35056 THE WORLD BANK ECONOMIC REVIEW Volume 15 * 2001 - Number 3 Capital Account-Liberalization: What Do Cross-Country Studies Tell Us? Barry Eichengreen Where Has All the Education Gone? Lant Pritchett Measuring the Dynamic Gains from Trade Romain Wacziarg Ownership and Growth Thorvaldur Gylfason, Tryggvi Thor Herbertsson, and Gylfi Zoega Infrastructure, Geographical Disadvantage, Transport Costs, and Trade Nuno Limao and Anthony J. Venables A NEW DEVELOPMENT DATABASE Deposit Insurance around the World Asli Demirguii-Kunt and Tolga Sobaci OXFORD ISI)N 02-58-6 7(0 THE WORLD BANK ECONOMIC REVIEW EDITOR FranSois Bourguignon, World Bank EDITORIAL BOARD Abhijit Banerjee, Massachusetts Institute of Ravi Kanbur, Cornell University, USA Technology, USA Elizabeth M. King, WorldBank Kaushik Basu, Cornell University, USA Justin Yifu Lin, China Centerfor Economic Tim Besley, London School of Economics, UK Research, Peking University, China Anne Case, Princeton University, USA Mustapha Kamel Nabli, World Bank Stijn A. Claessens, University ofAmsterdam, Juan Pablo Nicolini, Universidad di Tella, The Netherlands Argentina Paul Collier, World Bank Howard Pack, University ofPennsylvania, USA David R. Dollar, World Bank Jean-Philippe Platteau, Facultes Universitaires Antonio Estache, World Bank Notre-Dame de la Paix, Belgium Augustin Kwasi Fosu, African Economic Boris Pleskovic, World Bank Research Council, Kenya Martin Ravallion, World Bank Mark Gersovitz, The Johns Hopkins Carmen Reinhart, University ofMaryland, USA University, USA Mark R. Rosenzweig, University of Jan Willem Gunning, Free University, Pennsylvania, USA Amsterdam, The Netherlands Joseph E. Stiglitz, Stanford University, USA Jeffrey S. Hammer, WorldBank Moshe Syrquin, University of Miami, USA Karla Hoff, WorldBank Vinod Thomas, World Bank Gregory K. Ingram, World Bank L. Alan Winters, University of Sussex, UK The WorldBankEconomicReview is a professional journal for the dissemination of World Bank-sponsored and outside research that may inform policy analyses and choices. It is directed to an international readership among economists and social scientists in government, business, and international agencies, as well as in universities and development research institutions. The Review emphasizes policy relevance and operational aspects of economics, rather than primarily theoretical and methodological issues. It is intended for readers familiar with economic theory and analysis but not necessarily proficient in advanced mathematical or econometric techniques. Articles will illustrate how professional research can shed light on policy choices. Inconsistency with Bank policy will not be grounds for rejection of an article. Articles will be drawn from work conducted by World Bank staff and consultants and from papers submitted by outside researchers. Before being accepted for publication, all articles will be reviewed by two referees who are not members of the Bank's staff and one World Bank staff member. Articles must also be recommended by a member of the Editorial Board. Non-Bank contributors are requested to submit a proposal of not more than two pages in length to the Editor or a member of the Editorial Board before sending in their paper. 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 WorldBank Economic Review, The World Bank, 1818 H Street, Washington, DC 20433, USA, or wber@worldbank.org. For more information, please visit the Web sites of the Economic Review at www.wber.oupjournals.org, the World Bank at www.worldbank.org, and Oxford University Press at www.oup-usa.org. THE WORLD BANK ECONOMIC REVIEW Volume 15 * 2001 * Number 3 Capital Account Liberalization: What Do Cross-Country Studies Tell Us? 341 Barry Eichengreen Where Has All the Education Gone? 367 Lant Pritchett Measuring the Dynamic Gains from Trade 393 Romain Wacziarg Ownership and Growth 431 Thorvaldur Gylfason, Tryggvi Thor Herbertsson, and Gylfi Zoega Infrastructure, Geographical Disadvantage, Transport Costs, and Trade 451 Nuno Limao andAnthonyJ Venables A NEW DEVELOPMENT DATABASE Deposit Insurance around the World 481 As/h Demirgic;-Kunt and Tolga Sobaci Author Index to Volume 15 491 Title Index to Volume 15 493 The World Bank Economic Review (ISSN 0258-6770) is published three times a year by Oxford University Press, 2001 Evans Road, Cary, NC 27513-2009 for The International Bank for Reconstruction and Development / THE WORLD BANK. Communications regarding original articles and editorial management should be addressed to The Editor, The World BankEconomicReview, 66, avenue d'I6na, 75116 Paris, France. E-mail: wber@worldbank.org. Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide. SUBSCRIPTIONS: Subscription is on a yearly basis. The annual rates are USS40 (£30 in UK and Europe) for individuals; US$90 (£63 in UK and Europe) for academic libraries; US$110 (£75 in UK and Europe) for corporations. Single issues are available for US$17 (X13 in UK and Europe) for individuals; US$38 (£26 in UK and Europe) for academic libraries; US$46 (£31 in UK and Europe) for corporations. All prices in- clude postage. Individual rates are applicable only when a subscription is for individual use and are not available if delivery is made to a corporate address. Subscriptions are providedfree of charge to non-OECD countries. All subscription requests, single issue and back issue orders, changes of address, and claims for missing issues should be sent to: NorthAmerica: Oxford University Press,Journals Customer Service, 2001 Evans Road, Cary, NC 27513- 2009, USA. Toll-free in the USA and Canada: 800-852-7323, or 919-677-0977. Fax: 919-677-1714. E-mail: jnlorders@oup-usa.org. Elsewhere. Oxford University Press, Journals Customer Service, Great Clarendon Street, Oxford OX2 6DP, UK. Tel: +44 1865 267907. Fax: +44 1865 267485. E-mail: jnl.orders@oup.co.uk. ADVERTISING: Helen Pearson, Oxford Journals Advertising, P.O. Box 347, Abingdon SO, OX14 1GJ, UK. Tel/Fax: +44 1235 201904. E-mail: helen@oxfordads.com. REQUESTS FOR PERMISSIONS, REPRINTS, AND PHOTOCOPIES: All rights reserved; no part of this publica- tion may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, elec- tronic, mechanical, photocopying, recording, or otherwise, without either prior written permission of the publisher (Oxford University Press, Journals Rights and Permissions, Great Clarendon Street, Oxford OX2 6DP, UK; tel: +44 1865 267561; fax: +44 1865 267485) or a license permitting restricted copying issued in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923 (fax: 978- 750-4470), or in the UK by the Copyright Licensing Agency Ltd., 90 Tottenham Court Road, London WlP 9HE, UK. Reprints of individual articles are available only from the authors. COPYRIGHT: Copyright (C 2001 The International Bank for Reconstruction and Development / THE WORLD BANK. It is a condition of publication in the journal that authors assign copyright to The International Bank for Reconstruction and Development / THE WORI.D BANK. However, requests for permission to reprint material found in the journal should come to Oxford University Press. This ensures that requests from third parties to reproduce articles are handled efficiently and consistently and will also allow the article to be disseminated as widely as possible. Authors may use their own material in other publications provided that the journal is acknowledged as the original place of publication and Oxford University Press is noti- fied in writing and in advance. INDEXING AND ABSTRACTING: The World Bank Economic Review is indexed and/or abstracted by CAB Abstracts, Current Contents/Social and Behavioral Sciences, Journal of Economic Literature/EconLit, PAIS International, RePEc (Research in Economic Papers), and Social Sciences Citation Index. The microform edi- tion is available through UMI, 300 North Zeeb Road, Ann Arbor, MI 48106, USA. PAPER USED: The World Bank Economic Review is printed on acid-free paper that meets the minimum requirements of ANSI Standard Z39.48-1984 (Permanence of Paper). POSTAL INFORMATION: The World Bank Economic Review (ISSN 0258-6770) is published three times a year by Oxford University Press, 2001 Evans Road, Cary, NC 27513-2009. Send address changes to The World Bank Economic Review, Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513-2009. THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 341-365 Capital Account Liberalization: What Do Cross-Country Studies Tell Us? Barry Eic7hengreen Capital account liberalization, it is fair to say, remains one of the most controversial and least understood policies of our day. One reason is that different theoretical per- spectives have very different implications for the desirability of liberalizing capital flows. Another is that empirical analysis has failed to yield conclusive results. I. THEORETICAL PERSPECTIVES Models of perfect markets suggest that international capital movements benefit both borrowers and lenders. Because international investment is intertemporal trade, trade between periods and trade between countries have precisely analo- gous welfare effects. The case for free capital mobility is thus the same as the case for free trade but for the subscripts of the model.1 To put the point another way, the case for international financial liberalization is the same as the case for domestic financial liberalization. If domestic financial markets can be counted on to deliver an efficient allocation of resources, why can't international finan- cial markets? The answer, another influential strand of thought contends, is that this effi- cient-markets paradigm is fundamentally misleading when applied to capital flows. Limits on capital movements are a distortion. It is an implication of the theory of the second best that removing one distortion need not be welfare en- hancing when other distortions are present. There are any number of constellations of distortions, especially in develop- ing economies, for which this is plausibly the case. If the capital account is liber- alized while import-competing industries are still protected, capital may flow to Barry Eichengreen is at the University of California, Berkeley. For helpful comments, the author thanks Stign Claessens, Geoffrey Garrett, Michael Klein, Aart Kraay, David Leblang, Gian Maria Milesi- Ferretti, Dennis Quinn, Frank Warnock, Charles Wyplosz, and the editor and anonymous referees of this journal. I. The intertemporal approach to capital mobility owes its origins to Fisher (1930). Influential modern treatments that resuscitated this approach and summarized its implications include Sachs (1981) and Frenkel and Razin (1996). (C 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 341 342 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 sectors in which the country has a comparative disadvantage, with immiserizing effects (Brecher and Diaz-Alejandro 1977). If a downwardly inflexible real wage causes too many resources to be devoted to capital-intensive activities, a capital inflow may aggravate this misallocation, again reducing the incomes and wel- fare of domestic residents (Brecher 1983). If information asymmetries are en- demic to financial markets and transactions, then there is no reason to assume that financial liberalization, domestic or international, will be welfare improv- ing (Stiglitz 2000). Even if information asymmetries in domestic markets are judged insufficiently severe to undermine the case for domestic financial liberal- ization, the same may not be true of international financial liberalization to the extent that international financial transactions take place among agents sepa- rated by greater physical and cultural distance. Insofar as these problems are most severe when the transactions in question involve developing countries, where the capacity to assemble and process information relevant to financial transactions is least advanced, there can be no presumption that capital will flow to uses for which its marginal product exceeds its opportunity cost. But are restrictions on capital movements any better? Capital controls shelter financial intermediaries from foreign competition. They weaken the market dis- cipline on policymakers. They vest additional power with bureaucrats who may be even less capable than markets at delivering an efficient allocation of resources and open the door to rent seeking and resource dissipation by interest groups seeking privileged access to foreign capital. Although there is theoretical support for both positions, the unfortunate fact is that the evidence on them does not speak clearly. It is not simply quarrels among theorists that have rendered capital account liberalization controversial, in other words, but the failure of attempts to move beyond anecdote and assertion to systematic empirical analysis to yield conclusive results. The question is why. Have the questions been formulated poorly? Are the methods flawed? Or are the data not up to the task? A critical review of the lit- erature is the obvious first step toward answering these questions. The challenge is that the literature is large and varied. Some studies approach the phenomenon from a macroeconomic point of view, others from a microeconomic perspective. Some focus on the effects of capital account liberalization, others on the causes- that is to say, on the political economy of the decision to liberalize. Any survey of this extensive and varied terrain requires a focus. Here the focus is on cross- country studies of the causes and effects of capital account liberalization, be- cause this is where the big questions are asked and attempts are made to reach conclusions of general applicability to developing countries.2 2. This focus on cross-country ("large n") studies not only dictates what is reviewed and what is left aside. It also differentiates this survey from other reviews of the literature on capital controls and capital account liberalization (such as Dooley 1996; Williamson and Mahar 1998; Cooper 1999; Edwards 1999; Neely 1999). At the opposite end of the empirical spectrum lie case studies of particular epi- sodes. While this "small n" approach allows a particular episode to be considered in great detail, it Eichengreen 343 II. MEASURING CAPITAL ACCOUNT LIBERALIZATION A first reason why studies of capital account liberalization do not speak clearly is the difficulty of measuring the policy. This section considers three approaches to the problem: measures based on statute, on actual flows, and on asset prices. Efforts to establish the presence of capital account restrictions on the basis of statute typically build on the data published by the International Monetary Fund (IMF) in its Exchange Arrangements and Exchange Restrictions annual.3 Most studies focus on restrictions on payments for capital transactions. When capital account liberalization is related to a measure of economic performance like GDP growth over a period of years, the annual observations are transformed into a variable measuring the proportion of years when the country had restrictions in place. Some investigators supplement this information with the IMF'S measure of restrictions on payments for current transactions and sometimes with its measures of surrender or repatriation requirements for export proceeds, sepa- rate exchange rates for capital transactions or invisibles, and bilateral payments arrangements with members and nonmembers.4 is likely to run head long into an identification problem, because many things will have been changing in the country in question in the period under consideration. "Hybrid studies" attempt to strike a bal- ance between these approaches by pooling detailed information on the capital account regime for sev- eral countries and years. An example is Reinhart and Smith (1998), who focus on five cases in which restrictions on capital account transactions were imposed or tightened-Brazil in 1994, Chile in 1991, Colombia in 1993, the Czech Republic in 1995, and Malaysia in 1994-and analyze a four-year win- dow surrounding the event. Similarly, Edison and Reinhart (1999) use daily financial data to examine four capital control episodes: Brazil in 1999, Malaysia in 1998, Spain in 1992, and Thailand in 1997. Four countries offer more degrees of freedom than one, to be sure, but it is still hard to know how far one can generalize from a handful of cases. 3. Along with narrative accounts of the main changes in policies toward the exchange rate and cur- rent and capital account payments, starting in 1967 this report has included a table summarizing the exchange arrangements adopted by member countries, but without any detail on how the narrative accounts are converted to summary data. Prior to 1967, the publication provided exclusively qualita- tive descriptions of restrictions. Some investigators (for example, Quinn 1997) have built indices of capital account liberalization for the earlier period from this information. In the second half of the 1990s, the IMsF began providing more detailed breakdowns of policy measures. Starting in 1996, the report disaggregated controls on export proceeds into "surrender requirements for export proceeds" (requir- ing exporters to surrender to the authorities foreign exchange earned from exporting) and "repatria- tion requirements for export proceeds" (requiring them to surrender even payments made to overseas accounts). Starting in 1997 it distinguished controls on capital inflows and outflows. These changes create problems for investigators seeking to create time series for capital account liberalization. Thus Glick and Hutchinson (2000) use surrender requirements for export proceeds, which are more restric- tive than repatriation requirements for export proceeds, as equivalent to the pre-1996 export surrender measure, and code a country as having capital account restrictions in place in 1997 or 1998 when the report listed controls as in place for five or more of these capital account subcategories and "'financial credit" was one of the categories restricted. 4. Restrictions on current account transactions affect the ability of the private sector to obtain for- eign exchange for payments related to merchandise imports and to retain foreign exchange earned through exporting and the ability of foreign direct (and other) investors to repatriate interest earnings and other profits. The argument for using them is that current account transactions can be used to evade restric- 344 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 These data have limitations.' Data on "restrictions on payments for capital transactions" available before 1996, for example, may not reflect restrictions on capital transfers by nonresidents. In addition, drawing a line between mea- sures affecting the current account and those affecting the capital account is prob- lematic. Thus data on separate exchange rate(s) for some or all capital transac- tions, for instance, include measures affecting some or all invisibles, which may include payments on current as well as capital account. Bilateral payments ar- rangements with members and nonmembers include not just separate exchange rates for capital transactions, which are directly relevant to a consideration of capital account liberalization, but also the use of one unitary rate for transac- tions with one country but a different unitary rate for transactions for another country, where the second kind of multiple rate is often used to discriminate among transactions on current as well as capital account. Although the presence of current account restrictions, export-surrender re- quirements, bilateral payments arrangements, and separate exchange rates may convey information on the scope of efforts to deter the evasion of capital con- trols, deterrence is not their main purpose. Moreover, current account restric- tions are likely to have other important effects that the unwary investigator may conflate with their impact on capital mobility. They influence merchandise trade. They limit opportunities for repatriating interest and principal. And insofar as they tend to be imposed by countries suffering from serious policy imbalances, their "effects" will reflect the influence of these deeper policy problems as much as those of the capital controls themselves.6 Most studies "solve" the problem of measuring the intensity of controls by ignoring it. They settle for constructing a dummy variable for the presence or tions on capital account-related payments (by resort to leads and lags and over- and underinvoicing of exports and imports) and that surrender requirements, bilateral payments restrictions, and multiple exchange rates, which may then be used to close off these avenues of evasion, therefore contain infor- mation on the intensity of controls. 5. Leading in turn to creative attempts to supplement them. Some investigators have used such sources as the International Finance Corporation's Emerging Market Facts Book and World Bank country reports. Thus Levine and Zervos (1998) and Levine (1999), who are concerned to identify major changes in restric- tions on capital flows, consult all these sources and count only episodes corroborated in more than one publication and described there as "major" or "significant." Kraay (1998), relying exclusively on Exchange Arrangements and Exchange Restrictions, identifies major liberalization episodes as years that are pre- ceded by five consecutive years of capital controls and followed by five consecutive years of no controls. 6. Similar arguments are made about the black-market premium, which is sometimes used as a measure of restrictions-namely, that it distorts the pattern of trade, is associated with serious macro- economic policy imbalances, and tends to widen in response to political shocks. Thus, Sachs and Warner's (1995) measure of economic openness depends mainly on the black-market premium (one of its four components), as Rodriguez and Rodrik (1999) show. Rodriguez and Rodrik argue that this index is unlikely to be a good measure of openness per se because it tends to be associated with macroeconomic and political instability. Similar arguments can be made about capital controls themselves: Countries with serious policy imbalances are the most likely to resort to the instrument. The implication is that any effect superficially associated with the measure conflates the influence of those underlying condi- tions and that of the policy instrument itself. Eichengreen 345 absence of controls. In an attempt to go further, Quinn (1997) distinguishes seven categories of statutory measures for 56 countries for 1950-94 and for 8 more countries starting in 1954. Four are current account restrictions, two are capital account restrictions, and one captures international agreements constraining a country's ability to restrict exchange and capital flows, such as membership in the Organisation for Economic Co-operation and Development (OECD). For each category, Quinn codes the intensity of controls on a 2-point scale (from 0, most intense, to 2, no restriction) to produce a 0-14 index of current and capital account restrictions and a 0-4 index of capital account restrictions.7 Not sur- prisingly, Quinn's index has proven wildly popular and has been used by many subsequent investigators.8 The difficulty of deriving measures of the policy regime from information on statutes and policies has led investigators to experiment with alternatives. Kraay (1998) and Swank (1998) use actual capital inflows and outflows as a percent- age of GDP as a measure of the freedom of capital movements. The problem, as these investigators are aware, is that actual inflows and outflows will be affected by a range of policies and circumstances-monetary, fiscal, and exchange rate policies; the global economic and financial climate; and political circumstances, to name three-and not merely by restrictions on capital flows. Hence, this measure is unlikely to be an informative indicator of the capital account regime.9 Bekaert (1995) and Aherane and others (2000) use one minus the ratio of the market capitalizations of the International Finance Corporation's (IFC) Invest- able and Global Indices. The Investible Index consists of the stocks (or portions of stocks) in the Global Index deemed to be available to foreign investors. Thus, one minus the ratio of the two can be interpreted as a measure of the intensity of foreign ownership restrictions. The limitation of this measure, obviously, is that it captures only restrictions on equity inflows.10 7. Such a high degree of differentiation necessarily relies on the judgment of the coder. Quinn addresses this problem by having each observation coded by two coders and then reconciling the differences. 8. A more detailed index has been constructed by Johnston and others (1999) for 41 industrial, developing, and transition economies, but only for 1996. This uses the detailed breakdown of 142 individual types of exchange and capital controls (aggregated into 16 categories) first published in Exchange Arrangements and Exchange Restrictions in 1997. The existence and intensity of controls are measured by normalizing the number of actual categories of controls (separately for controls on current and capital accounts) by the number of feasible measures. The number of countries for which they provide these estimates is limited, reflecting the limited coverage of the 1997 edition of Exchange Arrangements and Exchange Restrictions. In addition, the time dimension is lost due to the absence of comparable data for prior years. 9. It is likely to be useful only for distinguishing countries wholly closed to capital flows, where payments on capital account will be zero, from more open countries, the notion being that only coun- tries with draconian controls that render them wholly closed to international financial markets will display neither inflows or outflows. 10. In addition, the measure captures more than statutory controls; for example, if a large firm that trades on, say, the Manila Stock Exchange is held mainly by one or two Filipino investors, their share would enter the Global Index but its weight in the 1FwC would be based on the portion of the shares available to foreigners. 346 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Several researchers have used the correlation of stock market returns across countries as a measure of the international integration of securities markets. But the correlation of raw returns says little about the integration of markets, be- cause returns will vary with the characteristics of the underlying assets, which depend on the characteristics and condition of the entities issuing the claims. Thus, in a study representative of the genre, Bekaert (1995) first regresses national returns in excess of the U.S. interest rate on five instrumental variables (lagged local and U.S. excess returns, local and U.S. dividend yields, and a transfor- mation of the U.S. interest rate, variations in which might create reasons why the excess returns on different markets might differ) to derive expected returns, before computing the correlation of the these expected returns with expected returns in the United States as a measure of market integration." Clearly, the resulting measure is only as good as the model that generates the expected re- turns.12 These studies show some markets to be more integrated than would be expected from the statutory restrictions placed on foreign ownership of domes- tic securities. A limitation of the approach is that it is hard to know whether the contrast reflects the limited effectiveness of the statutes, which results in a mis- leading picture, or problems with one or more of the assumptions needed to derive expected returns. Other researchers use onshore-offshore interest differentials and deviations from covered interest parity to measure capital mobility.'3 Unlike stock market returns, which must be purged of premia and discounts associated with the char- acteristics of the entities issuing them before they can be used to gauge market integration, short-term interest rates can be analyzed without transforming them in model-contingent ways.'4 However, interest differentials tend to be available only for a limited number of countries and years-specifically for countries im- portant enough to have well-developed offshore markets and advanced enough financially to have well-developed forward currency markets. Because industrial and emerging markets with these characteristics are not representative of the 11. A disadvantage of this simple implementation is that no changes in the estimated degree of market integration are allowed to occur over time. Harvey (1995) and Bekaert and Harvey (1995) implement rolling- and switching-regression methods that, subject to further assumptions, permit the degree of market integration to vary over time. 12. If assets are priced according to a multifactor model rather than the one-factor model with constant risk exposures that Bekaert assumes, emerging markets might display cross-section differences in risk exposures and in the correlation of expected returns with the U.S. market, even if those markets are otherwise integrated internationally. 13. See, for example, Frankel and MacArthur (1988), Giavazzi and Pagano (1988), Cody (1990), Obstfeld (1993), Marston (1993, 1995), and Holmes and Wu (1997). Dooley and Isard (1980), Ito (1983), and Wong (1997), among others, take a similar approach by using the black-market exchange rate premium. 14. Researchers justify their disregard of the country risk premium by focusing on high-quality debt securities for which default risk is close to zero. They disregard currency risk by focusing on covered interest parity. Eicben green 347 larger population of developing countries, drawing broad generalizations from these studies is likely to be problematic."5 Onshore-offshore interest differentials also have the inconvenient property of widening when there is an incentive for capital to move (when there is fear of a crisis, for example), while remaining narrower at other times. To put the point another way, differentials reflect not just the stringency of statutory controls but their interaction with ancillary policies and circumstances, making it difficult to separate the two influences. This observation points to a limitation of virtually all studies of capital con- trols. Controls tend to be imposed and removed as part of a larger package of policy measures.16 Clearly, then, it is important to control for the other elements of the reform package when studying the connections between capital account restrictions and economic growth, investment, and financial depth. Alas, this is easier said than done. Trade openness, financial depth, institutional development, and the like may be no easier to measure in an economically meaningful way than the presence or absence of capital controls. Developing adequate measures of capital account restrictions is a particular problem for the literature on the causes and effects of capital controls, but the more general problem of adequately capturing the economic, financial, and political characteristics of economies, which impinges on all cross-country empirical work of this sort, should not be overlooked. III. WHO USES CONTROLS, WHO LIBERALIZES, AND WHY? A large literature addresses the circumstances under which capital accounts are opened and the circumstances under which restrictions are retained. Perhaps the single most robust regularity in this literature is the negative association between per capita income and controls. Per capita income is typically interpreted in this context as a measure of economic development: The more developed the coun- try, the more likely that it will have removed restrictions on capital flows. The observation that all of today's high-income countries have removed their con- trols is consonant with the view that capital account liberalization is a corollary of economic development and maturation. But why is this the case? Does the more advanced development of institutions and markets in the high-income countries mean that these countries can better 15. In addition, focusing on cases where a significant onshore-offshore differential is quoted also has the consequence, not obviously desirable, of shifting attention from policies designed to limit capital mobility to policies effective in limiting capital mobility. Though many countries may put in place measures to limit capital flows, only where such policies are effective will a consequential offshore market develop and a significant onshore-offshore differential be observed. Focusing on cases where controls were effective-because, for example, the country had the administrative capacity to enforce them- again runs the risk of limiting the analysis to countries that are not representative. And it disregards much of what is interesting in the debate, namely, the capacity of the markets to neutralize the intended effects of statutory measures. 16. This is a theme of Ariyoshi and others (2000). 348 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 accommodate capital account liberalization-that well-developed markets and institutions shift the balance toward benefits and away from costs? Do these countries' well-developed political systems create avenues through which those who oppose restraints on their civil liberties-including their financial liberties- can make that opposition felt? Explaining why restrictions on international fi- nancial flows are more prevalent in some countries than others and why, in particular, they are less prevalent in the high-income countries is at the center of the literature on the political economy of controls. A specific development-related rationale for controls-on capital outflows in particular-is that they can usefully channel domestic saving into domestic in- vestment in countries where the underdevelopment of markets and institutions would otherwise result in a suboptimal supply of finance for investment. Thus Garrett and others (2000) find that there is a tendency to restrict capital account transactions in countries where domestic savings are scarce and that this effect is strongest for developing economies, where the premium on mobilizing sav- ings for domestic investment is presumably the greatest. Another strand of work pursues the association of controls with the exchange rate regime. Capital mobility increases the difficulty of operating a currency peg. Countries committed to pegging-China and Malaysia come to mind- may therefore support pegs with restrictions on capital flows. Contributors to the cross-country empirical literature generally find that countries with pegged exchange rates are less likely to have an open capital account (Leblang 1997, 1999; Milesi-Ferretti 1998; Bernhard and Leblang 1999; Garrett and others 2000).17 But it is not clear what should be regarded as endogenous and what as exog- enous in this analysis. Does a willingness to adopt a more flexible exchange rate determine the readiness of some countries to remove controls? Or do increases in capital mobility, associated perhaps with the removal of capital controls, lead to the adoption of a more flexible exchange rate, either voluntarily or as a result of crisis? Causality may run both ways, making it difficult to interpret an ordi- nary least squares regression coefficient on the exchange rate. As will become apparent, this difficulty of pinning down the direction of causality is a chronic problem in the literature on capital account liberalization (and a theme of this survey). Another line of thought portrays capital controls as an instrument of govern- ment revenue management. Controls limit the ability of residents to avoid the inflation tax on domestic money balances by shifting into foreign assets (Alesina and Tabellini 1989). They permit the authorities to raise reserve requirements on domestic financial institutions and thereby reduce their debt servicing costs 17. Similarly, countries with macroeconomic problems that may threaten the stability of a peg (a weak current account, a large budget deficit, sudden increases in interest rates, for example) have a disproportionate tendency to maintain controls, outflow controls in particular (Johnston and Tamirisa 1996). Eichengreen 349 without eroding the inflation tax base (Drazen 1989).Y This perspective sug- gests that controls are likely to be used where the domestic financial system is tightly regulated and reserve requirements can be used to compel financial insti- tutions to hold public sector liabilities. Consistent with this prediction, Leblang (1997) finds that governments that are less reliant on seigniorage are less likely to have capital controls. A further implication is that controls are less likely to be used where the inflation tax is not available because the central bank is inde- pendent and monetary policy is controlled by a conservative board. Epstein and Schor (1992), Alesina and others (1994), Quinn and Inclan (1997), Milesi-Ferretti (1998), and Bai and Wei (2000) all find that countries with more independent central banks are less likely to use controls. But does this pattern reflect the implications of central bank independence and domestic financial liberalization for the availability of inflation tax revenues, as these authors argue, or a common omitted factor-laissez-faire ideology, for example-associated with financial liberalization, central bank independence, and capital decontrol alike? Some investigators have sought to distinguish be- tween these alternatives by adding the political orientation of the government as a further determinant of the propensity to use controls. Once ideology is con- trolled for, they argue, any surviving correlation between central bank indepen- dence and domestic financial liberalization on the one hand and capital account liberalization on the other will reflect the implications of central bank indepen- dence and domestic financial liberalization for the seigniorage revenues prom- ised by controls. Though findings on the effect of government ideology are mixed, the effect of central bank independence survives this extension, consistent with the implications of the seigniorage-centered approach.19 A number of investigators pursuing this line have found democracy to be positively associated with capital account liberalization (see, for example, Quinn 2000 and Garrett and others 2000). Democracy may be a mechanism for resolving social conflicts that otherwise force resort to financial repression and the infla- tion tax (Garrett and others 2000). More generally, with democracy comes an increasing recognition of rights, including the international economic rights of 18. Moreover, by facilitating the use of rate ceilings and other administrative measures that cap interest rates, controls limit the cost of borrowing for those at the head of the financial queue, includ- ing the government and any private sector borrowers that it favors. 19. Epstein and Schor (1992) find that left-wing governments are more likely to maintain controls. While Garrett and others (2000) also conclude that left-wing governments are more likely to resort to controls, the effect is statistically insignificant at standard confidence levels. Only when high-income countries are removed from the sample is the association robust. While Quinn and Inclan (1997) also find some evidence that left-wing governments are more likely to retain controls, this effect is much more pronounced for the 1960s and 1970s than the 1980s. Alesina and others (1994) reach even more negative conclusions: They find little discernible effect of ideological orientation either before or dur- ing the 1980s after controlling for other characteristics of governments-coalition or majoritarian, cabinet durability and turnover-that plausibly reflect the time horizon of the government and therefore its propensity to put off tax increases to another day in favor of the inflation tax. 350 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 residents, and a greater ability to press for the removal of restrictions on their investment options (Dailami 2000). Several recent studies (Simmons and Elkins 2000; Garrett and others 2000) suggest that "policy contagion" affects the decision to open the capital account. Countries are more likely to liberalize when members of their peer group have done so, holding constant other factors. The pattern can be interpreted as policy emulation (governments are influenced by the initiatives of their neighbors) or signaling (when competitors have liberalized portfolio flows, it becomes harder to retain controls and, at the same time, remain an attractive destination for for- eign direct investment). But are such interpretations justified? It is a common problem in the litera- ture on contagion, financial and otherwise, that the simultaneity of policy ini- tiatives in different countries may reflect not the direct influence of events in one country on another countries but a tendency for decisionmakers to respond similarly to economic and political events not adequately controlled for in the analysis.20 Simmons and Elkins (2000) address this possibility by defining a country's economic neighbors as those that compete with it for foreign invest- ment (in the case of capital account restrictions) and those that compete with it in export markets (in the case of current account restrictions). These more so- phisticated proxies for policy contagion matter even when crude measures of common omitted factors (such as the share of countries in the same region that have liberalized their capital accounts) are also included in the specification. These findings go a good way toward explaining the recent trend toward capital account liberalization. Financial repression has given way to deregula- tion of domestic financial institutions and markets in a growing number of coun- tries. Governments and central banks have abandoned currency pegs in favor of greater exchange rate flexibility. The 1980s and 1990s were decades of democ- ratization in much of the developing world. As these developments led some countries to liberalize, the trend gathered momentum, as the literature on policy contagion suggests. Together these forces lent considerable impetus to the pro- cess of capital account liberalization.2' Before researchers congratulate themselves for their success and close up shop, it is worth noting other explanations that have been denied the same systematic attention. For example, capital controls may have become less attractive because information and communications technologies have grown more sophisticated, rendering controls more porous and their effective application more distortionary (Eichengreen and others 1998). The technical progress in question is hard to measure. A time trend intended to capture secular improvements in information 20. For a discussion of the problem of common unobserved shocks, see Eichengreen and Rose (1999). 21. At the same time, the research described in this section suggests the kinds of circumstances and events-disenchantment with financial liberalization, disaffection with flexible exchange rates, inef- fective democratic governance-that could conceivably reverse the trend toward capital account liber- alization sometime in the future. Eichengreen 351 and communications technologies would be contaminated by a variety of other omitted factors that were also changing over time. As is the case all too often in empirical economics, there may have been a tendency to focus on factors that are readily measured and quantified to the neglect of those that are more diffi- cult to capture. IV. CAPITAL MOBILITY AND GROWTH The most widely cited study of the correlation of capital account liberalization with growth is Rodrik (1998). Using data for roughly 100 industrial and devel- oping countries for 1975-89, Rodrik regresses the growth of GDP per capita on the share of years when the capital account was free of restriction (as measured by the binary indicator constructed by the IMF), controlling for determinants suggested by the empirical growth literature (initial income per capita, second- ary school enrollment, quality of government, and regional dummy variables for East Asia, Latin America, and Sub-Saharan Africa). He finds no association between capital account openness and growth and questions whether capital flows favor economic development. Given the currency of this article among economists, it is striking that the lead- ing study of the question in political science reaches the opposite conclusion. For 66 countries over the period 1960-89, Quinn (1997) reports a positive cor- relation between the change in his capital account openness indicator and growth. That correlation is robust and statistically significant at standard confidence levels. What explains the contrast is not clear. One difference that may matter is that Quinn's study starts earlier. Consequently, growth in his sample period is not dominated to the same extent by the "lost decade" of the 1980s (when there were virtually no capital flows to emerging markets to stimulate growth). With an earlier start, his sample may include more observations in which countries liberalized inflows of foreign direct investment, with positive effects on growth, and fewer in which they liberalized short-term portfolio flows, whose effects may have been more mixed. In addition, Quinn has more independent variables, and he looks at the change in capital account openness rather than the level. Edwards (2001) emphasizes that Quinn's measure of capital account liberalization is more nuanced and presumably informative. For example, Quinn's measure conveys information about whether capital account opening was partial or across the board, whereas the standard IMF measure does not.22 Quinn's country sample is also different, in that he considers fewer low-income developing economies. There are reasons to think that the effects of capital ac- count liberalization vary with financial and institutional development. Remov- 22. I return to the distinction between partial and comprehensive capital account openness and restrictions in section VI on crises. I ~ I I I I 352 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 ing capital controls may be welfare and efficiency enhancing only when there are no serious imperfections in the information and contracting environment, an implication of the theory of the second best, as noted at the beginning of this article. Portfolio capital inflows stimulate growth, this argument goes, only when markets have developed enough to allocate finance efficiently and when the contracting environment forces agents to live with the consequences of their investment decisions. The Asian crisis encouraged the belief that countries that open their economies to international financial transactions benefit only if they first strengthen their markets and institutions. Thus a positive impact on growth comes only if prudential supervision is first upgraded, the moral hazard created by too generous a financial safety net is limited, corporate governance and creditor rights are strengthened, and transparent auditing and accounting standards and equitable bankruptcy and insolvency procedures are adopted. Although these institutional prerequisites are difficult to measure, there is a presumption that they are most advanced in high-income countries. Edwards (2001) supports this view: Using Quinn's measure of the intensity of capital account restrictions, he finds that liberalization boosts growth in high-income countries but slows it in low-income countries.23 He shows further that the sig- nificance of capital controls evaporates when the IMF index used by Rodrik is substituted for Quinn's more differentiated measure. Thus it is tempting to think that the absence of an effect in earlier studies is a statistical artifact. And there is some suggestion that capital account liberalization is more beneficial in more financially and institutionally developed economies.24 But do these apparent differences between high- and low-income countries really reflect their different stages of financial and institutional development? Kraay (1998) attempts to directly test the hypothesis that the effects of capital account liberalization depend on the strength of the financial system, the effec- tiveness of prudential supervision and regulation, and the quality of other poli- cies and institutions.25 The results are not encouraging: the interaction of the quality of policy and institutions with financial openness is almost never posi- 23. Quinn's measure of capital account openness enters negatively, in other words, whereas the interaction between capital account openness and per capita income enters positively. 24. Using a different methodology, Quinn (2000) reaches a similar conclusion. He estimates bi- variate vector autoregressions using growth rates and his measures of capital account liberalization individually for a large number of middle- and low-income countries. He finds scant evidence that capital account liberalization has had a positive impact on growth in the poorest countries, but some positive evidence for middle-income countries, especially those that have other characteristics likely to render them attractive to foreign investors. 25. Kraay uses the ratio of M2 to GDP and the ratio of domestic credit to the private sector relative to GDP as ex ante proxies for the level of financial development, and one minus the average number of banking crises per year as an ex post indicator of financial strength. As an indicator of the strength of bank regulation, he uses a measure based on whether banks are authorized to engage in nontraditional activities, such as securities dealing and insurance. To capture the broader policy and institutional en- vironment, he uses a weighted average of fiscal deficits and inflation, the black-market premium, and indices of corruption and the quality of bureaucracy. Eichengreen 353 tive and significant, and it is sometimes significantly negative.26 Arteta and oth- ers (2001) similarly interact the level of capital account openness with the liquid liabilities of the financial system as a measure of financial depth and with Inter- national Country Risk Guide's index of law and order as a measure of institu- tional development. Again, the results are largely negative. There is little evidence that the growth effects of capital account openness are shaped in robust and predictable ways by a country's level of financial and institutional development. More important for shaping the effects of capital account liberalization, these authors suggest, is the sequencing of reforms. Countries that first complete the process of macroeconomic stabilization, allowing them to remove exchange controls and other distortions on the current account side, enjoy stronger growth effects of capital account openness. While some of the qualitative literature simi- larly suggests that sequencing is an important determinant of the effects of capi- tal account opening, systematic cross-country empirical analysis has barely begun. (In other words, there do not appear to be other "large-n" studies like that of Arteta and others 2001 that address this question.) One way of unraveling the mystery of why the growth effects of capital ac- count liberalization do not seem to vary as expected with institutional and fi- nancial development is to determine whether these results are sensitive to the measures of policies and institutions used. Here, it will be evident, work is al- ready under way. Another way is to pin down the mechanisms or channels through which capital account liberalization affects the economy, the approach examined next. V. CHANNELS LINKING CAPITAL ACCOUNT LIBERALIZATION WITH GROWTH The cross-country growth literature points to a number of factors that plausibly intermediate between capital account liberalization and growth. Investment, fi- nancial development, and the stability of macroeconomic policy, among other variables, have been shown to be positively related to an economy's rate of growth (see, for example, Levine and Renelt 1992; Levine 1997; Barro 1997). All of these variables create channels through which capital account liberalization can po- tentially exercise an effect. Studying the impact of capital account policy on these intermediate variables is thus a way of inferring its implications for growth. This section focuses on the impact of capital account policies on two of the channels that have received the most attention: investment and the depth and develop- ment of financial markets. Many attempts have been made to analyze the connections between capital account policies and investment. Rodrik (1998) relates the investment to GDP 26. Note that the test here is for whether the effects of capital account openness are conditional on these measures of institutional development. These measures are not simply used as additional controls in the growth equation; rather, they are entered interactively. 354 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 ratio to the IMF'S measure of capital account openness, again finding no trace of an effect. Kraay (1998) similarly finds no impact on gross domestic investment as a share of GDP, using the IMF index, the Quinn index, and gross inflows and outflows as alternative measures of financial openness. He considers the possi- bility that capital account openness positively affects investment only in countries where risk-adjusted returns exceed the world average-that is, where liberaliza- tion will cause capital to flow in rather than out. Using the average balance on the financial account of the balance of payments as a proxy for risk-adjusted returns, he reports a positive impact on investment when this variable is inter- acted with capital account openness. However, the coefficient in question dif- fers significantly from zero for only one of Kraay's three measures of capital account openness.27 Because the evidence on investment does not speak clearly, it is logical to strip off another layer and consider variables like real interest rates and financial depth-factors on which investment plausibly depends. Governments have used capital controls in support of administrative measures designed to keep interest rates low with the express purpose of stimulating investment. And a substantial number of studies have confirmed that capital controls are associated with lower real interest rates (see, for example, Alesina and others 1994; Grilli and Milesi- Ferretti 1995; Bordo and Eichengreen 1998; Wyplosz 1999). But whether there are benefits for growth is a separate question. The literature on financial repres- sion-especially the recent literature-is skeptical that interest rate ceilings, even if they reduce the cost of investment, succeed in nurturing growth. Although artificially low real rates reduce the required return on investment, they impede financial development. And financial development presumably increases the ef- ficiency of investment as well as financing and otherwise facilitating experimen- tation with new technologies.28 Klein and Olivei (1999) find that capital account openness stimulates finan- cial depth (measured variously as the change in the ratios of liquid liabilities to GDP, claims on the nonfinancial private sector, and bank domestic assets in de- posit money to the sum of bank domestic assets in deposit money and central bank domestic assets). But the correlation between capital account openness and financial deepening is limited to the OECD countries; the relationship dissolves 27. The measure in question is actual (gross) inflows and outflows. Because the interaction term is then gross inflows and outflows times net inflows and outflows, one suspects that it is dominated by cases where investment reacted to exceptional surges of capital inflows. In addition one worries about the near-tautological nature of using a variable that essentially captures whether or not capital flowed in as a way of determining whether the policy affected investment. Kraay's findings also appear to be sensitive to the estimator used and the sample period: He obtains different results depending on whether he estimates his investment equation by ordinary least squares or instruments his measures of capital account restrictions to control for their endogeneity. 28. The literature on the link between financial development and growth is vast-even more so than that on the topic surveyed here. Attempting to review the controversies and contributions would not be realistic. The reader may refer to Levine (1997) for a full-scale review of the topic. Eichengreen 355 when these countries are excluded from the sample. Thus where researchers like Kraay (1998) and Arteta and others (2001) find little evidence that an open capital account does more to stimulate growth in high-income countries,29 Klein and Olivei conclude that it may do more in the advanced industrial countries to stimu- late certain inputs into growth-specifically, well-developed financial markets. That the effect is indirect (an open capital account encourages financial devel- opment, which in turn encourages growth) and contingent (presumably) on a range of intervening factors may be why it has been so difficult to document a direct link from the capital account to growth that varies between high- and low- income countries. But not all investigators agree that the influence of capital account liberaliza- tion on financial development is limited to high-income countries. Levine and Zervos (1998) find for 16 developing economies that stock markets become larger and more liquid after the capital account is opened. To be sure, this study fo- cuses on a different aspect of financial development, namely, stock markets rather than bank intermediation. But why the evidence for different financial markets is apparently contradictory is not clear. It could be that Levine and Zervos's 16 countries, selected for having functioning stock markets, were already relatively advanced financially, so that capital account liberalization could then have a positive and powerful impact on their further deepening and development. Al- ternatively, it could be that banking systems typically are already relatively well developed when capital accounts are opened, so that the main effect of liberal- ization is on stock markets whose development is still at an earlier stage. Sorting through this controversy may require more sophisticated measures of capital account liberalization. Whether liberalization favors the development of banks or securities markets plausibly depends on how liberalization proceeds- on whether restrictions on offshore borrowing by banks are relaxed first, as in the Republic of Korea, or measures limiting foreign investment in domestic secu- rities markets are eased instead, as in Malaysia. Implementing such distinctions will also require measures of the development of the information and contracting environment, because asymmetric information and poor contract enforcement are thought to favor banks over securities markets.30 Another set of studies builds on the observation that controls are dispropor- tionately used by countries with chronic macroeconomic imbalances (see, for example, Alesina and others 1994; Grilli and Milesi-Ferretti 1995; Wyplosz 1999; Garrett 1995, 1998, 2000). The motivation is presumably to limit capital flight and contain the threat from these imbalances for the stability of financial mar- kets.3" By now it will be clear that more than a few studies advancing such con- clusions have identification problems. Whereas countries suffering from chronic 29. Edwards (2001) is an exception in this regard, as noted above. 30. The argument being that banks are in the business of internalizing transactions that cannot take place at arm's length due to such market imperfections (Baskin and Miranti 1997). 31. Along with the seigniorage-related rationale reviewed previously. 356 THE WORLD BANK ECONOMIC REVIEW, VOL. J5, NO. 3 macroeconomic imbalances are more likely to resort to controls, governments and central banks enjoying the additional policy autonomy that controls confer may indulge in more expansionary policies. That few studies have addressed this identification problem may reflect the difficulty of finding plausible instruments for the endogenous variables. One response by those concerned with the impact of controls on the public finances has been to move from the budget balance to its components (the ex- penditure and tax sides and different categories of taxes and spending), where the causality running from controls to budgetary outcomes is presumably easier to identify. Garrett and Mitchell (2000) find that public spending is lower when the capital account is open, which they interpret as capital mobility applying fiscal discipline.32 Garrett (2000) finds that this effect is specific to the exchange rate regime: Governments come under less pressure to limit spending when the ex- change rate is allowed to float, but the combination of fixed rates and an open capital account has a strong disciplining effect. A particular mystery is the impact of capital account liberalization on taxes on profits and other returns to capital. The idea that capital account liberaliza- tion, which increases the effective elasticity of supply of capital, should put down- ward pressure on the rate of capital taxation is one of the most fundamental corollaries of the theory of public finance. But the evidence to this effect is sur- prisingly weak. Quinn (1997), Swank (1998), Garrett (2000), and Garrett and Mitchell (2000) find that rates of capital taxation are unchanged or even higher in countries with open capital accounts. Because most countries with open capi- tal accounts are relatively high-income, it may simply be that they have large public sectors (by Wagner's Law) and high tax rates. But Quinn, Swank, Garrett, and others go to considerable lengths to control for income and other country characteristics that may independently influence the level of capital taxation, and none of their extensions makes this finding go away. Clearly, this is a puzzle requiring further study. Finally, a number of researchers, motivated by the association of short-term foreign debt with crises and, in particular, by the perception that debt runs played a role in many episodes of serious turbulence in emerging markets in recent years, have asked whether controls can be used to lengthen the maturity structure of foreign obligations.33 Using data for a cross-section of countries, Montiel and Reinhart (1999) find that controls reduce the share of portfolio and short-term capital flows in total inflows, while increasing the share of foreign direct invest- 32. Quinn (1997) reports a positive association between public spending and capital account liber- alization but concludes that the correlation is not robust. 33. On the association of short-term debt with crises, see Rodrik and Velasco (1999). Readers whose sensitivities will have been heightened by the preceding discussion to the causality problems arising in other contexts will not be surprised that the same issue arises here. Rather than short-term debt causing crises, in other words, it has been argued that anticipations of crises leads to a shortening of the matu- rity structure of the debt. Eichengreen 357 ment and leaving the overall volume of capital inflows unchanged. This general- izes the findings of detailed studies for Chile, many of which conclude that its holding period tax on capital inflows reduced the volume of short-term inflows but in a way that was fully compensated for by increased long-term flows. (In other words, only the maturity structure and not the level of the flows was af- fected by these controls.34) Controls like Chile's, with the potential to reduce the risk of currency and financial crises, have their advocates in the scholarly and policymaking commu- nities. But is this advocacy justified? Answering this question requires determin- ing whether controls in fact reduce crisis risk. VI. CRISES AND LIBERALIZATION OF THE CAPITAL ACCOUNT The currency and banking crises of the 1990s did much to encourage the belief that capital account liberalization raises the risk of financial instability. The relax- ation of capital controls in Europe following the implementation of the Single European Act made the realignment of currencies participating in Europe's Ex- change Rate Mechanism more difficult, allowing competitiveness problems to build up, exposing governments and central banks to speculative pressures, and culmi- nating in the crisis of 1992 (Eichengreen and Wyplosz 1993). Capital account lib- eralization was implicated in Asia's crisis insofar as the selective opening of capi- tal accounts allowed banks to respond to the moral hazard created by government guarantees and to lever up their bets (Furman and Stiglitz 1998). China's suc- cess in insulating itself from this instability by the use of capital controls is widely seen as the exception that proves the rule.35 These assertions are controversial; scholars continue to debate the causes of the European and Asian crises and the role of capital flows. But it is curious, given the intensity of the debate, how few cross-country studies have sought to systematically weigh the evidence. One reason may be that problems of reverse causality are severe in this con- text. Countries experiencing financial turbulence may impose or reinforce con- trols, as did Malaysia following the outbreak of the Asian crisis. Or they may relax their controls in an effort to restore investor confidence, as did Thailand in January 1998 and the Republic of Korea several months later. The absence of controls may or may not heighten crisis risk, but the fact that crisis risk some- times prompts changes in the capital account regime makes it hard to distinguish cause from effect. 34. Studies that reach this conclusion include Soto (1997), De Gregorio and others (1998), and Valdes-Prieto and Soto (1998). 35. China restricted borrowing by Chinese entities, restricted portfolio outflows by Chinese citi- zens and inflows by foreigners, and banned futures trading in yuan. While cautioning that controls were probably only one of several factors making for the resiliency of the Chinese economy, Fernald and Babson (1999) conclude that without a freely accessible onshore futures market, speculation against the yuan would be difficult and that controls on outflows make it harder for Chinese investors to con- vert their yuan if they expect the currency to weaken. 358 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 In fact, contrary to the intuition described at the beginning of this section, the cross-country evidence generally suggests that controls heighten currency crisis risk. Glick and Huchinson (2000) combine data on the presence or absence of controls at the end of one year (from the IMF'S Exchange Arrangements and Exchange Restrictions) with data on the occurrence of currency crises in the next. In both bivariate and multivariate analyses they find a positive correlation be- tween capital controls and crises. Leblang (2000) uses the narrative accounts in Exchange Arrangements and Exchange Restrictions to code changes in capital controls monthly and finds that controls are associated with an increased prob- ability of currency crises. He also finds evidence that controls influence the like- lihood that governments and central banks will successfully defend the currency against attack. An interpretation, following Bertolini and Drazen (1997a, b) and Drazen (1997), is that countries maintaining or imposing controls send a negative sig- nal to the markets. Investors may suspect a country that resorts to controls of reluctance to commit to the rigorous course of fiscal and monetary treatment to maintain stability. They may worry that a government inclined to resort to con- trols will be particularly willing to compromise investor rights. The signal may incite investors to flee and, if the control regime is less than watertight, enable them to do just that. But have these researchers identified the direction of causality? If govern- ments impose controls in anticipation of looming financial problems, then tim- ing cannot identify the direction of causality.36 And, even more than in other contexts, there is reason to question the conclusions of an analysis that lumps all controls together. Controls of different intensity may not be equally effec- tive in containing threats to currency stability, and different types of controls and different forms of liberalization may have different implications for finan- cial stability. Liberalizing banks' access to offshore funding but not also per- mitting foreign access to domestic equity and bond markets may be more destabilizing than doing the reverse; it may cause foreign funds to flow in through the banking system, the weakest link in the financial chain. This is a common conclusion drawn from the crisis in Korea, which liberalized offshore bank funding before permitting foreign access to its securities markets. Even if inflow controls can reduce crisis risk by preventing banks and firms from be- coming excessively dependent on short-term foreign debt, outflow controls, 36. For example, Thailand introduced partial controls in May 1997, prior to its crisis, and later extended their coverage several times: in June, July, and September 1997 and January 1998. That Glick and Hutchinson relate the presence or absence of controls in one year to crises in the next may convince some readers that they have finessed this problem; surely controls imposed fully a year before a crisis are not the response of the authorities to subsequent difficulties. In fact, however, Glick and Hutchinson relate the presence or absence of controls at the end of year t to the presence or absence of a crisis any time in year t + 1, so that the time between the observation of controls and the occurrence of a crisis is at most a year-and in practice can be considerably less. Eichengreen 359 except of the most draconian sort, may be incapable of restraining capital flight if panic breaks out.37 In addition, different controls may send different signals. Inflow controls like Chile's can be justified as prudential measures-a way of reinforcing regulations designed to stabilize the financial system (Eichengreen and others 1998). They may then be perceived as a signal that the authorities take seriously their commitment to currency and banking stability. Outflow controls, in contrast, may suggest only that the authorities are desperate. Using data for 15 developing countries, Rossi (1999) finds that outflow controls heighten the risk of currency crises but that inflow controls reduce it. Outflow controls similarly are associated with an increased risk of banking crises, whereas inflow controls have no discernible effect. VII. FROM RESEARCH TO POLICY AND FROM POLICY TO RESEARCH Turning from research to policy, one finds greater consensus on the lessons of international experience. That the G7 countries all have open capital accounts is regarded as telling. For those who emphasize this fact, capital account liberal- ization is just another manifestation of the policies of financial deregulation that countries adopt as they develop economically and institutionally, and specifi- cally as they acquire the capacity to operate market-led financial systems. In other words, the relaxation of statutory restrictions on international financial trans- actions and the growth of cross-border financial flows reflect the same forces that encourage the removal of repressive domestic financial regulations and fa- cilitate reliance on domestic financial markets to guide the allocation of resources. The same arguments suggesting that domestic financial deepening and devel- opment enhance the efficiency of investment, facilitate experimentation with new technologies, and encourage growth and efficiency generally similarly support the presumption that international portfolio diversification and cross-border portfolio investment should encourage efficiency and growth. Capital account liberalization can be counterproductive, to be sure, if it takes place before severe policy-related distortions have been removed and before domestic markets, in- stitutions, and the administrative capacity of the prudential authorities have developed enough to generate confidence that foreign finance will be channeled in productive directions. This qualification may be too frequently neglected-as the unconditional advocacy of capital account liberalization heard in the mid- 1990s and the Asian crisis that quickly followed remind us to our chagrin-but this caveat, too, is now an integral part of the conventional wisdom. But if caveats like this one complicate the journey, the destination, from all appearances, remains the same. Officials and their advisers may differ on pre- cisely when and how to liberalize international financial transactions so as to 37. The fact that outflow controls tend to be the dominant variety in crisis-prone countries may therefore be another part of the explanation for why previous cross-country studies have found a posi- tive association between controls and crisis incidence. 360 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 best ensure that capital inflows are channeled in productive directions, in other words, but there is little support for refusing to liberalize or (Malaysia in 1998- 99 notwithstanding) for reversing previous liberalization measures. International financial liberalization, to paraphrase Marx, may be just another instance of the more developed economies showing their less developed counterparts an image of their future. Given the breadth of support commanded by this synthesis, the lack of em- pirical substantiation of its fundamental tenets is worrisome indeed. If the evi- dence is really not there, then it is high time to rethink the conventional wisdom. With these stakes, priority should be attached to research with immediate promise for solving the key empirical puzzles. Empiricists need to better distinguish among controls-between inflows and outflows and between transactions involving banks and those involving securities markets. They need to develop more infor- mative measures of the legal, contracting, and information environments that plausibly shape the effects of capital account liberalization. They also need to construct better indicators of the other policy initiatives with which capital ac- count liberalization is sequenced. These extensions can be undertaken in the context of existing macro-oriented cross-country research. Admittedly, operationalizing them presumes a not incon- siderable investment in data, constructed in ways that are consistent across coun- tries and over time. The call for more and better data is standard fare in surveys like this one; here, however, is a case where it warrants its place of prominence. But could it be that the problem is with the framework and not with the data and methods used to operationalize it? The literature on capital account liberalization has been written by macroeconomists, for macroeconomists, with an emphasis on the macroeconomics of growth and crisis. Perhaps the micro- economic level offers more definitive evidence of the effects of capital account policies. A growing body of firm-level evidence and analysis, surveyed by Karoly (1998) and Stulz (1999), suggests that this may be the case. Some examples il- lustrate the kinds of questions asked and answers found. Tandon (1994) shows a reduction in the required rate of return on equity for firms offering bonds on international markets. Smith and Sofianos (1997) show that firms listing abroad experience an increase in trading volume, consistent with the argument that financial integration leads to greater liquidity and hence a lower cost of capital. Lins and others (2000) show that firms from emerging markets listing in the United States are able to relax capital constraints-that is, the cash-flow sensi- tivity of their investment declines-but no such change is evident for firms from industrial countries, where capital constraints are presumably less. More remains to be learned from a microeconomic perspective. That said, answering the big questions like how growth and crises are affected by capital account liberalization will ultimately require mapping the findings of micro- economic studies back into the macroeconomic framework adopted by the re- searchers whose work was reviewed in this survey. Eichengreen 361 REFERENCES The word "processed" describes informally reproduced works that may not be commonly available through library systems. Aherane, Alan, William Griever, and Francis Warnock. 2000. "Information Costs and Home Bias: An Analysis of U.S. Holdings of Foreign Equities." International Finance Discussion Paper 691. Board of Governors of the Federal Reserve System. Washing- ton, D.C. Alesina, Alberto, and Guido Tabellini. 1989. "External Debt, Capital Flight, and Politi- cal Risk." Journal of International Economics 27(3-4):199-220. Alesina, Alberto, Vittorio Grilli, and Gian Maria Milesi-Ferretti. 1994. "The Political Economy of Capital Controls." In Leonardo Leiderman and Assaf Razin (eds.), Capital Mobility: The Impact on Consumption, Investment and Growth. Cambridge: Cam- bridge University Press. Ariyoshi, Akira, Karl Habermeier, Bernard Laurens, Inci Otker-Robe, Jorge Ivan Canales- Kriljenko, and Andrei Kirilenko. 2000. "Capital Controls: Country Experiences with Their Use and Liberalization." IMF Occasional Paper No. 190. Arteta, Carlos, Barry Eichengreen, and Charles Wyplosz. 2001. "When Does Capital Account Liberalization Help More Than It Hurts?" NBER Working Paper No. 8414, Cambridge, Mass. Bai, Chong-En, and Shang- Jin Wei. 2000. "Quality of Bureaucracy and Open Economy Macro Policies." NBER Working Paper 7766. National Bureau of Economic Research, Cambridge, Mass. Barro, Robert J. 1997. Determinants of Economic Growth: A Cross-Country Empirical Study. Cambridge, Mass. MIT Press. Baskin, Jonathan Barron, and Paul J. Miranti Jr. 1997. A History of Corporate Finance. Cambridge: Cambridge University Press. Bekaert, Geert. 1995. "Market Integration and Investment Barriers in Emerging Mar- kets." World Bank Economic Review 9(t):75-108. Bekaert, Geert, and Campbell R. Harvey. 1995. "Time-Varying World Market Integra- tion." Journal of Finance 50(2):403-44. Bernhard, William, and David Leblang. 1999. "Democratic Institutions and Exchange Rate Commitments." International Organization 53(1):71-97. Bertolini, Leonardo, and Allan Drazen. 1997a. "Capital Account Liberalization as a Signal." American Economic Review 87(1):138-54. .1997b. "When Liberal Policies Reflect Shocks, What Do We Learn?" Journal of International Economics 42(3-4):249-73. Bordo, Michael, and Barry Eichengreen. 1998. "Implications of the Great Depression for the Evolution of the International Monetary System. " In Michael Bordo, Claudia Goldin, and Eugene White (eds.), The Defining Moment: The Great Depression and the Ameri- can Economy in the Twentieth Century. Chicago: University of Chicago Press. Brecher, Richard. 1983. "Second-Best Policy for International Trade and Investment." Journal of International Economics 14(3-4):313-20. Brecher, Richard, and Carlos Diaz-Alejandro. 1977. "Tariffs, Foreign Capital and Immiserizing Growth." Journal of International Economics 7(3-4):317-22. . .. .. || 362 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 Cody, Brian J. 1990. "Exchange Controls, Political Risk and the Eurocurrency Market: New Evidence from Tests of Covered Interest Rate Parity." International Economic Journal 42(2):75-86. Cooper, Richard. 1999. "Should Capital Controls Be Banished?" Brookings Papers on Economic Activity 1. Washington, D.C.: Brookings Institution. Dailami, Mansoor. 2000. "Managing Risks of Global Financial Market Integration." In Charles Adams, Robert Litan, and Michael Pomerleano (eds.), Managing Financial and Corporate Distress. Washington, D.C.: Brookings Institution. De Gregorio, Jose, Sebastian Edwards, and Rodrigo Valdes. 1998. "Capital Controls in Chile: An Assessment." Paper presented to the Interamerican Seminar on Macroeco- nomics, Rio de Janeiro, December. Dooley, Michael P. 1996. "A Survey of the Academic Literature on Controls over In- ternational Capital Transactions." IMF Staff Papers 43(4):639-87. Dooley, Michael P., and Peter Isard. 1980. "Capital Controls, Political Risk, and Devia- tions from Interest-Rate Parity." Journal of Political Economy 88(2):370-84. Drazen, Allan. 1989. "Monetary Policy, Capital Controls and Seigniorage in an Open Economy." In Marcello de Cecco and Alberto Giovannini (eds.) A European Central Bank? Cambridge: Cambridge University Press. .1997. "Policy Signaling in the Open Economy: A Re-Examination." NBER Work- ing Paper 5892, Cambridge, Mass. Edison, Hali, and Carmen M. Reinhart. 1999. "Stopping Hot Money." Board of Gover- nors of the Federal Reserve System, Washington, D.C., and University of Maryland, College Park. Processed. Edwards, Sebastian. 1999. "How Effective Are Capital Controls?" Journal of Economic Perspectives 13(4):65-84. . 2001. "Capital Flows and Economic Performance: Are Emerging Economies Different?" NBER Working Paper 8076, Cambridge, Mass. Eichengreen, Barry, and Andrew Rose. 1999. "Contagious Currency Crises: Channels of Conveyance." In Takatoshi Ito and Anne Krueger (eds.), Changes in Exchange Rates in Rapidly Developing Economies. Chicago: University of Chicago Press. Eichengreen, Barry, and Charles Wyplosz. 1993. "The Unstable EMS." Brookings Pa- pers on Economic Activity. Washington, D.C.: Brookings Institution. Eichengreen, Barry, Michael Musa, Giovanni Dell' Ariccia, Enrica Detragiache, Gian Maria Mihesi-Ferretti, and Andrew Tweedie. 1998. "Capital Account Liberalization: Theo- retical and Practical Aspects." IMF Occasional Paper 172, Washington, D.C. Epstein, Gerald, and Juliet Schor. 1992. "Structural Determinants and Economic Effects of Capital Controls in OECD Countries." In Tariq Banuri and Juliet Schor (eds.), Financial Openness and National Autonomy. Oxford: Clarendon Press. Fernald, John G., and Oliver D. Babson. 1999. "Why Has China Survived the Asian Crisis So Well? What Risks Remain?" International Finance Discussion Paper 633. Board of Governors of the Federal Reserve System, Washington, D.C. Fisher, Irving. 1930. The Theory of Interest. New York: Macmillan. Frankel, Jeffrey A., and Alan T. MacArthur. 1988. "Political vs. Currency Premia in International Real Interest Differentials: A Study of Forward Rates for 24 Countries." European Economic Review 32(5):1083-1114. Eichengreen 363 Frenkel, Jacob A., and Assaf Razin. 1996. Fiscal Policies and Growth in the World Economy. 3d ed. Cambridge, Mass.: MIT Press. Furman, Jason, and Joseph Stiglitz. 1998. "Economic Crises: Evidence and Insights from East Asia." Brookings Papers on Economic Activity 2. Washington, D.C.: Brookings Institution. Garrett, Geoffrey. 1995. "Capital Mobility, Trade, and the Domestic Politics of Economic Policy." International Organization 49(4):657-87. . 1998. Partisan Politics in the Global Economy. Cambridge: Cambridge Univer- sity Press. . 2000. "Capital Mobility, Exchange Rates and Fiscal Policy in the Global Economy." Review of International Political Economy 7(1):153-70. Garrett, Geoffrey, and Deborah Mitchell. 2000. "Globalization, Government Spending, and Taxation in the OECD." European Journal of Political Research (forthcoming). Garrett, Geoffrey, Alexandra Guisinger, and Jason P. Sorens. 2000. "The Political Economy of Capital Account Liberalization." Department of Political Science, Yale University, February. Processed. Giavazzi, Francesco, and Marco Pagano. 1988. "Capital Controls and the European Monetary System." In Donald E. Fair and Christian de Boissieu (eds.), International Monetary and Financial Integration. Dordrecht: Martinus Nijhoff. Glick, Reuven, and Michael Hutchinson. 2000. "Stopping 'Hot Money' or Signaling Bad Policy? Capital Controls and the Onset of Currency Crises." Federal Reserve Bank of San Francisco and University of California, Santa Cruz. Processed. Grilli, Vittorio, and Gian Maria Milesi-Ferretti. 1995. "Economic Effects and Structural Determinants of Capital Controls." IMF Staff Papers 42(3):517-51. Harvey, Campbell R. 1995. "The Risk Exposure of Emerging Equity Markets." World Bank Economic Review 9(1):19-50. Holmes, Mark J., and Yangru Wu. 1997. "Capital Controls and Covered Interest Parity in the EU: Evidence from a Panel-Data Unit Root Test." Weltwirtschftliches Archiv 133:76-89. IMF (International Monetary Fund). Various years. Exchange Arrangements and Ex- change Restrictions. Washington, D.C. Ito, Takatoshi. 1983. "Capital Controls and Covered Interest Parity." NBER Working Paper 1187, Cambridge, Mass. Johnston, R. Barry, and Natalia T. Tamirisa. (1998). "Why Do Countries Use Capital Controls?" IMF Working Paper no. WP/98/181, Washington, D.C. Johnston, R. Barry, Mark Swinburne, Alexander Kyei, Bernard Laurens, David Mitchem, Inci Otker, Susana Sosa, and Natalia Tamirisa. 1999. Exchange Rate Arrangements and Currency Convertibility: Developments and Issues. Washington, D.C.: Interna- tional Monetary Fund. Karolyi, Andrew. 1998. "Why Do Companies List Their Shares Abroad? A Survey of the Evidence and Its Managerial Implications." Salomon Brothers Monograph 1. New York University, Graduate School of Business. Klein, Michael, and Giovanni Olivei. 1999. "Capital Account Liberalization, Financial Depth, and Economic Growth." NBER Working Paper 7384, Cambridge, Mass. Kraay, Aart. 1998. "In Search of the Macroeconomic Effects of Capital Account Liber- alization." World Bank, Development Economics Research Group, Washington, D.C. Processed. 364 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 Leblang, David A. 1997. "Domestic and Systemic Determinants of Capital Controls in the Developed and Developing World." International Studies Quarterly 41(3):435-54. . 1999. "Domestic Political Institutions and Exchange Rate Commitments in the Developing World." International Studies Quarterly 43(4):599-620. Levine, Ross. 1997. "Financial Development and Economic Growth: Views and Agenda." Journal of Economic Literature 35(2):688-726. . 1999. "International Financial Liberalization and Economic Development." University of Virginia, Department of Economics, Charlottesville, VA. Processed. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Levine, Ross, and Sara Zervos. 1998. "Capital Control Liberalization and Stock Mar- ket Development." World Development 26(7):1169-83. Lins, Karl, Deon Strickland, and Mark Zenner. 2000. "Do Non-U.S. Firms Issue Equity on U.S. Stock Exchanges to Relax Capital Constraints?" Fisher College of Business, Ohio State University, Columbus. Processed Marston, Richard C. 1993. "Interest Differentials under Bretton Woods and the Post- Bretton Woods Float: The Effects of Capital Controls and Exchange Risk." In Michael Bordo and Barry Eichengreen (eds.), A Retrospective on the Bretton Woods System. Chicago: University of Chicago Press. . 1995. International Financial Integration: A Study of Interest Differentials be- tween the Major Industrial Countries. New York: Cambridge University Press. Milesi-Ferretti, Gian Maria. 1998. "Why Capital Controls? Theory and Evidence." In Sylvester Eijffinger and Harry Huizinga (eds.), Positive Political Economy: Theory and Evidence. Cambridge: Cambridge University Press. Montiel, Peter, and Carmen Reinhart. 1999. "Do Capital Controls and Macroeconomic Policies Influence the Volume and Composition of Capital Flows? Evidence from the 1990s." Journal of International Money and Finance 18(4):619-35. Neely, Christopher J. 1999. "An Introduction to Capital Controls." Federal Reserve Bank of St. Louis Review 81(6):13-30. Obstfeld, Maurice. 1993. "The Adjustment Mechanism." In Michael Bordo and Barry Eichengreen (eds.), A Retrospective on the Bretton Woods System. Chicago: Univer- sity of Chicago Press. Quinn, Dennis P. 1997. "The Correlates of Changes in International Financial Regula- tion." American Political Science Review 91(3):531-51. . 2000. "Democracy and International Financial Liberalization." McDonough School of Business, Georgetown University, July. Processed. Quinn, Dennis P., and Carla Inclan. 1997. "The Origins of Financial Openness: A Study of Current and Capital Account Liberalization." American Journal of Political Sci- ence 41(3):771-813. Reinhart, Carmen, and R. Todd Smith. 1998. "Too Much of a Good Thing: The Macro- economic Effects of Taxing Capital Inflows." In Reuven Glick (ed.), Managing Capital Flows and Exchange Rates: Perspectives from the Pacific Basin. Cambridge: Cam- bridge University Press. Rodriguez, Francisco, and Dani Rodrik. 1999. "Trade Policy and Economic Growth: A Skeptic's Guide to the Cross-National Evidence." NBER Working Paper 7081, Cam- bridge, Mass. Eichengreen 365 Rodrik, Dani. 1998. "Who Needs Capital-Account Convertibility?" In Peter Kenen (ed.), Should the iMF Pursue Capital Account Convertibility? Essays in International Finance no. 207, Princeton: Princeton University Press. Rodrik, Dani, and Andres Velasco. 1999. "Short Term Capital Flows." NBER Working Paper 7364, Cambridge, Mass. Rossi, Marco. 1999. "Financial Fragility and Economic Performance in Developing Countries: Do Capital Controls, Prudential Regulation, and Supervision Matter?" IMF Working Paper WP/99/66, Washington, D.C. Sachs, Jeffrey. 1981. "The Current Account and Macroeconomic Adjustment in the 1970s." Brookings Papers on Economic Activity 1. Washington, D.C.: Brookings Institution. Sachs, Jeffrey, and Andrew Warner. 1995. "Economic Reform and the Process of Global Integration." Brookings Papers on Economic Activity 1. Washington, D.C.: Brookings Institution. Simmons, Beth, and Zachary Elkins. 2000. "Globalization and Policy Diffusion: Explain- ing Three Decades of Liberalization." University of California, Department of Politi- cal Science, Berkeley. Processed. Smith, Katherine, and George Sofianos. 1997. "The Impact of a NYSE Listing on Global Trading of Non-U.S. Stocks." Working Paper 97-02, New York Stock Exchange. Soto, Claudio. 1997. "Controles a los Movimientos de Capitales: Evaluacion Empirica del Caso Chileno." Central Bank of Chile, Santiago. Processed. Stiglitz, Joseph. 2000. "Capital Market Liberalization, Economic Growth and Instabil- ity." World Development 28(6):1075-86. Stulz, Rene M. 1995. "Globalization and the Cost of Capital: The Case of Nestle." Eu- ropean Financial Mangment 8(1):30-38. Swank, Duane. 1998. "Funding the Welfare State: Globalization and the Taxation of Business in Advanced Market Economies." Political Studies 46(4):671-92. Tandon, Koshore. 1994. "External Financing in Emerging Economies: An Analysis of Market Responses." World Bank,Washington, D.C. Processed. Valdes-Prieto, Salvador, and Marcelo Soto. 1998. "The Effectiveness of Capital Con- trols: Theory and Evidence from Chile." Empirica 25(2):133-64. Williamson, John, and Molly Mahar. 1998. "A Survey of Financial Liberalization." Essays in International Finance 211. Princeton University, Department of Economics, Inter- national Finance Section, Princeton, N.J. Wong, Clement Yuk Pang. 1997. "Black Market Exchange Rates and Capital Mobility in Asian Economies." Contemporary Economic Policy 15(1):21-36. Wyplosz, Charles. 1999. "Financial Restraints and Liberalization in Postwar Europe." Graduate Institute of International Studies, Geneva. Processed. i THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 367-391 Where Has All the Education Gone? Lant Pritchett Cross-national data show no association between increases in human capital attribut- able to the rising educational attainment of the labor force and the rate of growth of output per worker. This implies that the association of educational capital growth with conventional measures of total factor production is large, strongly statistically signifi- cant, and negative. These are "on average" results, derived from imposing a constant coefficient. However, the development impact of education varied widely across countries and has fallen short of expectations for three possible reasons. First, the institutional/ governance environment could have been sufficiently perverse that the accumulation of educational capital lowered economic growth. Second, marginal returns to educa- tion could have fallen rapidly as the supply of educated labor expanded while demand remained stagnant. Third, educational quality could have been so low that years of schooling created no human capital. The extent and mix of these three phenomena vary from country to country in explaining the actual economic impact of education, or the lack thereof. To be a successful pirate one needs to know a great deal about naval war- fare, the trade routes of commercial shipping; the armament, rigging, and crew size of potential victims; and the market for booty. To be a successful chemical manufacturer in early twentieth century United States required knowledge of chemistry, potential uses of chemicals in different intermediate and final products, markets, and problems of large scale organization. If the basic institutional framework makes income redistribution (piracy) the preferred economic opportunity, we can expect a very different devel- opment of knowledge and skills than a productivity-increasing (a twenti- eth century chemical manufacturer) economic opportunity would entail. The incentives that are built into the institutional framework play the de- cisive role in shaping the kinds of skills and knowledge that pay off. -Douglass North (1990) Lant Pritchett is an economist on leave from the World Bank and is currently at the Kennedy School of Government. His e-mail address is lantpritchett@harvard.edu. The author is grateful for discussions with and comments from Harold Alderman, Jere Behrman, Bill Easterly, Deon Filmer, Mark Gersovitz, Dani Rodrik, Harry Patrinos, Marlaine Lockheed, Peter Lanjouw, David Lindauer, Michael Walton, Martin Ravallion, Jonathan Temple, Alan Krueger, Kevin Murphy, Paul Glewwe, Mead Over, and participants in the Johns Hopkins development seminar. ( 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 367 368 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 People with more education have higher wages. This is probably the second (after Engel's law) most well-established fact in economics. It would seem to follow naturally that if more individuals are educated, average income should rise; if there are positive externalities to education, average income should rise by even more than the sum of the individual effects. The belief that expanding education promotes economic growth has been a fundamental tenet of development strategy for at least 40 years.' The post-World War II period has seen a rapid, histori- cally unprecedented expansion in educational enrollments. Since 1960, average developing country (gross) primary enrollments have risen from 66 to 100 per- cent, and (gross) secondary enrollments from 14 to 40 percent. How has this experiment in massive educational expansion turned out? Is there now strong evidence of the growth-promoting externalities to education? This is an area where growth theory and empirical estimates are potentially impor- tant. Positive externalities should mean that the impact of education on aggre- gate output is greater than the aggregation of the individual impacts. To test for externalities, we need macroeconomic and microeconomic models of education's impacts that are consistent. The augmented Solow model is just such a model because it predicts that the "no externality" impact of education should be the share of educational capital in factor income. This impact can be estimated from microeconomic evidence on the wage increments to capital. Within the augmented Solow model, the estimated growth impact of education is consistently less than would be expected (rather than more) from the individual impacts. The cross- national data suggests negative externalities and present something of a "micro- macro" paradox. The path to resolving this paradox begins with an acknowledgment that the impact of education on growth has not been the same in all countries (Temple 1999). I discuss three possibilities for reconciling the macro and micro evidence and explaining the differences across countries. The first possibility is North's (1990) metaphorical piracy: Education has raised productivity, and there has been sufficient demand for this more productive educated labor to maintain or increase private returns, but the demand for educated labor comes, at least in part, from individually remunerative yet socially wasteful or counterproductive activities. In this case, the relative wage of each individual could rise with educa- tion (producing the micro evidence), even while increases in average education would cause aggregate output to stagnate or fall (producing the macro evidence). The second possibility is that expansion of the supply of educated labor when demand is stagnant could cause the rate of return to education to fall rapidly. In this case, the average Mincer returns (Mincer 1974) estimated in the 1960s and 1. The idea that either the "new" growth theory or the "neoclassical revival" has "discovered" the importance of human capital is belied by even a casual reading of Kuznets (1960), Lewis (1956), or Dennison (1967). Gunnar Myrdal's (I 975) Asian Drama, written mostly in the late 1950s, already treats the importance of human capital along with physical capital in development as the conventional wisdom. Pritchett 369 1970s overstated the actual marginal contribution to output from educational expansion in those instances where the demand for educated labor did not ex- pand rapidly enough. Third, schooling quality may be so low that it does not raise cognitive skills or productivity. This could even be consistent with higher private wages if education serves as a signal to employers of some positive char- acteristics, such as ambition or innate ability. I. EXPANSION OF EDUCATION AND GROWTH-ACCOUNTING REGRESSIONS The first approach is to do what we would do if we did not know it was not going to work. That is, we will take the standard production function specifica- tions of growth at the macroeconomic level, build aggregate measures of educa- tion capital from microeconomic data on education and its returns, and then examine the relationship between them. How Much Should Education Matter? The Augmented Solow Model Mankiw and others (1992) extend the Solow aggregate production function framework to include educational capital: (1) Y, = A(t) * K *X " Hi'l7 " L '* ' assuming constant returns to scale (Ok + 0h + als = 1), normalizing by the labor force, and taking natural logs to produce a linear equation in levels. But this "linear in log levels" specification can also be expressed in rates of growth. Because estimation in levels raises numerous problems (to which I return below), I focus on the relationship among percent per annum growth of output per worker (y = din (YIL)Idt), growth of physical capital per worker, and educational capital per worker:2 (2) =a+ ak k+ Xhh In the context of this model, a is the growth rate of the growth-accounting re- sidual-and I will reluctantly follow convention and call this total factor pro- ductivity (TFP), even though it is not (Pritchett 2000a). (3) TFP= - a - k-a *kh. The extended Solow approach facilitates simple nonregression-based estimates of how much the expansion of educational capital "ought" to matter. Because 2. Growth for each variable is calculated as the logarithmic least squares growth rate over the entire period for which the data are available. This makes the estimates of growth rates much less sensitive to the particular endpoints than if changes from the beginning period to the end period were calculated. This means the time period over which I calculate the growth rate does not always correspond exactly to the time period for the education data, but because both are per annum growth rates, this difference does not matter much. 370 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 the weights in the aggregate Cobb-Douglas production function represent the fac- tor shares of national income, the coefficient on educational capital in a growth- accounting regression ought to be equal to the share of educational capital in gross domestic product (GDP) that can be estimated based on microeconomic data. With constant returns to scale, labor share is one minus the physical capital share. A physical capital share of around 0.4 is somewhat high, but is consistent with a variety of evidence-the estimates from national accounts and from re- gression parameters-and with capital output ratios (if the capital-output ratio, KIY, is 2.5 and the rate of return to capital is 16 percent, then the share of capi- tal, rKIY, is 40 percent). This implies a labor share of 0.6. How much of the labor share is due to human (or educational) capital? One simple way of estimating the share of the wage bill attributable to human capi- tal is to use the ratio of the unskilled-or "zero human capital"-wage, w0, to the average wage, w: (4) HUMAN CAPITAL SHARE (FROM WAGES) = 1-Wo / W. A calculation based on the distribution of wages in Latin America estimates a human capital share of wages of between 50 and 75 percent. Mankiw, Romet, and Weil (1992) use the historical ratio of average to minimum wages in the United States to estimate that half of wages are due to human capital.3 Either of these calculations suggests a human capital coefficient (Uh) of at least 0.3. Another approach to estimating the educational capital share is to assume a wage increment to education (taking the micro evidence discussed below at face value), and then use data on the fraction of the labor force in each educational attainment category to derive the educational capital share. Table 1 shows the results of two calculations. The top half shows the fraction of the labor force in various educational attainment categories in various regions. One can calculate the share of the wage bill due to educational attainment by assuming a wage premium for each attainment category and applying equation 5: K W(Wj -wo) *ai (5) EDUCATIONAL CAPITAL SHARE OF WAGES BILL = L wL where i represents each of the seven educational attainment categories and i are the shares of the labor force in each educational attainment category. 3. Using data on the distribution of workers' earnings (World Bank 1993a), we take the ratio of the average wages up to the 90th percentile (to exclude the effect of the very long tails of the earnings dis- tribution) to the wage of those workers in either the 20th or 30th percentile (to proxy for the wage of a person with '"no" human capital). The estimates of human capital share of the wage bill are 62 and 47 percent, respectively. If the top 10th percentile is included (and I take the ratio of average wages to the 20th or 30th percentile), the estimates of human capital share are even higher-74 and 63 percent, respectively. Although these are considerably higher that other estimates, they are estimates of all hu- man capital, not just educational capital. In the United States, the ratio of the average to the minimum wage (taken as a proxy for the "unskilled" wage) has hovered around 2. TABLE 1. Share of Educational Capital in Wage Bill Wage premia by Share of work force by educational attainment, 1985 educational attainment (percent except where noted) under assumption set: Developing Sub-Saharan Latin American South A B countries Africa and Caribbean Asia OECD No schooling 1.00 1.00 49.7 48.1 22.4 69.0 3.3 Some primary 1.40 1.56 21.3 33.2 43.4 8.9 19.4 Primary complete 1.97 2.44 10.1 8.5 13.2 4.8 18.3 Some secondary 2.77 3.42 8.7 7.7 8.4 8.8 20.7 Secondary 3.90 4.81 5.9 1.6 5.5 5.3 20.1 Some tertiary 5.47 6.06 1.4 0.2 2.5 0.9 7.7 Tertiary 7.69 7.63 3.0 0.8 4.6 2.3 10.5 Average years of schooling 3.56 2.67 4.47 2.81 8.88 Calculated share of wtvge bill due to educational capital across regions under each asstmption .. '.. ui Assumption set A 36 26 43 30 62 (wage increment is constant at 10 percent) Assumption set B 49 38 56 42 73 (wage increments are: primary 16 percent, secondary 12 percent, tertiary 8 percent) Source: Data on educational attainment by region from Barro and Lee (1993). 372 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Under assumption set A (constant wage increment of 10 percent per year of schooling), the educational share of the wage bill varies across regions, from 26.3 percent in Sub-Saharan Africa (SSA) to 62.1 percent in the Organisation for Eco- nomic Cooperation and Development (OECD); and it is 36.4 percent for devel- oping countries as an aggregate. Under assumption set B (wage increments are proportionately higher for a year of primary than for a year of secondary, and higher for secondary than for tertiary, at 16, 12, and 8 percent, respectively), the share of educational capital in the total wage bill averages 49 percent-al- most exactly half-for all developing economies, varying from 38 percent in SSA to 73 percent in OECD. Both methods suggest that the educational capital share of the wage bill should be between 0.35 and 0.7. Hence the growth-accounting regression coefficient on educational capital (cxb) ought to be between 0.21 and 0.42-with 0.3 in the middle of the range. Data and Specification for Physical and Educational Capital Using two recently created cross-national time-series data sets, I create estimates of the growth rate of per worker educational capital. The two data sets use dif- ferent methods to estimate the educational attainment of the labor force. Barro and Lee (1993) estimate the educational attainment of the population age 25 and above using census or labor force data where available and create a full panel of five yearly observations over the period 1960-85 for a large number of coun- tries by filling in the missing data using enrollment rates. Nehru and others (1995) use a perpetual inventory method to cumulate enrollment rates into annual esti- mates of the stock of schooling of the labor force-aged population, creating annual observations for 1960-87. From these estimates of years of schooling of the labor force, I create a mea- sure of educational capital from the microeconomic specification of earnings used by Mincer (1974). I assume the natural log of the wage (or more generally, earn- ings per hour) is a linear function of the years of schooling: (6) ln(wN) = ln(wO) + r * N, where WN is the wage with N years of schooling, N is the number of years of schooling, and r is the wage increment to a year's schooling. The value of the stock of educational capital at any given time, t, can then be defined as the dis- counted value of the wage premia due to education: (7) HK(t) = St t * (WN- U'o), where w0 is the wage of labor with no education. Substituting the formula for the educational wage premia (equation 6) into the definition of the stock (equa- tion 7) and taking the natural log gives equation 8 for the log of the stock of educational capital, we get T (8) ln(HK(t)) = ln(Z 6t) + ln(wo(t)) + ln(erN - 1). t=O Pritchett 373 Therefore, the proportional rate of growth of the stock of educational capital is approximately4 (9) hk(t) _ dln(expRN(t) - 1) / dt. Based on existing surveys of the large number of micro studies,5 I calculate the growth of educational capital using equation 9, the data on years of schooling from either Barro and Lee (1993) or Nehru and others (1995), and an assumed r of 10 percent constant across all years of schooling.6 In addition to the measures of educational capital, I use two series created by a perpetual inventory accumulation of investment and an initial estimate of the "capital" stock, based on an estimate of the initial capital-output ratio (King and Levine 1994; Nehru and Dhareshwar 1993). As I have argued elsewhere, series constructed in this way cannot be treated as estimates of the physical capital stock relevant to the production function, because there is no underlying theoretical or empirical justification for doing so when governments are the main investors. Hence, they should be called by a purely descriptive acronym: cudie (cumulated, Depreciated Investment Effort) (Pritchett 2000a). The two CUDIE series are highly correlated and give similar results, with the principal difference being that King and Levine (1994) use investment data from the Penn World Tables, Mark 5 (PWT5; Summers and Heston 1991), while Nehru and Dhareshewar (1993) use invest- inent data from the World Bank. The dependent variable is growth of GDP per worker from PWT5. This is con- ceptually more appropriate in growth-accounting regressions than GDP per per- son or per labor force-aged person (but, as argued below, the findings are robust).7 4. There are two reasons this formula is only an approximation. First, the discount factor is as- sumed constant and hence is factored out in the time rate of change. It does depend on the average age of the labor force (because the discount is only until time T, retirement), which certainly varies system- aticallv across countries, but I am assuming that changes in this quantity over time are small. The sec- ond, potentially more serious problem is that I dropped out the growth rate of ln(w()(,)(-the evolution of the unskilled wage term. This means my growth rate of human capital is really that component of the growth of human capital due to changes in years of schooling. For instance, Mulligan and Sala-i- Martin (1997) estimate a human capital stock in which increases in unskilled wages reduce human capital; this is technically correct, but certainly counterintuitive. 5. A survey by Psacharopoulos (1993) shows wage increments by region: SSA 13.4 percent; Asia 9.6 percent; Europe, Middle East, and North Africa 8.2 percent; Latin America 12.4 percent; OECD 6.8 percent; and an unweighted average of 10.1. In any case, the cross-national differences in the growth rate of educational capital are very robust to variations in the value of r. 6. One confusion (among many) in this literature is hetween the wage increment and the rate of return to education. The often-repeated assertion that "returns are higher to primary schooling" (as reported, for example, by Psacharopolous [19931) seems true not because the increment to wages from a year of primary school is higher than for other levels, but because the opportunity cost of a year of primary schooling is much lower. This is due to the fact that the typical forgone wage attributed to a primary-age unschooled child is very low (Bennell 1996). What is relevant to growth accounting is the increment to wages, not the cost-inclusive return. 7. This output variahle does raise one problem. My estimates of human capital are hased on esti- mates of the educational capital of the labor force-aged population, whereas my output is output per estimated labor force (although not corrected for unemployment), so that systematic differences in the 374 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Regression Results for Growth and TFP The results for estimating the growth-accounting equation (2) for the entire sample of countries8 are reported in column 1 of Table 2. The partial scatter plot is displayed as figure 1. The estimates for cumulated physical investment (CUDIE) correspond reasonably well to national accounts-based estimates of the capital share (although 0.52 is somewhat on the high side) and are strongly sig- nificant (t = 12.8). Very much on the other hand, the estimate of the impact of growth in educational capital on growth of per worker GDP is negative (-0.049) and insignificant (t = 1.07). Adding the initial level of GDP per worker (column 2) has no impact on the negative estimates of the effect of education (-0.038). Columns 8 and 9 of table 2 show the results of regressing TFP growth on the growth of physical CUDIE and educational capital. In column 9, the assumed fac- tor shares used in creating TFP are 0.4 (physical) and 0.3 (educational). The growth of educational capital shows a large, statistically very significant (t = 6.91) and negative (-0.338) effect on TFP growth. In column 10, I make the educa- tional capital share as small as is consistent with growth accounting by assuming the physical capital share is on the high side (0.5) and the share of educational capital in the wage bill is on the low side (0.33), so that the educational capital share is as low as it can reasonably be (1/2 x 1/3 = 0.167). It is still the case that educational capital accumulation is strongly statistically significant and nega- tively related to TFP growth. Of course, except for fixing the physical capital share, this TFP regression is equivalent to a t-test that finds the estimated human capi- tal share equal to 0.167. Using the results of column 1, this hypothesis is easily rejected (t = [-0.049 - 0.167] / 0.046 = 4.72). These TFP results are a simple arithmetic trick, but this trick is useful because it changes a typically uninteresting "failure to reject" to a convincing rejection of an interesting and policy-relevant hypothesis. The findings are not a "low- powered" failure to reject zero-they are a "high-powered" failure to reject, because although the data do not reject zero, they do in fact reject a wide range of interesting hypotheses-including the hypothesis that the growth impact is as large as the microeconomic data would suggest. After all, the primary reason to use aggregate data to estimate the impact of schooling is to find out whether the evolution of the labor force versus the labor force-aged population (say, through differential female labor force participation) could affect the results. The question of whether or not changes in female labor force participation (cross-national level differences would not affect the results) are an important part of the story is beyond the scope of this article. With the currently available gender-disaggregated data, this is an active research question, with some arguing that female education is more important for growth, and others arguing that it is less important, than male schooling. 8. Four countries have been dropped from all regressions because of obvious data problems: Ku- wait, because PWT5 GDP data are bizarre; Gahon, because labor force data (larger than the population) are clearly wrong; Ireland, because the Nehru and others (1995) data report an average of 16 years of schooling (immigration has distorted these numbers); and Norway, because Barro and Lee (1993) re- port an impossible increase of 5 years in schooling over a period of 5 years. TABLE 2. Growth-Accounting Regressions of GDP per Worker Growth with Educational Capital and CUDIE per Worker Growth Per annum growth of GDP per worker (GDPPW) Level GDPPW TFP as defined in text 1 2 3 4 5 6 7 8 9 IV OLS OLS OLS (w/ Nehru (sample of OLS (factor (factor OLS OLS OLS and others IV countries (on level in shares, shares, (entire (with initial (on just iv [1995] educ. (w/ similar with test 1985, whole OK = 0.4, oK = 0.5, Dependent variable sample) GDPPW) sample) capital data) country) scores) sample) c11 = 0.3) (iH = 0.167) Growth of education -0.049 -0.038 -0.091 -0.120 -0.088 0.058 0.136 -0.338 -0.205 capital per worker, (1.07) (0.795) (1.61) (1.42) (0.593) (0.229) (1.97) (6.91) (4.19) Growth of 0.524 0.526 0.458 0.460 0.527 0.592 0.612 0.126 0.026 CUDIE per worker, (12.8) (12.8) (10.19) (10.18) (12.42) (6.78) (14.88) (3.08) (0.651) In (initial GDP 0.0009 0.0009 0.0009 per worker) (0.625) (0.625) (0.0625) Test score 0.014 (normalized, mean = 1) (1.31) Test score * EK -0.485 (1.27) Number of 91 91 70 70 77 25 96 91 91 countries R2 0.653 0.655 0.611 - - 0.71 0.909 0.419 0.205 Note: t-statistics in parentheses. aExcept in column 7, which uses levels. Source: Author's calculations. 376 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 FIGURE 1. Partial Scatterplot of Growth of GDP per Worker and Educational Capital per worker, Conditioned on CUDIE per Worker 0.1 *0.08 o I a, 0.06 1 ~~~~~~~~~~~~~~+ 0.04- 0. 9) 0*02 a-0.03 -0*2 0.01 001 0.02 0.03 E * * *+ -0.04 -0 06 GDP per worker growthICUDIE per worker growth impact is higher (or lower) than expected from the microeconomic data, and hence to provide some indication of the presence (or absence) of externalities. But to speak to this question, growth regressions using aggregate data must demon- strate not only that the educational capital coefficient is not zero but that it is higher than the value expected, given the microeconomic evidence applied to the same growth model. This is a seemingly modest standard, but one that has never been met. Before proposing explanations of this apparent micro-macro paradox of negative externalities, I first show that this result is robust to sample, data, and technique and that it is not the result of "pure" measurement error or failure to account for school quality.9 The estimated coefficient is not the result of a peculiar sample or a few ex- treme or atypical observations. To ensure robustness against outliers, individual observations identified as influential were sequentially deleted up to 10 per- cent of the sample size, with no qualitative change in results.10 The negative 9. I do not show that the results are robust to the introduction of other covariates (Levine and Renelt 1992). This is hecause I am interested in growth accounting within a specific growth model that takes a production function approach. Thus there is no scope to introduce other covariates arbitrarily, as in the "reduced form" literature. 10. An observation is identified as influential based on the difference in the estimate with and with- out the observation included (Belsley and others 1980). Temple (1999) working on a different data set, Pritcbett 377 coefficient on schooling growth persists if (a) only developing countries are used, (b) all observations from SSA are excluded, or (c) regional dummies are included. The results are also robust to variations in the data used for education, CUDIE, or GDP. All the regressions in table 2 were also estimated using Nehru and oth- ers' (1995) estimates of educational capital, and the educational capital coeffi- cient estimates are similar: consistently negative."1 Changing the data on growth and using World Bank growth rates of GDP in constant prices in local currency instead of the PWT5 GDP data gives similar results. Using growth of GDP per per- son or per labor force-aged population produces an even larger negative esti- mate for education. Relaxing the assumption of constant returns to scale does not alter the negative estimate on educational capital. Using weighted least squares with either (log of) population, GDP per capita, or total GDP because the weights also gives nearly identical results. The finding using level-on-level specifications of the augmented Solow equa- tion in table 2, column 7 shows a coefficient of 0.13 (t = 1.97)-which contin- ues to reject Ho : ath = 0.3, t = 2.37. However, there are good reasons to believe level-on-level coefficients will be biased upward. If this educational capital co- efficient is biased upward by as much as the CUDIE results appear to be (by about 0.1), then the small negative coefficient in the growth-on-growth regres- sions are consistent with the small positive coefficients in the level-on-level regressions. Although both sets of educational attainment data have been roundly criti- cized on a number of legitimate grounds (Behrman and Rosenzweig 1993, 1994), I use two different instruments to show that this particular result on educational capital is not the result of pure measurement error in the estimates of years of schooling. Using the growth of Nehru and others' (1995) educational capital as an instrument for Barro and Lee's (1993) educational capital (the correlation of the two series' growth rates is.67), the coefficient becomes slightly more nega- tive: -0.12 (column 4 of table 1) versus -0.091 for ordinary least squares (OLS) in the same sample (column 3). In addition, I also match each country with a similar country, usually picking the geographically closest neighbor, based on the idea that educational capital growth rates in similar countries are likely to be correlated (the actual correlation was p = 0.316), whereas the pure mea- surement error in similar countries' reported enrollment and attainment rates is plausibly uncorrelated (and certainly less than perfectly correlated). This IV coefficient in table 2, column 5 is also negative (-0.088). Correcting for pure finds that there is substantial parameter homogeneity and that a significant fraction the sample must be dropped to recover a significant positive coefficient on education. I take this to indicate not a lack of robustness but substantial parameter heterogeneity-a point to which I return below. 11. These are reported in Pritchett (1996), an earlier version of this article. In that paper, the basic ordinary least squares regression using the other data set was = c +o.5oik(,,4, - O.lO4(20T7h, N = 79, r2 = 0.557 (t-statistics in parentheses). 378 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 measurement error makes the estimates more negative (which is to be expected, as measurement error produces attenuation bias), and hence only deepens the puzzle.12 Recently, Krueger and Lindahl (2000) have criticized Benhabib and Spiegel (1994), based on the latter's older estimates of educational stocks. Krueger and Lindahl (2000) claim that Benhabib and Spiegel's (1994) findings are not ro- bust to pure measurement error. However, this criticism is not relevant to the present article (for which much of the work was done several years before the Krueger and Lindahl paper) for three reasons. First, I use newer data sets, not the Kyriacou (1991) data used in Benhabib and Spiegel (1994). Second, my use of iv to correct to measurement error is exactly the same conceptual approach as Krueger and Lindahl's (2000), and I do not find that iv reverses any findings. Third, Krueger and Lindahl (2000) focus particularly on the measurement error of growth rates over short (e.g., five-year) periods, and argue, rightly, that mea- surement error is a larger concern in differenced data. In any case, the results in Krueger and Lindahl's (2000) table 5, column 5, which are the most similar to those presented here (in that they control for physical capital with an uncon- strained coefficient and instrument for the education variable), find an empiri- cally modest but statistically insignificant impact of schooling (t = 0.41). The bound of two standard deviations on Krueger and Lindahl's estimate of the aggregate equivalent of the Mincerian rate of return ranges from negative 44 percent to positive 67 percent. The major difference between our results is that I use the percentage rate of growth in the value of educational capital (which is essentially a logarithmic specification; see equations 6-9), whereas they use ab- solute change in the years of schooling. A different, deeper notion of measurement error is that while the years of schooling are correctly measured, the true problem is that years of schooling do not reflect learning. However, while differences in educational quality can ac- count for heterogeneity in the impact of schooling, it should not explain a low average impact. In fact, due to the "general underlying positive covariance be- tween quantity and quality of schooling" (Schultz 1988), one would expect that excluding quality would bias the estimated return upward, as more schooling is accumulated where quality is high."l For lack of quality adjustment to explain the results or quantities in the aggregate, there would have to be a very strong inverse cross-national relationship between quality and the expansion of quan- tity-a relationship for which there is no evidence. The quality of schooling across countries is impossible to measure without internationally comparable test examinations of comparable groups of students, 12. Using instruments for physical CUDIE and educational capital simultaneously, to correct for measurement error in both has very little impact on the estimates of educational capital. 13. For instance, Behrman and Birdsall (1983) have shown, for Brazil, that not controlling for school quality leads to overestimating the impact of years of schooling by a factor of two. Pritchett 379 and these, unfortunately, exist for very few countries.14 Hanushek and Kim (1995) use test score data to show that test score performance has a positive and statis- tically significant coefficient as an independent variable in a growth regression.15 However, in this case the interest is in the impact of an increase in educational capital, and the expected functional form when schooling quality matters would be an interactive effect: the impact of an additional unit of educational capital is higher when the quality of schooling is higher. I estimate this functional form using a single observation on test scores for each of the 25 countries used by Hanushek and Kim (1995), normalized to a mean of one, to interact with the growth of the educational capital stock. As shown in table 2, column 7, while the estimated impact of education is higher with higher quality (although the interactive coefficient is statistically insignificant), it is still the case that, evalu- ated at the average level of quality (test score = 1), the education impact is sub- stantially less than zero (0.06 - 0.48 = -0.42). This suggests that, as expected, the lack of control for quality causes an upward bias, so the negative estimates that do not control for quality are not negative enough. Relationship to Other Empirical Results on Schooling As surprising as these negative results may seem, they are similar to what other researchers have found when they examined the relationship between education and growth using either growth-on-growth or level-on-level regressions. Benhabib and Spiegel (1994) and Spiegel (1994) use a standard growth-accounting frame- work that includes initial per capita income and estimates of years of schooling from Kyriacou (1990), and find a negative coefficient on growth of years of schooling.16 Lau and others (1991) estimate the effects of education by level of schooling (primary versus secondary) for five regions and find that primary edu- cation has an estimated negative effect in Africa and Middle East North Africa, 14. One possible way out of the lack of quality measures is to use proxies for quality. However, there is no particular reason to believe that physical indicators (such as teacher-to-pupil ratio or re- sources expended per student) will adequately proxy quality, and many reasons to believe they will not. Hanushek and Kim (1995) explore the connections between these indicators and test scores to extrapolate a quality when it is not available, but with little success. Because schooling is typically publicly provided, there is no reason to believe that dollars spent will be closely associated with output (that is, one cannot apply the usual theory about the relationship between inputs and outputs derived from production theory of profit maximizers). There is a huge amount of literature on the impact on achieve- ment of various physical and financial measures of resources expended per student, with generally ambiguous results (see, for example, Filmer and Pritchett [19991). 15. However, one could easily suspect that any variable-for example, test scores-on which coun- tries such as Singapore (the highest, 72.1) and Hong Kong (71.8) do well and countries such as Nigeria (38.9) and Mozambique (27.9) do poorly, might well be capturing more in a growth regression than just labor force quality. 16. Spiegel (1994) shows that the finding of a negative effect of educational growth is robust to the inclusion of a wide variety of ancillary variables (e.g., dummies for SSA and Latin America, size of the middle class, political instability, share of machinery investment, inward orientation), and to the inclu- sion of samples. 380 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 insignificant effects in South Asia and Latin America, and positive and signifi- cant effects only in East Asia. Jovanovich and others (1992) use annual data on a different set of capital stocks and Nehru and others' (1995) education data and find negative coefficients on education in a non-OECD sample. Behrman (1987) and Dasgupta and Weale (1992) find that changes in adult literacy are not significantly correlated with changes in output. The World Bank's World Development Report on labor also reports the lack of a (partial) correlation between growth and education expansion (World Bank 1995, figure 2.4). Newer studies using panels to allow for country-specific effects consistently find nega- tive signs on schooling variables (Islam 1995; Caselli and others 1996; Hoeffler 1999).17 Some very early studies used enrollment rates in growth regressions (Barro 1991; Mankiw and others 1992), but this approach had and has two deep prob- lems. First, especially in Mankiw and others (1992), secondary enrollment rates alone were used-but without any clear or compelling reasoning as to why both primary and tertiary enrollment rates should have been excluded. Second, en- rollment rates are a terrible proxy for growth in years of schooling."8 The as- sumption that current (or average) enrollment rates adequately proxy a country's steady-state stock is true only if enrollment rates are constant over time across countries-but this is contradicted by the massive recent expansion of school- ing in developing countries (Schultz 1988). The correlation between the growth of educational capital and secondary enrollment rates is -0.41. This is because the growth of educational attainment depends not on the current enrollment rate but on the difference in the enrollment rate between the cohort leaving the labor force and the cohort entering the labor force.'9 17. However, these studies are susceptible to the Krueger and Lindahl (2000) critique about exac- erbation of measurement error in short (five-vear) panels. Moreover, the dynamic properties of the educational series, which tend to have little time series variation within countries, make it difficult to identify impacts of education in any case (Pritchett 2000b). 18. This does raise the question of why, if they are not a valid proxy for accumulation of schooling, initial secondary enrollment rates are a reasonably robust correlate of subsequent growth rates. My conjecture is the nature of "conditional convergence" regressions-that is, both the initial level of in- come and initial secondary enrollment rate are on the right-hand side of the equation with growth on the left-hand side. It is not unreasonable to assume that high secondary enrollment rates conditional on income level may signal something good about a country's growth prospects (e.g., the government's provision of good schools might mean it does other things well, the country has a substantial middle class, or people anticipate the country will do well; but it could also mean income is temporarily low), quite independent of the impact via accumulation of educational capital. 19. Comparing Korea and Great Britain provides a simple illustration. Korea's secondary enroll- ment rate in 1960 was 27 percent, while Great Britain's was 66 percent. But the level of schooling of Great Britain's labor force in 1960 was 7.7 years, and the level of Korea's was 3.2 years. Subsequently, Great Britain's enrollment rate increased to 83 percent by 1975 and then remained relatively constant, whereas Korea's enrollment rate increased from 27 to 87 percent by 1983. Given these differences in initial stocks and the large changes in enrollment rates, Korea's average years of schooling expanded massively from 3.2 to 7.8 by 1985, but Great Britain's expanded only modestly from 7.7 to 8.6, even though Great Britain's enrollment rate was higher than Korea's for most of the period. Pritchett 381 Another section of the literature uses the initial level of the stock of education to explain growth of output per capita. Benhabib and Spiegel (1994) show that if the initial level of education is added to a growth-accounting regression, the initial level of education is positive, whereas the mildly negative impact of the growth of educational capital persists. This finding of a level effect is actually much more puzzling than is generally acknowledged, as the spillover effects of knowledge that might be captured by an effect of the level of education in the endogenous growth literature should be in addition to rather than instead of the usual direct productivity effects. Finding only a spillover impact is grossly in- consistent with the microdata: If the entire return to education at the aggregate level is spillover effects, then why is the wage premium observed at the individual level? Moreover, a regression with growth rates on the left-hand side and level of education on the right-hand side is either misspecified or a complicated way of imposing parameter restrictions. The obvious fact that growth rates are station- ary (without drift) while the stock of education is nonstationary and secularly increasing implies there cannot be a stable relationship between the growth of output and the level of education (Jones 1995).20 Growth regressions that in- clude initial levels of both education and output are only justified if education levels (nonstationary) are cointegrated with levels of income (nonstationary). But in that case, this specification still begs the original question, because to fully implement the error correction model one must still estimate the cointegrating relationship. II. WHY (AND WHERE) HAS SCHOOLING CONTRIBUTED TO GROWTH? So there is an apparent micro-macro contradiction. The microeconomic evidence is commonly (if naively) taken to mean that substantial wage increments from additional schooling are nearly universal and that additional schooling will lead to growth. The macroeconomic data in an entirely standard growth accounting model suggest that education has not uniformly had the growth impact the microeconomic data would suggest. The obvious resolution is that the impact of education has varied widely across countries (Temple 1999).2i The question 20. Ben-David and Papell (1 994) use Angus Maddison's historical data and find that growth rates are stationary after allowing for one structural break. This criticism applies to all endogenous growth models that make growth rates a function of any nonstationary variable (such as the magnitude of re- search and development or the stock of knowledge) while growth rates are stationary (Jones 1995). 21. Not surprisingly, the data, when unconstrained, do not say that schooling has contributed to output to exactly the same degree in Korea, Zaire, Paraguay, and Hungary. Parameter homogeneity does not change the fact that the unconstrained estimates are well below the expected level, on average. Hence, there must be a number of countries for which education appears to have had less than the expected "standard augmented Solow model no externality" growth impact if wage increments were on the order of 10 percent. 382 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 is why. In those countries that have had substantial improvements in educational attainment of the labor force yet still face declining real wages and slow eco- nomic growth, the question must be asked: Where has all the education gone? I do not propose a single answer, but put forward three possibilities that could account for the results: * The newly created educational capital has gone into piracy; that is, privately remunerative but socially unproductive activities. * There has been slow growth in the demand for educated labor, so the sup- ply of educational capital has outstripped demand and returns to school- ing have declined rapidly. . The education system has failed, so a year of schooling provides few (or no) skills. These possibilities are not mutually exclusive; all are likely to be present to varying degrees in every country. I will discuss each briefly, with some indication of the evidence that would support or contradict each approach in a given country. (For a more extensive discussion, see Pritchett 1996.) Are Cognitive Skills Applied to Socially Produictive Activities? Rent seeking in our [African] economies is not a more or less important phenomenon, as would be the case in most economies. It is the center- piece of our economies. It is what defines and characterizes our economic life. -Meles Zenawi, Prime Minister of Ethiopia, September 5, 2000 One way to reconcile high wage increments to schooling with a small (and dif- ferential) macroeconomic impact of education is to argue that social and private rates of return to education diverge due to distortions in the economy. North's (1990) powerful metaphorical comparison of piracy and chemical manufactur- ing in the introduction suggests the problem. Rent seeking and directly unpro- ductive activities can be privately remunerative but socially dysfunctional and reduce overall growth. If the improved cognitive skills acquired through educa- tion are applied to piracy, this could explain both the micro returns (rich pirates) and small macro impact (poor economies). Several pieces of evidence suggest this is at least part of the puzzle. In many developing economies, the public sector has accounted for a large share of the expansion of wage employment in the 1960s and 1970s (table 3). This is not to equate government or the magnitude or growth of government employment with the magnitude of rent seeking. Nor am I saying that the ex- pansion of education in government is necessarily unproductive. On the con- trary, the most successful of developing countries have had strong and active governments and highly educated civil servants hired through a very competi- Pritchett 383 TABLE 3. Share of Wage Employment Growth Accounted for by Public Sector Growth in Selected Developing Countries Average growth of wage employment Public sector (percent per annum) (percentage of Country Period Public Private Total total increase) Public sector employment growth positive, private wage employment growtb zero or less Ghana 1960-78 3.4 -5.9 -0.6 Zambia 1966-80 7.2 -6.2 0.9 418 Tanzania 1962-76 6.1 -3.8 1.6 190 Peru 1970-84 6.1 -0.6 1.1 140 Egypt 1966-76 2.5 -0.5 2.2 103 Brazil 1973-83 1.4 0 0.3 100 Public sector employment growth more than half of total wage employment growth Sri Lanka 1971-83 8 0.9 3.9 87 India 1960-80 4.2 2.1 3.2 71 Kenya 1963-81 6.4 2 3.7 67 Public sector growth faster, but less than half of total wage employment growth Panama 1963-82 7.5 1.8 2.7 45 Costa Rica 1973-83 7.6 2.8 3.5 34 Thailand 1963-83 6.3 5.5 5.7 33 Venezuela 1967-82 5.1 3.4 3.7 27 Unweighted mean 5.5 0.3 2.4 Source: Derived from Gelb and others (1991), table 1. tive process (World Bank 1994).22 The question is not whether educated labor flows into the government, but why the government hires educated workers (ac- tual need versus employment guarantee) and what they do once they are in the government (productive versus unproductive or rent-seeking activities). Murphy and others (1991) present a simple model of the allocation of talent in which, if returns to ability are the greatest in rent seeking, then economic growth is inhibited by drawing the most talented people away from productive sectors into rent seeking. Anecdotal evidence that rent seeking attracts educated labor abounds. There is the possibly apocryphal (but nevertheless instructive) story of one West African nation with an employment guarantee for all univer- sity graduates. In a year when the exchange rate was heavily overvalued (and hence, there was a large premium on evading import controls), 60 percent of university graduates in all fields designated the customs service as their prefer- ence for government employment. Explicit or implicit government guarantees of employment for the educated have been common and have led to large distortions in the labor market. In Egypt, 22. Wade (1990) asserts that college graduates are as likely to enter government service in Korea and Taiwan as in African economies. 384 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 government employment guarantees led to notoriously overstaffed enterprises and bureaucracies. In 1998, the government and public enterprises employed 70 percent of all university graduates and 63 percent of those with education at the intermediate level and above (Assaad 1997). Gersovitz and Paxson (1995) calculate that in 1986-88 in Cote d'Ivoire, 50 percent of all workers between age 25 and 55 that had completed even one grade of postprimary education worked in the public sector. Gelb and others (1991) built a dynamic general equilibrium model in which government responds to political pressures from potentially unemployed educated job seekers and becomes the employer of last resort for educated labor force entrants. They show that when both employment pressures are strong and the government is highly responsive to those pressures, the employment of surplus educated labor in the public sector can reduce growth of output per worker by as much as 2 percent a year (from a base case growth of 2.5 percent). Stagnant Demand for Educated Labor A second explanation for smaller growth returns from expanding education than from wage increments might suggest that the marginal return to adding an addi- tional year of schooling economy-wide can be dramatically different from the average returns estimated from a cross-sectional Mincer (1974) regression on wage employment at a single point in time. Depending on the shift in the de- mand for and supply of educated labor, and on the mechanism of labor market adjustment, the wage premia can rise or fall. In different countries there is evi- dence of rising, falling, stable, or vacillating returns to schooling. Mincer coeffi- cients in the United States have increased (at the median) from 0.063 to 0.096 (Buchinksy 1994). The returns to schooling in Egypt fell significantly in the 1980s (Assaad 1997). Funkhouser (1994) shows quite stable Mincer returns for five Central American countries over several years. Montenegro (1995) shows that the Mincer coefficient in Chile varied from 0.095 to 0.167 between 1960 and 1993-falling, then rising, then falling again over this period. There are two basic stories to explain the demand for educated labor (includ- ing by the self-employed). One is that education conveys skills that make labor more productive. In this case, the demand for educated labor will rise when the skill intensity of the economy rises. The second is that more educated individu- als are able to adapt more quickly to disequilibrium (Schultz 1975). In this case, the demand for educated labor will rise when there are greater gains to adapting to disequilibrium. These two stories of the source of returns to education are difficult to distinguish empirically, but both suggest that growth of educational capital would have a larger impact on output growth when policies are in place to ensure either that sectoral shifts lead to higher skill intensity, or that the cre- ation or assimilation of knowledge is higher (even within the same sector), or both. One can easily imagine a scenario in which a Mincer regression based on wage employment shows very high returns and yet, in the absence of expansion of the Pritchett 385 wage employment sector (assume, for now, this is the skill-intensive sector), these returns could fall very fast so that the marginal return to additional education is very small. Table 4 (adapted from Bennell 1996) shows that in many African countries, expansion of the number of newly educated laborers has often exceeded expansion of wage employment by more than an order of magnitude. Under these conditions, the returns to education could fall very fast. Even without sectoral shifts, the returns to education would be higher where technological progress was rapid, thus requiring constant adaptation to techno- logically induced disequilibrium. Schultz (1975) argues that in a technologically stagnant agricultural environment the production gains from education would be zero, as even the least educated could eventually reach the efficient allocation of factors. In this case, only when new technologies and inputs are available does education pay off, and then only in transition to the new equilibrium. Foster and Rosenzweig (1996) find that the return to five years of primary schooling versus no schooling in the average Indian district studied was a modest 11 per- cent (an average increase of 446 rupees in farm profits). However, returns to schooling were higher in those districts where agricultural conditions were in- trinsically conducive to the adoption of Green Revolution technologies 'which they proxy by the exogenous increase in average farm profits). In a district where farm profits are one standard deviation above the average due to technical TABLE 4. Growth of Enrollments and of Wage Employment in Selected Sub-Saharan African Countries Change in Change in Ratio, expansion Wage employment enrollments wage employment of enrollment to as percentage of Country (thousands) (thousands) wage employment total labor force Enrollment growth positive, wage employment falling Zambia 446 -4.3 13.1 C6te d'lvoire 323 -7.7 9 Enrollmtent growth exceeds wage employment growth by an order of magnitude Sierra Leone 257 8.9 29 4.9 Uganda 225 13.2 17 4.7 Ghana 1312 80 16 3.8 Burkina Faso 351 35.4 10 3.8 Lesotho 142 14.9 10 5.4 Enrollment growth higher by factor of 4 Senegal 180 45.4 4.0 5.5 Kenya 1709 436 3.9 14.1 Malawi 546 143 3.8 13.7 Rough equality of enrollment and wage sector growth Botswana 157 122 1.3 50.4 Zimbabwe 135 111.1 1.2 36.6 Note: Growth rates of enrollments and wage sector growth are calculated from the beginning date of the studv estimating Mlincerian return to 1990 (or the most recent data). Source: Bennell (1996), table 5. 386 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 progress, the return to primary schooling was 32 percent-almost three times higher. However, the converse of high returns with rapid progress is that the estimated returns to schooling were negative in those districts in which progress was low.23 Rosenzweig (1996) uses data across districts of India to show the pitfalls in cross- sectional regressions when technological progress varies exogenously. In a cross- section of Indian districts, education is correlated with economic growth. But Rosenzweig (1996) shows that once varying exogenous technical progress is in- troduced, this technological progress explains both the higher economic growth and higher returns to education (and the higher returns lead to greater expansion in the amount of education). Although schooling has paid off handsomely where the Green Revolution brought technological advances, education has not been an important determinant of local growth in technologically stable areas, and the apparent impact of education from cross-district regressions disappears. If some countries' policies are more conducive to the creation or assimilation of technical progress or to development patterns that are skill intensive, then one could expect that the output impact of a given expansion of schooling could be higher or lower. For instance, many argue that more open trade regimes in de- veloping countries would facilitate catch-up and lead to more rapid technical progress, and that the returns to education would depend, at least in part, on complementary policies such as reasonable outward orientation (World Bank 1994). Did Schooling Create Skills? Direct evidence from internationally comparable examinations shows substan- tial variation in schooling quality-and that children in some developing coun- tries lag far behind OECD and East Asian countries. Low quality of schooling is consistent with the macroeconomic evidence and is obviously consistent with the household evidence of little or no wage increment from additional schooling. However, in countries where there is a reliably demonstrated microeconomic return but no apparent macroeconomic impact of schooling, a more sophisti- cated "low quality" explanation of the paradox is needed. A signaling model of the labor market is consistent with schooling that creates few skills and yet sub- stantial observed wage impacts. If workers with high initial (or innate) ability have an easier time staying in school than workers with low initial ability, em- ployers will pay more for schooled workers even though schooling has no im- pact on skills or productivity (Spence 1976). 23. When average district farm profits were more than two-thirds of a standard deviation below the country average, the point estimate of education was negative. This explanation of the interaction of demand and supply for education due to different rates of technological progress might suggest the reason education appears not to have paid off in such places as SSA. Several recent studies have found very little return to education in farming in Africa (Gurgand 1995; Joliffe 1995). If there has been little exogenous change in the technical production functions appropriate for more educated farmers to adopt, it is because Green Revolution innovations were not appropriate for African agriculture. Pritcbett 387 There is mixed evidence of a signaling function of schooling. "Sheepskin" effects-in which the completion of a level of education has substantially more labor market impact than would be expected from the skills acquired at that level-are common and can be taken as indication of schooling as a filter. How- ever, there are at least three sources of evidence against an argument that the entire wage impact of schooling is signaling. First, several studies from develop- ing countries with data on ability, skills, and schooling suggest that signaling effects are small (Knight and Sabot [1990], containing data on Kenya and Tan- zania; Glewwe [1991] with data on Ghana; and Alderman and others [1996], with data on Pakistan). Second, the limited evidence of the impact of education on the productivity of farmers (Jamison and Lau 1982) or the self-employed is harder to explain by signaling. Finally, even for SSA countries, where one might suspect low educational quality, evidence from the Demographic and Health Surveys shows a 24 percent lower child mortality rate where women have a pri- mary education as opposed to no education (Hobcraft 1993). This is hard to explain if schooling has no impact on knowledge.24 III. CONCLUSION In the decades since 1960, nearly all developing economies have seen educational attainment grow rapidly. The cross-national data show, however, that on aver- age, education contributed much less to growth than would have been expected in the standard augmented Solow model. Where did all the education go? There are three possible explanations for the differences across countries in the impact of schooling on growth in economic output: * In some countries, schooling has created cognitive skills and these skills have been in demand, but to do the wrong thing. In other countries, the institu- tional environment has been sufficiently bad that the bulk of newly acquired skills has been devoted to privately remunerative but socially wasteful or counterproductive activities-that is, the expansion of schooling has meant the country just has better-educated pirates. . The rate of growth of demand for educated labor (due in part to different sectoral shifts, in part to policies, in part to exogenous differences in tech- nological progress) has varied widely across countries, so countries with the same initial individual returns and equal subsequent expansions in the supply of educated labor could have seen the marginal returns to educa- tion fall dramatically, stay constant, or rise. . In some countries, schooling has been enormously effective in transmitting knowledge and skills, while in others it has been essentially worthless and has created no skills. 24. But it is not impossible to explain, as the education-health linkage might be entirely the result of intergenerationally correlated endowments or preferences. 388 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 No two countries follow exactly the same pattern, and each explanation con- tributes a different amount to explaining the overall impact of schooling on growth in different countries. None of the arguments in this article suggest that governments should invest less in basic schooling, for many reasons. For one thing, most (if not all) societies believe that at least basic education is a merit good, so that its provision is not and need not be justified on economic grounds at all-a position with which I strongly agree. To deny a child an education because of a small expected economic growth impact would be a moral travesty. In addition, schooling has a large number of direct beneficial effects beyond raising economic output, such as lower child mortality. All education can raise cognitive skills, with everything that implies. The implication, therefore, of a poor past aggregate payoff from increased cognitive skills in a perverse policy environment is not "don't educate," but rather "reform now so that investments (past and present) in cognitive skills will pay off." REFERENCES The word "processed" describes informally reproduced works that may not be commonly available through library systems. Alderman, Harold, Jere Behrman, David Ross, and Richard Sabot. 1996. "The Returns to Endogenous Human Capital in Pakistan's Rural Labor Market." Economica 5(2):167-85. Assaad, Ragui. 1997. "The Effect of Public Sector Hiring and Compensation Policies on the Egyptian Labor Market." World Bank Economic Review 11:85-118. Barro, Robert. 1991. "Economic Growth in a Cross Section of Countries." Quarterly Journal of Economics 106(1):407-43. Barro, Robert, and Jong-Wa Lee. 1993. "International Comparisons of the Educational Attainment." Journal of Monetary Economics 32(3):363-94. Behrman,Jere. 1987. "Schooling in Developing Countries: Which Countries Are the Over and Underachievers and What Is the Schooling Impact?" Economics of Education Review 6(2):111-27. Behrman, Jere, and Nancy Birdsall. 1983. "The Quality of Schooling: Quantity Alone Is Misleading." World Bank Reprint Series (International) 311:928-46. Behrman, Jere, and Mark Rosenzweig. 1993. "Adult Schooling Stocks: Comparisons among Aggregate Data Series." Processed. .1994. " Caveat Emptor: Cross-Country Data on Education and the Labor Force." Journal of Development Economics 44:147-71. Belsley, David A., Edwin Kuh, and Roy E. Welsch. 1980. Regression Diagnostics: Iden- tifying Influential Data and Sources of Colinearity. New York: J. Wiley. Ben-David, Dan, and David H. Pappell. 1994. The Great Wars, the Great Crash, and the Unit Root Hypothesis: Some New Evidence about an Old Stylized Fact. Working Paper 4752. National Bureau of Economic Research, Cambridge, MA. Benhabib, Jess, and Mark Spiegel. 1994. "Role of Human Capital in Economic Devel- opment: Evidence from Aggregate Cross-Country Data." Journal of Monetary Eco- nomics 34:143-73. Pritchett 389 Bennell, Paul. 1996. "Rates of Return to Education: Does the Conventional Pattern Pre- vail in Sub-Saharan Africa?" World Development 24:183-99. Buchinsky, Moshe. 1994. "Changes in the U.S. Wage Structure 1963-1987: Applica- tion of Quantile Regression." Econometrica 62(2):405-58. Caselli, Francesco, Gerardo Esquivel, and Fernando Lefort. 1996. "Reopening the Con- vergence Debate: A New Look at Cross-Country Growth Empirics." Journal of Eco- nomic Growth 1(3):363-89. Dasgupta, Partha, and Martin Weale. 1992. "On Measuring the Quality of Life." World Development 20(1):119-31. Denison, Edward Fulton.1967. Why Growth Rates Differ. Brookings Institution: Wash- ington, D.C. Filmer, Deon, and Lant Pritchett. 1999. "What Education Production Functions Really Show: A Positive Theory of Education Expenditures." Economics of Education Re- viewv 18(2):223-39. Foster, Andrew D., and Mark R. Rosenzweig. 1996. "Technical Change and Human Capital Returns and Investments: Evidence from the Green Revolution." American Economic Review 86(4):931-53. Funkhouser, Edward. 1994. "The Returns to Education in Central America." Univer- sity of California, Santa Barbara, December. Processed. Gelb, Alan, J. K. Knight, and Richard H. Sabot. 1991. "Public Sector Employment, Rent Seeking, and Economic Growth." Economic Journal 101:1186-99. Gersovitz, Mark, and Christina Paxson. 1995. "The Revenues and Expenditures of Af- rican Governments: Modalities and Consequences." World Bank, Washington, D.C. Processed. Glewwe, Paul. 1991. Schooling, Skills, and the Returns to Government Investment in Education. Living Standards Measurement Survey Working Paper 76. World Bank, Washington D.C. Gurgand, Marc. 1993. "Les effets de l'education sur la production agricole: application a la C6te Ivoire." Revue de l'economie du developpement 10(4):37-54. Hanushek, Eric A., and Dongwook Kim. 1995. "Schooling, Labor Force Quality, and Economic Growth." National Bureau of Economic Research Working Paper 5399. Cambridge, MA. Hobcraft, J. 1993. "Women's Education, Child Welfare, and Child Survival: A Review of the Evidence." Health Transition Review: The Cultural, Social, and Behavioral Determinants of Health 3(2):159-75. Hoeffler, Anke Elisabeth. 1997. "The Augmented Solow Model and the African Growth Debate." Oxford University, Center for Study of African Economics. Processed. Islam, Nazrul. 1995. "Growth Empirics: A Panel Data Approach." Quarterly Jolurnal of Economics 110(4):1127-70. Jamison, Dean, and Lawrence Lau. 1982. Farmer Education and Farm Efficiency. Johns Hopkins University Press: Baltimore, MD. Jolliffe, Dean. 1998. "Skills, Schooling, and Household Income in Ghana." World Bank Economic Revieu' 12(1):81-108. Jones, Chad. 1995. "R&D-Based Models of Economic Growth." Journal of Political Economy 103(August):759-84. Joyanavich, Boyan, Saul Lach, and Victor Lavy. 1992. "Growth and Human Capital's Role as an Investment in Cost Reduction." New York University, New York. Processed. 390 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 King, Robert, and Ross Levine. 1994. "Capital Fundamentalism, Economic Development, and Economic Growth." Carnegie-Rochester Series on Public Policy 40:259-300. Knight, J. B., and Richard Sabot. 1990. Education, Productivity and Inequality: The East African Natural Experiment. New York: Oxford University Press for the World Bank. Krueger, Alan, and Mikael Lindahl. 2000. "Education for Growth: Why and for Whom?" National Bureau of Economic Research Working Paper 7591, Cambridge, MA. Kuznets, Simon. 1966. Modern Economic Growth: Rate, Structure, and Spread. New Haven: Yale University Press. Kyriacou, George. 1991. "Level and Growth Effects of Human Capital: A Cross-Coun- try Study of the Convergence Hypothesis." New York University Economic Research Report: 91-26, p. 25. Lau, Lawrence, Dean Jamison, and Larry Louat. 1991. "Impact of Education by Region." World Bank, Washington, D.C. Processed. Levine, Ross, and David Renelt. 1992. "Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Lewis, W. Arthur. 1955. The Theory of Economic Growth. London: Allen and Unwin. Mankiw, Gregory, David Romer, and David Weil. 1992. "A Contribution to the Empirics of Economic Growth." Quarterly Journal of Economics 107:407-37. Mincer, Jacob. 1974. Schooling, Experience, and Earnings. New York: Columbia Uni- versity Press. Montenegro, Claudio. 1995. "The Structure of Wages in Chile 1960-1993: An Applica- tion of Quantile Regression." World Bank, Washington, D.C. Processed. Mulligan, Casey B., and Xavier Sala-i-Martin. 1997. "A Labor-Income-Based Measure of the Value of Human Capital: An Application to the States of the United States." Japan and the World Economy 9(2):159-91. Murphy, Kevin M., Andrei Shleifer, and Robert Vishny. 1991. "Allocation of Talent: Implications for Growth." Quarterly Journal of Economics 106(2):503-30. Myrdal, Gunnar. 1968. Asian Drama: An Inquiry into the Poverty of Nations. New York: Twentieth Century Fund. Nehru, Vikram, and Ashok Dhareshwar. 1993. "A New Database on Physical Capital Stock: Sources, Methodology and Results." Revista de Analysis Economico 8(1):37- 59. Nehru, Vikram, Eric Swanson, and Ashutosk Dubey. 1995. "New Database on Human Capital Stock in Developing and Industrial Countries: Sources, Methodology and Results." Journal of Development Economics 46:379-401. North, Douglass. 1990. Institutions, Institutional Change and Economic Performance. Cambridge: Cambridge University Press. Organization for Economic Cooperation and Development (OECD) 1993. National Ac- counts. Paris: OECD. Pritchett, Lant. 1996. "Where Has All the Education Gone?" World Bank Policy Re- search Working Paper #1581. World Bank, Washington, D.C. .2000a. "The Tyranny of Concepts: CUDIE (Cumulated, Depreciated, Investment Effort) Is Not Capital." Journal of Economic Growth 5:361-84. .2000b. "Understanding Patterns of Economic Growth: Searching for Hills among Plateaus, Mountains, and Plains." World Bank Economic Review 14(2):221-50. Pritchett 391 Psacharopoulos, George. 1993. "Returns to Investment in Education: A Global Update." Policy Research Paper 1067. World Bank, Washington, D.C. Rosenzweig, Mark. 1996. "Schooling, Economic Growth, and Aggregate Data." Depart- ment of Economics, University of Pennsylvania. Processed. Schultz, T. P. 1988. "Education Investments and Returns." In H. Chenery and T. N. Srinivasan, eds., Handbook of Development Economics, vol. I. Amsterdam: Elsevier Science. Schultz, Theodore W. 1975. "The Value of the Ability to Deal with Disequilibria." Journal of Economic Literature 13(3):827-46. Spence, Michael. 1976. "Competition in Salaries, Credentials, and Signaling Prerequi- sites for Jobs." Quarterly Journal of Economics 90(1):51-74. Spiegel, Mark. 1994. "Determinants of Long-Run Productivity Growth: A Selective Sur- vey with Some New Empirical Results." Department of Economics, University of Rochester, Rochester, New York. Processed. Summers, Robert, and Alan Heston. 1991. "The Penn World Table (Mark 5): An Ex- panded Set of International Comparisons, 1950-88." Quarterly Journal of Econom- ics 106(2):327-68. Temple, Jonathan. 1999. "A Positive Effect of Human Capital on Growth." Economics Letters 65 (1):131-34. Wade, Robert. 1990. Governing the Market: Economic Theory and the Role of Govern- ment in East Asian Industrialization. Princeton, NJ: Princeton University Press. World Bank. 1993. The Story of Poverty in Latin America. Washington, D.C. . 1994. Priorities and Strategies for Education: A World Bank Sector Review. Education and Social Policy Department. Washington, D.C. . 1995. World Development Report 1995: Workers in an Integrating World. Washington, D.C.: Oxford University Press for the World Bank. i i i THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 393-4z9 Measuring the Dynamic Gains from Trade Romain Wacziarg This article investigates the links between trade policy and economic growth in a panel of 57 countries between 1970 and 1989. It develops a new measure of trade policy openness based on the policy component of trade shares, using it in a simultaneous equations system to identify the effect of trade policy on several determinants of growth. The results suggest a positive impact of openness on economic growth, with the accel- erated accumulation of physical capital accounting for more than half the total effect; enhanced technology transmission and improvements in macroeconomic policy account for smaller effects. This decomposition is robust with respect to alternative specifica- tions and time periods. The article also successfully tests whether the model exhaus- tively captures the effects of trade policy on growth. The relationship between trade openness and economic growth has been the subject of numerous empirical studies. Most uncover a positive empirical asso- ciation between trade openness and per capita income growth; until recently, few economists challenged the findings.' Although theories promoting inward- oriented development strategies flourished in the 1950s and 1960s, the policies' unsustainable effects had, by and large, discredited the idea that the costs of an open trade regime may outweigh its potential benefits. Recently, however, Rodrik and Rodriguez (2000) have questioned the em- pirical results on trade and growth, pointing to methodological problems asso- ciated with the measurement of openness and the specification of estimated equations.2 In particular, the collinearity between trade protection and other mea- sures of (possibly domestic) policy, such as the quality of macroeconomic policy, might lead researchers to conclude wrongly that trade protection depresses growth, when another omitted or poorly measured variable is in fact accounted Romain Wacziarg is with the Graduate School of Business, Stanford University. His e-mail address is wacziarg@gsb.stanford.edu. This article was written while the author was a visiting scholar in the World Bank's Development Prospects Group. The author thanks Francois Bourguignon, Milan Brahmbhatt, Francesco Caselli, Uri Dadush, David Dollar, Jean Imbs, Norman Loayza, Francisco Rodriguez, Jean- Francois Ruhashyankiko, Jose Tavares, Athanasios Vamvakidis, Alan Winters, and three anonymous referees for helpful comments. 1. See, for instance, Edwards (1992), Dollar (1992), Ben-David (1993), Sachs and Warner (1995), Frankel and Romer (1999), Alesina and others (2000), among many others. 2. Other recent studies casting doubt on a positive growth-openness link using macrodata include Rodrik (1998a) and Harrison and Hanson (1999). In a related literature, a study by Vamvakidis (1998) uncovers negative effects of regional arrangements, such as free trade areas, on growth in time-series data. ©3 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 393 394 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 for by trade openness.3 This challenge suggests two directions for research: im- proving existing measures of trade policy openness and being more explicit about how trade openness might affect growth by specifying more clearly the channels relating these variables. This allows for the possibility that negative channels may partially or fully outweigh positive ones. This article seeks to advance the litera- ture on both fronts. Theory points to a number of possible costs and benefits of trade openness, not mutually exclusive in general. Some theories stress technological spillovers and the international transmission of knowledge as a source of growth for open econo- mies.4 More traditional, static theories invoke allocative efficiency, which can be achieved more easily with an open trade regime even when factors of production are assumed to be immobile. Higher levels of output are attained when countries specialize according to comparative advantage, so growth rates can be expected to increase in the transition that follows a liberalization episode. The increased degree of market competition resulting from a wider scale of market interactions yields further gains in efficiency.5 More generally, by increas- ing the size of the market, trade openness allows economies to better capture the potential benefits of increasing returns to scale.6 Yet another set of theories points to the complementary aspects of virtuous policies: trade policy openness may create incentives for governments to adopt less distortionary domestic poli- cies and more disciplined types of macroeconomic management. On the cost side, some theories suggest that when comparative advantage patterns would lead a country to specialize in goods where technological innovations or learning by doing are largely exhausted, opening up to trade might actually reduce long-run growth (Young 1991). Another potentially negative channel was suggested in Rodrik's (1998b) findings on openness and government size: more open coun- tries may face incentives to increase the size of government to insure agents in the face of foreign shocks. In turn, a larger government may distort resource allocation, to the detriment of economic growth., There has been little empirical work to determine the relative roles of these dif- ferent factors in explaining the observed overall impact of trade policy openness on growth. The finding that trade openness spurs growth tends to be interpreted according to the observer's preferred theory, but two important possibilities are ignored: several forces may be operating simultaneously, and trade openness may also involve some dynamic costs, even if the benefits outweigh them. This be- 3. For example, Rodrik and Rodriguez (2000) criticize the Sachs and Warner (1995) contribution because much of the variance in their trade liberalization dummy is accounted for by the black market premium on the exchange rate, itself at least as much a measure of poor domestic policies as of a closed trade regime. 4. See, for instance, Grossman and Helpman (1991) and Barro and Sala-i-NMartin (1997). This relies on the notion that more open economies are better able to import advanced technologies. 5. See, for instance, Wacziarg (1997). 6. See Ades and Glaeser (1999) and Alesina and others (2000). 7. Barro and Sala-i-NMartin (1995) provide empirical evidence on this point. Wacziarg 395 comes especially important with increasing integration: by determining the source of the costs and benefits of trade liberalization, policymakers can hope to maxi- mize the benefits and to minimize the costs. This article employs a fully specified empirical model to evaluate the chan- nels through which trade policy might affect growth. To this end, it presents two innovations. The first is a new measure of trade openness based on a weighted average of several indicators (tariff revenues, nontariff barriers, and an indica- tor of overall outward orientation). This new measure of trade policy openness corresponds to the policy-induced component of an average country's trade to gross domestic product (GDP) ratio. The second innovation is a set of equations describing the incidence of trade policy on several determinants of growth. Moving away from single-equation, reduced-form growth empirics, these equations capture different theoretical ar- guments on the potential costs and benefits of trade policy openness. Various channel variables are included in a growth regression. By multiplying the effects of trade policy on the channel and the effect of the channel on growth, the effect of trade policy on growth through that specific channel can be identified. The results suggest a positive effect of trade policy openness on economic growth, with accelerated accumulation of physical capital accounting for more than half the effect. The article first analyzes the theoretical basis for the six channels and describes the empirical methodology for measuring the channel effects. It then discusses measurement issues, provides preliminary evidence on trade policy and growth, and describes the channel effects. The model's robustness and exhaustiveness are also examined. I. THEORY AND METHODOLOGY This section discusses the six channel variables and outlines the article's empiri- cal methodology. The Six Channels in Economic Theory The six links between trade policy and economic growth incorporated in the empirical model are meant to capture the dominant theories concerning dynamic gains (or possibly losses) from trade. The underlying assumption is that together these six channels adequately capture most of the effect of trade policy on growth. These channels are broadly classified under government policy, domestic allo- cation and distribution, and technology transmission. GOVERNMENT POLICY. Trade openness may create incentives for policymakers to pursue virtuous macroeconomic policies, either because of the threat of capi- tal flight or because of international agreements, implicit or explicit, that act as a check on policy. Preserving a competitive environment for domestic firms en- gaged in foreign transactions may also require policies that maintain macroeco- 396 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 nomic stability. In turn, macroeconomic stability is likely to favorably affect growth by reducing price uncertainty and moderating public deficit and debt levels, thereby reducing crowding out and the likelihood of future tax increases and furthering the ability of domestic firms to compete on global markets (Fischer 1993). Another way to capture the effects of trade openness on government activity is to consider the effect on the size of government. If more open economies are subject to larger exogenous supply and demand shocks, a larger government may be better able to provide insurance or consumption smoothing through redistri- bution or other forms of social programs (Rodrik 1998b). On the other hand, open economies may tend toward laissez-faire arguments and more limited taxa- tion to preserve the economy's price competitiveness and attractiveness to for- eign investors. The effect of trade policy openness on government size, measured by the public consumption of goods and services, is therefore theoretically am- biguous. On the other hand, theory points to a positive growth-maximizing size of government resulting from a tradeoff between the productive function of public activities and the distortionary nature of taxation (Barro and Sala-i-Martin 1992). In addition, Barro (1991) and Barro and Sala-i-Martin (1995) document the nega- tive impact of a larger government on growth in a cross-section of countries. ALLOCATION AND DISTRIBUTION. Open economies are less likely to have trad- able goods prices that differ substantially from world market prices because free trade should facilitate price convergence of traded goods across countries. Open countries will tend to specialize according to their comparative advantage, so once the effect of nontradable goods on deviations from purchasing power par- ity has been eliminated, countries with open trade policies would be expected to have lower overall price levels (relative to some benchmark country, such as the United States) than closed economies (Dollar 1992). Hence, theory points to a lower degree of price distortion in open economies, and price distortions have been shown to adversely affect factor accumulation and growth (Easterly 1989, 1993). Factor accumulation may also be of crucial importance. Much of the effect of trade policy on growth may well work through the domestic rate of physical investment, which is a determinant of economic growth in a nearly tautological sense (Levine and Renelt 1992; Baldwin and Seghezza 1996).8 The investment channel may capture several theories. First, investment may respond to open- ness through a size of the market effect. As first stressed by Adam Smith, market 8. However, some scholars question the direction of causality between investment and growth, based on Granger causality tests (see Blomstron and others 1996). But these tests are typically based on rela- tively high-frequency data, whereas this study examines long-term relationships between growth and its determinants. The Solow model predicts that the long-term relationship runs from investment rates to growth. This article also uses an instrumental variables estimator, which should limit the incidence of this type of endogeneity. Wacziarg 397 size imposes a constraint on the division of labor, so that more open countries are better able to exploit increasing returns to scale. Trade liberalization may thus provide the type of big push effect on capital accumulation that Murphy and others (1989) argued was required for less developed countries to move from a low growth equilibrium to a path of sustained industrialization.9 Using a related argument, Wacziarg (1997) shows that the extent of the market is an important determinant of product market competition. The entry of new firms in export markets after an episode of liberalization may well entail large fixed investments. Second, trade liberalization may simply allow domestic agents to import capital goods that were previously unavailable (or produced locally but at higher costs), thus removing structural constraints on investment. Capital goods imports, which make up sizable proportions of the imports of many recently liberalized devel- oping economies, also embody more recent technologies, a further source of growth. In a related argument, Baldwin and Seghezza (1996a, 2) state that "as- suming that traded goods are an input into capital formation, protection raises the cost of new capital goods and thereby tends to lower the rate of return on investment. With intertemporal optimization, this lowers the steady-state capi- tal stock and slows growth in the transition." TECHNOLOGICAL TRANSMISSIONS. The last two channels are drawn from the recent literature on endogenous growth: if knowledge spillovers are a driving force for sustained, long-run growth, and open economies are more exposed to a worldwide stock of productivity-enhancing knowledge, then trade openness can affect growth and convergence through technology transmissions (Barro and Sala-i-Martin 1997; Grossman and Helpman 1991). One way openness can increase the exposure of the domestic economy to tech- nology transmission is by making it easier, through more frequent and sustained international trade interactions, for domestic producers to imitate foreign tech- nologies and incorporate that knowledge in their own productive processes (Edwards 1992). This increased exposure can stem from direct imports of high- tech goods or from greater interaction with the sources of innovation (through enhanced international communication and mobility brought forth by economic integration). This should translate into a higher capacity to compete with more advanced economies on world markets. Such a pattern was certainly part of the East Asian growth miracle, characterized by broad transformations in the prod- uct composition of output and exports from agriculture to heavy industry and finally to high-tech goods, through the imitation of technology originating in industrial countries. 9. Ades and Glaeser (1999) provided preliminary empirical evidence showing that the extent of the market boosts growth largely through an increase in the rate of capital accumulation, thus lending support to big push theories. The working paper version of this study (Wacziarg 1998) contains further evi- dence on this point. 398 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 A second channel for greater technology transmission is foreign direct invest- ment (FD1), whether associated with joint ventures or not. FDI often transmits advanced types of technology, either through capital goods imports that are later imitated or through the diffusion of knowledge and expertise. However, it is unclear a priori that trade openness is associated with greater levels of FDI. FDI may act as a substitute for trade, because foreign investment is used to set up plants producing goods that cannot be imported because of trade restrictions (tariff-hopping). Or investors may view trade openness as a signal that a coun- try is committed to stable and market-oriented economic policies, whereas trade openness allows them to import the intermediate goods required to initiate projects, expect repatriation of some profits, and export the goods they produce. Falling transport costs may allow a slicing up the value-added chain, so that firms can produce a good in stages in several locations, adding a little more value at each stage (Krugman 1995). In that case FDI may complement rather than sub- stitute for trade openness. Indeed, evidence suggests that open economies attract more FDI than closed economies (Harrison and Revenga 1995). In turn, FDI is likely to spur growth. Because the share of FDI in GDP iS typi- cally small (averaging about 1 percent), it is hard to argue that FDI would spur growth through traditional physical capital formation. If there is any significant dynamic effect of FDI, it likely captures the incidence of a certain type of tech- nology transmission, an interpretation applied here to the FDI channel.10 Empirical Methodology The estimates presented in this article use a method first employed in a cross- country growth context by Tavares and Wacziarg (2001) to analyze the effects of democracy on growth.1" The underlying econometric theory is an extension of the three-stage least squares method of Zellner and Theil (1962) to panel data. THE STRUCTURAL MODEL. The basic framework for the cross-sectional analy- sis consists of a simultaneous equations model aimed at identifying the effects of trade policy on growth. The model consists of an equation for the growth of per capita income, one for determining the nature of trade policy, and six channel equations describing the effects of trade policy on several growth determining variables. This constitutes the structural model, derived from economic theory: the channel variables are included in the growth regression, but the measure of trade policy openness appears only in the channel relationships. 10. However, Aitken and Harrison (1999), using plant-level data for Venezuela, show that foreign ownership adversely affects the productivity of domestically owned plants. I will use macroeconomic data to evaluate whether this result holds at the aggregate level. 11. Baldwin and Seghezza (1996) also employ three-stage least squares to estimate a system for the joint determination of growth and investment rates, as a function, among other variables, of the trade regime. Taylor (1998) uses a structural approach to growth empirics to study the impact of outward orientation on growth in Latin America. Wacziarg 399 To better understand the foundations for the channel analysis, consider a neo- classical production function: Y = AKcH3L-1'Ai3, where A denotes the level of tech- nology, K physical capital, H human capital, and L labor.12 Dividing by L and totally differentiating with respect to time yields the traditional Solow decomposition: y A k h' where lowercase letters designate per worker quantities. Hence, the ultimate drivers of per capita growth are technological growth and the (per capita) growth of human and physical capital. Presumably, the nature of trade policy can affect either of these factors. The channel methodology therefore consists of excluding the trade policy index from the growth equation directly and examining its ef- fects on the ultimate drivers of growth instead. Limiting the number of ultimate growth determinants, however, may over- simplify the model. To enrich the stuctural model and allow for the effects of trade policy openness on growth through such factors as government policies or technology transmission (the latter being only part of the A/A term), the list of growth determinants can be augmented. For example, adding a measure of gov- ernment consumption and macroeconomic policy allows consideration of the corresponding channels. Hence, although the analysis here takes a step away from purely reduced-form growth empirics, it stops short of a fully structural model. Such a model would involve explicit consideration of the effects of, for example, government consumption on factor accumulation and technological progress, which are in turn the ultimate drivers of growth."3 In contrast, in the model developed here, government consumption and factor accumulation appear jointly in the growth equation, so any effects of openness mediated by government con- sumption (including those going through investment) will be reflected in the government consumption channel. An equation is formulated relating trade policy and other determinants to each channel variable under consideration, with the intention of fully exhausting the ways that openness could affect growth. (Formal evidence concerning this issue is provided in section III.) Finally, the equation describing the determinants of trade policy openness explicitly deals with the endogeneity issues having to do with the simultaneous determination of trade policy, growth, and the channel variables. In particular, several channel variables may appear on the right-hand side of the trade policy equation. 12. I am grateful to an anonymous referee for suggesting this interpretation. 13. In this case, the analysis would involve three rather than just two steps: a fully structural model would consider the effect of trade policy on government consumption, the effect of government con- sumption on technological and factor growth, and the effect of technological and factor growth on per capita income growth. However, such a system would be extremely cumbersome and would involve the estimation of a large number of parameters relative to the available data. I I~~~~~~~~~~~~~~~~~~~~~~ 400 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 ESTIMATION. The parameters of the structural model are estimated jointly using three-stage least squares. This method achieves consistency by appropriate instrumenting, and efficiency through optimal weighting. It combines features of instrumental variables, random effects, and generalized least squares models. Each of the eight equations in the structural model is formulated for four time periods under scrutiny (1970-74, 1975-79, 1980-84, 1985-89). Joint estima- tion allows the derivation of a large covariance matrix for the error terms of all 32 equations. Hence, both cross-period and cross-equation error correlations are allowed to differ from zero. This ensures the efficiency of the estimates. Taking cross-period error correlations into account is similar to assuming that the error terms contain country-specific effects uncorrelated with the right-hand-side vari- ables. The flexibility of the error covariance matrix allows for substantial effi- ciency gains relative to estimating each equation separately (that is, assuming zero cross-equation error covariances). Because several endogenous variables appear on the right-hand side of the structural equations, endogeneity bias is a concern. Consistency requires instru- menting for every endogenous variable that appears as a regressor. This is done by first rewriting every endogenous variable as a function of all the exogenous variables in the system in the model's reduced form. The fitted values of each endogenous variable from ordinary least squares estimation of the reduced form equations will provide suitable instruments for each corresponding endogenous variable in the structural form.14 Because of concerns about the endogeneity of per capita income levels in the context of a random effects estimator with a lagged dependent variable, per capita income was excluded from the list of instruments (see Caselli and others 1996). The second stage of the three-stage least squares procedure consists of esti- mating each equation in the structural model separately through instrumental variables (or two-stage least squares), using the instruments constructed in the first stage. This allows the derivation of a consistent covariance matrix for the error terms of the model. The third stage employs this covariance matrix as a weighting matrix as well as the instruments derived in the first stage to jointly estimate the equations in the structural model using instrumental variables- generalized least squares. IDENTIFICATION AND RESTRICTIONS. Some assumptions about specifications are required for this methodology to carry through. For each equation, enough instruments must be validly excluded for the order condition to be met: at least 14. Given the above specification of the baseline model, the instruments are male and female human capital, the island dummy variable, the log of population, the democracy index, the log of area, terms of trade shocks, population density, the secondary school completion rate, the share of population over age 65, the share of population under age 15, ethnolinguistic fractionalization, and postwar indepen- dence status, each taken at every time period when applicable. Wacziarg 401 as many exogenous variables must be excluded as regressors because there are endogenous variables included on the right-hand side. The chosen specification is based on empirical work on the determinants of the endogenous variables under study. For instance, the growth and in- vestment equations are based on common specifications used in the cross- country growth literature (Barro and Sala-i-Martin 1995). The specification of the government size equation is based on Rodrik (1998b) and Alesina and Wacziarg (1998). For other channels, priors were used to determine the set of exclusions."5 (Table C-1 in appendix C displays parameter estimates of each equation in the system for the baseline model, allowing readers to infer the specification of each of the equations in the system.)'6 To assess the long-run effects of trade policy on growth in a unified manner, cross-period parameter equality restrictions are imposed: none of the estimates of the parameters in the structural model is allowed to vary across time. This allows efficiency gains through higher degrees of freedom, as the number of es- timated parameters in the system is divided by four. Whether these cross-period parameter equality restrictions are justified is examined in section III. II. MEASUREMENT ISSUES AND PRELIMINARY EVIDENCE This section considers issues involved in measuring trade openness and the channel variables, and presents simple correlations between the main variables of interest. Existing Measures of Trade Openness Measuring the extent of trade openness is a major challenge for any study involving the analysis of trade policy, as suggested in Rodrik and Rodriguez (2000) and Pritchett (1996a). There are three broad categories of existing measures of trade openness. OUTCOME MEASURES. Outcome measures describe the volume of trade or its com- ponents. This type of indicator is most subject to endogeneity problems with re- spect to growth (Frankel and Romer 1999), but because it measures actual exposure to trade interactions, it may account quite well for the effective level of integra- tion. It may correlate only imperfectly, however, with attitudes or institutions re- lating to openness. Past research has tended to confuse outcome measures with policy attitudes (which are presumed to partly determine the outcome), largely because precise measures of actual trade policies were not widely available. Because most theories about dynamic gains from trade have to do with policy measures, contrasting free trade to restricted trade or autarky, an index of trade 15. Tavares and Wacziarg (2001) discuss in more detail the specification search for the type of sys- tem that is considered here. A previous report on this study (Wacziarg 1998) describes the specification of each equation in the system. 16. Because each equation is estimated for four time periods, with estimated parameters constrained to equality across periods, the table reports R2 statistics corresponding to each of these time periods. . . . . . . | | I I I I 402 THE WORLD BANK ECONOMIC REVIEW, VOL. IJ, NO. 3 policy had to be constructed for this study that adequately captures the nature of the policy regime for international trade.17 The use of outcome measures seems undesirable on these grounds, because they also reflect the gravity component of trade openness. The choice is then between direct policy indicators and effective protection measures. POLICY INDICATORS. Tariff rates, nontariff barriers, tariff revenues, and related matters describe the institutional features of a country's attitude toward the rest of the world with respect to trade and factor flows. As such, they are likely to be an important determinant of the outcome measures. However, there are endogeneity problems in their relationship with growth, and they tend to have lim- ited availability. Furthermore, they may not directly reflect the degree of effective protection faced by domestic agents, but only the legal framework they confront. The main drawback of such trade policy measures as tariff barriers, nontariff barriers, and broader measures of a country's liberalization status is that they are weakly correlated among themselves. Pritchett (1996a) showed that no such single policy measure adequately captures a country's outward orientation. Be- cause various measures may reflect different aspects of a country's trade policy, using a single indicator may not be very informative. This suggests combining the variation in several measures to obtain an indicator of trade openness. DEVIATION MEASURES. Deviations of observed trade volume from the predicted free-trade volume are also used to provide a measure of how restrictive the trade regime really is.1I Factor endowment and gravity models of trade generate predic- tions about a country's propensity to trade internationally. For instance, small country size, distance from major trading partners, and negative terms of trade shocks can be thought to affect trade volumes negatively. Similarly, relative en- dowments of skilled labor, unskilled labor, and capital and natural resources may have an impact on overall trade volumes. This type of variable can be used to predict a country's potential free trade volume of international commercial transactions. There are three drawbacks to measures based on deviations. First, some determinants of potential trade may have been omitted, so the predicted level of trade may not adequately measure the volume of commercial transactions that would prevail under complete free trade."9 Second, some gravity or endowment determinants of potential trade may be highly correlated with policy attitudes, so the deviation of observed from potential trade may exclude some valid infor- 17. The working paper on this study (Wacziarg 1998) presents empirical evidence in favor of this choice: the growth effects of trade openness seem mostly due to the trade policy regime, rather than to the gravity component of trade shares. 18. The classic reference on such residual measures is Leamer (1988). 19. Frankel and Romer (1999) presents a state-of-the-art method for computing the gravity com- ponent of trade volumes by regressing bilateral trade on exogenous characteristics of country pairs, such as distance and common language. Wacziarg 403 mation about policy. Third, as long as the observed volume of trade contains a white noise disturbance term, deviations from predicted volumes will also con- tain a white noise disturbance (whose share of the variance in the total variance of the measure has increased due to the differencing), and its use will result in increased downward bias associated with measurement error. Construction of the Trade Policy Openness Index The approach used here attempts to avoid these problems with existing mea- sures of trade openness. A country's trade to GDP ratio can be viewed as result- ing from policy, factor endowment, and gravity determinant variables. The trade policy index is computed by isolating the variation in trade shares attributable to a variety of trade policy measures. More specifically, trade shares (the ratio of imports plus exports to GDP) are regressed on several openness-determining variables, including policy, gravity, and endowment variables. The estimated coefficients on the policy variables are used as weights in constructing a weighted average of these variables. This weighted average is the index of trade policy openness, equal to the portion of observed trade shares attributable to the effective impact of trade policy. This procedure avoids both the problem of measurement error due to the construc- tion of the difference between observed and potential trade shares (because it is not constructed as a residual) and the problem of collinearity between gravity and endowment and policy factors. Components of the Openness Index The objective is to construct an openness measure that applies to a broad range of countries over the period 1970-89 and that adequately accounts for tariff barriers, nontariff barriers, and other policy attitudes toward international trade that capture outward orientation. TARIFF BARRIERS. The effects of tariff barriers are captured by the share of import duty revenues in total imports (from the International Monetary Fund's [IMF] Government Finance Statistics Yearbook). This has three advantages. First, it better captures the effective degree of tariff restrictions. Direct overall measures of tariff protection obtained from the U.N. Conference on Trade and Development (UNCTAD) are unweighted averages of goods-specific tariff rates. However, duty revenues are by construction weighted by the composition of imports. Second, officially declared tariff rates and effectively implemented rates may vary substantially. Duty revenues once again avoid this problem by mea- suring the tariff revenues actually collected. Third, data based on revenues are available for more countries and a wider time span than direct measures of tariff rates.20 20. Unweighted tariff rates were available for the period 1980-93 only, and for approximately 50 countries. 404 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 One potential limitation of the use of tariff revenues is that prohibitive tariff rates tend to reduce revenues through a Laffer curve effect applied to imports. However, the problem is likely greatly attenuated by the fact that duty revenues are treated as a share of total imports (high tariff rates work to reduce revenues by deterring imports, so the ratio of the two should roughly reflect effective tar- iff rates). Correlations between tariff revenues and tariff rates, for the dates and countries available for both measures, are relatively high, ranging from 66 (64?) percent to 83 percent (table 1). NONTARIFF BARRIERS. Existing measures of nontariff barriers are highly im- perfect, dealing mainly with coverage rates (percentage of goods affected by quo- tas, voluntary export restraints, and the like) and ignoring whether the constraints are binding. Furthermore, there is no consistent panel data set for nontariff bar- riers. The meausure used here is an unweighted coverage ratio for the pre-Uru- guay Round time period published by UNCTAD. Although the extent of nontariff barriers has no doubt varied across time, like tariffs it is likely to be highly autocorrelated within countries. The data do not permit accounting for this time- series variation, because there is only one observation for the 20 years under consideration. Presumably, this type of measurement error weakens the relation- ship of nontariff barriers with trade volumes and correspondingly reduces the weight of this indicator in the overall index. LIBERALIZATION STATUS. A third component for the index of trade policy was developed to capture the overall attitude of policymakers. Dummy variables were constructed for a country's liberalization status for each year using the list of trade liberalization episodes compiled by Sachs and Warner (1995) for a large sample of countries.21 These were then averaged over the four time periods under study. Insofar as this indicator receives some weight in the index, it captures factors other than just tariffs and nontariff barriers; in particular, it may help account for the effect of time variations in nontariff barriers, which cannot be explicitly accounted for because of data unavailability. Rodrik and Rodriguez (2000) have recently raised strong doubts about the indicator used in Sachs and Warner (1995), arguing that much of the variation in the liberalization dummy variable is attributable to the black market premium on the exchange rate (a proxy for distorted macroeconomic management as much 21. These dates were constructed by examining trade policy data and by conducting a systematic analysis of the literature concerning the trade regimes of specific countries. The sources for the dates for each country are reported in the appendix to their article. Note that the dates of liberalization com- puted by Sachs and Warner (1995) and their cross-sectional liberalization dummy (for the mid-1980s) are derived using different methodologies. In particular, because much of the tariff and nontariff data were not available for periods before the 1980s, Sachs and Warner (1995, 24, n. 44) resorted to a lit- erature search to determine when countries opened their trade regimes, rather than to the five formal criteria used to derive their well-known liberalization dummy variable (the latter is computed for the mid-1980s only). Wacziarg 405 TABLE 1. Correlations between Duty Revenues and Unweighted Tariff Rates Import duties Tariff rate 1980-84 1985-89 1990-94 1980-84 0.667 0.744 0.725 1985-89 0.638 0.754 0.717 1990-94 0.802 0.837 0.831 Note: 50 observations. as for the degree of openness) and the existence of an export marketing board (a characteristic mainly of slow-growing African economies). Hence, they argue that the Sachs and Warner variable is constructed in a way that is conducive to find- ing a positive effect of openness on economic growth. For this reason, results are also presented here based on an index of openness that excludes the Sachs and Warner liberalization status.22 Correlations between these underlying components of the trade policy index are displayed in table 2. The signs of the correlations are as expected. The nontariff barriers measure is most weakly correlated with the other indicators, suggesting either that its inclusion may provide useful information about trade policy or that it is a poor measure of openness. Insofar as the nontariff barrier measure poorly reflects the true orientation of trade policy, however, it should receive a small weight in the overall index. Trade Shares Regressions Estimates pertaining to the determination of trade shares are shown in table 3.23 The explanatory variables feature the three policy indicators (import duties as a share of total imports, the pre-Uruguay Round nontariff barriers coverage ratio, and the Sachs-Warner liberalization status indicator averaged over the relevant five-year time periods). The regression also features gravity components, such as log of land area and log of population, as well as the growth rate of per capita GDP.24 As expected, trade shares are positively affected by liberalization status and negatively affected by tariffs and nontariff barriers. The lack of precision of the 22. This study uses an indicator based on Sachs and Warner's liberalization dates rather than on their (purely cross-sectional) liberalization dummy. This may reduce the incidence of the Rodrik and Rodriguez critique, insofar as the liberalization dates are based on the broad survey of the literature on specific countries' trade regimes. Entirely removing this indicator from the index, however, allows the Rodrik and Rodriguez critique to be addressed more fully. 23. The three-stage least squares estimator described earlier is used to obtain these estimates. 24. The working paper on this study (Wacziarg 1998) provides evidence of reverse causation from growth to trade shares, justifying the inclusion of economic growth in the equation for the trade to GDP ratio. TABLE 2. Correlations between Underlying Components of the Trade Policy Index Duty Nontariff Liberalization Index component 1970-74 1975-79 1980-84 1985-89 barriers 1970-74 1975-79 1980-84 Duty 1970-74 1.000 Duty 1975-79 0.944 1.000 c. Duty 1980-84 0.825 0.887 1.000 Duty 1985-89 0.753 0.808 0.935 1.000 Nontariff barriers 0.190 0.232 0.157 0.212 1.000 Liberalization 1970-74 -0.467 -0.460 -0.470 -0.459 -0.120 1.000 Liberalization 1975-79 -0.471 -0.465 -0.474 -0.455 -0.085 0.994 1.000 Liberalization 1980-84 -0.470 -0.464 -0.473 -0.446 -0.048 0.978 0.994 1.000 Liberalization 1985-89 -0.429 -0.469 -0.469 -0.479 -0.182 0.900 0.890 0.870 Note: 57 observations. Wacziarg 407 TABLE 3. Trade Shares Regression (three-stage least squares estimates) Trade Policy 2: Trade Policy 1: excluding the Sachs Independent variable baseline index and Warner variable Constant 182.561 186.830 (9.70) (8.55) Growth of per capita income 0.322 0.444 (1.12) (1.40) Log of land area -8.029 -9.164 (-3.69) (-2.23) Log of population -9.121 -8.052 (-3.42) (-2.31) Import duties / total imports -34.733 -60.912 (-1.16) (-1.78) Pre-Uruguay Round -0.217 -0.239 nontariff barrier coverage (-0.73) (-0.74) Sachs/Warner 11.262 liberalization status (2.12) Adjusted R2 0.60 0.55 0.60 0.54 0.53 0.49 0.50 0.47 Number of observations 71 (4) 71 (4) (number of periods) Note: The dependent variable is imports plus exports as a share of GDP. Numbers in parentheses are t-statistics. Because each equation is estimated for four time periods, with estimated parameters constrained to equality across periods, the table reports R2 statistics corresponding to each of these time periods. The instruments used were: initial income; population density; dummy variables for religion, oil producers, postwar independence; log of popula- tion; share of population over 65; and log of area. estimates, largely due to collinearity between the policy measures, is not really a concern since the objective is only to generate rough weights for how the three components affect trade shares. Minor variations in these weights are not likely to influence the final results. Two indices of trade policy were computed using estimates from the two re- gressions in table 3 as weights on the various policy measures. For each period, the trade policy openness indices were computed as: Trade Policy 1 = -34.73(Import Duty Share) - 0.22(Nontariff Barriers) + 11.26*(Liberalization Status) Trade Policy 2 = -60.91(Import Duty Share) - 0.24(Nontariff Barriers) Correlating the baseline index (Trade Policy 1) with its three components (table 4) gives an idea of the relative weights attached to each. For all the compo- nents, correlations with the overall index are larger than 0.449 in absolute value, but the duty revenue component dominates with a correlation ranging from 0.634 to 0.790, depending on time period. As expected, the nontariff barriers compo- nent received the smallest weight. Correlations between the two indices of trade 408 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE 4. Correlations between Trade Policy 1 and Underlying Components Trade Policy 1 Index component 1970-74 1975-79 1980-84 1985-89 Duty 1970-74 -0.713 -0.703 -0.690 -0.643 Duty 1975-79 -0.705 -0.733 -0.725 -0.704 Duty 1980-84 -0.645 -0.673 -0.747 -0.737 Duty 1985-90 -0.634 -0.654 -0.724 -0.790 Nontariff barriers -0.507 -0.501 -0.449 -0.526 Liberalization 1970-74 0.867 0.862 0.860 0.752 Liberalization 1975-79 0.851 0.854 0.860 0.733 Liberalization 1980-84 0.826 0.837 0.850 0.706 Liberalization 1985-90 0.810 0.818 0.812 0.838 Note: 57 observations. openness used in this study correlations are always greater than 80 percent (table 5). Although high, this shows that the exclusion of the Sachs and Warner liberaliza- tion status from the index can be expected to have some impact on the results. Summary Statistics for Growth and the Openness Index Summary statistics for growth and the trade policy index provide preliminary insights into the relationship between them. Tables 6 and 7 display first and second moments for per capita GDP growth and the policy index for five-year averages during 1970-89. The simple contemporaneous correlations between growth and Trade Policy 1 are positive, but their magnitudes are somewhat small, especially for 1975-79, when the oil shock may have negatively affected the relationship between openness and growth (table 7). Furthermore, the simple correlations between growth and Trade Policy 2 are small in magnitude and negative in three out of four periods. Overall these correlations suggest that the relationship between trade policy openness and growth, if any, will be condi- tional on other determinants of growth. Measurement of Channel Variables Three of the channel variables considered in section I- FDI inflows as a share of GDP, government consumption as a share of GDP, and the domestic invest- ment rate-can be captured in fairly uncontroversial ways as far as measure- ment is concerned. The other three channels are captured by composite indices or approximated using available data. The quality of macroeconomic policy is captured by an index that gives equal weight to each of three decile rankings of policy characteristics for each country for each time period: level of public debt as a percentage of GDP, level of government deficit as a share of GDP, and growth of M2 net of total real output growth (higher numbers signal better policies). The rankings are averaged to obtain an index of overall macroeconomic policy quality, which Wacziarg 409 TABLE 5. Correlation between the Two Trade Policy Indices Trade Policy 1 Trade Policy 2 1970-74 1975-79 1980-84 1985-89 1970-74 1975-79 1980-84 Trade Policy 1, 1975-79 0.991 1 Trade Policy 1, 1980-84 0.967 0.982 1 Trade Policy 1, 1985-89 0.908 0.919 0.930 1 Trade Policy 2, 1970-74 0.806 0.795 0.758 0.763 1 Trade Policy 2, 1975-79 0.787 0.805 0.772 0.796 0.968 1 Trade Policy 2, 1980-84 0.763 0.782 0.817 0.846 0.889 0.927 1 Trade Policy 2, 1985-89 0.731 0.746 0.785 0.870 0.834 0.867 0.955 Note: 57 observations. reflects a country's position relative to others. This avoids the problem of hav- ing to characterize a "good" macroeconomic policy in absolute terms.25 The extent of technology transmission is approximated by the share of manu- factured exports in total merchandise exports, admittedly an imperfect proxy.26 Countries able to compete effectively on world markets for manufactured goods and to produce at world standards are likely to incorporate more of the existing modern technologies in their productive processes. The crucial point is that tech- nological advances and knowledge embodied in existing goods must make their way into production processes to truly qualify as technology transmission. The share of manufactured imports in merchandise imports, another possible mea- sure, was not used because imports of manufactures may act as a substitute rather than a proxy for technology transmission.27 The black market premium on the official exchange rate is used as a measure of price distortions prevailing within the economy, to capture the effect of trade policy on the efficiency of the price system. The black market premium is widely used in cross-country analyses. Barro and Sala-i-Martin (1995) argue that the black market premium on foreign exchange is a widely available and apparently accurate measure of a particular price distortion and can serve as a proxy for government distortions of markets more generally.28 25. The working paper on this study (Wacziarg 1998) presents greater detail on the method used to compute the macroeconomic policy quality index. 26. The share of manufactures in merchandise exports was used as a proxy for technology trans- mission in World Bank (1996). 27. It attempts to employ the share of manufactured imports to total merchandise imports as a proxy for technology transmission, instead of the share of manufactured exports, no statistically significant relationship was found between this variable and growth on the one hand and trade policy openness on the other, even when controlling for a diverse set of variables. 28. 1 am grateful to an anonymous referee for pointing out that the black market premium is a component of the Sachs and Warner dummy and that estimates of the coefficient on a variable that includes black market premium in a black market premium equation will be tainted by endogeneity bias. There are three answers to this objection in the context here: first, as stated above, this study employs an indicator based on the Sachs and Warner liberalization dates, rather than their dummy 410 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 TABLE 6. Summary Statistics for Growth and the Trade Policy Indices Standard Mean deviation Minimum Maximum Growth 1970-74 3.990 2.520 -0.499 12.351 Growth 1975-79 2.333 2.845 -6.688 10.433 Growth 1980-84 0.380 2.740 -8.277 6.018 Growth 1985-89 1.974 2.455 -3.063 8.770 Trade Policy 1, 1970-74 -1.305 8.496 -17.840 10.438 Trade Policy 1, 1975-79 -0.937 8.460 -18.716 10.781 Trade Policy 1, 1980-84 -0.712 8.663 -19.358 10.784 Trade Policy 1, 1985-89 -0.326 9.425 -26.000 10.781 Trade Policy 2, 1970-74 -9.659 6.518 -26.103 -1.136 Trade Policy 2, 1975-79 -9.151 6.704 -26.877 -0.535 Trade Policy 2, 1980-84 -8.896 7.188 -30.746 -0.528 Trade Policy 2, 1985-89 -9.605 8.835 -41.662 -0.535 Note: 57 observations. Simple statistics for openness, growth, and the channel variables, averaged over the period under consideration, provide preliminary evidence of the relevance of the choice of channels (tables 8 and 9). Unconditional correlations suggest that all of the channels involve a positive effect of trade on economic growth (first column of table 9). The largest correlations are in the investment and mar,u- factured exports channels. Overall, this shows that the trade policy index is positively related to FDI as a share of GDP, macroeconomic policy quality, manu- factured exports as a share of merchandise exports, and the domestic investment ratio. Each of these is positively correlated with growth. Trade policy openness is negatively related to the black market premium and government size, and each of these is negatively associated with growth. Although these simple correlations are suggestive, results obtained when controlling for other determinants of growth and the channel variables are likely to differ greatly. III. EMPIRICAL RESULTS Table 10 reports summary effects of each channel on growth, the effect of open- ness on each channel, and the product of the two coefficients for the baseline model for 57 countries for 1970-8 9.29 Trade policy openness works positively variables. Second, a full set of exogenous variables in the system is used to instrument for openness, which should eliminate the hias in the distortions equation. Last, section III shows that the estimated effect of trade policy openness on the black market premium is statistically indistinguishable from zero, so the endogeneity-induced concerns for an upward bias on the magnitude of this effect are not borne out in the estimates. 29. Appendix C, Table C-I contains the entire set of coefficient estimates for each equation in the system, from which the channel effects are obtained. The working paper on this study describes in great Wacziarg 411 TABLE 7. Correlation between Growth and the Trade Policy Indices Growth 1970-74 1975-79 1980-84 1985-89 Trade Policy 1, 1970-74 0.242 0.168 0.259 0.286 Trade Policy 1, 1975-79 0.241 0.168 0.270 0.284 Trade Policy 1, 1980-84 0.267 0.177 0.285 0.294 Trade Policy 1, 1985-89 0.325 0.101 0.118 0.223 Trade Policy 2, 1970-74 0.178 -0.102 -0.067 -0.129 Trade Policy 2, 1975-79 0.192 -0.109 -0.076 -0.125 Trade Policy 2, 1980-84 0.270 -0.084 -0.055 -0.076 Trade Policy 2, 1985-89 0.334 -0.095 -0.112 -0.066 Note: 57 observations. for growth through five out of six channels, three of which-investment, FDI, and macroeconomic policy quality-involve statistically significant effects at the 90 percent level. In each case these involve a positive effect of trade policy on the channel variable and a positive effect of the channel variable on growth. The remaining channel estimates are statistically insignificant at the 90 per- cent level, although government size comes close to being a significantly nega- tive channel (the p-value associated with the t-statistic on the channel effect is 13 percent). For price distortions, this is due to the absence of a significant ef- fect of trade openness on the black market premium once other determinants of this variable (such as per capita income) are held fixed. However, the black market premium was found to bear a negative relationship to economic growth. For manufactured exports, the absence of a statistically significant overall channel effect is due to the fact that this variable does not seem to affect growth in the model specification. However, trade openness was found to be positively asso- ciated with the share of manufactured exports in total exports. The overall effect of all the channels is significant at the 99 percent level. The magnitude of the effects is small for some channels: reduced distortions account for roughly 3 percent of the net effect of trade policy openness on growth and are statistically insignificant due to the absence of a significant estimated effect of trade policy on the black market premium. This is a surprising result in light of the importance accorded allocative efficiency in arguments about static and dynamic gains from trade. The same holds for manufactured exports, meant to capture technology transmission. Government size works negatively for growth, detail the specification choices for the channel equations, as well as the results for each equation in the baseline system (Wacziarg 1998). The t-statistics for the channel effects are obtained by computing linear approximations of the products of the parameters around the estimated parameter values and applying the usual formula for the variance of linear functions of random variables to this linear ap- proximation. Computing these standard errors is possible thanks to the joint estimation of all the equa- tions in the system, which allows the derivation of the covariance matrix for the estimated parameters pertaining to different equations in the system. 412 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 TABLE 8. Summary Statistics for the Main Variables (1970-89 Averages) Standard Mean deviation Minimum Maximum Growth 2.169 1.858 -1.798 7.513 Trade Policy 1 -0.820 8.588 -19.511 10.696 Trade Policy 2 -9.328 7.047 -30.076 -0.683 Macro policy quality 5.203 1.711 1.750 8.833 Black market premium 42.417 83.247 -0.471 437.182 Government consumption 15.591 6.681 7.731 33.962 Manufactured exports 36.933 25.138 0.421 83.664 Investment share 19.381 7.745 1.320 36.135 Foreign direct investment 0.871 1.217 -0.761 7.876 Human capital 1.515 1.163 0.084 5.343 Log income per capita 8.159 0.993 6.154 9.586 Note: 57 observations. although the effect is weak for both magnitude and significance. Differences in the quality of macroeconomic policy and in the ratio of FDI to GDP appear to be relatively important channels, each accounting for roughly 20 percent of the total effects of trade policy on growth. The most important channel by far is investment rate, which accounts for close to 63 percent of the total effect of trade policy on growth.30 Several theoretical arguments point to the potential direct impact of trade policy openness on in- vestment, such as those outlined in section I. It is also possible that measure- ment error in some of the channel variables leads to overstatement of the effect of trade policy through the investment rate. For instance, if investment is posi- tively correlated with technology transmission and if the share of manufactured exports in total merchandise exports is a weak proxy for the extent of technol- ogy transmission, part of this effect may be accounted for by the investment channel. However, the scope of this argument is somewhat limited by the use of a wide set of instruments for all of the channel variables: if measurement errors in the instruments are independent of measurement errors in the channel vari- ables, attenuation bias will be reduced. To summarize, this model provides evidence for a beneficial total effect of trade policy on growth. An 8.5 percentage point increase in the trade policy measure, corresponding roughly to one standard deviation, is associated with a 0.601 percentage point increase in the annual growth rate once all channels of influ- ence are brought into the picture. This effect is estimated with great precision. The most important channel by far seems to be through investment (63 percent of the total effect). Technology transmission explains 22.5 percent of the overall positive effect of trade on growth; macroeconomic policy quality accounts for 30. This is in line with empirical results in Baldwin and Seghezza (1996a) and Levine and Renelt (1992), who found evidence of trade-induced, investment-led growth. TABLE 9. Correlation Matrix for Main Variables Macro Black Trade policy market Government Manufactured Investment Human Growth Policy 1 quality premium consumption exports rate FDI capital Trade Policy 1 0.331 1 Macro policy quality 0.384 0.420 1 Black market premium -0.408 -0.404 -0.304 1 Government consumption -0.421 -0.265 -0.594 0.390 1 Manufactured exports 0.387 0.602 0.393 -0.484 -0.268 1 Investment rate 0.483 0.674 0.441 -0.498 -0.428 0.556 1 FDI 0.503 0.263 0.155 -0.255 -0.296 -0.012 0.342 1 Human capital 0.185 0.554 0.361 -0.357 -0.334 0.487 0.522 0.116 1 Log income 0.266 0.743 0.469 -0.530 -0.504 0.648 0.754 0.188 0.750 Note: 57 observations. 414 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE 10. Summary of the Channel Effects Using Trade Policy 1 Effect of Effect of Effect of Trade Channel variable channel on growth openness on channel Policy 1 on growth Price distortions -0.0066 -0.3445 0.0023 (-9.08) (-0.63) (0.63) Government consumption -0.0425 0.1539 -0.0065 (-1.57) (3.73) (-1.52) Manufactured exports 0.0036 0.6345 0.0023 (0.45) (4.59) (0.45) Investment rate 0.1425 0.3173 0.0452 (6.86) (6.72) (5.12) FDI 0.3203 0.0450 0.0144 (4.68) (4.01) (3.79) Macro policy quality 0.4887 0.0267 0.0130 (4.22) (2. 19) (1.90) Total effect 0.0707 (5.94) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. 18 percent of the effect. The only negative channel, government size, is signifi- cant at the 87 percent level only. Robustness Analysis The model was tested for robustness to the choice of liberalization indicator, to model specification, and to time coverage. EXCLUDING THE SACHS AND WARNER INDICATOR. Table 11 replicates the es- timation of the baseline model replacing Trade Policy 1 with Trade Policy 2 as a measure of openness. Trade Policy 2 excludes the Sachs and Warner liberal- ization status variable critiqued by Rodrik and Rodriguez (2000).31 The magni- tude and precision of the overall estimated channel effects fall, although the investment effect and the overall effect are still positive and statistically signifi- cant. A one-standard-deviation change in Trade Policy 2 (8 percentage points) is now associated with a 0.264 increase in the annual growth rate of per capita GDP. The proportional contributions of most channels remains roughly un- changed, with the investment channel accounting for the bulk of the effect. The main change in the channel effects is the disappearance of the macroeco- nomic policy quality channel, now statistically indistinguishable from zero. This is due entirely to the fact that Trade Policy 2 now bears no relationship to the 31. Referring explicitly to the earlier working paper on this study (Wacziarg 1998), they state that "we are skeptical that the Sachs-Warner measure, on which the Wacziarg indicator is partly based, is a meaningful indicator of trade policy.... We would have preferred to see estimates based only on tariff and [nontariff barrier] indicators." I am grateful to them and to anonymous referees for this suggestion. Wacziarg 415 TABLE 11. Summary of the Channel Effects (Using Trade Policy 2) Effect of channel Effect of Effect of Trade Channel variable on growth openness on channel Policy 2 on growth Price distortions -0.0068 0.4886 -0.0033 (-9.63) (0.65) (-0.65) Government consumption -0.0497 0.2030 -0.0101 (-1.69) (6.14) (-1.60) Manufactured exports 0.0033 -0.0653 -0.0002 (0.41) (-0.52) (-0.33) Investment rate 0.1365 0.2086 0.0285 (6.09) (4.39) (3.67) FDI 0.3066 0.0805 0.0247 (4.38) (5.41) (4.22) Macro policy quality 0.4989 -0.0129 -0.0064 (4.13) (-0.87) (-0.83) Total effect 0.0331 (2.50) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. index of macroeconomic policy. This is consistent with the Rodrik and Rodriguez (2000) critique of the Sachs and Warner indicator for proxying distorted do- mestic policies, which may not be the case for the other measures of trade open- ness used to construct the index. The result is also consistent with an alternative view, more favorable to the baseline model: that the Sachs and Warner liberal- ization dates reflect a component of a country's trade orientation that is only weakly related to direct measures of trade policy, such as tariffs and nontariff barriers, but is nonetheless causally linked to the quality of macroeconomic policy. ROBUSTNESS TO THE SPECIFICATION. Table 12 displays several modifications of the baseline model to examine its sensitivity to changes in specification and estimation method. It presents t-statistics and Wald tests for the significance of the products of coefficients. The Wald statistics are asymptotically distributed as !ables with 1 degree of freedom. As the table shows, the p-values implied by the t-tests and those obtained from the Wald tests are very similar. Figure 1 dis- plays the six channels graphically, for each of the five models in table 12. The third column examines the robustness of the model with respect to esti- mation method, employing the seemingly unrelated regression estimator. This estimator, though inconsistent (no instruments are used), is characterized by greater efficiency and may provide some indication of the model's robustness. It shows that changing the estimator does not greatly affect the sign or magnitude of the estimated effects. In fact, the overall effect of trade policy is roughly the same as in the baseline model. The fourth column restricts the sample to developing countries. The magni- tude of the effect of Trade Policy 1 on economic growth increases when the sample 416 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE 12. Channel Effects under Alternative Models Seemingly unrelated Regional Base model Using Trade regression Developing dummy 1970-89 Policy 2 estimates economies variables Distortions 0.0023 -0.0033 0.0046 0.0090 0.0072 (0.63) (-0.65) (1.73) (2.51) (1.71) Wald test 0.3986 0.4222 2.9825 6.3154 2.9240 (p-value) (0.53) (0.52) (0.08) (0.01) (0.09) Government -0.0065 -0.0101 -0.0044 -0.0009 -0.0107 consumption (-1.52) (-1.60) (-1.57) (-1.14) (-1.93) Wald test 2.3087 2.5534 2.4766 1.2913 3.7085 (p-value) (0.13) (0.11) (0.12) (0.26) (0.05) Manufactured 0.0023 -0.0002 0.0049 -0.0023 -0.0025 exports (0.45) (-0.33) (1.11) (-1.00) (-0.70) Wald test 0.2011 0.1103 1.2284 0.9943 0.4901 (p-value) (0.65) (0.74) (0.27) (0.32) (0.48) Investment rate 0.0452 0.0285 0.0326 0.0394 0.0222 (5.12) (3.67) (4.37) (5.20) (3.54) Wald test 26.1985 13.4690 19.0749 27.0762 12.5666 (p-value) (0.00) (0.00) (0.00) (0.00) (0.00) Foreign direct 0.0144 0.0247 0.0129 0.0231 0.0101 investment (3.79) (4.22) (3.46) (4.90) (2.37) Wald test 14.3848 17.8191 11.9667 24.0583 5.6374 (p-value) (0.00) (0.00) (0.00) (0.00) (0.02) Macro policy 0.0130 -0.0064 0.0161 0.0169 0.0040 quality (1.90) (-0.83) (2.84) (3.36) (1.18) Wald test 3.6089 0.6917 8.0783 11.2935 1.4016 (p-value) (0.06) (0.41) (0.00) (0.00) (0.24) Total effect 0.0707 0.0331 0.0667 0.0853 0.0303 (5.94) (2.50) (5.73) (7.85) (2.38) Wald test 35.3319 6.2294 32.8881 61.6244 5.6878 (p-value) (0.00) (0.01) (0.00) (0.00) (0.02) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. is restricted to developing economies. This is due to the fact that the distortions channel is now significant and represents roughly 10 percent of the overall ef- fect. The other channels are preserved. The last column shows the results of adding regional dummy variables to every equation to account for time-invariant region-specific effects that can covary with the right-hand-side variables. To account for the possibility that regional specificities might be the driving force of the results, regional dummy variables for Latin America, Sub-Saharan Africa, Southeast Asia, and coun- Wacziarg 417 FIGURE 1. Graphical View of the Channel Effects - ~~~~~~0 Macro Policy Qali,ty 01Inestinent (00 0 Manufactur,ed Exports Z Publ Cosumpoon 0 n04 * _f I Distortions 0 04 0 04 o 0 04 a, -c 0,b L _ J tries in the Organisation for Economic Co-operation and Development (OECD) were added to each of the channel equations and to the list of instruments. Be- cause accounting for fixed effects tends to wipe out much of the cross-sectional variation (a fixed-effects estimator uses only the variation within regions across time), the inclusion of regional dummy variables would be expected to lower the estimated effects of trade policy. The total effect of trade policy is reduced by the inclusion of region-specific dummy variables, but each channel's shares are roughly preserved. In particular, the dominant role of physical capital for- mation is maintained, and the overall effect remains statistically significant. ROBUSTNESS TO TIME COVERAGE. Three issues related to the study's time cov- erage were also examined (table 13). First, the cross-equation parameter equal- ity restrictions may not be warranted. Second, a wider time span, though reduc- ing the number of countries for which the data are available, might provide a further robustness check on the results. Third, the use of five-year averages, though increasing the number of data points in the estimation, may highlight short-term variability in the data (due, for example, to business cycle effects) and obscure the long-run relationships.32 32. Rodrik and Rodriguez (2000), referring to this article, state that "we are not sure that the regu- larities revealed by the data over time horizons of five years or less are particularly informative about the relationship between trade policy and long-run economic performance. It would be interesting to see if the results hold up with averages constructed over a decade or more." I am grateful to them for this suggestion. TABLE 13. Sensitivity to Time Period Coverage III III IV V VI Excl. 1970-74 Excl. 1975-79 Excl. 1980-84 Excl. 1985-89 1970-92 10-year averages Distortions -0.0067 -0.0015 0.0126 0.0008 0.0045 0.0431 (-1.15) (-0.37) (1.30) (0.07) (7.28) (1.78) Wald test 1.3148 0.1332 1.6789 0.0052 53.0421 3.1682 (p-value) (0.25) (0.72) (0.20) (0.94) (0.00) (0.08) Government -0.0057 0.0018 0.0050 -0.0105 -0.0066 0.0011 consumption (-1.09) (0.19) (0.62) (-1.53) (-5.85) (0.13) Wald test 1.1960 0.0350 0.3905 2.3511 34.1841 0.0177 (p-value) (0.27) (0.85) (0.53) (0.13) (0.00) (0.89) Manufactured 0.0129 0.0094 0.0037 0.0099 0.0009 0.0397 exports (1.83) (0.89) (0.54) (0.70) (0.53) (1.55) Wald test 3.3573 0.7916 0.2935 0.4940 0.2820 2.4178 (p-value) (0.07) (0.37) (0.59) (0.48) (0.60) (0.12) - Investment rate 0.0317 0.0933 0.0206 0.0349 0.0212 0.1078 (2.62) (5.05) (1.80) (2.07) (7.98) (3.34) Wald test 6.8634 25.5079 3.2294 4.2808 63.6389 11.1376 (p-value) (0.01) (0.00) (0.07) (0.04) (0.00) (0.00) FDI 0.0206 0.0040 0.0118 0.0155 0.0148 0.0232 (4.09) (1.17) (2.43) (2.41) (6.02) (2.23) Wald test 16.7045 1.3678 5.9239 5.8298 36.2359 4.9735 (p-value) (0.00) (0.24) (0.01) (0.02) (0.00) (0.03) Macro policy -0.0258 0.0009 0.0076 0.0089 0.0111 -0.0258 quality (-1.98) (0.11) (0.78) (1.08) (4.24) (-1.11) Wald test 3.9058 0.0126 0.6099 1.1630 17.9800 1.2340 (p-value) (0.05) (0.91) (0.43) (0.28) (0. 0) (0.27) Total effect 0.0271 0.1078 0.0612 0.0595 0.0459 0.1890 (1.48) (3.87) (3.62) (2.53) (11.71) (3.25) Wald test 2.2031 15.0079 13.1403 6.3990 137.2148 10.5605 (p-value) (0.14) (0.00) (0.00) (0.01) (0.00) (0.00) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. Wacziarg 419 First, each of the four time periods was excluded fronm the baseline model one at a time (columns one to four in table 13). 1 This should greatly reduce the pre- cision of the parameter estimates, because a quarter of the data is being excluded. Indeed, the t-statistics on most of the channel effects are considerably lower when only three time periods are used for estimation. For example, the macroeconomic policy and government size channels are no longer statistically significant. How- ever, both the signs and magnitudes of the estimates are remarkably close to those in the baseline model. The investment effect is preserved in all specifications, and in all but one case the overall effect of trade policy remains of the same magnitude. When the timespan is widened by adding 1990-92, distortions and govern- ment size become statistically significant channels, although still relatively small in magnitude. The addition of this time period reduces the number of observa- tions from 57 to 50, while raising the number of data points used to estimate each parameter, thus improving the precision of the estimates. The signs and relative magnitudes of most of the effects are maintained. The reduction in the overall effect, from 0.71 to 0.46, is almost entirely due to a reduction in the magnitude of the investment channel. With respect to the third issue, results are quite robust when 1 0-year averages of the data are used rather than 5-year averages (last column of table 13). In particular, the investment channel remains statistically significant and still ac- counts for over half the total effect of trade policy on economic growth. More- over, the total effect is more than double the previous magnitude, although as expected it is estimated with lower precision. One interpretation of the increase in magnitude is that data averaged over 5 years reflect to some extent short-term variability in the data, analogous to measurement error, whereas data averaged over 10 years are more likely to reflect long-term relationships. Exhaustiveness of the Model The last concern is whether the six channels fully capture the total effect of trade policy on growth. The omission of one or more channels could lead to an in- complete characterization of the effects of trade policy and to potential biases in the estimates of the included channels (insofar as the omitted channel variables covary with the included ones in the growth regression). Other Possible Chansnels Among other possible channels for the effect of trade policy on growth, this study looked briefly at human capital, income inequality, and corruption. HUMAN CAPITAL. The accumulation of human capital might be one of the chan- nels linking trade policy and economic growth. If trade openness modifies the relative returns to factors, it may create greater incentives to accumulate human 33. Furthermore, the exogenous variables corresponding to the excluded period were removed from the list of instruments. 420 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 capital. For instance, if an open trade policy spurs technology transmission and if technology and skills are complements, then trade openness will increase the returns to accumulating human capital. However, no significant linkage effect was found when a human capital channel was specified: the coefficient on the trade policy variable was essentially zero once other determinants of human capital formation, such as per capita income, were held constant. This was ro- bust with respect to the inclusion of a diverse set of controls. Furthermore, the effects of human capital on growth are not robust in the model's growth speci- fication, a problem compounded by the opposite signs of male and female human capital.34 Hence, human capital did not appear to be an important channel link- ing trade policy and growth. INCOME INEQUALITY. Neoclassical trade theory provides several tools for the analysis of income distribution in relation to trade openness. The simple factor endowments theory of Hecksher-Ohlin-Samuelson predicts that returns to un- skilled labor should increase in relative terms, with presumed positive effects on income distribution, when a relatively unskilled labor- abundant country moves from autarky to free trade. There are reasons to believe that inequality has an effect on growth, although the direction of this effect appears a priori ambigu- ous. Alesina and Perotti (1996), among others who have studied the issue of dis- tribution and growth, argue that when the poor have a larger weight in political decisionmaking, they tend to vote for transfer schemes that involve distortive (growth-reducing) taxation. Empirically, they report that more unequal societ- ies tend to display lower growth rates, once other determinants of growth are held constant. However, including a measure of income inequality (the Gini coefficient) in the basic growth regression gave rise to an insignificant effect. Furthermore, the effect of trade policy on income inequality, controlling for the level of per capita income, was found to be essentially zero. Hence, the income inequality channel does not appear to operate either, although the poor quality of cross-country inequality data may be the source of this result. CORRUPTION. Ades and Di Tella (1999) show convincingly that enhanced open- ness to international trade may limit corruption by increasing the degree of in- ternal market competition and reducing opportunities for local bureaucrats to demand bribes. Mauro (1995) provides evidence that corruption has adverse effects on growth. When Mauro's data (from Business International) were in- cluded, however, there was evidence of an effect of trade openness on corrup- tion, but the effect of corruption on growth, though negative, was insignificant and not robust to alternative specifications. This may be due to the fact that the 34. This is consistent with estimates in Barro and Sala-i-Martin (1995) and Pritchett (1 996b). Pritchett (1996b, 1) documents that "cross-national data on economic growth rates show that increases in edu- cational capital resulting from improvements in the educational attainment of the labor force have had no positive impact on the growth rate of output per worker." Wacziarg 421 index of corruption, based on survey methods (and hence likely subject to mea- surement error), was entered in the growth regression along with government size, the black market premium, and the quality of macroeconomic policy, each of which may proxy in part for corruption. Adding the corruption channel also resulted in a loss of degrees of freedom. The corruption data used in Mauro are available only for the 1980s, forcing abandonment of two of four time periods from the estimation and a loss of five countries.35 Unconditional Effect of Trade Policy Openness Further evidence of the model's exhaustiveness is provided by comparing the total effect under the channel methodology with the unconditional effect of trade policy on growth obtained by removing all of the channel variables from the growth regression and using only the trade policy index. The resulting estimates suggest a strong association between the trade regime and growth: a 10 percentage point increase in the trade policy index is associated with a 0.66 percentage point in- crease in the annual growth rate in the baseline model (table 14). With the exclusion of many variables from the growth equation, the trade policy index captures much of their effect on growth that is not necessarily linked to trade policy. However, this coefficient is useful in that it provides a rough order of magnitude against which to compare the total effect of trade policy computed above. Indeed, in all five models, the unconditional effect of trade policy is roughly of the same magnitude as the total effect of trade policy computed earlier. Tests Based on the Residuals from the Growth Equation A more formal test of exhaustiveness can be carried out by regressing the re- sidual vector obtained from the system estimates of the growth regression on the index of trade policy. A correlation between the estimated residual and the measure of trade openness could indicate that a significant channel has been left out of the growth regression. The results, based on a seemingly unrelated re- gression estimator, show that this is not the case (table 15).36 In most of the models, the residual effect of trade policy is generally positive but not signifi- cantly different from zero at any reasonable level of significance. This again re- inforces confidence in the exhaustiveness of the model. That the estimate is positive in the baseline model shows, if anything, that the channel methodology has uncovered a lower bound on the total effect of trade openness. In all cases, the residual effect is statistically insignificant. 35. Results for the income inequality, human capital, and corruption channels are available from the author on request. 36. Again, this should not be taken as an absolute proof of exhaustiveness. To the extent that po- tentially omitted channels covary with the included ones, the included variables will pick up the effects of trade policy that should be accounted for by the missing channels; this would be reflected by a lower correlation between the growth residual and trade policy openness. However, this test provides yet another indication that no major channel has been omitted. 422 THE WORLD BANK ECONOMIC REVIEW, VOL. i5, NO. 3 TABLE 14. Unconditional Effect of Trade Policy in the Growth Equation Seemingly unrelated Regional Baseline regression Developing dummy 1970-89 1970-92 estimates economies variables Intercept 2.6656 1.7436 4.1590 1.6855 4.7800 (1.42) (2.24) (2.34) (1.13) (1.61) Log of initial income -0.0777 0.0375 -0.2586 0.0056 -0.0857 (-0.32) (0.38) (-1.12) (0.03) (-0.23) Male human capital 0.7252 0.9481 0.6709 1.8925 -0.2851 (2.11) (5.30) (2.18) (13.54) (-0.92) Female human capital -0.9261 -1.2652 -0.8367 -1.8404 0.0190 (-3.04) (-8.02) (-2.99) (-7.48) (0.06) Trade policy openness 0.0659 0.0608 0.0908 0.0947 0.0729 (3.00) (7.18) (4.44) (5.97) (2.93) Latin America - - - - -2.1983 dummy variable (-6.74) East Asia - - - - 0.9702 dummy variable (1.77) Sub-Saharan Africa - - - - -3.0903 dummy variable (-5.70) OECD dummy variable - - - - -1.4381 (-3.71) R2 0.120 0.060 0.119 0.089 0.120 0.061 0.231 0.204 0.114 0.313 0.095 0.035 0.088 0.035 0.082 0.026 0.222 0.024 0.446 0.105 0.111 Number of observations 57 49 57 36 57 (number of periods) (4) (5) (4) (4) (4) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) stan- dard errors. Because each equation is estimated for four time periods, with estimated parameters constrained to equality across periods, the table reports R2 statistics corresponding to each of these time periods. Five time periods were reported for 1970-92. IV. CONCLUSION This article is a first attempt, in a cross-country context, to evaluate empirically various theories of dynamic gains from trade in explaining the observed positive impact of trade openness on economic growth. Trade openness affects growth mainly by raising the ratio of domestic investment to GDP. Depending on the specification, the rate of physical capital accumulation explains between 46 per- cent and 63 percent of the impact of trade policy on economic growth. FDI, as a proxy for technology transmission, and the quality of macroeconomic policies each account for roughly 20 percent of the overall effect. There is also weak evidence that the size of government, measured by the ratio of public consump- tion to GDP, constitutes a channel through which trade policy affects economic growth negatively. The lack of statistically significant results for manufactured exports and price distortions may be due to measurement problems. Measurement, although im- proving on past attempts, is still subject to considerable shortcomings. The black Wacziarg 423 TABLE 15. Regression of the Residuals from the Growth Equation on the Trade Policy Index Seemingly unrelated Regional Baseline Trade regression Developing dummy 1970-89 Policy 2 estimates economies variables Intercept 0.0318 -0.1668 0.0484 -0.1794 -0.1362 (0.18) (-0.67) (0.30) (-0.80) (-0.92) Trade policy openness 0.0135 -0.0193 0.0098 -0.0038 0.0188 (0.83) (-0.97) (0.64) (-0.24) (1.36) R2 0.001 0.015 0.011 0.008 0.000 0.006 0.068 0.065 0.002 0.009 0.025 0.0002 0.004 0.017 0.019 0.000 0.010 0.043 0.0003 0.002 Number of observations 57(4) 57(4) 57(4) 36(4) 57(4) (number of periods) Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. market premium may be a weak proxy for the overall efficiency of the price system. International technology transmission is extremely hard to measure as well, perhaps downwardly biasing estimates for this channel and overstating the others. Future research should seek to improve on the measures used in this study. The important role of investment in physical capital poses a theoretical chal- lenge. Some theories about gains from trade predict positive effects of openness on the rate of return to capital, but some of these effects should be captured either by the price distortions or technological transmission channel. Furthermore, theo- ries based on dynamic gains from technology transmission and efficiency improve- ment focus on the improvement of the overall productivity of factors, rather than on accelerated accumulation. If specialization is limited by the extent of the mar- ket, under increasing returns to scales trade openness should allow entrepreneurs to undertake previously unprofitable investments. Similarly, if trade liberalization involves procompetitive effects, the entry of new firms may entail large fixed capi- tal costs. Applying such theories to the study of the growth effects of trade open- ness may provide useful insights into the nature of dynamic gains from trade. REFERENCES Ades, Alberto F., and Edward L. Glaeser. 1999. "Evidence on Growth, Increasing Returns and the Extent of the Market." Ouarterly Journal of Economics 114(3):1025-46. Ades, Alberto F., and Rafael di Tella. 1999. "Rents, Competition and Corruption." American Economic Review 89(4):982-993. Aitken, Brian J., and Ann Harrison. 1999. "Do Domestic Firms Benefit from Direct Foreign Investment? Evidence from Venezuela." American Economic Review 89(3):605-18. Alesina, Alberto, and Roberto Perotti. 1996. "Income Distribution, Political Instability and Investment." European Economic Review 40(6):1202-29. 424 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Alesina, Alberto, and Romain Wacziarg. 1998. "Openness, Country Size and the Gov- ernment." Journal of Public Economics 69(3):305-21. Alesina, Alberto, Enrico Spolaore, and Romain Wacziarg. 2000. "Economic Integration and Political Disintegration." American Economic Review 90(5):1276-96. Baldwin, Richard E., and Elena Seghezza. 1996. "Testing for Trade-Induced, Investment- Led Growth." NBER Working Paper 5416. National Bureau of Economic Research, Cambridge, Mass. Barro, Robert. 1991. "Economic Growth in a Cross-Section of Countries." Quarterly Journal of Economics 106(2):407-43. Barro, Robert J., and Jong-Wha Lee. 1993. "International Comparisons of Educational Attainment." Journal of Monetary Economics 32(3):363-94. .1994. Data Set for a Panel of 138 Countries. Revised January 1994, unpublished, NBER. Available online at www.nber.org/pub/barro.lee. Barro, Robert, and Xavier Sala-i Martin. 1992. "Public Finance in Models of Economic Growth." Review of Economic Studies 4(59):645-61. 1995. Economic Growth. New York: McGraw-Hill. 1997. "Technological Diffusion, Convergence, and Growth." Journal of Eco- nomic Growth 2(1):1-26. Ben-David, Dan. 1993. "Equalizing Exchange: Trade Liberalization and Income Con- vergence." Quarterly Journal of Economics 108:653-79. Blomstrom, Magnus, Robert E. Lipsey, and Mario Zejan. 1996. "Is Fixed Invest- ment the Key to Economic Growth?" Quarterly Journal of Economics 111(1):269- 76. Caselli, Francesco, Gerardo Esquivel, and Fernando Lefort. 1996. "Reopening the Con- vergence Debate: A New Look at Cross-Country Growth Empirics." Journal of Eco- nomic Growth 1:363-389. Dollar, David. 1992. "Outward Oriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 1976-1985." Economic Development and Cultural Cbange 40:523-44. Easterly, William. 1989. "Policy Distortions, Size of Government, and Growth." NBER Working Paper 3214. National Bureau of Economic Research, Cambridge, Mass. . 1993. "How Much Do Distortions Affect Growth?" Journal of Monetary Eco- nomics 32(2):187-212. Edwards, Sebastian. 1992. "Trade Orientation, Distortions and Growth in Developing Countries." Journal of Development Economics 39(1):31-57. Fischer, Stanley. 1993. "The Role of Macroeconomic Factors in Growth." Journal of Monetary Economics 32(3):485-512. Frankel, Jeffrey A., and David Romer. 1999. "Does Trade Cause Growth?" American Economic Review 89(3):379-99. Grossman, Gene M., and Elhanan Helpman. 1991. Innovation and Growth in the Glo- bal Economy. Cambridge: MIT Press. Harrison, Ann, and Gordon Hanson. 1999. "Who Gains from Trade Reform? Some Remaining Puzzles." Journal of Development Economics 59(1):125-54. Harrison, Ann, and Ana Revenga. 1995. "The Effects of Trade Policy Reform: What Do We Really Know?" NBER Working Paper 5225. National Bureau of Economic Re- search, Cambridge, Mass. Wacziarg 425 Heston, Alan, and Robert Summers. 1995. Penn World Table Version 5.6. Available online at www.pwt.econ.upenn.edul. Krugman, Paul. 1995. "Growing World Trade: Causes and Consequences." Brooking Papers on Economic Activity 1. Brookings Institution, Washington, D.C. Leamer, Edward. 1988. "Measures of Openness." In R. E. Baldwin, ed., Trade Policy Issues and Empirical Analysis. National Bureau of Economic Research Conference Report Series. Chicago: University of Chicago Press. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Mauro, Paolo. 1995. "Corruption and Growth." Quarterly Journal of Economics 110(3):681-712. Murphy, Kevin M., Andrei Shleifer, and Robert W. Vishny. 1989. "Industrialization and the Big Push." Journal of Political Economy 97(5):1003-26. Pritchett, Lant. 1996a. "Measuring Outward Orientation in LDCs: Can It Be Done?" Journal of Development Economics 49(2):307-35. - 1996b. "Where Has All the Education Gone?" Policy Research Working Paper 1581. World Bank, Policy Research Department, Poverty and Human Resources Division, Washington, D.C. Rodrik, Dani. 1998a. "Globalization, Social Conflict and Economic Growth." World Economy 21(2):143-58. . 1998b. "Why Do More Open Countries Have Larger Governments?" Journal of Political Economy 106(5):997-1032. Rodrik, Dani, and Francisco Rodriguez. 2000. "Trade Policy and Economic Growth: A Skeptic's Guide to the Cross-National Evidence." In Ben Bernanke and Kenneth Rogoff, eds., NBER Macroeconomics Annual 2000. Cambridge, Mass.: MIT Press. Sachs, Jeffrey D., and Andrew Warner. 1995. "Economic Reform and the Process of Global Integration." Brookings Papers on Economic Activity 1. Brookings Institution, Washington, D.C. Tavares, Jose, and Romain Wacziarg. 2001. "How Democracy Affects Growth." Euro- pean Economic Review 45(8):1341-79. Taylor, Alan M. 1998. "On the Costs of Inward-Looking Development: Price Distortions, Growth, and Divergence in Latin America." Journal of Economic History 58:1-28. Vamvakidis, Athanasios. 1998. "Regional Trade Agreements versus Broad Liberaliza- tion: Which Path Leads to Faster Growth? Time-Series Evidence." IMF Working Paper No. 40, March 1998. Wacziarg, Romain. 1997. "Trade, Competition, and Market Size." Harvard University, Department of Economics, Cambridge, Mass. . 1998. "Measuring the Dynamic Gains from Trade." Policy Research Working Paper 2001. World Bank, Development Economics, Development Prospects Group, Washington, D.C. World Bank. 1996. Global Economic Prospects and the Developing Countries 1996. Washington, D.C. Young, Alwyn. 1991. "Learning by Doing and the Dynamic Effects of International Trade." Quarterly Journal of Economics 106:369-405. Zellner, Arnold, and Henri Theil. 1962. "Three-Stage Least Squares: Simultaneous Esti- mation of Simultaneous Equations." Econometrica 30(1):54-78. 426 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 APPENDIX A. TABLE A-1. List of Countries OECD Asia Latin America Africa Australia Cyprus Argentina Congo, Dem. Rep. Austria India Barbados Ghana Belgium Israel Brazil Kenya Canada Jordan Colombia Malawi Finland Korea Costa Rica Mauritius France Malaysia Dominican Republic Sierra Leone Germany, West Myanmar El Salvador South Africa Greece Pakistan Guyana Tanzania Ireland Philippines Mexico Gambia, The Italy Singapore Paraguay Tunisia Japan Sri Lanka Peru Zambia Netherlands Syria Venezuela, R.B. de New Zealand Thailand Norway Portugal Spain Sweden Switzerland Turkey United States United Kingdom APPENDIX B. DATA SOURCES AND DESCRIPTION Growth. Source: Heston and Summers (1995). Description: Growth rate of pur- chasing power parity (Ppp) adjusted GDP (percentage points). Import duties as a percentage of total imports. Source: IMF. Description: Import duties in local currency as a percentage of total imports in local currency (per- centage points). Pre-Uruguay Round nontariff barrier coverage. Source: UNCTAD. Description: Coverage rate of nontariff barriers pre-Uruguay Round (percentage points). Sachs and Warner liberalization status. Source: Sachs-Warner (1995). Description: For each year, a dummy variable was constructed based on the years of liberaliza- tion in Sachs and Warner (1995). Takes a value of 1 for liberalized countries and O for closed countries. The data were averaged over the relevant five-year subperiods. Manufactured exports share. Source: World Bank. Description: Share of manu- factured goods in merchandise exports (percentage points). FDI ratio. Source: IMF. Description: Ratio of gross foreign direct investment in- flows to GDP (percentage points). Democracy. Source: Gastil (Freedom in the World Reports). Description: Index of how democratic institutions are (regular elections, broad franchise, wide access to office, and relevance of elected officials). Takes values from 0 (nondemocracy) to 1 (country with fully developed democratic institutions). Wacziarg 427 Initial income. Source: Heston and Summers (1995). Description: Real GDP per capita in a given year (PPP adjusted) (log of per capita GDP in dollars). Human capital. Source: Barro and Lee (1993). Description: Average years of secondary and higher education in the total population over age 25. Secondary school completion rate. Source: Barro and Lee (1993). Description: Percentage of the total population that has completed secondary school. Macroeconomic policy quality. Source: World Bank and INIF. Description: In- dex of macroeconomic policy quality. Constructed by ranking countries accord- ing to their public debt to GDP ratio, deficit to GDP ratio, and growth of Ml net of total output growth and assigning values from 1 to 10 to each decile, then aver- aging the three resulting indicators. Index also ranges from 1 to 10. Higher num- bers signal better policies. Black market premium. Source: Tavares and Wacziarg (2001) data set, initially World Currency Yearbook and IMF. Description: Black market premium on the official exchange rate (black market rate minus official rate/official rate as a percentage). Public consumption. Source: Heston and Summers (1995). Description: Share of government consumption of goods and services in GDP, excluding transfers and public investment (percent). Population over 65. Source: Barro and Lee (1994). Description: Share of popu- lation aged over age 65 in the total population (percent). Population over 15. Source: Barro and Lee (1994). Description: Share of popu- lation over age age 15 in the total population (percent). Terms of trade shocks. Soturce: Tavares and Wacziarg (2001), initially from the World Bank. Description: Growth rate of manufactured export prices minus growth rate of manufactured import price (percent). Population. Source: Barro and Lee (1994). Description: Country population; log of population. Population density. Source: Barro and Lee (1994). Unit: Thousands of people per million square kilometers. Ethnolinguistic fractionalization. Source: Mauro (1994). Description: Probability that two randomly selected people from a given country will not belong to the same ethnolinguistic group. Postwar independence. Souirce: Barro and Lee (1994). Description: Takes on a value of 1 if the country gained independence after World War II and 0 otherwise. APPENDIX C. MODEL DETAILS TABLE C-1. Baseline Specification of the Structural System (57 countries, 1970-89) Government Manufactured Macro policy Growth Openness Distortions consumption exports Investment FDI quality Intercept 10. 5977 -53.8507 39.7204 57.7178 -75.7958 27.4932 1.1773 5.9803 (4.70) (-16.55) (0.83) (10.58) (-6.94) (3.72) (5.73) (5.14) Endogenous variables Trade policy - - -0.3445 0.1539 0.6345 0.3173 0.0450 0.0267 (-0.63) (3.73) (4.59) (6.72) (4.01) (2.19) Growth - 0.3215 - - - - -- (10.44) Log initial -1.6721 6.5481 -2.5352 -4.4392 7.2888 1.0034 - 0.1869 income (-5.81) (17.55) (-0.43) (-9.58) (5.18) (1.56) (1.42) Distortions -0.0066 - - 0.0084 -0.0131 -0.0101 -0.0008 -0.0016 00 (-9.08) (20.19) (-5.49) (-7.15) (-3.60) (-1.90) Government -0.0425 - 3.8212 - - - -0.0545 -0.1265 consumption (-1.57) (8.13) (-4.15) (-8.25) Manufactured 0.0036 - - - - exports (0.45) Investment 0.1425 - - - - rate (6.86) FDI 0.3203 - - - - - - (4.68) Macro policy 0.4887 - - - - 1.0265 quality (4.22) (6.97) Exogenous variables (instruments) Male human 0.4812 - - - - capital (1.59) Female human -0.3867 - - - - - - capital (-1.39) Secondary school - - - - 0.2907 enrollment (3.09) Democracy index - - -51.9867 (-4.69) Island dummy - -3.0493 - - - - 0.9878 (-2.37) (4.74) Log of land area - -0.8879 - (-2.20) Terms of - -7.1484 71.5887 - - - - -1.3179 trade shocks (-4.97) (2.87) (-1.86) Log population - 0.4201 - -0.9107 5.2154 (0.79) (-4.52) (5.68) Population - - -0.0253 -0.0033 0.0189 - - density (-3.37) (-5.87) (5.22) Population - - - 16.2617 - -88.3531 - over 65 (1.54) (-5.45) Population - - - 1.6533 - -38.3206 - under 15 (0.29) (-5.16) Ethnolinguistic - - - 0.0377 - -0.0471 - -0.0056 fractionalization (3.23) (-3.02) (-1.45) Postwar - - - - - - 0.9285 independence (3.96) R2 0.251 0.287 0.551 0.526 0.189 0.235 0.276 0.284 0.501 0.522 0.443 0.556 0.330 0.356 0.362 0.284 0.412 0.314 0.560 0.538 0.104 0.272 0.419 0.531 0.487 0.526 0.614 0.622 0.284 0.231 0.344 0.362 Note: Numbers in parentheses are t-statistics based on heteroscedastic-consistent (White robust) standard errors. Because each equation is estimated for four time periods, with estimated parameters constrained to equality across periods, the table reports R2 statistics corresponding to each of these time periods. Source: See Appendix B. i I i i i i I i t I i i i I THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 43 I-449 Ownership and Growth Tborvaldur Gylfason, Tryggvi Tbor Herbertsson, and Gylfi Zoega This article suggests how state enterprises can be incorporated into the theoretical and empirical growth literature. Specifically, it shows that if state enterprises are less effi- cient than private firms, invest less, employ less skilled labor, and are less eager to adopt new technology, then a large state enterprise sector tends to be associated with slow economic growth, all else remaining the same. The empirical evidence for 1978-92 indicates that, through a mixture of these channels, an increase in the share of state enterprises in employment by one standard deviation could reduce per capita growth by one to two percentage points a year from one country to another. The debate over private versus public enterprise has played an important part in the history of economic ideas and of the world. State ownership of all factors of production was a cornerstone of communism, as practiced in the former Soviet Union and its satellites. Even under capitalism, the state (especially European states) has sometimes been deeply involved in economic affairs. The state in developing economies has been particularly inclined to take a prominent role in producing goods and services and allocating resources to investment and other economic needs. Despite valiant efforts by many governments in recent years to get bureau- crats out of business, state enterprises remain prominent around the world. The unweighted average share of state enterprises in nonagricultural economic ac- tivity in 40 developing economies reporting to the World Bank (1995, table A2) was 13 percent in 1991, the same as in 1978. The comparable figure for eight industrial countries in 1988 was 7 percent, down from 8 percent in 1979. The unweighted average share of state enterprises in gross domestic investment in 55 developing countries was 18 percent in 1991, down from 23 percent in 1978. For 10 industrial countries it was 11 percent in 1988 and 13 percent in 1978. Thorvaldur Gylfason is with the University of Iceland and the Center for Business and Policy Stud- ies, Stockholm. Tryggvi Thor Herbertsson is with the Institute of Economic Studies, University of Iceland. Gylfi Zoega is with Burkbeck College, University of London, and the Institute of Economic Studies, University of Iceland. The authors would like to thank T6r Einarsson, Edmund Phelps, Ron Smith, Andreas Worgotter, two anonymous referees, and the editor for their valuable comments on earlier drafts. The usual dis- claimer applies. |© 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 431 432 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 I. A MODEL WITH STATE ENTERPRISES Recent worldwide interest in privatization derives, in part, from empirical evi- dence that seems to indicate that private enterprise is generally more efficient than state enterprise. This evidence was reviewed in detail in World Bank (1995). Phelps (1993) provides a useful classification by suggesting five main reasons for the superior efficiency of private enterprise. Private firms may be more entre- preneurial. Managers of private firms may find it easier to act on their intuition about what products or production processes will be successful. State enterprises may be more susceptible to pressure from interest groups, whereas private firms can focus solely on maximizing profits. Private investors generally have a long time horizon for acquiring assets that can be sold, whereas politicians' electoral assets tend to be more fleeting. Last, private firms may have more difficulty get- ting public assistance, so the penalty for failing to maximize profits is harsher, though the fruits of success may also be sweeter. In his presidential address at the 110th meeting of the American Economic Association, Arnold Harberger (1998, 23) airs similar views: "In most countries state-owned enterprises oper- ate under a series of constraints that seriously get in the way of real cost minimi- zation in a comparative-static sense and real cost reduction in a dynamic sense." Even so, several empirical studies have reported mixed evidence of the relative efficiency of public and private firms (see Stiglitz 1988 for a review of this evi- dence). The dearth of unambiguous empirical evidence is not surprising in view of the long-standing debate on the relative merits of public and private enter- prise, especially when the inefficiency that can arise from principal-agent (owner- manager) relations in private industry is taken into consideration.' The public sector has no monopoly on inefficiency in production. But if transferring state property to more productive uses in the private sector enhances efficiency, by replacing soft budget constraints with harder ones, for example, then the composition of corporate ownership would be expected to play a role in generating and sustaining long-term economic growth. This is a direct implication of the theory of endogenous growth: virtually anything that increases static efficiency stimulates growth. This result follows particularly clearly from endogenous growth models featuring constant returns to capital (the so-called AK model where A denotes the output to capital ratio, which may be viewed as a measure of macroeconomic efficiency, and K denotes capital stock). In these models, the long-run rate of growth of output per head equals the multi- ple of the saving rate, s, and efficiency, A, less the depreciation rate, 6: g = sA - 5. Any policy undertaking or external event that increases static efficiency by increas- ing the amount of output that can be made from given capital thus also increases the rate of economic growth, permanently. In the neoclassical theory of economic growth, increasing efficiency increases economic growth, possibly for a long time, 1. Important contributions to this debate include Vickers and Yarrow (1988), Laffont and Tirole (1993), and Stiglitz (1994), among others. Gylfason, Herbertsson, and Zoega 433 but eventually the rate of growth returns to its exogenously determined initial equilibrium value. Either way, this link between efficiency and growth explains why, for example, education is good for growth. It also explains why liberaliza- tion, stabilization, and-yes, why not?-privatization are probably also good for growth. Which brings up the question of private or public ownership and economic growth. Using an index of private ownership from Milanovic (1989), Palia and Phelps (2000) find for a sample of 43 countries that a strong private sector is good for growth. Rather than appealing to the simple framework of an AK-type model, which might mask the more complex interactions that have been debated in the literature on the efficiency of state enterprises (see Rama 1999), this article presents a more fully articulated model. It goes beyond the AK model to show how efficiency can be related to growth by incorporating into an endogenous growth framework the idea that state enterprises may be less inclined to invest and employ skilled labor and less innovative than private firms. State enterprises sometimes fail to adopt new products and processes invented in the private sec- tor, reducing their efficiency. The article places this hypothesis within a clear conceptual framework, to facilitate discussion of how growth may be affected by the form of corporate ownership. The model developed here bridges the ana- lytical literature on the static efficiency of state enterprises and the empirical tests of dynamic efficiency and economic growth reported below. This modeling strat- egy rests on a microeconomic foundation and derives testable macroeconomic hypotheses. These hypotheses about the linkages among state enterprises, efficiency, in- vestment, education, and economic growth are tested on new data from the World Bank (1995) on the share of state enterprises in employment for a cross-section of 34 developing economies for 1978-92. A significant inverse relationship emerges between the size of the state enterprise sector and economic growth, partly through investment and partly through education. In a similar attempt to find a relationship between the size of the state enter- prise sector and economic growth, the World Bank (1995, 52) reports that "there was insufficient time-series data on state enterprise sector size for enough coun- tries over a sufficiently long time to conduct satisfactory growth regression analy- sis." Even so, the World Bank concludes that "the microeconomic evidence, the experience of the centrally planned economies, and the strong negative effect SOEs [state enterprises] have on fiscal deficits all collectively support the premise that large SOE sectors can hinder growth. Moreover, because SOE sectors tend to be larger in low-income countries, SOEs are likely to be most costly in the coun- tries that can least afford them" (50-51). The empirical findings reported here support that conclusion. It is important in this kind of analysis to distinguish the adverse growth ef- fects of state involvement in production and in the allocation of resources from any effects of big government on growth. The effects of government spending, taxes, and transfers on growth depend on how the government spends tax rev- 434 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 enue (see Barro 1990). It is possible for increased government expenditure to boost growth despite a concurrent negative relationship between the size of the state enterprise sector and economic growth (on education, for example, see Glomm and Ravikumar 1992). The first step is to embed state enterprises in one model of economic growth to demonstrate that ownership can matter for growth and thus belongs in growth theory. In this model, derived from Romer (1990), growth arises from an ex- panding variety of inputs. Because the hypotheses will be tested using data for developing economies, it is natural to think in terms of the adoption or adapta- tion of leading-edge technology rather than the invention of new technology. The model is intended only to illustrate some-but by no means all-of the chan- nels through which efficiency influences economic growth. The intention is to show by example how state enterprises can be incorporated into the growth lit- erature and the standard determinants-of-growth regression framework. The empirical tests presented later are not meant as tests of the particular model se- lected as the vehicle, because other points of entry, such as the AK model, could as well have been chosen. In this version of the Romer model, output is produced in both private and public sectors, and in both sectors, output levels are set to maximize profits. Unlike private firms, however, state enterprises have to satisfy further constraints and objectives (regulations on work hours, on where to buy inputs, and the like) that affect labor productivity and the propensity to adopt new inputs. These firms are thought of as being run by bureaucrats on whom political authorities have imposed multiple goals and constraints. The model features full employment, free entry in the competitive private sec- tor, and infinite substitutability between private and public output. With free entry, the inherent static and dynamic inefficiency of state-owned firms means that they must be kept afloat by a government subsidy financed by a tax on pri- vate firms. State enterprises may produce goods and services (such as cars and computers, as in France, and banking services, as in India) that are no different from similar goods and services produced by more competitive, privately owned companies, but cost more to produce. So why do state enterprises exist? Because of their size, inefficient state enterprises may be important for the local economy: firms that employ workers that are not easily employable elsewhere are tempt- ing targets for politicians striving to gain popularity with job-saving measures. An even stronger motivation for public ownership is the strategic importance of certain industries, such as aircraft, utilities, and armaments. Other examples abound, especially in developing economies, where export and import-competing industries are often of great importance to the local economy, yet face stiff for- eign competition. In sum, this modeling strategy is intended to draw the attention of growth analysts to public versus private ownership and to shed some light on the conse- quences of state enterprises, which, while competing with the private sector at home or abroad, are saddled with an inefficient cost structure and social respon- Gylfason, Herbertsson, and Zoega 435 sibility for the local economy. The model is illustrative; it is not intended as a general framework for studying the raison d'etre of state enterprises. Preferences and Utility Maximization Consumers derive utility from the consumption of final output, which is sold in a perfectly competitive market. Public output Ys and private output YP are perfect substitutes. Total consumption equals C = CP + Cs. Though indifferent between consuming private and public output, the typical consumer maximizes the present discounted value of lifetime utility from total consumption. As in Blanchard ( 1985), workers face a constant probability of death 7, and new cohorts are continuously being born. This prevents Ricardian equivalence. This matters because inefficient state enterprises are often responsible for mounting public debt, which may reduce saving. Preferences are described by an isoelastic utility function, u = cl-'1 /(1 - 1/a), where c is per capita consumption and a is the elasticity of intertemporal substitu- tion. This gives the following Euler equation (equation 1) for the optimal aggre- gate consumption profile (C = cL, where L, the total labor force, is fixed). (1) C= f(r-p)-{Hp+7r-(c,-1)r] } W where r is the real interest rate, p is the pure rate of time preference, and W is total wealth, which consists of the total value of firms and outstanding public debt, D. This debt has been accumulated to sustain the operation of state enter- prises in the past. Technology and Profit Maximization Both sectors, private and public, use labor and other inputs, which are produced solely by private firms. With Romer (1990) as a starting point, production tech- nology in the two final-goods sectors is shown in equations 2 and 3. (2) YP = AP(ePLP) (XP) (3) Ys=As(esLs) j(Xj where Li is employment in sector i, ei is the efficiency of labor in that sector, XI, is the use of input j in that sector, i = s, p. N is the number of inputs produced and used in the private sector, and p is the probability that a new input will be adopted by state enterprises. Possible sources of (static) inefficiency in the public sector are: 1. Public enterprises may be less efficient-waste more resources-than pri- vate firms. This means that As < AP in equations 2 and 3. Managers of state enterprises may not have the same incentive as management in private firms 436 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 do to organize production efficiently and to invest in sound projects, partly because the penalty of failure is less threatening when the state coffers are within reach and partly because the rewards of success are typically smaller. 2. State firms may be less efficient in organizing labor within the firm and may employ less well educated labor than private firms. Therefore, e5 < eP. Wages are generally lower in state enterprises than in the private sector (see Gyourko and Tracy 1988). Also, wage setting in the public sector tends to be less flex- ible and thus less incentive compatible-less conducive to increased work effort and improved efficiency-than in the private sector (World Bank 1995). Moreover, in many countries, the public sector tends to be overstaffed be- cause state enterprises do not make hiring and firing decisions solely on the basis of profitability. Workers in state enterprises generally enjoy greater protection from cyclical layoffs than do workers in the private sector. 3. Each newly invented input is bought by private firms, but in the public sec- tor this occurs with probability p. The hypothesis is that state enterprises are not as innovative as private firms-and so not as likely to invest in new ma- chinery and equipment that embodies new and productive technology (Phelps 1993). Thus there is a fixed probability p < 1 that a new input will be adopted by state enterprises. For this reason, fewer types of inputs-less high-tech capital-may be used in the public sector than in the private sector: pN < N. With free entry, private firms enter until average profits in the sector are driven to zero. The question of the viability of public enterprises is bound to arise in light of the constant returns to scale nature of the production technology in both sectors. This issue is resolved by assuming that state enterprises receive a sub- sidy s per unit of output from the government financed by a tax t on the output of private firms. The effective subsidy s + t is then equal to the difference be- tween average long-run costs in public and private enterprises and can be writ- ten (see equation 4) for the case of N = 2 and p = 1. (4) s+t =e±s _l H ws ) 1 I JAP)§ where a a +| and P is the real price of an input. The government budget constraint is added next to solve for the subsidies and taxes (equation 5). (5) sYs + rD = tYP The effective subsidy is a decreasing function of eS and As and an increasing function of eP and AP as expected. If es = eP and As = AP, then s + t = 0 by equa- tion 4. This system of taxes and subsidies is the basis for the continued existence of state enterprises, given their presumed inefficiency. Firms in both sectors decide on employment and the use of other inputs to maximize profits. The first-order conditions for labor (in efficiency units) and other inputs are shown in equations 6 and 7. Gylfason, Herbertsson, and Zoega 437 (6) w 1-)YP (9-t) =(1 _oe) Y(1 +s) (7) XIP =ePLpA (1 t) =e` A where Pj is the real price of input j. The price of inputs is set by the monopolists that invented them, but the wage w is determined by supply and demand in labor markets, so that L=(l_a) Y (_-t)+_ _(_+_) where L, the labor force, is fixed. ePw eSwI Input Pricing, Output, and Growth Each intermediate input is produced by its inventor, who has a permanent mo- nopoly in production. The production technology involves turning one unit of the final good into a unit of input at zero cost. The monopolists' profits can be written as (Pi - 1)Xj where Pi is the (monopoly) real price of the input in terms of final goods. The monopolist then sets the price of the input to maximize current profits by taking factor demand (equation 7) into account; no intertemporal considerations enter the pricing decision. The monopoly price is P, = 1/a. Plug- ging this price into factor demand equations 7 yields the steady-state value of a new invention, assuming a constant rate of real interest, r (equation 8). (8) V= [ePLP(AP[1-t])b + pesLs(As[l±s])1l]( a Jt1 In a steady state with a growing variety of inputs and free entry, the expected value of a new invention has to equal the cost of inventing a new input, q. The total value of firms is therefore equal to Nq7. This gives the equilibrium interest rate and, through equation 1, the rate of economic growth (equation 9) g = jfeP( e -)L(AP[1-t>])j+pevL(As[l±sl])1 )0-p -(((p9) -(7-1)r)i) Nq7 +D [1-1 2_ where v = Lsl(LS + LP), Ls, and LP are determined by equation 6 and 0= oe". Apart from the usual effects of changes in the cost of innovation, 7I, the size of the total labor force, L = Ls + LP, and the rate of time preference, p, on growth, the equation suggests that 1. The rate of growth is a decreasing function of the size of the state sector, v, as long as AP(1 - t)/As (1+ s) > (pes/eP)1-1, because the transfer of labor from the private sector to the public sector reduces demand for inputs. Thus AP( 1 - t) > As( 1 + s) and eP > el are a sufficient but not necessary condition 438 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 for the expansion of the state enterprise sector from one time or place to another to reduce economic growth, as long as p < 1. This is the main hypothesis. 2. The rate of growth is an increasing function of the productivity of labor in state enterprises, eS (as well as in private firms, eP), for given taxes and subsidies. 3. The rate of growth is an increasing function of the level of technology and the efficiency of organization in state enterprises, As (as well as in private firms, AP), for given taxes and subsidies. 4. The rate of growth is an increasing function of the probability that state enterprises adopt newly invented inputs, p, which is interpreted as a sign of their willingness to invest. 5. The rate of growth is a decreasing function of public debt, D, which is as- sumed to have been, at least in part, accumulated by state enterprises in the past. A higher level of debt increases consumption and hence leaves less output for investment in research and development of new types of inputs. II. EMPIRICAL EVIDENCE Under ideal conditions, the next task would be to gather the data and test all the hypotheses derived from growth equation 9. This is an impossible task, how- ever, because several of the variables that drive economic growth in this model cannot be directly observed: the efficiency of organization (As and AP), the productivity of labor (eS and eP), and the probability that state enterprises adopt newly invented inputs (p). Hypothesis 1 can be tested directly by estimating the partial correlation be- tween the state enterprises' share in the labor force, v = LsI(LS + LP), for which data are widely available from the World Bank (1995), and economic growth, controlling for other potential determinants of growth. If the conjecture that this correlation is negative is confirmed, that is an indication that the state enterprise sector is less well organized, less efficient, or less innovative than the private sector in such proportions that AP(1 - t)IAs(1 + s) > (pesleP)1-i1. Hypotheses 2-5 can be tested only indirectly, however, because of lack of data. To test hypotheses 2 and 3, labor productivity and efficiency of organization are assumed to vary directly with investment and the education of the labor force. To test hypothesis 4, the propensity to invest is assumed to reflect, in part, the willing- ness to adopt newly invented inputs. This is the case when the intermediate inputs are capital goods. Therefore, if state enterprises are generally more prone than private firms to waste resources on unproductive investments ("white elephants") and to divert government spending from social needs, including education (Mauro 1998), and less willing to adopt new inputs, as conjectured, then this is an addi- tional link between the size of the state enterprise sector and economic growth. Together, hypotheses 1-4 imply that increased state enterprise activity can hurt economic growth directly as well as indirectly through investment and edu- Gylfason, Herbertsson, and Zoega 439 cation. A fifth hypothesis is that the impact of investment on growth varies in- versely with the size of the state enterprise sector. To test hypothesis 5, state enterprises are assumed to bear responsibility for a substantial part of public external indebtedness. The empirical results reported below need to be viewed in the light of these qualifications. A Preview These hypotheses are tested using cross-sectional data from the Penn World Tables (see Summers and Heston 1991) and the World Data Bank (World Bank 1997) covering 1978-92 (1978-91 for state enterprises). Table 1 reports sum- mary statistics for the share of state enterprises in employment (SOE/Labor) and in nonagricultural GDP (SOE/GDP) and for the external debt of state enterprises as a proportion of GDP (SOE/Debt). The share of state enterprises in employment was remarkably steady, averag- ing 12 percent in both the first and last years of the period.2 Several countries significantly downsized their state enterprise sector. Chile reduced the sector's share in employment from 4 percent to 1 percent and its share in nonagricul- tural GDP from 12 percent to 8 percent. Argentina reduced the employment share from 4 percent to 2 percent and the share in GDP from 6 percent to 2 percent, and Botswana reduced the public sector's share in employment from 3 percent to 2 percent and the sector's share in GDP from 9 percent to 6 percent. At the other end of the spectrum, Ghana increased the share of state enterprises in employment from 29 percent to 45 percent, while their share in GDP declined from 8 percent to 7 percent. It would be unwise, however, to ascribe rapid growth in Chile since the mid- 1980s and in Argentina since the early 1990s until recently in part to privatization (or, for that matter, to ascribe slow growth in Ghana in part to the failure to privatize). For one thing, causation can run both ways. Although the model sug- gests a link from privatization to growth, and privatization was an important ingredient of the reforms that started in Chile in the 1970s and in Argentina in the 1980s, it also seems reasonable to suppose that brisk growth in Chile and Argentina may have helped create conditions favorable to further privatization and other reforms. Even so, the high unemployment that accompanied the rapid growth in Argentina and Chile, by exerting political pressure not to endanger jobs in the state sector, seems likely to have weakened this reverse linkage from growth to privatization. By the same token, sluggish growth and high unemployment in Ghana (and elsewhere, no doubt) contributed to the expansion of employment in the state enterprise sector, even if the sector's share in GDP was declining. The main point, however, is this: if private enterprise is good for growth, as hypothesized, that does not mean that growth is not good for private enterprise. The same argument applies to other potential determinants of economic growth: trade, investment, education, and so on. The discussion that follows emphasizes 2. Due to gaps in the data, the first year and the last year vary from country to country. 440 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 TABLE 1. Summary Statistics Number of Mean SE Min Max countries SOE/Labor 0.13 0.14 0.008 0.698 41 SOE/GDP 0.15 0.14 0.013 0.717 76 SOE/Debt 0.07 0.06 0.000 0.289 82 Note: See World Bank (1995) for definition of variables. the link from privatization to growth, even though the relationship between the two may well be more complex. Correlation Analysis Correlation analysis can further illuminate the relationships between the size of state enterprises and growth. Some key bivariate correlations are between the relative size of the state sector, measured by state enterprises' share of employ- ment (SOE/Labor), and the accumulation of physical and human capital-two key determinants of economic growth. A scatterplot of state enterprise employment and the share of investment in GDP across countries, both measured as averages over the period 1978-1991/ 92, shows the relationship to be economically and statistically significant (fig- ure 1).3 The regression line in figure 1 is based on robust estimation to reduce the weight of potential outliers (the same applies to figures 2-3). An increase in the employment share of state enterprises by one standard deviation is associ- ated with a decrease in investment of 4.5 percent of GDP, all else remaining the same. The correlation r is -0.51 (t = 3.6). Similar results (r = -0.42, t = 2.9) obtain when the initial rather than average value is used to measure SOE/Labor. This suggests that causation runs from SOE/Labor to investment rather than the other way round. This result supports the hypothesis that state enterprises are less inclined than private firms to invest in new machinery and equipment and to adopt new technology and may thus impede economic growth. A second scatterplot shows the correlation of state enterprise employment during 1978-91 and the rate of enrollment in secondary schools in the base year, 1978, a commonly used measure of education in the growth literature (figure 2). An increase in the employment share of state enterprises of one standard devia- tion goes along with a decrease in secondary school enrollment of 1.5 percent- age points, all else remaining the same. The correlation is -0.58 (t = 4.1). The pattern is similar (r = -0.51, t = 3.4) when the initial rather than average value is used to measure SOE/Labor. This pattern seems consistent with the hypothesis that state enterprises are less inclined than private firms to employ skilled labor and perhaps less likely to adopt new technology and thus may inhibit economic 3. SOE/Labor is exceptionally large in Guinea (70 percent). This outlier is excluded from figures 1-3 and from equations 2 and 3 in table 2. Gylfason, Herbertsson, and Zoega 441 FIGURE 1. State Enterprises and Investment (percent) KOR 30 F ~~~ID~AW DZ rL CITLUR C 20 THA BWA ._ PER ~~~~~~~~~~~~~GRD en TGO E p 10 MUS C/) M~~~~~ ~ ~~~~LI COG - Q ZAR EGY SEN GHA MDG SLE 0 0 10 20 30 40 Share of State Enterprises in Employment 1978-91 Source: Penn World Tables and the World Bank. Note: Country abbreviations are defined in the Appendix. growth. Other interpretations are also conceivable; for example, low standards of education may generate unemployment and thus exert pressure on the au- thorities to create jobs through state enterprises. In sum, the data suggest that state enterprises may slow economic growth by discouraging investment (figure 1) and education (figure 2). Figure 3 confirms this: it shows an inverse correlation (-0.35, t = 2.2) between state enterprises' employment share and economic growth across countries. An increase in state enterprises' employment share of one standard deviation is associated with a decrease in the annual rate of economic growth of about 1 percent. The eco- nomic and statistical significance of this correlation is preserved when economic growth is regressed on the state enterprises' employment share and initial GDP using ordinary least squares (OLS) and when the initial rather than average value is used to measure SOE/Labor (r = -0.29, t = 1.8). These correlations suggest that a small state enterprise sector (where state enterprises account for 5 percent or less of employment) can be associated with both rapid growth, as in Indonesia, the Republic of Korea, Taiwan (China), and Thailand, and slow or even negative growth, as in Bolivia, Madagascar, and Peru. A large state enterprise sector, however, generally goes hand in hand with slow 442 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 FIGURE 2. State Enterprises and School Enrollment (percent) KOB PHL 60 TTO 0 p)EPpG o2 CHL LKA x MU EGY Cs ~ ~ ~ W cCO a) E 2- 40 c BOL GHA wT 0 3RA O THA D,Z,8~~D TGO 8 20 K ENCMR GI\ZM Us SLE SEN GNE\ MWI TZA BDI *0 0 10 20 30 40 Share of State Enterprises in Employment 1978-91 Source: The World Bank. Note: Country abbreviations are defined in the Appendix. growth, as in Ghana and Zambia. Except for Sri Lanka, all countries whose state enterprise sector accounted for 10 percent or more of total employment had economic growth of less than 2 percent a year on average over the period (a majority had negative growth). Does this inverse correlation between the employment share of state enter- prises and economic growth stem from inefficiency in the state enterprise sector, as hypothesized? Or does the size of the state enterprise sector simply reflect flaws in economic policy that hinder economic growth? If the second, the share of state enterprises in employment could be expected to be positively correlated with in- flation, a common measure of policy failure. That is not the case, however. The correlation between state enterprises' share in employment (SOE/Labor) and a 7T measure of inflation distortion (defined as + where 7c is the rate of inflation) is -0.31 (t= 1.9).4 4. The correlation between SOE/Labor and the share of government expenditure in GDP in the sample is 0.48 (t = 3.2), but government expenditure is not in itself a sign of policy weakness or inefficiency, certainly not if the government spends its tax revenue mostly on productive infrastructure, education, and health care. Gylfason, Herbertsson, and Zoega 443 FIGURE 3. State Enterprises and Economic Growth (percent) KOR 5 TF-A N~~~~~D 03 CHL ,w irlyv LKA _ ~~~~~~~~~T UN 2-, a- 1EG COBD_ Q DZo4BRAEN _L ° -1 ~~BOL 3 2~~~4G TTO ZARTG (D SLE N -3 r35LEGNUZM -5 0 10 20 30 40 Share of State Enterprises in Employment 1978-91 Source: Penn World Tables and the World Bank. Note: Country abbreviations are defined in the Appendix. Next, the simple correlations between the size of the state sector and invest- ment, education, and growth are subjected to closer econometric scrutiny. Regression Analysis The model is estimated as a system using seemingly unrelated regression (SUR).5 That allows the marginal processes for investment, education, and growth to be modeled simultaneously to investigate the direct and indirect effects of state enterprises on economic growth. First, however, a basic Barrovian growth re- gression is estimated to explain the growth rate alone. The results for regression 1 in table 2 are for a cross-sectional OLS regression of average growth on the logarithm of the initial level of GDP and the average share of investment in GDP. The negative coefficient on initial income (although not statistically significant) is a sign of /3-convergence, but quite slow: it implies a convergence speed of 0.3 percent a year rather than the 2-3 percent rate usu- ally reported in the literature (Barro and Sala-i-Martin 1995). However, this result 5. The models were also estimated independently using OLS (not reported). The results remained virtually the same. TABLE 2. Empirical Results Economic growth, 1978-92 Investment, 1978-92 Education, 1978 (1) (2a) (3a) (4a) (5) (2b) (3b) (4b) (2c) (3c) (4c) Initial GDP -0.003 -0.008 -0.010 -0.008 -0.007 0.024 0.023 0.053 0.147 0.147 0.240 (1.19) (2.12) (1.78) (1.74) (1.59) (1.73) (1.69) (7.99) (4.47) (4.47) (13.16) Investment 0.172 0.131 0.213 0.208 0.162 - - - - - - (4.74) (3.64) (4.08) (4.16) (3.87) Secondary education - 0.029 0.014 0.914 0.023 - (2.08) (0.56) (0.87) (1.34) SOE/Labor - - - - - -0.267 -0.278 - -0.419 -0.416 - (2.58) (2.68) (1.71) (1.70) SOE/GDP - - - - - - - 0.078 - - 0.036 (1.73) (0.30)) SOE/Debt - - - - -0.096 (2.48) SOE/Labor x Investment - - -0.695 - (2.06) SOEIGDP x Investment - - - -0.219 - - - (2.30) Constant 0.003 0.038 0.056 0.032 0.034 -0.016 -0.011 -0.272 -0.730 -0.731 -1.463 (0.16) (1.46) (1.42) (1.11) (1.11) (0.14) (0.10) (5.11) (2.79) (2.79) (10.00) SE 0.019 0.018 0.018 0.018 0.019 0.056 0.056 0.052 0.125 0.125 0.138 Adj. R2 0.22 0.25 0.36 0.26 0.22 0.33 0.33 0.44 0.55 0.55 0.72 Number of countries 96 88 34 67 71 39 39 74 35 35 69 Estimation method OLS SUR SUR SUR OLS SUR SUR SUR SUR SUR SUR Note: t values appear within parentheses below the coefficients. Gylfason, Herbertsson, and Zoega 445 is in line with other studies when such variables as human capital, trade, and political instability are excluded. The higher the share of investment in GDP, the more rapid is economic growth in all the regressions; this effect is quite robust. According to point estimates, increasing the investment ratio from 20 to 30 percent from one country to an- other increases growth by 1.3 to 2.1 percent, all else remaining the same. These estimates are broadly similar to those reported by Levine and Renelt (1992), Sachs and Warner (1995), Gylfason and Herbertsson (1996), and Gylfason (1999).6 The exclusive focus here on developing economies, where diminishing returns to capital have not yet set in fully, may explain why investment in some cases appears to have a slightly stronger effect on growth than in some of the above- mentioned studies (Sachs and Warner, in particular), which include industrial as well as developing countries. In regression 2a, the education variable is the usual secondary school enroll- ment rate from Barro and Lee (1993), measured at the beginning of the sample period (1978), as is customary to avoid simultaneity bias. The effect of educa- tion on growth is statistically-and economically-significant: an increase in the initial secondary school enrollment rate from 50 to 80 percent increases the average rate of growth by almost a whole percentage point, all else remaining the same. Regression 2a is estimated as part of a system of three equations, in which equations 2b and 2c describe the dependence of investment and secondary edu- cation on initial income and the share of state enterprises in employment (re- call figures 1 and 2). An increase in state enterprise employment discourages both investment (regression 2b) and education (regression 2c)-education only marginally, however-and thus reduces growth, as shown in regression 2a. The total indirect effect of an increase in the employment share of state enterprises on economic growth is 0.131 x (-0.267) + 0.029 x (-0.419) = -0.035 - 0.012 = -0.047 (t = 2.5).7 The indirect effect of state enterprises on growth through investment is statistically significant (t = 2.1), but the indirect effect through education is not (t = 1.3).8 When initial rather than average values of the SOEI Labor variable are used to guard against the possibility of reverse causation and omitted-variable bias, a broadly similar pattern emerges. This reduces the likelihood that the results are driven by the effects of economic growth and investment on the size of the state enterprise sector (for example, growth slow- downs that make governments more willing to expand state enterprise employ- ment, or investment booms that make state enterprise employment and growth increase simultaneously). 6. Barro and Sala-i-Martin (1995) report smaller and less significant effects of investment on eco- nomic growth. 7. The composite t values are computed by Taylor expansion following Staiger and others (1997). 8. The total effect of initial income on growth is, by similar arithmetic, smaller than the direct effect, as is reasonable: conditional convergence does not necessarily generate absolute convergence. 446 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Regression 3a adds the multiple of the employment share of state enterprises and the investment ratio to test for the direct impact of state enterprises' employ- ment share on growth. This makes the effect of investment on growth depen- dent on the size of the state enterprise sector. The idea is that state enterprises tend to buy inferior capital, which adds less to output. The coefficient on the SOE/Labor term is significant and implies that a one-standard-deviation increase in state enterprises' employment share (0.14) reduces economic growth by -0.695 x 0.153 x 0.14 = -0.015, or 1.5 percentage points, evaluated at the sample mean of the investment ratio (0.153). The investment rate survives the introduction of the interaction term involving employment share, but the education variable drops in both size and significance. Auxiliary regressions 3a and 3b are similar to re- gressions 2b and 2c. The total effect of an increase in state enterprise employ- ment on economic growth is -0.695 x 0.153 + 0.213 x (-0.278) + 0.014 x (-0.416) = -0.106 - 0.059 - 0.006 = -0.171 (t = 3.4).9 Therefore, when the share of state enterprises in employment increases by one standard deviation, economic growth decreases by 2.4 percentage points, all else remaining the same, directly as well as through investment. The indirect growth effect of state enterprises through investment is economically and statistically significant (t = 2.2), but the indirect effect through education is not (t = 0.5). The visual impression conveyed by figures 1 and 3 is confirmed. Again, a broadly similar pattern is observed when initial rather than average values of the SOEI Labor variable are used: the total effect of an increase in state enterprise em- ployment on growth is now -0.123 (t = 2.5). So far, the size of the state enterprise sector has been measured by its share in total employment rather than by its share in GDP. This is because the inefficiency associated with state enterprises is often manifested in overstaffing. (Recall the case of Ghana, where the share of state enterprises in employment rose by half during 1978-91, while their share in GDP declined.) It is nevertheless interesting to explore whether there is a significant relationship between the share of state enterprises in GDP (SOE/GDP) and economic growth. Regression 4a shows that an increase in the share of state enterprises in GDP has a significantly negative direct effect on economic growth, a result that also holds when initial rather than average values of SOE/GDP are used. There are no indirect effects, however, at least not through education (see regression 4c). True, the coefficient on SOE/GDP in investment regression 4b is marginally significant, but its sign is wrong in view of the model. Even entertaining the pos- sibility that an increase in the share of state enterprises in GDP stimulates invest- ment does not materially change the result: the total effect of SOE/GDP on growth is still significantly negative. 9. When SOE/Labor appears as an independent variable on its own in regression 3a, without inter- acting with investment, its direct effect on growth is still negative, but not significant (t = 1.1). In other respects, the results remain virtually unchanged. Gylfason, Herbertsson, and Zoega 447 Regression 5 includes the external debt of state enterprises as a proportion of GDP (SOE/Debt). This result also holds when initial rather than average values of SOE/Debt are used. III. CONCLUSION In the simple model developed here of endogenous growth in an economy with state enterprises as well as private firms, a large state sector tends to be associ- ated with slow economic growth, all else remaining the same, if state enterprises are less efficient than private firms, invest less, employ less skilled labor, and are less eager to adopt new technology. The main empirical finding is that, across countries, investment and economic growth during 1978-92 were inversely re- lated to the size of the state enterprise sector, measured by its share of total employment. Specifically, a one-standard-deviation increase in the state sector's share of total employment from one country to another reduces the ratio of in- vestment to GDP by about four percentage points and reduces per capita growth by about one to two percentage points, all else remaining the same. Thus, too great a reliance on state enterprises may stand in the way of both static and dynamic efficiency-and consequently also investment and economic growth. Even so, the results need to be interpreted with caution in view of the limited data coverage across countries and over time. APPENDIX. COUNTRY ABBREVIATIONS USED IN FIGURES 1-3. Code Name Code Name DZA Algeria MDG Madagascar ARG Argentina MWI Malawi BEN Benin MLI Mali BOL Bolivia MUS Mauritius BWA Botswana MEX Mexico BRA Brazil NAM Namibia BDI Burundi PER Peru CMR Cameroon PHL Philippines CHL Chile SEN Senegal COL Colombia SLE Sierra Leone COG Congo LKA Sri Lanka CIV C6te d'lvoire TAW Taiwan EGY Egypt, Arab Rep. TZA Tanzania GMB Gambia, The THA Thailand GHA Ghana TGO Togo GRD Grenada TTO Trinidad and Tobago GIN Guinea TUN Tunisia IND India TUR Turkey IDN Indonesia ZAR Zaire KEN Kenya ZMB Zambia KOR Korea, Rep. 448 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 REFERENCES Barro, Robert J. 1990. "Government Spending in a Simple Model of Endogenous Growth." Journal of Political Economy 98(5):S103-25. Barro, Robert J., and Xavier Sala-i-Martin. 1995. Economic Growth. New York: McGraw-Hill. Barro, Robert J., and Jong-Wha Lee. 1993. "International Comparisons of Educational Attainment." Journal of Monetary Economics 32(3):363-94. Blanchard, Olivier J. 1985. "Debt, Deficits, and Finite Horizons." Journal of Political Economy 93(2):223-47. Glomm, Gerhard, and B. Ravikumar. 1992. "Public versus Private Investment in Human Capital: Endogenous Growth and Income Inequality." Journal of Political Economy 100(4):818-34. Gylfason, Thorvaldur. 1999. "Exports, Inflation, and Growth." World Development 27(6):1031-57. Gylfason, Thorvaldur, and Tryggvi Thor Herbertsson. 1996. "Does Inflation Matter for Growth?" CEPR Discussion Paper 1503. Centre for Economic Policy Research, Lon- don (forthcoming in Japan and the World Economy). Gyourko, Joseph E., and J. Tracy. 1988. "An Analysis of Public- and Private-Sector Wages Allowing for Endogenous Choices of Both Government and Union Status." Journal of Labor Economics 6(2):229-53. Harberger, Arnold C. 1998. "A Vision of the Growth Process," American Economic Review 88(l):1-32. Laffont, Jean-Jacques M., and Jean Tirole. 1993. A Theory of Incentives in Procurement and Regulation. Cambridge, Mass.: MIT Press. Levine, Ross, and David Renelt. 1992. "A Sensitivity Analysis of Cross-Country Growth Regressions." American Economic Review 82(4):942-63. Mauro, Paolo. 1998. "Corruption and the Composition of Government Expenditure." Journal of Public Economics 69(2):263-79. Milanovic, Branko. 1989. Liberalization and Entrepreneurship: Dynamics of Reform in Socialism and Capitalism. Armonk, N.Y.: M. E. Sharpe. Palia, Darius, and Edmund S. Phelps. 2000. "The Empirical Importance of Private Own- ership for Economic Growth." In L. Paganetto and E. Phelps, eds., Finance, Research, Education and Growth. London: Macmillan. Phelps, Edmund S. 1993. "The Argument for Private Ownership and Control." In Annual Economic Report. London: European Bank for Reconstruction and Development. Rama, Martin. 1999. "Public Sector Downsizing: An Introduction." World Bank Eco- nomic Review 13(1):1-22. Romer, Paul M. 1990. "Endogenous Technical Change." Journal of Political Economy 98(5):S78-102. Sacks, Jeffrey D., and Andrew M. Warner. 1995, revised 1997, 1999. "Natural Resource Abundance and Economic Growth." NBER Working Paper 5398. National Bureau for Economic Research, Cambridge, Mass. Staiger, Douglas, James H. Stock, and Mark W. Watson. 1997. "The NAIRU, Unemploy- ment, and Monetary Policy." Journal of Economic Perspectives 11(l):33-50. Stiglitz, Joseph E. 1988. Economics of the Public Sector, 2nd ed. New York: Norton. Gylfason, Herbertsson, and Zoega 449 1994. Whither Socialism? Cambridge, Mass.: MIT Press. Summers, Robert, and Alan W. Heston. 1991. "The Penn World Table (Mark 5): An Expanded Set of International Comparisons: 1950-1988." Quarterly Journal of Eco- nomics 106(2):327-68. Vickers, John, and George Yarrow. 1988. Privatization: An Economic Analysis. Cam- bridge, Mass.: MIT Press. World Bank. 1995. Bureaucrats Business: The Economics and Politics of Government Ownership. New York: Oxford University Press. 1997. World Development Indicators. Washington, D.C.: World Bank. 0 ~~~~~~~~~~~~~~~~~~~~~~~~ i I I i i THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 451-479 Infrastructure, Geographical Disadvantage, Transport Costs, and Trade Nuno Limao and Anthony J. Venables The authors use different data sets to investigate the dependence of transport costs on geography and infrastructure. Infrastructure is an important determinant of transport costs, especially for landlocked countries. Analysis of bilateral trade data confirms the importance of infrastructure and gives an estimate of the elasticity of trade flows with respect to the trade cost factor of around -3. A deterioration of infrastructure from the median to the 75th percentile raises transport costs by 12 percentage points and re- duces trade volumes by 28 percent. Analysis of African trade flows indicates that their relatively low level is largely due to poor infrastructure. The real costs of trade-the transport and other costs of doing business interna- tionally-are important determinants of a country's ability to participate fully in the world economy. Remoteness and poor transport and communications infrastructure isolate countries, inhibiting their participation in global produc- tion networks.' For example, in 1995 landlocked countries on average had an import share in gross domestic product (GDP) of 11 percent, compared with 28 percent for coastal economies. Eight of the top 15 nonprimary export per- formers for 1965-90 are island countries, and none is landlocked (World Bank 1998).2 As liberalization continues to reduce artificial trade barriers, the effec- tive rate of protection provided by transport costs is now, in many cases, con- siderably higher than that provided by tariffs.3 To bring countries further into Nuno Limao is with Department of Economics at Columbia University, and Anthony J. Venables is with the Department of Economics at the London School of Economics and the Center for Economic Policy Research. Anthony J. Venables's e-mail is a.j.venables@lse.ac.uk Nuno Limao's e-mail address is ngl4@columbia.edu. Most of the work on this project was undertaken while Anthony J. Venables was at the World Bank. The authors thank David Hummels, Steve Redding, and anonymous referees for helpful comments. 1. Increasing trade in components and the geographical fragmentation of some production processes make transport costs even more important. See Feenstra (1998) and the references quoted therein for evidence of the increase in the importance of intermediate goods trade. Radelet and Sachs (1998) show how sensitive value added is to transport costs in a vertically fragmented activity. 2. Export performance corresponds to growth in exports of nonprimary manufactured products in 1965-90 (Radelet and Sachs 1998, table 1). 3. See Finger and Yeats (1976) for U.S. post-Kennedy round data on nominal and effective rates of protection afforded by tariffs and transport costs. See Hummels (1998b) for recent data on nominal rates for Argentina, Brazil, New Zealand, and the United States. ©D 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 451 452 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 the trading system, it is important to understand both the determinants of trans- port costs and the magnitude of the barriers to trade that they create. Here we study the determinants of transport costs and show how they de- pend both on countries' geography and on their level of infrastructure. The importance of geography has been established by Hummels (1998b) as well as by Moneta (1959).4 We focus on the distance between countries, whether they share a common border, whether they are landlocked, and whether they are islands. The infrastructure measures relate to the quality of transport and com- munications infrastructure. Although the importance of infrastructure for trans- port costs is well established in regional and transport economics, the few em- pirical studies of international transport costs often neglect this and focus on geographical and product characteristics.5 We show that infrastructure is quan- titatively important in determining transport costs, a finding with important policy implications for investment in infrastructure. Poor infrastructure ac- counts for 40 percent of predicted transport costs for coastal countries and up to 60 percent for landlocked countries. An improvement in own and transit countries' infrastructure from the 25th percentile to the 75th percentile over- comes more than half of the disadvantage associated with being landlocked. Our research uses different sources of transport cost data. The first is ship- ping company quotes for the cost of transporting a standard container from Baltimore, Maryland, in the United States, to selected destinations. The advan- tages of this measure are that it is the true cost of transporting a homogeneous good and that it gives both the city of landfall and the final destination city. This enables us to compare the transport costs of land and sea legs of a journey, find- ing that the former is around seven times more costly per unit distance. The dis- advantage of this data set is that it is not clear how the experience of Baltimore generalizes, because charges are affected by particular routes, frequencies, and opportunities for backhauling and exploiting monopoly power. Our second data set uses a cross section of the ratio of carriage, insurance, and freight (CIF) to free on board (FOB) values that the International Monetary Fund (IMF) reports for bilateral trade between countries. These are representative insofar as they cover the entire imports of each reporting country. However, the measure is an aggregate over all commodity types imported, and there are some questions, which we address, regarding the quality of the data. In addition to the determinants of transport costs, we want to know the ex- tent to which transport costs choke off trade. To do this we undertake a gravity modeling exercise, incorporating the same geographical and infrastructure mea- sures that we use in estimating trade costs. This analysis strongly confirms the 4. Hummels (1998b) has undertaken a thorough study of the implications of geography for freight rates on disaggregated commodity imports of New Zealand, the United States, and five Latin American countries. 5. An exception to this is Radelet and Sachs (1998), where port quality is entered as an explanatory variable for transport costs. Limao and Venables 453 importance of these variables in determining trade and enables us to compute estimates of the elasticity of trade flows with respect to transport costs. We find that this elasticity is large, with a 10-percentage-point increase in transport costs typically reducing trade volumes by approximately 20 percent. Taken together, our approaches provide a rather consistent picture of the determinants of transport costs, in particular the importance of infrastructure in source and destination countries and in any transit countries used by land- locked economies. We draw out the implications of our findings by looking in some detail at trade and transport costs in Sub-Saharan Africa. Our measures indicate that many of these economies have extremely high transport costs. We show how taking infrastructure into account explains part of the relative trade performance of these countries. In section I we discuss the determinants of transport costs and present esti- mates for the transport cost equation using the shipping data and the CIF/FOB data. In section II we present the gravity results. In section III we compare and contrast the results from the transport cost and gravity analyses and derive an estimate of the elasticity of trade flows with respect to transport costs. We show that improvements in the infrastructure of landlocked countries and their transit countries can dramatically increase trade flows. We analyze trade and trans- port costs in Sub-Saharan Africa in section IV, finding that infrastructure ac- counts for much of Africa's poor performance. Section V concludes and sum- marizes our main quantitative findings. I. TRANSPORT COSTS The Determinants of Transport Costs Let T,, denote the unit cost of shipping a particular good from country i to coun- try j. We suppose that it is determined by (1) T, = T(x,j, Xi, Xi, p,,), where xi, is a vector of characteristics relating to the journey between i and j, Xi is a vector of characteristics of country i, Xi is a vector of characteristics of country j, and pi, represents all unobservable variables. What are the relevant observable characteristics of countries and the journeys between them? For the journey, we use two types of measures, both standard in the literature. The first is whether the countries share a common border, and the second is the shortest direct distance between the countries. The importance of distance for transport costs is obvious, but why should sharing a border reduce transport costs after controlling for distance? First, neighboring countries typi- cally have more integrated transport networks that reduce the number of trans- shipments, for example, from rail to road or across different types of rail gauge. Second, neighboring countries are more likely to have transit and customs agree- ments that reduce transit times and translate into lower shipping and insurance costs. Finally, the higher volume of trade between neighboring countries dra- 454 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 matically increases the possibilities for backhauling, allowing the fixed costs to be shared over two trips. For country characteristics, we focus on geographical and infrastructure mea- sures. The main geographical measures are simply whether the country is land- locked and whether it is an island. The infrastructure measure (inf) we use is designed to measure the costs of travel in and through a country. It is constructed as an average of the density of the road network, the paved road network, the rail network, and the number of telephone main lines per person. In our regres- sions, we always work with an inverse measure of this index; so an increase in the variable inf is expected to be associated with an increase in the costs of trans- port. Details on the construction of this and other variables are given in the appendix. Shipping from Baltimore Our first results are based on the costs of shipping a standard 40-foot con- tainer from Baltimore to different destinations around the world. A firm that handles forwarding for the World Bank provided the data, which cover 64 destination cities, 35 of which are in landlocked countries (see appendix tables A-2, A-4, and A-5). This source of data has two major advantages. One is that journeys can be broken down into component parts-the data gives the land- fall city for each journey as well as the final destination city-allowing sepa- rate estimation of the effect of land and sea distance. The other is that the good shipped is homogeneous, avoiding compositional problems that can occur in aggregate data.6 We estimate a linear version of equation 1 both for the entire journey and for the journey divided into the sea journey (to the port) and the land journey (from the port). More specifically, we estimate: (2) Ti a = a + 3'x1, + -y'X' + 63Xj + VI, where i corresponds to Baltimore and j represents the destination city. The error term vi, is assumed to be independent of the explanatory variables and normally distributed. The most appropriate functional form is not clear a priori. On one hand, we are adding over the different legs of the trip. That is, the cost of going through the infrastructure of the importer and the exporter and the cost of shipping be- tween them suggests a linear form. On the other hand, it is possible that there are interactions among the cost variables that would make a nonlinear form more suitable. The simplest example is that an increase in land distance should increase the cost of going through a given infrastructure. For this reason, we also experi- 6. UNCTAD (1995, p. 58) presents similar data for a sample of four coastal countries and nine land- locked countries in Sub-Saharan Africa. Livingstone (1986) uses quotes made by regular shippers to the Crown agents from the United Kingdom to eight African countries. The small size of the sample in both studies does not allow for a systematic examination of the determinants of transport costs. Limdo and Venables 455 mented with some nonlinear forms, but they were rejected by the data.7 There- fore, table 1 presents the ordinary least squares (OLS) estimation results of the linear form given by equation 2. The first two columns in table 1 give results excluding the infrastructure vari- ables. There are three main conclusions. First, being landlocked raises costs by $3,450-compared with a mean cost for nonlandlocked countries of $4,620. Second, breaking the journey into an overland component and a sea component (the second column in table 1) considerably improves the fit of the equation. It also gives a much larger coefficient for the overland portion of the trip compared with the sea distance. An extra 1,000 km by sea adds $190, whereas a similar increase in land distance adds $1,380. When this value is compared with the $380 per 1,000 km predicted by total straight-line distance (the first column), it be- comes clear that using the latter measure leads to a large underestimate of the impact of overland distance on transport costs. Third, the additional transport cost from being landlocked is not fully explained by the extra overland distance that must be overcome to reach the sea. Although the final city destination for landlocked countries is on average four times further from the sea than the final city destination for coastal countries in this sample, the landlocked dummy re- mains significant after controlling for land distance. There are several possible reasons for this, arising from border delays or transport coordination problems, uncertainty and delays creating higher insurance costs, and direct charges that may be made by the transit country.8 The third and fourth columns in table 1 introduce our measures of the in- verse infrastructure of the destination (inf) and, for landlocked countries, the transit country (inftran) for the smaller sample covered by these data.9 The signs of these are as expected, inferior infrastructures leading to higher transport costs. We can also ask what proportion of the predicted value is explained by infra- structure versus distance. For coastal economies, own infrastructure explains 40 percent of the predicted cost; for landlocked countries, own infrastructure ex- plains 36 percent and transit infrastructure 24 percent of the cost. The final specification (the fourth column) breaks distance into the overland and sea components. The coefficients on these distance variables are very simi- lar to those in the full sample (the second column). Splitting the distance vari- able makes the coefficient for transit infrastructure smaller and insignificant because of the variable's high positive correlation with land distance. Moreover, transit and own infrastructure are also highly correlated. This multicollinearity 7. This is true even when quadratic distance terms are added to capture any nonlinearity. These terms are insignificant, further justifying the use of the linear land and sea distance measures. We also estimated equation 2 including the per capita income of the destination country, because low-income countries might have high transport costs for a variety of reasons other than infrastructure. It was not significant. 8. For example, Kenya charges a transit goods license for road transit of $200 (per entry or 30 days) and tolls on trucks (UNCTAD 1997, p. 11). 9. The landlocked dummy is not included because of its multicollinearity with transit infrastructure. 456 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE 1. The Cost of Shipping a 40-Foot Container from Baltimore, 1990 Variable 1 2 3 4 Infrastructure (in/a) 1.31 1.56* (2.51) (2.92) Infrastructure of 1.34** 0.67 transit country (inftrana) (1.93) (0.88) Landlocked country 3.45*** 2.17*** dummy (Idldummyb) (4.75) (2.94) Distance between 0.38** 0.29* trading partners (2.60) (1.84) Sea distance (distsea) 0.19** 0.18* (2.12) (1.74) Land distance (distland) 1.38*** 1.49* (4.66) (1.77) Constant 1.10 2.06* 0.11 -0.10 (0.95) (1.85) (0.093) (-0.07) Sample size 64 64 47 47 R2 0.32 0.47 0.38 0.43 F-test (p-values) inf, inftran 0.00 inftran, distland 0.03 *Significant at the 10 percent level. **Significant at the 5 percent level. -**Significant at the 1 percent level. Note: The dependent variable is transport cost, Ti,, in thousands of U.S. dollars. The sample used in specifications 3 and 4 is reduced to the countries for which the infrastructure variables are also available. For specifications 1 and 3, the standard errors were adjusted to correct for heteroskedasticity. t-statistics are in parentheses. The F-tests are for the pairs of variables indi- cated; the p-values show the level at which the null of no joint significance is rejected. See table A-2 for the countries included in the sample. aValues for the infrastructure variables are averages for 1990-95 (the latest year available). bldldummy = 1 if the country has no access to the sea, 0 otherwise. Source: Authors' calculations. poses problems for identifying the separate effects of the two variables. How- ever, the tests of significance at the bottom of table 1 confirm the importance of the transit variable when considered jointly with either own infrastructure or land distance. To reemphasize the relative importance of infrastructure, an im- provement of inf from the 75th percentile to the median is equivalent to a dis- tance reduction of 3,466 sea km or 419 land km.10 10. For 20 landlocked countries in the sample, we have both the costs of shipping to the port and the full cost of shipping to the landlocked destination (for example, the cost of shipping from Baltimore to Durban and that from Baltimore to Harare via Durban). This enables us to look at the determinants of the incremental costs associated with the final stage of the journey. Final destination infrastructure is significant and positive, although incremental distance and port infrastructure are not. This is due both to the small number of observations and to details that become apparent on inspection of the data. For example, shipping from Baltimore to Durban costs $2,500: shipping the 1,600 km further to Lusaka costs an additional $2,500, whereas the 347 km from Durban to Maseru (Lesotho) costs an additional $7,500. This points to the importance of details of geography, market structure, and trade volumes, in addition to the broader picture painted by the econometrics. Limdo and Venables 457 CIF/FOB Measures Our second set of experiments is based on the CIF/FOB ratio as derived from the IMF's Direction of Trade Statistics (IMF various years). Importing countries re- port the value of imports from partner countries, inclusive of CIF, and exporting countries report their value FOB, which measures the cost of the imports and all charges incurred in placing the merchandise aboard a carrier in the exporting port. Denoting the FOB price of goods shipped from i to j by pij, we define tij, the ad valorem transport cost factor, as (3) tij cifij / fobij = (pij + T,j) / pij = t(xij, Xi, Xj, Tij where the determinants of Tij are given in equation 1. The ratio CIF/FOB provides the measure of transport costs on trade between each pair of countries. In theory, the FOB and CIF prices are border prices, and thus it would seem that own and trading partner infrastructures as defined here should not affect these rates. There are three reasons why they are indeed rel- evant. First, road, rail, and telephone infrastructures are likely highly correlated with port infrastructure (for which we have no data) and the latter would be important even if the prices were pure border prices. Second, the insurance com- ponent reflects the total time in transit, that is, from door to door, not just bor- der to border; total transit time is likely to be a function of own and partner infrastructure. Finally, according to UN experts on customs data, the FOB and CIF figures rarely measure actual border prices, instead measuring the prices at the initial point of departure and final destination, respectively.1" Thus, own and partner infrastructure should be included in the estimation. Assuming that t can be approximated by a log linear function up to some measurement error, we can write the observed transport cost factors tij (4) Intij = a + $xij + y'lnXi + 3'lnX, + wj where the tildes distinguish this set of parameters from those in equation 2. The final term, wc, contains unobserved variables, which we assume are uncorrelated with the explanatory variables, and random measurement error. As in the previ- ous section, functional form is to a large degree an empirical question. There are good reasons why ti, may be nonlinear in its determinants. For example, if coun- try j does not have a container port, then country i will not benefit from its own container facilities in exporting to j.l2 We found that the log linear form fitted the CIF/FOB data considerably better than the linear one. Several questions have been raised about the use of this CIF/FOB transport cost data.13 The first is that the measure aggregates over all commodities imported, 11. E-mail correspondence with Mr. Peter Lee at the United Nations. 12. Even if the true transport cost function T* is linear, there is no reason for the reduced form of the transport cost rate t* to have the same functional form. The reason for this is that for small export- ers (facing a perfectly elastic demand) the FOB price, p, , will itself depend on the average transport cost between themselves and their importers, an effect captured by the reduced form of t',,. 13. See Hummels (1998a). 458 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 so it is biased if trade on high transport cost routes systematically involves lower transport cost goods. This suggests that our estimates in fact will underestimate the true magnitude of transport costs. 14 The second is the presence of measure- ment error, arising particularly from the fact that exports are not always accu- rately reported. To the extent that this measurement error is uncorrelated with the explanatory variables, this should not be a problem. We deal with three other data problems as follows. First, approximately 25 percent of potential bilateral trade flows are dropped because of missing data from one of the partner countries. Second, some countries had CIF import val- ues lower than the corresponding FOB export values, which would imply nega- tive costs; we dropped all such observations. Third, we also dropped values when they were imputed by the IMF for a CIF/FOB ratio of 1.10. Table A-1 provides further details on sample selection. In section III, we compare the results obtained using the CIF/FOB data with those from the shipping cost data. The comparison indicates that the CIF/FOB data contain information about the cross-sectional variation in transport costs that is consistent with the shipping cost data. The model is estimated with 1990 data for a sample of 103 countries. Delet- ing observations that are missing, estimated, or give negative transport costs leaves 4,615 observations. Approximately 22 percent of all country pairs in our sample are reported to have no trade. One important reason for this is that at high enough transport costs two countries will not find it profitable to trade. This implies that for these countries, the transport cost measure is censored at some upper limit and this motivates our use of an upper limit Tobit. We assume that for those countries that report zero trade, the transport cost of trading takes the value of the upper limit in the sample. Estimation Results Table 2 gives the results from the estimation of equation 4. The first two rows of the table are characteristics of the journey between i and j; the log of distance, (Indistance), and whether i and j share a common border (border). The remain- der are characteristics of the importer country and its trading partner; a dummy for an island (isldummy and pisldummy); the per capita income of the import- ing and exporting countries, (InY/cap and InpY/cap). Finally, the infrastructure measures (Ininf and Inpinf) and the infrastructure of transit countries (ln(1 + inftran) and ln(1+ pinftran)). The first column of the table gives the effect of distance alone, and the sec- ond column gives a specification with journey and country characteristics, apart from infrastructure. Distance and border effects are as expected. Being or trad- ing with an island reduces transport costs (although these effects are barely significant), and high per capita income reduces transport costs. The infrastruc- ture variables are included in the third column, and all are significant with the 14. Hummels (1998b) discusses the cross-commodity variation in transport costs using disaggre- gated data for four countries. Limao and Venables 459 TABLE 2. The Bilateral Transport Cost Factor, 1990 Variable 1 2 3 4 Distance (Indistance) 0.25*'* 0.23*S 0.21*** 0.38*;` (6.74) (6.02) (5.65) (10.17) Common border (border) -1.35` -1.36*** -1.02*** (-7.77) (-7.78) (-6.30) Island (isldummy) -0.12'* : -0.09 -0.06 (-1.73) (-1.23) (-0.94) Island (pisldummy) -0.16** -0.12 * (-2.18) (-1.65) Per capita income (InY/cap) -0.31*** -0.23*** -0.24**' (-19.97) (-9.64) (-10.78) Per capita income (InpY/cap) -0.45*** -0.30*** (-27.94) (-12.84) Infrastructure (Ininf) 0.34*** 0.36 -* (3.92) (4.47) Partner infrastructure (Inpinf) 0.66*;'* (7.64) Infrastructure of transit country 0.21*' 0.36*'* ln(1 + inftran) (2.15) (4.07) Infrastructure of partner's transit 0.24*** country ln(1 + pinftran) (2.51) Partner fixed effects Pseudo-R2 0.10 0.46 0.48 0.60 ol 1.92 1.70 1.69 1.53 *Significant at the 10 percent level. -Significant at the 5 percent level. -*Significant at the 1 percent level. Note: The dependent variable is In transport cost factor CIF/FOB, Int,,. All variables are in natu- ral logs, except for the border variables and the island dummies. The sample size is 4,516; Tobit estimates. The pseudo-R2 is given by the correlation of actual and predicted Int,. Constants are included but not reported. Exporter fixed effects are included in column four but not reported. ca is the standard error of the Tobit estimate. t-statistics are in parentheses. The Tobit coefficients cor- respond to the marginal effects for the full sample, including the zeros. See table A-1 for data de- scriptions and sources and table A-2 for the countries included in the sample. The original transit variable, inftran, ranges from 0 for the coastal economies to approximately 1.7. Before taking the log, we add 1 to the measure to correctly reflect that coastal economies bear no extra infrastructure transport cost. To compare the own and transit elasticities, we need to multiply the coefficient of Ininftran (reported above) by inftran /(1 + inftran). This ratio ranges from 0.40 to 0.63 for land- locked countries in this sample. Source: Authors' calculations. expected sign. The final column gives results when partner country variables are replaced by dummies for each partner country. As expected, this increases the explanatory power of the equation. The own-infrastructure effects continue to be highly significant. The results contain several important messages. The first is the quantitative importance of the infrastructure effects. If a country could improve its in- frastructure from the median to the top 25th percentile, then its CIF/FOB fac- tor would fall from 1.28 to 1.11, this being equivalent to becoming 2,358 km 460 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 closer to all its trading partners.15 Conversely, deterioration in infrastructure from the median to the 75th percentile raises the predicted CIF/FOB factor from 1.28 to 1.40, equivalent to becoming 2,016 km further away from all trading partners. We can ask a similar question for the border effect. How much closer must two otherwise identical countries be if they do not share a border and are to have the same transport costs? The answer is that they would need to be 932 km closer-compared with a mean distance between capitals of bordering countries of 1,000 km.16 Thus the positive border effect on trade-which is typically found in gravity model estimates-is very important for transport cost reasons other than distance, suggesting that trans-shipment costs and the integration of trans- port networks are quite important. We turn to the cost of being landlocked in more detail in section III. Finally, it is worth comparing our estimates with those using distance, the simple and most commonly used proxy for transport costs. As shown by the pseudo-R2, using distance alone explains only 10 percent of the variation of trans- port costs, compared with almost 50 percent when the remaining geography and infrastructure measures are added. Clearly, distance fails to explain a significant part of the variation in transport costs. II. TRADE VOLUMES Instead of looking directly at trade costs, we now look at the trade flows they support, by estimating a gravity model including the infrastructure variables used above. There are two main reasons for doing this. First, the variables identified as being important in the transport cost equations should also be important in the trade equations, and we want to confirm that this is so. Sec- ond, by using the same variables in estimating transport costs and trade equa- tions, we are able to compute estimates of elasticities of trade flows with re- spect to transport costs. The gravity equation is the standard analytical framework for the predic- tion of bilateral trade flows. Its empirical use in the context of international trade dates back to the early 1960s and theoretical underpinnings were devel- oped later.'7 Despite the abundant number of theoretical derivations of the gravity equation, the majority of the authors do not model transport costs explicitly, exceptions being Bergstrand (1985) and Deardorff (1998). More recently, Bougheas and others (1999) incorporate transport infrastructure in 15. This uses estimates from the fourth column in table 2, and evaluated at the median CIF/FOB ratio of 1.28 and the median distance of 7,555 km, respectively, so 1.11 = 1.28 * (0.95/1.41) A (0.36) and 2,358 = 7,555-7555 ** (0.95/1.41) ^ (0.36/0.38). 16. Evaluated at the mean distance for bordering countries of 1,000 km, as new distance = 1,000 exp(-1.02/.38). 17. See Frankel (1997) for a discussion of earlier references. For different theoretical underpinnings, see Anderson (1979), Bergstrand (1985). Limao and Venables 461 a two-country Ricardian model and show the circumstances under which it affects trade volumes.18 Bilateral imports, Mi,p depend on GDP in countries i and j (Y, and Y,) in the standard way, and on the transport cost factor, tij, which we model in terms of the geographical and infrastructure measures used in the preceding analysis. Therefore, we have (5) Mi, = kyjfl Yi¢2t,jT E, or InM,j = Oo + ¢1 InYi + ¢2 InYi + Tr['lnxij + 'IlnXi + 6'InXi] + qij, where the second equation is obtained by taking logs and substituting out the true transport cost rate as given by equation 4. We estimate the second equation in expression 5 in the form: (5') InMij = Oo + 01 InYj + 02 InYi + ¢3 Indistancei, + q4borderij + 05isldummyj + q56isldummyi + 07 Ininf, + g8 lninf1 + t;t9 ln(1 + inftran,) + 410 ln(1 + inftrani) + 1 ln(Y / cap,) + 012 ln(Y / capi) + qil, where Mi, represents country j's imports from i valued cif, Y, is GDP, distance is distance between countries, border is whether they share a border, isldummy is a dummy for island countries, inf is the infrastructure measure, inftran is the infrastructure measure for the transit country, and Y/cap is per capita GDP.19 The model is estimated by Tobit using the same data set as for transport costs. In the sample used, 22 percent of all observations are reported as zeros, in which case the import values are set equal to the censoring point, which is the mini- mum value in the sample. Estimation Results Table 3 contains the results of the estimation. Income, distance, border, and is- land effects have the expected signs, as usual in gravity estimates. The striking result is the strong performance of the infrastructure variables used in the pre- ceding analysis. First, all infrastructure variables (importer, exporter, and tran- sit if either country is landlocked) have the correct sign and are significant at the 1-percent level. Moreover, they have sizable effects on trade volumes. Moving from the median to the top 25th percentile in the distribution of infrastructure raises trade volumes by 68 percent, equivalent to being 2,005 km closer to other countries.20 Moving from the median to the bottom 75th percentile reduces trade 18. Bougheas and others (1999) estimate augmented gravity equations for a sample limited to nine European countries. They include the product of partner's kilometers of motorway in one specification and that of public capital stock in another and find that these have a positive partial correlation with bilateral exports. 19. The transit infrastructure variables are adjusted for neighboring countries, so if i and i are neigh- bors and j (i) is landlocked, then inftran, (inftran,) is set to zero since no transit country must be used. So, to be more precise, in equation 5' we should write for j inftran,*(1 - border,) not inftran,, and simi- larly for i. 20. This uses estimates from the fourth column, and evaluated at the median distance of 7,555 km, so 1.68 = (0.95/1.41) A (-1.32) and 2,005 = 7,555-7,555 ` (0.95/1.41) A (1.32/1.69). 462 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 TABLE 3. The Gravity Model of Bilateral Imports, 1990 Variable 1 2 3 4 Income (InY) 1.28** 1.054** 0.99*- 1.03* * (53.51) (30.30) (28.04) (31.30) Income of trading partner (InpY) 1.55*'- 1.35` * 1.28^** (60.57) (37.48) (34.67) Distance (Indistance) -1.65 ' ** -1.43* -1.37*` -1.69` (-24.07) (-18.70) (-18.03) (-22.40) Common border (border) 2.45>** 2.52*** 1.85* (7.03) (7.25) (5.67) Island (isidummy) 0.48* ** 0.35* * 0.41 * (3.23) (2.46) (3.06) Island (pislduminy) 0.48 0.40*** (3.34) (2.78) Per capita income (InY/cap) 0.41 ** 0.16** 0.12** (8.78) (2.96) (2.28) Per capita income (InpY/cap) 0.34` * 0. 16* (7.29) (3.04) Infrastructure (Inint) -1.32:`- -1.32 ** (-7.49) (-8.07) Partner infrastructure (lnpinf) -1.11 * (-6.26) Infrastructure of transit country -0.60-*' -0.77"** ln(1 + inftran) (-3.04) (-4.18) Infrastructure of partner's transit -0.45** country ln(1 + pinftran) (-2.26) Partner fixed effects Pseudo-R2 0.79 0.80 0.80 0.83 7 3.47 3.39 3.34 3.08 *Significant at the 5 percent level. *-Significant at the 1 percent level. Note: The dependent variable is bilateral imports, InM,,. The sample size is 4,516; Tobit esti- mates. The pseudo-R2 is given by the correlation of actual and predicted InM,,. Constants are in- cluded but not reported. a is the standard error of the Tobit estimate. All variables and sample selec- tion are as in table 2. t-statistics are in parentheses. Source: Authors' calculations. volumes by 28 percent, equivalent to being 1,627 km further away from trading partners. III. COMPARISON AND QUANTIFICATION In this section, we compare the results in a way that facilitates the assessment of the quantitative importance of infrastructure and geographical location for trans- port costs and trade. The Cost of Being Landlocked Table 4 shows the disadvantage of being landlocked, relative to being an aver- age coastal country, for different values of own and transit country infrastruc- Limdo and Venables 463 TABLE 4. The Cost of Being Landlocked, Relative to a Coastal Economy, 1990 Own infrastructure percentile Transit infrastructure percentile 25th Median 75th Shipping data: transport cost ratio 25th 1.33 1.48 1.67 Median 1.41 1.55 1.74 75th 1.51 1.65 1.84 CIF/FOB data: (CIF/FOB - 1) ratio 25th 1.31 1.43 1.65 Median 1.34 1.46 1.69 75th 1.37 1.49 1.72 Gravity model: trade volume ratio 25th 0.55 0.42 0.26 Median 0.53 0.40 0.25 75th 0.50 0.38 0.24 Note: The construction of the variables for the shipping and CIF/FOB data is as follows: we calculate the predicted transport cost for landlocked countries allowing inf and inftran to vary as well as the landlocked dummy, but keep- ing all other variables at the level of the representative coastal country (me- dian value over nonislands). This is then divided by the predicted transport cost (or by CiF/FOB - 1) for the representative coastal country. For the trade volume data, a similar procedure is used. The percentiles are taken over the sample of landlocked countries. The specifications used are column 3 in table 1, column 3 in table 2, and column 3 in table 3. See table A-2 for the countries included in the sample. Source: Authors' calculations. ture. The shipping data indicate that the median landlocked country has trans- port costs 55 percent higher than the median coastal economy. However, im- proving own infrastructure to the level of the best 25th percentile among land- locked countries cuts this cost penalty to 41 percent, improvement by the transit country cuts the penalty to 48 percent, and if both improvements are made the penalty drops to 33 percent. Using the CIF/FOB measure, table 4 reports ratios of cIF/FoB-1 for landlocked countries relative to the median coastal economy. This gives slightly smaller cost penalties, with the median landlocked economy's transport costs 46 percent higher than the median coastal economy's. Improv- ing own and transit country infrastructure to the 25th percentile reduces this penalty to 34 percent and 43 percent, respectively; if both are improved the pen- alty drops to 31 percent. Comparison of these results assures us that the estimates from our different data sources are consistent, and that the cross-sectional variation in the CIF/FOB measure does contain useful information regarding transport costs. Although the CIF/FOB data predict relative costs that are 9 percentage points lower than the shipping data at the median infrastructure values, the partial effects of the own and transit infrastructure variables are similar across the data sets, as illustrated 464 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 in figure A-1 in the appendix. The similarity between the predicted effects on relative transport costs is particularly striking in the case of own infrastructure. Table 4 undertakes an analogous experiment for trade volumes, asking how the volume of trade of representative landlocked economies compares with the average coastal economy at the same income levels and distance. The difference is dramatic, with the median landlocked economy having only 40 percent of the trade volume. Improvements in own infrastructure from the median to the 25th percentile increase the volume of trade by 13 percentage points, improvement in transit country infrastructure increase the volume by 2 percentage points, and a simultaneous improvement leads to an increase of 15 percentage points in the volume of trade. The Elasticity of Trade with Respect to Transport Costs It is natural to link our estimates of trade volumes and transport costs by com- puting the elasticity of trade volumes with respect to the transport cost factor as given by the parameter T in equation 5. In this subsection we offer two approaches to doing this, one based on comparison of the estimates of the CIF/FOB and gravity models, and the other based on regression of trade volumes on predicted trade costs. The estimates from the CIF/FOB and gravity models (equations 4 and 5) pro- vide overidentifying restrictions for T, one for each of the determinants in the transport cost equations. We focus on the estimates of distance, border, and own and transit country infrastructure.21 The elasticities previously found in the gravity estimation ($) and the CIF/FOB estimation (9) are reproduced in the first two col- umns of table 5. The last column gives the predicted elasticity of trade with re- spect to the transport cost factor, T, obtained as the ratio of the gravity and CIF/ FOB elasticities. The point estimates of T vary quite widely, from -6.47 on the distance vari- able, to -1.67 for the price of infrastructure. The likely reason for this is that some of the variables influence trade volumes through channels other than mea- sured transport costs. For example, distance and border effects might be expected to influence trade volumes through such channels as information flows and lan- guage and cultural ties, which would not show up in measured transport costs.22 Our second approach is to use predicted values of transport costs (from equa- tion 4) as independent variables in the gravity model (equation 5). In estimating this, we exclude variables that, a priori, we think only affect trade volumes 21. Of the other two variables, it is likely that income per capita may enter the gravity equation for reasons other than transport costs, and the island dummy is not significant. 22. Geracci and Prewo (1977) estimate t for a sample of 18 Organisation for Economic Co-opera- tion and Development (OECD) countries. They find a higher elasticity (T = -10) than the one we find. This is possibly because of the restriction of their sample to high-income countries. More important perhaps is the fact that they do not estimate an upper limit Tobit for the transport cost. This is likely to lead to an underestimate of the predicted transport cost factor and a consequent upward bias of the transport cost elasticity. Limnio and Venables 465 TABLE 5. Estimates of Import Elasticity with Respect to the Transport Cost Factor, 1990 Elasticity Gravitya CIF/FOB b Trade Variable E = / 6 I)istance (Indistance) -1.37 0.21 -6.47 Import country infrastructure (Ininf) -1.32 0.34 -3.86 Transit country infrastructure ln(1 + inftran) -0.60 0.21 -2.87 Common border (border) 2.52 -1.36 -1.85 Partner infrastructure (lnpinf -1.11 0.66 -1.67 Infrastructure of partner's transit country ln(1 + pinftran) -0.45 0.24 -1.84 Note: We also calculate upper and lower bounds for the trade elasticities using the 95 percent confidence intervals for the gravity and CIF/FOB coefficients. These are distance (-4.28, -10.98); Ininf (-1.90, -9.75); ln(1 + inftran) (-0.53, -53.65); border (-1.08, -3.15); Inpinf (-0.91, -2.94), and ln(1 + pinftran) (-0.14, -15.64). 'Gravity elasticities correspond to the estimates in column 3 in table 3. bCIF/FOB elasticities correspond to the estimates in column 3 in table 2. Source: Authors' calculations. through transport costs (the infrastructure measures), leaving in those that might affect trade volumes directly. Thus, table 6 reports regressions of trade volumes on predicted values of the transport cost factor, incomes, per capita incomes, and distance and border effects. The first column uses predictions of the transport cost factors from the third column in table 2, whereas the second column has partner fixed effects, so it uses predictions from the fourth column in table 2. The coefficient on the predicted transport cost factor, tij, measures the elastic- ity of trade volume with respect to the transport cost factor, T, and, in column 1 in table 6, this is -2.24.23 Distance remains highly significant, although the coef- ficient falls markedly compared with the gravity estimates in table 3. This sug- gests that distance affects trade volumes both through transport costs and inde- pendently through other channels, such as information, which could account for the large value of T associated with the distance coefficients in table 5. Of the other variables, the border coefficient is insignificant, while incomes per capita enter with a negative sign, suggesting that, controlling for transport costs, coun- tries with low per capita income trade more than countries with high per capita income. The second column reports analogous results when partner-country fixed effects are included. The main difference is that this increases the absolute value of the estimated elasticity T to -3.11, while reducing further the independent role of distance. Taking tables 5 and 6 together enables us to make an informed judgment about the quantitative importance of transport costs in determining trade flows. Re- 23. Because this is the transport cost factor, an increase from, say, 1.1 to 1.2 is a 9 percent increase, not a doubling. 466 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE 6. Trade Volumes and Predicted Import Costs, 1990 Based on Based on Variable full modela fixed-effects model" Transport cost factor ln(^,,) -2.24 -3.11 (-10.80) (-10.01) Import country income InY 1.01 1.03 (29.42) (31.28) Export country income InpY 1.26 (34.76) Import country per capita income (InY/cap) -0.25 -0.59 (-3.23) (-5.58) Export country per capita income (InpYlcap) -0.57 (-5.93) Distance (Indistance) -0.87 -0.51 (-9.99) (-3.74) Common border (border) -0.50 -1.39 (-1.14) (-3.02) Partner fixed effects Pseudo-R2 0.80 0.83 a 3.35 3.08 Note: The dependent variable is bilateral imports, InM,,. The standard error of ln(ki) is not adjusted for the fact that it is a predicted variable, and therefore underestimates the true estimate error. aThe dependent variable is from column 3 in table 2. bThe dependent variable is from column 4 in table 2. Source: Authors' calculations. sults suggest an elasticity of trade flows with respect to the transport cost factor in the range of -2 to -3.5. Taking a value of -3 means that doubling transport costs from their median value (that is, raising the transport cost factor from 1.28 to 1.56) reduces trade volumes by 45 percent. Moving from the median value of transport costs to the 75th percentile (transport cost factor 1.83) cuts trade vol- umes by two-thirds. IV. TRANSPORT COSTS, INFRASTRUCTURE, AND SUB-SAHARAN AFRICAN TRADE Our results show how poor infrastructure and being landlocked damage trade. We now extend the quantitative implications of our findings by applying them to Sub-Saharan African (SSA) trade.24 24. Evidence of the importance of transport costs for Africa's export performance is given by Amjadi and Yeats (1995) and Amjadi, Reincke, and Yeats (1996). In the former study, it is reported that, ac- cording to balance of payments statistics, SSA's net insurance and freight payments amounted to 15 percent of the value of the exports. By comparison, for all developing countries the payments averaged 5.8 percent. Collier and Gunning (1999, p. 71) provide a brief description of the quantity and quality of infrastructure in SSA. Limao and Venables 467 Is SSA Trade Too Low? There is a common belief that Africa trades "too little" both with itself and with the rest of the world. Frankel (1997) reports intraregional trade shares in 1990 of 4 percent for Africa compared with 44 percent for East Asia. Amjadi, Reincke, and Yeats (1996) discuss the marginalization of SSA in world trade. The poor performance is typically attributed to protectionist trade policies (Collier 1995; Collier and Gunning 1999) and high transport costs due to poor infrastructure and inappropriate transport policies (Amjadi and Yeats 1995). This view has been contested by Foroutan and Pritchett (1993), who show that the low level of intra-African trade is explained by the usual determinants of a gravity equation. Similarly, Coe and Hoffmaister (1998) conclude that bi- lateral trade between SSA countries and industrial countries in the 1990s was not unusually low. Finally, Rodrik (1998) finds that the trade/GDP ratios of SSA countries are comparable to those of countries of similar size and income, and that Africa's marginalization is mainly due to low income growth. What evidence does our data provide on this, and to what extent can it be accounted for by the infrastructure variables we have identified as being so im- portant? To answer this we reestimated the baseline and infrastructure specifi- cations of our transport cost and gravity models, augmenting them with African dummies: African importer (Africa), African exporter (pAfrica), African importer and exporter (AA), and an interaction of the latter with distance (AAdistance). Tables 7 and 8 provide the estimates for the transport cost equation and the gravity equation, respectively. Intra-sSA trade costs are substantially higher and trade volumes substantially lower than those for non-SSA countries. In tables 7 and 8, the Africa factor gives the combined effects of the Africa dummies. Intra-sSA transport costs are 136 percent higher (2.36 = exp(0.08 + 0.52 + 0.26) from table 7) and trade volumes are 6 percent lower (0.94 = exp(-0.23 - 0.59 + 0.76) from table 8). Thus the basic specification cannot account for the poor performance of African trade, even when it controls for both geographical variables (border and island dummies) and per capita income. In tables 7 and 8, the third and fourth columns add the infrastructure mea- sures. The key finding is that infrastructure accounts for nearly half the trans- port cost penalty borne by intra-sSA trade. The penalty attributable to the Africa dummies drops from 136 to 77 percent. The Africa penalty on trade flows is actually overturned, suggesting that, once we control for infrastructure intra- SSA trade is 105 percent higher than would be expected. It is sometimes claimed that poor communications infrastructure in Africa entails higher transport costs per kilometer within SSA than elsewhere. We in- vestigate this with the interaction variable AAdistance, which is zero for trade involving one non-African country, and equal to distance for trade between a pair of African countries. Foroutan and Pritchett (1993) use a similar variable and find that it is insignificant, which leads them to conclude that "the gravity model gives little evidence that in fact distance is a greater barrier to intra-SSA TABLE 7. Transport Costs of Sub-Saharan African Countries, 1990 Variable 1 2 3 4 Income Import country (InY) Export country (InpY) Distance (Indistance) 0.29... 0.23'* " 0.26*** 0.20*# (7.38) (5.67) (6.57) (4.88) Common border (border) -1.33- # -0.97' * - -1.35# #* -1.01 ' (-7.66) (-5.39) (-7.72) (-5.59) Island dummy Import country (isldummy) -0.13* -0.12* -0.10 -0.09 (-1.78) (-1.68) (-1.36) (-1.29) Export country (pisldummy) -0.12* -0.11 -0.11 -0.10 (-1.64) (-1.55) (-1.47) (-1.41) Per capita income Import country (InY/cap) -0.29*** -0.29 t* -0.23"'# -0.23k" (-15.31) (-15.36) (-9.36) (-9.36) Export country (InpY/cap) -0.36 * * * -0.36 - * * -0.28 * * * -0.28 ** (-18.98) (-19.12) (-11.56) (-11.66) Infrastructure Import country (Inint) 0.32#*# 0.32*# (3.47) (3.59) Export country (Inpinfl) 0.50 ** 0.51 * (5.54) (5.60) Infrastructure of transit countries Import country ln(1 + inftran) 0.21# 0.18* (2.13) (1.81) Export country ln(1 + pinftran) 0.14 0.11 (1.43) (1.09) Africa dummies African importer Africa 0.08 0.09 -0.02 0.00 (0.36) (1.15) (-0.26) (0.00) African exporter pAfrica 0.52*** 0.53*** 0.37*** 0.39'>* (6.52) (6.72) (4.37) (4.62) African importer and exporter AA 0.26* -6.05 * * 0.22 -6.00* * (1.79) (-6.57) (1.52) (-6.54) Interaction of AA and distance (In(1,000 km)) 0.81 ** 0.80* * AAdistance (6.93) (6.85) Pseudo-R2 0.47 0.48 0.48 0.49 Uf 1.69 1.68 1.68 1.68 Africa factor, 2.36 1.77 Africa (1,000 km) 1.18 0.92 Africa (3,000 km) 2.87 2.21 Critical distanceb 826 1,110 *Significant at the 10 percent level. - Significant at the 5 percent level. ###Significant at the 1 percent level. Note: The dependent variable is In transport cost factor CIF/FOB, Int,,. The sample size is 4,516. t- statistics are in parentheses. The pseudo-R2 is given by the correlation of actual and predicted imports. Constants are included but not reported. cy is the standard error of the Tobit estimate. All variables and sample selection are as in table 2. aAfrica factor = exp(Africa + pAfrica + AA), or exp(Africa + pAfrica + AA + AAdistance " ln(#km)). bCritical distance, x, is given by: I - exp(Africa + pAfrica + AA + AAdistance In(x)) = 0. Source: Authors' calculations. 468 TABLE 8. The Gravity Models for Sub-Saharan African Countries, 1990 Variable 1 2 3 4 Income Import country (InY) 1.05... 1.05... 1.02-' 1.02... (27.44) (27.45) (26.96) (26.99) Export country (InpY) 1.31-* 1.31*** 1.28-* 1.28... (33.47) (33.45) (32.69) (32.70) Distance (Indistance) 1.39*** -1.31*** -1.29*** -1.21*** (17.45) (16.06) (-16.29) (-14.93) Common border (border) 2.34*** 1.87*** 2.42*** 1.98*** (6.70) (5.14) (6.96) (5.49) Island dummy Import country (isidummy) 0.45*** 0.44*** 0.35** 0.34** (3.14) (3.07) (2.41) (2.37) Export country (pisidummy) 0.42** 0.41*** 0.37*** 0.37** (2.89) (2.83) (2.57) (2.53) Per capita income Import country (InY/cap) 0.41** 0.41** 0.16*"* 0.16*** (8.62) (8.64) (2.92) (2.90) Export country (InpY/cap) 0.32** 0.32* * 0.17* * * 0.17* * * (6.85) (6.93) (3.11) (3.16) Infrastructure Import country (Ininf) -1.44*** -1.45*** (-7.92) (-7.99) Export country (Inpinf) -1.10-* -1.10*** (-6.03) (-6.06) Infrastructure of transit countries Import country ln(1 + inftran) -0.62*** -0.58** (-3.13) (-2.91) Export country ln(1 + pinftran) -0.40** -0.36* (-2.02) (-1.80) Africa dummies African importer Africa -0.23 -0.25 0.15 0.13 (-1.29) (-1.43) (0.86) (0.71) African exporter pAfrica -0.59* -0.62* * -0.31* * -0.34* (-3.46) (-3.58) (-1.78) (-1.93) African importer and exporter AA 0.76** 9.18*** 0.88*"* 9.00*** (2.61) (4.92) (3.03) (-4.89) Interaction of AA and distance (ln(1,000 kin)) -1.08*** -1.04** AAdistance (-4.56) (-4.46) Pseudo-R2 0.79 0.79 0.80 0.80 e 3.38 3.38 3.33 3.33 Africa factora 0.94 2.05 Africa (1,000 km) 2.34 4.98 Africa (3,000 km) 0.71 1.59 Critical distanceb 2,196 4,684 *Significant at the 10 percent level. -Significant at the 5 percent level. ***Significant at the 1 percent level. Note: The dependent variable is bilateral imports, InM,,. The sample size is 4,516. t-statistics are in parentheses. The pseudo-R2 is given by the correlation of actual and predicted imports. Constants are included but not reported. ce is the standard error of the Tobit estimate. All variables and sample selec- tion are as in table 2. aAfrica factor = exp(Africa + pAfrica + AA), or exp(Africa + pAfrica + AA + AAdistance * ln(#km)). bCritical distance, x, is given by: I - exp(Africa + pAfrica + AA + AAdistance * In(x)) = 0. Source: Authors' calculations. 469 470 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 trade than it is for other countries. This result goes against the apparently com- mon feeling that the poor quantity and quality of communications and trans- port infrastructures between SSA countries is a major obstacle to intra-SSA trade." We find the opposite, with the second and fourth columns in table 7 indicat- ing that the variable is significant in raising transport costs, and the second and fourth columns in table 8 indicating that it is significant in reducing trade vol- umes.25 Thus, controlling for infrastructure, African transport costs are 8 per- cent lower on journeys of 1,000 km, but 121 percent higher on journeys of 3,000 km. One way to summarize the results, including the interaction variable, is to calculate the critical distance above which a pair of African countries faces a penalty compared with a pair of non-African countries. Looking at transport costs, the distance is 826 km, rising to 1,110 km once we control for infrastruc- ture. Looking at trade volumes, the distance is 2,196 km, rising to 4,684 km once infrastructure is included. It is interesting to note that including the infra- structure measures more than doubles the critical distance for trade and that the majority of country pairs in SSA on opposite coasts exceed that critical distance. Pulling our Africa results together, there are several main conclusions. First, intra-African transport costs are higher and trade volumes lower than would be predicted by a simple model (column one in tables 7 and 8). However, much of this can be attributed to poor infrastructure and to the particularly high cost of distance in Africa. Our results confirm the fact that intra-African trade is con- centrated at the subregional level with less east-west trade than would be ex- pected between a pair of otherwise similar countries in the rest of the world. V. CONCLUSION Transport costs and trade volumes depend on many complex details of geogra- phy, infrastructure, administrative barriers, and the structure of the shipping industry. In this article, we have used several sources of evidence to explain trans- port costs and trade flows in terms of geography and the infrastructure of the trading countries, and of countries through which their trade passes. Table 9 summarizes some of the main results on the impact of infrastructure, reporting levels and changes from the median infrastructure. The results are strongly consistent, although they come from different data sets and measure different things. Thus, deterioration in infrastructure from that of the median country to the 75th percentile raises costs, according to our shipping data, by an amount equivalent to 3,466 km of sea travel or 419 km of overland travel. Us- ing the CIF/FOB ratio, the equivalent distance is 2,016 km. The impact on trade volumes is equivalent to an extra 1,627 km distance. Linking transport costs to trade volumes, we estimate an elasticity of trade flows with respect to the transport cost factor of around -3. Table 10 summa- 25. The finding in Foroutan and Pritchett (1993) is most likely because the dummy for African coun- tries that export and import and the interaction variable are multicollinear and thus they are not able to identify either. In our sample the correlation between these variables is over 0.9. Limdo and Venables 471 TABLE 9. Predicted Effects of Infrastructure on Trade Costs and Trade Volumes, 1990 Infrastructure percentile Variable 25th Median 75th Shipping data Transport costs, US$ 4,638 5,980 6,604 Sea km, equivalent change -3,989 0 +3,466 Land km, equivalent change -481 0 +419 CIF/FOB CIF/FOB ratio 1.11 1.28 1.40 Kilometers, equivalent change -2,358 0 +2,016 Gravity Trade volume, percentage change +68 0 -28 Kilometers, equivalent change -2,005 0 +1,627 Note: Shipping data are from column 4 in table 1, CIF/FOB data are from column 4 in table 2, and gravity data are from column 4 in table 3. Source: Authors' calculations. rizes the implications of this. It indicates, for example, how a doubling of trans- port costs (from the median value) reduces trade volumes by 45 percent. The article also presents results on the disadvantages faced by landlocked countries and by African countries. From both the shipping and the cIF/FOB data sets, we see that landlocked countries are disadvantaged. The representative land- locked economy has transport costs 50 percent higher and trade volumes 60 percent lower than the representative coastal economy. However, landlocked countries are able to overcome a substantial proportion of this disadvantage through improvements in their own and their transit countries' infrastructure. Looking at SSA, we see that transport costs are relatively high, and that trade flows are lower than would be predicted by standard gravity modeling both for intra-SSA trade and for African countries' external trade. We find that most of this poor performance is explained by poor infrastructure and by a particular penalty on long-distance (typically cross-continental) trade in Africa. TABLE 10. Predicted Effects of the Transport Cost Factor on Trade Volumes, 1990 Transport cost factor, T, Predicted change in trade volume selected values from median (percent) 1.11 (25th percentile) +53 1.14 +42 1.28 (Median) 0 1.56 -45 1.83 (75th percentile) -66 Note: T = -3. Source: Authors' calculations. 472 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 Shipping costs of the magnitudes reported here have a major impact on in- come, both because of the direct cost they impose, and because of the gains from trade forgone. However, our results also point to the potential for reducing these costs through investment in infrastructure. APPENDIX. CONSTRUCTION OF VARIABLES Own Infrastructure Each country's infrastructure is measured by an index constructed from four variables: kilometers of road, kilometers of paved road, kilometers of rail (each per square kilometer of country area), and telephone main lines per person. These measures are highly correlated among themselves and identifying each of their influences on transport costs separately is not possible. One possibility would have been to build an index using principal components. However, we have data on all of the measures for only 51 countries. Thus, we first normalize the vari- ables to have the same mean, one, and then take the linear average over the four variables, ignoring missing observations. This is equivalent to assuming that roads, paved roads, railways, and telephone lines are perfect substitutes as in- puts to a transport services production function. Taking the mean over the available observations implicitly assumes that the missing variables take on average the same value as the available variables. This measure was raised to the power -0.3. The reason for this is that infrastructure is an input to a transport services production function that, if Cobb Douglas, might be written as Y = KILOIx where I, the index of infrastructure, is exogenous to the transport sector firm. Then for a given output the reduced form of the cost function will be T = 0Ix'(a+0) where 4 is a function of the factor prices of private inputs, the technology, and the target output. If there are constant re- turns to scale to the private inputs, K and L, then our assumption is that x = 0.3. According to the data, this value implies that the transport cost per kilometer of the worst infrastructure is approximately ten times that of the best one. In the log-linear specifications this scaling is only a choice of units. Transit Infrastructure Let L denote a given landlocked country and Lt the set of transit countries L uses to reach the sea (table A-3). Ideally, we weight transit countries' infrastructure by their share of the transit trade. However, available data report solely whether a country is used for transit, so if country L uses n transit countries, the variable inftran gives an equal weight of 1 /n to the infrastructure index of each of those countries. Two caveats should be noted. First, we are assuming that no trade (or the same share of trade for all countries) goes by air. Although this is clearly unrealistic and the share of trade that is airborne is rising, it is still small enough for land- locked countries to justify this assumption. Second, the transport cost from land- locked to neighboring countries should not include transit country costs and thus, when necessary, our variable is adjusted to reflect this fact. Lim,o and Venables 473 FIGURE A-1. The Transport Cost of Landlocked Countries Relative to an Average Coastal Country 2.5 A 2 A . 1.5 A*A S o 0 g 1 A' 0.5 0 0 1 2 3 4 5 6 Own infrastructure Index 17 A 1.6 1S A A . A "' AA o 1.4 cu 1.3 A 1.2 A 1.1 0 0.5 1 1.5 2 2.5 Transit infrastructure Index A Shipping A cif/fob data data TABLE A-1. Variable Descriptions and Sources of Data Variable Description Source Use distance Great circle distance between trading partners (1000s km unless In is used). Fitzpatrick (1986), authors' calculations All distsea Sea distance around continents from Baltimore to the sea port of landfall (lOOOs km). DMA (1985), authors' calculations Shipping distland Great circle distance from sea port of landfall to capital of destination (lOOs km). Authors' calculations Shipping border Dummy variable = 1 if two countries are contiguous or are separated by CIA (1998) CIF/FOB, gravity less than 40 km, 0 otherwise. inf Inverse of the index of road, paved road and railway densities and telephone lines per capita. A higher value indicates worse infrastructure (see below for Canning 1998, authors' calculations All more details). inftran Average value of infrastructure for the transit countries if a country is landlocked, Canning 1998, UNCTAD, All zero otherwise. Table A-3 below lists the landlocked countries with respective authors' calculations transit countries used. 4 ldldummy Dummy variable = 1 if the country is landlocked, 0 otherwise. CIA (1998) All 4 isldummy Dummy variable = 1 if the country is an island, 0 otherwise. CIA (1998) Gravity, CIF/FOB T,, Cost of shipping a 40' container from i = Baltimore to country j (lOOOs US$, 1999). Panalpina (private communication) Shipping The mode is surface (as opposed to air), type is freight (as opposed to household goods) and packing is loose (as opposed to lift van where the cargo is packed into wooden containers). The cost does not include insurance. M,l Aggregate imports (inclusive of insurance and freight, CIF) of country j from IMF (various years) Gravity country i 1000s current (1990) US$. lXI' Aggregate exports (free on board value ) of country i to country j 1000s current IMF (various years) Gravity (1990) US$. It, ml]M,IX,. IMF (various years) CIF/FOB Y GDP in current (1990) US$ market prices. World Bank (1998) Gravity Y/cap Y/population. World Bank (1998) All Note: In the text In variable stands for the natural logarithm of variable, pvariable stands for the trade partner's variable. There are 103 countries in the sample used in sections I and 11. This implies 10,712 potential bilateral pairs. The sample is greatly reduced because 2,759 of the pairs had missing import or export values; 555 had positive imports of j from i, but exports of zero from i to j; 2,494 had nonpositive transport costs; and 195 had CIF/FOB between 1.0909 and 1.101. Limdo and Venables 475 TABLE A-2. List of Countries in the Samples (sorted by quality of own infrastructure) Shipping data sample Belgium Swaziland Bolivia Georgia* Netherlands China Peru Russia* Switzerland Malawi Lesotho (75th) Luxembourg* Austria Argentina Burkina Faso Czech Republic* Italy Senegal Zambia Azerbaijan* Germany Uganda Benin Armenia* Hungary Kenya (50th) Nepal Belarus* Rwanda Botswana Burundi Kazakhstan:, Uruguay Cameroon Bhutan Kyrgyz Republic* Turkey Togo Mozambique Macedonia* India C6te d'Ivoire Mali Moldova* South Africa (25th) Ghana Central African Rep. Slovakia* Thailand Nigeria Niger Tajikistan'" Zimbabwe Congo, Rep. of Chad Turkmenistan* Brazil Paraguay Ethiopia'" Uzbekistan* Chile Tanzania Eritrea' IMF data sample Belgium Mauritius Oman C6te d'Ivoire Singapore Spain Guatemala Peru Netherlands Sweden Colombia Bolivia Switzerland Costa Rica Zimbabwe Benin Japan El Salvador Venezuela Gabon Hong Kong, China India Gambia, The Sierra Leone Denmark Turkey Iran, Islamic Rep. of Haiti Austria New Zealand Honduras Guinea United Kingdom Sri Lanka Togo Guinea-Bissau Germany Jamaica Ecuador Zambia Italy South Africa Malawi Angola France Norway Saudi Arabia Congo, Dem. Rep. of Ireland Mexico Egypt, Arab Rep. of Nicaragua Hungary Jordan Indonesia Papua New Guinea Poland Argentina Australia Congo, Rep. of Israel Tunisia Uganda Burkina Faso Trinidad and Tobago Malaysia China Nepal Portugal Syrian Arab Republic Kenya Lao PDR Finland Thailand Paraguay Madagascar Romania Panama Dominican Republic Mozambique Korea, Rep. of Bangladesh Senegal Mauritania Rwanda Chile Ghana Central African Rep. Greece Philippines Burundi Mali United Arab Emirates Brazil Algeria Niger Uruguay Canada Cameroon Chad United States (25th) Pakistan (50th) Nigeria (75th) *Excluded from columns 3 and 4 in table I due to missing data for own or transit infrastructure. Note: Not all country pairs were used due to missing data. Countries with infrastructure values closest to the corresponding sample values are labeled 25th, 50th, and 75th. Source: For shipping data, see text; for IMF data, IMF (various years). 476 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 TABLE A-3. List of Transit Countries for Landlocked Countries in the Shipping Data Sample and the IMF Data Sample (sorted by quality of transit country infrastructure) Landlocked countries Transit countries Shipping data samplea Austria Germany Hungary Germany Switzerland Germany Bhutan India Nepal India Botswana (25th) South Africa Lesotho (25th) South Africa Swaziland (25th) South Africa Zimbabwe (25th) South Africa Zambia South Africa, Zimbabwe Paraguay (50th) Brazil Bolivia Chile Malawi South Africa, Zimbabwe Burundi Kenya Uganda Kenya Rwanda (75th) Kenya, Tanzania Central African Republic Cameroon Chad Cameroon Burkina Faso Cote d'lvoire Mali C6te d'Ivoire Niger Benin IMF data sampleb Switzerland Germany, Italy, Netherlands Hungary Austria, Italy Austria Germany, Italy Laos PDR Thailand, Vietnam Zambia (25th) Mozambique, Tanzania, South Africa Zimbabwe (25th) Mozambique, Tanzania, South Africa Nepal Bangladesh, India Paraguay Argentina, Brazil, Chile, Uruguay Bolivia Argentina, Brazil, Chile, Peru Central African Republic (50th) Cameroon; Congo, Rep. of; Congo, Dem. Rep. of Burundi Kenya, Tanzania, Uganda Mali Burkina Faso, Cote d'lvoire, Senegal Rwanda Burundi, Kenya, Tanzania, Uganda Chad (75th) Cameroon, Nigeria Malawi Botswana, Mozambique, Zambia, Zimbabwe Niger Benin, Burkina Faso, Nigeria, Togo Burkina Faso C6te d'lvoire, Togo Uganda Kenya, Tanzania Note: 25th, 50th, and 75th denote the countries with transit infrastructure values closest to these percentile values. 'Transit countries coincide with the port of entry reported by the shipping company. In the case of Zambia and Malawi, Zimbabwe is also a transit country. The countries for which there are no transit or own-infrastructure data (see note in table A-2) are not included here as they were not used in the restricted sample. bWithout specific knowledge of the source of the import and transit route, we must take the average infrastructure measure over all the transit countries reported by UNCTAD (see table A-1). TABLE A-4. Summary Statistics for Shipping Data Sample, 1998 Full sample Mean Standard Landlocked Coastal Variable Mean deviation countries countries Whole sample Transport cost T 6.59 3.50 8.21 4.62 Distance Total 9.58 2.39 9.76 9.37 Over sea 10.5 3.75 10.10 10.90 Over land 0.979 1.27 0.979 0.353 Income per capita' 4.01 8.11 3.57 4.56 Number of countries 64 64 35 29 Restricted sampleb Transport cost, T 5.98 3.49 7.95 4.38 Distance Total 9.75 2.60 10.20 9.37 Over sea 11.20 3.92 11.60 11.00 Over land 0.63 0.57 1.00 0.34 Income per capitaa 4.21 8.24 3.54 4.76 Number of countries 47 47 21 26 ,Average for 1990-95. bCountries for which infrastructure data are available. Note that the in- frastructure data are available only until 1995. Here we use the average for 1990-95. Source: Authors' calculations. TABLE A-5. Summary Statistics for the IMF Data Sample, 1990 Standard Variable Mean deviation lnMa 2.89 2.80 Inta 0.49 0.62 Ininf 0.23 0.47 Ininftran 0.11 0.29 Indistance 8.49 1.54 border 0.03 0.16 isldummy 0.16 0.36 Idldummy 0.15 0.36 Africab 0.26 0.44 AAb 0.07 0.26 AAdistanceb 7.68 0.91 InY 24.29 2.26 InY/cap 7.80 1.66 Note: See table A-1 for variable descriptions and sources. See table A-2 for the country sample. aThese values correspond to the uncensored values of the variables. The sample size is 3,577 for those statis- tics. The median for t is 1.28. bThe statistics correspond to African partners only. Similar statistics hold for the partner country variables. Source: Authors' calculations. 477 478 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 TABLE A-6. Quartile Values for the IMF Data Sample, 1990 Percentile Variable 25th 50th 75th CIF/FOB (all sample) 1.11 1.28 1.83 CIF/FOB (coasral) 1.10 1.29 1.82 CIF/FOB (landlocked) 1.10 1.23 1.91 inf (landlocked) 1.48 1.82 2.61 inftran 1.18 1.37 1.59 inf 0.95 1.41 1.81 distance 4,536 7,555 10,729 distance (landlocked) 4,078 6,742 9,922 Note: See table A-1 for variable descriptions and sources. All vari- ables correspond to all country samples except for first and last row, which refer to landlocked importers only. Source: Authors' calculations. REFERENCES Amjadi, Azita, and Alexander Yeats. 1995. "Have Transport Costs Contributed to the Relative Decline of African Exports? Some Preliminary Evidence." Working Paper. World Bank, Washington, D.C. Amjadi, Azita, Ulrich Reincke, and Alexander Yeats. 1996. "Did External Barriers Cause the Marginalization of Sub-Saharan Africa in World Trade?" World Bank Discussion Paper no. 348. Washington, D.C. Anderson, James E. 1979. "A Theoretical Foundation for the Gravity Equation." Ameri- can Economic Review 69(1):106-16. Bergstrand, Jeffrey H. 1985. "The Gravity Equation in International Trade: Some Microeconomic Foundations and Empirical Evidence." Review of Economics and Statistics 67(3):474-81. Bougheas, S. and others. 1999. "Infrastructure, Transport Costs, and Trade." Journal of International Economics 47:169-89. Canning, David. 1998. "A Database of World Infrastructure Stocks, 1950-1995." World Bank Research Paper, Washington, D.C. CIA. 1998. World Factbook. Washington, D.C.: Central Intelligence Agency. Coe, David T., and Alexander W. Hoffmaister. 1998. "North-South Trade: Is Africa Unusual?" Working Paper of the International Monetary Fund, Washington, D.C. Collier, Paul. 1995. "The Marginalization of Africa," International Labour Review 134(4-5):541-57. Collier, Paul, and Jan Willem Gunning. 1999. "Explaining African Economic Perfor- mance." Journal of Economic Literature, 37:64-111. Deardorff, A. V. 1998. "Determinants of Bilateral Trade: Does Gravity Work in a Neo- classical World?" In J. A. Frankel (ed.), The Regionalization of the World Economy. Chicago: University of Chicago Press. Feenstra, Robert C. 1998. "Integration of Trade and Disintegration of Production in the Global Economy." Journal of Economic Perspectives 12(4):31-50. Limao and Venables 479 Fitzpatrick, Gary. 1986. Direct Line Distances. Metuchen, N.J.: Scarecrow Press. Finger, Michael J., and Alexander Yeats. 1976. "Effective Protection by Transportation Costs and Tariffs: A Comparison of Magnitudes." Quarterly Journal of Economics 90(1):169-76. Foroutan, Faezeh, and Lant Pritchett. 1993. " Intra-Sub-Saharan African Trade: Is It Too Little?" Journal of African Economies 2(1):74-105. Frankel, Jeffrey A. 1997. Regional Trading Blocs in the World Economic System. Wash- ington, D.C.: Institute for International Economics. Geracci, Vincent J., and Wilfried Prewo. 1977. "Bilateral Trade Flows and Transport Costs." Review of Economic Statistics 59(1):67-74. Hummels, David. 1998a. "Data on International Transport Costs: A Report Prepared for the World Bank." Development Economics Department, World Bank, Washing- ton, D.C. .1998b. "Towards a Geography of Transport Costs." Department of Economics, University of Chicago, mimeo. IMF. Various years. Direction of Trade Statistics. Washington, D.C.: IMF. Livingstone, Ian. 1986. "International Transport Cost and Industrial Development in the Least Developed African Countries," Industry and Development (19):1-54. Moneta, C. 1959. "The Estimation of Transport Costs in International Trade." Journal of Political Economy 67:41-58. Radelet, Stephen, and Jeffrey Sachs. 1998. "Shipping Costs, Manufactured Exports and Economic Growth." Harvard Institute for International Development, Cambridge, Mass. Mimeo. Rodrik, Dani. 1998. "Trade Policy and Economic Performance in Sub-Saharan Africa." NBER Working Paper no. 6562, NBER, Cambridge, Mass. UNCTAD. 1997. "Selected Basic Transport Indicators in the Landlocked Countries," UNCTAD/LDC/97, United Nations, New York. . 1995a. "A Transport Strategy for Land-locked Developing Countries. Report of the Expert Group on the Transport Infrastructure for Land-locked Developing Coun- tries." TD/B/453/Add.l/Rev.1, UN Publications, adopted by the UNCTAD Secretariat, United Nations, New York. .1995b. "Review of Maritime Transport." TD/B/C.4. New York: United Nations. World Bank. 1998. World Development Indicators. Washington, D.C.: World Bank. i i I i i I THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 48I-490 A New Development Database Deposit Insurance around the World Asli Demirguic-Kunt and Tolga Sobaci In the past two decades, in a series of banking crises around the world, banks have become systematically insolvent. These crises have occurred in developed and develop- ing economies alike. To make such financial system breakdowns less likely and to limit their costs if they occur, policymakers feel the need for financial safety nets. These in- clude such policies as implicit or explicit deposit insurance, a lender of last resort func- tion of the central bank, bank insolvency resolution procedures, and bank regulation and supervision. Of these policies, explicit deposit insurance has been gaining popu- larity in recent years. Since the 1980s the number of countries with explicit deposit insurance schemes almost tripled, with most OECD countries and an increasing num- ber of developing economies adopting some form of explicit depositor protection. In 1994 deposit insurance became the standard for the newly created single banking market of the European Union. Establishing an explicit deposit insurance scheme became part of the generally accepted best practice advice given to developing economies. I. THE ORIGIN OF THE DATABASE Given the complexities involved in safety-net design and operation, policymakers often request technical assistance from the World Bank, particularly on the de- sign of deposit insurance. Until recently, bank staff were unable to give sound policy advice because of the absence of a cross-country data set on deposit in- surance characteristics and a lack of empirical evidence on how different deposit insurance designs affect banking outcomes. A recent World Bank research project has started to fill this gap by collecting a cross-country data set and using it to develop much-needed empirical evidence (Demirguci-Kunt and Kane 1998). This article presents this data set on deposit insurance system arrange- ments currently in place around the world. A large section of the data set is con- structed using the survey results of an International Monetary Fund study by Garcia (1999) and earlier sources such as Kyei (1995) and Talley and Mas (1990). Additionally, information from other country sources is also compiled to double- check the data sets. Most of the data are coded through dummy variables to represent the presence or absence of the deposit insurance features. A few other features that are not suitable for binary coding are categorized using a range of Ash Demirgiiu-Kunt is Lead Economist of the Development Research Group at the World Bank. His e-mail address is ademirguckunt@worldbank.org. Tolga Sobaci is a student at Harvard University and his e-mail address is tolgasobaci@yahoo.com. 0 2001 The International Bank for Reconstruction and Development / THE WORLD BANK 481 482 THE WORLD BANK ECONOMIC REVIEW, VOL. I5, NO. 3 numeric values. The main motive for presenting the data in this format is to enable researchers to process the data sets for quantitative analysis using computer applications. The database (a Microsoft® Excel spreadsheet), further details on its construction, including data sources and individual country notes, are avail- able on the Web site www.worldbank.org/researchlinterests/confs/upcoming/ deposit_insurance/home.htm. II. FEATURES OF THE DEPOSIT INSURANCE SYSTEM DATABASE Table 1 provides information on deposit insurance design features for the 71 countries with explicit schemes. The following section describes deposit insur- ance features and information on the presentation methodology of the database. Explicit or Implicit Deposit Insurance TYPE. The first variable in the database identifies the form of the deposit insur- ance, whether explicit or implicit. Deposit insurance is explicit if some form of legislation, such as the central bank law, banking law, or the constitution, es- tablishes a guarantee scheme for deposits. In the absence of such formal arrange- ments, we assume that the country has an implicit deposit insurance system. Countries with explicit deposit insurance systems are coded 1, and the other countries are coded 0. The database has information on 178 countries, but the table reports only the design features of 71 explicit schemes. DATE ENACTED OR REVISED. This variable identifies the year in which an ex- plicit deposit insurance system was enacted and the year any revisions were made. Further details are provided in country notes. Coverage Variables Deposit insurance systems vary in the extent and amount of coverage that they provide depositors. The schemes specify the types of deposits, types of institu- tions, and the maximum amount of deposits guaranteed. EXTENT OF COVERAGE. Systems offering coverage to deposits denominated in foreign currencies are coded 1 and systems excluding such deposits are coded 0. Most countries with explicit insurance systems do not extend coverage to inter- bank deposits. Deposit insurance systems extending coverage to interbank de- posits are coded 1 and systems that do not are coded 0. AMOUNT OF COVERAGE. The amount of coverage provided by deposit insur- ance schemes varies. The database provides different coverage variables and a variable for the existence of coinsurance arrangements. COVERAGE LIMITS i. This variable provides the coverage limits of the insurance schemes that were in effect during the first half of 1999 in U.S. dollars or ECU. Demirgiu-Kunt and Sobaci 483 COVERAGE RATIOS. This variable shows the ratio of the coverage limits to 1998 GDP per capita. COINSURANCE. This variable is coded 1 if there is coinsurance, and 0 otherwise. Funding Variables FUNDING TYPE. Deposit insurance schemes fall into two categories in the way they are funded by banks. The most conventional type is the funded system, in which the member institutions make periodic contributions to an established, permanent fund. The alternative type, the unfunded system, has no permanently maintained fund, and members are required to contribute to the fund after a bank failure. Funded systems are coded 1, and unfunded systems are coded 0. ANNUAL PREMIUMS. This variable provides the banks' annual premiums that are applied to the assessment base (generally deposits or insured deposits). If con- tributions are made after banking problems, this is indicated. RiSK-ADJUSTED PREMIUMS. Banks contribute to the fund by paying periodic premiums at either variable or fixed rates. Risk-adjusted systems are coded 1; fixed premiums are coded 0. SOURCE OF FUNDING. In addition to the premiums collected from the banks, most insurance schemes can also resort to public funds when needed. Systems exclusively funded by the banks are coded 0. Systems exclusively funded by public funds are coded 2. Systems with access to both sources are classified as "jointly funded" and coded 1. Administration and Membership Variables ADMINISTRATIVE FORM. Systems administered by official authorities are coded 1, those administered by private authorities are coded 3, and those administered jointly are coded 2. Private administrators typically have limited authority. These are mentioned in the country notes. TYPE OF MEMBERSHIP. Membership in deposit insurance systems may be com- pulsory or voluntary. As of spring 1999 a majority of the deposit insurance schemes were compulsory. The compulsory systems are coded 1, and the volun- tary systems are coded 0. III. CONCLUDING REMARKS The cross-country database of deposit insurance design features described herein is part of a broader research project to understand the impact of deposit insur- ance design on bank stability, market discipline, and financial development. This database can be used to investigate a wide range of issues including when devel- TABLE 1. Explicit Deposit Insurance Schemes around the World Foreign Interbank Type currencies deposits explicit = 1 Date enacted/ yes = 1 yes = 1 Coverage limits-I Coverage Country implicit = 0 revised no = 0 no = 0 US$ or ECU ratios Argentina 1 1979/1995 1 0 30,000 3 Austria 1 1979/1996 1 0 $24,075 but 1 coinsurance for businesses Bahrain 1 1993 1 0 5,640 1 Bangiadesh 1 1984 0 0 2,123 6 Belgium 1 1974/1995 1 0 15,000 ECU until 1 year 2000 Brazil 1 1995 1 0 17,000 4 Bulgaria 1 1995 1 0 1,784 1 Cameroon 1 1999 0 1 5,336 9 Canada 1 1967 0 1 40,770 2 Central African 1 1999 0 1 3,557 13 Republic Chad 1 1999 0 1 3,557 15 Chile 1 1986 1 0 demand deposits in 1 full and 90% co- insurance to UF 120 of $3,600 for savings deposits Colombia 1 1985 0 1 in full until 2001, 2 then coinsurance to $5,500 Croatia 1 1997 1 0 15,300 3 Czech Republic 1 1994 0 0 coinsurance to 2 $11,756 Denmark 1 1988/1998 1 0 20,000 ECU 1 Dominican Republic 1 1962 1 0 coinsurance to 7 $13,000 Ecuador 1 1999 1 1 in full to year 2001 El Salvador 1 1999 1 0 4,720 2 Equatorial Guinea 1 1999 0 1 3,557 3 Estonia 1 1998 1 0 coinsurance 90% of 0 $1,383, but 20,000 ECU in year 2010 Finland 1 1969/1992/ 1 0 29,435 1 1998 484 Risk- Source of Co- Permanent adjusted funding Administration insurance fund premiums 0 = private official = I Membership yes = 1 funded = 1 Annual yes = 1 1 = joint joint = 2 compulsory =I no = 0 unfunded = 0 premiums no = 0 2 = public private 3 voluntary 0 0 1 risk-based, 1 0 3 1 0.36 to 0.72 1 0 pro rata, ex post 0 1 3 1 0 0 ex post 0 0 2 1 0 1 0.005 0 1 1 1 0 1 0.02 + 0.04 0 1 2 1 0 1 0.3 0 0 3 1 0 1 risk-based to 0.5 1 1 2 1 0 1 risk-based: 0.15% of 1 1 2 0 deposits + 0.5% of net non-performing loans 0 1 0.33 max 0 1 1 1 0 1 risk-based: 0.15% of 1 1 2 0 deposits + 0.5% of net nonperforming loans 0 1 risk-based: 0.15% of 1 1 2 0 deposits + 0.5% of net nonperforming loans 1 0 none 0 2 1 1 1 1 0.3 0 0 1 1 0 1 0.8 0 1 2 1 1 1 commercial banks 0.5, 0 1 1 1 savings banks 0.1 0 1 0.2 (maximum) 0 1 2 1 1 1 0.1875 0 1 2 0 0 1 0.65 0 n.a. 1 1 0 1 risk-based, 0.1 to 0.3 1 1 1 1 0 1 risk-based: 0.15% of 1 1 2 0 deposits + 0.5% of net nonperforming loans 1 1 0.5 (maximum) 0 1 2 1 0 1 risk-based: 0.05 to 0.3 1 1 3 1 (continued) 485 TABLE 1. (continued) Foreign Interbank Type currencies deposits explicit = 1 Date enacted/ yes = 1 yes = 1 Coverage limits-i Coverage Country implicit = 0 revised no = 0 no = 0 US$ or ECU ratios France 1 1980/1995 1 0 65,387 3 Gabon 1 1999 0 1 5,336 1 Germany 1 1966/1969/ 1 0 private: 30% of 1 1998 capital; official coinsurance 90% to 20,000 ECU Gibraltar 1 1998 1 n.a. lesser of 90% coinsurance or 20,000 ECU Greece 1 1993/1995 1 0 20,000 ECU 2 Hungary 1 1993 1 0 4,165 ECU or 1 $4,564 Iceland 1 1985/1996 1 0 20,000 ECU 1 India 1 1961 1 0 2,355 6 Indonesia 1 1998 Blanket guarantee Ireland 1 1989/1995 1 0 coinsurance 90% to 1 15,000 ECU Italy 1 1987/1996 1 0 125,000 6 Jamaica 1 1998 1 0 5,512 2 Japan 1 1971 0 0 $71,000, but in full until March 2001 Kenya 1 1985 1 1 1,750 5 Korea 1 1996 0 0 $14,600, but in full 0 until 2000 Latvia 1 1998 1 0 $830 until 2000 0 Lebanon 1 1967 0 1 3,300 1 Lithuania 1 1996 1 0 $6,250 then 2 coinsurance Luxembourg 1 1989 1 0 coinsurance 90% to 0 ECU 15,000 through 1999, then to ECU 20,000 Macedonia 1 1996 1 0 coinsurance 75% to 0 $183 Malaysia 1 1998 Blanket guarantee Marshall Islands 1 1975 1 1 100,000 Mexico 1 1986/1990 1 1 in full except sub- ordinated debt until 2005 Micronesia 1 1963 1 1 100,000 Netherlands 1 1979/1995 1 0 20,000 ECU 1 486 Risk- Source of Co- Permanent adjusted funding Administration insurance fund premiums 0 = private official = 1 Membership yes = 1 funded = 1 Annual yes = 1 1 = joint joint = 2 compulsory = 1 no = 0 unfunded = 0 premiums no = 0 2 public private = 3 voluntary 0 0 0 on demand but limited 0 0 3 1 0 1 risk-based: 0.15% 1 1 2 0 deposits + 0.5% of net nonperforming loans 1 1 official is 0.03 but can 0 0 3 1 be doubled 1 0 administrative expenses 0 0 2 1 and ex post contributions 0 1 decreasing by size: 0 0 2 1 1.250 to 0.025 0 1 risk-based to 0.3 1 1 2 1 1 1 0.15 0 0 1 1 0 1 0.05 0 1 1 1 1 1 0.2 0 0 1 1 0 0 risk-adjusted ex post 1 1 2 1 0.4 to 0.8 0 1 0.1 0 1 1 1 0 1 0.0048 + 0.036 0 1 2 1 0 1 0.15 0 1 1 1 1 0.05 0 1 1 1 0 1 0.3 0 1 1 1 0 1 0.05 0 1 2 1 1 1 1.5 0 1 1 1 1 0 ex post 0 0 3 1 1 1 1.5%, risk-based 1% 1 1 2 0 to 5% 0 1 risk-based, 0.00 to 0.27 1 0 1 0 0 1 0.3 (max 0.5) plus 0.7 0 1 1 1 as needed 0 1 risk-based, 0.00 to 0.27 1 0 1 0 0 0 ex post 0 1 1 1 (continued) 487 TABLE 1. (continued) Foreign Interbank Type currencies deposits explicit = 1 Date enacted/ yes = 1 yes = 1 Coverage limits-1 Coverage Country implicit = 0 revised no = 0 no = 0 US$ or ECU ratios Nigeria 1 1988/1989 0 1 $588 (at market 2 exchange rate), $2,435 (at official exchange rate) Norway 1 1961/1997 1 0 260,800 8 Oman 1 1995 1 0 Coinsurance 75% to 9 $52,630 Peru 1 1992 1 0 21,160 9 Philippines 1 1963 1 1 2,375 3 Poland 1 1995 1 0 1,000 ECU, then 90% 0 coinsurance for the next 4,000 ECU Portugal 1 1992/1995 1 0 15,000 ECU, co- 1 insurance to 45,000 ECU Congo, Rep. 1 1999 0 1 3,557 5 Romania 1 1996 1 0 3,600 2 Slovak Republic 1 1996 1 0 7,900 2 Spain 1 1977/1996 1 0 15,000 ECU through 1 1999, then 20,000 ECU Sri Lanka 1 1987 0 0 1,470 2 Sweden 1 1996 1 0 28,663 ECU, $31,412 1 Switzerland 1 1984/1993 0 0 19,700 1 Taiwan, China 1 1985 0 0 38,500 3 Tanzania 1 1994 0 0 376 2 Thailand 1 1997 Blanket guarantee Trinidad and 1 1986 1 1 7,957 2 Tobago Turkey 1 1983 1 0 in full Uganda 1 1994 0 0 2,310 8 Ukraine 1 1998 1 0 250 0 United Kingdom 1 1982/1995 1 0 Larger of 90% co- I insurance to $33,333 or 22,222 ECU United States 1 1934/1991 1 1 100,000 3 Venezuela, RB de 1 1985 0 0 7,309 2 Source: Demirgiiu-Kunt and Sobaci 2000. 488 Risk- Source of Co- Permanent adjusted funding Administration insurance fund premiums 0 = private official = 1 Membership yes = 1 funded = 1 Annual yes = 1 1 = joint joint = 2 compulsory = 1 no = 0 unfunded = 0 premiums no = 0 2 = public private = 3 voluntary = 0 0 1 0.9375 0 1 1 1 0 1 0.005 of assets and 0 1 3 1 0.01 of total deposits 1 1 0.02 0 1 1 1 0 1 risk-based from 0.65 to 1 1 2 1 1.45 0 1 0.2 0 1 1 1 1 1 not more than 0.4 0 1 1 1 1 1 risk-based, 0.08 to 0.12 1 1 1 1 + more in emergencies 0 1 risk based: 0.15% of 1 1 2 0 deposits + 0.5% of net nonperforming loans 0 1 risk-based: 0.3 to 0.6 1 1 2 1 0 1 0.1 to 0.3 for banks 0 1 2 1 0 1 maximum of 0.2 0 1 2 1 0 1 0.15 0 1 1 0 0 1 risk-based, 0.5 now, 1 1 1 1 0.1 later (future date is not available) 0 0 on demand 0 0 3 0 0 1 0.015 0 1 1 0 0 1 0.1 0 1 3 1 0 1 0.2 0 1 1 1 0 1 risk-based 1.0 to 1.2 1 1 1 1 0 1 0.2 0 1 1 1 0 1 0.5 plus special charges 0 1 1 1 1 0 on demand 0 0 3 1 0 1 risk-based, 0.00 to 0.27 1 1 1 1 0 1 2 0 1 1 1 489 490 THE WORLD BANK ECONOMIC REVIEW, VOL. 15, NO. 3 oping countries should adopt explicit insurance schemes and how these schemes should be designed (see Demirguc-Kunt and Detragiache 2000; Demirgiuc-Kunt and Huizinga 2000; Cull and others 2000; Kane 2000). REFERENCES The word "processed" describes informally reproduced works that may not be commonly available through library systems. Cull, Robert, Lemma W. Senbet, and Marco Sorge. 2000. "Deposit Insurance and Fi- nancial Development." World Bank, Development Economics Research Group, Wash- ington, D.C. Available online at www.worldbank.org/research/interest/confs/upcom- ing/deposit_insurance/home.htm. Demirgiiu-Kunt, Asli, and Edward J. Kane. 1998. "Deposit Insurance: Issues of Principle, Design and Implementation." Research Proposal. World Bank, Development Econom- ics Research Group, Washington, D.C. Processed. Demirguc-Kunt, Asli, and Enrica Detragiache. 2000. "Does Deposit Insurance Increase Banking System Stability? An Empirical Investigation." World Bank, Development Economics Research Group, Washington, D.C. Available online at www.worldbank.org/ research/interest/confs/upcoming/deposit insurance/home.htm. Demirgiuc-Kunt, Asli, and Harry Huizinga. 2000. "Market Discipline and Financial Safety Net Design." World Bank, Development Economics Research Group, Washington, D.C. Available online at www.worldbank.org/research/interest/confs/upcoming/ deposit insurance/home.htm. Demirgiiu-Kunt, Asli, and Tolga Sobaci. 2000. "Deposit Insurance around the World: A Database." World Bank, Development Economics Research Group, Washington, D.C. Available online at www.worldbank.org/research/interest/con fs/upcoming! deposit insurance/home.htm. Garcia, Gillian. 1999. "Deposit Insurance: A Survey of Actual and Best Practices." IMF Working Paper 99/54. International Monetary Fund, Washington, D.C. Kane, Edward J. 2000. "Designing Financial Safety Nets to Fit Country Circumstances." World Bank, Development Economics Research Group, Washington, D.C. Available online at www.worldbank. org/research/interest/con fs/upcoming/deposit_insurance/ home.htm. Kyei, Alexander. 1995. "Deposit Protection Arrangements: A Survey." IMF Working Paper 95/134. International Monetary Fund, Washington, D.C. Talley, Samuel H., and Ignacio Mas. 1990. "Deposit Insurance in Developing Countries." Policy Research Working Paper 548. World Bank, Washington, D.C. The World Bank Economic Review Author Index to Volume 15, 2001 Number 1: 1-176 Number 2: 177-340 Number 3: 341-494 Alderman, Harold. Multi-Tier Targeting of Social Assistance: The Role of Intergovernmental Transfers 33 Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh. New Tools in Comparative Political Economy: The Database of Political Institutions 165 Brock, William A., and Steven N. Durlauf. Growth Empirics and Reality 229 Clarke, George. See Beck, Thorsten. Collier, Paul, Anke Hoeffler, and Catherine Pattillo. Flight Capital as a Portfolio Choice 55 Demirguc-Kunt, Ash, and Tolga Sobaci. Deposit Insurance around the World 481 Denizer, Cevdet. See de Melo, Martha. Durlauf, Steven N. See Brock, William A. Easterly, William, and Ross Levine. It's Not Factor Accumulation: Stylized Facts and Growth Models 177 Eichengreen, Barry. Capital Account Liberalization: What Do Cross-Country Studies Tell Us? 341 Gelb, Alan. See de Melo, Martha. Glewwe, Paul, and Elizabeth M. King. The Impact of Early Childhood Nutritional Status on Cognitive Development: Does the Timing of Malnutrition Matter? 81 De Gregorio, Jose, and Rodrigo 0. Valdes. Crisis Transmission: Evidence from the Debt, Tequila, and Asian Flu Crises 289 Groff, Alberto. See Beck, Thorsten. Gunewardena, Dileni. See van de Walle, Dominique. Gylfason, Thorvaldur, Tryggvi Thor Herbertsson, and Gylfi Zoega. Ownership and Growth 431 Herbertsson, Tryggvi Thor. See Gylfason, Thorvaldur. Hoeffler, Anke. See Collier, Paul. Kaminsky, Graciela L., Richard K. Lyons, and Sergio L. Schmukler. Mutual Fund Investment in Emerging Markets: An Overview 315 Keefer, Philip. See Beck, Thorsten. King, Elizabeth M. See Glewwe, Paul. Klenow, Pete. Comment on "It's Not Factor Accumulation: Stylized Facts and Growth Models" 221 Levine, Ross. See Easterly, William. 491 492 Author Index Limao, Nuno, and Anthony J. Venables. Infrastructure, Geographical Disadvantage, Transport Costs, and Trade 451 Lyons, Richard K. See Kaminsky, Graciela L. de Melo, Martha, Cevdet Denizer, Alan Gelb, and Stoyan Tenev. Circumstance and Choice: The Role of Initial Conditions and Policies in Transition Economies 1 Pattillo, Catherine. See Collier, Paul. Pritchett, Lant. Comment on "Growth Empirics and Reality" 273 Pritchett, Lant. Where Has All the Education Gone? 367 Ravallion, Martin. The Mystery of the Vanishing Benefits: An Introduction to Impact Evaluation 115 Romer, Paul. Comment on "It's Not Factor Accumulation: Stylized Facts and Growth Models" 225 Sala-i-Martin, Xavier. Comment on "Growth Empirics and Reality" 277 Schmukler, Sergio L. See Kaminsky, Graciela L. Sobaci, Tolga. See Demirgiiu-Kunt, Ash. Solow, Robert M. Applying Growth Theory across Countries 283 Tenev, Stoyan. See de Melo, Martha. Valdes, Rodrigo 0. See Gregorio, Jose De. Venables, Anthony J. See Limao, Nuno. Wacziarg, Romain. Measuring the Dynamic Gains from Trade 393 van de Walle, Dominique, and Dileni Gunewardena. Does Ignoring Heterogeneity in Impacts Distort Project Appraisals? An Experiment for Irrigation in Vietnam 141 Walsh, Patrick. See Beck, Thorsten. Zoega, Gylfi. See Gylfason, Thorvaldur. The World Bank Economic Review Title Index to Volume 15, 2001 Number 1: 1-176 Number 2: 177-340 Number 3: 341-494 Applying Growth Theory across Countries 283 Robert M. Solow Capital Account Liberalization: What Do Cross-Country Studies Tell Us? 341 Barry Eichengreen Circumstance and Choice: The Role of Initial Conditions and Policies in Transition Economies 1 Martha de Melo, Cevdet Denizer, Alan Gelb, and Stoyan Tenev Comment on "Growth Empirics and Reality" 273 Lant Pritchett Comment on "Growth Empirics and Reality" 277 Xavier Sala-i-Martin Comment on "It's Not Factor Accumulation: Stylized Facts and Growth Models" 221 Pete Klenow Comment on "It's Not Factor Accumulation: Stylized Facts and Growth Models" 225 Paul Romer Crisis Transmission: Evidence from the Debt, Tequila, and Asian Flu Crises 289 Jose De Gregorio and Rodrigo 0. Valdes Deposit Insurance around the World 481 Asli Demirgiiu-Kunt and Tolga Sobaci Does Ignoring Heterogeneity in Impacts Distort Project Appraisals? An Experiment for Irrigation in Vietnam 141 Dominique van de Walle and Dileni Gunewardena Flight Capital as a Portfolio Choice 55 Paul Collier, Anke Hoeffler, and Catherine Pattillo Growth Empirics and Reality 229 William A. Brock and Steven N. Durlauf The Impact of Early Childhood Nutritional Status on Cognitive Development: Does the Timing of Malnutrition Matter? 81 Paul Glewwe and Elizabeth M. King Infrastructure, Geographical Disadvantage, Transport Costs, and Trade 451 Nuno Limao and Anthony J. Venables It's Not Factor Accumulation: Stylized Facts and Growth Models 177 William Easterly and Ross Levine 493 494 Title Index Measuring the Dynamic Gains from Trade 393 Romain Wacziarg Multi-Tier Targeting of Social Assistance: The Role of Intergovernmental Transfers 33 Harold Alderman Mutual Fund Investment in Emerging Markets: An Overview 315 Graciela L. Kaminsky, Richard K. Lyons, and Sergio L. Schmukler The Mystery of the Vanishing Benefits: An Introduction to Impact Evaluation 115 Martin Ravallion New Tools in Comparative Political Economy: The Database of Political Institutions 165 Thorsten Beck, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh Ownership and Growth 431 Thorvaldur Gylfason, Tryggvi Thor Herbertsson, and Gylfi Zoega Where Has All the Education Gone? 367 Lant Pritchett 10r.arv books from OXFORD Human Well-Being and the Natural Environment PARTHA DASGUPTA 327 pages Novembe, 2001 0-1 9-924788-9 E25 00 $35.00 Hardback Development Strategy and Management of the Market Economy Volume 1 Edmond Malinvaud, Jean-Claude Milleron, Mustapha K. Nabli, Amartya K. Sen, Arjun Sengupta, Nicholas Stern, Joseph E. Stiglitz, and Kotaro Suzumura oOUP/uorn DNOrIONS 315 pages November 2000 0-195924134-1 E14.99 21 .95 Paperback Governance, Equity, and Global Markets The Annual Bank Conference on Development Economics - Europe Edited by Joseph E. Stiglitz and Pierre-Alain Muet M L 352 pages August 2001 0-1 9-9241 55-4 E25.00 S39.95 Hardback Poverty and Undernutrition Theory, Measurement, and Policy PETER SVEDBERG WIDER STODIESIN LD, D-LPMENT ECONOMICS 378 pages October 2000 0-19-829268-6 E45.00 $74 00Hardback Resource Abundance and Economic Development Edited by Richard. M. Auty WIDER SIUIE5 IN D0...WOPM-'T ECONOMICS 356 pagesJune 2001 0-19-924688-2 E45.00 $70.00 Hardback NEW IN PAPERBACK Education and Development Measuring the Social Benefits WALTER W. McMAHoN 320 pages February 2002 0-1 9-924751-XE15.99 $24.95 Paperback NEW IN PAPERBACK The Economics of Regional Trading Arrangements RICHARD POMFRET 280 pages February 2002 0-1 9-924887-7 E20.00 $24.95 Paperback NEW IN PAPERBACK CwMoup.c Co Peasants versus City-Dwellers i t44,c36454534 Taxation and the Burden of Economic Development RAAI K. SAH AND JOSEPH E. STIGLITZ book.orders@oup co- 240 pages Febroary 2002 0-19-925357-9 E13.99 S20.00 Paperback I NI FI- WN I l 'lSt F S OX-FORD Introduce a Friend or Colleague to UNIVERSITY PRESS THE WORLD BANK ECONOMIC REVIEW Simply photocopy this page, fill in the name and address, and Oxford University Press will send a FREE sample copy of The World Bank Economic Review without obligation! Name: Address: City: State: Zip: Country: 01 04infrfa/wberfr Elsewhere: Oxford University Press _ 19)677-1714 Tel: +44 (0) 1865 267907 * . * * E-mail. jnl.orders@oup.co.uk iournals in Economics from X- } V.N I VEWSI I TY lPIRESS Journal of International Economic Law Dedicated to ixWmng thoughui and scholarhteo anon to the rmlation of law to tenanonal economic ami'ty - www.jiel.oupjournals.org _ow-TM-M-., Contemporary Economic Polic I A urrial or WVesten Economuc Assosiation International, CEP pub'shes vxtiQVS o ~\ quallty resarch and analyss on poik. aues of e.idesprasd conc,at www.cep.oupjournals.org Contributions to Political Economy A foirm for acadeiruc discussion of ongmal ideas and arg,uments dr3wn fThom bLnes oi [houg5hl ass. ciated wVith the works *o1 c1aisical pobtical \economisrs Ntarx. I;evnes. and Sr3ffa wwwxpe- _ wwwoep.oupojounnals.orgo_.g ,,>ttry~ ~9 _8' ' ' EeonomOifor Economic Pa ers peerreseo A general rconomrcs loumnal O publishes roiiee kW{r ' as wdl s applid rconoics and ecorvinetncs ECOUO ei .oupoep.oupjournall.orgorg: Econoniic Inqui o laThe(f A oma o Wsia Erioric s OxafoInrdraxl thnois apr pew-reviewedeq lnmA general e -conornics louma l p Es origsnad A loumal of Wei3ern a lasTnheb n Rveofinancial Stud es t p ros r o%ese sgcnrifian ne w eeac i innia nomics, sinasingises nIt'i pww.rfs.oupoournals.orgg esrablisn a balance between theoretical and empuxical studies §1||*s,,6 ; 0 - _ - = :agl Tel: (919) 67 * 4 * 4 4 (O) 1865 267485 Introduce Your Library to OXFORD THE WORLD BANK UNIVERSITY PRESS ECONOMIC REVIEW EDITOR: FranSois Bourguignon, The World Bank, Paris WBER is the most widely read scholarly economic journal in the world and specializes in quantitative development policy analysis. For more information, visit our website at: www.wber.oupjournals.org If you feel that a subscription would benefit students and colleagues, photocopy this page, fill in the name and address, and Oxford University Press will send a FREE sample copy of The World Bank Economic Review without obligation! IfnslilutiSon: _ _ ____ Librarian: AddfIress_ C ___ slate: Ziff eouRtr ____ - 101- - 1! l; ~~~~~~~~010411ntlib/inlwber : ~~~~~~~Elsewhere: Oxford University Press Great Clarendon Street, Oxford OX2 6DP. UK 19) 677-1714 Tel: +44 (0) 1865 267907 c E-mail: jnl.orders@oupco.uk Coming in the next issue of THE WORLD BANK ECONOMIC REVIEW Volume 16, Number 1, 2002 * Eliminating Excessive Tariffs on Exports of Least Developed Countries Marcelo Olarreaga, Francis Ng, and Bernard Hoekman * Imported Machinery for Export Competitiveness Ashoka Mody and Kamil Yilmaz * Trade Policy Options for Chile: The Importance for Market Access David G. Tarr, Glenn W. Harrison, and Thomas F Rutherford * Trade in International Maritime Services: How Much Does Policy Matter? Christina Neagu, Carsten Fink, and Aaditya Mattoo * Banking Risk and Deposit Insurance Luc Laeven * How Different Is the Efficiency of Public and Private Water Companies in Asia? Antonio Estache and Martin A Rossi THE NVORLD BANK 181S H Street, NWV Washilngtoni, DC 20433, USA World \NVide \'Veb: http:,//wwwvN.worldbanik.org/ E-liilil: '1 ! .t.i1;s 1_ 14 o 7801 983 386 5