I THEWORLDRANK ( ECONOMIC REVIEW EDITOR Jaime de Melo, University of Geneva EDITORIAL BOARD Kaushik Basu, CornellUniversity, USA Paul Gertler, WorldBank Jean-Marie Baland, University afNamur, Inderrnit Gill, World Bank Belgium Jan Willem Gunning, Free University, Chong-En Bai, TsinghuaUniversity, The Netherlands China Jeffrey Hammer, World Bank Timothy Besley, London Schoolof Graciela Kaminsky, George Washington Economics, UK University, USA Franfois Bourguignon, WorIdBank Ravi Kanbur, CornellUniversity, USA Kenneth Chomitz, World Bank Peter Lanjouw, World Bank Maureen Cropper, University ofMaryland, Thierry Magnac, Universite'de ToulouseI; USA France Jishnu Das, World Bank Juan-Pablo Nicolini, UniversidadTorcuatodi Klaus Deininger, World Bank Tella,Argentina Asli Demigurf-Kunt, World Bank Boris Pleskovic, WorldBank Stefan Dercon, Oxford University, UK Martin Ravallion, WorIdBank Ishac Diwan, World Bank Ritva Reinikka, WorIdBank Augustin Kwasi Fosu, Ui'? Economic Elisabeth Sadoulet, University of California, CommissionforA?ica (ECA), Ethiopia Berkeley, UUSA Alan Harold Gelb, World Bank Mark Gersovitz,Johns Hopkins Joseph Stiglitz, Columbia University,USA University, USA L. Alan Winters, World Bank The World Bank Economic Review is a professional journal for the dissemination of World Bank-sponsored and other research that may inform policy analysis and choice. It is directed to an international readership among economists and social scientists in government, business, international agencies,universities,and development researchinstitutions.The Review seeks to provide the most current and best research in the field of quantitative development policy analysis, emphasizing 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 illustrate how professional research can shed light on policy choices. Consistency with World Bank policy plays no role in the selection of articles. Articles are drawn from work conducted by World Bank staff and consultants and by outside researchers. Non-Bank contributors are encouraged to submit their work. Before being accepted for publication,articles are reviewed by three referees-one from the World Bank and two &om outside the institution. Articles must also be endorsed by two members of the Editorial Board beforefinal acceptance. For moreinformation,pleasevisittheWeb sitesof the Economic Reviewat Oxford UniversityPress at www.wber.ox6ordjournals.organd at the World Bank at www.worldbank.org/research/joumals. Instructionsfor authorswishingtosubmit articlesare availableonline atwww.wber.ox6ordjournals.org. Pleasedirect alleditorialcorrespondence to the Editor at wbefiorldbankorg. THE WORLD BANK ECONOMIC REVIEW Volume 20 2006 Number 2 The Doha Round and Preference Erosion: A Symposium Bernard Hoekman Doha Merchandise Trade Reform: What Is at Stake for Developing Countries? 169 Kym Anderson, Will Martin, and Dominique van der Mensbrugghe Preference Erosion and Multilateral Trade Liberalization Joseph Francois, Bernard Hoekman, and Miriam Manchin Trade Preferences to Small Developing Countries and the Welfare Costs of Lost Multilateral Liberalization 217 Nuno Lim2o and Marcelo Olarreaga Price Effects of Preferential Market Access: Caribbean Basin Initiative and the Apparel Sector 241 caglar 0zden and Gunjan Sharma Aid and the Supply Side: Public Investment, Export Performance, and Dutch Disease in Low-Income Countries 261 Christopher S. Adam and David L. Bevan Infrastructure, Externalities, and Economic Development: A Study of the Indian Manufacturing Industry 291 Charles R. Hulten, Esra Bennathan, and Sylaja Srinivasan SUBSCRIPTIONS: A subscription to The World Bank Economic Review (ISSN 0258-6770) comprises 3 issues. Prices include postage; for subscribers outside the Americas,issues are sent air freight. AnnualSubscription Rate (Volume 20,3Issues,2006):AcademicIibrarie~Printeditionandsite-wideonline access: US$137/X87/131, Printeditiononly:US$131/2233/125, Site-wideonline accessonly US$123/L78/ 117; Corporate-Priit edition andsite-wideonlineaccess:US$205/X130/195, Printedition only:US$196/ X124/186, Site-wide onlineaccessonly US$185K117/176; Personat-Priit edition and individual online access: US$49/X34/51. Pleasenote:XSterlingratesapplyinthe UK, in Europe,andUS$elsewhere.There may be other subscriptionrates available; for a complete listing, please visit www.wber.oxfordjournals.org/ subscriptions. Readers with mailing addresses in non-OECD countriesand in socialisteconomiesin transitionare eligible to receive complimentarysubsmptionson requestby writingto the UKaddressbelow. Full prepayment in the correct currency is required for all orders. Orders are regarded as firm, and payments are not refundable. Subscriptionsare accepted and entered on a complete volume basis. Claims cannot be considered more than four months after publication or date of order, whichever is later. All subscriptions in Canada are subject to GST. Subscriptions in the EU may be subject to European VAT. If registered, please supply details to avoid unnecessary charges. For subscriptions that include online versions, a proportionof the subscription price may be subject to UK VAT. Personal rates are applicable onlywhen a subscriptionis forindividual use and are not availableif delivery is madeto a corporateaddress. BACKISSUES: The current year and two previous years' issues are available from Oxford University Press. Previous volumes can be obtained from the PeriodicalsService Company,11Main Street, Germantown, NY 12526, USA. E-mail: psc@periodicals.com.Tel: (518) 537-4700. Fax: (518) 537-5899. CONTACT INFORMATION:Journals Customer ServiceDepartment,Oxford UniversityPress,GreatClarendon Street,OxfordOX2 6DP,UK. E-mail:jnls.cust.senr@oxfordjournals.org.Tel: +44(0)1865353907. Fax: +44 (0)1865353485. In tbe Americas, please contact:Joumals Customer Service Department,Oxford University Press, 2001 Evans Road, Cary, NC 27513, USA. E-mail: jnlorders@oxfordjournals.org.Tel: (800) 852- 7323 (toll-freein USAICanada) or (919) 677-0977. Fax: (919) 677-1714, InJapan, please contact:Journals Customer ServiceDepartment, W o r d UniversityPress,1-1-17-SF,Mukogaoka,Bunkyo-ku,Tokyo, 113- 0023,Japan. E-mail: okudaoup@po.iijnet.or.jp.Tel: (03) 3813 1461. Fax: (03) 3818 1522. POSTAL INFORMATION: The World Bank Economic Review (ISSN 0258-6770) is published by Oxford University Press for the International Bank for Reconstruction and DevelopmendTH~WORLD BANK. Send address changes to The World Bank Economic Review, Journals Customer Service Department, Oxford University Press, 2001 Evans Road, Cary, NC 27513-2009. Communications regarding original articles and editorial management should be addressed to The Editor, The World Bank Economic Review, The World Bank, 3, Chernin Louis Dunant, CP66 1211 Geneva 20, Switzerland. DIGITAL OBJECT IDENTIFIERS: For information on dois and to resolve them, please visit www.doi.org. PERMISSIONS: For information on how to request permissions to reproduce articles or informationfrom this journal, please visit www.oxfordjournals.org/jnls/permissions. ADVERTISING: Inquiriesabout advertisingshould be sent to Helen Pearson, OxfordJournals Advertising, PO Box 347, Abingdon OX14 lGJ, UK. E-mail: helen@oxfordads.com.Tel: +44(0)1235201904. Fax: +44 (0)8704296864. DISCLAIMER: Statements of fact and o~inionin the articles in The World Bank Economic Review are those of the respective authorsand contributorsand not of the InternationalBankfor Reconstructionand DevelopmendT~EWORLD BANK or W o r d University Press. Neither Oxford University Press nor the International Bank for Reconstruction and Develo~mendTHEWORLD BANK make any representation, express or implied, in respect of the accuracy of the material in this journal and cannot accept any legal responsibilityor liabilityfor any errors or omissionsthat may be made. The reader should make her or his own evaluation as to the appropriateness or otherwise of any experimental techniquedescribed. PAPER USED: The World Bank Economic Review is printed on acid-free paper that meets the minimum requirements of ANSI Standard 239.48-1984 (Permanence of Paper). INDEX~NGAND ABSTRACTING: The World Bank Economic Review is indexed andlor abstracted by C4B Abstracts, Current ContentdSocialand Behavioral Sciences, Journal of Economic LiteratureEconLit, PAIS International, RePEc (Research in Economic Papers), and Social Services Citation Index. COPYRIGHT O The International Bank for Reconstruction and Develo~mendT~E WORLD BANK 2006 AU rights reserved; no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without prior witten permission of the publisher or a license restricted copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1P 9HE, or in the USA by the Copyright Clearance Center, 222 Rosewood Drive, Danvers, MA 01923. The Doha Round and Preference Erosion: A Symposium Bernard Hoekman The trade and welfare impacts of multilateral liberalization on indivialual countries and groups within countries depend on many factors-including the depth of liberalization by trading partners, the extent of countries' own reforms, the responsiveness of investors to changes in relative prices and market opportunities, and actions by governments to reduce real trade costs. One consequence of multilateral liberalization is that it reduces the value of preferential access to markets that one or more countries have granted to other countries. Such preference erosion has become more of a policy con- cern for the least developed countries following initiatives by many Organi- zation for Economic Co-operation and Development (OECD) members to provide duty-free, quota-free access to their markets on a nonreciprocal basis. But erosion will also affect other developing countries that have received preferences, as well as economies that have signed reciprocal trade agreements. The magnitude of erosion will depend on a variety of factors, including the product and country coverage of preferential schemes, the level of most favored nation restrictions in the markets granting preferential access, the administrative costs associated with using preference programs, the incidence of any preference rents, the depth of liberalization realized in Doha, and the existence of and changes in reciprocal trade agreements. Recent studies of European Union (EU)and U.S. preference arrangements have concluded tlhat the value of preferences-measured by the product of the volume of dutiable exports and the preference margin-is significant for a relatively srn~all number of countries. Thus, U.S. preferences are equal to 5 percent or more of dutiable exports for some 27 countries (Deanand Wainio 2006), while :EU preferences exceed 6 percent of dutiable exports for 16 countries (excludnng preferential trade area partners; Candau and Jean 2006). These studies con- clude that apparel and some agricultural products-especially sugar and bana- nas in the EU-account for the largest share of the value of preferences. While for some countries, trade coverage is large relative to total dutiable exports to the markets concerned, the aggregate value of the preferences-and thus Bernard Hoekman is a member of the Development Research Group of the World Bank. THE WORLDBANK ECONOMICREVLEW,VOL. 20, NO. 2, pp. 165-168 doi:10.1093/wberilhj012 Advance Access publication May 12, 2006 @ The Author 2006. Published by Oxford UniversityPress on behalf of the International Bankfor Reconstructionand Development/ THEWORU) AN All rightsreserved.For permissions, B K. please e-mail: journals.perrnissions@oxfordjournals.org. potential losses-is relatively small and will diminish when the EU reforms in sugar and bananas are implemented.' By enabling disaggregated analysis at the tariff line level, the type of partial equilibrium analysis employed in these studies allows identification of the countries most affected by preference erosion risks and the products con- cerned-though analysis of the use of preferences by eligible countries is needed for a more complete understanding of the extent of preferential access. Such studies may be misleading, however, insofar as they ignore total exports of the countries concerned, the conditions of access offered to other countries for the same products, and the fact that preference margins are often low because most favored nation tariffs are low or zero. Moreover, the value of preferences is better measured by income earned-what matters is the impact on the price actually received by the exporters because the pass-through of preferential access is likely to be incomplete. Account should also be taken of the incidence of preferences on the costs of administering preference schemes, such as complying with origin requirements, which lowers the actual value of preferences. The four articles in this mini-symposium focus on the potential magnitude of-and possible solutions to-preference erosion caused by multilateral liberal- ization, taking into account the factors mentioned above. Francois, Hoekman, and Manchin consider the effects on developingcountries of only OECD coun- tries liberalizingon a most favored nation basis-the most appropriate measure of the scope for preferenceerosion. They find that the loss from full erosion in all OECD markets is some $250 million in real income terms for African least developedcountries and Bangladesh. Most of this is "caused by" the EU: the loss from full erosion in EU markets for the least developed countries that lose is some $600 million. The implication is that liberalization by other OECD coun- tries will benefit some least developed countries by reducing the losses in the EU market. ]?Jet benefits are likely to be greater if developing countries also liberalize. In another contribution, Anderson, Martin, and van der Mensbrugghe conclude that all low-income countries as a group would see real incomes' rise by some $16 billion followingfull global liberalization. For the subset of African least developed countries and Bangladesh, instead of the $250 million loss stemmingfrom OECD liberalization, there would be a gain of some $1.1billion. Thus, deep global reforms can do much to offset erosion losses in major markets for many countries. The magnitude of erosion losses also depends on administrative costs-such as rules of origin-and the prevalenceof nontariff barriers that constrain or raise the costs of market entry. Francois, Hoekman, and Manchin estimate that the ad valorem equivalent of administrative costs averages about 4 percent. Taking this into account lowers the value of preference programs and implies much lower 1. Similar findings are reported in Low, Piermartini, and Richtering (2005, 2006) for the Quad economies (Canada, the European Union, Japan, and the United States) as a group. Hoekman 167 erosion losses for recipient c~untries.~In a similar vein, Ozden and Sharma estimate that Caribbean exporters capture only two-thirds of the preference margin in the U.S. market, with importers capturing the remainder. Other recent analyses cited in the symposium articles have found that the share of rents captured by exporters under other programs may be much 10wer.~ These findings bolster the conclusionsof the disaggregated partial equilibrium studies mentioned above that the absolute magnitude of preference erosion losses is relatively small. Nonetheless, the impact for some countries of substan- tial liberalization is likely to be significant, raising the question of what coulcl be done to address potential losses. There are two broad options: seek a solution within the trading system (tietito trade and trade policy) or use nontrade instruments. The most obvious trade- based option is not to liberalize the products that are the most important source of preference rents. This would imply a significant opportunity cost in liberal- ization forgone and is undesirable from a global welfare perspective. A more efficient trade option is considered by LimHo and Olarreaga. They show that shifting from tariff preferences to a system of equivalent import subsidies; in OECD countries might encourage additional tariff liberalization and reduce distortions created by preferential trade. Essentially, their suggestion is one way that the concept of aid for trade might be applied to preference erosion. Francois, Hoekman, and Manchin argue in favor of aid for trade to assist countries in dealing with the adjustment costs associated with global trade reforms and improving their capacity to exploit trade opportunities and diversify their economies. As stressed in the literature, aid for trade should be seen as a complement zmd not as a substitute for global trade liberalization (Prowse 2006). Anderson, Martin, and van der Mensbrugghe show that the potential positive net effects of global trade reform are considerable and that the partial reforms that may emerge from the Doha Round may do little to benefit developing countri~es. Their findings that global free trade would benefit most developing countries and that developing countries' own liberalization is important in the context of the type of partial liberalization that is likely under the Doha Round are particularly relevant from a preference erosion perspective. They imply that losses to preference recipients from OECD liberalization can be offset by gains in other markets-those of other developing countries and those of OECD members, which do not already provide full duty-free and quota-free access to markets. 2. For example, Anson and others (2005)document complex rules of origin and low use rates for preferences under the North American Free Trade Agreement (NAFTA).In a detailedstudy of the costs of rules of origin for Mexicanexportersof textilesand clothingunderNAITA, Cadotand others (2005)find that about half the value of preferences is captured by U.S. importersand that U.S. producers of textile intermediateproducts sell at a higher price in the "captive" Mexican market. 3. Olarreaga and Ozden (2005) find that some exporters under the U.S. African Growth and Opportunity Act get less than 50 percent of the preference margin. The articles in this mini-symposium are by no means the final word on the value of global trade reform or on the potential magnitude of erosion and its effects. Clearly, such effects are multidimensional, and more research is needed on all of the issues addressed in the articles and on issues that are not. Of particular importance is incorporating into the analysis nontariff policies and the cost of complying with them. This extends beyond rules of origin and includes trade costs in the exporting country (and thus the payoffs to trade facilitation), as well as the differential costs of compliance with destination country product standards and regulatory requirements in each exporting coun- try. The literature effectively ignores trade in services and the policies that affect such services. Clearly, this needs to be remedied, given the increasing tradability of services and the importance of producer services as a determinant of competitiveness. Anson, J., 0. Cadot, A. Estevadeordal, J. de Melo, A. Suwa-Eisenmann, and B. Tumurchudur. 2005. "Rulesof Originin North-SouthPreferentialTrading Arrangementswith an Applicationto NAFTA." Review of International Economics 13(3):1501-17. Cadot, O., C. Carrere,J. de Melo, and A. Portugal-Perez. 2005. "Market Access and Welfare under Free Trade Agreements: Textiles under NAFTA." World Bank EconomicReview 19(3):379-405. Candau, F., and S. Jean. 2006. "What Are EU Trade PreferencesWorth for Sub-SaharanAfrica and Other Developing Countries?"Working Paper 2005-19. Paris: Centre #Etudes Prospectives et #Informa- tions Internationales. Dean, J., and J. Wainio. 2006. "Quantifying the Value of US Tariff Preferences." Washington, D.C.: United State International Trade Commission. Low, P., R. Piermartini,and J. Richtering. 2005. "MultilateralSolutionsto the Erosion of Non-Reciprocal Preferences in NAMA." Working Paper ERSD-2005-06. Geneva: World Trade Organization. [www.wto.org/english/res-elreset-elwpaps-e.htm]. -. 2006. "Non-Reciprocal Preference Erosion Arising from MFN Liberalization in Agriculture: What Are the Risks?"Working Paper ERSD-2006-02. Geneva: World Trade Organization. Olarreaga,M., and C. Ozden. 2005. "AGOA and Apparel: Who Capturesthe Tariff Rent in the Presence of Preferential Market Access?"The World Economy 28(1):63-77. Prowse, S. 2006. " 'Aid for Trade': Increasing Support for Trade Adjustment and Integration-A Proposal." In S. Evenett, and B. Hoekman, eds., Economic Development and Multilateral Trade Cooperation. Basingstoke,UK: Palgrave Macmillan. Doha Merchandise Trade Reform: What Is at Stake for Developing Countries? Kym Anderson, Will Martin, and Dominique van der Mensbrugghe The LINKAGE model of the global economy and the latest Global Trade Analysis Project (GTAP) database (version 6.05) are used to examine the impact of current merchandilse trade barriers and agricultural subsidies and possible reform outcomes of the Wor'ld Trade Organization's (wo's) Doha Development Agenda. The results suggest that moving to free global merchandise trade would boost real incomes in Sub-Saharan Africa proportionately more than in other developing countries or in high-income countries, despite the terms of trade loss in parts of that region. Particular attention is given to agriculture, as farmers constitute the poorest households in developing coun- tries but the most assisted in rich countries. Net farm incomes would rise substantiallly in Sub-Saharan Africa and other developing country regions, alleviating rural poverpy. Partial liberalization could move the world some way toward those desirable outcomes, the more so the more developingcountries themselvescut applied tariffs, particularly on agricultural imports. This article (a)summarizes the costs of current merchandise trade distortions to developingand other economies; (b)examinessome scenarios that might emerge as part of an eventual Doha agreement consistent with the 2005 Hong Kong Ministerial Declaration [World Trade Organization ( w o ) 20051, particu1,arly with respect to agriculture; and (c)draws implications for the strategies devel- oping countries might adopt in the me's Doha Round of multilateral trade negotiations. This article estimates what the world economy might look like in 2015 with- out and with a successful conclusion to the Doha Round, how far Doha could take the world toward an outcome with no distortions in merchandise trade, ,and what contribution various elementsof a Doha package could make. The analysis relies on a recursivemodel of the global economy known as LINKAGE (fordetails, Kym Anderson, Will Martin, and Dominique van der Mensbrugghe are Lead Economistsin the Devel- opment Economics Vice Presidency of the World Bank. Their email addresses are kanderson@worldban- k.org, wrnartinl@worldbank.org, and dvandermensbrugg@worldbank.org.The authors are grateful for helpful comments from seminar participants and journal referees, for tariff-cutting data from the staff of Centre dYEtudesProspectives et d'Infonnations Internationales ( a p n ) in Paris (with special thanks to David Laborde) and for funding from the U.K. Department for International Development. THE WORLDBANK ECONOMIC REVIEW, VOL. 20, NO 2, pp. 169-195 . doi:10.1093/wber/lhji009 Advance Access publication May 9, 2006 0The Author 2006. Published by Oxford UniversityPress on behalf of the International Bankfor Reconsmction and DevelopmentITHE WORLD BANK. All rightsreserved. For permissions, please e-mail: journals.pennissions@oxfordjoumals.org. see supplemental appendix S.2 posted at http://wber.oxfordjournals.org), which has formed the basis of the World Bank's standard decade-long projections of the global economy and its earlier trade analysis (see,for example, World Bank 2002, 2004). It also uses the latest version (6.05)of the Global Trade Analysis Project (GTAP) database, which includes the tariff preferences enjoyed by many developingcountries (http://www.gtap.org). The results distinguish between the effects on developing countries and those on more advanced economies. In doing so, it is necessary to consider both the World Bank's classification of economies by income level and the WTO practice of self-nominated developing country status, under which even economies as advanced as Hong Kong (China), Singapore, the Republic of Korea, and Taiwan (China)claim developing country status and so are eligible for special and differential treatment, including lesser tariff cuts and longer phase-in periods than are eventually agreed for developed countries following the Doha Round. The analysis suggests that most of the potential gains from multilateral reform would come from agriculture. But because of the large gaps between WTO'S bound and applied rates of protection, there would be little real agricultural reform globally as a result of the Doha Round-especiallyby developing countries-unless WTO members make substantial cuts to their bound tariff rates and domestic farm subsidy commitments. This article explores the effects of a more ambitious agricultural reform package over the next decade and of developing countries participating more fully in the Doha Round rather than invoking special and differential treatment to avoid reform. If WTO members insist on classifying even a small number of farm products as "sensitive" and subject to lesser tariff cuts, the gains from agricultural reform could be greatly diminished-and even disappear for developing countries. This article begins with an overview of the key elements of a prospective Doha agreement consistent with the Hong Kong Ministerial Declaration (WTO 2005) and focusing on the agricultural elements. It describes the model of the global economy used to analyze the consequences of such an agreement and of alternative, more-ambitious reforms including a move to complete free trade (whichprovides a helpful benchmark). The estimates of protection and subsidy rates for each region are a crucial part of the data in the global model, and so they are examined before turning to the key results of the simulations. After discussing some qualifications, this article draws out some implications for developing countries. To what extent are trade and subsidy reform commitments likely to emerge from the Doha Round? In addressing that question, it is important to know that WTO trade negotiators are seeking agreement on reductions not to applied tariffs and Anderson, Martin, and van der Mensbrugghe 171 subsidies but rather to members' legally bound import tariffs, agricultural export subsidies, and commitments on domestic support to farmers. These bound r;ates are higher than applied rates in nearly all countries, but especially in most developing countries, meaning that cuts in bound rates will have a wea~ker impact on market access. The Doha Round was launched at the wro Ministerial Meeting in Doha in late 2001, but the following Ministerial Meeting, in Cancun in September 2003, ended in acrimony and without agreement on how to proceed. At Cancun, developing countries made it abundantly clear that further progress would not be possible without a commitment by developedcountries to significalitlylower their agricultural subsidies (including, importantly, for cotton, despite its rela- tively minor role in developed country agriculture; see Sumner 2006). The so- called July Framework Agreement (mo2004) and the Hong Kong Ministerial Declaration (wro 2005) reiterate the importance of keeping development at the heart of the Doha agenda and stress agricultural reform as a key to doing that. Annexes to these documents provide guidance on how a Doha agreement might be structured, with frameworks for establishing modalities for agriculture and nonagricultural market access,.as well as providing recommendations for trade in services. The following sections highlight the key elements of a prospective Doha agreement focusing on agriculture and the state of negotiations to date. AgriculturalMarket Access The gap between bound and applied tariffs is the so-called bindingoverhang and can significantly blunt the impact of any negotiated outcome-so much so that in some Doha scenarios, some countries are not required to change their applied tariffs at all. Jean, Laborde, and Martin (2006)examined the consequences for applied tariff cuts of different bound tariff-cutting formulas, taking into account agricultural tariff rate quotas, the prevalence of preferences for developing countries (as described in Bouet, FontagnC, and Jean 2006), the need to accom- modate "sensitive"and "special"farm products, and the special and differenrial treatment outlined in the July Framework. Tariff cutting, implemented at the six-digit level of the Harmonized System (HS) of commodity disaggregation, involves a detailed comparison of each country's bound tariff, which is what negotiations focus on, with the applied most-favored nation tariff on a given bilateral trade flow, which is what affectseconomic outcomes. The applied tariff cuts vary not only by sector but also by trading partner-and may involve smaller or no cuts on imports from developing countries currently enjoy~ng nonreciprocal preferential access to richer countries' markets (Hoekman and Ozden 2005). Following the detailed tariff analysis, the results were aggregated up to the GTAP and INKA E models' regional and sectoral levels. L G Jean, Laborde, and Martin (2006) evaluated the consequences for 2001 applied rates of different approaches to liberalization, particularly different degrees of top-down progressivity in the bound tariff cuts, as well as different degrees to which developing countries participate in reform. They looked first at a proposal similar to the Harbinson progressive reduction formula (WTO 2003b), with marginal tariff rate reductions of 35 percentfor tariffs below 15 percent, 65 percent for tariffs above 90 percent, and 60 percent for tariffs within the 15-90 percent bracket.' Developing countries' tariff cuts also follow the progressive formula but with four rather than three brackets and with inflexion points at tariff levels of 20, 60, and 120 percent to be consistent with Harbinson's criterion of cutting by an average of 25, 30, 40, and 45 percent in those four brackets. That set of tariff cuts leads to very little import liberalization, because bound tariffs in many countries exceed applied rates by such large margins. As a result, Jean, Laborde, and Martin focused on another set of reforms that involve cuts in applied agricultural protection rates that are at least 10 percentage points greater: a 45, 70, and 75 percent cutting rule for bound rates for developed countries and a 35,40,50, and 60 percent cutting rule for developing ~ountries.~ These cuts are within the (wide)range proposed by key WTO membersin the lead- up to the Ministerial Meeting. Jean, Laborde, and Martin then examine the consequences of: Allowing lesser tariff cuts for self-nominated "sensitive" farm products, assuming that countries would take into account the importance of the commodity, the height of the tariff, and the gap between the tariff binding and the applied rate in deciding as to which products to grant such treat- ment, comparingsituations in which countries are allowed to treat 2 percent of agriculturaland food tariff lines as sensitiveand subject to just a 15 percent tariff cut; Including ccspecial"agricultural products just for developing countries, by adding another 2 percent of agricultural tariff lines as subject to just a 15 percent tariff cut; Adding a tariff cap of 200 percent, consistent with the suggestion in para- graph 30 of the July Framework Agreement that the role of a tariff cap be explored; These scenarios are also modeled in the analysis here. Agricultural Domestic Support Reductions in domestic support have been a particular concern of developing countries. Developed countries are the major providers of such assistance, and 1. This approach provides cuts in average tariffs-without the discontinuities created by the propor- tional cuts involved in the Harbinson formula-thatare more or less comparable with those generated by Harbinson's proportional reductions of 25, 30, and 60 percent, because the larger cuts on higher tariffs apply only on the portion of the tariff above 15 or 90 percent. 2. With no cuts in least developed countries, as specified in the July Framework Agreement and the Ministerial Declaration. Anderson, Martin, and van der Mensbrugghe 173 many developingcountries are concerned about the ability of their producers to compete with developed country farmers receiving large amounts of domestic support from their governments. While the marked asymmetry between devel- oped and developingcountries is a concern, there is evidence that the benefitsto developing countries from reductions in developed country domestic support may be substantially smaller than the potential gains from reductions in market access barriers (Hoekman, Ng, and Olarreaga 2004; Hertel and Keeney 2006; Anderson and Valenzuela forthcoming). Nonetheless, disciplining such support is crucial not only to prevent policy reversals but also to ensure that when tariffs are lowered, import protection is not simply replaced by equally or more distorting domestic measures. The Framework Agreement and the Ministerial Declaration propose tiered reductions in the total bound aggregate measure of support, with larger reduc- tions by members with higher initial levels of support. It turns out that extra- ordinarily large reductions in bound levels of support are required before any reductions in actual support would occur. If all countries with aggregate mea- sure of support notifications above 20 percent of the value of production cut their bound protection by 75 percent and all others by 60 percent, only four members would have to cut applied rates as of 2001: the United States by 28 percent, Norway by 18 percent, the European Union (EU) (thepre-expansion EU- 15)by 16 percent, and Australia by 10 percent. Agricultural Export Subsidies Export subsidies for nonfarm goods are outlawed in the WTO, so eliminating farm export subsidies would simply be bringing agriculture into line with other goods. The empirical analyses summarized in the works of Hertel and Keeriey (2006)and Anderson and Valenzuela (forthcoming)show that export subsidlies contribute only a small part of the welfarecost of agricultural support programs. That is true even when implicit subsidies in the form of food aid and export credits are included. A phase-out by 2013 of both explicit and implicit forms of farm export subsidies, as agreed at the Hong Kong Ministerial Meeting, shoilld therefore be a politically feasible component of a comprehensive Doha agree- ment. Their elimination in isolation could harm a few food-importing and aid- dependent developing countries, but the poor net buyers of food in those countries can be assisted through other more direct and more cost-effective forms of aid than through these measures. Nonagricultural Market Access Negotiations on nonagricultural tariffs have been lagging behind those on farm products. There has been a clear indication that developing countries wish to make smaller tariff cuts compared with developed countries and that the least developed countries expect not to have to make any cuts. A Doha Round is unlikely to involve cutting all nonagricultural bound tariffs by more than 50 percent, so the assessment assumes a 50 percent cut by developed countries, a 33 percent cut by developing countries, and no cut by least developed countries. However, becausecuts in bound rates may lead to very little reduction in applied rates by developing countries, a more ambitious scenario is also explored, with developing countries committing to more reform and in return seeking recipro- city in the form of further cuts in developed countries' agricultural and textiles tariffs. The most optimistic possibility considered is that developing countries (including least developed countries) agree to cut nonagricultural bound tariffs as much as developed countries (that is, by the 50 percent assumed for the analysis). Services Trade wro members have been very slow in coming forward with Doha proposals to reform services trade. At this stage, it seems likely that, as with the Uruguay Round, countries will make few meaningfulcommitments to genuinely open up their services sectors. For that reason, and because services trade is less ade- quately represented in trade models than is goods trade, reductions in this sector are not included in the analysis-despite the fact, as indicated by Hertel and Keeney (2006),that gains from services reform could well be enormous, includ- ing for developing countries. The analysis uses the LINKAGE model, a relativelystraightforward global compu- table general equilibrium (CGE) model but with some characteristics that distin- guish it from standard comparative static models such as the GTAP model [described by Hertel (1997)l. A key difference is that it is recursive, so that although it starts with 2001 as its base year, it can be solved annually through to 2015. This is important when evaluating a reform that is likely to take a decade or more to be fully implemented, because the structure of the world economy will be quite different in 2015 than it is in 2001. Economic expansion in the model is driven by exogenous population and labor supply growth, savings-dependent capital accumulation, and exogenous labor-augmenting technological progress (as used in the World Bank's Global Economic Prospects 2004 and as detailed in supple- mental appendix 52).In any given year, factor stocks are fixed. Producers minimize costs subject to constant returns to scale production technology, consumers maximize utility, and all markets-including for unskilled and skilled labor, which are both intersectorally mobile-are cleared with flexible prices. Also consistent with the focus on long-run adjustment to reform, the aggre- gate supply of farmland is defined by an overall upward slopingsupply function, Anderson, Martin, and van der Mensbrugghe 175 with land-abundant countries having a higher land supply ela~ticity.~Land is allocated across agricultural activities using a constant elasticity of transforma- tion function. There are three types of production structures: (a) crop sectors reflect the substitution possibility between extensive and intensive farming; (b) livestocksectors reflect the substitution possibility betweenpasture and intensive feeding; and (c)all other sectors reflect standard capitalllabor substitution. There is a single representative household per modeled region, allocating income to consumption using the extended linear expenditure system. Trade is modeled using a nested Armington structure in which aggregate import demand for each sector's product is the outcome of allocating domestic absorption between domestic goods and aggregate imports, and aggregate import demand is allocated across source countries to determine bilateral trade flows. There are various sources of protection in the model. The most important involves bilateral import tariffs. There are also bilateral export subsidlies. Domestically, there are subsidies only in agriculture, applied to intermediate goods, outputs, and payments to capital and land. Household consumption and savings are represented by the extenlded linear expenditure system, which provides a rigorous framework for mod- eling consumption and savings decisions and the allocation of consumption spending across commodities (Lluch 1973). Government fiscal balances are fixed in any given year, with government spending fixed as a share of GDP and the fiscal objective being met by changing the level of lump-sum taxes on household^.^ This implies that losses of tariff revenues are replaced by higher direct taxes on households. The current account balance is fixed, primarily for convenience in this recursive modelS but also consistent with the Feldstein-Horioka finding of limited international capital mobility (Feldstein and Horioka 1980; Ventura 2003). Finally, investment is driven by savings. With fixed public and foreign savings, investment is determined by changes in the savings behavior of households and changes in the unit cost of investment. The model solves only for relative prices, with the numeraire, or price ancbor, being the export price index of manufactured exports from high-income countries. Version 6.0 of the LINKAGE model is based on release 6.05 of the GTAP database, which has a 2001 base year instead of the 1997 base year of GTAP version 5, updated national and trade data, and a new source for the protec- tion data (see http://www.gtap.org for details). The new protection data are 3. Key elasticitiesfor the LINKAGE model are in supplemental appendix S.l, available online at http:ll wber.oxfordjournals.org. 4. For simplicity, they are fixed in U.S. dollar terms at their base year level, minimizing potential sustainabilityproblems. But this implies that they decrease over time as a percentageof GDPfor expanding economies. 5. Only with fixed financial inflows from abroad can utility changes be used to provide a money- metric measure of welfare changesresulting from a reform. from a joint Centre d'Etudes Prospectives et dYInformationsInternationales (Paris)/InternationalTrade Centre (Geneva)project. The product of this joint effort, known as MAcMaps, is a detailed database on bilateral protection at the HS six-digit level that integrates trade preferences, specific and compound tariffs, and a partial evaluation of nontariff barriers such as tariff rate quo- t a ~The new GTAP database has lower tariffs than the previous database . ~ because of the inclusion of bilateral trade preferences, major reforms between 1997 and 2001 such as continued implementation of the Uruguay Round agreements, and China's WTO accession, which alone caused the ratio of global exports plus imports to GDP to rise from 44 to 46 percent over those four years. The LINKAGE model used for this study comprises a 27-region, 25-sector aggregation of the GTAP database. There is a heavy emphasis on agriculture and food, which account for 13 of the 25 sectors, and on the largest commodity exporters and importers. The main source of protection is tariffs or border barriers, although some countries-particularly high-income countries-also have significant agricul- tural production and export subsidies. The average import tariff for agriculture and food in 2001 was 16.0 percent for high-incomecountries and 17.7 percent for developing countries, while for manufactures other than textiles and cloth- ing, it was 8.3 percent for developing countries and just 1.3 percent for high- income countries. The averages of course obscure large variations across countries and com- modities. For example, if high-income countries' tariffs on temperate farm products are at a near-prohibitive 100 percent but zero on tropical products such as coffee, the import-weighted average agricultural tariff could be quite low. Even at a relatively aggregated level, the variations can be quite sharp. For example, India has an average tariff on agriculture and food of 82 percent on imports from East Asia but only 20 percent on imports from Sub-Saharan Africa. Also, high-income countries' agricultural tariffs are lower on goods from low-income countries than on goods from high- and middle-income countries, while imports of textiles and clothing from low-income countries face a higher average tariff than imports from middle- or high-income countries. 6. More information on the MAcMaps database is available in the study by Bouet and others (2004) and at http://www.cepii.fr/anglaisgraph/bdd/macmap.htm.For a detailed analysis of the differences between the results resented here and those obtained by using the LMKAGE model and the earlier GTAP version 5 database and those obtained by using the same version 6 GTM database but the GTM model, see the works of Anderson, Martin, and van der Mensbrugghe (2006, appendix 12A) and van der Mensbrugghe (2006). Anderson, Martin, and van der Mensbrugghe 177 IV. ESTIMATES O F THE WELFARE IMPACT O F CURRENT PROTECTIO~V POLICIES The LINKAGE model provides a baseline projection of the world economy first to 2005 and then to 2015 assuming no other policy changes. Deviations from that baselinein 2015, due to phased partial or total liberalization from 2005, are rhen examined. One benchmark against which to measure the prospective benefits of the Doha Round is the gains that would come from completely freeing merchandise trade (including removing all agricultural producer and export subsidies) over the 2005-10 period. That leads to global gains by 2015 of $287 billion7 a year. Another benchmark is the reform incorporated in the presimulation experiment for the period 2001-04, reflecting the final stages of Uruguay Round implemen- tation including the phase-out of the Multifibre Arrangement, the accessioin of China and Taiwan (China)to the WTO, and the enlargement of the EU from 15to 25 members.' The impacts of those reforms on import tariffs are nontrivial. Had those three reforms not already been implemented, the gains in 2015 firom freeing global merchandise trade would have been $341 billion nnsteacl of $287 billion, or an extra $54 billion a year. Nearly half that difference is due to the removal of export quotas with the phase-out of the Multifibre Arrange- ment and the follow-on Agreement on Textiles and Clothing and so should be considered part of the Uruguay Round's legacy-assuming safeguards by high- income countries or export restraints by China do not replace textile and cloth- ing quotas after 2005.~ The distribution of the standard economic welfare or real-income (equivalent variation) effects of removing all merchandise trade distortions (including agri- cultural subsidies)shows that two-thirds of the $287 billion gain in income that global reform would generate each year by 2015 would accrue to high-income countries (table1).However, as a share of national income, developing countries (as self-defined by WTO members) would do twice as well, with an average 7. A billion is one thousand million. 8. These are the key internationally agreed and bound policy changes. Unilateral and unbound policy changes, such as recent reforms in EU and U.S. farm programs, are not included. 9. To get a sense of how important preferences are to developing countries and global welfare, the model was re-run for 2001, before the pre-simulation experiment and without those preferences in place. The estimated global welfare gains from reform are then $382 billion instead of $341 billion, and the developing country gains are $150 billion instead of $113 billion. That is, the inclusion of preferences in the database reduces estimated global welfare gains by 11 percent, developing country gains by 25 percent, and high-income country gains by 2 percent. Much of the difference is attributable to Sub- Saharan Africa, where the reduction is almost 50 percent. The reductions for developing countries are overstated, however, for two reasons. One is that no rules of origin or other impediments are assumed for developing countries fully utilizing their preferences. The second is that importers in the preference- providing rich countries are assumed not to use their power to gain a disproportionate share of the rent from that preferential access. In practice, neither of these assumptions holds, according to recent case studies (Ozdenand Sharma 2004; Olarreaga and Ozden 2005). TABLE 1. Impacts on Real Income from Removing All Global Merchandise Trade Distortions Including Agricultural Subsidies, without and with Own-Country Participation, by Country and Region, 2015 (Changes from Baseline) - - -- Annual income gain due to Total gain as share of Annual real income change in terms of trade baseline income gain (5billion) (5billion) (percent) From other From own plus From other From own plus From own plus countries' other countries' countries' other countries' other countries' Economy or region reforms reforms reforms reforms reforms - - Australia and New Zealand EU-25plus European Free Trade Association United States Canada r Japan 2 Korea, Rep. and Taiwan (China) Hong Kong (China)and Singapore Argentina Bangladesh Brazil China India Indonesia Thailand Vietnam Russian Federation Mexico South Africa Turkey (Continued) TABLE 1. Continued Annual income gain due to Total gain as share of Annual real income change in terms of trade baseline income gain ($ billion) ($ billion) (percent) From other From own plus From other From own plus From own plus countries' other countries' countries' other countries' other countries' Economy or region reforms reforms reforms reforms reforms Rest of South Asia Rest of East Asia Rest of Latin America and the Caribbean Rest of Europe and Central Asia Middle East and North Africa Selected Sub-Saharan African countriesa Rest of Sub-Saharan Africa Rest of the World High-income countries wro developingcountries Developingcountries (World Bank definition) Middle-incomecountries Low-incomecountries East Asia and Pacific South Asia Europe and Central Asia Middle East and North Africa Sub-Saharan Africa Latin America and the Caribbean World total 'Countries for which national modules are available in LINKAGE: Botswana, Madagascar, Malawi, Mozambique, Tanzania, Uganda, Zambia, and Zimbabwe. Source: Authors' World Bank LmKAGE model simulations. increase of 1.2 percent of national income over the baseline compared with 0.6 percent for high-income countries. The results vary widely, ranging from little impact for Bangladesh and Mexico to 4-5 percent increasesin parts of East Asia. As a percentage of national income, Sub-Saharan Africa (excluding South Africa) would gain twice as much as high-income countries, despite the adverse change in the terms of trade for many Africancountries due in part to the loss of their nonreciprocal tariff preferences.'' Policymakersare frequently interested in the extent to which gains come from their own liberalization compared with the gains from liberalization by their trading partners. This was estimated by solving the model once for each region, considering only liberalization by that region. Column 2 of table 1 summarizes the total gains from multilateral liberalization, while column 1 summarizes the gains that would result without liberalization by the region but from improved access to partners' markets, which Bagwell and Staiger (2002) suggest is the primary motivation for engaging in multilateral trade reforms. The analysis reveals that for regions with high agricultural protection (Western Europe, Northeast Asia, and Middle East), a large proportion of the gain comes from own-country reform. But for many more open economies, the bulk of the gain comes from increased market access in other countries. Even for Sub-Saharan Africa, 40 percent of the gain ($1.0 billion of $2.5 billion) would come from reforms in other regions, despite the much-lamented losses from preference erosion in industrial country markets. These results suggest that the benefits of a multilateral round could be substantially greater than the benefits from uni- lateral liberalization alone.'' Columns 3 and 4 in table 1 summarize the income effects of changes in the international terms of trade for each country. For developing countries as a group, the terms of trade effect is negative, somewhat reducing the gains from improved efficiency of domestic resource use (especially in China and India). A comparison of columns 3 and 4 reveals that it is mainly own-country reforms that are lowering developing countries' terms of trade. Under the Armington assumption used in this model, the terms of trade losses are overestimated 10. The gains would be even greater if African reforms were accompanied by complementary domestic policy reforms and investments in trade-facilitating infrastructure and institutions (funds for which may be forthcoming in the proposed aid for trade package that may accompany a Doha agreement; see Nielson 2006).For more detailed disaggregation of the results for Sub-Saharan Africa, see the study by Anderson, Martin, and van der Mensbmgghe (forthcoming-c). 11. This result depends heavily on the size of the models' Armington elasticities. Those in the LINKAGE model are about one-third larger than in the standard GTAP model on average, reflecting the focus on a longer (decade-long)adjustment period. The Armington elasticities would have to be even larger than those in theLWKAGE model to get more of a W Y D ~ G(what you do is what you get) result. There is a fundamental problem with the Armington approach, which assumes that countries export more of the same products following liberalization. Kehoe and Ruhl (2003) show that much of the expansion following liberalization is typically in new products, a response that reduces the adverse terms of trade impacts of export growth. Estimates of response elasticities that take adequate account of this phenom- enon are not yet available. Anderson, Martin, and van der Mensbrugghe 181 because no allowance is made for expansion in the number or quality of exported goods resulting from reform (Hummels and Klenow 2005). There are several other ways to decompose the real-income gains from full global trade reform so as to better understand the sources of the gains for each region. One way is to assess the impacts of developingcountry liberalization and those of industrial country liberalization in different economic sectors. Anotlier is to decompose by policy instrument. For agricultural reform, decomposing by policy instrument gives results very similar to those from the GTAP-AGR model reported by Hertel and Keeney (2006),who estimate that market access barriers explain 93 percent of the welfare effects of agricultural policies, domestic sup- port just 5 percent, and export subsidies just 2 percent.12 When decomposed by sector, the results suggest that global liberalization of agriculture and food contributes 63 percent of the global gains, which is similar to Hertel and Keeney's 66 percent (table 2). This is consistent with the higher tariffs in agriculture and food (17 percent global average) than in other sectors but is nonetheless remarkable given the low shares of agriculture in global GDP (4 percent) and global merchandise trade (9 percent). The share of gains from agriculture is even higher for Sub-Saharan Africa, at more than three-quarters of the total welfare gain. Seventy percent of the global gains from agriculture are accounted for by the farm policies of high-income countries, and those policies also account for the bulk of the overall gains to high-income countries. For developingcountries, as much of their gain from farm reform would come from agricultural liberal- ization in other developing countries as from getting unrestricted access to markets in high-income countries. The results are nearly the same for manufac- turing in aggregate, despite the large gains from reforms in textiles and clothing markets in high-income countries ($14 billion compared with $9 billion from growth in textile trade among developing countries).Thus, reform by developing countries is as important to the economic welfare gains of developing countries as reform by high-income countries. Notice also that developing country gains from high-income country reform are only half as large from textiles as from agricultural policies. What impact would the removal of cotton trade distortions and subsidies (which raise producer prices by more than 50 percent in the United States and even more in the EU) have in this context of freeing all merchandise trade and agricultural subsidies? The global price of cotton would rise an estimated 21 percent above the 2015 baseline on average because U.S. subsidies would no longer depress prices. However, the volume of U.S. cotton exports would shrink when the subsidies were removed, raising the price and volume of other 12. Hoekman, Ng, and Olarreaga (2004) reach a similar conclusion from estimating the effects; of halving each of the three types of agricultural distortions, in their case using partial equilibrium analysis. For an intuitive, nontechnical explanation of this result-which has surprised many observers-see the study by Anderson, Martin, and Valenzuela (2006). TAB LE 2. Sources of Regional and Sectoral Gains in Real Income from Full Liberalization of Global Merchandise Trade, Developing and High-Income Countries, 2015 (Changes from Baseline) Gains by region ($ billions) Share of regional gain (percent) All developing All high-income World All developing Middle-income Sub-Saharan Africa All high-income World Developing countries liberalize Agriculture and food Textiles and clothing Other merchandise r All sectors m NHigh-income countries liberalize Agriculture and food Textiles and clothing Other merchandise All sectors All countries liberalize Agriculture and food Textiles and clothing Other merchandise All sectors Note: Small interaction effects are distributed proportionately, and numbers are rounded to sum to 100 percent. Source: Authors' World Bank LINKAGE model simulations. Anderson, Martin, and van der Mensbrugghs 183 countries' exports. In particular, cotton output from Sub-Saharan Africa woiuld be 44 percent larger and cotton exports 73 percent larger under this full liberal- ization scenario, with the value of increases in both output and exports being greater than for any other region including Latin America and Australia (wh'ere there are more other agricultural expansion opportunities than in Africa, which would experience preference erosion). Indeed, cotton is so important in Sub- Saharan Africa excluding South Africa that it contributes one-quarter of Ithe region's net gain in agricultural value added from full global trade and subsidy liberalization. The share of all developing countries in global cotton exports would be 85 percent instead of 56 percent in 2015, further validating the efforts to ensure that cotton receives specific and substantial attention in the Doha negotiations (Baffes2005; Sumner 2006). All these results are for full trade liberalization. Smaller changes can be expected from partial reforms of the sort currently being negotiated under the Doha Development Agenda. These are explored in the next section. What will the Doha package ultimately contain? Agricultural export subsidies are assumed to be eliminated by 2013 and domestic support for agriculture is assumed to be cut relative to 2001 levels in just four economies: by an average of 28 percent by the United States,18 percent by Norway, 16 percent by the EU, and 10 percent by Australia. More difficult to determine are the likely nature and extent of reductions in market access barriers, so a number of scenarios are initially considered for agricultural and food products in isolation from nona- gricultural tariff cuts, before some nonagricultural market access reforms are also incorporated.13Throughout this section, the wro use of the term "develop- ing countries" applies, which means that Hong Kong (China),the Republic of Korea, Singapore, and Taiwan (China) are all able to enjoy reduced reform commitments under special and differential treatment despite their high-income status. Scenario 1 begins with a progressive or tiered reduction formula of marginal agricultural tariff rate reductions of 45, 70, and 75 percent for developed countries within each of the three bands defined by the Harbinson ( m o 2003b) inflection points of tariff rates of 15 and 90 percent, and reductions of 35, 40, 50, and 60 percent for developing countries within each of their four bands (least developed countries are not required to undertake any reduction commitments). Even with these large cuts to bound tariffs (whichare about h.alf way between those proposed by the United States and the EU in late 2005 in the lead-up to the Hong Kong Ministerial Meeting), average applied tariffs on agricultural and food products in 2015 would be only about one-third lower 13. As suggested in the Girard text (seewro 2003a), a good without a boundtariff is treatedas having double the applied most-favored nation rate. globally, at 10.0 percent instead of 15.2 percent, and 12.5 percent instead of 14.2 percent for developing countries. Scenario 2 examines the consequences of including "sensitive"farm products as allowed for in the July Framework Agreement. Developed countries are allowed to treat up to 2 percent of their agricultural tariff lines at the HS six- digit level as sensitive and subject to just a 15 percent tariff cut,14 and both developing and least developed countries are allowed to treat up to 4 percent of agricultural tariff lines as sensitive, in part to incorporate their demands for "specialn treatment of some products.1s Under this scenario, the average agri- cultural tariff falls to only 13.5 percent in both high-income and developing countries. Scenario 3 considers the effects of adding to scenario 2 a tariff cap of 200 percent. Any product with a bound tariff above that limit will be subject to a reduction to that rate, which leads to average cuts in agricultural tariffs of 18 percent for both developed and developing countries. The average agricultural tariff in 2015 would fall considerably more for high-income countries (to11.5 percent) and only slightly more (to13.3 percent) for developing countries. Scenario 4 adds to scenario 1 the cuts in nonagricultural tariff bindings of 50 percent in developed countries, 33 percent in developing countries, and zero in least developed countries. That lowers the average tariff on all merchandisefrom 2.9 percent in the baseline to 1.6 percent for high-incomecountries and from 8.4 percent to 7.5 percent for developing countries. Finally, scenario 5 makes developing (includingleast developed)countries full participants in the round, undertaking the same reductions in bound (but not necessarily applied) tariffs as the developed countries in scenario 4. That lowers the average tariff on all merchandise for developing countries from 8.4 to 6.8 percent instead of to 7.5 percent, a cut of almost one-fifth instead of one-ninth, as in scenario 4. Estimated Welfare and Trade Effects of the Scenarios as of 2015 The welfare consequences of implementing these reforms over the 2005-10 period and allowing the global economy to adjust to 2015 are summarized in table 3 in dollar terms and as percentage changes in real income in 2015. Agricultural liberalization using the harmonizing formula (scenario1)would generate a global gain of $75 billion even without the inclusion of nonagricul- tural tariff reform. But almost all those benefits accrue to the reforming high- income countries (amongwhich are included high-protection Korea and Taiwan (China) as well as Hong Kong (China) and Singapore in this and subsequent tables). Developing countries would gain only $9 billion because t6eir tariff 14. Some proposals involve larger cuts in tariffs on these goods. 15. As described by Jean, Laborde, and Martin (2006),"sensitive"farm products are chosen for each country by taking into account the importance of the product, the size of its existing tariff, and the gap between its bound and applied tariffs. TAB LE 3. Change in Real Income in Alternative Doha Round Scenarios, 2015 (Changes from Baseline) Billions of dollars Percent Economy or region Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Australia and New Zealand EU-25plus European Free Trade Association United States Canada Japan Korea, Rep. and Taiwan (China) Hong Kong (China) and Singapore Argentina g Bangladesh Brazil China India Indonesia Thailand Vietnam Russian Federation Mexico South Africa Turkey Rest of South Asia Rest of East Asia Rest of Latin America and the Caribbean (Continued) TAB LE 3. Continued Billions of dollars Percent Economy or region Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Rest of Europe and Central Asia Middle East and North Africa Selected Sub-Saharan African countries Rest of Sub-Saharan Africa Rest of the World High-income countries r m o developingcountries 00 a Developingcountries (World Bank definition) Middle-income countries Low-income countries East Asia and Pacific South Asia Europe and Central Asia Middle East and Noah Africa Sub-Saharan Africa Latin America and the Caribbean World total Source: Authors' World Bank LINKAGE model simulations. Anderson, Martin, and van der Mensbrugghe 187 binding overhang is so great as to lead to almost no cuts in applied tariffs. Were countries allowed to have smaller cuts for even just 2 percent of farm produ~cts declared to be "sensitive" (and another 2 percent in developing countries for "special" farm products), global gains would shrink to just $18 billion and developing countries as a group would be even worse off (scenario 2). If such exceptions are made, it would be important to exploit the opportunity-pro- vided for in the Ministerial Declaration-to cap bound tariffs. Scenario 3 sho~ws that even a cap as high as 200 percent would restore at least half the welfare gain forgone by allowing such exceptional treatment for sensitive and special farm products. The final two scenarios add nonagricultural tariff cuts to the agricultural reforms in the first two scenarios. In scenario 4, smaller cuts are provided for developing countries' nonagricultural tariffs, as is the case for all agri- cultural tariff cuts in the preceding scenarios. Even so, relative to scenario 1, where only agricultural tariffs are cut, the gains to developing countries double by adding these nonfarm reforms, contributing one-third of the extra boost to global welfare ($7.1 billion of the $21.6 billion difference between the global gains from scenarios 1 and 4). In scenario 5, the devel- oping (including least developed) countries fully engage in the reform pro- cess, forgoing the lesser cuts provided for in scenarios 1-4. Because this leads to considerably larger cuts in applied tariffs, that boosts their welfare substantially, and global welfare as well. Nonetheless, the global average merchandise tariff hardly changes with agricultural reform alone, whereas it falls by almost one-third or 1.5 percentage points when manufacturing is included in the reform package. Retaininglesser tariff cuts for developingcountriesas in scenario4 would yield a global gain of $96 billion from merchandise liberalization, which is a sizable one-third of what is in the table (thepotential welfare gain from full liberalization of $287 billion, reported in table1).But for developingcountries, the gain would be only $16 billion, less than a fifth of that group's potential gain from full liberalization ($86 billion; table 1).If developing countries forgo the option of reformingless than developed countries, their gain would rise by 42 percent, or an extra $7 billion (scenario 5). Much of those gains go to the largest developing economies, but in percentage terms Sub-Saharan Africa also gains substantially if it liberalizes more-contrary to the presumptions of many. By contrast, in sce- nario 4, the rest of Sub-Saharan Africa is not liberalizing enough to achieve sufficient efficiency gains to offset the terms of trade losses suffered as net foiod importers, as recipients of tariff preferences that have eroded with the decline in high-income countries' most-favorednation tariffs or as a result of the combined export growth of reforming economies with similar export compositions.16 16. Details of the results for Sub-SaharanAfrica can be found in the study by Anderson,Mamn, and van der Mensbrugghe (forthcoming-b). 1 8 8 THE WORLD BANK ECONOMIC REVIEW, VOL. 20, N O . 2 What would be the consequences for farm output and employment growth over the Doha implementation period of partial reform? If there were complete free trade, farm output would decline (instead of growing slightly)in just the EU and Japan, grow slower in a few other high-protection countries, and expand in most countries and regions. Doha scenario 4 would involve much less reform than a move to free trade, and hence a much slower loss of farm output for the EU and Japan but also less output growth than under free trade for the vast majority of countries where farm output would grow. For most protected economies, Doha scenario 4 would simply slow the growth of farm output slightly over the coming decade. The farm employment picture is somewhat different. Typically, economic growth leads to declines not only in the relative importance of agriculture but also in the absolute numbers employed in farming once a country reaches middle-income status. Thus, it is not surprising that numerous middle- and high-income countries are projected to lose farm jobs over the next decade in the baseline scenario. For the most protected farm sectors, the rate of farm employment decline would more than double if the world were to move to complete free trade, but it would increase only slightly under Doha scenario 4. For most developing economies, though, farm employment would grow a little faster under scenario 4 than under the baseline (Anderson, Martin, and van der Mensbrugghe 2006, table 12.17).17 How is this reflected in agricultural net income (value added by the farming sector)? Not surprisingly, agricultural value added would fall in regions with the highest agricultural protection-Europe, Northeast Asia, and to a lesser extent the United States (table 4). However, in the Doha reform scenario, none of the developing countries or regions would suffer a decline in agricultural net income, despite the lowering of their own agricul- tural tariffs even though the average agricultural tariff in developing countries is nearly as high as that in high-income countries (14.2 percent compared with 15.9 percent in the baseline). That is because a much larger proportion of developing country agriculture produces exportables, which do not have to be protected from imports. Because as much as 70 percent of the world's poor are in farm households in developing countries, this result has clear implications for poverty alleviation. The trade consequences of Doha scenario 4 are considerable as well. By 2015, annual exports by developing countries would be $41 billion greater for agricultural products, $25 billion for textiles and clothing, and $12 billion for other manufactures. The total increase of $78 billion is somewhat smaller than that for high-income countries ($135 billion), but the difference is less in percentage terms (a 2.6 percent increase for developing countries compared 17. This finding of only small intersectoral labor movements in response to partial trade reform is consistent with econometric evidence of adjustments to past trade reforms (see, for example, Wacziarg and Wallack 2004). Anderson, Martin, and van der Mensbrugghe 189 TABLE 4. Impact of Reform Scenarios on Agricultural Value Added, 2015 (Changes from Baseline) Billions of dollars Percent Full global Full global Economy or region liberalization Scenario 4 liberalization Scenario 4 Australia and New Zealand EU-25 plus European Free Trade Association United States Canada Japan Korea, Rep. and Taiwan (China) Hong Kong (China)and Singapore Argentina Bangladesh Brazil China India Indonesia Thailand Vietnam Russian Federation Mexico South Africa Turkey Rest of South Asia Rest of East Asia Rest of Latin America and the Caribbean Rest of Europe and Central Asia Middle East and North Africa Selected Sub-Saharan African countries Rest of Sub-Saharan Africa Rest of the World High-income countries Developing countries (World Bank definition) Middle-income countries Low-income countries East Asia and Pacific South Asia Europe and Central Asia Middle East and North Africa Sub-Saharan Africa Latin America and the Caribbean World total Source: Authors' World Bank LINKAGE model simulations. with 3.1 percent for high-income countries). This takes global merchandise trade one-fifth of the way to where it would be under complete free trade in merchandise.l8 Finally, of interest to those concerned that poor consumers would face higher food bills are the changes that might be expected in average food prices in international markets under the various Doha scenarios (table 5). For agriculture as a whole, prices would rise less than 2 percent over the 10-year phase-in period. The changes could be as high as 12 percent for dairy products, 6 percent for cotton, and 3-4 percent for coarse grains, oilseeds, sugar, and meat but well under 2 percent for other farm products. Thus, the annual change in basic food prices at the retail level would be hardly discernable even to poor consumers, thanks to the supply responsiveness of farmers to the increases in market access opportunities when agricultural subsidies and tariffs are reduced. TABLE 5 . Impact of Doha Reform Scenarios on Average International Product Prices, 2015 (PercentageChange from Baseline) Product Scenario 1 Scenario 2 Scenario 3 Scenario 4 Scenario 5 Rice Wheat Other grains Oilseeds Sugar Cotton Fruit and vegetables Other crops Vegetable oils and fats Livestock Processed meats Dairy products Other food, beverages, and tobacco All agriculture and food All primary agriculture All processed agriculture Textile and wearing apparel Source: Authors' World Bank LINKAGE model simulations. 18. It also raises the share of agriculturaland food production that is exported globally from 9.5 to 10.0 percent, which is one-seventh of the way toward its share of 13.2 percent under the free trade scenario. Even in the protected countries, this ratio rises a littleor, in the case of the EU, falls only slightly. This is because farm resource move from currently protected import-competing subsectors to more competitivefarmingsubsectors (Anderson,Martin, and van der Mensbrugghe 2006, table 12.18). Anderson, Martin, and van der Mensbrugghe 191 Some Caveats Results such as those presented here are always dependent on the underlying . - assumptions, data, and parameters and so are subject to numerous qualifica- tions. One particularly important qualification concerns the way preferencesare treated in the version 6.05 GTAP database. Previous versions of the database included only key reciprocal preferences (notably between members within the EU, North American Free Trade Agreement, Association of Southeast As,ian Nations, and Australia-New Zealand regional integration arrangements). Ver- sion 6.05 includes nonreciprocal tariff preferencesprovided by developed coun- tries to imports from developing countries under arrangements such as the Generalized System of Preferences, the EU'S Africa, Caribbean, and Pacific pro- gram and Everything but Arms agreement and the U.S. Africa Growth and Opportunity Act and Caribbean Basin Initiative. The analysis assumes that - - there are no rules of origin or other compliance requirements that raise the cost of using these preferences. It also assumes that the full benefits of the preferencesflow to developingcountries (eventhough developedcountry impor- ters often have more market power than developing country exporters of stan- dard commodities, who receive a smaller share of the rents).19This treatment overstates the extent of preference erosion that would occur, especially for least developed countries, and so understates their gains from multilateral trade reform. If these nonreciprocal preferences were instead excluded from the da.ta- base, the preference-receiving countries' gains from developed country trade reform would be ~verestirnated.~' Another important issue is the extent to which the model captures the supply- side constraints to adjustment by low-income countries to international price changes. The elasticities are aimed at representing adjustment to long-term changes but are still small compared with those used by some other analysts (for example, Harrison and others 2004). Other models, including GTAP-A,GR (Hertel and Keeney 2006), use smaller trade elasticities and generate smaller gains globally and for developing countries, with some regions (including parts of Sub-Saharan Africa) even losing slightly. The uncertainty about the values of 19. Evidencethat the preference margin is often eroded by complex rules of origin and that the rent is shared between importing and exporting countries with the exporters getting less the more trade is concentrated on standard commodities can be found in the works of Olarreaga and &den (2005)and Ozden and Sharma (2004).A recent partial equilibrium study found that in practice export revenue losses from preference erosion are likely to be limited to a small subset of countries, primarily small island economies dependent on exports of sugar, bananas, and, to a far lesser extent, textiles and clothing (Alexandraki and Lankes 2004). 20. The extent of overstatement would not be large though, because the differencein the low-income countries' estimated benefits even from full liberalization is only $2 billion a year when nonreciprocal preferences are excluded from the LINKAGE model's database, or $8 billion when middle-income countries are also included (van der Mensbmgghe 2005).A further complication is that the nonreciprocal pre- ferenceschemewith the African, Caribbean, and Pacificcountries is scheduled to be replaced in 2008 with reciprocal Economic Partnership Agreements between regional groupings of those countries and the riu. these elasticities is a fundamental problem associated with pervasive measure- ment errors and uncertainty about the true model structure (Hummels and Klenow 2005). Also to be kept in mind is that global CGE modelsnecessarily have to aggregate across sectors, thereby reducing the large variance in tariffs that are evident at the HS six-digit or greater levels of disaggregation. Because the welfare cost of a tariff is roughly proportional to the square of its height, this aggregation leads to underestimation of that cost. The analysis here does not include costs of adjustment to reform, but the structural changes that take place over time in the normal course of eco- nomic growth are typically very much larger than the small changes that would accompany gradual and partial trade liberalization [as shown in the study by Anderson, Martin, and van der Mensbrugghe (forthcoming-a)].Furthermore, adjustment assistanceschemes (financed by foreign aid in the case of low-income countries) to help fund adjustment to tariff and subsidy cuts are just one-off payments, whereas the benefits of reform continue into the future. Nor has this analysis taken into account the fact that trade reform typically boosts factor productivity and that not all sectors are subject to constant returns to scale and perfect competition. Most models that allow increasing returns and imperfect competition in some sectors generate higher gains from trade reform, although there is the possibility of the opposite outcome if reform induces resources to move back into an agricultural sector that has sufficiently fewer economies of scale than the rest of the economy.21 VI. IMPLICATIONS FOR DEVELOPING COUNTRIES The good news is that there are still large potential gains from liberalizing merchandise trade under Doha, with global gains on the order of $95 billion- $120 billion a year from an agreement consistent with the Hong Kong Minister- ial Declaration, even if no reforms are forthcoming in services, and with a disproportionately high share of that potential gain available for developing countries (relative to their share of the global economy). Moreover, it is the poorest people in developing countries who appear most likely to gain from global trade liberalization-farmers and unskilled laborers.22 To realize that potential gain, by far the greatest cuts in bound tariffs and subsidies are required in agriculture. However, the political sensitivity of farm support programs, coupled with the complexities of the measures introduced in the Uruguay Round Agreement on Agriculture and the modalities set out in the 21. An example is the study by Francois, van Meijl, and van Tongeren (2005), whose 50 percent global reform scenario yields only a 0.5 percent global income gain despite economies of scale, imperfect competition, and variety effects, and with agriculture contributing less to those gains than in the results here because farming, unlike other sectors, is assumed to be subject to constant returns. 22. For detailed analyses of the poverty consequencesof these Doha scenarios, see the study by Hertel and Winters (2006). Anderson, Martin, and van der Mensbrugghe 193 Ministerial Declaration, ensures that achieving consensus on the details of the final Doha agreement will remain challenging. Outlawing agricultural export subsidies is the obvious first step. That will help bring agriculture into line vvith other sectors and limit the extent to which governments encourage agricultural production by other means (asit would remove one option for surplus disposal). Concurrently, domestic support bindings must be cut substantially to reduce binding overhang. Even more important, agricultural tariff bindings must be slashed so that genuine market opening can occur. Allowing lesser cuts for even a few "se~nsi- tive" and "special"farm products would greatly reduce the gains from reform, given the tariff peaks currently in place. If it turns out to be politically impos- sible not to designate some products as sensitive and special, the resulting welfare cost could be reduced by imposing a tariff cap at, say, 100 percent. Expanding market access for nonagricultural products at the same time as reforming agriculture would increase the prospects for a successful conclusion to the Doha Round. An essential part of the Doha development agenda is "concessions" between developing countries because that is where half their potential benefits lie. That means reconsideringthe extent to which developingcountries liberalize.Because developingcountries trade so much more with each other now than in the 198:0s, they are the major beneficiaries of reforms within their own regions. Even the least developed countries need to consider reducing at least their tariff binding overhang, becausedoing that in the context of the Doha Round gives them mlore scope to demand concessions (or compensation for preference erosion or otiher contributors to terms of trade deterioration) from richer countries than if they hang on to the opportunity not to engage in reform, as provided for in the July Framework Agreement. What ultimately emerges from the analysis is that developing countries would not have to reform very much following the Doha Round because of the large gaps between their tariff bindings and applied rates. But to realize more of the potential gains from trade, they would need to commit to addi- tional trade reforms and complementary domestic reforms and to invest more in trade facilitation. High-income countries could encourage them to do so riot only by opening up their markets to developing country exports but also by providing more targeted aid. A new proposal has been put forward to reward developing country commitments to greater trade reform with an expansion of trade-facilitating aid through a major expansion of the Integrated Framework, now operated by a consortium of international agencies for the least developed countries (Hoekman and Prowse 2005; Nielson 2006). This may provide an attractive path for developing countries seeking to trade their way out of poverty. In addition, it is potentially a far more efficient way for developed countries to assist people in low-income countries than the current systems of tariff preferences. 1 9 4 THE WORLD B A N K ECONOMIC REVIEW, VOL. 20, N O . 2 Alexandraki, K., and H. P. Lankes. 2004. "The Impact of Preference Erosion on Middle-Income Coun- tries." IMF Working Paper 041169. International Monetary Fund, Washington, D.C. Anderson, K., and E. Valenzuela. Forthcoming. "Do Global Trade Distortions Still Harm Developing Country Farmers?" Review of World Economics 143(1). Anderson, K., W. Martin, and E. Valenzuela. Forthcoming. "The Relative Importance of Global Agri- cultural Subsidies and Market Access." World Trade Review 5(3). Anderson, K., W. Martin, and D. van der Mensbrugghe. 2006. "Market and Welfare Implications of the Doha Reform Scenarios," In K. Anderson and W. Martin, eds., Agricultural Trade Reform and the Doha Development Agenda. New York: Palgrave Macmillan and the World Bank. -. Forthcoming-a. "Distortions to World Trade: Impacts on Agriculnual Markets and Incomes." Review of Agricultural Economics 28(2). -. Forthcoming-b. "Would Multilateral Trade Reform Benefit Sub-Saharan Africa?" Journal of African Economies 15. Baffes, J. 2005. "The 'Cotton Problem.' " World Bank Research Observer 20(1):109-43. Bagwell, K., and R. Staiger. 2002. The Economics of the World Trading System. Cambridge, Mass.: MIT Press. Bouet, A., Y. Decreux, L. Fontagnk, S. Jean, and D. Laborde. 2004. "A Consistent ad valorem Equivalent Measure of Applied Protection across the World: The MAcMap-HS6 Database." Centre &Etudes Prospectiveset d'Informations Internationales, Paris. Bouet, A., L. Fontagn6, and S. Jean. 2006. "Is Erosion of Preferencesa Serious Concern?"In K. Anderson and W. Martin, eds., Agricultural Trade Reform and the Doha Development Agenda. New York: Palgrave Macmillan (co-published with the World Bank). Feldstein, M., and C. Horioka. 1980. "Domestic Saving and International Capital Flows." Economic Journal 90(358):31429. Francois, J., H. van Meijl, and F. van Tongeren. 2005. "Trade Liberalization in the Doha Development Round." Economic Policy 20(42):349-91. Harrison, G. W., T. F. Rutherford, D. G. Tarr, and A. Gurgel.2004. "TradePolicy and Poverty Reduction in Brazil." World Bank Economic Review 18(3):289-317. Hertel, T., ed. 1997. Global Trade Analysis: Modeling and Applications. New York: Cambridge Uni- versity Press. Hertel, T. W., and R. Keeney. 2006. "What's at Stake: The Relative Importance of Import Barriers, Export Subsidies and Domestic Support." In K. Anderson and W. Martin, eds., Agricultural Trade Reform and the Doha Development Agenda. New York: Palgrave Macmillan and the World Bank. Hertel, T. W., and L. A. Winters, eds. 2006. Poverty and the wro: Impacts of the Doha Development Agenda. New York: Palgrave Macmillan and the World Bank. Hoekman, B., F. Ng, and M. Olarreaga. 2004. "Agricultural Tariffs versus Subsidies: What's More Important for Developing Countries?" World Bank Economic Review 18(2):175-204. Hoekman, B., and C. Ozden. 2005. "Trade Preferences and Differential Treatment of Developing Countries: ASelectiveSurvey."Policy Research WorkingPaper 3566. World Bank, Washington, D.C. Hoekrnan, B., and S. Prowse. 2005. "Policy Responses to Preference Erosion: From Trade as Aid to Aid for Trade." Paper presented at the World Bank Conference on Preference Erosion: Impacts and Potential Policy Responses,June 13-14, Geneva. Hummels, D., and P. Klenow. 2005. "The Variety and Quality of a Nation's Exports." American Economic Review 95(3):704-23. Jean, S., D. Laborde, and W. Martin. 2006. "Consequences of Alternative Formulas for Agricultural Tariff Cuts." In K. Anderson and W. Martin, eds., Agricultural Trade Refom and the Doha Devel- opment Agenda. New York: Palgrave Macmillan and the World Bank. Anderson, Martin, and van der Mensbrugghe 195 Kehoe, T., and K. Ruhl. 2003. "How Important Is the New Goods Margin in International Trade?" Federal Reserve Bank of Minneapolis Staff Report 324. Minneapolis, Minn. Lluch, C. 1973. "The Extended Linear Expenditure System." European Economic Review 4(1):21-32. Nielson,J. 2006. "Aid for Trade." In R. Newfarmer, ed., Trade, Doha and Development:A Window into the Issues. Washington, D.C.: World Bank. Olarreaga, M., and C. Ozden. 2005. and Apparel:Who Captures the Tariff Rent in the Presenceof Y~~~~ Preferential Market Access?"The World Economy 28(1):63-87. Ozden, C., and G. Sharma. 2004. "Price Effects of Preferential Market Access: The CBI and the Apparel Sector." Policy Research Working Paper 3244. World Bank, Washington, D.C. Sumner, D. A. 2006. "Reducing Cotton Subsidies: The DDA Cotton Initiative." In K. Anderson and W. Martin, eds., Agricultural Trade Reform and the Doha DevelopmentAgenda. New York: Palgrave Macmillan and the World Bank. van der Mensbrugghe, D. 2005. "The Doha DevelopmentAgenda and Preference Erosion: lvfodelingthe Impacts."Paper presented at the World Bank Conferenceon PreferenceErosion: Impacts and Potelntial Policy Responses, June 13-14, Geneva. 2006. "Estimatingthe Benefits:Why Numbers Change." In R. Newfarmer, ed., Trade, Doha and Development: A Window into the Issues. Washington, D.C.: World Bank. Ventura, J. 2003. "Towards a Theory of Current Accounts." The World Economy 26(4):483-512. Wacziarg, R., and J. S. Wallack. 2004. "Trade Liberalization and Intersectoral Labor Movements." Journal of International Economics 64(2):411-39. World Bank. 2002. Global Economic Prospectsand the DevelopingCountries 2002: Making Trade Work for the Poor. Washington, D.C. . 2004. Global Economic Prospects: Realizing the Development Promise of the Doha Agenda. Washington, D.C. WTO (World Trade Organization). 2003a. "Negotiating Group on Market Access: Report by the Chairman." TNIMAI12 (The Girard Text). Geneva. . 2003b. "Negotiations on Agriculture: First Draft of Modalities for the Further Commitmeras." TNIAGIWIlIRev.1 (The Harbinson Draft). Geneva. . 2004. "Decision Adopted by the General Council on 1 August 2004." WTLI579 (The July Framework Agreement).Geneva. .2005. "Doha Work Program: Draft Ministerial Declaration: Revision." Ministerial Confereilce, Sixth Session, December 13-18, Hong Kong, China. WTIMIN(O5)IWI3/Rev.2.Geneva. Preference Erosion and Multilateral Trade Liberalization Joseph Francois, Bernard Hoekman, and Miriam Manchin Because of concern that tariff reductions in Organisation for Economic Co-operation and Development (OECD)countries will translate into worsening export performance for the least developed countries, the erosion of trade preferences may become a stumbling block for multilateral trade liberalization. An econometric analysis of actual preference use shows that preferencesare underused because of administrative burdens-estimated to be equivalent to an average of 4 percent of the value of goods traded. To quantify the maximum scope for preference erosion, the compliance cost estimates are used in a model-based assessment of the impact of full elimination of OECD tariffs. Taking into account administrative costs eliminates erosion costs in the aggregate and greatly reduces the losses for countries most affected by preference erosion. Nonreciprocal trade preferences have long been granted by Organisation for Economic Co-operation and Development (OECD) countries to various develop- ing countries. In the early history of the General Agreement on Tariffs and Trade, the pattern of these preferences reflected old colonial trade ties. In 1968, the United Nations Conference on Trade and Development (UNCTAD) recommended creation of a Generalized System of Preferences under which high-income countries would grant trade preferences to all developing countries on a nonreciprocal basis. The jury remains out on whether trade preferences have actually made a substantive difference in the welfare of recipient countries. The developing countries that were granted the fewest preferences at the incep- tion of the program in the 1960s, those in East Asia, have subsequently grovvn Joseph Francois is professor of Economics in the Department of Economics at Erasmus University Rotterdam, fellow at the TinbergenInstitute,and researchfellow at the Centrefor EconomicPolicyResearch; h s email address is francois@few.eur.nl. Bernard Hoekman is research manager in the Development Research Group at the World Bank and research fellow at the Centre for Economic Policy Research; his email address is bhoekman@worldbank.org.Miriam Manchin is a research fellow at Centro Studi Luca d'Agliano; her email address is miriam.manchin@guest.unimi.it. The authors are indebted to Nuno Limiio, Marcelo Olarreaga, Sheila Page, and Susan Prowse; to the journal editor;and to three anonymousreferees for constructive comments and suggestions. They also gratefully acknowledge support from the U.K. Depart- ment for International Development project"Global Trade Architecture and Development" and from the EW Research Training Network "Trade, Industrialization,and Development." THE WORLD BANK ECONOMIC REVIEW, VOL. 20, NO. 2, pp. 197-216 doi:l0.1093/wber/lhj010 Advance Access publication May 22, 2006 O The Author 2006. Published by Oxford UniversityPress on behalf of the International Bankfor Reconstructionand Development/ THE WORLDBW. All rightsreserved.For permissions, please e-mail: journals.permissions@oxfordjournals.org. the fastest. Those granted the deepest preferences, including the least developed countries in Sub-Saharan Africa, have not managed to increase their per capita incomes or diversify their exports significantly in the last 40 years. To a large extent, outcomes in both regions are due more to domestic policies and institu- tions than to trade policiesin OECD countries. Most analysts would agree that the major constraints on export diversification and expansion in Africa are the investment climate and supply capacity. Whatever the intended and actual impacts of trade preferences, they are a central issue in the ongoing efforts to negotiate further multilateral trade liberal- ization. Middle-income countries are increasingly concerned about the discrimi- nation they confront in OECD markets as a result of free trade agreements and the better access granted in these markets to poorer or "more preferred" developing countries on a nonreciprocal basis. Conversely, preferences are used as an argument by the least developed countries and African countries against a general liberalization of trade and removal of trade-distorting policies in agri- culture by OECD members, because of fears about the potential negativeeffects of an erosion of preferential access.' This article explores the economic relevance of trade preferences in the context of World Trade Organization (wro)-basedmultilateralliberalization. This involves both an econometric assessment of the extent to which preference schemes are actually used and a numericalassessment of the dollar valueof potentialpreference erosion associated with further wro-based, nondiscriminatory tariff reductions. The numerical analysis assumes that the principle of nonreciprocity continues to prevail--developing countries are assumed to benefit from preferential access to OECD markets without reducing their own protection.2 While many analysts have argued that nonreciprocity imposes costs on developing countries because it reduces the incentives of partners to liberalize while maintaining their own trade barriers, from the perspective of quantifying the magnitude of potential preference erosion, nonreciprocityis an appropriate constraint to impose. What matters is the loss of benefits stemming from the removal of an explicit development-motivated policy that has been put in place by OECD countries. A key question when evaluating the value-of trade preferencesto beneficiaries is the cost of obtaining the preferences. Traders requesting preferences must com- ply with administrative and technicalrequirements. The most important require- ments are related to compliance with rules of origin (see Cadot and others 1. They are also concerned about the potential negativeterms of trade effects of multilateral liberal- ization insofar as this raises the price of their imports, especially of goods that currently benefit from subsidies and protection in OECD markets, by more than the price or quantity of their exports. 2. See Anderson, Martin, and van der Mensbrugghe (2006)for an analysisof the relative contribution of liberalization by developing countries to global welfare. Francois, Hoekman, and Manchin 199 forthcoming), which define the conditions that a product must satisfy to be considered as originating from the exporting country that has been granted preferential access. Rules of origin are intended to prevent trade deflection- routing exports from nonbeneficiaries through a participating country to avoid customs duties (Brenton and Manchin 2003). When products are produced in a single stage, origin is relatively easy to establish. For all other cases, the rules of origin define the criteria that determine whether the product has been suffi- ciently processed in the exporting country to qualify for preferential access. Rules of origin have become more important as technological progress and globalization increasingly fragment production processes into stages that are undertaken in different locations. In an early study, Herin (1986)argued that the costs of documentation and administration of origin rules applied by the European Economic Commurdty imposed costs on exporters in European Free Trade Association countries equivalent to some 3 percent of the value of the goods traded. More rec,ent work on the North American Free Trade Agreement (NAFTA) by Carrhre and de Melo (2004) finds that total compliance costs for Mexican exporters to the United States averaged 6.2 percent in 2001. Using double-censored Tobit esti- mation techniques, they obtain a compliance cost estimate of 3.9 percent for products where the use rate is below 100 percent.3 For developing countries, these costs are expected to be even higher because of information disadvantages and institutional weaknesses. The Estimating Framework A threshold technique is used to find the minimum preference margin (the difference between the preferential and nonpreferential tariff) below which traders have little or greatly reduced incentive to ask for preferences because the costs of obtaining them exceed the benefits. Data availability constrains the analysis to the preferential trade relations between African, Caribbean, and Pacific (ACP) countries and the European Union (EU)under the Cotonou agree- ment. Nevertheless, the assessment should provide a more general proxy for the costs traders from developing countries have to bear for requesting or obtaining preferential acces4 Because factors other than preferential margins may influence decisions to ask for preferences, Hansen's (2000)technique is used to endogenously determine a threshold in the relationship of interest while controlling for other factors. The threshold estimation technique is ideal when data must be split into subsamples to examine the relationship of interest. This approach allows the estimated 3. See also Anson and others (2005),who estimate that average compliance costs in NAFTA are about 6 percent, offsetting the preferential tariff differential of about 4 percent. Administrative costs chewed up about half the value of preferential access for Mexican firms. 4. Section III does take into account that for a subset of least developed countries, the U.S. African Growth and Opportunity Act (AGOA)has substantially reduced the costs of rules of origin. 200 T H E W O R L D BANK E C O N O M I C REVIEW, VOL. 20, NO. Z regression coefficients to differ dependingon the value of a particular parameter. The threshold estimation model can be written as follows: In equation (I),y is the threshold parameter that splits the sample into two subsamples-here the threshold value under which traders have no incentive to request preferences-@and 6 are sets of coefficients applying to explanatory variablesx and xi(y)= xidi(y).The term di(y)is a dummy variable identifyingthe threshold value for the splitting parameter, which is the difference between preferential and most favored nation tariffs. The first step is to identify the threshold value of y and the other coefficients. This is done with an algorithm from Hansen (2000) that searches through the values of y until the splitting value is found [this is the value of y that minimizes the concentrated sum of squared errors based on an ordinary least square (o~s) regression]. The decision to use preferences depends on the net benefit from requesting them. At the margin, this requires balancing the costs of obtaining preferences against the resulting benefits. This is influenced by source market characteristics, the characteristics of the market being supplied, and transportation and other transaction costs. Thus, use rates are expected to map to the same variables that determine trade itself. Therefore, in identifying y, standard gravity model vari- ables are used. As a proxy for the trading countries' market size, the gross domestic product (GDP) (GDP; is the level of income in country i) of both partner countries is included in the regression.5 GDP data are taken from the World Bank's (2005)World Development Indicators database. The level of develop- ment of the country or the quality of its economic environment can also influ- ence the net revenue from using the preferences. Therefore, an indicator of economic freedom (Freedomindex) was included in the regression. The index was obtained from Freedom House's Freedom in the World databa~e.~The lower a country's index, the more economic freedom the country has. The more distant trading partners are from each other, the higher transport costs will be. Distance (Dii)is therefore a proxy for transport costs. The data originate from the Paris-based Centre d'Etudes Prospectives et d'Informations Internationales (CEPII) distance database.' The transaction costs for using the preferences might be lower between countries with historical links. To proxy for these links, two zero-one type dummy variables were included in the regression. FrenchExcolonyii and Non- FrenchExcolonyiitake a value of 1 if the exporting country (i)was a colony of France or other partner country (j).Colonial links often reflect not only histor- ical ties but also that the traders speak the same language. A separate dummy 5. As a robustness check, further specifications were estimated using population and income per capita as a proxy for the size of the economies. All yielded the same threshold value. 6. For details see http://www.freedomhouse.org/research/freeworld/2003/methodoIogy.htm. 7. For details see http://www.cepii.fr/anglaisgraph/bdd/d~stances.htm. Francois, Hoekman, and Manchin 201 variable is included for non-French former colonies and for French former colonies because there might be differences in the intensities of the trade links for French former colonie~.~ In many aspects, such as size of the economy or level of development, South Africa differs from most of the other ACP countries in the sample. To avoid having specificities of South Africa drive the results, a dummy variable taking a value of 1 if the exporting country is South Africa is included in the regressions. SECDUMijkis a set of k dummy variables for agriculture, textiles, clothing, footwear, machinery, and mineral products.9 Finally, net revenues that can be earned from using the preferences are assumed also to depend on the size of preferences available as measured by the difference between most favored nation and preferential tariffs for each product k (Dutydifferencek).This is the variable for which a threshold value of :i is identified. To identify the threshold value in the preferential margin, the following equation is estimated using OLS: InUTILijk= a +PIIn GDPi +,5~In GDPj +,83 In Dii +P4FrenchExcolonyi+~sNonFrenchExcolonyj (2) + P6Freedomindexi+,&SouthAfrica +PsDutydifferenceijk+ ,L$kSECDUMiik eijk~ + ijk In equation (2),UTILijkis the use rate of Cotonou preferences for product k, or the percentage of country i's imports of product k from country j for which preferential access was requested. As the use rate ranges from 0 to 1, a logistic transformation of the observed use rate is applied, setting zero rates to 01.001and full use rates to 0.999. The data are import flows for 2001 from Eurostat at t:he eight-digit level of product disaggregation. Estimation Results The 95 percent confidenceinterval (CI) for the threshold estimates (cut-offvalue) places the threshold between the 48th and 52nd percentiles. This impliesthat the preferential tariff must be 4 4 . 5 percentage points lower than the most favored nation tariff for traders to request preferences. This ais plausiblytight: only 3,41 of 23,685 observations fall within the 48th and 52nd percentiles. With 1,0100 bootstrap replications, the p-value for the threshold model is significant at the 8. Only four countriesin the sample were not colonies. 9. Of the 23,685 observations, 9,015 are not covered by any sectoral dummy variable. These observations belong to sectors such as metals, vehicles, optics, chemicals, plastics, stones, and glass. 202 THE WORLD BANK ECONOMIC REVIEW, VOL. 20, N O . 2 2.5 percent level. In terms of use rates, 70 percent of observations where preferences were requested were above the threshold and 30 percent were below. The average use rate is 16 percent when the duty reduction is less than 4 percent and 43 percent when it is more than 4 percent. That some traders request preferences when benefits are very limited could reflect lack of information about the preference scheme or the fact that costs are only measurable ex post.10 To test the robustnessof the results, the threshold regressionswere also run with country-specificfixed effects instead of the country-specific variables. The same threshold values were obtained for allthe specifications: the preferentialtariff must be lower than the most favored nation tariff by 4-4.5 percentage points.11 Table 1 reports OLSestimates using the logistic transformation for the thresh- old model based on the threshold value of a 4 percent duty difference for country-specific explanatory variables and country-specific fixed effects. The variable belowthreshold measures duty reduction when the difference between the most favored nation and preferential duty is smaller than the threshold, whereas the variable abovethreshold measures duty reduction when the differ- ence is higher than the threshold. The results confirm that the threshold value was correctly identified. The coefficient of the variable belowthreshold is nega- tive. When the duty reduction is above the 4 percent threshold, the probability of using the preference scheme rises. Thus, there is a different relationship between tariff reduction above the threshold and use of preferences and between tariff reduction below the threshold and the use of preferences. While not reported because of space constraints, the threshold was also estimated separately for a split sample that distinguishes agricultural from non- agricultural products. Both estimates of y remain significant at the 5 percent level. The manufacturing threshold is identical to the full-sample threshold, whereas the agricultural threshold is somewhat higher, at 15 percent. Higher rates for agriculture appear counterintuitive in that the consensus in the litera- ture is that it is easier to prove origin for such products (Stevens and Kennan 2004).12The higher threshold for agriculture may reflect, in addition to rules of origin, compliance costs related to technical and sanitary requirements and the procedural requirements associated with the administration of quota-guaranteed access to protected markets. These procedural requirements are important for 10. Several observations in the datasetindicatedthat some traders requested preferenceseven when preferenceswere notavailable,which also confirmsthatthoroughunderstandingof the preference scheme might be lacking. 11. The p-value for the threshold model using country-specificfixed effects was not significant. 12. Candau, FontagnC, and Jean (2004)find that underuse of preferences is highest in textiles and garments (for EU imports under both the Generalized System of Preferences and Everything but Arms programs).In the Everything but Arms program,exporters in principle benefit from 100 percentduty-free access but are found to pay up to 6.5 percent average tariffs. Francois, Hoekman, and Manchin :203 TABLE 1. Endogenous Threshold Regressions: Logistic Transformation of Use Rate Independent Model 1 Country-Specific Model 2 Country-Specific Variables Explanatory Variables Fixed Effects Lngdp-i Lngdpj Lndist SouthAfrica Freedomindex FrenchExcolony -0.056 (0.016)" NonFrenchExcolony 0.057 (0.016)'~ DUMagri 0.263 (0.007)" DUMtext 0.205 (0.012)*' DUMfoot 0.109 (0.017)*' DUMrnach -0.109 (0.006)" DUMwood 0.346 (0.014)'* DUMmineral 0.06 (-0.038) DUMcloth 0.197 (0.009)** Belowthreshold -1.046 (0.211)** Abovethreshold 0.161 (0.033)" Constant 3.74 (l.OSE+O'l) Number of observations 23,631 R-squared 0.36 Country-specific fixed effects Included *Significant at the 5 percent level. *+Significantat the 1 percent level. Note: Estimates of threshold values are based on least squares estimation (see Hansen 2000). Numbers in parentheses are standard errors. Source: Authors' analysis based on data described in the text. commodities such as sugar and bananas in the EU.'~It may also be that the lower unit values for agricultural products imply that the percentage equivalent foir a given set of administrative requirements is higher. To illustrate the implications of the estimated threshold level for trade pre- ferences, consider EU imports from least developed countries for 2001. Table 2 provides estimates of the rate of most favored nation protection that would be applied against least developed country exports to the EU (weighted by th'eir exports), the share of imports by sector reported as actually entering the EU duty free, and the actual underlying trade flows. Note that for least developed countries, the most important exports (by value) are manufactured products, followed by mining products (generally duty free on a most favored nation basis). On average, many EU tariffs in manufacturing are below the estimated threshold. Yet there are some peak rates that make preferences worthwhile in 13. De Gorter and Kliauga (2006) document the complexity of the administrative requirements associated with the implementationof quota regimes (tariffrate quotas). TABLE 2. EU15 Imports from the Least Developed Countries, 2001 Total Imports Duty-Free Imports Sector Share of Duty-Free Share Most Favored Nation Product (US$ Thousands) (US$ Thousands) Total (%) of Sector (%) Tariff Rate (%) Agriculture Forestry, fisheries Mining Processed foods Animal products Vegetable oils and fats Dairy products Processed rice Sugar Other food products Beverages and tobacco Manufactures (nonfood) Textiles Clothing Leather Wood products Paper products Petroleum and coal products Chemicals, rubber, plastics Nonmetallic minerals Iron and steel Nonferrous metals Fabricated metal products Motor vehicles and parts Other transport equipment Electrical machinery Other machinery Other manufactures Total - - - - -- Note: Duties are weighted by least developed country exports to the ELI.Duty-free imports are because either most favored nation tariff rates are zero or because imports otherwise received duty-free treatment. Source: WTO integrated database. Food estimates for the last column are from GTAP (version 6). Francois, Hoekman, and Manchin :205 specific categories. Thus, the benefits of preference regimes hinge on the margin of preference at the tariff-line level.14 The highest use of preferences in 2001, as proxied by duty-free-eligible imports, was in agriculture and processedfoods-productswith rates of protec- tion that are generally well above the identified thresholds. That agricultural protection is high and that preferences in this sector are often reflected in tariff quotas implies that high observed use rates are not inconsistent with the rela- tively high average split-sample threshold estimate: the implication is that the associated rents are high enough to cover the costs. The results presented in this section indicate that traders require a certain minimum preference margin before they request preferences. If the difference between preferential and most favored nation tariff rates is less than that amount, there are no incentives for traders to request preferences, because the costs of obtaining the preferencesare expected to be higher than the benefits of obtaining the preferences. This threshold for ACP countries in their preferential trade with the EU was between 4 and 4.5 percent overall, with higher costs for agriculture-based products. Although this figure pertains to a specific group of developing countries, it provides an approximation of the costs associated with preferential schemes for other countries as well, because the requirements are similar. The estimate is broadly consistent with others reported in t:he literature-see, for example, Cadot and others (2005, 2006, forthcoming). Next, a global general equilibrium model is used to provide a numerical assess- ment of the likely magnitude of preference erosion if OECD countries were to implement further multilateral tariff reductions. As mentioned previously, it is assumed that developing countries do not liberalize. In the model, preferences are included as part of the benchmark data, and the estimates of the adminis- trative costs of preferences in the previous section are integrated in the assess- ment of the overall benefit of preferences. The Mechanics of Erosion To examine the basic mechanicsof preferencesand preferenceerosion, start with figure1, in which an archetype OECD country imports varieties of good X from a least developed country supplier, SLDC,and from a non-least developed count:ry supplier, Snon-LDC.Trade preferences are represented by a reduction in the tariff applied to imports from the least developed country. The result is a new equili- brium in which there is an increase in exports by the least developed count:ry supplier from XLDC,Oto XLDC,~.The benefit for the least developed count:ry 14. Simple regression analysis of the data in table 2 confirms that the share of duty-free trade, a~nd hence the implicit use of preferences, is indeed significantly and positively correlated with the peak tariff rates in the table. FIGU E 1. Impact of a Tariff Preference R exporter is represented by area A. At the same time, there will be a shift in demand away from imports from the nonpreferential supplier, resulting in a cost represented by area Bythe loss in exporter surplus. The magnitude of these costs and benefits depends on underlyingsupply and demand responsiveness to price changes as well as on the degree of substitution between preferential and non- preferential suppliers. The impact on the importer depends on a mix of effects- terms of trade, trade creation, and trade diversion. On net, therefore, trade preferences involve a mix of benefits for preferential exporters, costs imposed on third-country exporters, and potential losses for the importer.15 Basically, trade preferences are a beggar-thy-neighbor type of for- eign aid-robbing Peter to pay Paul. This is why there has been tension between the least developed countries, the "not-quite" least developed countries, and other developing countries in the context of WTO negotiations to liberalize trade. Preference erosion involves the reduction or elimination of tariffs on the nonpreferential supplier, as shown in figure 2. Elimination of the tariff on the remaining third-country suppliers, given the duty-free access already granted to preferential suppliers, means that third-country exporters see their exports increase from Xnon-LDC,l to Xnon-LDC,Z. In the new equilibrium, there is a gain in exporter surplus of area E, which may be greater or less than the original loss of exporter surplus resulting from the preferences, area B in figure 1. The 15. The impactcan be represented in stages:an expansionof least developed country exports results in a subsequentfall in non-least developedcountryexportsand price.Shownhere is thefinal result of such an adjustment process. Francois, Hoekman, and Manchin 207 FIGURE 2. Impact of Liberalization by a Preference-Granting Country P~~~ P n o n - ~ ~ ~ preferential supplier sees a drop in demand for its exports from DLDC,lto DLDC,2.This results in a partial loss of the benefitsfrom the original preference scheme. This is represented by area C, which is shown as being less than area A in figure 1. The reason the loss is not complete is that preferences include the benefits relative to the original tariff-ridden equilibrium from a nondiscrimina- tory tariff reduction by the importer. Thus, preference erosion generallyyields a partial, not a full, loss of the original benefits of the preference scheme. At the same time, third-countries recover some of the costs originally imposed by the preference scheme. A few caveats are in order here. First, to the extent that either importers (Francoisand Wooton 2005) or the transport and logistics sector (Francois a:nd Wooton 2001) exercise market power, they can be expected to capture at least some of the benefits of tariff reductions rather than the exporters. There is evidence, based on the U.S. African Growth and Opportunity Act (AGOA) pre- ference scheme that the pass-through of preference margins is indeed partial at best. Olarreaga and Ozden (2005) find that exporters received an average of one-third of the tariff rent, with poorer and smaller countries tending to obta.in lower shares-as low as 13 percent in ~ a 1 a w i . lIn~ addition, based on the analysis in the previous section, administrative costs related to these programs 16. See Ozden and Sharma (2006)for an analysis of the Caribbean Basin Initiative program. Cadot and others (forthcoming)estimate the share of NAFTA tariff preferences that accrue to Mexican exporters of apparel at about 50 percent, once account is taken of the compliancecosts associated with rules of origin. can be expected to chew up some of the benefits. Although not shown in the figure, this implies deadweight losses involving parts of areas A and C. In the case of market power, the result is a simple redistribution of the benefits of preferences (rents)to importers. With administrative costs, however,the share of the gains that is lost is not redistributed but is a deadweight loss. In both cases, the trade effects of preference programs will be less as well. A Numerical Assessment While considerable political weight has been attached to this issue, the debate has occurred largely in a vacuum, without real information on costs and bene- fits. The exceptions have focused almost exclusively on the effects of preference erosion on the exports of beneficiary c~untries.~'This section uses a global multiregion general equilibrium model of trade to provide a numerical assess- ment of the likely magnitudes of costs and benefits on nondiscriminatory trade liberalization by OECD members. The model includes 34 regions and countries and 24 sectors.'* The social accounting data are from the Global Trade Analysis Project (GTAP) database version 6.0 benchmarked to 2001. The data include national production and international trade flows. The import protection data are based on a thorough effort to include use of preferences in a matrix of global import protection data (Bouet and others 2004). These data are the product of a joint effort of the U.N. International Trade Centre, UNCTAD, the m o , and CEPII. An important contribution of this project has been the exhaustive coverage of preferential trade arrangements across the world and calculations of the ad valorem equivalents of specific duties. Combined with differences in the bilateral composition of trade, the result is that protection varies by sector and partner for each importer. These data have in turn been integrated with the GTAP database for 2001. For this study, the data were further modified to assume full use of the 2001 EU Every- thing but Arms initiative and the U.S. AGOA for African countries benefiting from more liberal rules of origin, as this has been implemented over a period extend- ing beyond the benchmark year of the original protection data. Elimination of quotas on textiles and clothing trade has also been imposed on the benchmark as 17. See, e.g., IMF (2003),Alexandralu and Lankes (2004),and Brenton and Ikezuki (2005). 18. The 34 regions are the European Union (EU25), Turkey, Russian Federation, Other Europe, Middle East, North Africa, Botswana, Madagascar, Malawi, Mozambique, South Africa and Namibia, Tanzania, Uganda, Zambia, Other Sub-Saharan Africa, Canada, Mexico, United States, Central America, Caribbean Islands, Argentina, Brazil, Other South America, Japan, High-income Asia, China, Vietnam, Other Southeast Asia, Bangladesh, India, Sri Lanka, Other Central and South Asia, Oceania, Australia, and New Zealand. The 24 sectors are rice, wheat, other cereals, horticulture and other crops, sugar, intensive livestock and products, cattle and beef products, milk and dairy, cotton, other agriculture, processed food products, textiles, clothing, leather goods, extraction industries, petroleum and chemicals, ferrous and nonferrous metals, motor vehicles, metal and electro-technical, other industries, construction, trade services, transport services, business services, and other services. Francok, Hoekman, and Manchin 209 required by the implementation of the wro Agreement on Textiles and Cloth.ing as of January 1, 200.5.'~ The model is a standard general equilibrium model, with Cobb-Douglas consumer demand over broad product categories and constant elasticity of substitution-based demand within product categories. The overall theoret:ical structure of the model follows Francois, van Meijl, and van Tongeren (2005). For primary sectors, this is Armington-based trade. Manufacturing and busir~ess service sectors are modeled as monopolisticallycompetitive, whereas the extrac- tion and construction sectors are modeled with industrywide scale economies. Scale elasticityestimates are based on Antweiler and Trefler (2002)and Francois (2001). Factor supplies are fixed nationally and are allocated between sectors through factor markets2' The experiment involves elimination,on a multilateral basis, of all OECD import tariffs on all goods. This includes the ad valorem equivalents of specific tariffs and tariff-rate quotas and takes into account the prevailing preference programs as reported by CEPII. A subexperiment was also conducted, with these tariffs elimi- nated first for the EU alone, to provide an opportunity to identify the full magninude of preference erosion for a sample of least developed and low-income countries (conceptuallyrepresented by area C in figure 2) and to identify the share of the effects that are due to EU preferences. Finally, the estimate of EU preferenceerosion was recalculated after adjustingfor the administrativecost threshold (4percent for manufactures and 15 percent for agriculture) identified in the previous section. This has a substantial impact on the estimated scope for preference erosion. The recalculationinvolves two steps. For sectors where the most favored nation tariiff is below the threshold, no preference use is modeled, and so there is no erosion. :For sectors where use is modeled, recovery of the deadweight costs proxied by the threshold values is allowed for once tariffs are eliminated. Estimates of the annual dollar impact of full preference erosion on real national income are shown in table 3, which includes the impact on the least developed countries in Sub-Saharan Africa as well as on other low-incolme countries in the sample (using the World Bank classification of countries by income). The tables reveal that EU preferences are very important, as a bilateral measure, for Sub-Saharan African countries. Given the current trade policy landscape, and before adjusting for use rates, EU preferences are estimated to be potentially worth some $460 million a year to African least developed countries. Asian countries benefit less, with the exception of Bangladesh ($100 million).These countries therefore stand to lose-all other things equal-from a move by the EU to lower most favored nation trade protection. 19. This is of course an important dimension of preference erosion in its own right, insofar as the constraint on the most efficient producers under the Agreement on Textiles and Clothing implied there was an "implicitn preference for the non- or less constraineddeveloping country exporters. The impacts are assessed in greater detail in Francois and Woertz (2006). 20. The full model can be downloaded at http:llwww.intereconomics.com~francoisld. TABLE 3. The Impact of Full Preference Erosion: National Income Changes (Millions of U.S. Dollars) Preferences Unadjusted for Administrative Burden Preferences Adjusted for Administrative Burden Effects of EU Effects of Other Total Effects of EU Effects of Other Total Full Trade OECD Full Trade Preference Full Trade OECD Full Trade Preference Liberalization Liberalization Loss Liberalization Liberalization Loss African least developed countries Botswana Madagascar Malawi Mozambique Tanzania 2 Uganda 0 Zambia Other Sub-Saharan African least developed countries Asian and other least developed countries Bangladesh Other Central and South Asian least developed countries Other low-income countries India Vietnam Total Note: Least developed country and low-income country classification is based on World Bank designations. Source: Authors' analysis based on data described in the text. Francois, Hoekman, and Manchin 211 Other developingcountry groups stand to gain-these are the "less preferred" in the overall hierarchy of preferences.21Indeed, the gains for other low-income countries reveal why developing countries are at odds over preferences in dis- cussions in the WTO. One group gains at the expense of the other. Although the potential preference rents probably do not all accrue to the exporting countries (Olarreaga and Ozden 2005; Ozden and Sharma 2006), the estimates give a sense of what is at stake.22 The welfare estimates here cannot be compared directly with the results from recent partial equilibrium-based analyses of erosion (such as IMF 2003 and Alexandraki and Lankes 2004), which focus on trade effects. The International Monetary Fund (IMF 2003) estimates a potential export revenue loss of some $530 million from preference erosion resulting from a 40 percent cut in protec- tion by Canada, the EU, Japan, and the United States. This assumes that prefer- ences are fully used and that developing countries get all the associated rents. Alexandraki and Lankes (2004), focusing only on middle-income countries, conclude that potential erosion impacts are less than 2 percent of total exports for the most preference-dependent countries. Limfo and Olarreaga (2006)esti- mate that income transfers to least developedcountries of $266 million would be needed to equal the transfers implied by existing preference programs. This is a one-year, short-run effect-all else equal, the net present value is argued to be several times higher. Their results are in line with those reported here, assuming away the compliancecosts associated with preferenceprograms (whichthey do). If preference erosion is viewed in the broader context of potential tariff reduction by all OECD countries, not just EU members, the magnitude of the total losses is reduced. In part, this is because the EU has been the most aggressive in using preferences as a tool for development assistance-such programs in other OECD countries have been subject to greater exceptions (for example, the exclusion of apparel from the U.S. Generalized System of Preferences programs and the fact that AGOA does not cover all products). Thus, the gains associa~ted with non-EUmost favored nation tariff reductions will partially offset losses due to EU liberalization. For Sub-Saharan Africa, the overall losses will be reduced significantly. In addition, low-income countries in Asia stand to gain substan- tially from other OECD tariff reduction^.^^ What are the implications of taking into account the threshold estimates of compliance costs?The right panel of table 3 reports a second set of estimates for preferenceerosion tied to EU preferencesand all other OECD preferences. The left panel shows that the EU preferencesare the dominant issue at play. The estimates 21. The income effects are mirrored in the trade effects, which reveal that export reductionsmap to income reductions (not reported).This is fully consistent with the earlier discussion of figure 1. 22. As discussedin Hoekman and Prowse (2005),a case can be made that even if exporters do not get the rents they should get them. In a normative discussion that focuses on offsetting the loss ikom preference erosion, one can argue that account should be taken of any "missing" rents. 23. As noted by Anderson, Martin, and van der Mensbrugghe (2006),the gains are even greater if developing countryreforms are also considered. 212 THE WORLD BANK ECONOMIC REVIEW, VOL. 20, N O . 2 in the right panel are based on the earlier estimate of the compliance costs for EU trade preferences. These costs are eliminated as part of the experiment. Prefer- ence use is also removed in the base data when most favored nation rates are below these compliance costs. The magnitude of preference erosion changes dramatically overall, with the change varying across countries. For Bangladesh, which specializesin high-tariff manufacturing categories such as clothing that are subject to restrictive rules of origin, the income effects of potential erosion change from a loss to a gain. For Madagascar too, potential losses turn into potential gains. The reason for these results is that the compliance costs associated with implementing preference programs bias estimates of the value of preferences upward. Overall, allowing for compliance costs, there are no longer significant losses for African least developed countries as a group, though there are for individual countries. What this says then is that on net, EU preferences do not really offer collective benefits to African least developed countries. To the extent that individual countries benefit, blocking multilateral reductions to maintain these benefits involves hurting some (neighboring) countries with no real net benefits for the region as a whole. These results also point to the need for country-by-country analysis and assessments of the potential impacts of preference erosion.24 The resultsin table 3 imply that the magnitude of any transfer needed to offset (orcompensate for) the effect of erosion is much smaller when all OECD countries liberalizethan when the EU alone liberalizes. Individual least developedcountries stand to lose from tariff reductions in sectors or products where preferences matter. However, they also stand to benefit from improved overall access to OECD markets, a process that may outweigh any more direct losses on a bilateral basis from erosion of preference margins. They also stand to gain from better access to other developingcountry markets. While not modeled in this study-as the more advanced developingcountries have not granted significantpreferential access to low-income countries-other research has shown that much of the potential market access gains for the poorest countries are in other developing countries (Francois, van Meijl, and van Tongeren 2005; Anderson, Martin, and van der Mensbrugghe 2006). Becauseof concern that OECD tariff reductionswill translate into worsening export performance for the least developed countries, trade preferences may be a stum- bling block to obtaining broad-based support for deep liberalization by OECD 24. IMF (2003) concludes that individual least developed countries may suffer more than average because of the concentration of their exports in products that enjoy deep preferences. Of the least developed countries, Cape Verde, Haiti, Malawi, Mauritania, and SZo Tom6 and Principe are the most vulnerable to preference erosion. Alexandraki and Lankes (2004) conclude that six middle-income countries-Belize,Fiji, Guyana, Mauritius, St Kitts and Nevis, and St Lucia-would also be significantly affected, with predicted export declines ranging from 11.5 percent for Mauritius to 7.8 percent for Fiji. Francois, Hoekman, and Manchin 213 countries in the WTO. This article examined the potential magnitude of full preference erosion, through an econometric assessment of actual use of prefer- ences and numericalmodeling of full elimination of OECD tariffs. Strongstatistical evidence was found that administrative burdens lead to underuse of preferences. This presumably reflects rules of origin and related hurdles in the way of actually using trade preferences. These substantiallyreducethe actual value of preferences. What are the policy implications of these findings? Preferences can have an impact only if there is a nonzero tariff in the importing market. Two- third:^ of the major items Africa exports to Canada, for example, face zero most favored nation tariffs, and 69 percent of EU imports from Africa (byvalue) in 2000 faced zero most favored nation duties (Stevensand Kennan 2004). One policy option sometimessuggested is to raise trade barriers to increase the value of preferential access. This would clearly be globallywelfare reducing, however.More comnnon is the argument that preferred developing countries should not lose any more preferential access to highly distorted OECD markets. The result would b'e a potential status quo bias, with potentially significant opportunity costs of 1.iber- alization forgone. Evidence also suggests that interests in developed countries may be extracting a significant share of the rents from preferential access (Francois and Wooton 2005; Olarreaga and Ozden 2005; Ozden and Sharma 2006). This points to a political economy of support for preferences that has nothing to do with genuine development objectives. In U.S. dollar terms, the largest negative impact of erosion follows from EU liberalization. This suggests that the erosion problem is primarily a bilateral concern2' and might therefore best be addressed bilaterally through develop- ment assistance financing.26This is not to deny that the recent extensionl of deeper preferences for least developed countries and other Sub-Saharan African countries by other OECD members has created new preference erosion conce:rns for beneficiary countries. When Canada, for example, granted duty-free access to clothing imports for certain countries in 2003, exports grew threefold from Bangladesh and sevenfold from Cambodia. But access to the other major OECD market-the United States-remains restricted for key items such as apparel. Indeed, the 2005 wro Ministerial Meeting in Hong Kong illustrated how strong the resistance in the United States is to granting full duty-free access for all 25. The bilateral nature of the erosion problem is clearest in the case of specific products such as slugar and bananas, where a select subset of preferred countries has been granted large rents through quota- guaranteed access to the highly protected EU market. 26. Options could be to use instruments such as the European Development Fund or a program such as that adopted in the context of reform of the EU Common Agricultural Policy to provide affected producers in ACPcountries with income support and adjustment assistance. A small step in this direction was taken in the recent EU reform of its sugar regime, under which ACP producers were allocated 40 million in income transfers. Considering that compensation payments to EUsugar producers under this reform are some 1.5 billion, there should be scope to expand the amounts allocated to developing country producers in the context of such adjustment programs. products-particularly those that matter most in terms of export potential.27 Thus, the conclusion that multilateral liberalization by all OECD countries has the potential to offsetthe aggregate amount of erosion created by the removal of EU trade barriers appears robust to the recent deepening of preferential access programs in other OECD countries. Recent researchcited above has shown that in practice, rules of origin may be a binding constraint on the ability of poor countries to exploit preferentialaccess opportunities. For example, Brenton and Ozden (2005) document the large supply response that followed the relaxation of origin requirements for eligible countries under the U.S. AGOA program. This supply response is not taken fully into account in the simulations here, although the database was updated to reflect the preferential access granted to least developed countries and other Sub-Saharan African countries. The extension of duty-free access programs to more countries and products implies that the erosion problem is growing. More- over, if the EU were to follow Canada and the UnitedStates in relaxingits rulesof origin for least developed countries, the supply response could well be substan- tial, further reducing the incentive of these countries to support most favored nation liberalization. Indeed, the analysis suggests that relaxation of rules of origin and measures to reduce administrative costs would both enhance the benefits of existing duty-free access schemes and increase the costs of erosion. Actions to achieve this are therefore potentially useful as part of an effort to reduce the short-run impacts of tariff liberalization on preference-dependent countries.28 Considering the systemic downsides, limited benefits, and historical inability of many poor countries in Africa and elsewhere to use preferences, a decision to shift away from preferential "trade as aid" toward more efficient and effective instruments to support poor countries could both improve development out- comes and help strengthen the multilateral trading system (Hoekman and Prowse 2005; Zedillo and others 2005).More effectiveintegration of the poorest countries into the trading system requires instruments aimed at improving the productivity and competitivenessof firms and farmers in these countries. Supply constraints are the primary factors that have constrained the ability of many African countries to benefit from preferences. This suggests that the main need is to improve trade capacity and facilitate diversification. In part, this can be pursued through a shift to more (and more effective) development assistance 27. The final compromise was to allow 3 percent of all tariff lines-the most politically sensitive items-to be excluded from duty-free access programs for the least developed countries. With exports frommost developingcountries concentratedin just a few tariff lines, this can implythatthe majority of a country's exports could be excluded.For example,more than 70 percent of Bangladesh'sexports to the United States are in 70 tariff lines that together account for less than 1 percent of U.S. tariff lines. 28. Annex F of the Hong Kong m o Ministerial Declaration dealing with Least DevelopedCountries Agreement-specific proposals specifies that m o members must ensure that preferential rules of origin applicable to imports from least developed countries are transparent and simple and contribute to facilitating market access. Francois, Hoekman, and Manchin 215 that targets domestic supply constraints as well as measures to reduce the costs of entering foreign markets. Actions in these areas would assist firms and communities that stand to lose from preference erosion while benefiting a much larger set of agents by enhancing the gains from trade opportunities. Alexandraki, K., and H. P. Lankes. 2004. "The Impact of Preference Erosion on Middle-Income Coun- tries." IMF Working Paper 041169. International Monetary Fund, Washington, D.C. Anderson, K., W. Martin, and D. van der Mensbrugghe. May 9, 2006. "Doha Merchandise Trade Reform: What's at Stake for Developing Countries?" World Bank Economic Review 2.0(2), 10.1093lwber/lhj009. Anson, J., 0. Cadot, A. Estevadeordal, J. de Melo, A. Suwa-Eisenmann, and B. Tumurchudur. ;!005. "Assessing the Costsof Rules of Origin in North-South PTAS with an Applicationto NAF~A." Review of International Economics 13(3):501-17. Antweiler, W., and D. Trefler. 2002. "IncreasingReturns and All That: A View from Trade." American Economic Review 92(1):93-119. Bouet, A., Y. Decreux, L. Fontagne,S. Jean, and D. Laborde. 2004. "A Consistent, ad-valorem Equivalent Measure of Applied Protection across the World: The MAcMap-HS6 Database." c~pnDiscussion Paper 2004-22. Centre &Etudes Prospectiveset &Informations Internationales, Paris. Brenton, P., and M. Manchin. 2003. "Making EU Trade AgreementsWork: The Role of Rules of Origin." The World Economy 26(5):755-69. Brenton, P., and T. Ikezuki. 2005. "The Impact of Agricultural Trade Preferences, with Particular Attention to the Least Developed Countries." In M. Aksoy and J. Beghin, eds., Global Agricul'tural Trade and Developing Countries. Washington, D.C.: World Bank. Brenton, P., and C. Ozden. 2005. "Trade Preferences for Apparel and the Role of Rules of Origin: The Case of Africa." World Bank, Washington, D.C. Cadot, O., C. Carrtre, J. de Melo, and A. Portugal-Perez.2005. "Market Access and Welfare Under Free Trade Agreements:Textiles Under NAJTA." World Bank Economic Review 19(3):379-406. Cadot, O., A. Estevadeordal,A. Suwa-Eisenmann,and T. Verdier,eds. 2006. The Origin of Goods: Rules of Origin in Preferential Trading. New York: Oxford University Press. Cadot, O., C. Carrtre, J. de Melo, and B. Tumurchudur. Forthcoming. "Product Specific Rules of Origin in EU and us Preferential Trading Agreements: An Assessment."World Trade Review. Candau, F., L. Fontagni, and S. Jean. 2004. "The Utilization Rate of Preferences in the EU." Centre dYEtudesProspectives et &Informations Internationales, Paris. Carrtre, C., and J. de Melo. 2004. "Are Different Rulesof Origin EquallyCostly?Estimatesfrom NA~FTA." CEPR DiscussionPaper 4437. Centre for Economic Policy Research, London. De Gorter, H., and E. Kliauga. 2006. "Reducing Tariffs versus Expanding Tariff-Rate Quotas.." In K. Anderson and W. Martin, eds., Agricultural Trade Reform and the Doha Developnzent Agenda. New York: Palgrave Macmillan and World Bank. Francois, J. 2001. "The Next wro Round: North-South Stakes in New Market Access Negotiations." University of Adelaide, Centre for International EconomicStudies,Adelaide, Australia, and Tinbergen Institute, Amsterdam and Rotterdam. Francois,J. F., and I. Wooton. 2001. "Trade and Competition in ShippingServicesand the GATS."Review of International Economics 9(2):249-61. .2005. "Market Structure in Servicesand Market Access in Goods." EPR DiscussionPaper 5135. C Centre for Economic Policy Research, London. Francois, J., and J. Woertz. 2006. "Rags in the High Rent District: The Evolution of Quota Rents in Textiles and Clothing." Yale University, Yale Center for the Study of Globalization, New Haven, Conn. Francois, J., H. van Meijl, and F. van Tongeren. 2005. "Trade Liberalization in the Doha Development Round." Economic Policy 20(42):349-91. Hansen, B. 2000. "Sample Splitting and Threshold Estimation." Econometrics 68(3):575403. Herin, J. 1986. "Rules of Origin and Differencesbetween Tariff Levelsin EFTA and in the EC." Occasional Paper 13. European Free Trade Association, Economic Affairs Department, Geneva. Hoekman, B., and S. Prowse. 2005. "Economic Policy Responses to Preference Erosion: From Trade as Aid to Aid for Trade." Policy Research Working Paper 3721. World Bank, Washington, D.C. IMF (InternationalMonetary Fund). 2003. "Financing of Losses from Preference Erosion, Note on Issues Raised by Developing Countries in the Doha Round." WTEFICOW14. Washington, D.C. Limio, N., and M. Olarreaga. May 17, 2006. "Trade Preferences to Small Countries and the Welfare Costs of Lost Multilateral Liberalization." World Bank Economic Review 20(2), 10.1093/wber/ lhj013. Olarreaga, M., and C. Ozden. 2005. "AGOA and Apparel: Who Captures the Tariff Rent in the Presence of Preferential Market Access?"World Economy 28(1):63-77. Ozden, C., and G. Sharma. May 4, 2006. "Price Effects of Preferential Market Access: The Caribbean Basin Initiative and the Apparel Sector." World Bank Economic Review 20(2),10.1093/wberAhj008. Stevens, C., and J. Kennan. 2004. "Making Preferences More Effective." Briefing Paper. Institute for Development Studies, Sussex. World Bank. 2005. World Development Indicators Database. Washington, D.C. Zedillo, E., J. Audley, S. Evenett,J. Francois, E. Fuller, G. Helleiner, B. Hoekman, F. Ismail, H. P. Lankes, R. Melendez-Ortiz,P. Messerlin, D. Njinkeu, H. Pack, S. Page,S. Prowse,J. Roy, K. Saggi,J. M. Salazar, S. Siphana, T. Verdier, and I,. A. Winters. 2005. "Strengthening the Global Trade Architecture for Economic Development: An Agenda for Action." Yale University, Yale Center for the Study of Globa- lization, New Haven, Conn. Trade Preferences to Small Developing Countries and the Welfare Costs of Lost Multilateral Liberalization Nuno Limiio and Marcelo Olarreaga The proliferation of preferential trade liberalization over the last 20 years has raised the question of whether it slows multilateral trade liberalization. Recent theoretical anid empirical evidence indicates that this is the case even for unilateral preferences that developed countries provide to small and poor countries, but there is no estimate of the resulting welfare costs. This stumbling block effect can be avoided by replacing the unilateral preferences with a fixed import subsidy, which generates a Pareto improve- ment. More importantly, this paper presents the first estimates of the welfare cost of preferential liberalization as a stumbling block to multilateral liberalization. Recent estimates of the stumbling block effect of preferences with data for 170 countries and more than 5,000 products are used to calculate the welfare effectsof the European Union, Japan, and the United States switching from unilateral preferences for least developed countries to an import subsidy scheme. In a model with no dynamic gains to trade, the switch produces an annual net welfare gain for the 170 countries that adds about 10 percent to the estimated trade liberalization gains in the Doha Round. It also generates gains for each group: the European Union,Japan, and the United States ($2,934 million), least developed countries ($520 million), and the rest of the world ($900 million). One pillar of the multilateral trading system is nondiscrimination across trading partners: the most favored nation (MFN) principle in the first article of the General Agreement on Tariffs and Trade. But nearly all World Trade Organiza- tion (mo)members also participate in preferential trade agreements. 'The Nuno Lima0 is an assistant professor of economics at the University of Maryland; his email address is limao@econ.umd.edu.Marcelo Olarreaga is a senior economist at the World Bank; his email address is molarreaga@worldbank.org.Bothauthors are affiliatedwith the Centre for EconomicPolicyResearch. 'They thank StephanieAaronson, Piyush Chandra, Bernard Hoekman, Kyle Bagwell, George Bermann, Bill Davy, Aaditya Mattoo, Petros Mavroidis, Stefano Inama, Chris Stevens, Alan Winters, three anonymous referees, and participants of the seminar on the World Trade Organization and developing countries at Columbia Law School for helpful comments and discussions. They also thank the U.K. Department for International Development for financial support. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. THE WORLD BANK ECONOMIC REVIEW, VOL.20, NO. 2, pp. 217-240 doi:10.1093/wber/lhi013 Advance Access publication May 17,2006 O The Author 2006. Published by Oxford UniversityPress on behalf of the International Bankfor Reconstructionand Development1THE WORLD BANK.All rightsreserved. For permissions, please e-mail: joumals.permissions@oxfordjoumals.org. potential for preferential trade agreements to be a stumbling block to multi- lateral trade liberalization (MTL) was an important concern during the Uruguay Round (Bhagwati1991). This article quantifies the welfare effects of the more recent concern during the Doha Round that even unilateral trade preferences provided to small and poor developing countries can slow multilateral liberal- ization by the large developed countries that provide them. This problem prompted the International Monetary Fund (IMF) to create a special lending program aimed at developing countries "to mitigate concerns... that their bal- ance of payments positions could suffer, albeit temporarily, as multilateral liberalization changes their competitive position in world markets. Chief among these concerns is that broad-based tariff liberalization might erode the value of their preferential access to important export markets" (IMF 2004b).' The concern with preference erosion could affect the level of liberalization, particularly in the Doha Round, for two reasons. First, preference beneficiaries oppose MFN liberalization by countries that grant preferences. Even though erosion is critical only for a specific set of goods and countries, many other countries perceive such losses to be important. These countries can influence the current round because it was designated a "development round," creating the expectation that it would benefit developing countries.' Second, developed countries providing preferences may want to maintain them because they can be used as a side payment for cooperation on nontrade issues. Here are two examples. First, in 2000 the European Commission argued that a cut in its price support for sugar was untenable because it would reduce income for developing countries that export sugar to the EU under preferences (European Commission 2000). Second, the United States recognizes that its MFN tariff reductions hinder its ability to extract concessions in terms of enforcing labor, environmental standards, and the like from countries that export under a preference to the United States (Mendelowitz 1994). 1. See also IMP (2004a). According to the wro's director-general, the IMF'S program is "a welcome contribution to the Doha Round, in particular to attaining ambitious market access results"(Panitchpakdi 2004). For an interesting discussion, see Winters (2004). The possibility that preference erosion would reduce ~m liberalization was actually noted long ago; it was a concern voiced by opponents of the Generalized System of Preferenceswhen it was originally proposed (Johnson 1967). 2. This points to the importance of further research and dissemination of basic facts about gains from preferences. There is some debate regarding the effectiveness of preferences in generating additional exports. Haveman and Shatz (2004)provide evidence that the unilateral preferences of the triad econo- mies (the European Union (EU),Japan, and the United States) considerably increased their imports from least developed countries in 2000. Earlier estimates of the Generalized System of Preferencesscheme are provided by Sapir (1981)for the EU and by Sapir and Ludenberg (1984) for the United States. UNCTAD (2003),Hoekman, Michalopoulos, and Winters (2004),and Stevens and Kennan (2004)propose several recommendations to fully realize the net benefits of preferencesfor least developed countries, including ways to improve utilization rates, to relax and harmonize rules of origin, to increase the predictability of preferences by binding them in the wro, and so on. Many of these recommendations also apply to the alternative of import subsidies proposed here. Limiio and Olurreaga 219 Limso (2002)provides a model in which preferencesthat are extended to small countries can cause large countries to maintain higher MFN tariffs. This stumblling block effect arises because of an important feature of EU and U.S. preference schemes: they require cooperation in nontrade issues such as labor and envilron- mental standards, intellectual property protection, drug enforcement, irnmi;gra- tion, and human rights. The appendix contains references to these and other side conditions in various preferential trade agreements. The preferential margins extended to small countries are often a payment for cooperation, which implies that a reduction in MFN tariffs that lowers the preferential margin will be resisted by both the country that receives preferences and the country that grants The following example illustrates the intuition behind the problem and the solution explored here. The U.S. MFN tariff on flowers is 10 percent, but Colom- bia can export them to the United States at a preferential tariff equal to zero- that is, it has a 10 percent preference margin in the United States. If the United States set its MFN tariff on flowers equal to zero in the Doha Round, it would be unable to provide a preference margin to Colombia to extract cooperation on nontrade issues. Thus, the model predicts that the United States will not reduce this MFN tariff to zero. There is a simple solution to this problem: allow a preferential import subsidy set at a fixed rate, for example, at the level of the current preference margin, 10 percent. If the U.S. charges the MFN tariff on Colombian flowers but pays a fixed subsidy rate of 10 percent to the Colombian producers, all participants are initially indifferent. But the United States coluld then maintain the fixed subsidy rate and reduce its MFN tariff without the opposition that results from preference erosion and the U.S. desire to extract cooperation through preferences. In this article the welfare costs of preferencesin terms of lost MTL are calcula.ted using the recent estimates of the stumbling block effect generated by U.S. prefer- encesfrom LimZo (2006).Similar estimates are obtained for the EU by Karacaovali and Limso (2005).These estimates are combined with data for 170 countries and approximately 5,000 goods to provide the first welfare estimates of preferential trade agreements as a stumbling block. The welfare effects are calculated from a baselineMFN tariff reduction of 33 percentexpected in the Doha Round relative to the subsidy counterfactual, which entailsadditional reductions by the triad econo- mies (theEU, Japan, and the United States).Given the results in Limso (2006),this additional liberalizationis estimated to be about 8 percentage points for good:$in which the triad economies offer preferences. The focus here is on the preferences that the triad economies extend to least developed countries: a subset of preferential trade agreements that fit the assumptions of the theory well. Total annual welfare gains for the 48 least developed countries are estimated at $520 million, with all countries but one 3. The fear that preferential trade agreements could be a stumbling block to MTL has generated a considerable theoretical literature but no consensus.See Bagwelland Staiger (1998),Krishna (1998),and Levy (1997).Winters (1999)provides an excellent review. 220 T H E W O R L D B A N K E C O N O M I C REVIEW, V O L . 20, N O . 2 gaining. The maximum gain is 6.7 percent of DP, and the average is 0.38 G percent. More important, the countries that most oppose MTL because of its preference erosion effect are estimated to gain the most from the switch. Welfare for the triad economies would increase by nearly $3 billion a year, mainly because of the additional MFN tariff reductions allowed by the subsidy. For the same reason, the rest of the world experiences an annual welfare gain of $900 million, which is important since one concern with preferential trade agreements is their costs on outsiders (Chang and Winters 2002). In relative terms, least developed countries as a group gain the most-over 0.5 percent of GDPa year. Similar gains arise for the least developed countries even if the subsidy causes only 2.8 percentage points of extra liberalization. Both the triad economies and the rest of the world continue to gain under this last scenario. The aggregate annual welfare gain of $4,354 million adds almost 10 percent to the gains that the model predicts from goods liberalization in the Doha Round without switching to subsidies-that is, relative to the baseline reduction of 33 percent. Moreover, this estimate is likely to be a lower bound of the stumbling block effect of preferential trade agreements. At one extreme, the concern with preference erosion could prevent the completion of the round, costing an addi- tional $47 billion. Section I discusses recent theoretical and empirical evidence on how preferential treatment to small countries can generate a stumbling block and how it can be addressed by an import subsidy. In Section I1 the welfare effects are calculated. Section I11 addresses issues related to the implementa- tion of the import subsidy. Section IV discusses the results. The appendix provides details on the theoretical model and discusses side conditions in preferential trade agreements. The supplemental appendix (available at http://wber.oxfordjournals.org/) provides details regarding the trade model used to simulate the effects, discusses data sources, and provides some descriptive statistics. I. PREFERENCES AS A STUMBLING BLOCK TO MTL Preferential trade agreements can affect MFN tariffs through several channels. They can divert scarce negotiation resourcesand alter the number of negotiating parties and their bargaining power. In the context of unilateral preferences analyzed here, these effects are not as relevant as the concern with preference erosion. That concern has created an important stumbling block to MTL, as can be shown in a model where unilateral preferences are exchanged for cooperation in nontrade issues. The appendix providessuch a model, based on LimHo (2002), that shows two things. First, the unilateral preferences that large countries use can cause them to maintain higher multilateral tariffs even if those preferences are extended to countries that are small from a trade perspective. Second, an import subsidy resolves this problem. Limiio and Olarreaga 221 Three main criteria were used in choosing the counterfactual to analyze the cost of preferences as a stumbling block. It should remove the stumbling block, generate a Pareto improvement in the context of a well-defined model (sothat it can be expected to gain the support of w o members) and be simple enough to permit estimation and implementation. When the import subsidy is designed to mimic trade preferences, it fulfills these criteria.Eliminatingunilateral preferencesor replacingthem withdirecttransfersto reward cooperation would remove the stumbling block. However, eliminating preferences altogether is opposed by several countries so it is not as interesting a counterfactual. Cash transfers may not be the most efficient way to transfer resources to other countries, as the aid versus trade literature highlights, or to reward their c~o~eration.~Political economy constraints that reduce the effectlive- ness of cash transfers relative to preferences are present in practice; otherwise it would be difficult to explain the existence of many preference schemes. For example, preferences may trigger investment in a particular sector and create a longer-lasting constituency when that country lobbies its government to do what- ever is necessary to maintain the preferences. Lump-sum transferscould end up in anonymous bank accounts in Switzerland.So, as an alternativeto preferences, the subsidy scheme may also dominate direct transfers. Before the welfare costs of preferences are estimated, some direct evidence for preferentialtrade agreementsas a stumbling block to ~ nis.provided that is subse- quently used in the quantification. The model generates specific testable predic- tions. At an extreme, if the MI.T\Ttariff is zero, no tariff preference can be offered and so no preferential agreement is possible without subsidies. Therefore, to the extent that the EU and the United States value the cooperation in the side conditions of those agreements, reductionsin MFN tariffs on products imported from the prefer- ential partner are more costly than reductionsin tariffs on other products. LimZo (2006)estimates the effect of U.S. preferential trade agreements on its MFN tariffs by exploring whether the effect occurs only with goods that ithe United States imports from its preferential trade agreement partners and not with goods that it imports only from the rest of the world. Using tariff data for more than 5,000 products, he finds that U.S. preferential trade agreements generated a stumbling block in the Uruguay Round. He estimates that the U1.S. average reduction in MFN tariff for goods exported by any of its preferential trade agreement partners was about half that for other goods. The effect is stronger for products that are exported under all preferential trade agreemeints or that constitute larger shares of a given preferential trade agreement partner's exports to the United States. These estimates control for several determinants of U.S. tariff changes and carefully establish the direction of causality from the 4. Thus, the proposal here differs from the others that suggest temporary development assistance to the governments of developing countries that are hurt by preference erosion-for example, Hoeknnan (2004).See McCulloch and Pinera (1977)on the issue of trade and aid issue and Adam and 0'Con:nell (2004)for a more recent analysis. 222 THE W O R L D BANK E C O N O M I C REVIEW, VOL. 20, N O . 2 preferential trade agreements to the MFN tariff changes using instrumental variables. A switch by the triad economies to a subsidy after the Doha Round would cause them to reduce their tariffs by an additional 8.3 percentagepoints on the goodsfor whichtheyprovidepreferencestoleastdevelopedcountries.Thisresultisobtainedby assuming a 33 percentMFN tariff liberalization in Doha on all goods-the average for developed countries in the Uruguay Round-and dividing it by a stumbling block factor of 80 percent calculated from LimBo (2006). According to his esti- mates, the reductionin MFN tariffsfor goods that the United States imports from its Generalized System of Preferencesbeneficiarieswas only about 80 percent of what it would have been without such preferences.Thus under the subsidy scheme, triad economies' liberalization in goods where they provide a preference to least devel- oped countries is estimated to be 41.3 percent, an additional 8.3 percentagepoints relative to the 33 percent baseline liberalization. The sensitivity of the welfare results is analyzed with respect to this e~timate.~ Karacaovali and LimHo (2005) provide similar estimates for the stumbling block effect that apply to the EU'S preferential trade agreements. The EU average MFN tariff cut for goods exported by any of its preferential trade agreements was about half that for other goods. They also provide evidence that directly supports the use of an import subsidy as a solution when preferential tariffs are close to zero. Their model predicts that if the preferential tariff is positive, the MFN tariff can still be reduced without affecting the preference margin simply by lowering the preferential tariff by the same amount. This prediction is tested and confirmed for the EU, strongly supporting the argument that by removing the non-negativity constraint for preferential tariffs an import subsidy would indeed eliminate the stumbling block effect. 11. WELFARE ESTIMATES OF PREFERENCES AS A STUMBLING BLOCK TO MTL This section first describes the methodology used to quantify the welfarecosts of preferences as measured by the gainsfrom switching to an import subsidy. Once 5. More specifically,the stumbling block factor isdefined asf = (Atp/tO)/(Ats/tO),where Atp/to is the growth in the MFN tariffs of a triad economy when preferences are in place and At,/to is the growth when they are absent. Lirn5o (2006)takes the changes in tariffs on the nonpreferential trade agreement good as the counterfactual for the outcome in preferential trade agreement goods if preferential trade agreements were absent and calculates Aln(1 +tp)/Aln(l +t,) w Atp/Ats = f. Using his estimates leads to A ln(1 + tp)/Aln(1+t,) (i+p* TOTLIB+c$)/(ii+p* TOTLIB)= = 0.8. The variable TOTLIB represents the liberalization by U.S. parmers, so p captures the reciprocity effect; TOTLIB was 51.3 on average in his sample, so p * TOTLIB = 0.018 * (-51.3). The estimates for p = 0.018 and 4 = 0.658 are from the last column of table1in LimZo (2006),where he controls for the existenceof other preferential trade agreements. Using his estimate for the Generalized System of Preferences in table 2, 0.74, yields ri = -2.53 = 0.658/(0.74 - 1). The 95 percent confidence interval for the stumbling block factor described above is (0.697, 0.923), which is then used to calculate the high and low scenarios of additional liberalization of 14.3 percentage points and 2.8 percentagepoints in table 2. Limiio and Olarreaga 223 the computed measures are conceptually clear, the estimation is standard. (The formal details of the estimation can be found in the supplemental appendix.) The empirical results are then discussed. Methodology First, the budgetary cost for each subsidy-granting country (SGC) is calculated. This is the amount paid to each subsidy-receiving country (SRC),which is an input into the welfare calculation and a measure that may preempt concerns with the budgetary costs of implementation. An import subsidy rate for each good and each SGC must be chosen. Here, this rate is the current preference margin, which provides a convenient benchmark to calculate welfare changes because the initial switch leaves all prices and quantities unchanged. Naturally, alternative subsidy rates are possible and would deliver different results. The first two columns of table 1 list the cost of this subsidy for all goods under current levels of tariffs and trade and under those predicted after the Doha Round, with MFN tariffs assumed to be 33 percent lower. The key budgetary measure is the cost of switching to the subsidy, which is obtained by subtract- ing the forgone tariff revenue under the current preferences (see column 3 of table 1). TA BLE 1. Budgetary Impact on Triad Economies of Switching to a Subsidy Scheme ($ Millions) - - - - - - - -- Cost of Subsidy Cost of Switch Preference Erosion Current Most Post-Doha Most Post-Doha Most Post-Doha Most Favored Nation Favored Nation Favored Nation Favored Nation Economy Tariffa ~ a r i f f Tariff ' ~ariff* European Union 669 674 208 Japan 31 32 6 United States 63 64 21 Total 763 768 235 Note: A decomposition of subsidies and preference erosion by subsidy-receiving countries is provided in appendix table A.1. "Calculated using a subsidy rate for each good equal to the current preference margins in each subsidy-granting country using current prices, quantities, and most favored nation tariffs; see equation (S.l) in the supplemental appendix. b~ssumesa 33 percent reduction in most favored nation tariffs by all World Trade Organization members; see equation (S.2) in the supplemental appendix. 'Assumes a 33 percent reduction in most favored nation tariffs by all World Trade Organization members; see equation (S.3)in the supplemental appendix. *uses current preferences and assumes a 33 percent reduction in most favored nation tariffs by all World Trade Organization members; see equation (S.12) in the supplemental appendix. Source: Authors' analysis based on data discussed in the supplemental appendix. Second, the preference erosion measure is calculated for each preference- receiving country in each preference-grantingcountry market. If current prefer- ences are maintained, this measure is simply the difference between the extra export revenue due to preferences before and after a 33 percent reduction in MFN tariffs without a subsidy. These values are listed in column 4 of table 1 and by country in appendix table A . I . ~ Third, and most important, the welfare cost of preferences is calculated by country (table2 and appendix table A.2).This is the additional welfare gain that would be obtained in the Doha Round if the triad economies switched to an import subsidy. To calculate all three measures, we need an estimate of the changes in world prices and quantities imported and exported associated with the tariff reductions in the 170 countries in the sample. This is calculated by combining estimates of import demand and export supply elasticitieswith a simple trade model, which is described in detail in the supplemental appendix. A simple partial equilibrium model is employed for each six-digit tariff line in the Harmonized System. It assumes that each tariff line includes a homogeneous good and that the world market for each is in equilibrium. Neither substitution effects across goods nor income effects are modeled on the demand or the supply side. Although such effects could be incorporated here, recent work suggests that they have little impact on aggregate welfare gains.7 The aggregate welfare estimate here is similar to those generated using more complex computable general equilibrium models.* Empirical Results The focus here is on the preferences granted by the triad economies to 48 of the 49 least developed countries under the Generalized System of Preferences (data are not available to compute welfare gains for Kiribati). While these are not the only small countries that the triad economies offer preferences to, they form an interesting subgroup. Data on products at the six-digit harmonized standard level that receive preferences from triad economies 6. These preferences are not fully utilized (Inama 2003),so it is likely to also be the case under the subsidy. Therefore, the estimates of the cost of the subsidy and preference erosion are likelyto be an upper bound for the true effects. However, this should not have a large effect on the net welfare calculations since the qualitative effect of imperfect utilization is similar under the subsidy and preference scenarios, and the net welfare effects are the difference between the two. 7. Hoekman, Nicita, and Olarreaga (2006)find that the world welfare gain from a 40 percent cut in MFN tariffs among wro members is $51 billion without cross-price effects and $59 billion with them. This difference is statistically insignificant when they account for the standard errors associated with the measurement of the elasticities of import demand and export supply. Thus, the exposition of the model in the supplemental appendix is kept simple and clear by abstracting from cross-price effects. 8. For example, assuming a 33 percent liberalization for all WTO members leads to a yearly gain of about $47 billion in our estimation. Francois, van Mejil, and van Tongeren (2005)estimate it to be $45 billion when they apply a 50 percent cut in bound tariffs using a static computable general equilibrium model with constant returns to scale and with an Armington setup. Limdo and Olarreaga 225 TABLE 2. Annual Welfare Cost of Least Developed Country Preferences as a Stumbling Block Economy Share of GDPa (%) Intermediate Low High Triad economies 0.013 2,934 1,048 5,045 European Union 0.027 2,336 849 4,023 Japan 0.006 237 79 415 United States 0.004 361 120 607 Least developed countriesC 0.518 520 513 526 Bangladesh 0.36 176 173 179 Cambodia 0.60 24 22 25 Lesotho 2.09 18 18 18 Madagascar 1.16 53 53 53 Malawi 6.66 117 117 3117 Maldives 0.47 3 3 3 Mauritania 0.61 6 6 6 Mozambique 0.25 9 9 10 Myanmar 1.23 20 19 21 Slo Tom6 and Principe 1.32 1 1 1 Senegal 0.33 17 17 17 Sierra Leone 0.65 5 5 5 Solomon Islands 0.67 2 2 2 Tanzania 0.32 30 30 30 Uganda 0.11 7 7 7 Others 33 31 32 Rest of the world 0.010 900 315 1,721 Total 0.010 4,354 1,888 7,2!70 - Note: See appendix table A.2for welfare estimates for other least developed countries. Estimates are based on equations (S.6) and (S.ll)in the supplemental appendix. aIntermediate welfare gains/^^^, where GDP is for 2002 and in current U.S. dollars. bAssumesa baseline reduction in most favored nation tariffs of 33 percent by all World Trade Organization members plus an additional amount by the triad economies in the products in which they provide preferences to least developed countries. The additional amount is 8.3 percentage points for intermediate, 2.8 percentage points for low, and 14.3 percentage points for high. These three figures are calculated using the estimates in Lirnlo (forthcoming); 8.3 percentage points is calculated from a point estimate, and the extremes represent the 95 percent confidence interval. See the text for details. 'The estimates for least developed countries are insensitiveto different scenarios because most of their welfare gain is due to the avoided preferenceerosion under the 33 percent baseline common to all three scenarios. Source: Authors' analysis based on data discussed in the supplemental appendix. and on bilateral trade flows and MFN tariffs for all countries are used to calculate changes in prices and quantities due to liberalization. The import and export elasticities at the tariff line for every country are from the article :by Kee, Nicita, and Olarreaga (2004).The exact data sources and years are in the supplemental appendix. To provide an idea of trade coverage, we should note that in 2001-03 imports from least developed countries totaled on average $8 billion for the EU, $1 billion for Japan, and $7 billion for the United States. According to UNCTAD (2003),99.8 percent of dutiable EU imports from least developed countries are covered by preferences, as are 53 percent of Japanese imports and 44 percent of U.S. imports. Among least developed countries, Bangladesh and Cambodia are the top beneficiaries of EU and Japanese tariff preferences, and Madagascar and Malawi are the top beneficiaries of U.S. preferences. The first column of table 1 lists that the cost of the subsidy for the triad economies is $763 million without tariff reductions. This is equal to the tariff revenue currently forgone due to preferences, and it is borne mostly by the EU for three reasons. First, the EU imports larger quantities from least developed coun- tries. Second, EU preferences to least developed countries tend to be more generous, as seen in the supplemental appendix table S.1. Third, the EU MFN tariffs on the type of goods exported by least developed countries tend to be higher. The second column of table 1 lists the amount of the subsidy after a 33 percent reduction in bound tariffs by all WTO members, $768 million. It is higher than the currently forgone tariff revenue because exports from SRCSincrease with the rise in world prices caused by the multilateral tariff reductions. The third column of table 1 lists the estimates of the budgetary cost to triad economies implied by switching to the subsidy scheme. This cost arises because the tariff charged on imports from s ~ c falls but the subsidy rate does not. s Moreover, this shortfall applies to a larger value of imports because the MFN tariff reductions increase world prices. The total budgetary cost for the triad economies is only $235 million, and the share borne by Japan and the United States is minimal. This value excludes the impact of any additional MFN tariff reduction that may be possible under the subsidy. The fourth column of table 1lists the preference erosion costs that preference- receiving countries suffer in each triad economy market under a 33 percent reduction in MFN tariffs without a subsidy. These erosion costs would be much larger than the budgetary costs for the triad economies to move toward a , subsidy. The least developed countries' loss is $624 million, mostly in the EU market. Previous estimates were around $530 million (for example, Subrama- nian 2004). There are two main reasons for the larger results here. First, the average estimated elasticity of export supply in least developed countries is 5, which is larger than the value assumed by Subramanian for all goods, 1. Second, the MFN tariffs here include ad valorem equivalents of specific tariffs, which leads to larger preference erosion. Lima0 and Olarreaga 227 Estimates bycountryaresummarizedinappendixtable A.1. They are calculated to help identify the countries that are, or should be, most opposed to MTL based on its impact on preference erosion. If these countries gain considerably by a switch to a subsidy, a large obstacle to liberalization may be removed. The three countries that would face the largest losses in absolute terms due to erosion are Bangladesh ($202 million), Malawi ($151 million), and Madagascar ($63 rnil- lion). The countries that face the largest losses in share of DP are Malawi (8.6 G percent), Lesotho (2.7 percent), and SBo Tom6 and Principe (1.6 percent). The results also show that losses from preference erosion are concentrated, with 26 of 48 countries losing less than 0.1 percent of DP. G The triad economies should not be opposed to switching to a subsidy simply because of the budgetary cost; switching would also allow further liberalization. table 2 summarizes estimates of the net welfare impact, including this extra liberalization by the triad economies on products subject to least developed country preferences.The focus here is on the intermediate estimates in the first two columns, which are based on an estimated additional 8.3 percentage points of liberalization; the sensitivityof the results to alternative estimates is discussed below. The annual gain for the triad economies is $2,934 million, meaning that their governments should be in favor of switching. The largest gain, $2,336 million, is for the EU; Japan would gain $237 million and the United States $361 milli~n.~ The least developed countries also benefit from switching to the subsidy, with a total gain of $520 million a year. The change in welfare is much larger for least developed countries when measured relative to GDP -about50timeslargerthan that for the triad economies. There are 27 least developedcountries that will see net welfare changes of less than 0.1 percent of GDP. And Djibouti experiences a marginal loss of $0.07 million due to deteriorating terms of trade a~ssociated with liberalization. Table 2 also lists the least developed countries that are among the top 10 countries that gain the most either in absolute value or as a share of GDP. These countries are also among those that gain the most per capita. The other 33 countries together gain only $33 million. The largest absolute gains for individual least developed countries are for Bangladesh ($176 million), Malawi ($117 million), and Madagascar ($53 million). The largest gains in share of GDP are for Malawi (6.7 percent), Lesotho (2.1 percent), and Siio Tom6 and Principe (1.3 percent). Recall that these were the same countries that stood to lose the most from preference erosion. So if faced with a choice between the preference and the subsidy scheme, least devel- oped countries should support the subsidy scheme and the additional MTL that it entails. 9. The results do not appear to be driven by specific products such as sugar, which, as discussed in the introduction, provides an important example of the effects discussed here. Excluding sugar, the welhre gain for the European is only $72 million less. The aggregate welfare gain for least developed countries is lower than the preference erosion value in table 1 because preference erosion measures the change in export revenues and ignores any extra costs of production. For individual countries, the gains in table 2 may also be lower than the preference erosion effect in appendix table A.2 because additional MTL by the triad economies occurs in many goods, some of which the SRC may import, imply- ing that it will face a deterioration in its terms of trade, for example, Djibouti and Rwanda. The rest of the world gains $900 million: 108 countries experience a gain and 11 a loss. But the largest loss, which is for Nigeria, is only $0.6 million. By contrast, gains are as high as $167 million for China. The gains for the rest of the world occur as a result of the additional liberalization from switching to the subsidy.10 The aggregate welfare gain of $4,354 million adds nearly 10 percent to the gains that the model predicts from goods liberalization in the Doha Round without switching to a subsidy. Nonetheless, this is likely to be a low estimate of the cost of preferential trade agreementsas a stumbling block for MTL because the focus here is on the preferences to least developed countries. The estimates in LimZo (2006)suggest that the stumbling block effect is greater when it applies to several preferential trade agreements; if the additional liberalization was about 14 percentage points, the aggregate gains increase to more than $7 billion, as shown in the last column of table 2. Even under the 8.3 percentage points scenario, the welfare estimates are likely to be a lower bound, for several reasons. First, they refer to an annual effect. With the static model used here, if there were no further shocks after the implementation of the subsidy scheme, the discounted welfare effect would be several times higher. Second, to calculate the welfare effects, we employed a static perfect competition model with no externalities, which is well known to provide relatively small gains from trade liberalization. Third, when the effect of the additional liberalization by the triad economies is computed, reciprocity effects are not added. In practice, the triad economies would be expected to attempt to negotiate further reductions from other countries.'' Since the additional liberalization of 8.3 percentage points is an estimate, the sensitivity of the results was tested by considering two extreme alter- natives that represent its confidence interval: 2.8 percentage points and 14.3 percentage points. Two points stand out. First, most of the change in the aggregate gains, which now range from $2 billion to $7 billion, is due to the 10. For non-least developed countries, the estimates do not include the preference erosion effect caused by the extra 8.3 percent liberalization in the triad economy markets. 11. Underthe preferencescheme the MFN tariff reductions would result in a lower level of cooperation in nontrade issues by the small countries, which would entail an additional cost to the triad economies that is not computed here. Limrio and Olarreaga 229 triad economies-which is not surprising since they are the countries liberal- izing.12 Second, the estimates for the least developed countries remain nearly unchanged because most of their gains arise from the avoided preference erosion under the 33 percent baseline reduction common to all three scenarilos. Therefore, from the perspective of least developed countries, the proposal here should be attractive even if the amount of additional liberalization by the triad economies is very small.13 We analyzed the small budgetary cost of switching to the subsidy. The logistilcal costs of running it are similar to those for preferences because many of the procedures are already in place. However, we now discuss some potential differences in running each scheme and possible refinements to the subsidy proposal. When the value of the subsidy and that of tariff rate are identical, they have equivalent effects on prices, quantities, and implementation. When the prefer- ential tariff is zero, the exporter receives the domestic price in the preference- granting country, pw +t, where pw is the world price and t is the MFN tariff, and pw +t is also the price that the importer pays. Under the subsidy, the preferential exporter receives pw +s, the same as before when t = s. The importer pays pw +t+s before receiving a rebate equivalent to the amount of the subsidy per unit, s. If this occurs at customs, the buyer's price is simply pw + t. Here, the subsidy is implemented as an immediate drawback scheme, which already exists in the triad economies to rebate import duties when t:he 12. The triad economies' total welfare change can be decomposed into an efficiency and a budgetary effect. The budgetary effect has two components: the first captures the cost of the subsidy relative to the preference under a 33 percent m tariff reduction (equation S.5 in the supplemental appendix), and the second captures the extra cost of the subsidy due to the additional liberalization (the term in the last parenthesis of equation S.8 in the supplemental appendix). The efficiency component is given by the remaining terms on the right side of equation (S.8). Under the intermediate scenario of an additional 8.3 percentage points, the first component of the budgetary cost accounts for 8 percent of the $2,934 million welfare change of the triad economies. This increases to 14 percent for the total budgetary cost (14percent for the EU, 5 percent for Japan, and 24 percent for the United States). Thus, most of the change in their welfare is driven by the efficiency effect. 13. To understand the insensitivity in the welfare effect for the least developed countries, note that it can be decomposed in two terms. As seen from equation (S.ll) in the supplemental appendix, the first effect arises from the additional liberalization that occurs only under the subsidy, given by equation (510).The second effect is given by the second term in equation (S.11),which captures the difference in welfare changes under the subsidy versus the preference when tariffs fall by 33 percent. Under the preferences, a reduction in m tariffs causes a reduction in the welfare for the least developed coun~ries because of erosion, as summarized in table 1. Thus, the second effect captures the role of the subsidy in avoiding preference erosion. In the estimates here this last effect dominates the effect from additional ~n by the triad economies, which accounts on average for only about 2 percent of the total welfare change for individual least developed countries. But there is some heterogeneity across countries in this, and for a few least developed countries, the effect from additional liberalization by the triad economies is actually negative because the additional ~noccurs in many goods, some of which the s ~may import. c imported good is re-exported or used as an input in the production of an exported good. The discussion above should also make clear that the subsidy is as transpar- ent as the current preferences. Therefore, political economy arguments that the subsidy is more transparent, and thus less likely to be adopted, are not convincing.14 But the subsidy may become an issue after an MFN tariff reduction, that is, when t'c s. Two differences arise when the MFN tariff rate falls. The first is the direct budgetary cost: customs collects t' per unit and must rebate s. Estimates show this cost is small. However, when the subsidies involve expenditure above the tariff collected on the product from that country, they may need to be in the budget. This could be financed directly from tariff revenue collected on imports from other countries. Moreover, from the domestic legal perspective of the SGC, the budget implications of subsidies are no different from those of the preferences.1s The second issue that may arise when t'c s is re-exports. The buyer in the SGC can ship the good back to an agent in the SRC that will re-export it to collect the subsidy again. This is profitable if the excess subsidy, s - t', is high enough to offset the transaction cost, which includes the cost of forging the certificate of origin of the good and the two-way transport cost. For most goods, though, transport costs make re-exporting unprofitable, particularly when the origin country is a least developed country, where transport costs tend to be higher (LimBo and Venables 2001). If interest forgone on the sale of the good because of the time in shipment is added, re-exports are very unlikely to be profitable. If re-exports became a problem, one solution would be to implement the subsidy rebate only when the good is sold to the final consumer in the SGC. Since internal sales often incur a sales tax, this would serve as an additional hurdle that would have to be overcome for re-exports to be profitable. In fact, exempting imports from least developed countries from 14. One political problem that governments in SGCS may face is justifying to the import competing producers why they are subsidizing a foreign producer. This is not an issue before the MFN tariff reduction since t = s and so the subsidy is offset by the tariff. But it may be an issue after an MFN reduction, that is, when t'e s. This should not be an issue if the MFN reduction was given since theSRCS we consider are small and so the domestic price in the SGC would remain almost unchanged for a given reduction in the MFN tariff under the preference or subsidy. So import competing producers would only oppose the subsidy if indeed the subsidy led to additional MFN liberalization. But this opposition is exactly identical to the import competing sector's opposition to MFN liberalization that arises because of other motives. The countervailing force here is therefore the same: exporters in each of the triad countries would support the subsidy precisely because it allows additional MTL in the remaining triad markets to which they export. 15. In the United States, for example, any foregone tariff revenue from a new preference must be estimated and a replacement in the budget suggested. There are also precedents for the use of import subsidies in the EU-for example, when supplies of sugar available within the Community or a major consuming region are no longer sufficient (Council Regulation 126012001,June 19, 2001). Limdo and Olarreaga 2:31 paying such a tax could, under certain conditions, mimic the subsidy and thus be an alternative.16 This proposal may apply better to certain countries that are "strategically large," that is, whose cooperation in certain nontrade issues is importa~nt. Proposals for a "buy-out" of preferences, either through a lump-sum payment or temporary adjustment assistance through grants (Hoekman 2004) or loans (IMF 2004a and 2004b),are unlikely to work for those countries.17For example, the United States had the opportunity to offer only cash for Colombia's coopera- tion on the war on drugs, and it chose not to. However, the skeptics of the subsidy scheme may want to first experiment by switching only in one sector or a subset of one sector. For example, there may be a stronger case for U.S. plre- ferences to Colombia in the agricultural sector as a way to provide alternatives to growing coca. Moreover, sector-specificexperimentation can show whether unexpected implementation problems may arise. Choosing a fixed subsidy rate equal to the current preference margin yields a convenient benchmark to evaluate the welfare effects of a switch. But alternative subsidy rates can help address other problems with preferences. Preferences may have led to investment in sectors where countries do not have a comparative advantage (Tangermann 2002). A modified subsidy scheme would address this. A fixed subsidy rate that is equal for all products exported to all SGCS could be used to align the pattern of export specialization with relative world prices. This would keep relative prices in the SRC equal to world relative prices. IV. CONCLUSION The recent wave of preferential trade agreementssuggests that policymakers find them desirable. But their effect on multilateral liberalization must be estimated and taken seriously. This article describes recent theoretical and empirical evi- dence that shows how preferential trade agreements can slow multilateral liber- alization even when the countries receiving the preference are small. It argues that import subsidiescan be used to decouple preferences from MFN tariff liberal- ization and that subsidies would be supported by the countries receiving aind granting preferences. 16. A sufficient condition for this exemption to work exactly in the same way as a subsidy rate of value s is for the existing sales tax rate to be at least as high as the subsidy rate. This condition is not satisfied for some goods with high preference margins (which were chosen as the initial subsidy rate). 17. The w ' s proposed program is in the form of loans to finance temporary adjustment costs if the MFN tariff reductions have a significant effect on the balance of payments in the beneficiarycountries. Even if this program compensated for the full amount of preference erosion, it would not be able to deliver -the same MFN tariff reduction as the subsidy proposed because after the MFN tariff reduction the granting countries can only threaten to remove a smaller amount of preferences. Thus, they will be able to extract less in the form of side conditions than before the MFN reduction (orsubsidy scheme). This implies that lthe granting countries themselves will offer fewer MEN reductions under the LMF program than under the subsidy scheme. We provide the first estimates of the welfare costs of preferential trade agreements as a stumbling block to multilateral liberalization. It is $2,934 million per year for the TRIAD if they switched their preferences to least developed countries into fixed import subsidies in the context of the liberal- ization anticipated in the Doha Round. This occurs because under the subsidy plan triad economiescan further reduce their MFN tariffs. The switch would also increase annual welfare in the rest of the world by $900 million. For least developed countries receiving preferences from the triad economies, the gain is $520 million-due mostly to the preferenceerosion avoided under the subsidy. The aggregate annual welfare estimate of more than $4 billion adds about 10 percent to the gains that the model predicts from goods liberalization in the Doha Round. This is likely to be a lower bound of the stumbling block effect of preferential trade agreements. At one extreme, the concern with preference erosion could prevent the completion of the round, costing an additional $47 billion a year. In terms of implementation, the budgetary costs of the subsidy are small, and the logistical costs are in principle not significantly higher than those associated with the current preferences. However, two issues must be addressed: legality in the WTO and the scope of the subsidies. From a legal perspective the subsidies conflict with the MFN principle; but this could be addressed in the same way as preferences: through waivers, the enabling clause, or article XXIV, for example. In termsof thescopeof the subsidy, the questionis whether the proposalshould extend to all countries that receive unilateral preferences. The basic argument should apply to them as well. Naturally, the budgetary cost for the triad econo- mies would be higher, but the additional MFN tariff liberalizationthat would result may be enough to offset it. Even more broadly, extending the scheme to members of all preferential trade agreements could be considered. The concern with pre- ference erosion also applies to them, and there is evidence that they have slowed U.S. and EU MFN tariff liberalization. Future research should address this broader question and calculate the welfare costs of other preferential trade agreements. First a simplified version of the model in Lim3o (2002)is presented to show how unilateral preferences to small countries can cause economiessuch as the EU and the United States to maintain higher multilateral tariffs and how this incentive disappears when an import subsidy is allowed. The role of side conditions in preferential trade agreements is then discussed. Preferences to Small Countries as a Stumbling Block to MTL Assume each regional bloc contains a large and a small country, denoted by L and S. Two externalities exist within each bloc. First, L is affected by the level of Livniio and Olarreaga 233 an action e that S can undertake at a cost. This is a general way to capture Ithe demand that L has for cooperation in labor, environmental, immigration issues, improvements in governance, and so on. Second, there is a terms of trade externality, that is, Lcan use a tariff to depress the price of S's exports. Countries in a bloc can internalize these effects through a preferential trade agreement where L lowers its tariffs on S's exports in exchange for an increase in S's provision of e. The preferential trade agreement is modeled so that its only direct trade effect is to increase the price that S receives for its exports. There are two import tariffs that L chooses: t, the multilateral tariff that it applies to the rest of the world, and t p 5 t, the preferential tariff on that good applied to imports from S. The good that L exports to the rest of the world faces a tariff tr.So the objective function that L maximizes is where the partial effects are W: > 0 and, due to the terms of trade effects, W,L > 0 for x = tP,t when evaluated at x = 0 and W: < 0. The objective maximized by S depends on e and the tariffs set by L, which affect the price that S receives for its exports. ( A 4 wS= wS(e,tp,t). Under a preferential trade agreement Sreceives an export price of pW(t) + t - tp, where pw(t) is the equilibrium world price, so pW(t) t is the price in L. There- + fore, wfPc 0 and ~f > 0. But in the absenceof a preferentialtrade agreement, S faces the multilateral tariff and receives only pW(t), which falls when t is raised and so W: c 0. For simplicity, it is assumed that S has no trade in the good exported by L to the rest of the world and thus is indifferentto the level oft'. The balance of payments condition is satisfied through a numeraire good that enters utilities in a quasi-linear way and that L uses to pay for its imports from S. The crucial assumption that generates a motive for L's preferential treatment for Sis that it values e. Take the neutral case where S places neither a positive nor a negative weight on the direct effect of e on itself and assume that this action requires some expenditure by S, which implies a negative margiinal benefit in terms of the numeraire. So, if S and L do not cooperate, S does not supply e and L does not provide a preference. They can improve on this outcome through a bargain where L sets tp < t and Ssupplies e.18 18. Note that for a given t, this bargain may be just as efficient as a lump-sum transfer frorn L assuming that preferencesare quasi-linear so that the tariff revenue that L gives up through the has the same effect as an equivalent lump-sum transfer. This would not be the case if S had an upward sloping export supply as assumed in the simulation. However, there can be realistic political economy constraints that would deliver the preference instead of the lump-sum transfer as the constrained first-lbest policy to be used in exchange for e; here, it is simply assumed that the preference is the only available instrument, and the implications for the multilateral tariff are then analyzed. In most of the agreements analyzed here L has nearly all the bargaining power relativeto S. Therefore, it is sensibleto focus on the outcome of a take-it-or-leave-it offer that leaves S at the status quo welfare level. If S is a WTO member, the maximum tariff that L can set is tP = t, due to the MFN rule. Therefore, the status quo welfare will be determined by evaluating tp at t and e at the level that is optimal for S. More specifically, the bargain that L offers must at least satisfy WS(e = eb, tP = tb c t,t) > Ws(e= 0,tP = t,t). SO for a given level of t and tP, the equilibriumlevel of eb can be written as a function eb(tp,t) that is decreasing in the preferential tariff and increasing in the MFN tariff because either movement raises wSand thus allows L to extract a higher level of e. In the simple case the focus is on the fact that net exports of Sare constant, implying that, in the function eb(tp,t), S cares only about the preferential margin t - tp, so e: = -e:p > 0. WTO countries negotiate reciprocal tariff reductionswith their principal suppli- ers. So if the rest of the world (ROW)is the main supplier of the good that L also imports from S, L negotiates with ROW. To capture this, the equilibrium multi- lateral tariffs are modeled as the solution that maximizes the joint objective of L and ROW. Moreover, it is assumed that L chooses the preferential tariff simulta- neously and that ROW is a mirror image of L, although neither of these assump- tions is essential for the result. This implies that the focus can be on solving for t, since it will be equal to tr, and that the effect oft on Wr is identical to that of tr on wL.Imposing the equilibrium conditions of symmetry, t = tr, and e = eb(tp,t) the joint optimum for L and ROW under a preferential trade agreement is given by the following program and necessary first-order conditions (A.3) (2, 8')= arg,,, max wL(e= eb(tp,t),tp,t,tr = t) To see how a stumbling block can arise when import subsidies are not allowed, let us constrast this with the condition for the tariff in the absence of a preferential trade agreement. Now L has no incentive to provide a - preference to S, so t = tP and e = 0, which yields the following solution: (A.6) t' arg, max WL(e = 0,tP= t,t, P = t). Using the first-order condition derived for this last problem, Wb + Wf;+ Wk = 0, to evaluate the first-order condition for t when a prefer- ential trade agreement is present (equation A.5), it can be determined whether 't > t'. That will be so if the following expression is positive: L i d o and Olarreaga 2:35 where the inequality in the first line reflects the use of equation (A.4). If import subsidies were allowed, t P could continue to be lowered below zero and the inequality above would disappear since an interior solution could be found. But otherwise it is possible to obtain a corner solution in the preferential tralde agreement, that is a situation where, at tP = 0, L would like S to increase e. In this case, equation (A.4) holds with a strict inequality. The second line of equation (A.7) is zero in the case considered here, where -e,p = ek-that is, when an increase in the MFN tariff has the same effect as a decrease in the preferential tariff, both simply increase the preferential margin. This implies that Z > t' if [w; + ~ ~ e< 0.~So ~import subsidies completely eliminate ] ~ ~ = ~ the need to distort the MFN tariff to maintain a preference margin. Side Conditions in Preferential Agreements The Generalized System of Preferencesand other unilateral preferences provided by the EU and United States often have side conditions attached that are valued by the preference-granting country and potentially costly to the recipient. That is, these unilateral preferences are not free to developingcountries. The Generalized System of Preferences, for example, was designed to promote the development:of poorer countries based on an infant industry argument (UNCTAD 1964),but "dur- ing the last twenty-fiveyears or so the experience of the GSP in the GATT system has been that for a number of reasons the preference-granting national entities (that:is, the industrializedcountries)often succumb to the temptation to use the preference systems as part of 'bargaining chips' of diplomacy" (Jackson 1997, p. 16#0). International trade lawyers have also warned of the possibility that the General- ized System of Preferenceswill retard the MTL of the developedcountriesrhat grant such preferences because of a similar mechanism (Trebilcockand Howse 199'9). Failure of Generalized System of Preferences beneficiariesto comply with some of the nontrade issues has cost them their preferentialaccess to the United states.19 19. For example, according to USTR (2005),some countries have lost eligibility for trade preferences under Generalized System of Preferences because of worker rights or intellectual property concerns. In 2005 the United States revoked C6te dYIvoire'seligibility for trade preferences under the African Growth and Opportunity Act because it failed to comply with the U.N. cease fire resolution. Morocco enactted a comprehensive new labor law recently, and according to U.S. trade negotiators, it was "the prospect of a free trade agreement with the United States [that] helped to forge a domestic consensus for labor law reform in Morocco, spurring reform efforts that had been stymied for more than 20 years" (USTR2004). Both the EU and the United States explicitly offer reductions in trade barriers in exchange for cooperation on various nontrade issues such as labor, environ- ment, drug trafficking, and intellectual property protection. Examples of these have included the Eastern European and Mediterranean agreements signed by the EU; the U.S. agreements with Jordan, Mexico, and other Latin American and Caribbean countries; and the preferential treatment that the EU and the United States extend to most developing countries through Generalized System of Pre- ferences. This type of agreement is increasingly prevalent, as the new U.S. preferences to Middle Eastern countries make clear.20 TABLE A.1. PreferenceErosion and Subsidy by BeneficiaryCountry ($ Millions) European Union Japan United States Subsidy-Receiving Preference Preference Preference Country Erosion Subsidy Erosion Subsidy Erosion Subsidy Afghanistan Angola Bangladesh Benin Bhutan Burkina Faso Burundi Cambodia Cape Verde Central African Republic Chad Comoros Congo Djibouti Equatorial Guinea Eritrea Ethiopia Gambia, The Guinea Guinea-Bissau (Continued) 20. See USITC (1994,1996)for conditions applying to the AndeanTrade Preference Act. See Bayard and Elliot (1994)and LINCTAD (2000)for details on conditionality in the GeneralizedSystem of Preferences program. See Perroni and Whalley (2000)for details on conditionality in the North American Free Trade Agreement. See Winters (1993)for details on the EU'SEastern European, Mediterranean, and Generalized System of Preferences programs; the Generalized System of Preference programs are also described in UNCTAD (2002). Limdo and Olarreaga 237 TABLE A.1. Continued European Union Japan United States - Subsidy-Receiving Preference Preference Preference Country Erosion Subsidy Erosion Subsidy Erosion Subsidy Haiti Kiribati Lao PDR Lesotho Liberia Madagascar Malawi Maldives Mali Mauritania Mozambique Myanmar Nepal Niger Rwanda Samoa SHo Tom6 and Principe Senegal Sierra Leone Solomon Islands Somalia Sudan Tanzania Togo Tuvalu Uganda Vanuatu Yemen Zambia n.a., not available. Source: Authors' calculations based on data discussed in the supplemental appendix. Tab1e A.2. Changes in Net Welfare by Least Developed Countries Subsidy-ReceivingCountry Total($ Millions) Per Capita($) Share ofGDP~(%) Afghanistan Angola Bangladesh Benin Bhutan Burkina Faso TAB LE A.2. Continued Subsidy-ReceivingCountry Total($ Millions) Per Capita($) Share of GDP~(%) Burundi Cambodia Cape verdeh Central African Republic Chad comorosb Congo, Rep. Djibouti Equatorial Guinea Eritrea Ethiopia Gambia, he' Guinea Guinea-Bissau ~ a i t i ~ Kiribati Lao PDR ~ e s o t h o ~ Liberia Madagascar Malawi Maldives Mali Mauritania Mozambique Myanmar Nepal Niger Rwanda samoah SZo Tomt and Principe Senegal Sierra Leone Solomon Islands Somalia Sudan Tanzania Togo Tuvalu Uganda Vanuatu Yemen, Rep. Zambia n.a., not available. aGDPis for 2002 and in current U.S. dollars. b~xcludeschanges in tariff revenue due to the lack of tariff data. The direction of the bias is unclear, but it is probably small since changes in tariff revenue represent on average less than 1percent of the total welfare change. Limiio and Olarreaga 239 Adam, C., and S. O'Connell. 2004. "Aid versusTrade Revisited." The Economic Journal 114:150-'73. Bagwell, K., and R. Staiger. 1998. "Regionalismand Multilateral Tariff Cooperation." In J. Piggott and A. Woodland, eds., International Trade Policy and the Pacific Rim. London: MacMillan. Bayard, T., and K. Ann Elliott. 1994. Reciprocity and Retaliation in US Trade Policy.Washington, D.C.: Institute for International Economics. Bhagwati, J. 1991. The World Trading System at Risk. Princeton, N.J.: Princeton University Press. Chang, W., and A. Winters. 2002. "How Regional Blocs Affect Excluded Countries: The Price Effects of Mercosur." American Economic Review 92(4):889-904. European Commission. 2000. "CommissionProposes Overhaul of Sugar Market." IPl00/1109, April 10. Brussels. Francois, J. H. van Mejil, and F. van Tongeren. 2005. "Trade Liberalization in the Doha Development Round." Economic Policy 20(42):349-39. Haveman, J., and H. Shatz. 2004. "Developed Country Trade Barriers and the Least Developed Countries: The Economic Results of Freeing Trade." In B. Guha-Khasnobis, ed., The WO, Devel- oping Countries and the Doha Development Agenda. New York: Palgrave-MacMillan. Hoekman, B. 2004. "Overcoming Discrimination against Developing Countries: Access, Rules and Differential Treatment." Paper presented at the Cornell Law School and Cordell Hull Institute conference "The Role of the wro System in the World Economy," July 9-10, Paris. Hoekman, B., C. Michalopoulos, and A. Winters. 2004. "Special and Differential Treatment of Devel- oping Countries in the wro: Moving Forward after Cancun." The World Economy 27(4):481-5116. Hoekman, B., A. Nicita, and M. Olarreaga. 2006. "Estimating the Effects of Global Trade Reform." In B. Hoekman and M. Olarreaga, eds., Global Trade Liberalization and Poor Countries: Poverty Impacts and Policy Implications. Washington D.C.: Brookings Institution Press, and Paris: Institut de Sciences Politiques. UIF (International Monetary Fund). 2004a. "Fund Support for Trade-Related Balance of Payments Adjustments." Policy Development and Review Department, Washington, D.C. [http:llwww. imf.org/external/np/pdr/tim/20041eng/022704.pd. 2004b. "IMFExecutive BoardApprovesTrade Integration Mechanism."Press Release04/73, April 13. Washington, D.C. Inama, S. 2003. "Trade Preferences and the World Trade Organization Negotiations on Market Access." Journal of World Trade 37(5):959-76. Jackson, J. 1997. The World Trading System: Law and Policy of International Economic Relations. Second edition. Cambridge, Mass.: MIT Press. Johnson, Harry G. 1967. Economic Policies toward Less Developed Countries. Washington, D.C.: Brookings Institution. Karacaovali, B., and N. LitnZo. 2005. "The Clash of Liberalizations:Preferentialvs. MultilateralTrade Liberal- ization in the EuropeanUnion." Policy Research Working Paper 3493. World Bank, Washingto4D.C. Kee, H. L., A. Nicita, and M. Olarreaga. 2004. "Import Demand Elasticities and Trade Distortions." Policy Research Working Paper 3452. World Bank, Washington, D.C. Krishna, P. 1998. "Regionalism and Multilateralism: A Political Economy Approach." Quarterly Jozirnal of Economics 113(1):227-51. Levy, P. 1997. "A Political-EconomicAnalysis of Free-Trade Agreements." American Economic Review 87(4):506-19. LimPo, N. 2002. "Are Preferential Trade Agreements with Non-trade Objectives a Stumbling Bloclc for Multilateral Liberalization?"Working Paper 02-02. University of Maryland, Department of Econom- ics, Center for International Economics, College Park. 2006. "PreferentialTrade Agreements as Stumbling Blocks for Multilateral Trade Liberalization: Evidence for the US." American Economic Review. LimHo, N., and A. Venables. 2001. "Infrastructure, Geographical Disadvantage, Transport Costs and Trade." World Bank Economic Review 15(2):451-79. McCulloch, R., and J. Pinera. 1977. "Trade as Aid: The Political Economy of Tariff Preferences for Developing Countries." American Economic Review 67(5):95947. Mendelowitz, Allan. 1994. "International Trade: Issues Concerning the Generalized System of Prefer- ences." GAOfl-GGD-94-174, June 20. U.S. General Accounting Office, Washington, D.C. OECD (Organisation for Economic Co-operation and Development). 2003. Tariffs and Trade: OECD Query and Simulation Package. Paris. Panitchpakdi, Supachai. 2004. "Dr. Supachai Lauds IMF'S New Policy." Statement before International Monetary and Financial Committee, April 24, Washington, D.C. Perroni, C., and J. Whalley. 2000. "The New Regionalism:Trade Liberalization or Insurance?"Canadian journal of Economics 33(1):1-24. Sapir, A. 1981. "Trade Benefits under the EEC Generalized System of Preferences." European Economic Review 15(3):339-55. Sapir, A., and L. Lundberg. 1984. "The US Generalized System of Preferencesand Its Impacts." In A.O. Krueger and R.E. Baldwin, eds., The Structure and Evolution of US Trade Policy. National Bureau of Economic Research Conference Report. Chicago, Ill.: Universityof Chicago Press. Stevens, C., and J. Kennan. 2004. "Making Preferences More Effective." Briefing Paper. Institute of Development Studies, Brighton, U.K. Subramanian, A. 2004. "Financing of Losses From Preference Erosion." WTflFlCOW14. Communica- tion from the International Monetary Fund. World Trade Organization, Geneva. Tangemann, S. 2002. "The Future of Preferential Trade Arrangements for DevelopingCountries and the Current Round of wro Negotiations on Agriculture." Rome: Food and Agriculture Organization. Trebilcock, M. J., and R. Howse. 1999. The Regulation of International Trade. Second edition. New York: Routledge. UNCTAD(United Nations Conference on Trade and Development). 1964. "Towards a New Trade Policy for Development." Report by the Secretary-Generalof UNCTAD. New York. 2000. "GeneralizedSystemof Preferences: Handbook on the Scheme of the USA." UNCTAD/ITCD/ TSBIMisc.58. Geneva. . 2002. "Handbook on the Scheme of the European Community." UNCTAD/ITCD/TSBIM~SC.~~/ Rev.2. Geneva. . 2003. "Trade Preferences for LDCs: An Early Assessment of Benefits and Possible Improve- ments." UNCTADIITCD/TSBI2003/8. Geneva. USFC (United States International Trade Commission). 1994. "Andean Trade Preference Act: Effect on the U.S. Economy and on Andean Drug Crop Eradication and Crop Substitution." Washington D.C. .1996. "Andean Trade PreferenceAct: Effect on the U.S. Economy and on Andean Drug Crop Eradication and Crop Substitution." Washington, D.C. USTR (Office of the United States Trade Representative). 2004. "Morocco FTA Leads to Progress on Labor Reform." Fact sheet. Washington, D.C. [http://www.ustr.gov/Document~Library/Fa~t~Sheets/ 2004/Morocco~FTA~Leads~to~ Progress-on-Labor-Reform.html]. .2005. "U.S. GeneralizedSystemof Preferences Guidebook."Washington, D.C. [http://www.ustr.gov/ assets/Trade~Developmen~referen~e~rogramdGSP/assetupoadle2678359.p. Winters, A. 1993. "Expanding EC Membership and Association Accords: Recent Experience and Future Prospects." In K. Anderson and R. Blackhurst, eds., Regional Integration and the Global Trading System. New York: St. Martin's Press. .1999. "Regionalismvs. Multilateralism." In R. Baldwin, D. Cohen, A. Sapir, and T. Venables, eds., Market Integration, Regionalism and the Global Economy. Cambridge, U.K.: Centre for Eco- nomic Policy Research. .2004. "Adjustment Assistancefor Trade Liberalization." World Bank, Washington, D.C. Price Effects of Preferential Market Access: Caribbean Basin Initiative and the Apparel Sector caglar Ozden and Gunjan Sharma Preferential trade arrangements should be evaluated by their effecton prices rather than by their effect on the total value of trade. This point is emphasized in the theoretical literature but rarely implemented empirically. This article analyzes the U.S. Caribbean Basin Initiative's (CBI'S)impact on the prices received by eligible apparel exporters. The CBI'S apparel preferences are the most important and heavily used unilateral preferences because of high trade barriers imposed on exports from the rest of the world. A fixed- effects generalized least squares (GLS) estimation is used to isolate the effects of other factors (such as quality, exchange rates, and transaction costs)and to identify the effects of tariff preferences. CBIexporters capture only about two-thirds of their preference margin despite the high degree of competition among importers. This translates into a 9 percent increase in the relative prices they re'ceive,with some variance across countries and years. Countries specializing in higher value items capture more of the preference margin, and the implementation of the Nortlh American Free Trade Agreement (NAF~A) has a negative effect. Removing Multifibre Arrangement quotas significantly lowers the benefits of CBIpreferences. Preferential trade arrangements have prioliferated in both number and impor- tance in recent years. Among them, rec.iproca1agreements, such as free trade agreements, receive more attention in academic and policy debates than unilat- erally granted preferences do. However, enhancing unilateral market access became one of the centerpieces of devel'opingcountry agendas in recent trade negotiations.* Furthermore, certain unilateral programs dramatically irnp~:ove eligible developing countries' access to highly protected markets such as aigri- culture and apparel. Thus, they have significantimpact on beneficiary,excluded, and granting countries. Caglar Ozden is an economist in the Development IKesearch Group, International Trade Division of the World Bank; his email address is cozden@worldbank..org. Gunjan Sharma is a graduate student un the Department of Economicsat the University of Maryland; her e-mail address is sharma@econ.bsos.umdl.edu. The authors thank Alan Winters and Daniel Lederman for their detailed comments and Bernard Hoekman, Hiau Looi Kee, Will Martin, Marcelo Olarreaga, Carlos Felipe Jaramillo, and Eric Reinhardt for their numerous suggestions. 1. SeeOzden and Reinhardt (2005)onhow unilateral preferencesaffectrecipient countries' own trade policies. THE WORLD BANK ECONOM~CREVIEW, VOL .20,NO . 2, pp. 211-259 doi:l0.1093/wberllt1j008 Advance Access publication May 4, 2006 O The Author 2006. Published by Oxford University Press on behalf of the International Bank for Reconstructionand Development/ TEEworn BANK. All rights reserved. For permissions, please e-mail: joumals.permissions@oxfordjournals.org. This article analyzes the impact of unilateral preferenceson the prices received by apparel exporters under the U.S. Caribbean Basin Initiative (CBI).The CBI was initiated in 1983 and grants duty- and quota-free market accessto a wide range of exports from 24 eligible countries in Central America and the Caribbean. Apparel preferences are the most valuable and heavily used ones because of high trade barriers imposed by the United States on exports from the rest of the world. The effects of discriminatory arrangements have been widely discussed and studied in the empirical literature. The focus has generally been on the total value of trade, even though the theoretical literature emphasizesthat prices are a more appropriate instrument for evaluating trade policies.2 However, quantity and price data are either not widely collected or not made available by most countries. Furthermore, even if these data were available, it is difficult to compare unit prices for differentiated products (such as machinery) that are aggregated in the same category. In short, the availability of very detailed and disaggregated data, together with the high utilization of the preferences, makes apparel an ideal sector for studying price effects of preferential arrangements. This article answers two related questions: How much do the prices received by CBI exporters increase due to preferences? What share of the preference margin is captured by exporters through higher prices? The prices received by CBI exporters naturally depend on the preferential tariffs they face and the most favored nation tariffs paid by excluded countries, as well as on many other market characteristics. One innovation here is the use of country-, industry-, and year-fixedeffects in a generalized least squares (GLS) estimation to isolate the effects of other factors, such as quality variation, exchange rates, transaction costs, and other market characteristics. CBI exporters capture about two-thirds of their preference margin, which translates into roughly a 9 percent increasein the relative prices they obtain. But this is an upper bound. Eligibility for preferences requires compliance with complicated rules of origin requirements that entail significant administrative and production costs.3Two recent innovative articles address this issue within the context of the North American Free Trade Agree- ment (NAFTA).Anson and others (2005) show that a large share of Mexico's preferential access was eroded by the rules of origin compliance costs. Cadot and others (2005)specifically analyze apparel imports from Mexico using a metho- dology similar to the one used here. They find that the border price of Mexican exports rises about 12 percent, with one-third of this increase being compensa- tion for the cost of complying with NAFTA'S rules of origin. In short, when these issues are all taken into account, the net benefits of preferentialmarket access are likely to be much lower. 2. See Winters (1997)for a convincing argument to this effect. 3. For example, CBI rules require the use of fabric and yarn from U.S.or domestic sources instead of inputs from third countries, which are generally cheaper. Another issue is the presenceof intermediaries in certain markets, who might also capture some of this preference rent. Ozden and Sharma 243 The results in this article exhibit significant variation across countries ;and years. More specifically, the implementation of NAFTA leads to a decline in the share of the preference margin captured bsy the beneficiaries, whereas specializa- tion in higher value items leads to an increase. Finally, the benefits from CBI preferences are significantly lowered when the Multifibre Arrangement quotas imposed on third countries such as China and India are removed. The results have important policy implications for the future of preferential trade arrangements. Among the main goals of such programs are the integration of developing countries into the world trading system and long-term economic growth through international trade. Many unilateral preferenceprograms fail to deliver the promised gains for a variety of reasom4 However, CBI programs provide significant advantages over other programs-such as the inclusion of the apparel sector-and are considered a silccess based on the rapid growth of exports from beneficiary countries. Nevertheless, eligible exporters do not cap- ture the full benefits of preferential access,,and the benefitsseem to be even lower in low-value products and are likely to be further restrained by the rules of origin restrictions. Another important point is that preferential access is valuable as long as excluded countries face high trade barriiers. When these barriers are lowered, as in the case of the implementation of N A ~ Aor the removal of Multifibre Arrangement quotas, the value of the preferences to the beneficiaries are con- siderably eroded. Recipient countries should not rely on preferences to deliiver long-term rents but use them as a transition stage to an environment where trade flows are determined by comparative advantage rather than by preferential access. Possible options are moving to higher quality products and taking advantage of geographic proximity to the U.S. market. CBI exporters can com- mand higher prices by providing rapid deliveries to U.S. retailers who are implementing just-in-time inventory management systems. This would require tighter integration of production facilities with the supply chain networks of the consumers, as Evans and Harrigan (2005)emphasize. The rest of the article is organized as follows: Section I reviews the literature. Section I1presents a brief history of the CBI followed by some stylized facts that motivate the article. Section I11explains the analytical model that forms the b.asis of the estimation, as well as the data and the methodology. Section IV presents the main results, along with how the effectof preferenceson export prices varies across countries and across years and how it is influenced by the Multifibre Arrangement quotas. Conclusions follow. The empirical literature on the impact o:F trade policies, especially preferential arrangements, on prices is not very large. In one of the earliest studies, Kreinin 4. See Hoekman, Michalopoulos, and Winters (2003)for a review. (1961)shows how the reductions of most favored nation tariffs by the United States influenced the export prices of its trading partners. With respect to the effects of discriminatory policies, the initial focus has been on voluntary export restraints. Crandall (1985)and Feenstra (1985)analyze the effect of U.S. volun- tary export restraints on Japanese and domestic automobile prices. In a different approach, Dinopoulos and Kreinin (1988)investigate the effects of voluntary export restraints on nonrestricted European exporters' prices. Thepriceeffectsof preferentialarrangements begantoreceivemoreattentiononly recently. Winters and Chang (2000) find that, as predicted, European exporters' prices increased relative to non-Europeanexportersafter Spain joined the European Community. Later, Winters and Chang (2002)also find that the relative prices of exports from excluded countries declined after Mercosur was implemented. Speci- fically, they show that, among the excluded countries, Chile and Japan fully pass through their own tariffs, Germany and the United States do so partially, and the Republic of Korea does so nominally. These articles, especially that of Winters and Chang (2002),are the most closely related to this article. Their empirical metho- dology and results are discussed in more detail in the following section. Olarreaga and Ozden's study (2005) seems to be the only study of the price effects of unilateral preferences. They show that, in the case of African Growth and Opportunity Act (AGOA) preferences, beneficiary countries' prices increased by only one-third of the preference margin, with the rest captured by importers. They then provide empirical evidence to show that the market power enjoyed by importers contributes to this division of the preference margin. However, limited data prevent them from conducting an in-depth analysis that fully controls for other factors. Krishna, Erzan, and Tan (1994) and Krishna and Tan (1998)also find wide evidence of rent sharing between exporters and importers in the context of apparel quotas from var- ious East Asian countries. This article is also related to the pass-through literature, which focuses mostly on the effect of exchange rate fluctuations on exporter and importer prices. Goldberg and Knetter (1997) provide a comprehensive review of this vast literature. The most relevant study is by Feenstra (1989), who estimates the effect of tariffs and exchange rates on U.S. prices of Japanese cars and finds that the long-run pass-through is identical. His estimation equation, which is similar to the one used here, is discussed in the methodology section. The CBI is a general term used to refer to the Caribbean Basin Economic Recovery Act of 1983 (CBERA), the Caribbean Basin Economic Recovery Expansion Act of 1990 (CBERA Expansion Act), and the Caribbean Basin Trade Partnership Act of 2000 (CBTPA).The aim of the CBI is "to assist in the achievementof a stable political and economic climate by stimulating the development of the export potential of Ozden and Sharma 245 the region."' Its main feature is quota- and tariff-free market access gra:nted unilaterally by the United States to exports from eligible countries. Textile and apparel articles subject to textile agreements (such as the Multi- fibre Arrangement) were initially exempt from preferential treatment. In June 1986, a special access program, called the Super 807, was implemented for imports of textile apparel assembled in CBI beneficiaries from fabric formed and cut in the United States. It granted partial duty-free treatment on the domestic value added and on inputs from the United States. Export processing zones rapidly appeared in the region for the production of eligible products. Initially, a sunset provision was included (Section218)that terminated duty-.free treatment on August 5,1990. But the CBEIU Expansion Act, signed on August 29, 1990, extended the initial preferences. Once NAFTA was implementedin 1994:,csr countries began to worry about the erosion of their preferences. The CBTFJA w.asa response to these concerns. Section 211 specified the new regime for textiles and apparel preferences. In contrast to the previous regime, which granted partiial duty-free treatment, the CBTFJA allows textile and apparel articles to enter the United States without any tariffs or other restrictions when certain rules of origin a~ndother requirements are satisfied. As before, these rules of origin favor the use of materials formed in the United States or CBI countries. In short, the CBTFJA has provided CBI members with N A F T A - ~ ~ ~ ~ treatment without the burden of reciprocity prescribed by NAFTA. Stylized Facts The CBI preferences had a large impact on the aggregate volume of apparel exports from the benefi~iaries,~which accounted for a steadily rising share of total U.S. imports during 1989-2002, the span of data used here. In 1989, exports from CBI countries were valued at $1.7 billion, 7.8 percent of total U.S. imports. In 2002, their exports increased to $9.5 billion, 16.2 percent of total U.S. imports (figure1). As stated earlier, the main focus herleis on the prices received by apparel exporters due to the preferences.The average price of U.S. apparel imports peaks at $83 in 1991 and declines rapidly to $52 in 2002 (figure2).These are nomlinal prices, so the decline in real prices is even steeper. The average price of CBI apparel exports is always below the average U.S. import price, probably because of quality and other differences. However, the gap narrows considerably over time. To put it differently, the price ratio starts at 84 percent in 1989 and increases to 90 percent in 1993. It stays quite stable at about 90 percent until the CBTPA is implemented, after which it rises to 92 percent in 2002. 5. The US.-Caribbean Trade Partnership Act of 2000, Title 11: Trade Benefits for Caribbean Basin, Subtitle A, Section 202. 6. Somecountries aremore successfulatincreasingtheirexports. Eightof the24eligibleexportersCosta Rica, the Dominican Republic,ElSalvador, Guatemala, Haiti, Honduras,Jamaica, and Nicaraguaaccount for more than 99 percent of the total. FIGURE 1. U.S. Imports of Apparel, 1989-2002 Total U.S. imports Imports from Caribbean Basin Initiative countries Source: United States Intenational Trade Commission. FIGURE 2. Average Unit Prices for U.S. Apparel Imports and Caribbean Basin Initiative (CBI)Apparel Exports, 1989-2002 $ 4 0 1 1CaribbeanInitiative exportsBasin Source: United States International Trade Commission. Although the average most favored nation tariffs gradually declinefrom 20.6 percent to 18 percent over 1989-2002, the preferential tariffs paid on CBIexports drop rapidly from 20.7 percent in 1989 to 9 percent in 1993, where they remain stable until the CBTPA is implemented (figure 3). They then fall to 5 percent in 2002. In other words, in 2002, CBI countries enjoyed an average preference margin of about 13 percent. The relative prices of exports from beneficiary countries and their preference margins both increased from 1989 to 2002, indicating a strong correlation. But Ozden and Sharma 247 FIGURE 3. Most-Favored Nation and Preferential Tariffs, 1989-2002 - 20 - \Most \ favored nation tariffs 15 s \ Q 0 & 10- \ I Caribbean n Basin 5 h. Initiative Note: Tariffs are weighted by Caribbean Basin Initiative (CBI) export volume. Source: United States International Trade Commission. price increases can be caused by many factors, such as quality upgrades by exporting firms, declines in transportation and other transaction costs, fluctua- tions in exchange rates, and declines in tariffs. The key question is whether (and what share of) this export price increase is due to the preferential market access. And a related question is what share of the preference the exporters capture through higher prices. The next sectio~nprovide an analytical and empirical framework to investigate the preliminary evidence presented. The previous section showed that CBI countries increased their exports to the United States considerably and obtained higher prices for their exports aftel-the CBERA and CBTPA were implemented. This;section attempts to identify the extent of the export price change that is due to the preferential access. The first estimation equation is based on the examples in the pass-through literature. Let pi denote the price of procluct k from country i inclusive of tariffs and transport costs. Similarly, let ptoW denote the final average import price if k was imported from the rest of the world. Based on these definitions, a typical estimation equation in the pass-through literature would be where t is the time subscript and is the tariff rate (Goldbergand Knetter 1997). 7 This equation implies that ph would depend on the price of imports from other countries (orthe domestic price index in ithat category), the tariff rate, the w;ages w (or another proxy for costs), and the exchange rate e of country i. The main coefficients of interest would be al, the tariff rate pass-through coefficient, and a 4 , the exchange rate pass-through coefficient. With perfect pass-through, both coefficients would equal 1. Other control variables and fixed effects can also be included. Winters and Chang (2000)derive a similar equation based on an imperfect competition model to analyze the effect of Spain's accession to the European Union (EU)on the prices of excluded countries' exports to Spain, includingthose from the United States. However, they estimate a relative price equation of the following form: where the product subscript k is suppressed; pi, pRoW,7', and Pow are as defined above; z' and zRoWare the costs of exporters and are functions of wages and exchange rates, and Y and P are the income and price levels in Spain used to capture demand conditions. The focus of attention is the coeffi- cients a1 through a 4 as well as certain restrictions implied by theory. Their results imply that a 1 percent decline in the tariffs faced by i (EU countries) causes a 0.56 percent decline in the relative prices of U.S. exporters to Spain. They estimate a similar equation for Mercosur in the article by Winters and Chang (2002) where the dependent variable is the ratio of U.S. (excluded country) export prices in Brazil to export prices in the rest of the world. In contrast to the above studies, the focus here is on the prices received by exporters, net of tariffs, and other transaction costs. The pi (@OW) here denotes the net prices received by exporters of k from i (ROW)without tariffs or other costs, < represents the preferential tariff imposed on i, and TYis the most favored nation tariff rate imposed by the United States on the rest of the world. The estimating equation here is similar to equation (2),but the dependent variable is the ratio of pretariff prices: where In (~h,/pf('~) is the approximate difference (in percentage) in net prices received by exporters from the CBI and the rest of the world, and (rkK,Ow - 7Lt) is the average preferencemargin enjoyed by the exports of beneficiary country i. Because the tariff imposed by the United States on the rest of the world, %Ow, did not vary considerably in the sample over time, the tariff difference in estimation was used instead of two separate tariffs. Also included are the total Ozden and Sharma 249 export volume of country i in category k, denoted by xi,,and total U.S. imports in that category, denoted by M;:~. Country, product, and year dummy variables, denoted 0,a, and 9,respec- tively, are added to capture variables that are missing from the estimation equation.7 These include exchange rates and wages that are included in equa- tions (1)and (2)as well as variables such as differences in quality and transport costs that are unrelated to the effects of preferential market access programs.8As mentioned earlier, these dummy variables help isolate all these effects that influence prices and allow the focus to be the impact of preferential prograjms. Implementation of a preference program is equivalent to a decline in the tariff rate faced by the beneficiaries, 7;. If eligible countries capture all the benefits of the tariff reduction (that is, 01= I), the prices they receive for their exports should increase by the amount of the tariff decline. However, if the increase in export prices is less than the tariff decline (that is, PI< I),the importers are capturing a share of the tariff rents created by the preferential market access. The traditional tariff pass-through effect, denoted by a1in equations (1)and (2), can be found from the estimation. More specifically, given the definitions of prices used here, a1=1- pl.If ,Bl = 0.25, this implies that a 1 percent tariff increase would decrease the pretariff price by 0.25 percent (or increase the posttariff price by 0.75 percent). Thus, th.e tariff pass-through rate is 75 percent. Data and Mt?thodology The United States International Trade Commission collects and pub1i:shes detailed and disaggregated customs data, including the customs value, unit prices, and duties paid in a given eight-digit Harmonized System category f.rom any country for 1989-2002.~The data are further classified by whether the imports entered the United States under a specific preference program (suclhas AGOA, the CBTPA, Generalized System of Preferences, or NAFTA) or no program (that is, under most favored nation statu:s).1° 7. Other specifications with country-product fixed effects and the like are also tested; they are identified in more detail in the Results section. 8. Quotas imposed on other exporting countries, shocks to demand in the United States, and supply shocks in the apparel sector can also be included. 9. Data were obtained from the United States International Trade Commission Web site at http:// www.dataweb.usitc.gov. The customs value data exclude insurance and freight. 10. Until the CBTPA was implemented in 2000, the main preference scheme under the CBI was duty-free treatment on the portion of the value-added created in the beneficiary country and the inputs (fabricand yarn) imported from the United States. In other words, the most favored nation tariff was paid only on the portion of the inputs imported from third countries. ]However, all shipments to the United States i~nthe same eight-digit category were compiled together and listed under the most favored nation category..The tariffs reported for a given eight-digit category are thus an average rate of the tariff paid on individual shipments. Because shipment-level data are not available, these averages are relied on here. Since 2.000, separate data are available for exports entering under the c ~(wherethe tariff is zero) and under the old n ~ scheme (listed as most favored nation). The results in table 5 indicate that the coefficients of the tariff difference variable are not statistically different under the two regimes. The data are customs value, quantity, and duties collected from each country in the sample for 1989-2002 disaggregated at the eight-digit level of the Harmonized System. The prices received by the exporters, denoted as pi,, are unit prices, calculated as the ratio of customs value to the number of units of category k in year t from country i. The average U.S. import price, pfoW, is the average unit price received by exporters from the rest of the world, excluding CBI countries.ll Tariffs imposed on the exports of country i in category k in year t, denoted as T:,,were calculated as the ratio of collected duties to customs value. In addition, most favored nation tariffs, $OW, were calculated as the ratio of collected duties to customs values from all exporters to the United States, excluding beneficiaries of preference programs (such as AGOA, the CBI,General- ized System of Preferences,and NAF~A).Both most favored nation tariffs and U.S. prices vary over products and years but not over beneficiary countries. The analysisis conductedfor the eight largest exportersof apparel to the United States from the Caribbean and Central America: Costa Rica, the Dominican Republic, El Salvador, Guatemala, Haiti, Honduras, Jamaica, and Nicaragua. Of 24 eligiblecountriesunder the CBI, only14 actually exported apparel to the United States during 1989-2002. The data set used here for the eight countriescovers 99 percent of the $82.3 billion worth of apparel imports into the United States from all eligible CBI countries during the period. The data have 211 eight-digit cate- gories, which are grouped into 32 four-digit categories (table1). The product dummy variables are at the four-digit level rather than at the eight-digit level.12Most eight-digit categories within a four-digit category are very similar, and four-digit product dummy variables are likely to capture most of the effects (quality, margin effects, and demand and supply shocks) targeted TABLE 1. Sample Statistics Mean Standard Deviation Ratio of Caribbean Basin Initiative price to U.S. price (percent) Tariff difference (percent) (post-1991) Export value (millions of dollars) Total U.S. imports (millionsof dollars) Number of countries Number of years Number of eight-digit product categories Number of four-digit product categories Number of observations Source: Authors' analysis based on data described in the text. 11. Unitpricesareproductspecific,suchasperdozenshirtsorpants;theyare not measuredbyweight or amount of fabric used, as some apparel data are reported. 12. Eight-digit categories are extremely narrow and detailed. For example, 6105 is men's or boys' knitted shirts, whereas 61052020 is men's or boys' cotton knitted shirts. Ozden and Sharw 251 here, whereas eight-digit dummy variables unnecessarily reduce the degrees of freedom of the estimation. Despite the highly disaggregated data, fixed effects may not be enough to capture systematic differences between ]products. In particular, heteroskedasti- city in residuals is a concern. For example, heteroskedasticity may exist across ~anels-that is, the variance of the error may be different for each panel (the product k). This may be due to the variation of scale in imports of different products. It could also be due to specific features of a product that system- atically affect the error. To correct for heteroskedasticity, a two-step feasible GLS estimation procedure is adopted. In the first step, ,Dis estimated using ordinary least squares and used to calculate the residuals. These are in turn used to construct a consistent estimator for tlhe variance matrix. Each variab1.e is reweighted by the inverse of the commodity-specific residual standard devia- tions from the variance matrix, and PCLSis estimated. Thefixed-effectsGLS estimator allows product-, country-,and year-specific uinob- served error terms to be estimated as parameters while allowing the idiosync.ratic component of the error to have a more general structure. Robust standard errors can also be used on ordinary least squares estimators. The feasible GLS estimatlor is more efficient than the fixed-effectsestimator obtained by ordinaryleast squares as the number of panels (that is, the product categories)as K -+ cm,T fixed. IV. ESTIMATION RESULTS The resultsof the main estimation using the full sampleare summarized in table 2. As described earlier, equation (3)is estimated using feasible GLS, which provides consistentand efficient estimates. The second column reports resultswith separate country-fixed, year-fixed, and four-digit product category-fixed effects. The third column presents the results from a similar regression with joint country-ycar- product category dummy variables. Instead of separate dummy variables for each country, four-digit category, and year (denoted by Ri, Qk, and Prrespectively),a single dummy variable denoted as Ti,is used. This much more general structure allows for quality effects in category k to vary across years and countries simul- taneously. The disadvantageis that it requires additional dummy variables, and it substantially reduces the degrees of freedom.13 All the variables have very significantcoefficientswith the expected signs. 'The variable of most concern is Tariff Difference, which is the difference between the most favored nation tariff imposed by ithe United States and the preferential tariff enjoyed by the CBI beneficiaries. The coefficient is 0.663 in the second 13. For example, the data set coverseight countries, 14 years, and 32 four-digitproduct categories.In the first regression, this leads to a total of 51(7+13 i- 31) dummy variables, because one variable from each group is dropped. But in the second estimation, there are 3,584(8 x 14 x 32 - 1) potential vari- ables. In the actual estimation, many of these are dropped because of collinearity, but 2,413 dulmmy variables remain. Significantlyincreased computation time is an additional cost. TABLE 2. Effect of Preference Margin on Caribbean Basin Initiative Benefici- aries' Export Prices All Observations Quality Controlled Tariff difference 0.663" (0.056) Log of export value 0.042" (0.003) Log of total U.S. imports -0.026" (0.004) Constant -0.010(0.072) Product group-fixed effects Yes Country-fixed effects Yes Year-fixed effects Yes Country-year-product-fixed effects No Yes Number of observations 7,784 7,784 X* 4,585.31" 10,332.94" "Statistically significant at the 1 percent level. Note: Dependent variable is the ratio of Caribbean Basin Initiative price to average U.S. price. Numbers in parentheses are standard errors. The second column includes separate product group-, country-, and year-fixed effects; the third column includes a combined dummy variable. Source: Authors' analysis based on data described in the text. column and 0.642 in the third column, implying that the CBI beneficiariescapture about two-thirds of the preference margin (or the tariff rent). Another way to interpret this result is to look at the price increase received by the CBI benefici- aries. Although it varies across years, countries, and products, the average preference margin in the sample in 2002 is 13 percent. This means that the exporter price increase due to preferences is about 8.5 percent. The other 4.5 percent is captured by the importers, who now enjoy lower import prices.14 In a perfectlycompetitive market with homogeneousgoods, the exporters would be expected to captureall this potential rent. The main estimationequationincluded additionalvariablestocapturemarket power effectsthat might explainwhythe tariff rent is being shared between exporters and importers. The variables are the natural log of the total exports of country i and the natural log of total U.S. imports in category k in year t.The coefficientsof both variables are significant. The results in the second column imply that doubling the exports of i (withconstant U.S. imports so that the market share of i is also doubled) is associated with a 4.2 percent increase in the relative export prices received. But doubling U.S. imports (with constant exports from i so that its market share is halved) is associated with a 2.6 percent decline in export prices. The coefficients in the third column are smaller, indicating that the additionaldummy variables are capturingsome of these effects. The results do not depend on the inclusion of the market power effect variables. 14. As mentioned in the Introduction, the rules of origin create large administrative costs for compliance and production costs from having to use more expensive U.S. inputs instead of possibly cheaperinputs from third countries. If the CBI beneficiary firmscannot pass these costsonto the buyers, the real benefits of preferential access are likely to be lower than the 8.5 percent price increase. Ozden and Sharma 253 If both are dropped from the estimation, the coefficient of Tariff Difference falls to 0.58 and retains the same level of significance. Variation Across Countries The next question is whether the tariff rent captured and export price incrlease due to preferential access vary across countries. Equation (3) is estimated for each country, with year and four-digit product dummy variables. The coefficient of Tariff Difference is reported in the second column of table 3, although the market share variables are included in the estimation. The countries are ordered in terms of decreasing export volumes to the United States in the data set, with the Dominican Republic as the largest exporter and Nicaragua as the small~est. The results for the three largest exporters are similar: the Dominican Reyub- lic, Honduras, and Guatemala capture 72-79 percent of the tariff rent. A similar result is found for Costa Rica, the fifth largest exporter. But El Salvador, the fourth largest, captures 46 percent. Whein the average preference margins faced by the countries over time are calculated, the Dominican Republic, Honduras, El Salvador, and Costa Rica have similar trade-weighted averages: 10-11.5 percent (after1991) compared with only 6 percent for Guatemala. This implies that the average export price increases due to preferential market access are 8.5--9.5 percent for the Dominican Republic, !Honduras, and Costa Rica but only about 4.5 percent for El Salvador (due to a smaller share of the rents being captured) and Guatemala (dueto lower preference margins).ls TABLE 3. Variation Across Countries, 1992-2002 Tariff Total Imports Number of Difference (Millions of Dollars) Observations Dominican Republic 0.794" (0.110) 23,250 Honduras 0.729" (0.117) 17,226 Guatemala 0.719" (0.117) 11,814 El Salvador 0.458' (0.127) 10,764 Costa Rica 0.750' (0.171) 9,297 Jamaica -0.692" (0.232) 4,686 Haiti 1.315* (0.217) 2,193 Nicaragua 0.098 (0.224) 2,101 Product-fixed effects Yes Country-fixed effects No Year-fixed effects Yes *Statistically significant at the 1 percent level. Note: Dependent variable is the ratio of Caribbean Basin Initiative price to average U.S. price. Numbers in parentheses are standard errors. Each estimation includes separate product group- and year-fixed effects and has a X2 statistic significant at the 1 percent level. Source: Authors' analysis based on data described in the text. 15. Detailedcountry-level data are not presented dueto spacelimitations. They are available from the authors upon request. The other three countries (whichexport significantlylower volumes than the top five)exhibit different patterns. Haiti has a coefficient that is larger than but not statistically different from 1, which means that it is capturing all 15 percent of the preferencemargin. Thus, it is no surprise that Haiti had one of the fastest growing export volumes to the United States over the time period. Nicaragua's coefficientis not statistically different from 0, which means that it receives none of its average5.7 percent tariff preference rent. The most perplexing outcome is Jamaica's negative and significant coefficient. Jamaica has the worst export performance, with volumes declining from a peak of $500 million in 1994 to $120 million in 2002. The decline in export volumes and failure to capture the preference margins should be related. Also, the vast majority of plants in Jamaica are subsidiaries of U.S. firms. The surprising result might be the out- come of transfer pricing issues. What can explain the different patterns in the capture of tariff preferencerents? Larger exporters naturally capture a larger share of the rent. This result is also found in Olarreaga and Ozden's (2005)analysis of AGOA preferences in apparel. The category of exports in which these countries specialize also matters. The average export unit price for the Dominican Republic, Honduras, Guatemala, and Costa Rica is about $58, but the average price for all U.S. imports in categories where these countries export is $63. By contrast, the average export pricefor El Salvador is $37, and the averageU.S. import pricein thesecategoriesis $44. These values imply that the first four countries, when compared with El Salvador, are specializing in higher value categories and are producing higher quality products relative to average U.S. imports in these categories. These are consistent with the results of Evans and Harrigan (2002), who emphasize the increasing importance of just-in-time manufacturing and retailing in apparel. By specializing in high-value categories where other product features (suchas quality, timely delivery, and production flexibility) become more impor- tant, these countries are able to extract better prices from importers. Both Haiti and Nicaragua alsospecializein low-pricecategories, but they performdifferently. Haiti's average preference margins are similar to those of the larger exporters, whereas Nicaragua's margins are much lower. This difference indicates that a larger share of Haitian exports enter tariff-free and that Haitian exporters are better at complying with the rules of origin requirementsand at taking advantage of the preferences, which requires a certain level of legal and business expertise. Variation Across Years Another interesting issue is the variation in the level of the preference margins captured by the exporters over time. The main equation is estimated for each year separately starting in 1992, when the preferences first appear in the data (figure 3), using country and product category dummy variables. All the coeffi- cients on Tariff Difference are highly statistically significant, starting with a value of 1.05 in 1992 and declining to 0.685 in 1998 (table 4). &den and Sharma 255 TA BLE 4. Variation Across Years Tariff Total Imports Number of Difference (Millions of Dollars) Observations 1992 1.052' (0.289) 1993 0.787' (0.207) 1994 0.812' (0.198) 1995 0.686' (0.192) 1996 0.728" (0.173) 1997 0.658" (0.182) 1998 0.685' (0.192) 1999 1.183' (0.203) 2000 1.287' (0.202) 2001 0.592' (0.156) 2002 0.815' (0.163) Product-fixed effects Yes Country-fixed effects Yes "Statistically significant at the 1 percent level. Note: Dependent variable is the ratio of Caribbean Basin Initiative price to average U.S. price. Numbers in parentheses are standard errors. Each estimation includes separate product group- and country-fixed effects and has a X2 statistic significant at the 1 percent level. Source: Authors' analysis based on data describled in the text. One explanation is the NAFTA effect: these are the years when NAFTA entered into force, which led to Mexico becoming a significantcompetitor to CBI coun- tries, the only countries with preferential ;accessto the U.S. apparel market u.nti1 NAFTA.'~ However, the rents captured by .the beneficiary countries seem to have sharply increased in 1999 and 2000 before the CBTPA went into force. In fact, the beneficiaries seem to be capturing all the tariff rents, because the coefficient is larger than but not statistically different from1.The coefficient is 0.592 in 2001 and 0.815 in 2002. But the net benefit stays the same, because prefere:nce margins also increase due to the CBTPA (figure 3). For example, the aver.age preference margin in 2002 is 13 percent, which means that the average export price increase due to preferential access is;still about 10.5 percent. The beneficiariescapture a largershare of the rentsunder the original CBIregi.me (71 percent) than they do under the new CBTPA regime (64 percent) (table 5).17 However, the average preference margin under the old regime is 9.5 percent compared with 12.6 percent under the CBTPA. This means that the averageincrease in export prices due to preferences is 6.7 percent under the old regime and 8.1 percent under the CBTPA. This once more explains why the CBI countries were so keen on implementing the CBTPA. 16. As mentioned earlier,the competitionfrom Mexicoactuallyled the CBIbeneficiariesto extensively lobbythe U.S. government to implementthe CBPA and greatly increase CBI benefitswhilerelaxingthe rules of origin requirements. 17. The CBTPA portion of the data includes only the exports that enter under zero tariffs. TABLE 5. Effect of the Caribbean Basin Initiative (CBI) 1992-2000 (CBI) 2001-2002 (CBTPA) Tariff difference 0.709" (0.071) 0.642" (0.166) Log of export value 0.036" (0.004) 0.034"* (0.008) Log of total U.S. imports -0.019" (0.005) -0.009(0.012) Constant -0.014 (0.084) -0.383(0.259) Product group-fixed effects Yes Yes Country-fixed effects Yes Yes Year-fixed effects Yes Yes Number of observations 3,890 986 x2 2,441.53" 1,432.12'' CBTPA, Caribbean Basin Trade Partnership Act of 2000. "Statistically significant at the 1 percent level. Note: Dependent variable is the ratio of CBIprice to average U.S. price. Numbers in parentheses are standard errors. Each estimation includes separate product group-year- and country-fixed effects. Source: Authors' analysis based on data described in the text. Impact of Quotas The final question addressed here is the impact of U.S. apparel quotas (imposed on third countries, mainly Asian countries such as China, India, and the Repub- lic of Korea)on the export prices and tariff rents received by the CBI beneficiaries. Because quota data and import data (with unit prices and preference programs) are collected under different classification schemes and in most cases a quota category covers multiple eight-digit categories and the coverage might vary by exporting country, the data had to be adjusted. First, a list of the top 16 apparel exporters to the United States that face significant quotas was compiled.18 Second, data on the country-level quota size and the fill rate (assumed to be the same for all Harmonized System categories within that quota category for that country) were collected. Third, various measures of quota restriction were constructed to analyze the impact of the Multifibre Arrangement regime on the CBI beneficiaries. Fourth, equation (3) was esti- mated for the sample 1998-2002 with country-fixed, year-fixed, and four-digit product group-fixed effects (table 6). The first variable is percent of Exports from Quota Countries. It has a mean of 52.2 percent and is the market share of the quota countries for a given year in a given Harmonized System category. The market share of quota countries has no significant impact on the prices received by CBI beneficiaries. The second variable, percent of Exports under Quotas, is the total value of imports entering under quotas, whether they are binding or not. It has a mean value of 22 percent 18. These are China, Turkey, India, Pakistan, Bangladesh,Sri Lanka, Thailand, Vietnam, Cambodia, Malaysia, Indonesia, the Phillipines, Macao (China), the Republic of Korea, Hong Kong (China),and Taiwan (China).The others in the top 20 exporters of apparel are Mexico, Canada, and CBIcountries. Ozden and Sharma ;!57 TABLE 6. Effect of Quotas - Percent of Exports Percent of Percent of Exports from Quota Exports under under Binding Countries Quotas Quotas Tariff difference 0.656" (0.056) 0.661" (0.056) 0.645" (0.056) Quota restriction variable 0.021 (0.019) 0.066' (0.022) 0.078' (0.03;') Product group-fixed effects Yes Yes Yes Country-fixed effects Yes Yes Yes Year-fixed effects Yes Yes Yes Number of observations 3,883 3,883 3,883 x2 2,679.96" 2,690.31 2,682.96" Statistically significant at the 1 percent level. Note: Dependent variable is the ratio of Caribbean Basin Initative price to average U.S. pr~~ce. Numbers in parentheses are standard errors. Each estimation includes separate product group-, year-, and country-fixedeffects. Source: Authors' analysis based on data describe'din the text. and a coefficient of 0.066, which implies that CBI countries receive about 3..3 percent higher relative prices in an average category where quotas are imposed on other countries. This is an important benefit of preferential access. Finally, a third variable, percent of Exports under Binding Quotas, is constructed, with a quota defined as binding if the fill rate is aLbove 80 percent. The mean value lor this variable is 7.2 percent, and the coefficientis 7.8 percent, which is the relative price increase obtained by CBI beneficiaries due to binding Multifibre Arrange- ment quotas. These different results imply that the quotas and other nontariff barriers imposed on a group of countries have an important effect on the prices received by other countries, especially the benefic~iariesof preferential market access. These results are intriguing and need to be further explored. The theoretical literature emphasizes that .tradepolicies should be evaluated by looking at their effect on prices rather than on the value of trade. However, this rarely occurs empirically, except in the pass-through literature and recent work on regional agreements by Winters and Chang (2000,2002).The CBI preferences in apparel are ideal for the analysis of th~eseissues. First, because of barriers imposed on excluded countries, these preferencesare highly valued and heavily utilized by the beneficiaries. Second, detai.ledand disaggregated unit value and quantity data of more than a decade are available. When country-, year-, and product category-fixed effects are used, tlie effects of other variables (quality changes, exchange rates, and the like)can Ibe isolated, and the focus can shift to the price effects of preferences. This has not been done as extensively as in t:he literature. The results here indicate that CBI beneficiariescapture about two-thirds of the preference margin, which causes their relative prices to increase by about 9 percent. The net benefits to exporters are likely to be lower, because they face additional administrative and production costs to comply with the rules of origin. The rest of the benefits go to importers through lower prices. More interesting,strong variation occurs over time and across countries. For example, NAFTA has a negative effect, whereas specializing in higher value products has a positive effect on capturing the preference margin. Furthermore, eliminating Multifibre Arrangement quotas is likely to significantly decrease the benefits of preferential access. There are severalimplications for exporters from beneficiary countries. First, they need to be aware that preferences do not necessarily have a positive effect on the prices they receive.Especiallyif they specialize in low-quality or low-price categories,they are likely to capture only a small share of the preferencemargin. Second, the price effect of preferential accessis quite sensitive to the extent of the barriers imposed on the excluded countries. As these barriers are removed, the preferences are going to become less valuable. Several issues remain. The results here indicate that the prices received by the excluded countries in the United States relative to prices of beneficiaries have declined. But how do the prices that they obtain in the United States change relative to the prices in the world markets? This is especially important if such unilateral preferences harm excluded countries. Also, the effect of preferences on quality upgrading has not been explored empiricallydespite the fact that policy- makers often claim positive effects. Anson, Jose ,Olivier Cadot, Antoni Estevadeordal, and others. 2005. "Rules of Origin in North-South Preferential Trading Arrangements with an Application to NAF~A." Review ofInternatioml Economics 13(3):1501-17. Cadot, Olivier , Celine Carrere, Jaime de Melo, and Alberto Portugal-Perez. 2005. "Market Access and Welfare Under Free Trade Agreements: Textiles under NAFTA." World Bank Economic Review 19(3):379-405. Crandall, Robert W. 1985. "Assessing the Impact of the Automobile Voluntary Export Restraints upon US Automobile Prices." Washington, D.C.: Brookings Institution. Dinopoulos, Elias , and Mordechai E. Kreinin. 1988. "Effects of the U.S.-Japan Auto VER on European Prices and U.S. Welfare." Review of Economics and Statistics 70(3):484-91. Evans, Carolyn, and James Harrigan. 2005. "Distance, Time, and Specialization." American Economic Review 95(1):292-313. Feenstra, Robert C. 1985. "Automobile Prices and Protection: The U.S. Japan Trade Restraint." Journal of Policy Modeling 7(1):49-68. .1989."SymmetricPass-through of TariffsandExchangeRatesunderImperfect Competition: An Empirical Test." Journal of International Economics 27(1-2):2545. Goldberg, Pinelopi Koujianou ,and Michael M. Knetter. 1997. "Goods Pricesand Exchange Rates: What Have We Learned?"Journal of Economic Literature 35(3):1243-72. Ozden and Sharma 259 Hoekman, Bernard , Constantine Michalopoulos, and L. Alan Winters. 2003. "Development and More Favorable and Differential Treatment of DevelopingCountries." Washington, D.C.: World Bank. Kreinin, Mordechai E. 1961. "Effect of Tariff Changes on the Prices and Volume of Imports." American Economic Review 51(3):310-24. Krishna, Kala, and Ling Hui Tan. 1998. Rags and Riches, Implementing Apparel Quotas under the Multifiber Arrangement. Ann Arbor, Mich.: University of Michigan Press. Krishna, Kala, Refik Erzan, and Ling Hui Tan. 1994. "Rent Sharing in the Multifiber Arrangement: Theory and Evidence from the U.S. Apparel Imports From Hong Kong." Review of International Economics 2(1):62-73. Olarreaga, Marcelo, and Caglar Ozden. 2005. "AGOAand Apparel:Who Captures the Tariff Rent in the Presence of Preferential Market Access?"World Economy 28(1):63-77. Ozden, Caglar, and Eric Reinhardt. 2005. "Perversity of Preferences: GSP and DevelopingCountry Trade Policies, 1976-2000." Journal of Development Economics 78(1):1-21. Winters, L. Alan. 1997. "Regionalism and the Rest of the World: The Irrelevance of the Kemp-Wan Theorem." Oxford Economic Papers 49(2):228-34. Winters, L. Alan, and Won Chang. 2000. "Regional Integration and Import Prices: An Empirical Investigation."Journal of International Economics 51(2):363-77. 2002. "How Regional Blocs affectExcludedCountries: The Price Effectsof Mercosu~."American Economic Review 92(4):889-904. Aid and the Supply Side: Public Investlment, Export Performance, and Dutch Disease in Low-Income Countries Christopher S. Adam and David L. Bevan Contemporary policy debates on the macroeconomics of aid often concentrate on short-run Dutch disease effects, ignoring the possible supply-side impact of aid- financed public expenditure. In the simple model of aid and public expenditure presented here, public infrastructure generates an intertemporal productivity spil- lover, which may exhibit a sector-specific bias. The model also provides for a learning-by-doing externality, through which total factor productivity in the trad- able sector is an increasing function of past export volumes. An extended compu- table version of this model is used to simulate the effect of a step increase in net aid flows. The simulations show that beyond the short run, when conventional demand-side Dutch disease effects are present, the relationship between enhanced aid flows and real exchange rates, output growth, and welfare is less straightfor- ward than simple models of aid suggest. Public infrastructure investment that generates a productivity bias in favor of nontradable production delivers the largest aggregate return to aid, but at the cost of a deterioration in the income distribu- tion. Income gains accrue predominantly to skilled and unskilled urban households, leaving the rural poor relatively worse off. Under plausible parameterizations of the model, the rural poor may also be worse off in absolute terms. Recent global initiatives on debt relief and development assistariceanticipate a significant increase in overall aid flows to the poorest countries and, at least in the medium term, a concentration of these flows on a small number of reci- pients (e.g., Commission for Africa 2005; United Nations Millennium Project Christopher S. Adam is reader in development economics at the University of Oxford; his email address is christopher.adam@economics.ox.ac.uk. David L. Bevan is emeritus research fellow at St John's College in the University of Oxford; his email address is david.bevan@economics.ox.ac.uk. The article is based on work originally carried out for the World Bank and the U.K. Department for International Development ( ~ ~ ~ U g a n dThea ) .authors gratefully acknowledge the support of both institutions. They thank Tony Killick, Simon Maxwell, Catherine Pattillo, Luca Ricci, and seminar participants at the universities of Oxford, Western Ontario, Brunel, and Clermont-Ferrand. They also thank the Overseas Development Institute, and the International Monetary Fund for helpful comments. The article has also been much improved in response to comments by three anonymous refereesand by the journal editor. Supplemental appendixes to this article are available at http://wber.oxfordjournals.org. THE WORLD BANK ECONOMICREVIEW, VOL.20, NO. 2, pp. 261-290 doi:l~D.l093/wber/lhjOll Advance Access publication May 17,2006 O The Author 2006. Published by Oxford UniversityPress on behalf of the International Bank for Reconstructionand DevelopmentITHE WORLDBANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org. 2005). Accompanying these pressures for a scaling-up of aid, however, is a heightened anxiety among some donors and potential recipients that large increases in aid may jeopardize macroeconomic stability and growth (Rajan and Subramanian 2005). Not surprisingly, these concerns are most acute in already aid-dependent countries, such as Tanzania and Uganda, whose recent track records on growth, policy reform, and poverty reduction mean they are best placed to take advantage of donors' willingness to increase aid but, arguably, where there might be most to lose if further aid increases were to undermine long-run growth. In part, this anxiety reflects reservations about the absorptive and man- agerial capacity of overstretched public sectors to deliver higher public expenditure without a serious decline in quality, and in part it reflects deeper reservations about aid dependency and the impact of foreign aid on the domestic political economy (Adamand O'Connell1999; Svensson 2000). However, more traditional concerns about the macroeconomics of aid also figure large, and these are the focus here. Dominating these concerns is the fear that the Dutch disease effects of aid will inhibit development of the tradable goods sector and reduce growth in the recipient economy. Research has tended to focus on the tax-like distortion of aid or resource discoveries on the competitiveness of the tradable sector, typically where that sector enjoys learning-by-doing pro- ductivity effects (Van Wijnbergen 1984; Sachs and Warner 1995; Gylfason, Herbertsson, and Zoega 1997; Elbadawi 1999; Adam and OYConnell2004). This conventional perspective may be overturned when productivity spillovers accrue in both tradables and nontradables. In this paper, the case is examined in which public infrastructure investment generates an intertemporal produc- tivity spillover for both tradable and nontradable production, but in a poten- tially unbalanced manner.' For example, public investment in rural roads is likely to affect the production of (nontradable)food crops more than urban- based (tradable)manufactures, while the reverse is likely for, say, telecommu- nications infrastructure. A second source of concern is that the distributional effects of higher public expenditure may run counter to inequality and poverty-reduction objectives. There are two elements here. The first is that the immediate beneficiaries of higher public investment expenditure tend to be the nonpoor working in the services and manufacturing sectors as opposed to the poor producing mainly food and cash crops. The second is that if public expen- diture is devoted to infrastructure that enhances productivity in nontradable sectors, this may shift the domestic terms of trade against net producers of nontradables and, to the extent that the poor are located ih these sectors, 1. That productivity externalities accruing to the production of nontradables might reverse conven- tional Dutch disease results is not new. Torvik (2001),for example, makes the same point, although he does not explore specific mechanisms through which these externalities may emerge. Adam and Bevan 263 worsen the distribution of income. This is shown to be a distinct possibility in circumstances where preferences are nonhomothetic so that the income elasticity of demand for nontradable output (in this case bask food) is low.2 Section I outlines a simple two-sector, two-good model to highlight these Dutch disease effects in the presence of aid-financed public infrastructure invest- ment and a learning-by-doing production externality. This model is highly stylized, and so section I1 presents a more detailed, calibrated simulation model, loosely based on data from Uganda, which permits examination of the magnitudes likely to prevail in reality as well as of the distributional pressures likely to arise under alternative aid-financed public expenditure strategies. Two core versions of the model are examined-with public infrastructure as the only dynamic externality and with this mechanism interacting with a learning-by- doing externality that captures the productivity spillovers associated with increased nontraditional export production-and subjected to s~ensitivityanaly- sis. Both the core simulation results and the sensitivity analysis are discussed in section 111. The results suggest that for reasonable parameter valuesgoverningthe supply- side response to public expenditure, traditional Dutch disease effects are not present beyond the short run and are likely to be dominated in the medium term by the positive supply-side effects of aid. Somewhat paradoxically, growth in aggregate exports and total output in the medium term are strongest when the productivity effects of public investment expenditure are skewed in favor of nontradable production, reflecting the aggregate dynamic gaiins arising from improvements in nontradable supply. These effects remain even if the country is assumed to be well endowed with public infrastructure and its productivity on the margin to be relatively low. Moreover, these results rema.in qualitatively unchanged in the presence of plausibly scaled learning-by-doing externalities in nontraditional exporting. The simulation model also highlights important dis- tributional tensions that disadvantage rural households relative to urban house- holds and that may even lead to an absolute decline in rural incomes. It is difficult and rather tedious to set out and give intuition to the characteristics of a full-scale simulation model. This section takes a shortcut by developing a simple stylized model that highlights the key features embedded in the full-scale simulation model employed in the remainder of the article. Think of this as a model of the model. 2. The implications of this combination are also explored by Matsuyama (1992) in his analysis of industrial takeoff, where the low-income elasticity of demand for agricultural output allows agricultural productivity growth to generate both the labor surplus and the declining price of the wage good (food) that fuel industrialization. Consider a two-period Ricardo-Viner small open economy in which the representativeagent produces and consumesone nontraded good and one traded good. Private capital stocks are fixed and sector specific, and they do not depreciate, while a fixed endowment of labor, L, moves freely between sectors to equalize real consumption wages. The economy faces fixed external terms of trade, and there are no tariffs or taxes. Aid, represented by a fully fungible transfer of (tradable) resources, is the only international capital flow in the model. To focus on the mechanismsof interest, aid is received in the first period only, although in the simulation model applied in section 11,aid flows are treated as permanent. Total aggregate expenditure consists of private expenditure on tradable and nontradable goods and public expenditure on infrastructure. All values are expressed in terms of tradable goods, where PT = 1. Hence, defining the real exchange rate as PN/PT = Q and using superscripts P and G to denote private and government expenditure, the first-period income-expendi- ture balance is given by: where U is the private utility, K the public infrastructure capital, and A is the aid. E'(Q,u), EG(Q,K),and R(Q;L)represent the publicexpenditurefunction, private expenditure function, and the revenue function, respectively. Letting the supply and compensated demand functions for nontraded goods be RQ, E&, and EE , respectively, first-period equilibrium in the nontraded goods market is given by: Equations (1)and (2)imply that the trade balance is equal to the exogenous aid flow, thus, E;(Q, U) +E:(Q, K) - RT(Q;L) = A. Finally, the government budget constraint is defined as: The government's role in this model is simply the conversion of donor aid into public infrastructure. Since infrastructure is composed of tradable and nontrad- able goods, the quantity of public investmentactually realized will depend on the real exchange rate and the elasticity of substitution between tradable and non- tradable goods in investment demand.3Public investment takes place in the first period (atfirst-period prices) but augments productive capacity in either or both the tradable and nontradable sectors only in the second period.4 3. At thisstage, no restrictions are imposed on this elasticity, although the simulation model in section I1 assumesa Leontief structure for public investment demand. 4. Notice that in this model, the first-period equilibrium embodies a latent externality, in the sense that the public capital stock is not optimized. Implicitly, the government is assumed to lack access to the tax or borrowing instruments required to raise K sufficiently to exhaust the return from public capital. Adam and Beuan 265 This completes the characterization of the first period. Two potential extern- alities come into play in the second period. First, firms in both,the sectors may enjoy productivity gains from public infrastructure investment and, if forthcom- ing, these gains are sector specific but not appropriable by individual firms. Second, as noted in the introduction, an important strand in the debate about aid and Dutch disease has been the concern that aid-induced appreciations of the real exchange rate dilutes positive learning-by-doing externalities arising from tradable goods production. This is reflected by assuming that firms in the export sector benefit from learning-by-doing externalities, which, as with the effect of infrastructure, are not appropriable by individual firms. With second-period values denoted by lowercase letters, production in period 2 therefore depends on the real exchange rate, q, the size of the public capital stock, K, installed from period 1, and the volume of first-period exports, RT. Second-period GDP and sectoral equilibrium conditions are given by: It is assumed that r~ = qrqR + rtR > 0, where r , ~5 0 and r , ~ > 0. Spillovers therefore create their own biased shift in the production possibility frontier in the second period so that at fixed relative prices, the output of nontradables will fall in absolute terms (theRybczynski theorem) in the face of higher first-period tradable production. First-Period Equilibrium Equations (I),(2),and (3) fully determine the first-period equilibrium. Total differentiation of these three equations produces the following expressions for the proportional change in the real exchange rate, private utility, and public infrastructure in terms of the increase in aid, where a "hat" denotes a propor- tional change (see supplemental appendix S.1, available at htt~?://wber.oxford- journals.org) where EQ is total (private plus government) demand for nontradables, and EQQ>O, ACQc0,and AZac0 are the real exchange rate elasticitiesof supply and (private and government) demand for nontradables, respectively. The three parameters, 4 , y, and v, describethe composition of government expenditure: 4 is the share of government expenditure in total expenditure and y is the share of government expenditure on nontradables in total expenditure, so that (714) is the nontradable share in government expenditure and y is its share in total demand for nontra- dables. Finally, hP and hGdenote the (uncompensated) income elasticities of demand for nontradables of the public and private sectors. Expressions (7)to (9)deliver the standard demand-side Dutch disease results. First, notice that unless hP is very large relative to hG,QQ, and QQ, the expression for B will be positive. Letting AQQ= (1- y)AP yAG be the overall real exchange rate elasticity of demand for nontradab es, fQ+B wil71e positive provided5 Therefore, for reasonable values, an increase in aid will cause the real exchange rate to appreciate and will increase the first-period private welfare. The welfare result may at first seem counterintuitive, but the private sector is a net seller of the nontradable goods to the public sector so that the aid-induced real exchange rate appreciation generates a favorable movement in the private- public terms of trade. Finally, aid will increase public infrastructure as long as B > y(y/+)AG,which requires that Assuming AG> 0, this is a stricter condition than that required for increased aid to cause the real exchange rate to appreciate and increase private welfare although, for the reasons stated in note 3, this condition will be satisfied in most circumstances. 5. In the simulationmodel below,y w 0.10, q w 0.125 and q5 % 0.20,so that the first term on the right side scales the sum of the real exchangerate demand and supply elasticitiesby a factor of about 9.Since it is reasonable to expect that A' will be less than unity, then even if were very low B would still be positive. Adam and Beuan 267 In all three cases, the magnitude of these effects is determined by the structure of the economy. Consider, for example, the responsiveness of the real exchange rate to the aid inflow [equation (7)].The tendency for the real exchange rate to appreciate moderates the higher are CQQ, A& , and A& (in absolute value) but increases with the private and government income elasticitiesof demand for n~ntradables.~similar set of comparative static results can be derived for the A private welfare and public expenditure effects of aid. Since these are not of central importance here, they are not discussed. Notice that if there is no public investment response to the aid inflow [so that E ~ ( .=)0 in equation (1)and aid resourcesaccrue directly to the private sector as an income transfer], equation (3)disappears, yielding: and which confirm the simple demand-side results of a pure consumption transfer, which emerge from any standard model (Devarajan, Lewis, and Robinson 1993). In this case, the aid flow is strictly welfare increasingand will unambigu- ously cause the real exchange rate to appreciate, with the extent of the apprecia- tion being determined by the income elasticity of demand and the elasticities of demand and supply in the nontradable ~ e c t o r . ~ Second-Period Equilibrium The second-period equilibrium is derived in an analogous fashion by totally differentiating conditions (4)and (5)to solve for dq and du in terms of dK, the productivity of investment in the two sectors, and dRT, as follow^.^ First, notice that from the properties of conditions (4),(5),and (6),the value of the marginal product of infrastructure capital is given by r~ = qrqK+ r t ~Then, letting 0 = qeq/e . 6. In the case of the private-sector expenditure elasticity, the effect is unambiguous; in the case of the government elasticity, the responsivenessof the real exchange rate elasticity is increasing in AG provided condition (la)is satisfied. 7. Notice, also, that if public investment is entirely composed of tradables, so that y = 7= AG =0, the obvious result is that dQldA = dUldA = 0 and dKldA = l l E ~in , other words the aid inflow has no consequences for the first-period real exchange rate or private utility and public capital increases indirect proportion to the aid inflow. 8. The results that follow are expressed in terms of dK, the increase in public infrastructure, rather than solving for dK from expression (9) since from the second-period perspective, the relationship between the original aid flow and the volume of additional infrastructure it financed is immaterial. Though not done here, it would be a simple matter to solve the donor's optimal aid allocation as a function of the second-period productivity, given the donor's welfare function and budget constraint. be the share of nontradables in total expenditure, the following expressions are obtained for the changes in second-period utility: and in the second-period real exchange rate where /ZP is the second-period private-sector income elasticity of demand for nontradables, and o,, > 0 and 6 < 0 are the second-period real exchange q4 rate elasticities of supply and (private-sector) demand for nontradables, respectively.9 To interpret equations (15)and (16),consider first the case where there are no learning-by-doing effects (rR= rqR = TtR = 0). Three key results may be noted. The first is that the change in second-period utility depends on the value of the aggregate product of public capital; it does not depend on the presence or absence of any bias in productivity. Second, and by contrast, the evolution of the real exchange depends on the scale of infrastructure invest- ment and the relative bias in productivity spillover between the tradable and nontradable sectors. Thus, noting that (oqq- fiqq >0),it follows that the higher the impact on nontradable productivity, the more likely is the real exchange rate to depreciate, and the higher the impact on tradable productivity, the more likely is the real exchange rate to appreciate. Third, these effects are moderated by the income elasticity of demand for nontradables. For given values of rqK and rtK, the lower the income elasticity, /Zp, the weaker the tendency for the real exchange rate to appreciate. Specifically, solving equa- tion (16),it follows that and vice versa for q 0 are the second-period real exchange rate elasticities of supply and demand for tradabies. Adam and Bevan 269 The simulation model in section I1 considers only extreme-bias cases where alternately q r q = ~ 0 and r , =~0. In the first case, where productivity gains are located exclusively in the tradable sector, the real exchange rate will unambigu- ously appreciate for any non-negativeincome elasticity,while in the second case, where productivity gains are located exclusively in the nontradable sectors, condition (17) becomes: These three results highlight the principal aggregate effects of aid that are explored in the remainder of the article. They indicate that, other things equal, in the presence of productivity effects, the evolution of the equilibrium real exchange rate is, in general, ambiguous. However, where aid-financed public expenditure is targeted at improving the productivity of the nontradable sector and where income elasticities of demand for nontradable goods such as basic food are low, the initial appreciation is likely to be followed by a subsequent equilibrium depreciation of the real exchange rate. Consider, next, the effect of introducing the learning-by-doing spillover. In the natural case where rqR = 0 (learning by doing does not affect ithe productivity of the nontradable sector), it follows that with rtR > 0, an aid inflow that lowers the first-period net exports (so that dRT < 0) will lower the second-period welfare relative to equation (15) and will lead to a more appreciated real exchange rate (and hence a lower level of net exports) relative to equation (16).This effect is larger, other things equal, the higher is the income elasticity of demand for nontradables and the larger is the share of nontradables in total expenditure. Whether this second externality could reverse the sign of h or q will, of course, depend on the relative size of the two externalities and the changes triggering them (dK and dRT).As is shown in section 111, the positive effects flowing from public infrastructure investment dominate the negative learning-by-doing effects for reasonable calibrations of the simulation model, which is described in the next section. The analytical model is necessarily highly stylized. It assumes fixed private resource endowments and a highly simplified government structure, and it focuses only on aggregate production and consumption. The simulation model presented here provides a sense of the magnitude of the possible effects policymakers are likely to confront and unpacks some of the first- order distributional consequences of the aid and public expenditure interac- tion. This is a recursively dynamic real computable general equilibrium model of a small open economy calibrated to reflect the principal features of an archetypical low-income aid-dependent economy.10The equations of the model, along with the calibration data, are detailed in the supplemental appendix. Private Production and Consumption Producers and consumers are assumed to enjoy no market power in world markets, so the terms of trade are independent of domestic policy choices and are, for convenience, held constant across all simulations. Firms in each of the four sectors (foodcrop agriculture, cash crops, manufacturing, and services)are assumed to be perfectly competitive, producing a single good that can be sold to either the domestic or the export market. Production in each sector i is deter- mined by a Cobb-Douglas function of the form where S is land, KP is sector-specific private capital, KG is infrastructure, and L is a composite labor input. Only production in the rural sectors requires land, which is fixed in perpetuity. Private-sector-specificcapital stocks are fixed in each period but evolveover time through depreciation and gross investment.The labor composite, L, is constructed as a constant elasticity of substitution aggre- gation of skilled and unskilled labor, with fixed supplies that are intersectorally mobile. Labor markets are competitive so that composite labor is employed in each sector up to the point that it is paid the value of its marginal product. Private-sectoroutput is also determined by the level of infrastructure, KG, which is provided by the government. Constant returns to scale prevail in the private factors of production, but increasing returns are possible in the presence of public infrastructure. The distributional consequences of aid and public expenditure are tracked through their impact on three household types differentiated by factor owner- ship and patterns of consumption and saving. The first is a rural household, which is involved in food crop and cash crop agriculture and owns the land and capital in these two sectors. This household is outside the direct tax net and has zero net savings.'' The second is an urban unskilled household, whose only factor of production is unskilled labor, which it supplies to the manufacturing, services, and government sectors. It owns no capital or land and has zero gross and net savings but, in contrast to the rural household, it pays direct taxes. Finally, the urban skilled household supplies skilled labor to the manufacturing, services, and public sectors and owns the remainder of the capital in the econ- omy. This household pays direct taxes to government at a higher rate than the 10. The underlyingsocial accounting matrix isloosely based on data from Uganda around the turn of the century but offers a reasonable representation of many similar cash crop agriculture-based economies. 11. The rural household's grosssavingsare constrained to beequal to the depreciation of agricultural capital. Adam and Beuan 271 unskilled household, earns interest on its net holdings of government domestic debt, and has positive net savings in the initial equilibrium. Consumption for each household type is defined by a constant elasticity of substitution linear expenditure system, which allows for the income elasticity of demand for different goods to deviate from unity. In the simulations reported in the next section, attention is restricted to the case where only food consumption is subject to a subsistence threshold. This implies that the marginal income elasticity of demand is less than unity for food and greater than unity for all other goods (manufacturedgoods and services).12 Macroeconomic Closure and Dynamics The default is a neoclassical closure in which total private investment is con- strained by total savings net of exogenous public investment, where household savings propensitiesare exogenous.This rule, broadly consistent ~ ~ iconditions t h in the poorest countries where unrationed access to world capital markets is virtually zero and domestic private saving is relatively interest inelastic, means that the shortfall of government savings relative to the cost of government capital formation, net of exogenous foreign savings, directly crowds out private investment (and the excess of government savings directly crowds in private investment). There is a risk, however, that this closure rule exaggerates the private investment response to public investment (either positively or negatively). Therefore, the simulation experiments are also run under an alternative closure that defines an independent, return-sensitive, private aggregate invest- ment function and that allows for the marginal savings propensity of the urban skilled household to adjust endogenously. As discussed in more derail in the next section, the key insights delivered by the simulation model are not greatly altered by the choice of closure rule, but for completeness simulations for the alternative Kaldorian closure rule are reported in the supplemental appendix S.111 available online. The model has a simple recursively dynamic structure. Each solution run tracks the economy over 10 periods from the initial policy change, and each period may be thought of as a fiscal year. Within-year public and private capital stocks are fixed, and the model is solved given the parameters of the experiment (e.g., the change in aid flows and the corresponding public expenditure response). This solution definesa new vector of prices and quantities for the economy, including the level of public- and private-sectorinvestment, which feed into the equations of motion for sectoral capital stocks: 12. Since cash crops are produced solely for export, final household consumption is defined over food, manufactures, and services only. where Ki= {KPi,KG}, pidenotes the sector-specific rate of depreciation, and j measures the gestation lag on investment. In the simulations presented below, the default setting is j = 1, although the effects of assuming that public invest- ment augments the stock of infrastructure capital only with a longer lag are also examined. To focus exclusively on the impact of increased aid flows on the economy, the model is calibrated to an initial static steady-state equilibrium in which net public and private investment is zero (gross investment exactly matches depreciation),and there is no growth in the labor supply. The final element is the learning-by-doing externality. Learning by doing is assumed to generate a Hicks-neutral innovation to total factor productivity in the manufacturing sector (the nontraditional export sector). Specifically, equa- tion (20) assumes that Ait = Ai for nonspillover sectors, while in the spillover sector, denoted s, total factor productivity evolves according to where Ef = Cim_, @Et-jis the (discounted.)sum of exports in the spillover sector up to and including t - 1 under the simulation experiment, and E; is the correspondingly defined cumulative exports under the baseline trajectory for the economy. The term $ >0 measuresthe extent of the spillover, fi = (1+ p)-l < 1 is the discount factor, and Asois the value of As, in the baseline calibration. Hence, the higher the p, the lower is the impact of experience on current productivity, but for any p < oo, there will always be some persistence in (In~f - InEf)so that temporary policy reforms will have at least some perma- nent consequence for productivity. Aid and Government Expenditure To focus on the principal mechanisms of interest, aid is assumed to accrue to government and is used exclusively to finance increased public investment expenditure.13Two further assumptions are made. The first is that an increased public capital stock entails a higher level of operations and maintenance (O&M) expenditure. This is calibrated on the basis of evidence on the recurrent expenditure requirements of World Bank- financed capital projects compiled by Hood, Husband, and Fu (2002).Recur- rent O&M is set to 3.5 percent of the additional capital stock (the Hood, Husband, and Fu weighted average across all projects). A higher O&M rate is also considered corresponding to Hood, Husband, and Fu's highest estimated rate of 7.5 percent (for education). The baseline assumption is that these additional O&M costs are financed out of the additional aid flow so that the 13. Hence, there is no examination of the consequences of changes to the structure of taxation, the level of reserves,or the volume of real recurrent expenditure (other than those arising directly from public investment, such as operations and maintenance; see below), all of which are kept constant across all simulations. Adam and Bevan 273 domestic budget deficit is (ex ante) unchanged. In the sens~ltivityanalysis reported in the supplemental appendix S.111, the case is also examined in which aid flows finance only the installation of public capital, and O&M expenditures are met through increases in the domestic budget deficit.14 In both cases, it is assumed that the government takes into account price changes in determining the volume of expenditure that can be financed with the addi- tional aid. Second-order changes to household incomes, demand, and relative prices arising from inframarginal government activities are not, however, internalized in the government's decisions, so the experiments are not necessa- rily budget neutral ex post, even when O&M costs are aid financed. The second assumption concerns the nature of public investment expenditure. In keeping with much of the evidence on Poverty Reduction Strategy Papers, the baseline simulations assume that aid-financed increases in public investment expenditure are more intensive in nontradable inputs on the margin than for both private investmentsand inframarginal government expenditure. Scaling-up is therefore assumed to skew aggregatedemand toward nontradablesin the short run. The sensitivity analysis considers two key variations, however. One allows for the possibility that public investment demand is less nontradable intensive than average (which may be the case when government infrastructure investment is geared toward, say, upgrading telecommunications technologiesor employing more technical assistance).The other is that an aid-financed scaling-up of public investment entails additional demand for labor, beyond that entailed by O&M requirements reflecting, for example, greater management and coordination burdens placed on the public sector. Finally, although it is reasonable in practice to assume that public investment in areas such as health and education will augment human capital and thereby generate a source of extensiveproductivity growth, these effects are not included in the simulations. This feedback is relativelyslow, and the simulations reflect a medium term in which adjustment to the physical capital stock takes place but where changes to the human capital stock have not yet materialized.15 This section describesthe simulation experiments and discusses the core simula- tion results and the sensitivity results. 14. There are of course other ways in which this issue could be handled. The first is through matching increasesin domestictaxes, and the second is to set the levelof O&M as a choice variable, with the rate of effective depreciation of the public capital stock being a (negative) function of the level of O&M expenditure.In this case, failing to meet 0&M requirements serves to accelerate the rate of depreciation of public infrastructure. 15. The model by Aginor, Bayraktar, andAynaoui (2005),bycontrast,providesan explicittreatment of the links between public investment and human capital formation. Simulation Experiments The policy experiment consistsof an increase in public infrastructure investment financed by a permanent 12.5 percent increase in the net (grant)aid inflow to the economy. This increase is equivalent to just under 2 percent of baseline GDP, a step increase roughly equivalent to the size of the increase in net aid flows to Mozambique, Tanzania, and Uganda at the end of the last century related to the Heavily Indebted Poor Countries Debt Initiative. In all cases, the additional aid flow is used exclusively to finance an increase in public infrastructure invest- ment, holding tax rates and all other components of publicexpenditure (withthe exception of O&M expenditure)constant. Consequent changes in the domestic budget balance after grants therefore reflect general equilibrium effects arising from the increased public spending and are accommodated through adjustments to private saving or investment, depending on the macroeconomic closure rule. Table1summarizes the core simulations and a set of variations around these. The potential simulation space is vast: in principle, each core simulation can be implemented under any of the variants and many combinations of the variants. TABLE 1. Simulation Experiments Experiment Description Core simulations 1 No productivity spillovers from infrastructure capital 2 Neutral productivity spillovers 3 Export-biased productivity spillovers 4 Domestic-biased productivityspillovers 5 As 4, with subsistence threshold for food Variants Learning-by-doingspillover = 0.20 Learning-by-doing spillover = 0.45 Learning-by-doing spillover = 0.00 with three-year gestation lag on public investment Learning-by-doingspillover = 0.20 with three-year gestation lag on public investment Elasticity of substitution between skilled and unskilled labor = 0.50 Elasticity of substitution between skilled and unskilled labor = 2.00 Public investment demand is tradable-good intensive High initial public capital (KG = 75% of optimal value and spillover = 0.25) Operations and maintenance (O&Mi= 0.0% O&M = 7.5% O&M = 3.5% plus additional public-sectorlabor demand O&M = 3.5% financed through higher domestic budget deficit As1with additional public-sector labor demand Kaldorian closure As m with Kaldorian closure "Reported in supplementalappendix, availableat http://wber.oxfordjournals.org. Adam and Bevan 275 Only a small number of simulations illustrating the key features ad the results are presented here; the sensitivity analysis offers some support to their robustness. The core simulations are presented in table 2. Simulation 1 is the benchmark. The infrastructure investment has no effect on private-sector productivity: the economy's total capital stock is increased, but the increased public capital does not sustain higher private output. This allows the pure demand-sideeffects of the aid flow to be isolated. Simulation 2 examines the case where the infrastructure investment enhances private-sector productivity, but the effects are uniform across all sectors of the economy and are represented by a balanced outward shift in each sector's production possibilityfrontier between domestic (nontradable)and export (trad- able) variants of the good. The remaining permutations on the basic experiment (simulations3-5)exam- ine three central cases in which the productivity impact is still felt across all sectors but now embodies a bias such that within each sector the shift in the production possibility frontier is skewed in favor of either export- or domestic- !good production. Specifically, only the "extreme-bias"cases described in equa- tions (17) and (20) are considered, represented by a rotation in the frontier around either end point. Simulation 3 considers the case of an export bias in the productivity effects of government infrastructure, and simulations 4 and 5 a domestic-good bias. Simulations 1-4 assume that the subsistence component in consumption is zero so that the consumption is homothetic in income across all goods and households. In simulation 5, however, a subsistence component is imposed for food consumption so that the income elasticity of demand for food falls below one. These runs are all based on a common set of assumptions, which are varied in subsequent simulations to assess the robustness of the central findings. The core assumptions and principal variations (reported in table 1)are explained below, where each variation can be applied to any or all of the core simulations. First, to reflect the idea that there is ofiten a severe shortage of (functional) infrastructure capital in countries to which this model applies, public infrastruc- ture capital stock is assumed at only half its "optimal" value.16 Second, there is very little empirical consensus on the size of the productivity effects of infrastructure investment in low-income countries. The assumed value for this parameter is a~ = 0.50, comparatively higher than the values estimated in Hulten's (1996) study of infrastructure capital and economic growth. This higher baseline value reflects in part the expectation of a higher marginal product of public capital for countries with a severely depleted capital stbck 16. The optimal public capital stock is defined as the amount at which the marginal product of public capital is equal to the average marginal product of private capital, given the initial endowments of private factors (land,labor, and capital) and the assumed parameters of the production function. At such a point, the output gain from a tax- or deficit-financed increase in infrastructure capital would exactly offset the loss arising from the crowded-out private capital. TAB LE 2. Simulation Results of the Effect of a 12.5 Percent Increase in Net Aid Flows (percent) Experiment Period 1 2 3 4 5 Productivity biasa Neutral Export Domestic Domestic ~ l ~ h a ~ ~ 0.5 0.5 0.5 0.5 EPSLc 0.99 0.99 0.99 0.99 O & M ~ 3.5 3.5 3.5 3.5 Initial public capital as percent of "optimal" 50.0 50.0 50.0 50.0 public capital' Subsistence food share in consumption 0.00 0.00 0.00 0.90 f Change in KG (at initial prices) 2.4 2.4 2.4 2.4 14.3 14.2 14.3 14.1 Prices and quantities Export-weighted real -2.8 -2.8 -2.8 -2.4 exchange rateg -1.1 -2.1 0.5 1.7 -0.2 -1.3 1.3 3.0 Total exports -6.9 -6.9 -6.9 -6.4 1.4 0.7 2.4 8.5 8.7 7.7 10.1 18.5 Manufacturing exports -6.6 -6.6 -6.6 -6.5 1.2 1.9 0.3 -1.5 8.6 9.1 8.1 5.3 Cash crop exports -7.7 -7.7 -7.7 -7.0 1.1 -0.8 3.7 12.6 8.7 6.4 11.9 24.2 Real GDP 0.08 0.08 0.08 0.10 4.16 4.01 4.37 4.71 7.64 7.37 8.01 8.57 Private investment -3.7 -3.7 -3.7 -3.4 7.4 5.1 10.4 14.7 16.8 14.2 20.3 26.2 (Continued) TA BLE 2. Continued Experiment Period 1 2 3 4 5 Domestic budget balance (percent of GDP) Real disposable income Rural Total h, -4 -4 Factor markets (average real wage; walcpi) Skilled Unskilled "Denotes whether the productivity enhancement is neutral (Neutral)or biased toward domestic production (Domestic)or export production (Export). b~lasticityof public infrastructure in private production. "Eiasticityof substitution between skilled and unskilled labor. *operations and maintenance (O&M)costs (as percentage of additional capital stock). 'Size of initial infrastructure capital stock relative to optimal given initial private capital stocks and labor. *1ndicatesthe presence of a sector-specific subsistence level of consumption (as percentage of baseline consumption). gThereal exchange rate is defined as (pelpd)where pe denotes the domestic price of exports and pd denotes the price of domestic goods, so that negative values indicate an appreciation. Note: All experiments consider a permanent increase in net aid inflows of 12.5 percent, equivalent to 1.97 percent of initial GDP. Values reported as changes relative to baseline except for fiscal measures, which are reported as percentage points of GDP. Source: Authors' analysis based on model described in text. and in part the likelihood that the contemporary marginal productivity of public infrastructure expenditure may in fact be higher than the historical point esti- matessuggest. Both the size of the initial public capital stock and its productivity are exogenous parameters of the model, and both could be either high or low. The sensitivityanalysis reported in the supplemental appendix S.111examinesthe robustness of the core results to the case where the economy is endowed with a larger infrastructure capital stock and a lower return on the margin (denoted as variant "h").17 Third, scaling-up is assumed to entail additional O&M expenditure equiva- lent to 3.5 percent of the increase in the public capital stock, with this additional expenditure met from aid inflow. The sensitivity analysis allows for lower and higher values of the O&M rate, for additional public labor demand in support of scaling-up, and for an alternative mechanism for financing this additional expenditure. These possibilities are reflected in variants "i" to "m." Fourth, it is initially assumed that there are no dynamic externalities to exporting. The evidence base for the learning-by-doing spillover is far from conclusive, and the sensitivity analysis experiments with plausible alternative values. The central value for the elasticity of total factor productivity in the manufacturing sector with respect to nontraditional exports [the parameter in equation (22)]is set to + = 0.20. Since this value is highly contested, a value of + = 0.45 is also considered. These possibilities are reflected in variants a and b. Fifth, the initial simulations set the elasticity of substitution between skilled and unskilled labor to unity [so that equation (20)becomes Cobb-Douglasin all factors]. This assumption is then relaxed by examining the cases where the elasticity of substitution is low (oL= 0.50) and where it is high (oL= 2.00). These possibilities are reflected in variants e and f. Lastly, infrastructure investment is initially assumed to augment the capital stock with a lag of one year. The sensitivity analysis considers the case where public infrastructure investment has a longer gestation, taking three years to augment private productivity. These possibilitiesare reflected in variants c and d. For each experiment, the impact effect (year1)and the cumulative evolution of the economy after 5 and 10 years are reported. To simplify the presentation, the focus is on changes in only a small number of key aggregates: the export- weighted real exchange rate, the volume and composition of exports, real GDP, private investment, the fiscal balance, and the real disposableincome of the three household types, measured in terms of the household-specificconsumption price index. For a given level of government expenditure, real disposable income is a direct measureof household welfare. Figures1-4 also report the evolution of the real exchange rate, total exports, total real disposable income, and the rural household's share in this. 17. The sensitivity analysis not reported here suggests that the variation in the behavior of the economy between these points is fairly regular. Adam and Bevan 279 FIGURE 1. Export-Weighted Real Exchange Rate Response to Aid-Financed Public Investment Simulation period +Base case +Neutral +Export bias ++ Domesticbias FIGURE 2. Total Export Response to Aid-Financed Public Investment Simulation period +Base case +Neutral +Export bias ++ Domesticbias FIGURE 3. Real Household Disposable Incomes in Response to Aid-Financed Public Investment Simulationperiod tNeutral +Exportbias *Domestic bias FIGURE 4. Rural Share in Total Income in Response to Aid-Financed Public Investment Simulationperiod -r-Basecase tNeutral +Exportbias +Domesticbias Adam and Beuan 281 Results UNPRODUCTIVEINFRASTRUCTURE. Experiment 1 provides a benchmark. Here, the infrastructure investment confers no benefits on private productivity so that in terms of the model in section I, q r , = ~r& = 0. Hence, the aid flow has little initial impact on GDP, but it does lead to an appreciation of the export real exchange rate and a sizable contraction in exports in favor of higher production of domestic goods. In contrast to the endowment model of sectionI, the evolutionof the simulated economy over the medium term points to a progressive deterioration in overall economic performance because of the decline in real private-sector investment. This reflects both a decline in total savings as the fiscal balance deteriorates, in turn a reflection of the adverse effects of the real exchange rate on the budget,18 and the effect of the real exchange rate appreciation on the cost of capital goods (since capital formation is intensive in nontradable services). This means that although the appreciation of the real exchange rate moderates over time, the deterioration of the capital stock ensures that the decline in export performance does not reverse and hence that initial welfare gains weaken over time. Finally, while total real income increases, rural households actually suffer a decline in income, both absolutely and relative to urban households. The princi- pal reason is that the demand effectsfrom increasedgovernment expenditure fall disproportionately on urban skilled and unskilled labor and on intermediate goods from the manufacturing and servicessectors. Backward linkagesfrom the formal urban sectors (manufacturing, services, and government) to the rural sectors (food and cash crops) are extremely weak. As later results show, these demand effects may be largely offset when the aid inflow is used productively but may re-emerge and be exacerbated when relative price effects turn against the rural sector and the income elasticity of demand for food is low. PRODUCTIVEINFRASTRUCTURE. By contrast, in experiments 2-5, government infrastructure investment raises private-sector productivity. In experiment 2, this productivity effect is uniform across sectors and between production for domestic and export markets. There is now fairly substantial cumulative growth in GDP over the 10 years, some improvement in the fiscal balance, and a marked increase in private investment.19As a consequence, while the impact effect? on the real exchange rate and on exports are identical to those in experiment 1, because of the time lag before the productivity effects kick in, the impacts diverge sharply over time. Virtually all of the real exchange rate appreciation 18. Since government in this model is a net seller of foreign exchange, the real exchange rate appreciation reduces the domestic value of the budget balance and therefore increases the domestic financing requirement. 19. Government revenue grows as real incomes and expenditures grow while, after the inltial step change, real government spending does not. Savings available for private investment grow partly withGDP but also because of crowding-in from the improvement In the fiscal balance. It is a consequence of the closure rule mentioned earlier that these resources are duly invested. has been unwound by the end of the 10-year period. Moreover, even though the real exchange rate remains somewhat appreciated relative to its baseline value, the initial 6.9 percent fall in export volumesis reversed, moving to an 8.7 percent increase over the baseline by the end of the simulation. While the impact effects on household incomes are the same as in the previous experiment, so that rural income again initially falls, matters now improve over time. Not only is total real income 6.5 percent higher over the long run, but rural householdsenjoy asimilarincreasein realincomeovertimeinthisexperiment,even though their proportionate gain is slightly lower than that of urban households. Experiments 3 and 4 consider the outcomeif the productivitygainswitnessedin experiment 2 are biased toward the production of either tradable (exportable)or nontradable (domestic)goods. In the case of experiment 3, while the productivity effect is again positive and uniform acrosssectors, it is now biased within the food and manufacturingsectors in favor of export production. As expected, when there is no increase in the productivity of nontradable production, this leads to a more appreciated path for the real exchange rate than in the neutral experiment 2. Hence, although manufacturing export performance is stronger because of the productivity bias, traditional cash crop exports are hit relatively hard, some 2.3 percentage points lower than when productivitygains are neutral. When the productivitygain is biased entirely toward productionof the domestic good, as in experiment 4, outcomes are markedly different.The bias in production (whichincreasesthesupply of nontradablegoods)issufficientlystrongto morethan offset the demand effectsof the increased aid flows so that the initial real exchange rate appreciationis reversedwithinfive years.20The effects on exports are symme- trical with those in experiment 3; cash crop exports recover more strongly than in earlier experiments, but the domestic bias in manufacturingproductivity resultsin a more sluggish recovery in manufacturing exports. Overall export performance is stronger with a domestic biasthan with an export bias, reflectingthe real deprecia- tion induced by the domestic bias. The domestic-biasedsupply response also leads to a larger improvement in the long-runfiscalbalance(of0.8 percentagepointsof GDP), reflectingfavorable relative price movements (seenote 19) as well as the effects of higher growth and invest- ment than in either the neutral or export-biasedforms of productivity growth. The most striking difference between these two experiments, though, is the effect on real household disposable incomes. Compared with the case of a neutral supply response, a strong export bias in the productivity gain induced by infrastructure expenditure sharply moderates real income growth in the economy. Long-run total income rises only 4.1 percent over its baseline com- pared with 6.5 percent when the supply response is neutral between export and domestic production. However, the income gain is spread somewhat differently 20. The model in section I premcts that the real exchange rate change should be exactly zero. That it is not so in the simulation model reflects its richer structure, including the fact that the government budget is not invariant to changes in the real exchange rate. Adam and Bevan 283 across household groups, with urban unskilledworkers now doing less well than the other two groups. This contrasts sharply with the domestic-biased supply response, which generates a markedly higher aggregate real income gain of 9.9 percent in the long run but one that is disproportionately skewed in favor of urban households. As noted above, demand-side effects imply a tendency for urban households to gain disproportionately from aid-financed increases in infrastructure because of low backward linkages from government expenditure to the rural sector of the economy. The relative price movements underpinning experiment 4 exacerbate theseweak linkages.As the economy's increased ability to producedomesticgoods reverses the real exchange rate appreciation, this shifts the domestic terms of trade in favor of those consuming the now relatively cheaper domesticgoods (allhouse- holds) and against those producing them (therural household). Rural households thus share more or less equally in the consumption gain from lower cost domestic goods but share disproportionately in the income loss from producing them. In experiments2 and 4, these adverse distributionaleffects are weak enough that theyonly partiallyoffset rural households' share in the aggregateincomegainfor the economy.This is not thecase, however, in experiment5.Thisexperimentrepeatsthe previousone but assumesa highsubsistencerequirementin food consumptionfor all households. The implication is that, once households have met this requirement, positive income gains will be allocated disproportionately away from food expen- diture so that on the margin the income elasticity of demand for food will be less than unity-increasinglyso the higher is the subsistencethreshold and vice versa for the other sectors. The effect of this adjustment to assumed consumer behavior is dramatic. After its impact appreciation, the real exchange rate depreciates sharply and becomes more depreciated over time. Similarly, export volumlesincrease sub- stantially after their initial fall, as do the fiscal balance, private investment, and real GDP. In allcases, the gains aregreater than in any of the other experiments. Thesame holdsfor aggregate real income, which increases by 10.6 percentover the baselinein the long run. The distributionalimpactin this experiment is rather unpleasant,though. Urban householdsenjoysubstantialrealincomegains becauseof the declinein food prices, while rural households experience large income falls. The reason is simple: the adverse shiftin the internal termsof trade against rural households inoted in experi- ment 4 is magnified by the low-income elasticityof demand in food consumption from all households.As net producers, rural householdssuffer twice over: the fall in food prices caused by the increase in supply is exacerbated by the weakness in the demand for food because of the low-income ela~ticity.~' 21. The size of these effects clearly reflects the subsistence threshold; the lower the subsistence food share in private consumption, the larger the local income elasticity of food and the smaller the quantitative difference between experiment 5 and experiment 4. Although the effects are not everywhere proportional, reducing the subsistence share in food consumption from 90 to 45 percent produces a simulated outcome that lies roughly mid-way between experiments 4 and 5 regardless of which variants are examined. Learning by Doing and Gestation Lags Table 3 introduces two factors that might be expected to modify the results presented in table 2. First, a learning-by-doing externality is introduced from nontraditional (manufactured) exports. This externality is assumed to be symmetric, in the sense that while cumulative growth in exports relative to the baselineaugments total factor productivity in manufacturing, sluggish export performance reduces it.22The same five basic cases are retained as in table 2, combined with four variations involving learning by doing and gestation lags, indexed by the letters "a" to "d." Thus, letter "a" always refers to a variation with a learning-by-doing elasticity set at $ = 0.20 but no other changes from the assumptions of table 2. Variation "b" is similar, except that this elasticity is set at the very high level of $ = 0.45. Variation "c" reverts to setting the learning-by-doing spillover at zero but increases the gestation lag on public investment to three years. Variation "d" combines a learning-by-doing elasti- city of 0.2 with the three-year lag. The first point to note is that simply inserting learning by doing into the unproductive case (experiment la) has fairly strong adverse effects. What was a relatively slow deterioration in the original unproductive case is now acceler- ated. The costs of acceptingaid but then wasting it are markedly raised. Turning to the productive cases, two features stand out. The first is that, as the model in section I anticipates, this second spillover pulls in the opposite direction to the infrastructure effect, at least over the horizon of these simulations. Second, however, even at what is arguably a rather high learning-by-doing elasticity of $ = 0.20, the "positive" impact flowing from the aid-funded infrastructure investment still dominates. For example, when productivity effects are neutral, the learning-by-doing effect lowers medium-term (10-year) GDP growth only from 7.6 to 7.3 percent and total real income growth from 6.5 to 6.2 percent (experiment 2a compared with experiment 2). Obviously, manufacturing exports bear most of the cost (fallingfrom a medium-term growth of 8.6 percent to one of 5.1 percent), but this is partly offset by stronger growth in traditional exports. As experiment 2b indicates, however, a substantially larger learning-by- doing elasticity $ = 0.45 would inhibit recovery in manufacturing exports (still down by 3.1 percent after 10 years) and knock a further 1 percent off growth in real GDP. The second reason why the results in table 2 may be seen as painting a relatively positive picture is the assumption that public infrastructure invest- ment in period t augments the public capital stock in t +1. Experiments c and d 22. Since zero total factor productivity growth is assumed in the baseline, this relative decline manifests itself as an (ratherunrealistic) absolute decline in total factor productivity. This has no material bearing on the qualitative nature of the results, however. TA BLE 3 . Simulation Results of the Effect of a 12.5 Percent Increase in Net Aid Flows: Learning by Doing and Gestation Lags (percent) Experiment Period 1a 2a 2b 2d 3a 4a 5a 5c 5d Productivitybiasa Neutral Neutral Neutral Export Domestic Domestic Domestic Domestic ~ l ~ h a ~ ~ 0 0.5 0.5 0.5 0.5 0.5 0.5 0.5 0.5 EPSLc 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 0.99 O & M ~ 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 3.5 Initial public capital as percent of optimal 50.0 50.0 50.0 50.0 50.0 50.0 50.0 50.0 50.0 public capital' Subsistencefood share in consumption 0.00 0.00 0.00 0.00 0.00 0.00 0.90 0.90 0.90 f Public capital gestation lag (year@ 1 1 1 3 1 1 1 3 3 Learning-by-doing spilloverh 0.20 0.20 0.45 0.20 0.20 0.20 0.20 0.00 0.20 Change in KG (at initial prices) t = 2 2.4 2.4 2.4 2.4 2.4 2.4 2.4 2.4 2.4 Ll t = 1 0 14.0 14.3 14.3 12.7 14.2 14.3 14.1 12.7 12.7 Prices and quantities Export-weighted real exchange rate' to t =1 -2.8 -2.8 -2.8 -2.8 -2.8 -2.8 -2.4 -2.4 -2.4 to t = 5 -2.3 -1.2 -1.5 -1.7 -2.3 0.3 1.6 1.0 0.8 to t = 10 -3.0 -0.6 -1.3 -1.2 -1.6 0.8 2.5 2.6 2.0 Total exports t o t = l -6.9 -6.9 -6.9 -6.9 -6.9 -6.9 -6.4 -6.4 -6.4 to t = 5 -7.9 0.9 0.3 -3.2 0.3 1.9 7.8 3.5 2.7 to t = 10 -10.3 7.7 5.3 4.3 6.9 8.7 16.4 15.6 12.6 Manufacturing exports tot =1 -6.6 -6.6 -6.6 -6.6 -6.6 -6.6 -6.5 -6.5 -6.5 to t = 5 -9.5 -0.5 -2.8 -4.9 0.4 -1.6 -3.6 -4.8 -7.3 to t = 10 -16.9 5.1 -3.1 -1.0 6.4 3.5 -0.9 3.2 -6.0 Cash crop exports tot =1 -7.7 -7.7 -7.7 -7.7 -7.7 -7.7 -7.0 -7.0 -7.0 to t = 5 -7.2 1.7 2.5 -2.3 -0.3 4.4 13.2 6.9 7.7 to t = 10 -5.2 9.9 12.6 8.8 7.3 13.4 25.9 21.0 23.5 (Continued) TAB LE 3. Continued Experiment Period 1a 2a 2b 2d 3a 4a Sa 5c 5d Real GDP t o t = l 0.08 to t = 5 -0.31 t 0 t = 1 0-1.25 Private investment to t =1 -3.7 to t = 5 -5.0 to t = 10 -8.1 Total factor productivity (manufacturing)jto t = 1 -0.2 to t = 5 -1.6 to t = 10 -6.3 Domestic budget balance (%of GDP) to t = 1 -0.47 w to t = 5 -0.53 00 0\ to t = 10 -0.68 Real disposable income Rural to t =1 -1.6 to t = 5 -2.1 to t = 10 -3.5 Urbanunskilled t o t = l 2.7 t o t = 5 2.3 to t = 10 1.5 Urban-skilled t o t = l 2.0 t o t = 5 2.0 t o t = 1 0 2.1 Total to t = 1 0.5 0.3 to t = 10 -0.4 (Continued) TAB LE 3 . Continued Experiment Period l a 2a 2b 2d 3a 4a 5a 5c 5d Factor markets (average real wage; walcpi) Skilled t o t = 1 2.7 2.7 2.7 2.7 2.7 2.7 3.2 3.2 3.2 t o t = 5 3.2 5.7 5.9 4.5 2.4 10.2 24.5 20.4 20.5 to t = 10 4.4 8.0 8.8 7.8 4.6 12.8 31.7 30.0 29.8 Unskilled t o t = 1 0.9 0.9 0.9 0.9 0.9 0.9 1.0 1.0 1.0 t o t = 5 0.5 4.5 4.3 2.7 1.8 8.5 12.9 10.4 10.0 to t = 10 -0.6 7.5 6.7 6.3 4.6 11.6 17.9 17.3 15.8 "Denotes whether the productivity enhancement is neutral (Neutral)or biased toward domestic production (Domestic)or export production (Export). b~lasticityof public infrastructure in private production. 'Elasticity of substitution between skilled and unskilled labor. doperations and maintenance (O&M) costs (aspercentage of additional capital stock). 'Size of initial infrastructure capital stock relative to optimal given initial private capital stocks and labor. f~ndicatesthe presence of a sector-specific subsistence level of consumption (as percentage of baseline consumption). gGestation lag for public infrastructure investment. h~earning-by-doingparameter tj[see equation (22)l. iThe real exchange rate is defined as (pelpd) where pe denotes the domestic price of exports and pd denotes the price of domestic goods, so that negative values indicate an appreciation. 'Percentage change in As$[see equation (22)l. Note: All experiments consider a permanent increase in net aid inflows of 12.5 percent, equivalent to 1.97 percent of initial DP. Values reported as G changes relative to baseline except for fiscal measures, which are reported as percentage points of GDP. Source: Authors' analysis based on model described in text. allow public investment in t to augment the capital stock only in t + 3. This rather naturally elongates the "J-curve" effects seen throughout these simula- tions for exports and lowers the rate of GDP and real income growth but does not eliminate the recovery in total exports or the growth in income (e.g., in experiment 2d, the growth in total exports is roughly halved relative to experiment 2). It is worth highlighting one important feature of the results for all variants of experiment 5: the decline in rural incomes in these experiments is immediate and persistent. In contrast to what is happening elsewhere in the economy, it is the demand effects rather than the supply factors that drive rural incomes in both the short and the medium term. This is seen very clearlyfrom the fact that changes in supply-side factors across all variants of experiment 5 alter the pattern of rural incomes very little indeed. Sensitivity and Robustness Checks The sensitivity of these central results to a battery of robustness checks is discussed in the supplemental appendix S.111. Broadly speaking, these experi- ments suggest that the qualitative character of the simulations is unchanged by such variations in specification as the effects of altering the elasticity of substitu- tion between skilled and unskilled labor, varying the nontradable intensity of public investment, altering the assumed initial endowment and productivity of public infrastructure capital, changing the assumptions concerning O&M expenditures, and altering the closure rule. IV. SUMMARY AND CONCLUSIONS Six key conclusions emerge from the simulations presented in this article. First, when public infrastructure augments the productivity of private factors, and when there is an initial scarcity of public infrastructure, there are potentially large medium-term welfare gains from aid-funded increases in public invest- ment, despite the presence of short-run Dutch disease effects of aid. Second, the dynamic and distributional consequences of this investment are highly sensitive to the location of productivity effects and the characteristics of demand. Third, the presence of a domestic bias in the aggregate supply response (experiments 4 and 5) is broadly beneficial to the economy in terms of boosting aggregate growth and investment, welfare, and exports and moderating appre- ciation of the real exchange rate. Fourth, across most experiments, particularly when there is a domestic-good bias in the supply response, the rural household does not share proportionately in the aggregate income gains to the economy. In particular, if a domestic bias in productivity is combined with a high subsistence requirement in food (experi- ment 5),the economy as a whole enjoys a large supply response that dominates Adam and Bevan 289 the other cases, but at the cost of falling rural incomes and a sharp worsening in the income distribution. Fifth, there are potentially substantial payoffs through an ilmproved fiscal balance and increased private investment, regardless of the presence or absence of bias (experiments 2-5). Lastly, the results suggest that while it is certainly possible to identify config- urations of parameters such that aid-funded increases in public investment leave the econorny worse off than without aid, this requires very low values for the productivity of public expenditure in circumstances where the public capital stock is already very close to its optimum and high values of the learning-by- doing externality. These conclusions must, of course, be qualified by a number of caveats. First, the modeling of the labor market here has been vestigial. In particular, it permits no migration from rural to urban sectors so that improved productivity in traditional agriculture becomes problematic, condemning rural households to declining real incomes rather than stimulating migration into the urban tradable sectors. Similarly, there is no scope in the model for rural households to shift to tradable forms of production. Second, the model does not allow for any form of human capital formation. Future work will extend the model to address both these shortcomings. However, one general conclusion can be drawn from the analysis with con- siderable confidence: serious analysis of the impact of aid must pay close atten- tion to supply-side issues, which are likely to be specific to the uses to which aid is put. It should not seem paradoxical that a proper assessment of the macro- economic impact of aid depends closely on the underlying microeconomics of the associated public expenditures it finances. Adam, C. S., and S. O'Connell. 1999. "Aid, Taxation, and Development in Sub-Saharan Africa." Economicsand Politics 11(3):225-53. 2004. ''Ad Versus Trade Revisited: Donor and Recipient Policiesin the Presence of Learning-by- Doing." EconomicJournal 114(492):150-73. Agtnor, P.-R., N. Bayraktar, and K. El Aynaoui. 2005. "Roads Out of Poverty? Assessing the Links Between Aid, Public Investment, Growth, and Poverty Reduction." Policy Research Paper 3490. World Bank, Washington, D.C. Commission for Africa. 2005. Our Common Interest. London. Devarajan, S., J. D. Lewis, and S. Robinson. 1993. "External Shocks, Purchasing Power Parity, and the Equilibrium Real Exchange Rate." World Bank Economic Review 7(1):45-63. Elbadawi,I. 1999. "External Aid: Help or Hindrance to Export Orientation in Africa?"Journal ofAfrican Economies 8(4):578-616. Gylfason, T., T. T. Herbertsson, and G. Zoega. 1997. "A Mixed Blessing: Natural Resources and Economic Growth." CEPR DiscussionPaper 1668. Centre for Economic Policy Research, London. Hood, R., D. Husband, and F. Yu. 2002. "Recurrent Expenditure Requirements of Capital Projects." Policy Research Working Paper 2938. World Bank, Washington, D.C. Hulten, C. 1996. "Infrastructure Capital and Economic Growth: How Well You Use It May Be More Important than How Much You Have." NBER Working Paper 5847. National Bureau of Economic Research, Cambridge, Mass. Matsuyama, K. 1992. "Agricultural Productivity, Comparative Advantage, and Economic Growth." Journal of Economic Theory 58(2):317-34. Rajan, R., and A. Subramanian.2005. "What UnderminesAid's Impact on Growth?"IMF Working Paper 05/126.International Monetary Fund, Washington, D.C. Sachs, J. D., and A. M. Warner. 1995. "Natural Resource Abundance and Economic Growth." NBER Working Paper 5398. National Bureau of EconomicResearch, Cambridge, Mass. Svensson,J. 2000. "Foreign Aid and Rent Seeking." Journal of International Economics 51(2):437-61. Torvik, R. 2001. "Learning by Doing and the Dutch Disease." European Economic Review 45(2): 285-306. United Nations Millennium Project. 2005. Investing in Development: A Practical Plan to Achieve the Millennium Development Goals. New York. Van Wijnbergen, S. J. 1984. "The 'Dutch Disease': A Disease after All?"EconomicJournal 94(373):41-55. Infrastructure, Externalities, and Economic Development: A Study of the Indian Manufacturing Industry Charles R. Hulten, Esra Bennathan, and Sylaja Srini:lvasan If infrastructure tends to generate spillover externalities, as has been the assumption in much of the development literature, one may reasonably look for evidence of such indirect effects in the accounts of manufacturing industries. Empirical support for this assumption has so far been ambiguous. This analysis of Indian data, however, reveals substantial externality effects from the states' infrastructure to manufacturing produc- tivity. The analysis separates the direct effects of roads and electricity, as mediated by the infrastructure services purchased by manufacturing industries along with other intermediate inputs, from the indirect effects, as measured by the impact of infrastruc- ture capacity on the Solow productivity residual. In the 20 years from 1972 to 1992, growth of road and electricity-generating capacity seems to have accounted for nearly half the growth of the productivity residual of India's registered manufacturing. In what are now classics in the theoretical literature on growth and economic development, infrastructure investments are associated with significantspillover externalities, with benefits that accrue outside the target area of the investrhent (Young 1928; Rosenstein Rodan 1943; Hirschman 1958). This view also)fits well with endogenous growth theory that sees externalities as the sourck of endogenous feedback effects on output growth (Romer 1986; Lucas 1988; Barro 1990). Empirical support for the existence of significant infrastructure externalities has, however, been far from unanimous. Aschauer's estimates (1989a, 1989b) of the macro effects of infrastructure investment support,the hypothesis of significant spillovers. Not so the estimates of the infrastruckure Charles R. Hulten is a professor of economicsat the University of Maryland and a research associQteat the National Bureau of EconomicResearch;his emailaddress is hulten@econ.umd.edu.Esra Bematha! is an emeritus professor of economicsat the University of Bristol, UK; his email address is ebemathan2@aol!com. Sylaja Srinivasan is a senior economist at the Bank of England; her email address is saUy.srinivasan@danko fengland.co.uk. This amcle reports on a part of the research produced under a World Bank research pTject, "Growth and Productivity Effects of Infrastructure: Regional Study in India" (RPO 681-S4),funded Ijy the Research Support Budget and sponsored jointly by the Public Economics Division, Development ~eskarch Group, Infrastructure Group, and the South Asia Department. This work was done while Sylaja Srinipsan was at the University of Southampton, UK. The viewsexpressedin this article are those of the authors and do not necessarily reflect those of the institutions with which they are affiliated. THE W O ~ BDANK ECONOMIC REVIEW, VOL. 20, NO. 2, pp. 291-308 doi:10.1093/wber/lhj007 Advance Access publication May 4,2006 O The Author 2006. Published by Oxford University Press on behalf of the International Bankfor Reconstruction and Development/ T= WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org. effects based on a growth-accounting model. In that model, spillovers from infrastructure should show up as increases in total factor productivity. Young (1992,1995)found only a limited role for total factor productivity as a source of growth in four East Asian economies, implicitly limiting the role of infrastruc- ture spillovers operating through increased productive efficiency. Other studies, however, have found significant total factor productivity effects for Japan (Nishimizu and Hulten 1978) and East Asia (Hsieh 1999). None of these growth-accounting studies links infrastructure explicitly to growth externalities. This step was taken by Hulten and Schwab (1984, 1991, 2000) in their studies of regional total factor productivity in U.S. manufacturing. Modifying the conventional growth-accounting model to isolate the effect of infrastructure externalitieson growth, they found no evidence of externalities in explainingthe growth of total factor productivityof U.S. manufacturingindustry. This article joins this debate by investigatingthe effects of infrastructure in the manufacturing sector of a large low-income developing economy. India experi- enced a rapid increase in highway infrastructure and electricity-generatingcapa- city during 1972-92, and when the Hulten-Schwab framework is applied to data developed from the Indian Annual Survey of Industries (ASI) for these years, the resultssuggest that, unlike the case of U.S. manufacturing, spilloversaccount for a large part of total factor productivity growth. Because the focus is on the industrial development effects of infrastructure, infrastructure is placed directly in the production function for manufacturing output as an unpaid factor of production (Meade 1952). A general form of the production function can be written as: where Q denotes gross output, B infrastructure stock, K privately owned (non- infrastructure) capital, L labor input, and M intermediate inputs. The term A(B,t)is a standard Hicks-neutral efficiency function that allows for exogenous shifts in the production function. This technology may exhibit diminishing, constant, or increasing returns to scale.' 1. Grossoutput isusedin equation (I),becauseit isthe appropriate measureof theactual product made by manufacturing firms-the actual quantity of textiles, steel, autos, and so on. Value added is, in fact, a measureof primary input (Hulten2001).Real valueadded is sometimes used in industry studies, becauseit is easily related to aggregateoutput. This is rarely an appropriate procedure for productivitystudies, because it requires the assumption of a separable production function. [Pradhanand Barik (1998)tested for separability with Indian industrial data and concluded that the value added function did not exist.] Furthermore, the value-addedapproach assumes that the efficiencychange affects only labor and capital but not intermedate inputs (a dubious assumption, at best, and inappropriate in the current context because of the role of intermediate inputs in transmitting the effects of infrastructure). However, to facilitate comparison with other research, results are displayed based on the value-added approach at various points in t b article. Hulten, Bennathan, and Sriniuasan 293 This specification allows infrastructure systems to affect the manufacturing industry through two channels. First, changes in the stock of in:rastructure are reflected in th; intermediate input variable, M(B).The intermediate goods and services purchased by manufacturing firms include transportation services and electricity,each produced by its own industry with its own particular infrastruc- ture. An increase in either road capacity or electricity-generatingcapacity tends to lower the cost of producing the corresponding services and, from the stand- point of the manufacturing sector, lowers the price paid for road transport and electricity-the cost of acquiring M(B).This is the market-mediated effect of infrastructiure on manufacturing output. The second channel through which an increase in the stock of infrastructure capital may affect manufacturing output is through an outward shift in the ~icksianefficiency term, A(B,t),-caused by the efficiency-promoting external- ities associated with an increase in infrastructure, which may lead to an increase in manufacturing output. This is the type of effect envisioned by the endogenous growth literature and by the new economic geography. Lower 1:ransport costs, for example, may lead to economies of scale and agglomeration and to better inventory management. Similarly, an increase in electricity-generating and dis- tribution capacity promotes continuous supply and more stable voltage and thus allows more sophisticated machinery in the manufacturing industry and reduces the need for firms to provide their own generating capacity (Leeand Anas 1992). These are nonmarket-mediated infrastructure effects, operating through the A(B,t)channel in the manufacturing production function.-a his second channel permits isolating the spillover effects from the market-mediated effect of infra- structure and measuring them using well-established techniques for estima~ting the Hicksian shift term. Assuming that the Hicksian efficiency term and its components are multiplicative, production function 1 can be written as with subscript t denoting time and i denoting region, the dimensions relevant for the empirical work. The parameter Ace indicates the initial level of productive efficiency, hiis the exogenous rate of productivity change, and the parameter yi measuresthe infrastructurespillover effect and is assumed to be constant over time but can vary among regiom2The infrastructure stock, B,, has both a time and a 2. The cost of producing output tends to fall, other things remaining equal, as the production function shifts outward. Thus, under conventional restrictions on the production function, -y will also be the elasticity of the cost of production with respect to infrastructure: each 1 percent increase in the stotk of infrastructure reduces cost by y percent, other things remaining equal. The time-shift term, h, is usually interpreted as the average rate of costless technical change. However, it is well known that h also includes the effects of omitted variables. Explicit allowance has not been made for human capital variables, education and health, nor implicit allowance for the "quality" of labor input. These effects may therefore be suppressed into the h term. However, to the extent that expenditures for education and health are correlated with expenditures for hard infrastructure, the infrastructure spillover term, y, could pick up some of the human capital effects. 294 THE W O R L D B A N K E C O N O M I C REVIEW, VOL. 20, N O . 2 regional dimension, with the index i referring to the region in which the stock is located. This formulation can be expanded to allow the infrastructure in other regions to generate cross-regionalspillovers. Although the problem of estimating the externality parameter y is not familiar in the empirical growth analysis, the problem of estimating the Hicksian shift term, of which y is a component, has been solved by the Solow (1957) model of residual total factor productivity growth. Total factor productivity is defined as the ratio of output to the direct inputs used in production. At the level of the manufacturing sector, intermediate inputs also have to be included, in addition to labor and capital. The relevant ratio (fortotal productivity) is then TPi,,= Qi,J F(Ki,,,Li,,, Mi,,).Substituting from equation (2)shows that total productivity is directly associated with the parameters of interest in the analysis This expression, in logarithmic form, is the basis for the estimate of the infrastructure externality parameter, y. The variables on the right side of equa- tion (3),infrastructure and time, can be measured directly. The left-side variable, total productivity, must be estimated. The first step in estimating TPi,,follows Solow in measuring productivity as the residual output not attributable to the inputs of labor, capital, and inter- mediate inputs. Analytically, the Solow residual is the growth rate of output less the growth rates of these inputs weighted by their shares in total cost (xK,TL, xM).This procedure yields the expression AlnQ AlnK AlnL AlnM - AlnTP --- (4) - TK--~L-- TM- At At At At At . For ease of exposition, the time and region subscripts associated with each variable are omitted here. When input prices are assumed to be proportional to marginal products, the output elasticities of K, L, and M are equal to the corresponding cost shares, and the residual measures the shift in the production f~nction.~Each item on the right side of equation (4) can be measured or imputed from published data, yielding an estimate of total productivity growth 3. As shown, equation (4)is a continuous time differential equation, referred to in the literature as the growth rate of the Divisia Index of total productivity. In the actual computation of the Solow residual, the discrete time Tornquist approximation to the Divisia index is used, in which case the shares (xK,TL,rM) are estimated using the average share from 1 year to the next [e.g., (112)(T ~ , ~ , ~ - I ) ] , + x ~ , ~ , i ~and the growth rates by the year-to-year change in the logarithm (e.g., AlnQi,t = InQi,t- InQi,t - I ) . Hulten, Bennathan, and Srinivasan 295 that can be used, in the context of equation (3),to estimate the infrastructure externality parameter, y. The data needed for implementation of the Solow residual [equation (4)]are obtained from India's ASI, which include annual estimates of gross output, intermediate inputs, labor input, and the book value of capital stocks for "registered" manufacturing firms by state.4 Numerous empirical studies of Indian manufacturing have used this data set, and these stuclies formed the starting point for the aggregate and regional estimates of output and productiv- ity reported here. The ASI data refer to manufacturing firms registered under the Factory Act, which are the larger manufacturing enterprises.The 1992/93survey presentsdata for 24 industries that include both manufacturing and other related activities. The surveys are constructed from a probability sample, with large firms enumerated every year but smaller firms included according to a sampling probability. Thus, the time series constructed from the ASI does not necessarily represent the same firms over time, and year-to-year sampling variation can introduce volatility. The estimates used in this study are ba~sedmainly on industry data at the one-digit (all manufacturing) level of detail. Studies of productivity and, more generally, of the structure of production require estimates of output and input in constant (real)prices. The ASI data are in current prices and therefore require deflation. Alternative approaches have been hotly debated (Ahluwalia 1991, 1994; Balakrishnan and Pushpangadan 1994, 1995; Dholakia and Dholakia 1994, 1995; Rao 1996a, 1996b).The Divisia-type deflation approach used by Rao (1996a, 1996b) is applied here, because it is consistent with equation (4),but unlike in Rao, the wholesale price index (WI)is not used here as the deflator for intermediate goods. The WI approach to price deflation combined with the other data yields a suspicious pattern of total productivity growth, with dramatic growth from 1973 to the mid-1980s and then a sharp and prolonged decline through the early 1990s.~A prolonged inward shift in the manufacturing production function is intuitively implau~ible and hard to square with the events of that period, which include the movement toward liberalization of planning controls. Based on an empirical regularity in the data on manufacturing output and inputs, in which output and material inputs tend to be highly collinear, the procedure applied here is therefore to assume that the ratio of intermediate input to output quantity is constant over time for each two-digit manufacturing 4. See Hulten, Bennathan, and Srinivasan (2000)for a more detailed description,of the data used in this study. 5. Such a pattern is often associated with mismeasurement in one or more data series. When the price deflation strategy used here is substituted for the WI, the resulting pattern of total productivity shows a steady and moderate increase over the entire period. The endpoint, however, is the same in both price deflation approaches. See Hulten and Srinivasan (1999)for further discussion. industry in each Indian state. This assumption is a matter of expediency rather than choice, forced by the data, but the resulting price indexes do capture changes in the mix of industrial production over time. Moreover, the procedure produces the same result as the WPI approach for the period as a whole but changes the pattern in the years between the endpoints of the sample period. For this period, the Indian manufacturing industry seems to deviate from the finding by Kubo and others (1986)that intermediate inputs rise faster than output in the manufacturing sectors of industrializing economies. The difference may be due to the difference in procedures, because Kubo and others used input-output data. In any event, this assumption permits deriving price deflators for inter- mediate goods that vary over time and across regions. Real gross output grew at a respectable average annual rate of more than 7 percent a year for the two decades of the sample (table 1).Growth in inputs explains most of the growth in output, with material the primary explanatory factor. Growth in materials was steady over the period at roughly the growth rate of output (hardly surprising in view of the way it was estimated), and the share was almost 80 percent of total product. Labor's share was smaller than that of capital (although some of the return to capital is probably a return to the labor of entrepreneurs and family), and it declines in the second half of the period. Of greater significance is the low rate of growth of labor input (total employment)and its decline from 2.8 percent to 1.4 percent in the second half of the period. Capital grew faster and accelerated in the second half of the period. The 0.5 percent annual growth in total productivity may seem unusuallysmall but recall that it measures the impact of innovation and infrastructure invest- ment on a very broad base of inputs. Most other studies express the result of TABLE 1. Sources of Growth of Gross Output in the Indian Manufacturing Industry (AverageAnnual Percentage Rates of Growth) Source of growth 1973-92 1973-82 Gross output 7.3 7.2 Materials 7.4 7.3 Labor 2.1 2.8 Capital 6.8 5.9 Total input 6.8 6.7 Total productivity 0.5 0.5 Material'sshare 77.9 77.2 Labor's share 9.0 10.0 Capital'sshare 13.1 12.8 Note: Detail may not sum to total because of rounding. Source: Authors' analysis based on data from India's Annual Survey of Indus- tries;see descriptionin text. Hulten, Bennathan, and Srinivasan 297 TABLE 2. Sources of Growth of Real Value Added in the Indian Manufacturing Industry (AverageAnnual Percentage Rates of Growth or Ratios) Source of growth 1973-92 1973-82 1983-92 Real value added 7.1 6.8 7.5 Labor 2.1 2.8 1.4 Capital 6.8 5.9 7.7 Total factor input 5.0 4.6 5.3 Total factor productivity 2.2 2.2 2.1 Ratio total factor productivity 31 32 28 to real value added Labor's value added share 41 44 37 Capital's value added share 59 56 63 Note: Detail may not sum to total because of rounding. Source: Authors' analysis based on data from India's Annual Survey of lndus- tries; see description in text. total productivity change in terms of its value-added counterpart, total factor productivity, which equals total productivity divided by the sum of capital'sland labor's share of income (in effectdividing total productivity by about 0.20). To facilitate comparison, we also calculated total factor productivity estirriates (table 2). At 2 percent, the Solow total factor productivity residual is of a more conventional magnitude. The acceleration in real value-added growth in the second half of the period is apparent, but again it is caused by the increase iri the capital-labor ratio not in productivity. IV. TOTAL PRODUCTIVITY BY REGION Tables 1 and 2 present estimates of the annual growth rate of productivity (rota1 productivity and total factor productivity) for registered manufacturing for all the states of India combined and therefore do not have a regional dimension. Because transport and electricity generation and transmission systems are het- works with a spatial dimension, differences in productivity and infrastruciure across geographic regions are a potentially important source of variation that should not be excluded a priori when trying to pin down spillover effects. Consequently, sources of growth estimates were calculated for each ofthe 16 states in the regional sample (table 3). As table 3 summarizes, there was much regional variation in the growth rates of real gross output and total productivity. Growth rates by themselves give an incomplete picture of the comparative growth dynamics of the various states. It is also important to know the levels of total productivity. For exam- ple, the situation in which a state has a more rapid rate of productivity growth than another state and starts with a lower level of productivity is quite different from the case in which a state has both a higher growth rate and a higher initial TAB LE 3. Average Annual Growth of Output and Productivity and Relative Productivity Levels in Manufacturing Industry in India's States 1973-92 - Average annual growth, Total productivity 1973-92 (percent) level State Gross output Total productivity 1973 1992 Andhra Pradesh Bihar Gujarat Haryana Himachal Pradesh Jammu and Kashmir Karnataka Kerala Madhya Pradesh Maharashtra Orissa Punjab Rajasthan Tamil Nadu Uttar Pradesh West Bengal Rank by 1973 level of total productivitya Top five Middle five Bottom five "Excludes Kerala because, alone among the states, real value-added growth in manufacturing was negative in the first part of the sample, an oddity that could well have been caused at least in part by sampling variation and deflator problems. Source:Authors' analysis based on data from India's Annual Survey of Industries; see descrip- tion in text. level.of productivity. In the first case, productivity levels in the two states are converging, whereas in the second case, one state is pulling away from the other. These estimates of the level of total productivity are obtained using the translog index procedure developed by Jorgenson and Nishimizu (1978) and extended by Caves, Christensen, and Diewert (1982). This method computes total productivity in each state in some base year as the output of the state relative to the output index for all of India, less the private inputs in the state relative to the all-India index, weighted by the relative cost shares: TPi Qi - Ki - Li Mi In-= In- -xKln- -xLln- -~ ~ l n - TP* Q* K* L* M* where Hulten, Bennathan, and Sriniuasan 299 Because total productivity is an index number, it must be normalized to the base value of some year and place. The initial year of the sample is used for the base year, and the average level of total productivity across all states is used as the base place (set equal to 100). This gives the initial conditions, A,,o, in equations (2)and (3).The is then "grown" by the rate of total productivity growth calculated using the Solow residual [summarized in column (2)of table 31. The result is an estimate of the level of total productivity by state and year that can be used as the left-side variable in a regression based on equation (4)to produce estimates of y and h that exploit differences in infrastructure among states and over time. Columns (3)and (4)of table 3 present summary estimates of the relative level of total productivity for each state for 1973 and 1992 using the Caves- Christensen-Diewertmethod. Maharashtra started the sample period with the highestlevel of total productivity and Himachal Pradesh with the lowest, lagging 27 percent behind. Growth rates of total productivity, however, were not even across states [column (2)],and total productivity levels were converging by the end of the sample period [column (4)].To show this, we grouped states into terciles according to their initial levels of total productivity. (Kerala is omitted, because, alone among the states, real value-added growth in manufacturing was negative in the first part of the sample, an oddity that could well have been caused at least in part by sampling variation and deflator problems.) The bottom five states ranked by this criterion experienced a more rapid rate of both output and total productivity growth, implying that the states that started with the lowest levels of total productivity had narrowed (but not eliminated) the gap with the leaders by 1992. This finding is in accord with the findings of Mitra, Varoudakis, and V6ganzoni.s (1998), who report convergence of total factor productivity across India's two-digit-level industries over a similar time period (but with a wider list of industries). The parameters of equation (3)are estimated by regressing the natural logarithm of the annual estimates of total productivity levels by state on the natural logarithm of each state's own infrastructure, time, and a coristant term. In addition, the correction terms suggested by Hall (1988)are included to allow for the possibility of increasing returns and for departures from marginal cost pricing. The Hall modifications appear in specification (3) as the additional variables lnK, and X = T K ln(M1K)+ nln(L1K);the corresponding regression coefficients areE - 1and p -1. Constant returns toscale obtain when E - 1is 0, increasing returns when it is greater than 0, and decreasing returns when it is less than 0. The parameter p is equal to 1 when price equals marginal cost and greater than 1 when price exceeds marginal cost.6 The resulting equation is This equation is the basic regression model for estimating infrastructure elasticity, y. Time and region subscripts have been omitted for clarity, but the variables lnTP, lnB, lnK, and InX have both a time dimension (20 years) and a regional dimension (16 states), and 1nA is the logarithm of the initial level of productivity in each region. The parameters of equation (7) are estimated from the total productivity statistics and the data underlying table 3, combined with data on the infrastruc- ture systems of interest: paved roads and electricity. Data on paved roads are from annual issues of the Ministry of Transport's Basic Road Statistics of India. The variables consist of lengths of several categories of paved roads: national highways (arterialroads for interstate movement),state highways (arterial roads for interdistrict movement, linking up with national highwaysand adjacent state highways),and district roads (otherPublic Works Department roads). Adequate data on road capacity (lanes)were not available. For electricity, time-series data on generating capacity in megawatts are from the energy reports of the Centre for Monitoring Indian Economy, which cover both the state utilities (State Electricity Boards) and centrally controlled capacities. State road lengths were normalized by state area and generating capacities by census data on state populations. To calculate rates of return from the regression estimates of equa- tion (7),we derivedestimates of the construction cost per kilometer for each type of road and of the cost per megawatt of generating and transmission capacity for 1985, based on World Bank data and project information. The infrastructure variables display strong growth over the sample period. The correlation between the main road variable (national and state highways) and time is 0.76, and the correlation between time and electricity capacity is 0.96. There is also a high degree of correlation between the two infrastructure variables, at 0.72, suggesting the possibility of a multicollinearity problem. VI. REGRESSION RESULTS The parameters of the infrastructure-productivity link expressed in equation (7) were estimated using the sample of 320 observations-20 years and 16 states (table 4). A fixed effect approach was used to allow for differences in the initial 6. The Hall (1988) corrections are needed for two reasons. First, the standard Solow residual [equation (4)]is computed under the assumption of constant returns to scale, which limits the generality of the approach in economies such as India, where scale economies are a potential source of growth. Second, the markup term helps ameliorate the possibility of noncompetitive pricing, noted by Tybout (2000) as characteristic of markets in developing economies. TABLE 4. Determinants of Total Productivity in Indian Manufacturing: Parameter Estimates of Basic Model Variable (1)" (2)" (3)" (4)" (sib (6)' Scale variable 0.038 (4.12) 0.033 (3.55) 0.033 (3.53) 0.030 (3.15) -0.015 (-1.21) -0.011(-0.89) Markup variable 0.082 (7.31) 0.086 (7.64) 0.083 (7.43) 0.086 (7.69) 0.076 (5.44) 0.079 (5.66) Time 0.004 (4.81) 0.003 (4.53) 0.002 (2.52) 0.002 (2.58) 0.007 (4.40) 0.003 (1.22) Log of highway variable 0.044 (2.71) 0.039 (2.37) 0.038 (1.90) 0.059 (1.92) w s Log of electricity variable 0.024 (2.19) 0.019 (1.76) 0.022 (1.26) 0.068 (1.91) R2 0.809 0.814 0.812 0.816 0.901 0.901 "Estimated using annual data for 1973-92 for 16 states (320observations). b~stimatedusing annual data for 1981-90 for 16 states (144 observations). 'Estimated using annual data for 1981-90 for 16 states but includes adjacency variable. Note: The dependent variable is the log of total productivity. The numbers in parentheses are t-statistics. State fixed effects are not shown. Source: Authors' analysis based on data from India's Annual Survey of Industries and Centre for Monitoring Indian Economy; see description in text. levels of technical efficiency (A) among the states and different initial endow- ments of infrastructure. The baseline regression without any infrastructure vari- ables [column (I)] reveals a slight degree of increasing returns to scale (3.8 percent) and an 8.2 percent markup of price over marginal cost. The estimate of the scale parameter is within the range reported by Fikkert and Hasan (1998) in their study of scale elasticity in a panel of Indian manufacturing firms. The implied rate of pure technicalchange is 0.4 percent. All estimates are statistically significant at conventional levels. The introduction of the national and state highways [column (2)]produces a statistically significant estimate of the key spillover elasticity (y) of 4.4 percent. Inclusion of highways lowers the estimated rate of technical change but generally has only a small effect on the size and significanceof the other parameters.7 The introduction of electricity (by itself, without highways) yields an estimated spil- lover elasticity (y)of 2.4 percent and lowers the rate of technical change but has little effect on the other variables [column (3)].Inclusion of highways and elec- tricity together [column (4)]produces estimated ys that are smaller than either of the separate estimates (and the electricity y is marginally significant).The strong intercorrelation between infrastructure variables and time may be at work here, but it should be noted that the two ys yield a combined elasticity of 5.8 percent. Without consistent time series of the two-digit industries within each state before 1980, state-specificprice indexes could not be constructed for intermedi- ate goods by the methods discussed earlier, and therefore, the corresponding all- India price had to be used for the period before 1980. Equation (7)was thus re-estimated using the more reliable data from 1980 onward, omitting the recession years of 1991 and 1992 to avoid a cyclical bias in the estimates. The results reveal constant or slightly decreasing returns to scale during this period, with little change in the markup estimate [column (5)of table 41. The rate of technical change is appreciably higher, but although the point estimates of infrastructure elasticities of highways do not change much, their statistical significance is lower, possibly reflecting multicollinearity. The joint infrastruc- ture parameter remains about 6 percent. The analysis has thus far assumed that spillover effects occur within the boundaries of each state, with no allowance for spillover effects to neighboring states. However, one obviousrole of highways is to connect the various regionsof a country to promote commerce and population movement. Moreover, India's electricity grids extend across state boundaries, and some states get substantial quantities of energy from their neighbors. Variants of the basic approach were used to attempt to capture such superstate indirect effects. First, neighborhood indirect effects, yii, which allow for spilloversfrom a neighboring state j to state i, 7. Experiments with different definitions of the road variables found that those with a broad regional reach produced results similar to those reported in table 4. However, the measure of district roads yielded estimates that were statistically insignificant, suggesting that this sort of road does not generate the indirect spillover effects in the manufacturing sector associated with national and state highways. Hulten, Bennathan, and Srinivasan 303 are distinguished from own indirect effects, output elasticities, y", which capture spillovers within state i. The inclusion of 7'' in the analysis does not imply double counting when the states are summed to an all-India total but rather that the whole is greater than the sum of its parts viewed in isolation. Several neighborhood approaches were attempted. An "extended neighbor- hood" definition of infrastructure adds the infrastructure in immediately adja- cent states to a state's own infrastructure. Estimates of y" + for highwaysand electricity for the shorter period 1980-91 comparable to that in column (5)are summarized in column (6).The implied spilloverelasticitiesare much larger than in previous cases-a combined 12.7 percent-although the levels of significance are quite thin. The rate of technical change falls from 0.7 percent for the period to 0.3 percent, and it becomes statistically insignificant. Again, this implausible result probably reflects multicollinearity, which increases when adjacent infra- structure is included in the analysis. VII. THE SIZE OF THE SPILLOVER EFFECT Because total productivity is used as the measure of productive efficiencyrather than total factor productivity, the estimated spillover elasticities, y, appear absolutely small. It is therefore instructive to compare them with the size of the implied output elasticities of private capital and labor employed in manu- facturing. These can be approximated by their value shares, .ir~and KL,which are 13.1 percent and 9 percent, respectively (seetable 1).Compared in this way with the "private" inputs, the combined indirect infrastructure elasticity of 5-6 percent is by no means small. Another way to assess the relative importance of the infrastructure spillovers is to compare marginal products. The marginal product of infrastructure, AQ/ AB, can be computed from the corresponding elasticities that are equal to (AQ1AB)BIQ. In the regressions, B is measured as a physical quantity. To express the physical stock in constant prices (to be symmetrical with the estimates of Q), 1985 unit values were used (obtained from World Bank sources) and extrapolated to other years. Estimates of the marginal product of labor and private capital in manufacturing were also developed using similar methods. These estimates of marginal product can be interpreted as the gross of depreciation return to the different types of capital in terms of manufacturing output. The results are summarized in table 5 using the results from coluhns (21, (3),and (4)of table 4. The gross rate of return to highways rises over the sample period as does the overall return to the infrastructure aggregpte, whereas electricity and private capital exhibit a more or less constant return. By 1992, the combined return to infrastructure was 9 percent, almost one-third of the direct return to private capital. Recall that the return to infrastructure is an indirect effect, over and above the return attributable to the direct use of highways and electricity. 304 T H E WORLD BANK E C O N O M I C REVIEW, VOL. 20, N O . 2 TABLE 5. Comparison of Gross Marginal Product: All-India Average for Manufacturing Industry (Average Gross Return Per Rupee of Capital) - - - Highways alone 0.02 0.04 0.05 Electricity alone 0.05 0.05 0.05 Highways and electricity 0.06 0.07 0.09 Private capital 0.29 0.26 0.29 Note: Detail may not sum to total because of rounding. Source: Authors' analysis based on data from India's Annual Survey of Indus- tries and Centre for Monitoring Indian Economy; see description in text. A third way to assess the importance of the indirect infrastructure effect is to examine its relative contribution to the growth of the overall total productivity residual. The first stage of the sources of growth approach decomposes the growth of output into the contributions of labor, capital, intermediate inputs, and residual total productivity, as in table 1. Equation (7)permits a second stage in which total productivity is further decomposed, first, into the indirect effects of highways and electricity capacity, a pure time effect (costless technical change), and second, into the errors that arise from assuming that returns to scale are constant when they are not, an error made by assuming price equals marginal cost when it does not, and a pure residual error. Table 6 quantifies this second decomposition by multiplyingthe means of the variables in equation (7)by the corresponding elasticity estimates from table 4. TABLE 6. Decomposition of the Growth Rate of Total Productivity: All-India Average for Manufacturing Industry (Average Annual Percentage Growth Rates) Using estimates of table 4 column True productivity Highways Electricity Subtotal Scale effect Markup effect and residual error Subtotal Total productivity Note: Detail may not sum to total because of rounding. Source: Authors' analysis based on data from India's Annual Survey of Indus- tries and Centre for Monitoring Indian Economy; see description in text. Hulten, Bennathan, and Srinivasan 305 This yields the mean percentage contribution to total productivity of the infra- structure, scale, markup, and "true" productivity variables, with a residual error term accounting for the balance. Resultsare shown for the estimates of columns (2),(3),and (4)of table 4 (the base case 1973-92, without adjacency effects). For the combined effects of highways and electricity summarized in the last column, the average annual growth rate of the true productivity residual is 0.47 percent, of which highwaysand electricityaccount for nearly half. This suggests that infrastructure is an important contributor to productivity growth and hence to a reduction in the cost of production. VIII. FINAL REMARKS Most macro studies of infrastructure and economic growth have sought to measure the overall impact of infrastructure on growth. This analysis differs by attempting to isolate the component of the overall impact that is associated with the indirect effects of infrastructure, as they affect Inldia's registered manufacturing sector. These indirect effects are traditionally viewed, in the broader literature on economic development, as an important dimension of industrialization. However, while widely assumed to be important, such externalities have not been quantified at the industry-wide level of growth. The findings here suggest, first, that these externalities exist and, second, that they are an important part of productivity growth in India's modern manufac- turing industry. When investments can be shown to have external effects, standard reasoning sees this as evidence of underinvestment. If this conclusion is to apply to the results here, it can hold only within the strict limits of the study. The physical infrastructure found to be exerting indirect effects on the productivity of registered manufacturing will have been laid down for the benefit of a wider constituency than registered manufacturing. Thus, nothing can be said about infrastructure externalities in general or in terms of gross benefits and certainly not in terms of net benefits, which would have to account for environmental impacts. With the meaning of the study thus narrowed, what are the implications of the results for industrialization policy, specifically investments intended to support the growth of India's modern manufacturing sector? From the strict point of view of this sector, the results suggest that there has been underinvestment in the kinds of infrastructure covered in the study. Still left open is the question of the point at which the results of this analysis ought to enter the process of India's investment decision. Externalities are not generally considered in project evalua- tion, but World Bank project criteria require successful infrastructure projects to pass a 12 percent (internal) rate of return test. The indirect rates of return summarized in table 5 are one-half to three-quarters of this test rate, suggesting 306 T H E W O R L D BANK E C O N O M I C REVIEW, VOL. 20, NO. 2 that conventional project analyses may lead to significant underinvestment in infrastructure systems. More research is needed. The inadequacy of the data, particularly on intermediate inputs, is fully acknowledged, and with that the possibility of biases arising from model mis- specification. Also, because the analysis is based on industry data at the one-digit level, aggregation biases due to changes in the composition of the manufacturing industry within any state cannot be ruled out.8 These problems are common to many econometric studies and are likely to be even greater when models devel- oped for high-income industrial economies are applied to a developing economy. Moreover, the fact that the infrastructure systems are lumpy networks of inter- locking investments may cause specification problems. The network feature suggests, for example, that the externality parameter, y, may change over time as the network evolves rather than remaining fixed over the period of analysis. Also, there is a tendency to build capacity in advance of demand, causing a divergence between the measures of the stock of infrastructure and the corre- sponding flows that determine the volume of output. There may also be impor- tant omitted variables, such as human capital, that are highly correlated with expenditures on roads and electricity and whose effects may be present in the estimates of y. However, these caveats notwithstanding, the results are quite consistent with the micro evidence from Lee and Anas (1992), and with the assessments of Ahluwalia (1998,2000)and Acharya (2002),as well as with the overwhelming impressionistic evidence about the inadequacy of transport and electricity sys- tems in India. Also, although the results stand in stark contrast to those reported for the U.S. manufacturing industry by Hulten and Schwab (1991, 2000), the difference is comforting in one sense. Their results and those reported here were obtained by essentially the same method, so that one may conclude that the model itself is not predetermining the results. Rather, the difference may point to asymmetrical effects in which infrastructure investments play a larger role in developing economies than in developed economies with more mature and denser infrastructure systems. 8. As an example, a bias might arise from a clientele effect in which infrastructure-intensive industries within the manufacturing sector are drawn to regions in which infrastructure is plentiful. If these industries also have the highest levels of total factor productivity, estimates of the externality parameter, y,may be biased upward, because they are based on industry data at the one-digit level that do not take industry composition into account. However, composition effects do not necessarily create a selection bias in the model. If infrastructure-intensive industries within the manufacturing sector are drawn to infrastructure-rich regions, because its infrastructure causes their total factor productivity to increase, this is a relocation effect and is commonly regarded in location theory as a system externality (recallthat the direct effect of infrastructure on output is reflected in the purchase of intermediate services from the transport and electricity generation sectors). Hulten, Bennathan, and' Srinivasan 307 Acharya, Shankar. 2002. "India's Medium-Term Growth Prospects." Economic and Political Weekly (July 13):2897-906. Ahluwalia, Isher Judge. 1991. Productivity and Growth in Indian Manufacturing. Delhi: Oxford Uni- versity Press. . 1994. "TFPG in Manufacturing Industry." Economic and Political Weekly (October 22):2836. Ahluwalia, Niontek S. 1998. "Infrastructure Developmentin India's Reforms."In IsherJudge Ahluwalia, and I. M. D. Little, eds., India's Economic Reforms and Development: Essays for Manmohan Singh. Delhi: Oxford University Press. 2000. "Economic Performanceof Statesin Post-ReformsPeriod." Economic and Political Weekly (May 6):163748. Aschauer, David A. 1989a. "Is PublicExpenditure Productive?"Journalof Monetary Economics 23:197- 200. . 1989b. "Public Investmentand ProductivityGrowth in the Group of Seven." Economic Perspec- tives 13:17-25. Balakrishnan, P., and K. Pushpangadan. 1994. "Total Factor Productivity Growth in Manufacturing Industry: A Fresh Look." Economic and Political Weekly (July30):2028-35. .1995. "Total Factor Productivity Growth in Manufacturing Industry." Economic and political Weekly (March 4):46244. Barro, Robert J. 1990. "Government Spending in a Simple Model of Endogenous Growth." Jourhal of Political Economy 98(5):S1034125. Caves, D. W., L. R. Christensen, and W. E. Diewert. 1982. "The Economic Theory of Index Nurjbers, and the Measurement of Input, Output and Productivity." Econometrica 50(6):1393414. Centre for Monitoring Indian Economy. Various years. "Energy reports." Mumbai. [www.cmie.com]. Dholakia, Baku1 H., and Ravindra H. Dholakia. 1994. "Total Factor Productivity Growth in Indian Manufacturing." Economic and Political Weekly (December 31):334244. .1995. "Total Factor Productivity Growth in Indian Industry." Economic ~rndPolitical weekly (July 15):1786-7. Fikkert, Brian, and Rana Hasan. 1998. "Returns to Scale in a Highly Regulated Economy: ~vidence;from Indian Firms." Journal of Development Economics 56(1):51-79. Hall, Robert E. 1988. "The Relation between Price and Marginal Cost in U.S. Industry." Joudal of Political Economy 96(5):92143. Hirschman, Albert 0.1958. The Strategy of Economic Development. New Haven, CT: Yale ~nivbrsity Press. Hsieh, Chang-Tai. 1999. "Productivity and Factor Prices in East Asia." American Economic Rkview 89(2):133-8. I Hulten, Charles R. 2001. "Total Factor Productivity: A Short Biography." In Charles R. Hulten, Edbard Dean, and Michael J. Harper, eds., New Developments in Productivity Analysis. Chicago: ~nivkrsity of Chicago Press. Hulten, Charles, and Robert M. Schwab. 1984. "Regional Productivity Growth in U.S. ~anufactJrin~: 1951-1978." American Economic Review 74(1):154-62. . 1991. "Public Capital Formation and the Growth of Regional Manufacturing Indusqies." National Tax Journal 44(4):121-34. -. 2000. "Does Infrastructure Investment Increase the Productivity of Manufacturing Industry in the U.S.?" In Lawrence J. Lau, ed., Econometrics and the Cost of Capital. Cambridge, MA: MIT Press. Hulten, Charles R., Esra Bennathan, and Sylaja Srinivasan. 2000. Indirect Effects of Infrastructure: Effects of Infrastructure on Productivity in Manufacturing. Regional Study in India. Washington, D.C.: World Bank. Hulten, Charles, and Sylaja Srinivasan. 1999. "Indian Manufacturing Industry: Elephant or Tiger?" NBER Working Paper 5569. Cambridge, MA: National Bureau of Economic Research. India, Ministry of Planning, Central Statistical Organization. Various years. AnnualSurvey of Industries. New Delhi. India, Ministry of Transport and Highways. Various years. Basic Road Statistics of India. New Delhi. Jorgenson, Dale W., and Mieko Nishimizu. 1978. "U.S. and Japanese Economic Growth, 1952-1974: An International Comparison." Economic Journal 88(352):707-26. Kubo, Yuji,Jaime De Melo, Sherman Robinson, and Moshe Syrquin. 1986. "Independenceand Industrial Structure." In Hollis Chenery, Sherman Robinson, and Moshe Syrquin, eds., Industrialization and Growth: A Comparative Study. New York: Oxford University Press. Lee, Kyu Sik, and AlexAnas.1992. "Costs of Deficient Infrastructure: The Case of Nigerian Manufactur- ing." Urban Studies 29(7):1071-92. Lucas, Robert E. Jr. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22(1):342. Meade, James E. 1952. "External Economies and Diseconomies in a Competitive Situation." Economic Journal 6254-67. Mitra, Arup, Aristomtne Varoudakis, and Marie-Ange Viganzonss. 1998. "State Infrastructure and Productive Performance in Indian Manufacturing." Development Centre Technical Paper 139. Orga- nisation for Economic Co-operation and Development, Development Centre, Paris. Nishimizu, Mieko, and Charles Hulten. 1978. "The Sources of Japanese Economic Growth, 1955-71." Review of Economics and Statistics 60(3):351-61. Pradhan, Gopinath, and Kaustuva Barik.1998. "Fluctuating Total Factor Productivity in India: Evidence from Selected Polluting Industries." Economic and Political Weekly (February 28):M25-M30. Rao, J. Mohan. 1996a."ManufacturingProductivity Growth: Method and Measurement." Economic and Political Weekly (November 2):2927-36. -. 1996b. "Indices of Industrial Productivity Growth: Disaggregation and Interpretation." Eco- nomic and Political Weekly (December 7):3177-88. Romer, Paul M. 1986. "Increasing Returns and Long-Run Growth." Journal of Political Economy 94(5):1002-52. Rosenstein Rodan, P. N. 1943. "Problems of Industrializing Eastern and South-Eastern Europe." Eco- nomicJournal 53(2):202-11. Solow, Robert M. 1957. "Technical Change and the Aggregate Production Function." Review of Eco- nomics and Statistics 39:312-20. Tybout, James. 2000. "Manufacturing Firms in Developing Countries: How Well Do They Do, and Why?"Journal of Economic Literature 38(1):1144. Young, Allyn. 1928. "Increasing Returns and Economic Progress." Economic Journal 38(152):52742. Young, Alwyn.1992. "A Tale of Two Cities: Factor Accumulation and Technical Change in Hong Kong and Singapore." In 0.J. Blanchard, and S. Fisher, eds., NBER Macroeconomics Annual 1992. Cam- bridge, MA: MIT Press. .1995. "The Tyranny of Numbers: Confronting the Statistical Realitiesof the East Asian Growth Experience."Quarterly Journal of Economics 110(13):641-80. Signup for CiteTrack today - go to www.hlghwire.org,and click on 'My EmallAlerts' for more information. OXFORD UNIVERSITY PRESS I Forthcomingpapersin THE WORLD BANK ECONOMIC REVIEW Volume20, Number 3,2006 The Primacyof Institutions Reconsidered:Direct Income Effects of Malaria Prevalence Kai Carstensen and Erich Gzmdlach Micro-Firm Dynamics in Less Developed Countries: How Similar are they to those in the Industrialized World? Evidence fiom Mexico Pablo Fajnzylber,William I? Maloney, and Gabriel Montes Rojas When Is External Debt Sustainable? Aart Kraay and Vikram Nehru How Endowments, Accumulations and Choice Determine the Geography of Agricultural Productivity in Ecuador Donald I? Larson and MauricioLedn The"Glass of MiW' SubsidyProgram and Malnutrition in Peru David Stifeland Harold Alderman