owned by large-scale producers. Let pv and Pq be the prices of vegetables and manufac- tures, respectively, and let manufactures be the numeraire, so that Pq = 1. Prices of vegetables and manufactures are assumed to be the same in urban and rural areas. Abstracting from transport costs, this is a fair representation of Mexico, because there is no agency like Conasupo that explicitly intervenes to set a wedge between the two. Finally, let wr be the rural wage rate. Consider now the income effects on each of the four groups mentioned above of removing produc- tion and consumption subsidies to maize. Subsistence producers. Subsistence producers must allocate their total labor, Lp (where an overbar denotes an exogenous variable), between cultivating maize on their rain-fed land, Lrf, and participating in the labor market (Lp - Lrf) Writing Qrf for the quantity of maize produced by subsistence producers on rain-fed lands, and E(-) for the expenditure function, the budget constraint for subsistence producers is (1) E(pr,pv, 1, U) = pr Q4(Lrf) + Wr. (Lp - L) where U is utility level. We ignore leisure, given the valuation difficulties in rural environments; see Braverman, Hammer, and Ahn (1987) for further discussion. Differentiating equation 1 with respect to the rural maize price, pt,, yields (2) Epr. *dpr + Eu- dU = pr * (dQrfIdpr )dpr + Qrf - dpr + (Lp *dWr/dpr -wr-dLrf/dpr -Lf -dwr/dpr)dpm where we use the fact that dpl/dpr = 0. Noting that Epr is the compensated demand for maize, Cm, that at an optimal labor allocation pr . aQr/ aLrf = wr and that dQ;4/dpr = (aQrf/aLrf) - dLrf/dpr, we can rearrange equation 2 to yield (3) Eu * dU = (Qf- Cm) * dpr + (Lp - Lrf)(dwr/dpr)dpr. Eu is the inverse of the marginal utility of income, so the left side of equation 3 provides a "money" measure of the change in real income associated with a small change in the rural maize price. Hence, equation 3 shows that changes in the rural maize price affect subsistence producers through two channels, which we label the direct price effect and the indirect wage rate effect (the first and second terms on the right side of equation 3, respectively). The direct price effect shows that without changes in the rural wage rate, the real income effect on subsistence producers of changes in the rural maize price depends only on their net maize position: subsistence producers who are net sellers (buyers) lose (gain) with a fall in the rural maize price. The indirect wage rate effect, however, shows that subsistence farmers gain (lose) on the amount of their marketed labor as the rural wage rate increases (decreases). 486 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Labor allocated by subsistence producers to on-farm maize production is obtained by solving the equation P- aQrf/aLrf - wr = 0. Let the solution be Lmf(wr, p^) and note that aLm/ pm > 0. Hence, reducing the rural maize price increases subsistence farmers' participation in the labor market. This increase in the rural labor supply causes a change in the rural wage rate, which in turn affects subsistence farmers by changing the income they receive on their mar- keted labor. The term dwr/dp/ captures this indirect effect. The direction of change in the wage rate, however, also depends on what happens to the demand for rural labor as the rural maize price falls. If dwr/dpr > 0 there are two effects. First, subsistence producers who are net sellers lose both because their marketed maize is worth less and because their marketed labor is worth less. Second, subsistence producers who are net buyers gain from lower maize prices because of the maize they buy, but lose from lower wages for their labor, so the impact on their real income is ambiguous. Conversely, if dWr/dpr < 0, subsistence producers who are net sellers face an ambiguous real income change (losing on their maize sold but gaining on their labor sold), and subsistence producers who are net buyers unambiguously gain (paying less for the maize they buy and getting more for the labor they sell). Landless rural workers. By definition landless rural workers own no land; market all their labor, Lr; and purchase all the maize they consume. The change in their real income is given by (4) Eu .dU = -Cm * pm + Lr- (dwr/dpM) dpM. Clearly, the direct effect of lower maize prices is beneficial to landless rural workers. However, the indirect effect may hurt them if the rural wage rate falls as a result of the increased labor market participation by subsistence producers and if such a fall is large enough to eliminate the gains associated with a lower price for the maize they consume. When pr falls the wage rate measured in terms of maize increases, but if dWrldpr > 0 the wage rate in terms of manufac- tures decreases. Large-scale farmers. We assume that large-scale farmers derive all their in- come from the (irrigated) land they own. Their problems are to allocate their total land, T, between maize and vegetables (Ti and T,, respectively,) and to determine how much labor to employ in each crop (Li and L,). Hence, their budget constraint is (5) E(pm, p, 1, U) = pr * Qm(Ti, Li) + pv * Q,(T,, L,) -Wr - (Li + L,) where Qi is maize output on irrigated land, and Qv is vegetable output. Effi- cient allocation of land requires dTl/dpr = - dT;/dpr, given the land con- straint T = T, + Ti. Differentiating equation S and using this condition yields (6) Eu * dU = (Qi - Cm) dp - (L, + L) (dwr/dp) dpr. Levy and van Wijnbergen 487 Large-scale farmers certainly consume less maize than they produce. Hence, the direct price effect lowers their real income. The indirect wage rate effect again depends on the sign of dWr/dprM. If it is positive, large-scale producers gain because their total wage bill is less. If the wage fall is large enough, these gains can offset the loses on their marketed maize, thus implying that large-scale farmers may actually benefit from lower maize prices. Urban consumers. The impact on the real income of urban workers is given by an expression similar to equation 4, except that the urban wage rate, wu, and the urban price of maize are the relevant variables in this case. Since pu < pw a full liberalization of the maize market would lower urban workers' real income. And urban workers could be affected through the indirect wage rate effect if rural employment contracts as a result of a large fall in the rural maize price and this, through migration, lowers the urban wage. Urban employers will also see their real income lowered when the urban maize price is increased through the direct price effect, although its significance is probably minimal because maize is relatively unimportant in their diets. However, to the extent that the indirect wage rate effect puts downward pressure on the urban wage rate through migra- tion, the product wage in manufacturing falls, thus leading to an increase in employment and quasi-rents on the capital stock employed in manufacturing. Real Income Effects: Indirect Wage Rate Effects Lowering the rural price of maize increases participation by subsistence farmers in the rural labor market. But the resulting change in the rural wage rate also depends on the change in the demand for rural labor as the rural maize price falls and on the change in the size of the rural labor force resulting from migra- tion to urban areas. In the appendix we discuss the relationship between urban and rural labor markets. Here we consider the isolated rural labor market described by the equilibrium condition in equation 7, (7) (Lp-Lrf) + Lr=L + Ln + Lg. The bracketed term on the left side of equation 7 is marketed labor by subsis- tence farmers, and L, is marketed labor by landless rural workers. The right side is the demand for rural labor, made up of employment in vegetable and maize production on irrigated land and a term (Lg) that represents an exogenous component of rural labor demand (associated, say, with government infrastruc- ture projects). When the rural maize price falls, marketed labor supply of subsistence farmers increases, labor demand in irrigated maize falls, and labor demand in vegetables increases. Thus, the pressures on the rural wage rate hinge on: (8) 1 aLmf + aL I < aLK, apr+ apr > apr that is, on whether the additional employment created in vegetable production 488 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 can absorb the employment displaced from maize cultivation. The appendix derives conditions that determine the direction of the inequality. Here we discuss the case where there is a net release of labor (the left side of expression 8 exceeds the right side); if this is not the case, the analysis needs to be mnodified accordingly. Equilibrium in the rural labor market can be restored through different mech- anisms. First, ignoring migration, the rural wage rate would fall, with employ- ment increasing in both maize and vegetable production until excess labor disap- pears. Second, still ignoring migration, released labor could be absorbed through direct interventions like public work programs (increase Lg). Third, equilibrium can be restored through migration. If the only policy change is a decrease in the rural maize price, the labor released from maize would reduce the rural wage rate and widen the rural-urban wage differential. This would induce rural workers to migrate to urban areas, which would mitigate the decline in the rural wage rate but would lower the urban wage rate. A lower urban wage rate, in turn, increases employment and the marginal product of capital in manufac- turing. A fourth possibility arises if the decrease in the rural maize price is accompanied by an increase in the urban maize price (as urban subsidies are eliminated). In this situation migration incentives are reduced because there are offsetting changes. The outcome in terms of migration and wage rate changes depends on the magnitude of the different effects. Clearly, however, policy- makers can influence migration flows (and hence the changes in the rural and urban wage rates) through both direct interventions, which hire labor for rural public work programs, and the size of programs like the tortibonos in Mexico, which is targeted on the urban poor and operates through coupons. Aggregate Social Welfare In the rural sector we treat the difference between the rural and the world price of maize as an ad valorem tariff, denoted by tr, so that pr = p W(1 + trT). It is convenient to take 1989 as the base year and calculate any deviations from the base year as a tariff, tr, that is additional to the tariff ruling in 1989 (denoted by to). Thus, the wedge between the rural and world maize price is given by (1 + tr ) = (1 + tr)(1 + tro), with the convention that in the base year tr = 0. We linearize the expressions for welfare change around the values observed in 1989, rather than around the free trade values (as is customary) because the free trade values are not known and we would not be able to calculate the linearization constants without a potentially large approximation error. This approach allows us to compute the welfare effects of any tariff, including as special cases a tariff that would produce self-sufficiency (denoted by tss) and the tariff-equivalent that would produce free trade (denoted by tFT). We assume tariff revenues are plowed back into the rural sector. (See Newbery 1987 for a discussion of how distributional weights could be incorporated.) This leads to the following aggregate rural budget constraint: (9) Rrr[pW(l + trT)** .] + prw(Er - Rr)tr- = Er[p'W(j + tr). . X Ur] Levy and van Wijnbergen 489 where Rr and Er are the rural revenue and expenditure functions, respectively. Hence, Rr is the quantity of maize supplied and Em the compensated demand for maize, so that the second term in parenthesis in the left side of equation 9 is net rural maize exports. Differentiation of equation 9 yields: (10) Er dU = [(Rr - Er)pr 0 + pAW(j + tr)(Er - Rr)] dtr + Pw(Eu,m -Rrm) trpnodtr + (1 +t&)rnr U p Er0 ED - Rr0E) FPn (mo e-m e) t dtr L(1 1 Vmr) j where Vmr is the marginal value share of maize in total rural expenditure; ED and ES are the price elasticities of demand and supply of maize, respectively; and the subscript 0 refers to base year (1989) values. If we linearize around the base situation, the term in parentheses on the right side of equation 10 becomes a constant, to be evaluated at base year prices and quantities. Integration of equation 10 yields the change in welfare due to a change in the tariff from its value in the base year to any specific target tariff. pr tr eD Rr ES) ttr2 (11) Er [Ur(tr ) - Ur(tr)] = 0 (EMO mO K to2 tr 2 (1- Vm r) 1+ t6 - _L1r 2 tr + (t-2l A formula similar to equation 11 can be derived for urban areas by setting Rrm (and hence Es) equal to zero and replacing tr by tu: _____ tu)2 (12) EUu[u(tj-) - Uu(to)] = _ t + (t )] Expressions 11 and 12 measure the "dead-weight loss" in the rural and urban sectors associated with any tariff and are equivalent to the sum of triangles AWQ and BRZ in panel c and to triangle HGF in panel d of figure 1, respectively. As long as price elasticities are not zero, maize protection has positive welfare costs. These costs increase with the square of the tariff, so increasing protection be- comes progressively more costly. II. FISCAL COSTS, REDISTRIBUTIVE IMPACT, AND AGGREGATE WELFARE We now apply the model to an assessment of the fiscal costs and distributive consequences of Mexico's policies on maize pricing and of the welfare gains of moving to free trade. 490 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 1. Estimated Supply and Demandfor Maize, 1989 Quantity Category (millions of tons) Supply Domestic production in rain-fed lands 8.8 Domestic production in irrigated lands 1.5 Imports 2.5 Total 12.8 Demand Rural own consumption 3.6 Intermediate usea 2.7 Urban consumption 6.5 Total 12.8 a. This includes animal feeds and seeds. Source: Rivera (1990). Fiscal Costs Table 1 gives the 1989 quantities of maize associated with each of the panels in figure 1.1 The domestic production of 10.3 million tons exceeds rural con- sumption, so that the rural sector is a net exporter (4 million tons). Because rural exports exceed production on irrigated land, it follows that the subsistence sector as a whole is a net maize exporter. And urban consumption of 6.5 million tons is met by importing 4 million tons from the rural sector and 2.5 million tons from abroad. The ratio (pr - pW)/pW was 54 percent in 1989; this gives a subsidy of US$72.8 per ton. In 1989 the rural maize price was US$208.5 per ton, and the world maize price (using the yellow maize Gulf price as a proxy) was US$135.5 per ton. Thus for domestic production of 10.3 million tons, the gross producer subsidy is US$749.8 million. But not all of this subsidy is paid by the govern- ment, since part of domestic output is consumed within the rural areas at the rural price. The net producer subsidy is only US$291.2 million, which is ob- tained by multiplying rural exports of 4 million tons by the US$72.8 per ton subsidy; this is area AIJB in panel d of figure 1. The US$458.6 million difference between the US$749.8 million and US$291.2 million gross and net subsidy is the tax paid by rural maize consumers. This is made up as follows: a tax of US$196.5 million paid by intermediate users (2.7 million tons times US$72.8) and a tax of US$262.1 million (3.6 million tons times US$72.8) paid by final consumers in the rural areas (landless rural workers and subsistence producers who are net buyers). Large-scale producers sell 1.5 million tons; therefore they appropriate 109.2 million (or 38 percent) of the US$291.2 million net production subsidy. The remaining US$182 million (or 62 percent) goes to producers on rain-fed land who are net sellers. But the ratio (pu - pw)/pw was -37.1 percent in 1989, 1. The following is based on data presented in Rivera (1990). Levy and van Wijnbergen 491 because the average urban price of maize was US$85.2 per ton (the weighted average of the different urban prices; see Rivera 1990 for more details). The subsidy per ton is US$50.37, which when multiplied by the 6.5 million tons of urban consumption gives a subsidy of US$327.40 million (area EFGI in panel d of figure 1). Redistributive Impact The redistributive impact works directly through prices and indirectly through the impact of maize prices on labor markets. Direct price effects. Figure 2 serves as departure point to estimate the direct price effects of lower maize prices on the net income of various groups. The vertical axis measures maize output and consumption by all producers. The horizontal axis lists all maize producers: K large-scale producers and N subsis- tence producers. Next to N we append a total of M other maize consumers in the rural areas (mainly landless rural workers but also other non-maize pro- ducers). Maize producers are listed in order of decreasing output. The underly- ing assumption is that all large-scale producers on irrigated land have larger maize output than any subsistence producer on rain-fed land. Letf(Qm) be the distribution of production, so that the total area underf(Qm), the surn of A + B + C, is total maize output. The distribution of maize con- sumption is denoted by g(Cm); area B + C + D is total own-consumption by maize producers. If maize is a normal good and if all producers have the same tastes, g(-) should be a decreasing function, although probably much flatter than -(-). In addition, the area under g(.) may also include some consumption by other rural workers as well as some animal consumption. Masera (1990, table 4.5, Figure 2. Maize Consumption and Production Maize output, f(Q,), and consumption' g (C",) (Q-) A- g (C,) Maize producers and __________________ Dconsumers, M °-53I1 MM, I Larg- ~et elle~] Net buyer Landless workersT01 scale '4 ~~~~~~~~~' ~ and other ~,;roducers , g =N+± (C- Q ). Figure 2, however, makes clear that N'" depends critically on the shapes off(-) and g(-). Hence a condition like inequality 13, by itself, provides insufficient information for policy. In particular, if we take K + N + M to be the total rural population, it follows that there are K + N: losers and N - N: + M gainers from the direct price effect of lowering the price of maize. If f( ) is relatively steep, then K + N*' will be a relatively small number, and, while all large-scale producers and subsistence producers as a whole lose, most of the rural popula- tion may directly benefit from lower maize prices. The available data are insufficient to completely trace f(Qm) and g(Cm) or to determine the exact numbers for K, N, and M. Nevertheless, piecing together estimates in the literature, we find information in two areas. First, the rural population is estimated to be 21.8 million people (or 27 percent of the total); out of this about 6 million are estimated to be the economically active population. In turn, out of the 6 million rural workers, about 2.5 million to 3 million are estimated to be landless, with the remaining 3 million to 3.5 million small producers (Salinas 1990, p. 817; Montanez 1988, p. 684). Montanez and Warman (1985) mention that there are about 2 million subsistence maize pro- ducers; the same figure is quoted by Masera (1990, p.39). INEGI (1988, p. 20) also states that there are more than 2 million such producers. However, Masera (1990, table 1.6, p. 40) cites a total of 2.2 million maize producers (of all types). Although some of these figures relate to different years, it seems reasonable to make an estimate of 2.25 million maize producers (out of which 2 million are subsistence and 0.25 million are large scale), about 3 million landless rural workers, and about 0.75 million other rural producers dedicated to non-maize activities. In terms of figure 2 this suggests K = 0.25, N = 2.0, and M = 3.75 million. Second, data to trace the production and consumption distributions of maize are also scarce. Montanez (1988, p. 679, our translation) states that: "Of all maize producers in the spring-summer cycle, 55 percent do not cultivate more than 2.5 hectares, which yield, on average, 1.35 tons per hectare. Individually, 58 percent obtain, at most, 2.5 tons. It is estimated that 66 percent keep all of its output and that, of the total amount produced, only 50 percent enters the market." Andrade (1988, pp. 11-15) points out that only 5.8 percent of pro- ducers obtained more than 10 tons, with average yields of 2.2 tons per hectare; he also shows that 66 percent of all producers harvested between 0 and 2.5 hectares, 32.3 percent between 2.5 and 10 hectares, and only 1.73 percent more than 10 hectares. Levy and van Wijnbergen 493 These figures imply that out of the 2.25 million maize producers assumed above, only 787,000 (34 percent) are net sellers and that about 135,000 of these (or 6 percent of all producers) account for the bulk of the sales. Thus, the distribution of maize production falls rather steeply at first and then flattens out. With own-consumption requirements falling as well with lower production (less maize required to pay for hired labor and for animal feed), this implies that area D in figure 2 is relatively small. An immediate observation is that the 66 percent of producers who are net buyers probably buy a small share of their total maize consumed, so the benefit they receive from lower maize prices is positive, but probably small. Conversely, there is a group of producers who are net sellers, but whose sales are also relatively small, so their loses from lower maize prices would also not be significant. This analysis suggests that the indirect wage rate effect of lower maize prices is more important to subsistence producers who are on the margin of being net sellers or net buyers than is the direct price effect. It also indicates that the direct impact of lower maize prices would be strong only for producers in the initial segment of the distribution. A plausible estimate is that this group of significant sellers represents, at most, 15 percent (or 330,000) of all producers (250,000 large-scale producers on irrigated land and 80,000 subsistence producers on rain-fed land). Groups that are definitely net gainers would be those close and to the right of point M2 in figure 2, or about 3.75 million other rural workers, made up mostly by landless workers. About 1.92 million subsistence producers would lose or gain little. Indirect labor market effects. To provide an assessment of expression 10, the first step is to calculate the output reduction implied by free trade in maize. A recent survey of econometric estimates of supply elasticities in Mexico shows that the aggregate price elasticity of supply of maize, es, is in the order of 1.1 (Nathan Associates 1989, table 4.1). (Since we analyze nonmarginal price changes, we assume that the supply curves of maize are isoelastic.) Unfor- tunately, these studies failed to distinguish between supply elasticities for irri- gated and rain-fed lands, (Ei and Erf, respectively). Since eS = aei + (1 - a)erf, where a is the share of maize cultivated on irrigated land, it is necessary to exogenously assume one of the two individual elasticities (satisfying Ei > Erf). Lacking additional information, we set ei equal to 1.5, such that erf is 1.03 (since a = 0.15; see table 1). A move toward free trade in maize would reduce the rural price of maize by, roughly speaking, 50 percent, thus implying a cut in maize output from irrigated land of 0.97 million tons (from 1.5 million to 0.53 million) and of 4.5 million tons from rain-fed land (from 8.8 million to 4.3 million tons). The second step is to estimate the land and labor released from maize cultiva- tion. Assuming average maize yields from irrigated and rain-fed lands of, respec- tively, 2 and 1.4 tons per hectare, the output contraction in maize releases 3.21 million hectares of rain-fed land and 0.32 million hectares of irrigated land. Table 2 then implies that 59.1 million worker-days are released from rain-fed 494 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 2. Ratios of Land to Labor, by Crop and Kind of Land Other Fruits and Kind of land Maize grains vegetables Other Pasture Irrigated 51.5 25 165 23 n.a. Rain-fed 18.4 9 58 18 5 n.a. Not applicable. Note: Number of worker-days employed per year per hectare of land. Source: Norton and Solis (1983). maize cultivation and 16.48 million worker-days from irrigated maize cultiva- tion, so that I aLrfa/pr + aLi/lapr = 75.58 million worker-days. The third step is to calculate average labor requirements on non-maize irri- gated land and estimate the increase in non-maize employment. Assume the released irrigated land is used to produce other grains, fruits, and vegetables as well as other crops in the same proportions found in table 3. This implies that 62.05 worker-days are required per hectare of non-maize irrigated land. As a result, when the 0.32 million hectares of irrigated land are turned over to other crops, we have that aL,/8pr equals 19.85 million worker-days. Consider now what happens to the rain-fed land released from maize. According to agri- cultural experts, a plausible scenario is that half of the released rain-fed land is devoted to pasture, with the remaining half equally divided between other grains and other crops (and no fruits and vegetables). Table 2 indicates that this use of the released rain-fed land requires 29.6 million worker-days, implying a net release of labor of 26.13 million worker-days. Assuming the average rural worker works 180 days a year in agricultural activities, this translates into 145,000 workers. Thus, under the assumptions made above and without any other government intervention, free trade in maize would put downward pressure on the rural wage rate. Of course, the wage reduction also depends on migration assump- tions, as well as on the elasticities of labor demand in non-maize crops. How- ever, the number of workers released by free trade in maize is small when compared with the total rural labor force (approximately 6 million workers). Table 3. Allocation of Land, 1989 (thousands of harvested hectares) Crop Irrigated Rain-fed Total Maize 915 5,553 6,468 Other grainsa 1,554 1,620 3,174 Fruits and vegetablesb 1,110 796 1,906 Other, 1,367 5,295 6,662 Total 4,946 13,264 18,210 a. Rice, sorghum, wheat, and barley. b. Including perennial crops and sugar. c. Cotton, tobbaco, beans, and others. Source: Direccion General de Estadistica (unpublished data); Secretaria de Agricultura y Recursos Hidraulicos (1984). Levy and van Wijnbergen 495 Thus, very small migration responses and very small elasticities of labor demand would be required to generate a significant fall in the rural wage rate. Social Welfare Costs In this section we look at the aggregate welfare cost of maize policies in 1989 and estimate the welfare gains of free trade in maize. We also evaluate the long- standing Mexican policy objective of maize self-sufficiency. It is difficult to ascertain the economic merit of this objective because its appeal is political and therefore outside the scope of this article. However, we can estimate the eco- nomic costs of self-sufficiency. We apply expressions 11 and 12 to measure the cost of current policies by adding the total welfare gain to both urban and rural groups in moving from the present tariff equivalent, to, to the free trade tariff equivalent, tFT = 1/(1 + to) - 1. Beginning from the base situation, the free trade tariff equivalent would make all producers and consumers face the world price of maize.2 To measure the welfare costs of self-sufficiency requires an estimate of the tariff equivalent that would reduce maize imports to zero. This tariff equivalent can be derived from the import demand equation. By definition, at self- sufficiency total maize imports, M, equal zero. To estimate the tariff that yields this outcome, we must also make an assumption about which consumers face the increased maize price. In what follows we assume that self-sufficiency would be reached while maintaining constant the price of maize to urban consumers; this is consistent with the current situation where only rural consumers face the producer price of maize. Thus the self-sufficiency tariff needs to provide enough of a producer subsidy to generate net exports from the rural to the urban areas high enough to offset the effect of urban subsidies. The import demand function for maize is (14) M = Eu + Erm-Rrn The implicit rural self-sufficiency tariff tsrs needs to solve (1S) - Eu = Er [pW(l + tr) (1 + tss), . .]- Rr [pW(l + tro) (1 + tSrS)....*] Differentiation while ignoring income effects then yields (16) Atr= _(Er o ED-Rr o ES)-l Mo. Table 4 provides the data to carry out the computations. The supply elasticity is much higher than the demand elasticity, something that will affect the relative cost of rural versus urban distortions. Moving to free trade. Table 5 shows that moving to free trade from the 1989 base year configuration yields welfare benefits of US$154 million per year, most of which is due to a reduction in rural distortions. To put this number in 2. By definition of t, tFT has to satisfy (1 + tE,)(1 + t,) = 1. 496 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 4. Basic Maize Statistics, 1989 Rest of the Urban Rural world Basic data Consumption, C_ (millions of tons) 6.5 3.6 n.a. Output, Qm (millions of tons) 0.0 7.6a n.a. Net imports M_ (millions of tons) 6.5 -4.0 -2.5b Price (US$ per ton) 85.23 208.50 135.60 Implicit tariff in 1989 (percent) -37 54 n.a. Implicit tariff to produce self- sufficiency (percent) -37 94 n.a. Basic parameters A0 (US$ billion), 0.185 1.99 n.a. Price elasticity of demand of maize, eD d 0.334 0.334 n.a. Price elasticity of supply of maize, eS d n.a. 1.1 n.a. Marginal income share of maize v_,, (assumed) 0.01 0.01 n.a. n.a. Not applicable. a. Output, Q_, equals gross output minus intermediate use (see table 1). b. World imports of maize minus Mexico's total maize imports. c. Derived from basic data, above. d. Nathan Associates (1989). Source: Authors' calculations. perspective, assume the 1989 intervention structure is maintained forever. In that case, the costs of the subsidy will rise at the growth rate of gross domestic product (GDP), say 5 percent on average over the medium term. However, future distortionary costs need to be discounted; the relevant discount rate is the margi- nal real cost of borrowing in foreign markets. Taking recent, post-debt-deal market flotations as a guide, this marginal real cost is estimated at 7.6 percent in real terms, when using a long-term inflation estimate of 5 percent. This makes for a growth-adjusted discount rate of 2.5 percent ([1.076]/1.05 = 1.025). Applying this discount rate yields the results summarized in table 6: the total, Table 5. Annual Recurrent Welfare Costs of Maize Price Interventions (millions of U.S. dollars) Areas of welfare costs Intervention Rural Urban Total Move from free trade to 1989 price structure 122 32 154 Move from 1989 price structure to self-sufficiency 251 n.a. 251 Move from free trade to self-sufficiency 374 32 406 n.a. Not applicable. Note: Totals may not add because of rounding. Source: Authors' calculations. Levy and van Wijnbergen 497 Table 6. Permanent Growth-Adjusted Welfare Costs of Maize Price Intervention (net present value, billions of U.S. dollars) Areas of welfare costs Intervention Rural Urban Total Move from free trade to 1989 price structure 4.9 1.3 6.2 Move from 1989 price structure to self-sufficiency 10.0 n.a. 10.0 Move from free trade to self-sufficiency 14.9 1.3 16.2 n.a. Not applicable. Note: Totals may not add because of rounding. Source: Authors' calculations. permanent, but discounted, welfare costs of current maize policies equals about US$6.2 billion, or 3 percent of 1989 Mexican GDP. Moving to self-sufficiency. Combining data from table 4 with expression 16 yields the self-sufficiency tariff tss of 0.26. This means that to reach self- sufficiency in maize while maintaining urban consumer subsidies, the rural price needs to be raised by 26 percent above the 1989 level of 54 percent over world prices. Thus the total self-sufficiency tariff would be 94 percent. The second and third rows of table 5 indicate that the associated welfare costs are substantial. To move from the 1989 configuration to a price configuration that achieves self-sufficiency increases the welfare costs by US$251 million a year. The permanent costs of this policy, using the same growth adjusted dis- count rate of 2.5 percent, are US$10 billion, or around 5 percent of 1989 GDP. But the real costs are higher, because the relevant comparison should not be the 1989 base year, but free trade; this almost doubles the yearly cost estimate of self-sufficiency to US$406 million. If maintained forever, the total discounted welfare costs in 1989 dollars of self-sufficiency are equal to US$16.2 billion, or 8 percent of 1989 GDP! Fiscal subsidies and aggregate welfare costs. The analysis shows that in 1989 the government spent US$618.6 million in maize subsidies divided between a net subsidy to the rural areas of US$291.2 million and an urban subsidy of US$327.4 million. Netting out redistributions, we find that the aggregate wel- fare cost of this policy is US$154 million, divided between US$122 million in the rural areas and US$32 million in the urban areas. For the country as a whole every dollar of subsidy generated only US$0.75 of welfare gain. Thus, the country as a whole loses 25 cents per dollar spent on maize price intervention. For the rural areas every dollar of subsidy generated only US$0.58 of welfare gain, making for a net loss of 42 cents per dollar, while for the urban areas the corresponding amount is 0.90 (a net loss of 10 cents per dollar spent). The difference between rural and urban losses is explained by the fact that in the rural areas subsidies induce a production and a consumption distortion, while in the urban areas the production distortion is absent. 498 THE WORLD BANK ECONOMIC REVIEW. VOL. 6, NO. 3 These welfare costs of course depend on the values of the elasticities of supply and demand of maize, parameters that are very difficult to estimate with preci- sion. Nevertheless, it is worth pointing out that the values used are probably on the low side (particularly so with respect to the elasticity of supply of maize),3 so that these estimates would appear to be a lower bound on the welfare costs of current policies. This observation is reinforced by noting that in the analysis we also have ignored the welfare costs of raising the fiscal revenues required to cover the maize subsidies, a potentially important additional cost (Browning 1987). Unfortunately, there are no studies of the welfare cost of raising fiscal revenues in Mexico. To the extent that this cost is substantial, our estimates of welfare loses would have to be increased accordingly. III. WHY MAIZE POLICIES NEED TO BE CHANGED A recent study of poverty found that 67 percent of the extremely poor popula- tion in Mexico lives in rural areas (Levy 1991). The importance of urban pov- erty is reduced further if account is made of the fact that not all the extremely poor are equally poor and that the distribution of poverty is not the same across regions. When poverty measures that are sensitive to the depth and distribution of poverty are used, the proportion of people in extreme poverty accounted for by the rural areas in Mexico increases to 76 percent. It is thus very difficult to justify, on poverty alleviation grounds, a subsidy to maize consumption in the urban areas. Moreover, the reduced urban maize price represents an across-the-board sub- sidy to all urban consumers. Hence, although poor urban inhabitants benefit, part of the benefit spills over into groups that clearly do not need the subsidy. Since the fiscal cost of this policy is substantial (US$327.4 million in 1989), a targeted program can be equally effective in transferring income to the urban poor while at the same time reducing the fiscal burden and the dead-weight loss. An infra-marginal targeted subsidy (using coupons) also breaks the link between income and the size of benefits received and liberates resources to help the rural poor. Targeted programs are not without problems, however. When benefits are made a function of income, a negative incentive to work is created: if partici- pants realize that benefits fall when their incomes increase they effectively face very high marginal tax rates. From this perspective it is better not to make benefits conditional on income. As poverty programs in Mexico move away from generalized subsidies to means-tested programs, the tension between incen- tives to work and means-tested targeting will become sharper. This article has shown that the policy of setting the rural price of maize above the world price is inefficient. The policy precludes workers from being employed in other areas where the value of their marginal product at world prices can 3. The econometric studies surveyed by Nathan Associates (1989) are all based on single equation estimates of maize supply elasticities. We know of no econometric study for Mexico where individual crop elasticities are derived from a profit function approach that incorporates cross-price effects. Levy and van Wijnbergen 499 eventually be higher, while it induces some of the scarce factor, high-quality land, to be used in activities for which the value of its marginal product at world prices is lower. The labor allocation is also distorted because, by keeping subsis- tence farmers employed on their own land, higher maize prices reduce the supply of labor and put upward pressure on the rural wage. This is a very indirect and inefficient mechanism for supporting the rural wage rate. The same amount of resources currently spent on maize subsidies could be used for rural infrastruc- ture programs. These programs would provide rural employment and would create the necessary infrastructure that, over the medium term, is required to open more earnings possibilities for the rural poor. We have shown that the distributional effects of the higher maize price are mixed. It directly benefits subsistence producers who are net sellers. But a sub- stantial number of subsistence producers are net buyers; and even while the policy helps a subset of subsistence producers, larger rents are transferred to large-scale producers. And finally, landless rural workers are hurt because they must pay a higher price for the maize they consume. But perhaps the fundamental problem with the policy of protecting maize production is that it focuses on the wrong objective. Rather than helping those who produce maize, the objective should be to help poor rural inhabitants, regardless of where they work. Of course reaching the poor directly is noto- riously difficult, especially in rural areas. Indirect methods often need to be applied, which use an observable variable that one suspects is highly correlated with the degree of poverty as a "screening device." We have shown that, because of its differing effects on subsets of the rural poor, the rural price of maize is a very ineffective screening device. There is in fact a strong presumption that it screens out the relatively poor rather than the relatively rich. With distributional arguments discredited, the central policy objective for maize, therefore, should be to move toward eventual free trade in this commodity. The FTA provides an ideal opportunity to pursue this objective. It is likely to provide freer entrance into the U. S. market for other agricultural products in which Mexico does have a comparative advantage (such as sugar and fruits and vegetables), as well as for the broadest possible range of manufactured products. Insuring secure and sustained access for labor-intensive agricultural and manu- factured products can help ease the transition away from subsistence maize cultivation. If this opportunity is taken, Mexico, unlike many other countries, will avoid the need to increase protection to agriculture as real incomes increase in the rest of the economy and will be able to provide sustainable increases in living standards over the medium term to the currently rural poor. APPENDIX. THE RELATION BETWEEN THE URBAN AND RURAL LABOR MARKETS To analyze the wage rate effect of reducing the producer price of maize, we first consider the impact on the rural labor market, assuming there is no migra- tion. We then integrate the urban and the rural labor markets. 500 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 The initial rural wage rate, wr,O, is obtained by solving (A-1) Lf(wr, pm; Trf) + LI(wr, pm; Tm) + LjEwr, pm; (Ti - Tx)] - Lr-Lp = 0 where the superscript 0 denotes the value of the variable before the producer price of maize is reduced, Ti is the total endowment of irrigated land, T; is irrigated land allocated to maize, and (T, - Tm = TV) is irrigated land allocated to vegetables, and Trf is the total endowment of rain-fed land. Differentiating the total demand for rural labor, DLR, and noting that a Tm/apm = - a T, /ap., equation A-1 can be rearranged to yield (A-2) dDLR/dPm = aLrf/apm + (aL,/aT,-dla%T) - aTl/apm. The first term on the right side of equation A-2 is obviously positive. The second term, however, depends on the comparison between the marginal labor- land ratios in vegetable and maize cultivation on irrigated land. If at the margin vegetables are less labor intensive than maize, the second term will be positive since aTv/apm < 0. Expression A-2 thus tells us that if vegetables are less labor intensive than maize on irrigated land, when pm falls, the total demand for rural labor unambiguously falls. Conversely, if on irrigated land vegetables are more labor intensive than maize, the change in the demand for labor is ambiguous. Clearly, without migration the contraction in rural labor demand translates directly into lower rural wages. Rural employment stays constant, although its composition changes. Subsistence producers now devote less of their labor to grow maize in their own land, so labor is shed from subsistence maize cultiva- tion. We can calculate the amount of labor that needs to be shed from the rural areas to keep wages constant. Thus, if Lg increased by this amount, the rural wage would remain the same. Differently put, we can calculate the size of a rural employment program that neutralizes the wage effect of reducing the price of maize. The impact of migration on this outcome depends on the assumptions made about the determinants of rural-urban wage differentials. Before the change in the price of maize, equilibrium wage rates are wr,O and w-.°, implying an urban- rural wage differential measured by j0 = wu,/Wr, ° > 1. The fall in the price of maize shifts the rural labor demand downward. With- out migration the rural wage falls. But this increases the differential between urban and rural wages, which in turn may induce rural migration. If migration occurs until the initial wage differential is reestablished, some rural workers migrate, thus reducing the labor supply in the rural areas and increasing it in the urban areas. Under a constant wage differential assumption part of the wage rate effect of reducing the price of maize is absorbed by lower wages and part by shifting labor from rural to urban areas. Of course, migration need not restore the initial wage differential. But as long as there is some (positive) migration, the fall in the rural wage is mitigated. The counterpart to this is some fall in the urban wage, which in turn increases manufacturing employment. Hence, under Levy and van Wijnbergen 501 the assumptions stated above and without rural employment programs (AL, = 0), free trade in maize would increase the marginal productivity of capital in manufacturing.4 REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Andrade Ascencio, J. 1988. "El Autoconsumo de Maiz y Frijol en Mexico." Programa de las Naciones Unidas para el Desarollo. Mexico City. Processed. Braverman, Avishay, Jeffrey S. Hammer, and Anne Gron. 1987. "Multimarket Analysis of Agricultural Price Policies in an Operational Context: The Case of Cyprus." The World Bank Economic Review 1 (2): 337-56. Braverman, Avishay, Jeffrey S. Hammer, and Choong Yong Ahn. 1987. "Multimarket Analysis of Agricultural Pricing Policies in Korea." In David Newbery and Nicholas Stern, eds., The Theory of Taxation for Developing Countries. New York: Oxford University Press. Browning, Edgar. 1987. "On the Marginal Welfare Cost of Taxation." American Eco- nomic Review 77 (March): 11-23. INEGI (Instituto Nacional de Estadistica, Geografia, e Informatica). 1988. Abasto y Comercializacion de Productos Basicos: Maiz. Mexico City. Levy, Santiago. 1991. "Poverty Alleviation in Mexico." PRE Paper 679. The World Bank, Latin America and the Caribbean Country Department II, Washington, D.C. Processed. Masera Cerutti, 0. 1990. Crisis y Mecanizacion de la Agricultura Campesina. Mexico City: El Colegio de Mexico. Montanez Villafana, Carlos. 1988. "Las Condicionantes de la Politica Agropecuaria." Comercio Exterior 38 (8): 679-85. Montanez Villafana, Carlos, and Arturo Warman. 1985. Los Productores de Maiz en Mexico: Restricciones y Alternativas. Mexico City: Centro de Ecodesarollo. Nathan Associates, Inc. 1989. "Comermax: A Multimarket Model of Mexico's Agricul- ture." Prepared for the Inter-American Development Bank. Washington, D.C. Pro- cessed. Newbery, David. 1987. "Identifying Desirable Directions of Agricultural Price Reform in Korea." In David Newbery and Nicholas Stern, eds., The Theory of Taxation for Developing Countries. New York: Oxford University Press. Newbery, David, and Nicholas Stern, eds. 1987. The Theory of Taxation for Developing Countries. New York: Oxford University Press. Norton, Roger D., and Leopoldo Solis M. 1983. The Book of CHAC: Programming Studies for Mexican Agriculture. Baltimore, Md.: Johns Hopkins University Press. 4. Of course, the free trade agreement also shifts the demand for urban labor. If such an agreement results in increased investment (foreign or domestic), demand for urban labor increases, thus increasing urban (and, through migration, rural) wages. 502 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Rivera, Ivan. 1990. "Mexico Agsal II: Regulatory Framework of Food Agroindustries." World Bank, Latin America and the Caribbean Country Department II, Washington, D.C. Processed. Secretaria de Agricultura y Recursos Hidraulicos. 1984. Anuario Estadistico de la Pro- duccion Agricola Nacional 1984. Mexico City. Sah, Raaj Kumar, and Joseph Stiglitz. 1987. "The Taxation and Pricing of Agricultural and Industrial Goods in Developing Economies." In David Newbery and Nicholas Stern, eds., The Theory of Taxation for Developing Countries. New York: Oxford University Press. Salinas de Gortari, R. 1990. "El Campo Mexicano Ante el Reto de la Modernizacion." Comercio Exterior 40 (9): 816-29. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 503-528 The Optimal Currency Composition of External Debt: Theory and Applications to Mexico and Brazil Stijn Claessens The changes in exchange rates, interest rates, and commodity prices during the past decades have had large impacts on developing countries. Many developing countries have limited access to already incomplete international long-term hedging markets. Thus the question arises whether the currency composition of external debt can be used to minimize exposure to external price risk. Using a utility-maximizing framework, this article shows that, by choosing the optimal currency composition, a country can indeed manage its external exposure. The optimal, risk-minimizing currency composition de- pends on the relation between export receipts and the costs of borrowings in each currency and on the relations among the costs of borrowings in different currencies. A simple methodology can be used to derive the optimal shares of individual currencies and is applied to Mexico and Brazil. The results show that Mexico and Brazil could have lowered their external exposure to a limited degree by continuously altering the currency composition of their debts. The low correlations between the costs of borrow- ings and export and import prices make the currency composition of debt a very imperfect hedging tool, and it is likely that hedging instruments directly linked to prices are preferable. Other things being equal, a strengthening of the dollar will worsen the terms of trade of net commodity exporters and hence reduce their welfare. For net com- modity importers the reverse pattern will hold (Dornbusch 1985, p. 335). . . . for some developing countries, the fall in the dollar increased the burden of debt relative to their economies (World Bank 1987, p. 49). Which of these statements about the effect of exchange rate changes on the welfare of developing countries is correct? Even though placed out of context, both quotations illustrate some of the unresolved issues regarding the effect of exchange rate changes. This article aims to clarify some of these issues and to Stiin Claessens is with the International Economics Department of the World Bank. He would like to thank Stanley Fischer, Ron Duncan, Kathy Mann, Brian Pinto, Darius Malekpour, the referees, and participants in a seminar at New York University for their comments. ( 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 503 504 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 present conceptual and practical guidelines that may help with external debt management in general. The issue of the optimal currency composition of external debt can be ap- proached from a narrow or a broad perspective. Given the existence of external debt, the narrow perspective attempts to determine its optimal composition. Given not only the presence of debt but also the availability of a range of other international financial contracts, the broader perspective attempts to identify the welfare-maximizing liability structure and the subsequent optimal currency composition. Here we take the broader perspective. Developing countries in particular have been affected by large changes in exchange rates, interest rates, and prices of international goods during the past decades. The degree of uncertainty in these international variables is illustrated in figures 1, 2, and 3. Figure 1 plots the nominal effective U.S. dollar exchange rate and a measure of the volatility of this rate for 1977-90. Exchange rate volatility, which had increased after the movement to floating exchange rates in the early 1970s, did not decline in the 1980s. Figure 2 plots the nominal interest rate most relevant for developing countries from 1965 to 1990 as well as the coefficient of variation (CV) of the interest rate over the preceding 24 months at each point in time. Nominal interest rates have experienced large fluctuations, and there have been few periods of tranquillity. Figure 3 plots measures of commodity prices and commodity price volatility during 1962-90. Even though part of the price movements can be explained by shifts in the demand for commodities and supply factors, the large fluctuations in commodity prices resulted in large risks to both producers and consumers. Figure 1. Nominal Effective US. Dollar Exchange Rate: Index and Volatility, 1977-90 Index Volatility 110 50 105-1- 45 100 Nomidnal index I -40 95- '3 90 ( 3 85 - - 25 80 - N 20 75 - ~I15 70 N.10 765°I~ >/ I Volatility 65 j 1977 1979 1981 1983 1985 1987 1989 1990 Note The weights used to create the index of effective exchange rates are the IMF weights. Exchange rate volatility is calculated as the coefficient of vaTiation of the effective exchange rates over the preceding 24-month period. Real effective exchange rates show a very similar pattem for level as well as volatility. Source: IMv, Inenationa Financia Statistics (various years). Claessens 505 Figure 2. Nominal Interest Rate: Index and Volatility, 1965-90 Index Volatility 20 90 18 - I Volatility - 80 16 - -l 70 - 60 -50 4~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 40 8 ~~~~~~~~~~V20 6 Nominal inex10 1965 1970 1975 1980 1985 1990 Note: The nominal rate is the London interbank offer rate (lIBOR) on six-month U.S. dollar deposits (period averages in percent per annum) on which much commercial bank debt is indexed. Interest rate volatility is measured as the coefficient of variation over the preceding 24 months. No distinctaon is made between expected movements and actual deviations from these expectations; overall variability is considered as risk. Source. IMF, Internatfonal Financial Statistis (various years). Figure 3. Nominal Price Index for Developing Countries' Non-Oil Commodity Exports, 1962-90 Index Volatility 160 100 i 90 Nominal index - 80 120 70 100 - ~~~~~~~~~~~~~~~~~60 -50 80 40 60 -/%I I 30 40 Volatility ~~~~~~~~~~~~~10 20 0 1962 1967 1972 1977 1982 1987 1990 Note; The figure plots the commodity price volatility as the coefficient of variation of the IMPf index of 34 nonfuel commodity prices for developing countries in the preceding 24 months. Source: IMP, International Financial Statistfcs (various years). 506 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 For industrial countries, external exposures are small, partly because the export and import patterns of these economies are often diversified. In addi- tion, exposures are largely private, and volatilities are thus not an issue for public-budget management. These volatilities have of course been important issues for industrial countries in the larger context of macroeconomic manage- ment, but seldom in the narrow context of their financial impact on the public sector. Only a few industrial countries, such as Norway and the Netherlands, have felt a direct financial impact of commodity price and exchange rate move- ments on their public sector's budget. But for many developing countries external exposures are large because of a more concentrated export and import pattern. They are also primarily public or quasi-public (as in the case of public external debts and the exports and imports of state enterprises). The high volatilities have had serious implications for the government budgets of these countries and other costs through their disruption of the economy and associated resource misallocations. This is demonstrated by Indonesia, in which the ratio of debt service to exports rose from 8.2 percent in 1981 to 27.8 percent in 1987. More than 65 percent of the increase could be explained by the depreciation of the U.S. dollar after 1985 and the fall in oil and other commodity prices in 1986 and 1987. If commodity prices and cross- currency exchange rates had remained at their end-of-1982 values, the ratio of debt service to exports for all developing countries would have been approxi- mately 17 percent as opposed to the actual ratio of 24 percent in 1987. The fact that we observe large volatilities in several external prices (exchange rates, interest rates, and commodity prices) and corresponding large changes in measures of debt service burden draws attention to the importance of properly measuring external exposures. Large volatilities of external variables do not necessarily have to affect a firm or a country adversely. The issue is whether and to what extent the interaction between external price movements and overall external transactions affects the firm or country adversely. In order to determine this, a framework for measuring a country's overall (net) exposure to external price uncertainty is needed; the magnitude of external exposures can then be measured. Once external exposure is quantified, one can try to manage it by using financial instruments to transfer exposures to other parties more able to absorb them. What type of financial instruments are available to manage external risks? Firms in industrial countries have access to and use many financial instruments to reduce and transfer risks. Examples are futures, options, and swaps on cur- rency and interest rates, and more recently on commodities. In many developing countries, however, neither the private sector nor the public sector has used these instruments to the same degree as firms in industrial countries have to shift risks to (international) financial markets in line with comparative advantage. Some hedging instruments-such as currency, interest, and commodity futures and options-are in principle available to most developing countries. However, these have short maturities and, even when rolled over, provide limited hedging Claessens 507 value over longer periods. Furthermore, the problem with rolling-over coverage based on short-term maturing instruments is that this usually implies a large exposure in these instruments. This can imply large margin calls or large option premiums for these instruments, which makes them less attractive for foreign- exchange-constrained countries and reduces them largely to self-insurance in- struments. The markets for longer-term hedging instruments-such as currency, commodity, and interest swaps, and especially commodity price-linked instruments-are relatively thin. And these longer-term hedging instruments tend to be unavailable to most developing countries due to institutional, credit, and other constraints. The presence of large exposures, for which limited or potentially expensive hedging tools are available, raises the question whether a developing country can use alternative means to manage its external exposure. One possibility, of course, is that the country actually diversifies through the sourcing, producing, and exporting of a broader mix of products. However, many developing coun- tries depend heavily on primary exports and may have little room to diversify into other export products as a means to hedge risks. And self-insurance through diversification may not be the most efficient option (because it may not be in line with the country's comparative advantage) and may take a long time to achieve. A possibly more effective mechanism is to use the currency composition of existing and new external long-term liabilities to reduce external exposure. For example, Kahn (1988) mentions that a number of rescheduling agreements be- tween developing countries and commercial banks have provided currency con- version options for non-U.S. banks. This indicates that, in addition to the possibility of incurring new loans in the currencies of choice, developing coun- tries may be able to alter the currency composition of their existing external debt in rescheduling negotiations, possibly to their own and the commercial banks' advantage. The currency composition of external debt may be able to achieve hedging advantages similar to those of a portfolio of short-term hedging instru- ments. Changes in external earnings may be offset by simultaneous shifts in the costs of borrowings, but without possible adverse margin calls or premiums. The effectiveness of the currency composition strategy compared with, say, a rolling-over of futures strategy, will depend on the relation among the different external risks. Thus exchange, interest, and commodity price risk management and determination of the currency composition must be integrated. Several questions follow if a country wants to use the currency composition of external liabilities to manage external exposures. What is the overall objective function to be managed (maximized), and what is the definition of risk that follows from the chosen objective? Which factors play a role in choosing the optimal currency? What kind of rules follow for managing the external liability of a country? The existing literature provides little guidance on these questions. Some objective functions and rules have been suggested for determining the currency denomination of borrowings by a developing country: matching the currency composition of debt with the trade direction or currency composition 508 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 of export revenues, matching it with the composition of the basket of currencies with which the domestic currency is pegged or managed, and matching it with the currency composition of the foreign exchange flows into the country (see further Lessard and Williamson 1985). However, most of these rules have not explicitly been related to a specific goal or objective, and the risks have not been explicitly defined. In general, most decision rules proposed have been of an ad hoc nature, and detailed guidelines for implementation have not been developed. This article develops therefore an integrated model for deriving the optimal currency composition for a small open economy facing external uncertainties. The model uses international portfolio theory and identifies the factors that determine the optimal currency decision. The optimal currency composition for a country to hedge itself against commodity price, interest, and exchange rate movements-and not to speculate on relative exchange rate movements- depends only on the covariances between the effective costs of borrowings and the country's exports and on the covariances among the effective costs of bor- rowings. A simple operational procedure, which relies on the coefficients of appropriately specified ordinary least squares regressions for calculating the optimal currency shares, is developed and applied to Mexico and Brazil. I. THE PORTFOLIO MODEL The optimal currency composition of external debt is determined for a small open economy that acts as a price taker in international goods markets and that faces perfect world capital markets. We will use here a simple, two-period model to determine the optimal portfolio investment and consumption decisions. Alter- natively, an intertemporal capital asset pricing model-as in Merton (1971), Breeden (1979), and Stulz (1981)-could be used. In that case the model would have features similar to Krugman (1981), Macedo (1982 and 1983), Fraga (1986), and Stulz (1984). Such a model is derived in Claessens (1988). It is assumed that the country can be represented by one domestic individual who lives two periods, or, alternatively, the government acts benevolently in choosing the optimal fiability composition for its own citizens. This approach is appropriate if the private citizens have limited access to foreign capital markets. For many developing countries this is the case and is confirmed by the small amount of private, nonguaranteed debt compared with total public and publicly guaranteed debt (about 15 percent). In cases where the private sector has more access to foreign capital markets, it may be more appropriate for the govern- ment to concentrate on managing its own external liabilities optimally with respect to its fiscal revenues and expenditures. The country has a fixed first-period endowment and can invest in activities that produce goods for export. In the first period the consumer makes invest- ment, consumption, and borrowing decisions based on its expectations of second-period variables. In the second period, the consumer receives payments on its exports, which it uses to finance debt service payments and imports of Claessens 509 goods. Export receipts and import payments are uncertain because international commodity prices are uncertain. The consumer maximizes its welfare function defined over this and next period's consumption of each good, ci, i = 1, . . . K. The country does not face any borrowing or lending constraints and can denominate its liabilities (borrow) and invest its wealth (lend) in N currencies. The amount borrowed or lent in currencyj is denoted by Bj, which is a represen- tative element of B, the N x 1 vector of demand for foreign bonds. The effective cost (or returns) of Bj in terms of the domestic currency depends on the foreign interest rate and the exchange rate. Each of the N foreign interest rates, rj, is assumed to be uncertain. Similarly, each of the N exchange rates, ej, defined in terms of units of domestic currency per unit of foreign currency (for example, pesos per dollar), is also uncertain. The effective (gross) interest rate on Bj is defined as Rj, and R* is the N x 1 vector of the expected effective costs on the N foreign loans. Rj"is thus given by the (gross) foreign interest rate rj-times (one plus) the rate of depreciation of the domestic currency in relation to currencyj, that is, (1 + r',%) (ej,2/ej1,). It is assumed that the effective costs of liabilities denominated in the currency of each foreign country do not necessarily equal each other. And the standard finance assumption of no transaction costs is made. Due to certain barriers, domestic investors are prevented from investing in foreign firms and stocks. This restriction prevents domestic investors from using foreign equities as hedging instruments against unanticipated changes in exter- nal prices. Foreign investors cannot own domestic firms. We assume that there are no nontraded domestic assets (such as human capital). (See Svensson [1989] for the case of nontraded assets.) Domestic assets that are traded are domestic bonds, denoted by D, which are in zero net supply, with a gross interest rate of R, and shares of export firms. Export goods are produced by domestic, profit-maximizing firms whose shares are not traded abroad. In the first period investment levels, denoted by Ih, h = 1, ... L are determined for production of each of the L export goods in the second period. The production functions, denoted by Q(Ih), exhibit decreasing returns to scale, and Q(I) denotes the vector of second-period outputs. First- period prices of all goods are normalized at one, and second-period prices in the domestic currency for the export goods are denoted by Ph. In the second period the country will receive payments per unit of exports in domestic currency equal to P . The word "price" should be interpreted broadly here and includes all factors that determine the value of one unit of an export good. The exact currency in which the payments are received is immaterial to our results: what matters is that the payments have a stochastic unit value in terms of domestic currency. Second-period import good prices are Pi. The first- and second-period asset flow constraints for the representative con- sumer are defined in equations 1 and 2, respectively: K N L (1) Zc1 i = YBj - LIsI + D I i h S10 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 K N L (2) RPi62,i Rj:Bj + EPQ(Ih)-RD where first-period exchange rates are normalized to one. It is assumed that the (representative) domestic investor maximizes an ex- pected utility function, which is von Neuman-Morgenstern and depends only on the consumption of the K commodities. The expected utility function of the investor is given by (3) U( Ict T) + 3E [ U( Ac i)2 where II is the standard multiplication function, E is the expectation operator conditional on all information available at time 1, and : is the factor of time preference. The utility function is characterized by constant expenditure shares: the investor will always spend a share cxi of total expenditures on good i. This allows us to represent the objective function in terms of one composite good for which we will use the notation Ci, i = 1,2, with average price P (geometrically weighted with weights ai). The investor will maximize utility subject to the constraints on income. (4) max U(C1) + 3E[U(C2)] Bj, D,I,, N L subject to Cl - YBj + E 1/2 - D = 0 j h N L and PC2 + jR1B, - P bQ(Ih) + RD = 0. j h2 To simplify the solution for the optimal amounts borrowed in each foreign currency, invested in domestic bonds, and held in first-period investment levels, we assume that the utility function U is quadratic, that is, U = aC - (b/2)C2. For the more general case, see Claessens (1988) or Svensson (1988). We define Vr, as the N x N variance-covariance matrix of the effective costs of foreign borrowings deflated by the consumer price index and VMP as the variance- covariance matrix of those costs with the changes in the unit values of the export goods (again expressed in domestic currency and deflated by the consumer price index). After imposing that the domestic market for bonds clears, the optimal borrowings are, where 1 is a N x 1 vector of ones (see further the appendix): (5) B = V-1 + V1-rEV)p3 Q(I). E( UI) Equation 5 implies that the demand for foreign bonds consists of two mutual funds. The first fund is the speculative portfolio: Claessens 511 (6) B5 = Vr-rE [(1 / P)R; -1 T] where X = U'/(/E(U2)) is the effective rate of intertemporal substitution. The composition of the speculative portfolio of foreign bonds depends on the ex- pected effective real costs of excess borrowings. In equilibrium these costs will depend on the real rate of return on investment, adjusted for risks, and the inverse of the variance-covariance matrix of effective real costs. The demand for the fund depends on a risk aversion parameter, y, where -y = - bIE(UI). The higher the aversion against risks, the higher the y, and the lower the amount borrowed on account of the speculative fund. The second portfolio is the minimum variance hedge: (7) Bm = Vt-VrpQ(I) and is independent of the level of risk aversion. The investor will borrow in foreign currencies to insulate against changes in the domestic currency value of future receipts from exports. These changes are caused not only by movements in the domestic exchange rate at which export receipts are converted into domes- tic currency, but also by movements in the unit values of export goods. Since cross-currency movements can be related to unit value movements, foreign bor- rowings can serve as a hedge, and the stronger the correlation between domestic currency values and the cost of borrowings the larger the demand for foreign borrowings. The equation for the optimal amount of loans is specified with respect to the effective, real cost of foreign liabilities, relative to the domestic borrowing costs. Alternatively, the portfolio can be written in nominal terms and then split into three funds: a nominal speculative fund, a minimum variance fund, and a price hedge (see the appendix). In addition, we need to assume that the distribution of all prices is lognormal. The nominal fund will be V-1 R, and the minimum variance fund will be V -I V pQ(I), and the matrices are now based on the nominal costs of borrowings and the nominal unit values of exports. The price hedge fund will be - V-1 VrP, where VYP is the vector of covariances between the effective costs of borrowings and the consumer (import) price index. The de- mand for the price level hedge will be -(1 - 1 /by) (see the appendix). It is useful to look at the price level hedge when we assume that the foreign interest rates are constant and thus that the uncertainty in the effective foreign interest rates is only a result of exchange rate uncertainty. This implies that the variance-covariance matrix of effective costs (Vrr) would be equal to the variance-covariance matrix of exchange rates, and the vector of covariances between the effective costs of borrowings and the consumer price index (Vrp) is equal to the vector of covariances between exchange rates and consumer prices. Here we look at the price level hedge in two cases. In the first case changes in the prices of imported goods are perfectly correlated with the exchange rate, that is, the law of one price holds for imported goods. Holding foreign bonds provides then a perfect hedge against unanticipated changes in the domestic price of foreign goods. The demand for each foreign bond will be determined by the 512 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 share of total consumption expenditures spent on the imports from each cur- rency area. In the second case purchasing power parity holds. Foreign bonds are perfect hedges against unanticipated domestic price movements as exchange rate movements offset perfectly relative price movements. With no unanticipated inflation abroad there will be no demand for foreign bonds on account of the price level hedge. To summarize, the model outlined above indicates that the optimal external liability composition of a country depends on the following factors: * The expected costs of borrowing in each of the foreign currencies relative to the domestic cost of funds * The variance-covariance matrix of the expected costs of borrowing in each foreign currency * The variance-covariance matrix of domestic goods prices and expected costs of borrowings in each foreign currency * The shares of consumption expenditures spent on the different goods in the country * The export receipts of the country * The vector of covariances between export receipts and expected costs of borrowings in each foreign currency • The level of risk tolerance in the country. When purchasing power parity holds and foreign prices are (relatively) stable, the rules for the optimal currency composition of external liabilities can be further simplified. A country may be very risk averse and want to hedge itself against commodity prices and interest and exchange rate movements. Or it may take the view that the expected real costs of borrowing in each of the foreign currencies are equal to the domestic real cost of funds (that is, it may not want to speculate on the relative costs of different instruments). In these cases the opti- mal composition will depend only on the vector of covariances between export receipts and the expected costs of borrowing in each foreign currency, and the variance-covariance matrix of the expected costs of borrowing in each foreign currency. II. PRACTICAL RULES FOR THE CURRENCY COMPOSITION OF EXTERNAL DEBT From the theoretical analysis it follows that if a country does not want to take an active view on exchange and interest rate movements-and associated costs of borrowings-or if it is relatively risk averse, then its optimal net borrowing portfolio (gross borrowings minus foreign assets, that is, reserves) will be the risk-minimizing hedge portfolio. This is a very familiar result from simple one- period mean-variance hedging models such as those used to determine the opti- mal amounts of futures to buy or sell to hedge an exposure (see, for example, Gemmill 1985). Techniques introduced by Adler and Dumas (1980) and later on Claessens 513 refined by others for the operational measurement of the economic exposure of a firm to external risks are also similar. We will pursue the latter similarity further. Economic exposure to external risks for a firm has been defined in terms of the sensitivity of its objective function with respect to unanticipated changes in external variables. The firm's objective function may be defined in terms of the net present value of future expected cash flows or in terms of near-term cash flows or profits. When the hedging instruments available are foreign borrow- ings, the operational way of measuring exposure as a cash flow sensitivity can then be reduced to measuring the covariance between cash flows and the effec- tive costs of borrowings relative to the variance of total cash flows. In other words the measure is the ordinary least squares (OLS) regression coefficient of cash flows on the relevant effective costs of borrowings: cov(CF,,R,)/Var(CF,), where CF, is the cash flow in terms of domestic currency in period t and R, the relevant effective cost of borrowings at time t. The absolute values of these estimated coefficients provide then the minimum variance hedging quantities. These are the amounts of foreign borrowings (multiple exposure measures fol- low if one regresses the cash flows on multiple cost of borrowings). A related exposure measure is discussed by Oxelheim and Wihlborg (1987). The portfolio model above is consistent with this approach because it indi- cates that-when the expected foreign and domestic real costs of funds are equal and purchasing power parity holds between foreign and domestic goods-the optimal composition is simply determined by the minimum variance fund (V-1 VP.Q(I)). In the practical approach we consider neither the relative costs of funds in different currencies nor the benefits of using foreign borrowings to hedge against domestic price uncertainty. It is unlikely that a developing country will be able to successfully exploit (speculate) at an acceptable risk the small differences between costs of funds. The international financial capital markets at large are better equipped for this and will assure that these differences remain small. Domestic price uncertainty is more often the result of endogenous government policy choices than of exogenous influences; consequently it is unlikely that it can be hedged using foreign borrowings. It is not assumed that the law of one price holds among foreign goods (that is, Pi is not necessarily equal to ehPi/h for all i and h, where Pi,b is the price of traded good i in terms of foreign currency h) and foreign exchange rates can influence the prices of foreign goods. We leave the precise mechanisms through which nominal foreign exchange rate move- ments influence prices of foreign goods unspecified. For the effects of currency movements on the behavior of absolute and relative prices, see Giovannini (1986), Dornbusch (1987), and Varangis and Duncan (1987). More generally, the risk-minimizing borrowing portfolio would be based on the relation between the ability of a country to service its debt and the effective cost of debt service in each of the relevant borrowing currencies. The main problem with this approach is measuring the ability of a country to pay, some- 514 THE WORLD BANK ECONOMtC REVIEW, VOL. 6. NO. 3 thing that has been the subject of extensive research in the sovereign debt litera- ture. In addition, ability and willingness to pay may differ, thus resulting in actual payments that may be determined by a bargaining game between a debtor and its creditors (see Eaton [1990] for a survey). In that case the factors deter- mining the payment behavior belong on the right side of the OLS equation. We suggest here that total exports of a country is the relevant measure to hedge. (Healy [1981] uses a similar simple regression technique for the optimal diversification of foreign exchange reserves.) Total exports measured in domes- tic currency are thus regressed on the effective costs of foreign funds (both expressed as percentage changes), and the coefficients are used to calculate the optimal liability portfolio. The equation to be estimated becomes then (8) ln(P"Q),+i - ln(P*Q), = D3i{lnL(1 + riD()ei,t±) In [( i,t1)( e1t-)3 ± Ut. Further disaggregation, for example by commodity groups or direction of trade, can also be used to calculate the optimal portfolio shares if one expects that the disaggregated relationships are more stable. For instance, prices of some com- modities tend to have a close relationship with a single currency because the supply (or demand) tends to come from (or go to) that currency area. For example, the price of coniferous timber products has been closely associated with the Nordic countries' currencies because they are large suppliers and influ- ence the dollar price of timber. Similar relationships may exist for other com- modities (see further Lessard and Williamson 1985). III. APPLICATIONS TO MEXICO AND BRAZIL In applying the model, the objective is to find the currency composition of net liabilities (gross liabilities minus reserves) that minimizes the variability in do- mestic currency of export earnings net of foreign liability debt service. We as- sume that at the beginning of each period new net liabilities are incurred, which are then paid off at the end of the period. We have thus a rolling portfolio of liabilities. If the composition of the optimal portfolios obtained is now stable over time, then little rebalancing of the portfolio would be required, and the rolling portfolio would essentially mimic a portfolio of long-term liabilities that has debt service payments falling due each period. We would then be justified to use the currency composition of long-term liabilities (which is inherently diffi- cult to change over short periods) to hedge the short-term exposure arising from changes in export earnings. Large reductions in overall variability would indi- cate that the currency composition of external debt can be an effective tool for risk management, especially since it would avoid the drawbacks of rolling over short-dated instruments (margin calls and premiums). Claessens 515 We use monthly data for 1973-89 for Mexico and for 1973-86 for Brazil. In accordance with the model, we use as the dependent variable Mexico's and Brazil's total exports, expressed in domestic currency. (All data are from the International Monetary Fund's International Financial Statistics.) As possible borrowing currencies we choose the U.S. dollar, Japanese yen, and German deutsche mark (DM). These three currencies were selected for the borrowing portfolio because Mexico and Brazil's current debt is largely denominated in these currencies (respectively, 70 and 72 percent U.S. dollar, 15 and 11 percent DM, and 8 and 10 percent yen); the DM and other European currencies are highly correlated, and the three chosen currencies are therefore good proxies for the diversification available in international currency markets; and these coun- tries' future access to financial markets is likely concentrated in these currencies (or currency blocks). The estimations were based on the (monthly) percentage change in the costs of borrowing in the different currencies expressed in local currency. The foreign interest rates used were the monthly quotes of the respective six-month LIBOR rates, which were then multiplied with the exchange rate depreciation over the next month. Table 1 presents some data on the average annualized costs in pesos during 1973-89 and in cruzados during 1973-86. The variability in the local cost of foreign borrowings was quite high. The variability of foreign costs can be compared with the monthly coefficient of variation of export earnings in local currency: 62 percent for Mexico and 233 percent for Brazil. The estimate for the optimal portfolio for the whole sample period was based on regression equation 6. The estimates for the optimal amount borrowed in each currency lead to the portfolio compositions for Mexico and Brazil given in table 2. The absolute amount to be borrowed as a percentage of export earnings is 0.25 percent for Mexico and 5.62 percent for Brazil. The small fractions reflect the fact that the variability in export earnings is much less than the variability in effective costs of borrowings and that the correlation between the two is low. The results indicate that, of the absolute amount to be borrowed, Table 1. Average Annual Costs of Borrowings (percent) Cost in domestic currency Coefficient Currency of debt Mean of variation Minimum Maximum Mexico, 1973-89 (pesos) U.S. dollar 46 256 -156 1,133 DM 46 266 -138 1,059 Yen 48 249 -144 1,100 Brazil, 1973-86 (cruzados) U.S. dollar 70 90 6 475 DM 70 106 -92 467 Yen 72 104 -105 444 Source: Author's calculations. 516 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Table 2. Optimal Portfolio Compositions for Mexico and Brazil Percentage of total amount borrowed in Amount borrowed Durbin- U.S. as a percentage of Watson dollars DM Yen export earnings R2 statistic Mexico, 1973-89 111.0 16.3 -27.4 0.25 0.068 2.00 Brazil, 1973-86 98.0 3.4 -1.5 5.62 0.053 2.04 Source: Author's calculations. Mexico should borrow more than 100 percent in U.S. dollars, borrow some additional DM, and hold some of the foreign funds in yen reserves. Brazil should also borrow U.S. dollars and DM and hold yen in reserves. If Mexico and Brazil had borrowed and invested their reserves in this fashion during 1973-89, the variance of monthly exports net of debt service would have been slightly lower compared with no borrowings and no hedging. The small reduction in uncertainty reflects the fact that the fits of the estimated equations are relatively poor (R2 of only 0.068 and 0.053 for Mexico and Brazil, respec- tively) and that the underlying relationships between export earnings and costs of borrowings are not stable over time. The latter implies that portfolios should be more frequently adjusted and thus that the hedging effectiveness of a constant portfolio is limited. The results so far were calculated using historical data and then applied over the previous sample period. The results therefore amount to an in-sample test. However, such historical comparisons cannot indicate the effectiveness of the strategy as a planning tool. After all, in reality, the country would only have had information up to time t with which to determine the composition for time t + 1. To show whether or not this technique is effective in terms of risk minimiza- tion, we perform a number of "out-of-sample" tests. We construct portfolios for period t + 1 that are based on a sample of observations up to time t. The sample is then updated to include time t + 1 information, portfolios are constructed for period t + 2, and so forth. The relative effectiveness of this strategy of rolling hedges is then calculated as the risk reduction realized over the whole hedging period compared with the risk reduction realized by the other strategies. Specifically, we calculated optimal portfolios for Mexico and Brazil for each quarter using the last 48 monthly observations up to that quarter. This results in 51 portfolios for Mexico and 39 for Brazil. Table 3 summarizes the results. Annex tables A-1 and A-2 list the individual portfolios, indicating the mean square residual, the (adjusted) R2, and the Durbin-Watson statistic. For some individual portfolios the total borrowing amount as a percentage of export earnings is negative, thus implying that a net lending strategy would have been preferred at some points in time. The summary table indicates that the standard deviation of the portfolio shares is quite large. Portfolio shares vary between -3,622 and 2,997 percent for Mexico and between -123 and 202 percent for Brazil. Some of the large Claessens 517 individual portfolio shares are caused by the very low absolute amount to be borrowed in a specific quarter, which inflates the individual portfolio shares. For example, the total borrowing amount as a percentage of export earnings of portfolio 37 for Mexico is only 0.000087. Some of the outliers in the case of Mexico are also caused by the two large discreet valuations that occurred in 1982, which distort the estimation for all portfolios that include these data points. This is for instance reflected in the amounts to be borrowed for portfolios 23 through 39 (1982-87 through 1986- 87, which thus include the 1982 data), which are all negative. To control for this, we have also excluded from the sample the negative sums. We also solved for the optimal shares while imposing some constraint on the absolute amounts to be borrowed, for example, a certain ratio of debt service to exports. The results for both approaches were, however, not significantly better. Similar ex- planations exist for Brazil for portfolios after 1983-10. Excluding these periods, the estimates for the portfolios indicate a consistent strategy of borrowing in U.S. dollars combined with some minor DM or yen borrowings or reserve holdings. The absolute amounts to be borrowed are lower than the actual borrowings by these two countries and are closer to the net transfers these countries pay on their debts, expressed as a fraction of exports, consistent with the idea that the portfolios are rolled over. With these individual portfolios, we performed the following out-of-sample test. For any portfolio one can calculate the residual in each of the next three months that was not hedged as u, = AE, - Zi ARi, , where AE, is the realized monthly percentage change in export earnings, ARi,t is the realized monthly Table 3. Summary Statistics of Optimal Portfoliosfor Mexico and Brazil (percent) Standard Value Variable Mean deviation Minimum Maximum Mexico Share of debt in U.S. dollars 257 499 -89 2,997 DM 63 174 -72 940 Yen -309 679 -3,622 29 Total borrowing amount as a percentage of ex- port earnings 2 3 -1 8 Brazil Share of debt in U.S. dollars 56 93 -97 202 DM -15 23 -123 7 Yen 2 10 -8 38 Total borrowing amount as a percentage of ex- port earnings 2 11 -20 17 Note: For Mexico there were 51 observations; for Brazil there were 39. Source: Author's calculations. 518 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 percentage change in the costs of borrowing (for i = U.S. dollar, DM, and yen, all expressed in local currency), and fi are the shares, which change every three months. A new portfolio is then used for the next three months to calculate the residual, and so forth. This allows us to compare the effectiveness of a particular borrowing strategy with any other borrowing strategy. This was done here by comparing the strategy where portfolio shares are adjusted every quarter according to the estimates derived from the observations during the previous 48 months-here called the rolling strategy-with three alternative strategies. The first alternative strategy was Mexico's and Brazil's actual debt composition in 1988 (reported above); the second one was the composition that was historically optimal for 1977-89 and 1977-86 (89.8 and 126.6 percent U.S. dollar, 11.5 and -25.0 percent DM, and -1.4 and -1.6 percent yen, respectively, for Mexico and Brazil); and the third one was the composition that was historically optimal for 1973-89 and 1973-86 (reported in table 2). For all three alternatives only the composition of debt was changed. The total amount to be borrowed (or loaned) was assumed to be equal to the rolling strategy (otherwise using the optimal amounts of the historically optimal portfolios would amount to an in-sample test). For both countries the residuals of all four portfolios were calculated; in total there were 152 residuals for Mexico and 117 for Brazil. The results are shown in table 4. For Mexico the results are very encouraging. The rolling strategy achieved approximately a 46 percent lower coefficient of variation of residuals compared with the actual borrowing strategy. The compositions that were his- torically optimal for 1977-89 and 1973-89 achieved less or no risk reduction compared with the actual portfolio, and the rolling portfolios outperformed all three portfolios. For Brazil, the results are less encouraging. The rolling strategy achieved only a 2 percent lower coefficient of variation than the actual borrow- ing strategy. Much risk reduction was achieved with the rolling portfolio com- pared with the 1977-86 historically optimal strategy (a 69 percent reduction), but not compared with the 1973-86 historically optimal strategy, which per- Table 4. Risk Reduction Achieved with Optimal Portfolios (compared with the actual portfolio) Mexico Brazil Coefficient Reduction Coefficient Reduction Portfolio of variation (percent) of variation (percent) Actual 686.0 n.a. 367.0 n.a. Rolling strategy 371.9 46 358.9 2 Optimal composition for 1977-89 565.8 18 n.a. n.a. 1973-89 916.2 -33 n.a. n.a. 1977-86 n.a. n.a. 1,153.0 -214 1973-86 n.a. n.a. 356.0 3 n.a. Not applicable. Source: Author's calculations. Claessens 519 formed overall the best. For both countries, all strategies did much better than a strategy using the actual amounts as well as the actual shares in which the countries borrowed (statistical results are not reported here). None of the strategies performed better than a strategy where both composi- tion and amounts were determined in an historically optimal way. This may be expected since historical portfolios will always lead to risk reduction given the benefits of hindsight. The instability in the covariances between changes in costs of borrowings and exports implies that planned portfolios will always result in less hedging. Only with more stability can better results be expected from planned portfolios compared with historical ones. This is, however, not a fair comparison, because, as noted above, in-sample tests of effectiveness of hedging are misleading. Other out-of-sample tests were also performed. One of these was based on a rolling hedge strategy using the past 24 months of observations. This led to much larger standard deviations in borrowing shares, a result of the fact that the sample period for the estimation was halved. Compared with the actual and historical optimal composition this strategy led to little or no risk reduction. IV. CONCLUSIONS This article shows that an integrated analysis of the external risks an economy is facing is necessary to measure external exposures correctly. The article uses international portfolio theory for such an analysis and finds that in particular exchange and commodity price exposure need to be integrated. A developing country wishing to implement more active management with respect to the currency composition of its net liabilities will need to consider its objective function very carefully. Nominal dimensions (such as direction of trade flows and currency composition of cash-flows receipts) do not necessarily provide correct indications for the optimal currency composition of debt. Determining the optimal currency composition requires a careful empirical investigation of relationships between cross-currency exchange rates and indicators of the coun- try's external account, such as terms of trade, exports, and the non-interest current account. Determining the optimal liability structure requires facts such as the historically observed inverse relationship between the value of the dollar and dollar commodity prices. This relationship indicates for instance that non- dollar currencies can perform a hedging function for (primary) commodity ex- porters because commodity prices, and thus export earnings, increase when nondollar currencies rise in value and vice-versa. The article applies a reduced form of the analytical model developed to Mex- ico and Brazil and finds some, albeit limited, reduction in external exposures from constructing optimal, rolling portfolios. The results indicate, however, that portfolios are unstable over time as a result of time-varying covariances. Thus transactions costs can significantly reduce the benefits of adjusting portfo- lios frequently. Results obtained elsewhere (Kroner and Claessens 1991) indicate 520 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 that even when using Generalized AutoRegressive Conditional Hetero- skedasticity (GARCH) to deal with time-varying covariances-and to achieve less time-varying portfolios-instability remains an issue and that using the currency composition of long- term debt in a cost-effective way to hedge against real shocks will be difficult. In general, the results indicate that the currency composition of debt is an imperfect hedging tool against external price uncertainty. Other types of contin- gent contracts (such as commodity price linked instruments) are likely better able to transfer risk from developing countries to the international capital mar- kets. Unfortunately, the longer-term spectrum of commodity risk management instruments is not yet well developed for many commodities of interest to devel- oping countries and will in any case be difficult to access because of credit constraints. In principle, developing countries should then also be using the short-dated commodity-linked markets, such as options, forwards, and futures, to hedge directly against price risks. These will provide the countries with some risk reduction benefits when longer-term markets become available or the coun- tries achieve some real diversification. Unless strong relationships are found between the prices of export and import products, the currency choice of exter- nal debt may for many developing countries be largely immaterial as it relates to reducing overall exposure to external factors. APPENDIX. DERIVATION OF THE OPTIMAL PORTFOLIO MODEL For simplicity we will derive first the optimal portfolio amounts in case only one foreign bond and one domestic investment opportunity are available. The generalization to N foreign bonds and L investment opportunities will then follow easily. The Simplified Case of One Foreign Bond and One Domestic Investment Opportunity From the maximization problem in equation 4 the following first-order condi- tions with respect to the amount of investment, foreign bonds, and domestic bonds can be derived. (A-1) 0 -U' + OEUt Q) (w.r.t. I) (A-2) 0 =-U' + j3E(Ujy) (w.r.t. B) (A-3) 0 =-U' + OE(U,p) (w.r.t. D) Given the assumption of a quadratic utility function, we can rewrite the first- order conditions as Claessens 521 (A-1') 0 = - U, + I E(UU)E (Q ) - bcov ( p ,Q p) tP; Q, R 0 P :Q' R + bcovt p , p B + bcovV ~p , p DI (A-2') 0 U1 + F[E(U2)ER p ) R bcv(p P Q) + bcov( p, p!)B + bcov( p, - )D] (A-3') 0= - U, + 3E(U2)E( p) - bcov(p, pQ) + bcov(p, -.p )B + bcov( p,P)D1. Imposing that in equilibrium the amount of domestic bonds will be in zero net supply, equation A-3' can be solved for the domestic equilibrium nominal inter- est rate R as (A-4) E) R ___ b Ro v 'R Q (A-4) E(p) = OE(UI)+ E(U') p p - E(U') p pB. After some manipulation equation A-2' can be used to find the expression for the optimal amount of foreign bonds as ER :F\) - 1 c/ R5 P*\ (A-5) B - J-E(U') + c / Q(p), b ,)VAR (Rp) VAR p) The equilibrium domestic interest rate-derived in equation A-4-can alter- natively be used in equation A-5, in which case the demand for foreign bonds will depend on the real interest differential and the covariance of the real interest differential with foreign interest rates. The Generalized Case of N Foreign Bonds and L Domestic Investment Opportunities In case of L domestic investment opportunities and N foreign bonds, there will be 1 + L + N first-order conditions: one for domestic bonds, one for each domestic investment opportunity, and one for each foreign bond. The N differ- Table A-1. Composition of Rolling Portfolios for Mexico, 1977-89 Amount to be borrowed as a Mean Durbin- Portfolio percentage of square Adjusted Watson number Year Month U.S. dollar DM Yen export earnings residual R2 statistic 1 1977 1 4.2936 0.8412 -6.135 -0.0018 0.04 0.24 1.97 2 1977 4 4.6652 0.9101 -6.575 -0.0016 0.05 0.23 2.04 3 1977 7 4.6664 0.9008 -6.567 -0.0016 0.05 0.23 2.11 4 1977 10 4.9198 0.9982 -6.918 -0.0015 0.05 0.22 2.12 5 1978 1 13.9847 2.9492 -17.934 -0.0005 0.04 0.21 2.09 6 1978 4 29.9681 7.2515 -36.220 0.0002 0.05 0.20 2.09 7 1978 7 0.8349 0.3283 -0.163 0.0044 0.05 0.17 2.10 8 1978 10 0.8376 0.3278 -0.165 0.0043 0.05 0.16 2.09 9 1979 1 0.8645 0.3724 -0.237 0.0042 0.04 0.14 1.91 10 1979 4 0.8470 0.3726 -0.220 0.0042 0.04 0.16 2.06 11 1979 7 1.4379 -0.0401 -0.398 0.0027 0.03 0.13 2.07 12 1979 10 0.8864 0.3306 -0.217 0.0040 0.03 0.14 2.06 13 1980 1 0.6550 0.2366 0.108 0.0045 0.03 0.13 1.99 14 1980 4 1.0901 -0.2329 0.143 0.0029 0.03 0.14 1.84 15 1980 7 1.0343 -0.1959 0.162 0.0031 0.02 0.16 1.97 16 1980 10 -0.8930 -0.3971 0.290 -0.0015 0.02 0.02 1.97 17 1981 1 1.0446 -0.0808 0.036 0.0078 0.02 0.10 1.91 18 1981 4 1.0866 -0.0834 -0.003 0.0078 0.02 0.15 1.66 19 1981 7 1.0078 -0.0326 0.025 0.0184 0.02 0.11 1.81 20 1981 10 1.0071 -0.0132 0.006 0.0373 0.03 0.06 1.92 21 1982 1 1.0094 -0.0114 0.002 0.0459 0.03 0.10 2.00 22 1982 4 1.0747 -0.0926 0.018 0.0074 0.03 0.06 1.84 23 1982 7 -0.5712 -0.0595 -0.369 -0.0105 0.03 0.06 1.97 24 1982 10 -0.6164 -0.0429 -0.341 -0.0139 0.04 0.06 1.97 25 1983 1 1.1022 -0.4414 -1.661 -0.0013 0.04 0.06 1.97 26 1983 4 1.3839 -0.5198 -1.864 -0.0012 0.04 0.07 1.96 27 1983 7 1.6296 -0.6995 -1.930 -0.0010 0.04 0.06 1.97 28 1983 10 1.6611 -0.7202 -1.941 -0.0010 0.04 0.06 1.92 29 1984 1 1.6979 -0.6765 -2.021 -0.0009 0.04 0.06 1.97 30 1984 4 1.8010 1.0433 -3.844 -0.0008 0.04 0.08 1.97 31 1984 7 1.7718 1.0413 -3.813 -0.0008 0.04 0.08 1.96 32 1984 10 1.8518 1.0763 -3.928 -0.0008 0.04 0.08 1.97 33 1985 1 1.8328 1.0542 -3.887 -0.0008 0.04 0.08 1.95 34 1985 4 3.1815 1.9637 -6.145 -0.0004 0.04. 0.07 1.97 35 1985 7 1.4076 0.7766 -3.184 -0.0011 0.04 0.09 1.80 36 1985 10 4.3707 2.1911 -7.562 -0.0004 0.03 0.08 1.97 37 1986 1 17.6909 9.4037 -28.095 -0.0001 0.03 0.06 1.97 38 1986 4 0.4354 0.1885 -1.624 -0.0045 0.03 0.07 2.00 39 1986 7 0.4248 0.1910 -1.616 -0.0047 0.03 0.08 2.01 40 1986 10 2.4271 1.5531 -2.980 0.0006 0.02 0.09 1.94 41 1987 1 0.9899 0.0126 -0.003 0.0740 0.01 0.30 1.88 42 1987 4 1.0161 0.0128 -0.029 0.0754 0.01 0.33 1.82 43 1987 7 1.0080 0.0127 -0.021 0.0758 0.01 0.33 1.97 44 1987 10 1.0047 0.0127 -0.017 0.0752 0.01 0.32 1.95 45 1988 1 1.0116 0.0120 -0.024 0.0773 0.01 0.36 1.98 46 1988 4 1.0376 -0.0060 -0.032 0.0701 0.01 0.32 1.95 47 1988 7 1.0019 -0.0151 0.013 0.0718 0.01 0.32 1.95 48 1988 10 1.0050 -0.0192 0.014 0.0740 0.01 0.32 1.96 49 1989 1 1.0056 -0.0194 0.014 0.0745 0.02 0.32 1.93 50 1989 4 1.0051 -0.0199 0.015 0.0707 0.02 0.32 1.97 51 1989 7 1.0031 -0.0227 0.020 0.0591 0.01 0.31 1.92 Source: Author's calculations. Table A-2. Composition of Rolling Portfoliosfor Brazil, 1977-86 Amount to be borrowed as a Mean Durbin- Portfolio percentage of square Adjusted Watson number Year Month U.S. dollar DM Yen export earnings residual RZ statistic 1 1977 1 1.0178 0.0672 -0.0850 0.0762 1.31 0.21 2.05 2 1977 4 1.0267 0.0578 -0.0846 0.0744 1.37 0.20 2.11 3 1977 7 1.0074 0.0508 -0.0582 0.0899 1.30 0.23 2.13 41 4 1977 10 1.0079 0.0494 -0.0573 0.0932 1.24 0.26 2.12 5 1978 1 1.0588 0.0008 -0.0597 0.0626 1.04 0.14 1.90 6 1978 4 1.0809 -0.0247 -0.0562 0.0703 0.97 0.21 1.98 7 1978 7 1.0551 -0.0278 -0.0274 0.1306 0.88 0.27 1.97 8 1978 10 1.0480 -0.0191 -0.0289 0.1665 0.84 0.26 2.04 9 1979 1 1.0480 -0.0184 -0.0296 0.1571 0.81 0.25 1.96 10 1979 4 1.0388 -0.0226 -0.0162 0.1518 0.73 0.27 1.99 11 1979 7 1.0347 -0.0287 -0.0060 0.1378 0.74 0.28 2.01 12 1979 10 1.0582 -0.0505 -0.0076 0.1079 0.77 0.26 1.99 13 1980 1 1.0674 -0.0591 -0.0083 0.0877 0.77 0.30 1.86 14 1980 4 1.0925 -0.0854 -0.0072 0.0846 0.64 0.31 1.99 15 1980 7 1.1042 -0.0943 -0.0099 0.0820 0.63 0.32 2.02 16 1980 10 1.1144 -0.1047 -0.0097 0.0811 0.63 0.31 2.02 17 1981 1 1.1259 -0.1161 -0.0098 0.0814 0.60 0.32 2.03 18 1981 4 1.0628 -0.0531 -0.0097 0.0828 0.56 0.33 2.03 19 1981 7 1.0632 -0.0528 -0.0104 0.0787 0.60 0.29 1.99 20 1981 10 1.0693 -0.0626 -0.0067 0.0776 0.65 0.25 1.98 21 1982 1 1.1303 -0.1212 -0.0091 0.0744 1.81 0.12 1.91 22 1982 4 1.0617 -0.0521 -0.0095 0.0685 1.91 0.10 2.04 23 1982 7 1.1157 -0.1059 -0.0098 0.0682 1.91 0.11 2.09 24 1982 10 1.1387 -0.1286 -0.0101 0.0687 1.91 0.11 2.09 25 1983 1 1.1494 -0.1395 -0.0098 0.0687 1.95 0.10 2.07 26 1983 4 1.3053 -0.2942 -0.0111 0.0452 3.14 0.08 1.79 27 1983 7 1.4530 -0.5423 0.0893 0.0246 3.36 0.03 2.00 28 1983 10 2.0162 -1.2338 0.2176 0.0097 3.43 0.03 2.00 29 1984 1 -0.8810 -0.1323 0.0133 -0.1614 3.11 0.09 1.91 30 1984 4 -0.8935 -0.1175 0.0109 -0.1747 3.17 0.09 1.99 31 1984 7 -0.8876 -0.1233 0.0109 -0.1923 3.13 0.09 2.00 32 1984 10 -0.9029 -0.1446 0.0475 -0.1877 3.09 0.10 2.01 33 1985 1 -0.9721 -0.0697 0.0417 -0.1994 3.71 0.01 1.88 34 1985 4 -0.8184 -0.2692 0.0876 -0.1253 5.29 0.05 1.76 35 1985 7 -0.8112 -0.2730 0.0842 -0.1349 5.27 0.05 2.00 36 1985 10 -0.7615 -0.2835 0.0450 -0.1330 5.34 0.05 1.95 37 1986 1 -0.9655 -0.4183 0.3838 -0.1192 4.75 0.05 1.84 38 1986 4 -0.9275 -0.3841 0.3116 -0.1233 4.71 0.05 1.98 39 1986 7 -0.7465 -0.4378 0.1843 -0.1279 4.70 0.05 1.97 Source: Author's calculations. 526 THE WORLD BANK ECONOMIC REVIEW, VOL. 6. NO. 3 ent first-order equations for the foreign bonds in equation A-2' can be written in vector notation as one equation. Thus in equation A-6 VRr represents the vector of covariances of the real domestic costs with the real foreign effective costs, VYP represents the covariance matrix of real domestic returns with the domestic unit value of export receipts, and Vr, represents the covariance matrix of real foreign returns. (A-6) 0 = -U1l + 3E(U;)IE[(l/P)R:-] - bVrp*Q(I) + bVr,B + bVRr1DI. Imposing that the domestic bond market is in zero net supply (D = 0), it can easily be shown that this equation can be solved for the optimal amount of foreign bonds as reported in equation 5 in the text. The Speculative, Minimum Variance, and Price Hedges When variables are lognormally distributed, we can split the speculative hedge and minimum variance hedge further and derive the price hedge. In the case of one foreign bond and one domestic investment opportunity we first combine equations A-2' and A-3', setting D = 0, to derive the following expression: (A-7) E(U')LE(p) - E = bcov (PQ p , p ) - bcov(P Q p ,p). Next we redefine variables in terms of percentage changes instead of second- period values. This implies that E(R/P) can be written as E(r) + E(1/p) + cov (r, 1lp), where lower case symbols are percentage changes. This also implies that the covariances can be rewritten. For instance: (A-8) cov (p, p)* -cov(p -, p) + cov(p, p) + cov(p*, r;) - cov(p, r*). Similar implications can be used for other covariances. Substituting these sim- plified expressions for the covariances in equation A-7, we see that some of the covariances will drop out. This allows us to rewrite equation A-7 and solve for the optimal amount of foreign bonds as a combination of a speculative, price, and minimum variance hedges, as in equation A-9: (A-9) B = [E(r") - E(r)]E(U2) + cov(r P) Q(I) - bVAR(r*) VA R(r:') cov(rato, p)s 1 C Ls E(18) VAR(r- ) I b I' For an alternative derivation, see Claessens (1988). Claessens 527 REFEREN CES The word "processed" describes informally reproduced works that may not be com- monly available through library systems. Adler, Michael, and Bernard Dumas. 1980. "The Exposure of Long-Term Foreign Cur- rency Bonds." Journal of Financial and Quantitative Analysis 15: 971-94. Breeden, Douglas. 1979. "An Intertemporal Asset Pricing Model with Stochastic Con- sumption and Investment Opportunities." Journal of Financial Economics 7: 265-96. Claessens, Stiin. 1988. "The Optimal Currency Composition of External Debt." PRE Working Paper 14, World Bank, International Economics Department, Washington, D.C. Processed. Dornbusch, Rudiger. 1985. "Policy and Performance Links between LDC Debtors and Industrial Nations." Brookings Papers on Economic Activity 2: 303-56. . 1987. "Exchange Rates and Prices." American Economic Review 77: 93-106. Eaton, Jonathan. 1990. "Debt Relief and the International Enforcement of Contracts." Journal of Economic Perspectives 4(1):43-56. Fraga, Arminio. 1986. "The Exchange Rate Premium and the Micro-Foundations of Asset Markets Models of the Exchange Rate." Journal of International Economics 20: 179-85. Gemmill, Gordon. 1985. "Optimal Hedging on Futures Markets for Commodity- Exporting Nations." European Economic Review 27: 243-61. Giovannini, Alberto. 1986. "The Macroeconomics of Exchange Rate Changes and Price-Level Interactions: Empirical Estimates for West Germany." Working paper, Co- lumbia University, Graduate School of Business, New York. Processed. Healy, James. 1981. "A Simple Regression Technique for the Optimal Diversification of Foreign Exchange Reserves." International Monetary Fund, Research Department, Washington, D.C. Processed. iMF (International Monetary Fund). Various years. International Financial Statistics. Washington, D.C. Kahn, R. B. 1988. "Currency Diversification of Developing-Country External Debt with Liquidity Constraints?" Federal Reserve System, Division of Research and Statistics, Washington, D.C. Processed. Kroner, Kenneth, and Stijn Claessens. 1991. "Optimal Dynamic Hedging Portfolios and the Currency Composition of External Debt."Journal of Money and Finance 10: 131- 48. Krugman, Paul. 1981. "Consumption Preferences, Asset Demands, and Distribution Effects in International Financial Markets." NBER working paper 651. National Bu- reau of Economic Research, Cambridge Mass. Processed. Lessard, Donald, andJohn Williamson. 1985. Financial Intermediation beyond the Debt Crisis. Policy Analysis In International Economics 12. Washington, D.C.: Institute for International Economics. Macedo, J. B. de. 1982. "Portfolio Diversification across Currencies." In R. N. Cooper and others, eds., The International Monetary System under Flexible Exchange Rates. Cambridge, Mass.: Ballinger. . 1983. "Optimal Currency Diversification for a Class of Risk-Averse Interna- tional Investors." Journal of Economic Dynamics and Control 5: 173-85. 528 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Merton, R. C. 1971. "Optimum Consumption and Portfolio Rules in a Continuous- Time Model." Journal of Economic Theory 3: 373-413. Oxelheim, Lars, and Clas Wihlborg. 1987. "Exchange-Rate-Related Exposure in a Mac- roeconomic Perspective." In Sarkis Khoury and Alo Ghosh, eds., Recent Develop- ments in International Banking and Finance. Lexington, Mass.: Lexington. Stulz, Rene. 1981. "A Model of International Asset Pricing." Journal of Financial Eco- nomics 9: 383-406. .1984. "Optimal Hedging Policies." Journal of Financial and Quantitative Analy- sis 19: 127-40. Svensson, L. E. 0. 1989. "Portfolio Choice and Asset Pricing with Nontraded Assets." NBER working paper 2774. National Bureau of Economic Research, Cambridge, Mass. Processed. Varangis, Panayotis, and Ron Duncan. 1987. "Exchange Rates Pass-through in Primary Commodity Prices." World Bank, International Economics Department, Washington, D.C. Processed. World Bank. 1987. The World Bank Annual Report. Washington, D.C. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 529-547 Household Saving in Developing Countries: First Cross-Country Evidence Klaus Schmidt-Hebbel, Steven B. Webb, and Giancarlo Corsetti Although most studies have relied on domestic or private sector saving data, this article uses household data available from the U.N. System of National Accounts for a sample of 10 countries. Household saving functions are estimated using combined time-series and cross-country observations in order to test households' responses to income and growth, rates of return, monetary wealth, foreign saving, and demographic variables. The results show that income and wealth variables affect saving strongly and in ways consistent with standard theories. Inflation and the interest rate do not show clear effects on saving, which is also consistent with their theoretical ambiguity. Foreign saving and monetary assets have strong negative effects on household saving, which suggests the importance of liquidity constraints and monetary wealth in developing countries. Households are responsible for a substantial part of the saving in both industrial and developing countries. Accordingly, most economic models treat the motiva- tion for saving from the household's perspective. Because of the shortage of household data, however, most empirical work on saving in developing coun- tries has used only aggregate saving data, so that the results may not be useful for predicting household behavior. This study tests several hypotheses about saving behavior using panel data from the U.N. System of National Accounts. This is the first study to use cross- country variation to estimate household saving behavior in developing countries and has the advantage that the data are defined consistently across countries. The study tests how household saving in developing countries responds to in- come and growth, rates of return, monetary wealth, foreign saving, and demo- graphic variables. Section I reviews the evidence in the empirical literature on the main determi- nants of saving in developing countries. Section II presents an empirical frame- Klaus Schmidt-Hebbel and Steven B. Webb are with the Country Economics Department at the World Bank; Giancarlo Corsetti was with the department when this article was written. They are indebted to Ricardo Caballero, Vittorio Corbo, Stanley Fischer, Mark Gersovitz, and two anonymous referees for helpful comments and suggestions and thank Heidi Zia for assistance with some of the research for this article. © 1992 The International Bank for Reconstruction and Development/THE WORLD BANK 529 530 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 work for analyzing household saving in developing countries, discusses the data set and estimation methods, and presents results. Section III concludes with a discussion of policy implications. 1. DETERMINANTS OF SAVING IN THE EMPIRICAL LITERATURE A number of studies have estimated saving behavior in developing countries. (For general surveys, see Mikesell and Zinser 1973, Gersovitz 1988, and Deaton 1989.) Typically, studies use cross-section, time-series data on national saving rates; see for instance Collins (1989); Fry (1978, 1980, 1988); Giovannini (1983, 1985); and Gupta (1987). The advantage of using national saving rates is that more years of data are available for more countries. Even when private or household data are available, however, as in Collins (1989), they may have limited use because of lack of comparability across countries. The use of na- tional saving is rationalized with the argument that private saving is a large and typically predominant part of total saving. To extrapolate from total saving behavior to household saving behavior, how- ever, requires the assumption that the latter is a perfect substitute for both private corporate saving and public sector saving. Assuming substitutability between private and public saving requires that Ricardian equivalence holds. If income and saving data for just the household sector are used, then household saving decisions are not necessarily assumed to offset saving decisions made elsewhere in the economy, unless households respond to determinants such as disposable income, public saving, or private wealth in a way consistent with the Ricardian hypothesis. A few studies have used private sector saving or household consumption data, but none of the studies with aggregate data from developing countries has focused on household saving. Because data of this type are only gradually be- coming available, the data sets vary widely from study to study. Rossi (1988) uses a cross-section, time-series data set for 49 countries covering 10 years. Saving is implicit because per capita private consumption is estimated as a function of per capita private income, among other things. Private income in- cludes the profits of public as well as private enterprises, so that the implicit saving includes saving by public sector enterprises. Lahiri (1988) uses time-series data for private consumption to run separate regressions for 8 Asian countries with about 20 years of data for each. Several older studies, such as Singh (1975) and Williamson (1968), also used private saving data, but they investigated Keynesian theories of saving and overlooked many of the issues currently of interest. Despite differences in the data, most studies using private sector data produce consistent results on most of the major issues. The main determinants of private saving or consumption considered in the literature fall into five groups: income and wealth, public saving, rates of return, foreign saving, and demographic variables. As discussed below, there is a broad consensus that faster economic growth Schmidt-Hebbel, Webb, and Corsetti 531 and high incomes contribute to higher saving rates. Evidence on the pervasive- ness of liquidity constraints suggests that an increase in public saving is not neutralized by an offsetting decline in private saving. There is still controversy and lack of sufficient evidence about the effects of monetary asset holdings, foreign capital inflows, the real interest rate, and inflation, as well as demo- graphic variables. Income and Wealth Income and wealth are the basic determinants of consumption. Most studies use per capita income levels or growth rates and total or financial wealth as the relevant income and wealth variables. The level of per capita income is hypothesized to have a positive effect on the saving rate because richer people can afford the luxury of saving to assure their future consumption. The poor are more likely to be at their biological or social minimum level of current consumption. This does not mean zero saving by the poor in all years, for they will attempt to cushion themselves against fluctuations in current income. But they will have relatively smaller cushions and will more frequently find themselves with zero wealth and no opportunity to borrow in order to sustain consumption when income is low (Deaton 1989; Zeldes 1989). All of the empirical studies mentioned earlier find a strong positive effect- measured in various ways-of the current income level on the saving rate. Saving may also depend on income fluctuations. Simple consumption- smoothing models predict that temporary fluctuations in income should primar- ily affect saving. If households are not credit constrained and the temporary fluctuation does not change much permanent income, consumption would re- spond only marginally to temporary income fluctuations. Households, particu- larly in developing countries, tend to be credit constrained, however, which would imply that consumption would respond significantly to temporary in- come changes. Campbell and Deaton (1989) also argue that, at least in indus- trial countries, a household's perception of its permanent income is strongly affected by current shocks, with no evident distinction made between current and permanent income flows. This implies that households will consume out of current shocks, saving much less out of shocks than predicted by consumption- smoothing models. Most empirical studies of developing countries have not looked closely at the effect of income fluctuations on saving. An exception is Gupta (1987), which consistently finds that saving responds significantly and positively to temporary income shocks. The growth rate of income is also typically included in recent saving studies for developing countries, but its effect cannot be signed a priori. Intertemporal models of consumption or saving (that is, permanent income or life-cycle models) predict that an increase in the growth rate of an individual household's income would lower its saving rate; households would save less now if they expected higher incomes in the future, which would allow both higher consump- tion and higher saving in the future. Faster growth of average per capita (or 532 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 average household) income could also have positive effects on saving. More rapid income growth would raise the average household saving rate if the growth were concentrated in households with high saving rates, such as rich or middle-age households. Collins (1989) develops these concepts in a simple model. A positive coefficient for the growth rate could also reflect the slowness with which people change their consumption habits or indicate that people have regressive expectations about their level of income. The studies that investigate real growth of gross domestic product (GDP) as a determinant of saving-such as Collins (1989); Fry (1978, 1980); Giovannini (1983, 1985); Mason (1987, 1988); and Mason and others (1986)-find positive and usually significant effects on the saving rate.' Wealth is usually a key determinant of consumption or saving in theoretical models of intertemporal optimization. Of course, permanent income can be viewed as the stream of income from total wealth, but a narrower definition of wealth would be the assets that can be exchanged for current consumption. Theory unambiguously predicts that greater wealth would reduce saving out of current income. Since most concepts of wealth are not directly observable, wealth has not been used in most empirical studies of saving in developing countries. Schmidt-Hebbel (1987) uses five alternative measures of total wealth in an empirical intertemporal consumption model for Chile, based on different assumptions regarding expectation formation. Behrman and Sussangkarn (1989) use household-level data on wealth and saving in Thailand. In both studies, wealth has a strong negative effect on saving. Monetary or financial assets also lessen a household's dependence on current income sources when income declines temporarily, because consumers can draw on the assets to maintain their consumption levels. Hence, holding a higher stock of assets allows a household to maintain a higher consumption rate on average, thus depressing the saving rate. In addition, monetary asset holdings are an important component of total consumer wealth. This implies that mone- tary holdings have a second, negative influence on saving rates; previous studies have not tested for this. Government Deficits and Government Saving Fiscal stabilization programs aim to lower public sector deficits or raise public saving, which is equal to public investment minus the deficit. Changes in govern- ment deficits or government saving may change households' perceived perma- nent income and therefore affect private saving rates. A voluminous empirical literature has been devoted to the Ricardian equivalence proposition as reformu- 1. There might be some simultaneity bias-both high saving and high growth reflecting the effect of good investment opportunities, which are not fully reflected in the real deposit interest rates. Also, with the public sector included in aggregate saving, one would expect revenues to adjust more automatically to income increases than current expenditures. Furthermore, government investment (which counts as saving) might be driving both faster growth and higher aggregate saving. Schmidt-Hebbel, Webb, and Corsetti 533 lated by Barro (1974), which states that public debt issues are macro- economically indistinguishable from tax increases, and hence a change in public saving should be offset by an equal and opposite change in private saving. For Ricardian equivalence to hold, the following conditions must be satisfied jointly: no liquidity constraints; equal discount rates for the public and private sectors; and certainty of future income, tax, and public expenditure flows. For a further discussion of these conditions and surveys of empirical studies, see Hay- ashi (1985), Hubbard and Judd (1986), Bernheim (1987), and Leiderman and Blejer (1988). Ricardian equivalence has been widely rejected in empirical studies of industrial countries. The pervasiveness of borrowing constraints is most often cited as the main reason for its rejection. For a sample of developing countries, Haque and Montiel (1989) also find that borrowing constraints are the main cause of the deviations from Ricardian equivalence. They also test for differences between public and private discount rates but do not find that such differences explain the deviations from Ricardian equivalence. Credit con- straints, proxied by current income levels or financial asset holdings, are also major determinants of private saving in the cross-developing-country study by Rossi (1988). Corbo and Schmidt-Hebbel (1991) explicitly test the Ricardian hypotheses by regressing private consumption on permanent government saving in 13 developing countries. They conclude that the Ricardian hypothesis does not explain consumption behavior. Because it implies that private and public saving are not perfect substitutes, empirical rejection of Ricardian equivalence is a powerful argument in favor of using household saving data instead of aggre- gate saving data. Rates of Return Because the income and substitution effects of higher interest rates work in opposite directions, the effect of rates of return on saving cannot be predicted. In addition to the well-known income and substitution effects, rates of return also affect saving through a wealth effect. A higher real interest rate reduces the present value of future income streams from human capital or fixed-interest financial assets. Consumption is therefore depressed even if the substitution and income effects cancel each other. Economists have heatedly debated the inter- pretation of the empirical evidence. Based on noneconometric country studies, McKinnon (1973) and Shaw (1973) argued that the rate of return on saving, as measured by the real interest rate, would have a positive effect on saving rates. This view was also held by Balassa (1990) in his survey on the issue. Fry (1978, 1980) found statistical evidence to support this contention, although he con- ceded that the magnitude of the effect is small (Fry 1988). Thus, only large changes in real interest rates would be economically important. Giovannini (1983, 1985) revisited Fry's earlier work and found that two observations (on the Republic of Korea in 1967 and 1968) accounted for the entire result. With an expanded data set, Giovannini found that the interest rate did not signifi- 534 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 cantly affect saving. Both Fry and Giovannini used aggregate data, which is especially problematic for testing the effects of interest rates. Giovannini (1985) implicitly recognizes this; he points out that much of the increase of Korean saving in the late 1960s resulted from an increase in government saving. Gupta (1987) finds some support for a positive effect of interest rates on saving in Asia, but not in Latin America. Schmidt-Hebbel (1987) and Arrau (1989) estimate intertemporal consumption substitution elasticities for Southern Cone countries and find values close to 1.0. With an elasticity of 1.0 and for given wealth stocks, consumption is insensitive to the interest rate because the substitution effect is canceled by the income effect. Even if higher real interest rates are unlikely to raise private saving and hence total private wealth, they can substantially alter the portfolio composition of private wealth. Negative real interest rates on deposits (resulting from high inflation), for example, cause substitution into real assets, especially consumer durables, and into foreign currency assets through capital flight. Both higher purchases of durables and capital flight tend to reduce private saving as mea- sured by official national accounts, without affecting private saving defined as total private wealth accumulation. Inflation may affect saving independently from its effect through the real interest rate. High inflation often contributes to stagnation in output or outright recession, which will be reflected in the income variable. Higher inflation also increases economic instability and uncertainty about future rates of return on real assets and income levels. Theoretically, the effect of inflation on private saving is ambiguous, as uncertainty about future asset values could either dis- courage saving because of the substitution effect of the lower real rate of return or encourage saving for precautionary motives. Similarly, the increase in uncer- tainty of future income streams affects private saving ambiguously, depending on the form of the underlying utility function. A comprehensive treatment of the effects of different sources of uncertainty on saving can be found in Gersovitz (1988). Gupta (1987) and Lahiri (1988) include the expected and unexpected compo- nents of the inflation rate as separate determinants of saving, and Gupta also includes the nominal interest rate. Gupta's results differ sharply by region. In Asia both expected and unexpected inflation have positive and significant coeffi- cients. In Latin America neither coefficient was significant with the preferred estimation technique. In Lahiri's sample, comprised only of Asian countries, the signs on both inflation variables are mixed for the eight separate country regres- sions. Therefore, there is no clear empirical evidence that inflation generally affects saving in developing countries. Foreign Saving If access to foreign borrowing at international interest rates is unlimited, foreign saving passively fills the gap between domestic investment and national saving. In this case, foreign saving (or borrowing from abroad) is simply the Schmidt-Hebbel, Webb, and Corsetti 535 result of national saving decisions and not one of its determinants. If foreign borrowing is rationed, however, either by the lenders or by government regula- tion in developing countries, then domestic savers (and investors) are con- strained in their intertemporal choices by the amount of available foreign fi- nance, and foreign saving becomes a determinant of domestic saving. During most of the post-World War II period, developing countries have not had unrestricted access to private commercial lending, a partial exception being the period from 1976 to 1981, before the debt crisis erupted. Even during that short time, governments in developing countries restricted access to foreign borrowing. When restrictions were partially lifted, governments and corpora- tions typically made use of their freer access to foreign borrowing to increase investment, but households were the last and least likely sector to have unlimited access to sources of foreign loans. Hence, even during 1976-81, households were effectively constrained by foreign lending: for given income levels, a frac- tion of total foreign lending finances higher private consumption, hence reduc- ing household saving. A number of empirical studies have included foreign saving as a determinant of domestic saving rates. Fry (1978, 1980) and Giovannini (1985) find the effect of foreign saving is significant and negative, although its size is small. With a noneconometric analysis, Chenery and Strout (1966) also find a negative initial impact of capital inflows on domestic saving, although the secondary effects on capacity growth tend to increase saving. Giovannini (1983) finds coefficients on foreign saving to have mixed signs and to be insignificant. Gupta (1987) finds positive coefficients on foreign saving, which are significant for Latin America but not for Asia. The results of these studies seem to depend on the sample and model specification. All of the studies with foreign saving as an explanatory variable used total saving as the dependent variable, so the results may reflect the extent to which capital inflows went straight to public and corporate invest-. ment, which appears as an increase in saving. Demographic Variables The life-cycle models of saving imply that demographic variables should affect saving rates, and demographic influences on saving have been widely researched (Collins 1989; Leff 1969; Mason 1987, 1988; Mason and others 1986; Rossi 1989; Webb and Zia 1990; see Hammer 1986 for a survey). The dependency ratio-those under age 15 or over 65 as a share of total population-is the most commonly used explanatory variable. In the life-cycle model, older people work less and consume out of their savings. Households with more children at home may also save less because saving for retirement would be deferred until the children left home, which would raise the per capita income of the household, or because parents would expect old-age support from their children. Thus, one would expect saving rates to depend negatively on the dependency ratio. Early work on the topic, especially that of Leff (1969), found that the depen- dency ratio had a strong negative effect on saving. Subsequent studies challenged 536 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 the robustness of this result and have examined both the theory and the mea- surement of demographic variables more carefully (Mason and others 1986; Mason 1987, 1988). The results seem to depend on the data used and on the other explanatory variables included. The models of Mason and others (1986) and Collins (1989), applied to cross-country samples of Asian economies, add to the dependency ratio and the growth rate of per capita income (among other variables) an interaction term between the two. Although Mason and others find that higher population dependency lowers saving unambiguously, Collins's re- sults show an ambiguous effect of the dependency ratio, with a negative (posi- tive) influence on saving in countries with high (low) growth rates. II. NEW EVIDENCE ON HOUSEHOLD SAVING To test for the determinants of saving with household data, a behavioral function for household saving was estimated. The function incorporates vari- ables to address the major unresolved issues in the literature. The saving rate (the ratio of household saving to private disposable income), rather than the absolute level of saving, is used as the dependent variable for three reasons. First, there is no adequate deflator for saving that can be used to obtain constant-price saving series. Second, by using ratios instead of levels, cross- country comparisons can be made without having to choose appropriate ex- change rates. Third, saving rates tend to be stationary, whereas absolute saving flows grow over time, so that, by using rates, spurious correlation with time- trended explanation variables can be minimized. The specification for the household saving rate is consistent with the con- sumption hypotheses reviewed in the previous section but is not derived from an explicit optimizing framework. The reason for the latter is the difficulty of deriving a closed-form saving equation with the host of economic and structural determinants considered in this study. Hence the following general saving func- tion is implemented in a linear form: S sr HT MQM FS1 (1) I I [LITP, GITP, (LIP - LITP), I, R, INF, , , DEP, URB] where S is household saving, I is household disposable income, LITP is the natural logarithm of trend per capita household disposable income, GITP is the growth rate of trend per capita household disposable income, LIP is the natural logarithm of per capita household disposable income, HT is transfers to house- holds, R is the real interest rate, INF is the inflation rate, MQm is money plus quasi-money at the end of the previous period, I* is the average of household disposable income in the current and previous years, FS is foreign saving, DEP is the dependency ratio, and URB is the urbanization rate. Signs below the vari- ables indicate the expected coefficient signs, given the discussion of the preced- ing section. Equation 1 generalizes the saving functions estimated for developing countries in the recent empirical literature by including explanatory variables Scbmidt-Hebbel, Webb, and Corsetti 537 from each of the five main groups: four income determinants, two rates of return, broad money, foreign saving, and two demographic variables. The first three income variables are the natural logarithm of trend per capita private disposable income, its growth rate, and the deviation of log current from log trend per capita private disposable income. Unlike previous studies, both the trend and the deviation of current from trend per capita income are introduced as saving determinants in order to discriminate between the permanent and cyclical influences of income on the saving rate. Because the sum of log trend income and the deviation of log current from log trend income equals log current income, we can test the proposition, implicitly assumed in some earlier studies, that the coefficients on the two components of income are the same. Growth of per capita trend income enters as a separate, third determinant. Finally, in order to test whether saving out of transfers differs from saving out of other income, transfers to households is included. Since transfers are already counted as part of total household income, the coefficient on transfers would be zero if consump- tion out of that income were the same as that out of other income. A significant negative coefficient would indicate that households saved less out of transfers than out of other income. Domestic real interest and inflation rates, both with ambiguous signs, could affect intertemporal consumption and portfolio composition decisions, with consequences for household saving. Monetary wealth (which reflects both con- sumer wealth and liquidity constraints) and foreign saving should depress house- hold saving rates. Finally, the two demographic variables have ambiguous signs. Although the dependency ratio should reduce saving due to life-cycle patterns, its net influence on saving seems to depend on its measurement and the inclusion of other variables. The urbanization variable was added to control for the effect of differences in the measurement of urban and rural saving, as well as for structural differences in the underlying saving behavior of urban and rural households. In estimating equation 1, simultaneity biases could arise from the potential endogeneity of some explanatory variables. For this study the potentially most important simultaneity is between household saving and the real interest rate, which will be addressed by instrumenting the latter variable. The responsiveness of foreign saving to household saving appears to be minimal in the sample of developing countries used, for the reasons discussed in section I. Also, house- hold saving does not appear to have important contemporaneous effects on household income, because even the effects on GDP, from which household income derives, are dominated by other influences. Hence neither foreign saving nor household income will be instrumented. Data The data set is especially well suited for investigating household saving behav- ior. It is based on household saving and disposable income series for 10 econ- omies for which at least 7 and as many as 13 consecutive annual observations were available during 1970-85. The economies are Botswana, Colombia, Ecu- 538 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 ador, Honduras, Republic of Korea, Philippines, Paraguay, Thailand, South Africa, and Taiwan. Except for Taiwan, the data come from the U.N. System of National Accounts, which breaks down income and consumption into general government, corporate, and household sectors. From these, household dispos- able income, household saving, and transfers from the general government to households were calculated. The data for Taiwan come from the Statistical Yearbook of the Republic of China, compiled by the Taiwanese government using the U.N. method. All household data (saving, disposable income, and transfers) are aggregate data obtained from national accounts sources, not mi- croeconomic data from household surveys. Interest rates come from a data set developed by the Country Economics Department of the World Bank. The remaining data (inflation, urban population share, dependency ratio, current account balance, and money balances) are from the International Monetary Fund's International Financial Statistics and Government Financial Statistics and the World Bank Economic and Social Database. Household disposable income includes all current receipts by households, less taxes and social security contributions. One potential problem is that excluding the retained earnings of (private) corporations owned by households may distort the picture of household decisionmaking. The household sector does include all agricultural firms and firms in the informal nonagriculture sector. Excluding the income of private corporations will not affect the results as long as most of the variation of household income and saving is accounted for by households that would not consider the corporate income and saving as part of their own budget and would not make household saving decisions to offset corporate sector deci- sions. Another potential problem is that in the System of National Accounts net saving is defined as the difference between current receipts and current outlays and is therefore a balancing item. Hence measurement errors in receipts and current outlays are reflected in measured saving. As long as these errors are systematic and stable over time for each country, they are completely accounted for by the use of country-specific intercepts in the empirical model introduced below. To calculate the three income variables, the log of household disposable in- come was regressed on a time trend using five-year overlapping data series, up to and including the current year. The estimated value for the current year gives the trend value of current income, the coefficient for time is the trend rate of growth, and the deviation from the estimated value in the current year is the temporary fluctuation in income. This simple deterministic time trend method was used instead of more sophisticated time-series procedures because of lack of data. Although one should view the results with caution, this method avoids some of the pitfalls of using simple moving averages. To obtain results for the first sample observations for each country, household income for the four years before the start of the sample was estimated using the growth rate of GDP in each year. Transfer income includes social security, unemployment insurance, and transfers from abroad; it does not include indirect transfers such as farm price supports or subsidies that permit food to be sold at below-market prices. Schmidt-Hebbel, Webb, and Corsetti 539 The nominal interest rate is the annual rate on three-month time deposits. The inflation rate is the annual (December-to-December) change in the log of the household consumption deflator. The sample includes countries with low to medium inflation, with annual average inflation rates ranging from 7 percent in Honduras to 23 percent in Colombia. Household income, which includes net nominal interest income, is therefore only slightly distorted by including the inflationary component of interest income. Data to correct for this bias are not available for the sample of countries used. The real interest rate is calculated as: R = [(1 + NOI)/(1 + INF)] - 1, where R and INF were introduced in equation 1 above and NOI is the nominal interest rate. Money and quasi-money holdings at the end of the previous period were used as a measure of liquid wealth available for consumption in the current period. To obtain its value as a share of income, the sum of money and quasi-money holdings was divided by the geometric average of nominal disposable income in the current and previous years, which is an adequate measure for low- to medium-inflation countries. Foreign saving is the current account balance, again as a share of household disposable income. The two demographic variables, the dependency and urbanization rates, change little for each country during the observation period and in many cases are not known on an annual basis. Because both variables vary widely across countries but little over time, collinearity between country dummies and these variables is unavoidable. The dependency ratio is the population below 15 years and above 65 years as a percentage of total population. The urbanization rate is the share of population in cities, originally put together by the United Nations; the population cutoff for an area to be designated as a city varies from country to country. Estimation Methods The base specification is a fixed effect model of the form: N K (2) Yit= i° + EY YWXt + E AkXk2t + Eit i=2 k=1 where the subscript i refers to countries, t refers to time, and k refers to indepen- dent variables. Yh. is the vector of saving rates, Wit is the vector of country dummy variables (such that Wit is equal to 1 for the ith country, and 0 other- wise), Xkit is the matrix of independent variables, and Ei, is the vector of errors, which are assumed to satisfy the assumptions of the classical normal linear model. This basic model incorporates fixed country-specific effects in the intercept term. In fixed-effects models, the empirical results are conditional on the par- ticular sample used in the estimation. Alternative estimators, such as the error- components (or random-effect) model, treat the available observations as a random sample from a population. Given the small number of countries in- cluded here and the marked differences in their economic features, the fixed- effect estimator seems to be an appropriate choice. Table 1. Determinants of Household Saving, with Household Sector Saving as a Percentage of Household Disposable Income as Dependent Variable Independent variable Beginning of period Income Transfers money and growth Deviation to house- quasi- Foreign Trend rate of income holds Real money saving Urban Estimation income (five-year from trend (ratio to interest Inflation (ratio to (ratio to Depen- population Adjusted technique (log) average) (log) income) rate rate income) income) dency ratio ratio R2 1. OLS' 0.03 0.75 0.12 -0.32 -0.10 -0.06 -0.03 -0.23 -0.48 -0.001 0.646 (1.4) (6.2) (0.8) (-1.7) (-(.8) (-0.6) (-1.0) (-4.3) (-3.2) (-2.1) 2. Fixedeffect 0.26 0.54 0.30 -0.40 -0.08 -0.14 -0.19 -0.14 0.83 0.0001 0.811 (5.4) (4.2) (2.5) (-2.1) (-0.7) (-1.5) (-3.1) (-2.8) (2.5) (0.1) 3. Random 0.04 0.70 0.19 -0.38 -0.13 -0.13 -0.06 -0.18 -0.54 -0.001 0.703 4 effectb (1.6) (5.8) (1.3) (-1.9) (-1.1) (-1.2) (-1.7) (-3.4) (-3.1) (-1.6) 4. Fixed 0.25 0.51 0.27 -0.42 -0.19 -0.23 -0.18 -0.12 0.75 0.0004 0.809 effectwith (4.6) (3.5) (2.0) (-2.1) (-0.7) (-1.0) (-3.0) (-2.2) (2.0) (0.2) instrumen- tal variables 5. Fixed effect 0.22 0.56 0.26 -0.19 -0.15 n.a. -0.16 -0.14 0.56 -0.0004 0.765 with instru- (3.1) (3.9) (1.8) (-0.7) (-0.5) n.a. (-2.3) (-2.7) (1.0) (-0.2) mental vari- ablesc 6. Fixedeffectd 0.26 0.56 0.31 -0.38 n.a. -0.08 -0.18 -0.14 0.87 -0.0001 0.812 (5.7) (4.4) (2.8) (-2.0) n.a. (-1.9) (-3.2) (-3.1) (2.7) (-0.04) n.a. Not applicable. Note: t-statistics are in parentheses. a. Breusch-Pagan test for the null hypothesis of the absence of individual effects in the errors: x,2 = 15.99 (P-value = 0.00006). b. Hausman specification test for the null hypothesis of random-effect estimators equal to fixed-effect estimators: xi = 28.4 (P-value = 0.002). c. Inflation rate is omitted variable. d. Real interest rate is omitted variable. Source: Authors' calculations. Schmidt-Hebbel, Webb, and Corsetti 541 This choice has also been verified by the usual set of specification tests. (For a detailed description of these tests, see Kmenta 1986 and Hsiao 1986.) The test results are reported in table 1. The first of these is the Breusch-Pagan test for the presence of both individual (cross-section) and time effects in the residuals of a simple ordinary least squares (OLS) regression on pooled data. The OLS specifica- tion was rejected, which implies that either the fixed effects model or the ran- dom effects model is superior. The second test applied is the Hausman specifica- tion test, which compares fixed-effect and random-effect estimators. Random- effect estimators are efficient but are consistent only in the absence of correlation between the included regressors and the errors. Fixed-effect estimators are not efficient but are still consistent even when the regressors and errors are corre- lated. The random effects estimates were found to be inconsistent, implying that a fixed-effects specification is superior. Finally, regression t-tests for the presence of trend- and time-specific effects showed such effects were never significant in any of the models specified here. An important caveat to the results below is that the number of cross-sectional units considered in the estimation is small, which is problematic because the properties of panel data estimators are based on asymptotic results for a large number of cross-sectional units. This problem could not be avoided because the construction of a data set useful for testing household saving behavior limits the sample to only those countries for which disaggregated data on saving are available over a sufficiently long period. Its implication is that the empirical results reported below could change in the future when more household data at the country level become available. Results Table 1 summarizes the regression results with different estimation methods and with different sets of regressors. Four estimation techniques were used: OLS, country-specific fixed effects, country-specific random effects, and fixed effects instrumental variables to instrument the real interest rate. It was expected that the country-specific, fixed-effects method-with or without instrumental variables-would be best, and the results in table 1 confirm this. Instrumental variable estimation corrects for the possible simultaneity bias stemming from the interaction between saving and domestic real interest rates. LIBOR, corrected for world inflation, was used as the instrument for the domes- tic real interest rate. Other sets of instrumental variables, such as the lagged independent variables, were also tried but did not improve upon the reported results. Regression 4 in table 1-fixed-effects estimation with instrumentaliza- tion of the real interest rate-thus constitutes the primary set of results. The estimated regressions fit the data well, as can be seen in the adjusted R2 coefficients. Starting at 0.65 for the OLS estimation, the adjusted R2 increases to 0.81 under fixed-effects estimation. This overall fit compares favorably with other panel data estimations of saving functions in developing countries, such as Corbo and Schmidt-Hebbel (1991), which explain roughly half of the variation of the dependent variable. 542 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 The income variables all have a positive effect on saving rates, which accords with most of the previous studies for developing countries. The growth rate of trend per capita disposable income has a strong effect on the household saving rate in the sample countries: a 1 percentage point increase in per capita income growth raises the household saving rate by about 0.5 percentage points. An increase in the growth of per capita income seems the best way to raise the private saving rate, which suggests the presence of a virtuous circle between growth and saving. The deviation of income from its trend level has a positive influence on the saving rate. Its coefficient is in the neighborhood of 0.30, however, which is much lower than that predicted by the permanent income theory. Given the saving function in equation 1, it is easy to show that this coefficient is equal to the difference between the average and marginal propensities to consume out of current (transitory) income. Only if the estimated value of this coefficient were close to 0.75, which is the average propensity to consume out of current income in the sample, would the marginal propensity be close to zero, as predicted by the permanent income hypothesis. This effect of transitory income on consump- tion reflects the importance of borrowing constraints faced by households or their use of current income in estimating permanent income levels. The dominance of transitory over trend (permanent) income in determining household consumption is confirmed by the coefficient of the log of the level of per capita trend disposable income. Although it is highly significant, one cannot reject the hypothesis that it is identical to the estimated coefficient of the devia- tion of current from trend income. According to the permanent income hypoth- esis, the marginal propensity to save out of permanent income should be close to -1. For the saving equation specified here, this would require that the coeffi- cient of trend income should be equal to the difference between the coefficient of the deviation of current from trend income and 1-a far cry from the actual results. This implies that, for the present sample, trend income could be omitted as a determinant of household saving, using instead just current income rather than decomposing income into its trend and fluctuation. With regard to the last income variable, the coefficient on transfers is negative and usually large and significant, thus indicating that households consume more out of transfers on average than out of other sources of disposable income. The domestic real interest rate has a small, mostly negative, and not signifi- cant influence on household saving rates. This result, confirming many recent studies that show a negligible role of interest rates in determining consumption or saving, reflects either that income and substitution effects offset each other or that liquidity constraints weaken the effects of intertemporal relative prices on intertemporal consumption choices. The important role of liquidity constraints in our sample is corroborated by the significance of current income, monetary wealth, and foreign saving in affecting household consumption and saving. Inflation has a negative but statistically not significant effect on saving. Be- cause of collinearity between the real interest rate and the inflation rate, these Schmidt-Hebbel, Webb, and Corsetti 543 variables were entered alternately, as reported in regressions 5 and 6. The infla- tion rate is significant at about the 10 percent level when the real interest rate is excluded, and its coefficient is negative. Omitting inflation, however, does not make the effect of the real interest rate significant. In other words, reducing inflation seems to encourage household saving, but raising the deposit rate relative to inflation has no discernible effect on saving. Monetary assets play a dual role here: first, they constitute a stock variable signaling the extent of domestic liquidity constraints; second, they are related to household financial wealth. For both reasons monetary stocks were expected to have a negative influence on saving. The results show a negative and significant coefficient for the ratio of money to income in the fixed-effect estimations. Foreign saving, which acts as an external liquidity constraint, boosts private consumption, as shown by its significantly negative influence on saving. Because of the lack of variation in the demographic variables over time and because the fixed-effect estimation technique does consider cross-country varia- tion only in its country dummies, one should not attach much significance to the coefficients on the demographic variables. The urbanization rate has no discern- ible effect on saving under fixed-effect estimation, probably because of the collinearity between this variable and the country dummies. With OLS and random-effect estimation, the urbanization rate exerts a small negative and barely significant effect on household saving, which indicates a weak difference between urban and rural behavior that is consistent with higher consumption opportunities in cities. The dependency ratio has widely varying effects depend- ing on the estimation technique. An interaction term between income growth and the dependency ratio was included, in the spirit of Mason and others (1986) and Collins (1989), but was found to be insignificant. III. CONCLUSIONS AND POLICY IMPLICATIONS The surprisingly strong results of this study, considering the small number of countries sampled, verify the value of using household data. These results should, of course, be reevaluated with a larger sample when more data become available. The empirical findings of this study confirm the central role of income and wealth in determining household saving in developing countries. House- holds save a larger share of their income when that income is higher and when it is growing faster. They save less the greater is their monetary wealth (compared with other forms of wealth). Borrowing constraints are also major determinants of household saving: in addition to domestic liquidity constraints (affected by current income and monetary wealth), consumers seem to face foreign liquidity constraints on the use of foreign saving. Real interest rates do not encourage saving in the countries investigated here. This may be because the substitution and income effects of higher real interest rates offset each other or because intertemporal consumption decisions are limited by liquidity constraints. As is often expected, the effect of inflation on saving is negative. This effect is at best 544 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 marginally statistically significant, however, which may result from the lack of high-inflation countries in the sample. Table 2 summarizes the empirical results by calculating the changes in signifi- cant variables required to raise the household saving rate by one percentage point, based on the fixed-effect instrumental variables regression (equation 4 in table 1). Raising the household saving rate 1 percentage point, for example, would require an increase of 2.0 percentage points in the trend growth rate of per capita disposable income. Because the business cycle has a relatively small influence on the saving rate, a high (3.7 percentage point) increase in the ratio of current to trend disposable income is required to achieve a 1 percentage point rise in the saving rate. Public policies can affect national saving either directly, by aiming fiscal poli- cies at changing public sector saving, or indirectly, by inducing changes in vari- ables that affect private saving. Although other studies have focused on the effectiveness of the first types of policies (see Corbo and Schmidt-Hebbel 1991), the results of this study may be used to derive implications for the effectiveness of indirect policies in raising household saving. Successful stabilization policies reduce public sector deficits and hence domes- tic financing requirements of deficits, which lowers domestic real interest rates and inflation. Although the change in real interest rates itself has no significant effect on household saving, price stabilization could raise saving, possibly due to the reversion of past flight into consumer durables when inflation was high. A cyclical downturn caused by the stabilization measures, however, reduces house- hold saving as long as the temporary income decline lasts. Financial liberalization lifts interest rate controls, relaxes credit constraints, and deregulates domestic financial intermediation. Frequently this results in increased domestic real interest rates and increased monetary or financial wealth. Although the former has no discernible effect on household saving, financial deepening could depress saving as monetary wealth is raised. As shown in table 2, a 5.6 percentage point rise in the ratio of money to income would depress the saving rate by 1 percentage point. If the buildup of monetary wealth comes mostly from a portfolio substitution away from consumer durables, how- ever, household saving could rise. It may be concluded that financial liberaliza- tion has no clear effect on household saving. Table 2. Changes in Saving Determinants Required to Raise Saving Rate by 1 Percentage Point Percentage change in the level of trend per capita disposable income 4.0 Percentage point change in the trend growth rate of per capita disposable income 2.0 Percentage point change in the ratio of current per capita disposable income to the trend level 3.7 Percentage point change in the ratio of household transfers to private disposable income -2.4 Percentage point change in the ratio of monetary assets to private disposable income -5.6 Percentage point change in the ratio of foreign saving to private disposable income -8.3 Source: Authors' calculations. Schmidt-Hebbel, Webb, and Corsetti 545 Foreign capital inflows tend to reduce household saving in the short run; a fraction of each additional dollar of foreign borrowing finances private con- sumption. A rise in the ratio of foreign saving to income of 8.3 percentage points will result in a fall in household saving of 1 percentage point. The long-run effect of increased foreign lending on household saving is very likely positive, however, once the income growth from the investment financed by foreign lending is realized. Successful structural adjustment and growth policies are the most effective ways to raise household saving. Reforming the incentive structure in developing countries to spur growth will feed back into higher household and, hence, national saving, thus allowing for even more economic growth. Structural re- forms that move an economy beyond its initial conditions of low resource mobil- ization and income growth give rise to a virtuous cycle of mutually reinforcing saving and growth efforts. REFERENCES The word "processed" describes informally reproduced works that may not be com- monly available through libraries. Arrau, Patricio. 1989. "Intertemporal Monetary Economics: Evidence from the South- ern Cone of Latin America." Ph.D. dissertation. University of Pennsylvania, Depart- ment of Economics, Philadelphia. Processed. Balassa, Bela. 1990. "The Effects of Interest Rates on Saving in Developing Countries." Banca Nazionale del Lavoro Quarterly Review 172 (March): 101-18. Barro, Robert J. 1974. "Are Government Bonds Net Wealth?" Journal of Political Econ- omy 82 (6): 1095-117. Behrman, Jere, and Chalongphob Sussangkarn. 1989. "Do the Wealthy Save Less?" Department of Economics, University of Pennsylvania. Processed. Bernheim, B. Douglas. 1987. "Ricardian Equivalence: An Evaluation of Theory and Evidence." NBER Macroeconomics Annual 1987. Cambridge, Mass.: MIT Press. Campbell, John, and Angus Deaton. 1989. "Why Is Consumption So Smooth?" Review of Economic Studies 56 (3): 357-74. Chenery, Hollis B., and Alan M. Strout. 1966. "Foreign Assistance and Economic Devel- opment.' The American Economic Review 56 (4, September): 679-733. Collins, Susan M. 1989. "Savings Behavior in Ten Developing Countries." NBER Confer- ence on Savings, Maui. National Bureau of Economic Research, Cambridge, Mass. Processed. Corbo, Vittorio, and Klaus Schmidt-Hebbel. 1991. "Public Policies and Saving in Devel- oping Countries.' Journal of Development Economics 36 (1): 89-115. Deaton, Angus. 1989. "Saving in Developing Countries: Theory and Review." Proceed- ings of the World Bank Annual Conference on Development Economics, 1989. Wash- ington, D.C.: World Bank. Fry, Maxwell J. 1978. "Money and Capital or Financial Deepening in Economic Devel- opment?" Journal of Money, Credit, and Banking 10 (4, November): 464-75. 546 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 .1980. "Saving, Investment, Growth, and the Cost of Financial Repression." World Development 8 (4): 317-27. .1988. Money, Interest, and Banking in Economic Development. Baltimore, Md.: The Johns Hopkins University Press. Gersovitz, Mark. 1988. "Saving and Development." In Hollis Chenery and T. N. Sri- nivasan, eds., Handbook of Development Economics. Vol. I. New York: North- Holland. Giovannini, Alberto. 1983. "The Interest Elasticity of Savings in Developing Countries: The Existing Evidence." World Development 11 (7): 601-07. . 1985. "Saving and the Real Interest Rate in LDCS." Journal of Development Economics 18 (August, 2-3): 197-218. Gupta, Kanhaya L. 1987. "Aggregate Savings, Financial Intermediation, and Interest Rate." Review of Economics and Statistics 69 (2): 303-11. Hammer, Jeffrey S. 1986. "Population Growth and Savings in LDCS: A Survey Article." World Development 14 (5, May): 579-93. Haque, Nadeem U., and Peter Montiel. 1989. "Consumption in Developing Countries: Tests for Liquidity Constraints and Finite Horizons." Review of Economics and Statis- tics 71 (3): 408-15. Hayashi, Fumio. 1985. "Tests for Liquidity Constraints: A Critical Survey." NBER Work- ing Paper 1720. National Bureau of Economic Research, Cambridge, Mass. Pro- cessed. Hsiao, Cheng. 1986. Analysis of Panel Data. Cambridge: Cambridge University Press. Hubbard, R. Glenn, and Kenneth L. Judd. 1986. "Liquidity Constraints, Fiscal Policy, and Consumption." Brookings Papers on Economic Activity 1: 1-50. IMF (International Monetary Fund). Various issues. International Financial Statistics. Washington. D.C. - =. Various issues. Government Financial Statistics. Washington, D.C. Kmenta, Jan. 1986. Elements of Econometrics. New York: Macmillan. Lahiri, Ashok K. 1988. Dynamics of Asian Savings: The Role of Growth and Age Structure. IMF Working Paper 88/49. Washington, D.C. Leff, Nathaniel H. 1969. "Dependency Rates and Savings Rates." American Economic Review 59 (5): 886-96. Leiderman, Leonardo, and Mario I. Blejer. 1988. "Modeling and Testing Ricardian Equivalence: A Survey." IMF Staff Papers 35 (1): 1-35. Mason, Andrew. 1987. "National Saving Rates and Population Growth: A New Model and New Evidence." In D. Gale Johnson and Ronald D. Lee, eds., Population Growth and Economic Development: Issues and Evidence. Madison: University of Wisconsin Press. . 1988. "Saving, Economic Growth, and Deniographic Change." Population and Development Review 14 (1): 113-43. Mason, Andrew, and others. 1986. Population Growth and Economic Development: Lessons from Selected Asian Countries. Policy Development Studies 10. New York: United Nations Fund for Population Activities. McKinnon, Ronald I. 1973. Money and Capital in Economic Development. Washington D.C.: Brookings Institution. Schmidt-Hebbel, Webb, and Corsetti 547 Mikesell, Raymond F., and James E. Zinser. 1973. "The Nature of the Savings Function in Developing Countries: A Survey of the Theoretical and Empirical Literature." Jour- nal of Economic Literature 11 (1): 1-26. Rossi, Nicola. 1988. "Government Spending, the Real Interest Rate, and the Behavior of Liquidity-Constrained Consumers in Developing Countries." IMF Staff Papers 35 (1): 104-40. - 1989. "Dependency Rates and Private Savings Behavior in Developing Coun- tries.' IMF Staff Papers 36 (1): 166-81. Schmidt-Hebbel, Klaus. 1987. "Foreign Shocks and Macroeconomic Adjustment in Small Open Economies'" Ph.D. dissertation. Massachusetts Institute of Technology, Department of Economics, Cambridge. Processed. Shaw, Edward S. 1973. Financial Deepening in Economic Development. New York: Oxford University Press. Singh, Sateesh K. 1975. Development Economics: Theory and Findings. Lexington, Mass.: Lexington Books. Taiwan, Government of. Various years. Statistical Yearbook of the Republic of China. Taipei. Webb, Steven B., and Heidi Zia. 1990. "Lower Birth Rates=Higher Saving in LDCS." Finance and Development 27 (2): 12-14. Williamson, Jeffrey G. 1968. "Personal Savings in Developing Nations: An Inter- temporal Cross-Section from Asia." Economic Record 44(106, June): 194-210. Zeldes, Stephen P. 1989. "Consumption and Liquidity Constraints: An Empirical Inves- tigation." Journal of Political Economy 97 (2): 305-46. THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 549-550 Cumulative Index of Titles VOLUME 6 (1992) "Alleviating Transitory Food Crises in Sub-Saharan Africa: International Altruism and Trade," Victor Lavy (1, January): 125-3 8 "Are High Interest Rates Effective for Stopping High Inflation? Some Skeptical Notes," Guillermo A. Calvo (1, January): 55-70 "The Business Cycle Associated with Exchange Rate-Based Stabilizations;' Miguel A. Kiguel and Nissan Liviatan (2, May): 279-305 "Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia," Martin Ravallion (1, January): 109-24 "The Dynamics of Optimal Gradual Stabilizations,' Alex Cukierman and Nissan Liv- iatan (3, September): 439-58 "The Economics of Farm Fragmentation: Evidence from Ghana and Rwanda," Benoit Blarel, Peter Hazell, Frank Place, and John Quiggin (2, May): 233-54 "A Framework for Evaluating the Impact of Pricing Policies for Cocoa and Coffee in C6te d'Ivoire,' Pravin K. Trivedi and Takamasa Akiyama (2, May): 307-30 "Household Saving in Developing Countries: First Cross-Country Evidence," Klaus Schmidt-Hebbel, Steven B. Webb, and Giancarlo Corsetti (3, September): 529-47 "Inflation and the Transition to a Market Economy: An Overview,' Simon Commander (1,January): 3-12 "Lessons from Experiences with High Inflation;' Rudiger Dornbusch (1, January): 13- 31 "Linking Trade and Productivity: New Research Directions ," James R. Tybout (2, May): 189-211 "Maize the Free Trade Agreement between Mexico and the United States," Santiago Levy and Sweder van Wijnbergen (3, September): 48 1-502 "The Measurement of Central Bank Independence and Its Effect on Policy Outcomes," Alex Cukierman, Steven B. Webb, and Bilin Neyapti (3, September): 353-98 "On the Transmission of World Agricultural Prices;' Yair Mundlak and Donald F. Larson (3, September): 399-422 "The Optimal Currency Composition of External Debt: Theory and Applications to Mexico and Brazil,' Stijn Claessens (3, September): 503-28 "Price-Wage Dynamics and Inflation in Socialist Economies: Empirical Models for Hun- gary and Poland;' Simon Commander and Fabrizio Coricelli (1, January): 33-53 "Project Evaluation and Uncertainty in Practice: A Statistical Analysis of Rate-of-Return Divergences of 1,015 World Bank Projects," Gerhard Pohl and Dubravko Mihaljek (2, May): 255-77 "How Small Enterprises in Ghana Have Responsed to Adjustment," William F. Steel and Leila M. Webster (3, September): 423-38 "The Short- and Long-Run Effects of Fiscal Policy," Edward F. Buffie (2, May): 331-51 549 550 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 "Social Security and Private Transfers in Developing Countries: The Case of Peru;' Donald Cox and Emmanuel Jimenez (1, January): 155-69 "Stagflationary Effects of Stabilization Programs in Reforming Socialist Countries: Enterprise-Side and Household-Side Factors," Guillermo A. Calvo and Fabrizio Cori- celli (1,January): 71-90 "Transportation Policy and Panterritorial Pricing in Africa," Mark Gersovitz (2, May): 213-31 "Uniform Commercial Policy, Illegal Trade, and the Real Exchange Rate: A Theoretical Analysis' Stephen A. O'Connell (3, September): 459-79 "Voluntary Choices in Concerted Deals: The Menu Approach to Debt Reduction in Developing Countries," Ishac Diwan and Kenneth Kletzer (1, January): 91-108 "The Willingness to Pay for Education for Daughters in Contrast to Sons: Evidence from Rural Peru," Paul Gertler and Paul Glewwe (1, January): 171-88 "The Working Behavior of Young People in Rural C6te d'Ivoire," Rob Alessie, Paul Baker, Richard Blundell, Christopher Heady, and Costas Meghir (1,January): 139-54 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3: 551-552 Cumulative Index of Authors VOLUME 6 (1992) Akiyama, Takamasa (see Trivedi, Pravin K.) Alessie, Rob, Paul Baker, Richard Blundell, Christopher Heady, and Costas Meghir, "The Working Behavior of Young People in Rural C6te d'lvoire" (1, January): 139-54 Baker, Paul (see Alessie, Rob) Blarel, Benoit, Peter Hazell, Frank Place, and John Quiggin, "The Economics of Farm Fragmentation: Evidence from Ghana and Rwanda" (2, May): 233-54 Blundell, Richard (see Alessie, Rob) Buffie, Edward F., "The Short- and Long-Run Effects of Fiscal Policy" (2, May): 331-Si Calvo, Guillermo A., "Are High Interest Rates Effective for Stopping High Inflation? Some Skeptical Notes" (1, January): 55-70 Calvo, Guillermo A., and Fabrizio Coricelli, "Stagflationary Effects of Stabilization Programs in Reforming Socialist Countries: Enterprise-Side and Household-Side Fac- tors" (1,January): 71-90 Claessens, Sti;n, "The Optimal Currency Composition of External Debt: Theory and Applications to Mexico and Brazil" (3, September): 503-28 Commander, Simon, "Inflation and the Transition to a Market Economy: An Overview" (1,January): 3-12 Commander, Simon, and Fabrizio Coricelli, "Price-Wage Dynamics and Inflation in Socialist Economies: Empirical Models for Hungary and Poland" (1, January): 33-53 Coricelli, Fabrizio (see Calvo, Guillermo A., and Commander, Simon) Corsetti, Giancarlo (see Schmidt-Hebbel, Klaus) Cox, Donald, and Emmanuel Jimenez, "Social Security and Private Transfers in Develop- ing Countries: The Case of Peru" (1, January): 155-69 Cukierman, Alex, and Nissan Liviatan, "The Dynamics of Optimal Gradual Stabiliza- tions" (3, September): 439-58 Cukierman, Alex, Steven B. Webb, and Bilin Neyapti, "The Measurement of Central Bank Independence and Its Effect on Policy Outcomes" (3, September): 353-98 Diwan, Ishac, and Kenneth Kletzer, "Voluntary Choices in Concerted Deals: The Menu Approach to Debt Reduction in Developing Countries" (1, January): 91-108 Dornbusch, Rudiger, "Lessons from Experiences with High Inflation" (1, January): 13- 31 Gersovitz, Mark, "Transportation Policy and Panterritorial Pricing in Africa" (2, May): 213-31 Gertler, Paul, and Paul Glewwe, "The Willingness to Pay for Education for Daughters in Contrast to Sons: Evidence from Rural Peru" (1, January): 171-88 Glewwe, Paul (see Gertler, Paul) Hazell, Peter (see Blarel, Benoit) Heady, Christopher (see Alessie, Rob) SS1 552 THE WORLD BANK ECONOMIC REVIEW, VOL. 6, NO. 3 Jimenez, Emmanuel (see Cox, Donald) Kiguel, Miguel A., and Nissan Liviatan, "The Business Cycle Associated with Exchange Rate-Based Stabilizations " (2, May): 279-3 05 Kletzer, Kenneth (see Diwan, Ishac) Larson, Donald F. (see Mundlak, Yair) Lavy, Victor, "Alleviating Transitory Food Crises in Sub-Saharan Africa: International Altruism and Trade" (1, January): 125-38 Levy, Santiago, and Sweder van Wijnbergen, "Maize and the Free Trade Agreement between Mexico and the United States" (3, September): 481-502 Liviatan, Nissan (see Cukierman, Alex, and Kiguel, Miguel A.) Meghir, Costas (see Alessie, Rob) Mihaijek, Dubravko (see Pohl, Gerhard) Mundlak, Yair, and Donald F. Larson, "On the Transmission of World Agricultural Prices" (3, September): 399-422 Neyapti, Bilin (see Cukierman, Alex) O'Connell, Stephen A., "Uniform Commercial Policy, Illegal Trade, and the Real Ex- change Rate: A Theoretical Analysis" (3, September): 459-79 Place, Frank (see Blarel, Benoit) Pohl, Gerhard, and Dubravko Mihaljek, "Project Evaluation and Uncertainty in Prac- tice: A Statistical Analysis of Rate-of-Return Divergences of 1,015 World Bank Proj- ects" (2, May): 255-77 Quiggin, John (see Blarel, Benoit) Ravallion, Martin, "Does Undernutrition Respond to Incomes and Prices? Dominance Tests for Indonesia" (1, January): 109-24 Schmidt-Hebbel, Klaus, Steven B. Webb, and Giancarlo Corsetti, "Household Saving in Developing Countries: First Cross-Country Evidence" (3, September): 529-47 Steel, William E, and Leila M. Webster, "How Small Enterprises in Ghana Have Re- sponded Adjustment" (3, September): 423-38 Trivedi, Pravin K., and Takamasa Akiyama, "A Framework for Evaluating the Impact of Pricing Policies for Cocoa and Coffee in C6te d'Ivoire" (2, May): 307-30 Tybout, James R., "Linking Trade and Productivity: New Research Directions" (2, May): 189-211 van Wijnbergen, Sweder (see Levy, Santiago) Webb, Steven B. (see Cukierman, Alex, and Schmidt-Hebbel, Klaus) Webster, Leila M. (see Steel, William E) -~~~I- "Byfocusing more attention on this age group, we can hope to improve its health status and productivity and ultimately the quality of life of the entire population." --Hiroshi Nakajima, Director General World Health Organization The Health of Adults in the Developing World Edited by Richard G.A. Feachem, Tord Kjellstrom, Christopher J. L. Murray, Mead Over, and Margaret A. Phillips The editors have put together an exceptional volume documenting the burden of adult ill-health in the developing world. It examines problems that lead to adult ill-health, such as infectious diseases and poor nutrition, and suggests ways to deter them. The book also discusses cost-effective preventive options in case management and primary and secondary care, along with ways to reduce the severity of many diseases. The 14 contributors include the five editors and nine other researchers representing the World Health Organization, the Centers for Disease Control, the Harvard School of Public Health, the London School of Hygiene and Tropical Medicine, as well as several other universities and the World Bank. Publishedfor the World Bank by Oxford University Press 360 pages / ISBN 0- 19-520879-X / US$34.95 Order Stock #60879 More Titles on Health Organizing and Managing Rfopical Disease Control Programs Lessons of Success Bernhard H. Liese, Paramjit S. Sachdeva, and D. Glynn Cochrane Provides guidelines to improve tropical disease programs and tells how to organize and deliver the essential medical technology. Gives an intercountry comparison based on case studies published in a companion volume. 62 pages /ISBN 0-8213-1959-0/US$6.95 / Order Stock #11959 Case Studies Reviews seven programs in Brazil, China, Egypt, and Philippines. Includes background, strategy, organization, management, and results of successful tropical disease control programs. 131 pages / ISBN 0-8213-2086-6 / US$8.95 / Order Stock #12086 Disease and Mortality in Sub-Saharan Africa Richard G. Feachem and Dean T. Jamison, editors Documents the significant changes in the health situation in Sub-Saharan Africa during the past 20 years. Presents useful insights concerning the patterns of health and disease, including such frightening challenges as AIDS. Publishedfor the World Bank by Oxford University Press 376 pages / ISBN 0-19-520826-9 / US$49.95 / Order Stock #60826 To ordler an) of these titles please use the coupon that follows. Emerging Stock Markets Factbook 1992 Gives instant access to 10 years of data on emerging - and influential - stock markets in the developing world. Annual data are presented by month, plus an expanded and updated directory of stock exchange opera- tions for more than 50 developing countries. 190 pages / ISBN 0-8213-2122-6 / US$60.00 / Order Stock #12122 Corporate Financial Structures in Developing Countries Alanyzes how corporate capital is structured and financed to show which financial markets and institutions will support long-term growth. 161 pages / ISBN 0-8213-1874-8 / US$10.95 / Order Stock #11874 Financing Corporate Growth in the Developing World Analyzes corporate capital structures of industrialized and developing countries, compares financing patterns, and reviews recent progress. 46 pages/ISBN 0-8213-1871-3 /US$6.95 / Order Stock #11871 TIends in Private Investment in Developing Countries 1992 edition Provides statistical data on private investment trends in 40 countries. 55 pages / ISBN 0-8213-2099-8 / US$6.95 / Order Stock #12099 Distributors of World Bank Publications Prices and credit terms vary from country to country. Consult your local distributor before placing an order. ARGENTINA ELSALVADOR KENYA SOUTH AFRICA, BOTSWANA Carlos Finch SRL Fusado Afica Book S (vice(A Ld. F-rirg tit CGretia Goms Alam Dr. M nue Broique Amy4o 43530 Quora HoM Mfianga,n Strt OCsd Ur-ivamity Ps FlPida 165,4th Floor-Of. 43/465 EdifddoSISA, I- PIo P.O. Boo 45245 Southe nAfica 1333 Braes Air Sa Salvador 011 Nairbi PO. Bxo 1141 CapeTown80 AUSTRALIA, PAPUA NEW GUINEA, FINLAND KOREA, REPUBUC OF FIR,SOLOMONISLANDS1 Akateadin, Kijkauppa Pa Korea BookCr Cp-torr Farna orodmu VANUATU, AND WESTERN SAMOA PO Box 128 PO. Bo ll, Kwa-gwhs-un hanatiuoral Subsiption Senor DilA. Books AJoirsoals SF452101 HdsnkiO0 Smul P.O. Box 41095 648 Whithorse Road Craghall Mitdchs 3132 FRANCE MALAYSIA Joho-ebug 2D24 Victoria Wrrld SnkPublcaties Univety f Maya CoopsMien 66,vovarad'lAt B-okohop. Lmited SPAIN AUSTRIA 75116 Pars P.O. ox 1127, ols. PetIaBaris M,odi,-P n Ubro, SA Ge,rd and Ca 59700 K.ala Lmpur Ctdlo 37 Ghaten31 GERMANY 2SDDI Madrid A-l01 Wia UNO-Veiag MEXICO Popp.dtxf-e AEee SS INFOTEC Labrei. lerosoon AEDOS BANGLADESH D-530 Bann Apartado Potal 22-560 C-rs1 ddeCaa, 391 MicroIndurtrio Dev-lopmrt 14060 Tlpi, Mdo D.F. UWD9Br-skiea Assistne Socety (MIDAS) GUATEMALA Houe 5, Road 16 Libre-n Pierl Santa NIETBERLANDS SRI LANKA AND THE MALOIVES DhIroondi R/Are Si. Call 7-55 Dv Undeb&ar/lrOr-Pub1katis Lake He Bookhop Dhaka 1209 Z`ona 1 P.O Box 202 PO Boo 244 Coatesas City 74SD AE Hlaksbeg 100, Sir Chittampalar A. BmC s en~C aGrdine- Masth. 156, Nur Ahed Sarak HONG KONG MACAO NEW ZEALAND Cokmbo2 Chdttbrg 40C Asia 20D Lid. EBSCO NZ Ltd 46-48 Wyndham Stbet Pivate Mil Bag 99914 SWBDEN 76, KDA. Averse Winning Cate Na Market Fosigl ileti Kr9100 2nd floor Auddland FPitz FPrdokaoretaget Cantal HFlg Kong Roge-ingagatan 12, Bar 16356 BELGIUM NIGERIA S-103 27 Stodkh1ol jean D Lannoy INDIA Unie,sity P,re lsinited Av d RrS 202 Allied Phlihes Pivt, Ltd. Th-ee Co-sBddi,gjei,,o Forclc'4tt ir 1060 Bnusxl 751 Morul Road Priv-te Mail Bag s1s5 W-g-Wilhzas, AS Madis--600002 rbada- B 3OD4 CANADA S-104 25 Storchdm IA Diffs-eur rs-hg NORWAY CP. 55,15B rue- Anpbre 15)N. H-rdiMarg Narve- nfornationCarte- SWITZERLAND Bouraiervile Qicibs Ballard EStabt Book DOartat Forsi-gletitle JAB 516 Bobay-4003038 P.O. Boo 612 Ett.stad Lfreisi Payt N-0602 Or16 1,rtrede. B-g CHINA 13/14 Asaf Alt Soad CH 10D2 Laoaanne Chi Pi-candal& Bn-sic New Deld- 110002 PAKISTAN Publishing HFoue Mica Book Ageny F-sbript rAoR S, Da F. SiDcscgjie 17 ChittarjanAv-uoe 65, Shahrah--Quaid--A Libradie Paynt Setjiog Ckutt.ta-700072 PO BoxNo 729 S-evire dAbratninat. LatSe 54, Caspostale 3312 COLOMBIA Jyadeva HatalBildtg CH 11D2 Lonne Inf-a,,e Ltds. Sth Main Road Gandtinagar PERU Apa tado Aers 34270 Barnolgse - 560D 09 Editorial Desarsoli SA TANZANIA Bogota D.E Apartado 3824 Oxford University Prs 3-5-1129 Khuigada Crnis Road Lsts I P.O. Bo 5299 COTE D'IVOIRE Hyde-abad - SW02027 Mktatb Rod Catred'Edic to dst Diffur PIHLIPPINES DOres. ESLam Africans. (CEDA) Pehana Rat6 2nd Flosr Ilste-naita Bock Cene-, 04BP. 541 Near ThakoreSau5 Nav-angpra Fifth Floor, lipin Life Building THAILAND Atbidjsn 04 Plateau Ahunedahad -30009 Ayala Aveue, Mke Canal oqpaotnert Store met'.sManila 3D6 SIte Pod CYPRUS Paliala Houre ,angkok Cypru CoUle Bookstore 16-AArhokMwrg POLAND 6, Diogener Street, Eng,i Lrknssw- 226001 Ontaniara l PubMhring S-ce TRINIDAD & TOBAGO, ANTIGUA PO. Boxo21D6 Uti Piebna 31/37 BARBUDA. BARBADOS, Nicodc Cantra zaar Road D8677WW va DOMLNICA, GRENADA, GUYANA, 60 Bli' Nage, JAMAICA, MONTSERRAT, ST. DENMARK Nagpur 440D10 FOrYO O orders KITTS & NEVIS, ST. LUCIA, Sa-nf-ndsLittrat-r IPSJoumb ST. VINCENT&GRENADINES RosasAnoE6Aid 11 INDONESIA UI. Ok,,reana 3 SytanaticsboadissUnit DK-1970 Fredailrsbe,g C PL Indira Limcited 02-916 Warnawsa 9 Watts Stit Ji S- Ratlangi 37 Curepe DOMINICAN REPUBLUC PO. Box 181 PORTUGAL Trindad, West ndbis EditDrasTalle, C porA Jakarta Pust Liaria Portugal Rtstaurrs.n , Isabel 1 Ct6lica 309 Rus Do Carmt 70-74 TURKEY Apaetado do Cmes. 7219D Z-l ISRAEL 1200 Lisba- Ifddo Santo Daruogo Ym-t Lite-aatreLd. N.taho,leSok No 15 P.O Box 5605 SAUDI ARABIA.QATAR Cagalogiu EGYPT, ARAB REPUBLIC OF Td AviV 61560 atr BookStbr kIsanbut Al Ahrlm less! P.O. B.o 3196 Al Gals Stret Riyadh 11471 UNITED KINGDOM Cairo ITALY Mi-fiso Ltd. 1cols C,runiaiaruria Sanstif SPA SWIGAPORE,TAIWAN, PO Boo 3 The Sfiddle East 0n,ve- Via Dra Di Calbria, 1/1 MYANMAREBRUNEI Altar, HarnphireGU34 2PG 41. Slerif Street Casella Folsa 552 hf ation Pobhbcaios England Cliro 28125 Firers Priva, Ltd. Golda Whed Building VENEZUELA JAPAN 41, Ka11ag Puddings 04-03 Libria del Sate Eart Book Se-vice Sigapore 1334 Apedo.60337 Hongo 3-Ghosts, Bunkyoku 113 Caracas 1060-A Tokyo CUSTOMERS IN THE UNITED STATES: CUSTOMERS OUTSIDE Complete this coupon and return to THE UNITED STATES: World Bank Publications Contact your local World Bank Publications Box 7247-8619 distributor for information on prices in local Philadelphia, PA 19170-8619 currency and payment terms. (A complete U.S.A. list of distributors follows this coupon.) If no To have your order shipped faster, charge distributor is listed for your country, use this by credit card by calling (908) 225-2165 order form and return it to the U.S. address. or send this completed order coupon by facsimile Orders that are sent to the U.S. address from by dialing (908) 417-0482. countries with distributors will be returned to the customer. Title Stock Number Price Subtotal US$ Postage and Handling* US$ Total US$ * If purchase order is used, actual postage will be charged. If payment is by check or credit card, postage and handling charges are $3.50 per order. For air mail delivery outside the U.S., include an additional US$6.00 per item. CHECK METHOD OF PAYMENT O Enclosed is my check payable to World Bank Publications. O Charge my I) VISA O MasterCard O American Express Credit card account number Expiration Date Signature O Bill me. (Institutional customers only. Purchase order must be included.) PLEASE PRINT CLEARLY Name Firm Address City State Postal Code Country Telephone Coming in the next issue of THE WORLD BANK ECONOMIC REVIEW January 1993 Volume 7, Number 1 Articles on ... * External Shocks, Purchasing Power Parity, and the Equilibrium Real Exchange Rate by Shantayanan Devarajan, Jeffrey D. Lewis, and Sherman Robinson * Moderate Inflation by Rudiger Dornbusch and Stanley Fischer * Theoretical Implications of Imperfect Competition on Quota License Prices and Auctions by Kala Krishna * Obstacles to the Development of Indigenous Small and Medium Enterprises by Brian Levy * Estimating Returns to Scale with Large Imperfect Panels: An Application to Chilean Manufacturing Industries by James R. Tybout and M. Daniel Westbrook