The Limits of Stabilization I N F R A S T R U C T U R E , P U B L I C D E F I C I T S , A N D G R O W T H I N L AT I N A M E R I C A Edited by William Easterly Luis Servén THE WORLD BANK The Limits of Stabilization The Limits of Stabilization INFRASTRUCTURE, PUBLIC DEFICITS, AND GROWTH IN LATIN AMERICA Edited by William Easterly Luis Servén A COPUBLICATION OF STANFORD SOCIAL SCIENCES, AN IMPRINT OF STANFORD UNIVERSITY PRESS, AND THE WORLD BANK © 2003 The International Bank for Reconstruction and Development/The World Bank 1818 H Street, NW Washington, DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org E-mail: feedback@worldbank.org All rights reserved. 1 2 3 4 06 05 04 03 A copublication of Stanford Social Sciences, an imprint of Stanford University Press, and the World Bank. Stanford University Press The World Bank 1450 Page Mill Road 1818 H Street, NW Palo Alto, Calif. 94304 Washington, DC 20433 The findings, interpretations, and conclusions expressed herein are those of the au- thor(s) and do not necessarily reflect the views of the Board of Executive Directors of the World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of the World Bank concerning the legal sta- tus of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The World Bank en- courages dissemination of its work and will normally grant permission promptly. For permission to photocopy or reprint any part of this work, please send a re- quest with complete information to the Copyright Clearance Center, Inc., 222 Rose- wood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, www.copyright.com. All other queries on rights and licenses, including subsidiary rights, should be ad- dressed to the Office of the Publisher, World Bank, 1818 H Street, NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. ISBN 0-8213-5489-2 (World Rights except North America) ISBN 0-8047-4972-8 (North America) Library of Congress Cataloging-in-Publication Data The limits of stabilization : infrastructure, public deficits and growth in Latin America/ edited by William Easterly and Luis Servén. p. cm. ­ (Latin American development forum) Includes bibliographical references and index. ISBN 0-8213-5489-2 1. Infrastructure (Economics)--Latin America. 2. Economic stabilization-- Latin America. 3. Public-private sector cooperation--Latin America. I. Easterly, William Russell. II. Serven, Luis. III. World Bank. IV. Series. HC130.C3L56 2003 339.5 098--dc21 2003053802 Latin American Development Forum Series This series was created in 2003 to promote debate, disseminate infor- mation and analysis, and convey the excitement and complexity of the most topical issues in economic and social development in Latin Amer- ica and the Caribbean. It is sponsored by the Inter-American Develop- ment Bank, the United Nations Economic Commission for Latin America and the Caribbean, and the World Bank. The manuscripts chosen for publication represent the highest quality in each institu- tion's research and activity output, and have been selected for their rel- evance to the academic community, policymakers, researchers, and in- terested readers. Advisory Committee Members Inés Bustillo, Director, Washington Office, Economic Commission for Latin America and the Caribbean, United Nations Guillermo Calvo, Chief Economist, Inter-American Development Bank José Luis Guasch, Regional Adviser, Latin America and Caribbean Region, World Bank Stephen Haber, A. A. and Jeanne Welch Milligan Professor, Depart- ment of Political Science, Stanford University; Peter and Helen Bing Senior Fellow, the Hoover Institution Eduardo Lora, Principal Adviser, Research Department, Inter- American Development Bank José Antonio Ocampo, Executive Secretary, Economic Commission for Latin America and the Caribbean, United Nations Guillermo E. Perry, Chief Economist, Latin America and Caribbean Region, World Bank Luis Servén, Lead Economist, Latin America and Caribbean Region, World Bank About the Contributors César A. Calderón is a senior economist with the Research Depart- ment at the Central Bank of Chile. He received his Ph.D. in economics from the University of Rochester. His research focuses on growth, de- velopment, and open economy macroeconomics. Javier Campos is a professor of economics in the Department of Ap- plied Economic Analysis at the University of Las Palmas de Gran Canaria. He specializes in the empirical analysis of competition and regulation issues in network industries. Since his sabbatical at the World Bank, he has continued to advise governments on reform and regulation of the transport sector in Latin America. William Easterly is a professor of economics (joint with Africana Stud- ies Program) at New York University, where he is also codirector of the Development Research Institute. He is also a senior fellow at the Center for Global Development and the Institute for International Economics. He worked for 16 years as a research economist at the World Bank. He is author of The Elusive Quest for Growth: Econo- mists' Adventures and Misadventures in the Tropics (2001), as well as of numerous academic articles. His research specializations include economic growth, foreign aid, ethnic conflict, political economy, and macroeconomic policies. He holds a Ph.D. from the Massachusetts Institute of Technology. Antonio Estache is a senior economic adviser at the World Bank and a research fellow at the European Center for Advanced Research in Eco- nomics and Statistics at the Université Libre de Bruxelles. He special- izes in industrial organization and public sector economics. For the past 20 years, he has advised governments in Africa, Asia, and Latin America on infrastructure sector reform and regulation. viii ABOUT THE CONTRIBUTORS Noelia Martín is a researcher in the Department of Applied Econom- ics at the University of Las Palmas de Gran Canaria. She holds an M.A. in economics from the University of York, where she is currently com- pleting her Ph.D. Her research focuses on the interactions between macroeconomic performance and infrastructure sector investments and reforms. Sheoli Pargal is a senior economist at the World Bank, with experience in the Bank's research department as well as in Bank operations. She has worked on infrastructure, environment, and social development is- sues, focusing on institutional, incentive, and regulatory aspects. She received her Ph.D. in economics from Northwestern University. Luis Servén manages the research program of the Latin America and Caribbean Region of the World Bank. He has published numerous ar- ticles in academic journals and several monographs in the fields of macroeconomics and international finance. He has taught at the Uni- versidad Complutense de Madrid, Centro de Estudios Monetarios y Financieros (CEMFI), and the Massachusetts Institute of Technology. His research areas are saving, investment, and open economy macro- economics. He holds a Ph.D. from the Massachusetts Institute of Tech- nology. Lourdes Trujillo is a professor of economics and the director of the Department of Applied Economic Analysis at the University of Las Palmas de Gran Canaria. She specializes in the empirical analysis of the infrastructure industries. She has advised governments throughout Latin America on transport sector reform and regulation. Contents Acknowledgments xv 1 INTRODUCTION 1 William Easterly and Luis Servén 2 LATIN AMERICA'S INFRASTRUCTURE IN THE ERA OF MACROECONOMIC CRISES 21 César Calderón, William Easterly, and Luis Servén 3 THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 95 César Calderón and Luis Servén 4 INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY IN LATIN AMERICA 119 César Calderón, William Easterly, and Luis Servén 5 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION IN INFRASTRUCTURE 139 Javier Campos, Antonio Estache, Noelia Martín, and Lourdes Trujillo 6 REGULATION AND PRIVATE SECTOR PARTICIPATION IN INFRASTRUCTURE 171 Sheoli Pargal Acronyms and Abbreviations 199 Index 205 ix x CONTENTS FIGURES 2.1 Comparative Performance in Infrastructure Stocks 24 2.2 Power Generating Capacity by Region, 1980­97 26 2.3 Power Generating Capacity per Worker by Country, 1980 and 1997 27 2.4 Length Comparisons of Transportation Routes 28 2.5 Road Length per Worker by Country, 1980 and 1997 29 2.6 Access to Clean Water, 1985­93 29 2.7 Infrastructure Quality and Excess Demand 30 2.8 Power Losses by Region, 1980­97 31 2.9 Power Losses by Country, 1980 and 1997 32 2.10 Comparisons of Surface Transportation Quality 33 2.11 Total Investment in Infrastructure in Selected Latin American Countries, 1980­98 34 2.12 Investment in Infrastructure in Selected Latin American Countries, by Sector, 1980­98 35 2.13 Public Investment in Infrastructure in Selected Latin American Countries, 1980­98 37 2.14 Public Investment in Infrastructure and Noninfrastructure, by Country 40 2.15 Public Investment in Infrastructure, by Sector and by Country 43 2.16 Private Investment in Infrastructure in Selected Latin American Countries, 1980­98 47 2.17 Private Investment in Infrastructure and Noninfrastructure, by Country 48 2.18 Private Investment in Infrastructure, by Sector and by Country 51 2.19 Private Investment Per Capita around the Date of Reform in Selected Country, by Sector 55 2.20 Infrastructure Quality and the Private Share of Investment in Infrastructure 65 3.1 Infrastructure Accumulation and Growth, 1960­97 97 6.1 Annual Average Share of Private Investment in Total Infrastructure Investment in Selected Latin American Countries, by Sector 174 6.2 Average Annual Investment in Infrastructure in Selected Latin American Countries, by Sector 175 TABLES 2.1 The Contribution of Infrastructure Compression to Fiscal Adjustment, Average 1980­84 versus Average 1995­98 37 CONTENTS xi 2.2 Regression of Public Infrastructure Investment/GDP on the Primary Balance/GDP 39 2.3 Infrastructure Reform Dates 46 2.4 Correlation between Public and Private Infrastructure Investment, by Sector 57 2.5 Regressions of Public Infrastructure Investment/GDP on Private Infrastructure Investment/GDP 58 2.6 Relationship between Physical Stocks and Investment Spending in Infrastructure 61 2.7 Relationship between Physical Stocks, Public and Private Investment Spending in Infrastructure 63 2.8 Private Participation and Infrastructure Quality 64 2A.1 Telephone Service Variables 69 2A.2 Summary of Coverage and Availability of Telephone Service Indicators 70 2A.3 Telecommunications Indicators: Time-Series Coverage by Region 72 2A.4 Variables Used as Proxies for Energy 73 2A.5 Energy Indicators: Time-Series Coverage by Region 74 2A.6 Sanitation and Sewerage Indicators: Time-Series Coverage by Region 74 2A.7 Indicators for Roads 75 2A.8 Other Indicators for Roads 75 2A.9 Indicators for Irrigation 75 2A.10 Irrigation: Time-Series Coverage by Region 76 2A.11 Roads and Railways: Time-Series Data for Selected Regions 77 2A.12 Public Sector Definitions Used in the Figures of Public Investment in Infrastructure 77 2A.13 Definition of the Transport Sector 78 2B.1 Infrastructure Reform Laws 82 2B.2 Sale and/or Concession of Public Enterprises in Infrastructure Sectors 82 2B.3 Greenfield Projects in Infrastructure Sectors 83 3.1 The Widening Infrastructure Gap, Latin America versus East Asia 96 3.2 Sample Correlations 103 3.3 Infrastructure-Augmented Production Function: Alternative Estimates 104 3.4 Alternative GMM Estimates 106 3.5 First-Difference GMM Estimates of Alternative Specifications 108 xii CONTENTS 3.6 Elasticity of Output per Worker with Respect to Capital per Worker 110 3.7 The Infrastructure Gap and the Output Gap: Contribution of Various Inputs to the Change in Relative GDP per Worker, Latin America versus East Asia, 1980­97 111 3.8 The Infrastructure Gap and the Output Gap: Contribution of the Change in Relative Infrastructure Stocks to the Change in Relative GDP per Worker, East Asia versus Selected Latin American Countries, 1980­97 112 4.1 Taxes and Growth: Panel Data Regression Analysis 126 4.2 Government Spending and Growth: Panel Data Regression Analysis 127 4.3 Impact on the Annuity Value of Net Worth of a Cut in Infrastructure Investment by 1 Percent of GDP 129 4.4 Partial-Equilibrium Effect of Actual Infrastructure Investment Changes on the Annuity Value of Public Net Worth, 1980­84 versus 1995­98 131 4A.1 Panel Unit Root Tests, Government Revenues, Government Spending, and Real Output 135 5.1 Main Macroeconomic Variables: Levels and Ranking 146 5.2 First Year for Private Participation in Utilities and Transport 147 5.3 Average Value of the Institutional Variables between 1985 and 1998 149 5.4 Effects of PPI on GDP Per Capita (Model 1) 151 5.5 Effects of PPI on GDP Per Capita (Model 2) 152 5.6 Effects of PPI on Private Investment (Model 1) 155 5.7 Effects of PPI on Private Investment (Model 2) 156 5.8 Effects of PPI on Private Investment Defined as Gross Domestic Investment Minus Public Investment (Model 1) 158 5.9 Effects of PPI on Public Investment (Model 1) 160 5.10 Effects of PPI on Public Investment (Model 2) 162 5.11 Effects of PPI on Recurrent Public Expenditures (Model 1) 164 5.12 Effects of PPI on Public Expenditures (Model 2) 166 5.13 Summary of Signs of Average Macroeconomic Effects of PPI 168 6.1 Data Sources 179 6.2 Country Fixed Effects Estimation 183 6.3 Fixed Effects Regressions by Sector 187 CONTENTS xiii 6A.1 Descriptive Statistics 189 6A.2 Descriptive Statistics if Regulatory Body Exists 189 6A.3 Correlation Matrix for Complete Data Set 190 6A.4 Correlation Matrix for Regulatory Variables 190 6A.5 Public and Private Investment before and after the Passage of Reform Legislation 190 6A.6 Aspects of Regulatory Structure, by Country and by Sector 192 Acknowledgments THIS BOOK EMERGED AS A RESULT OF research developed under the Re- gional Studies program of the World Bank's Latin America and Caribbean region. We are very grateful to Guillermo Perry and Danny Leipziger for suggesting this research topic to us and for their generous support and encouragement throughout this project. A major challenge for the analysis presented in this book was the collection of suitable data. We are especially grateful to Pilar Blanco for her valuable assistance in this effort, and to Luis Guasch, Graciela Moguillansky, and Quentin Wodon for generously sharing their data with us. We are also indebted to Patricia Macchi for her valuable re- search assistance. We have benefited from comments and suggestions from many col- leagues at various stages of this research. In particular, we would like to thank Ezra Bennathan, Raquel Carrasco, Alex Galetovic, Luis Guasch, Marianne Fay, Greg Ingram, Jose Luis Irigoyen, Danny Kauf- man, Phil Keefer, Michael Kerf, Martin Rodriguez-Pardina, Arijit Sen, and Nils Tcheyan. We would also like to acknowledge the useful com- ments received from participants at the September 2001 BNDES (Banco Nacional de Desenvolvimento Econômico e Social)­World Bank Infrastructure Conference held in Rio de Janeiro and the March 2002 CLAI­Organization of American States Conference on Energy in Latin America held at The George Washington University in Washington, D.C. Finally, we would like to stress that the views stated in this book are only those of the authors and should not be attributed to the institu- tions with which they are affiliated. xv 1 Introduction William Easterly and Luis Servén SUPPOSE THAT A DEVELOPING-COUNTRY policymaker proposed the fol- lowing adjustment program to the International Monetary Fund (IMF) and the World Bank: her government would repay public external debt, which carries an interest rate of 9 percent a year, by substituting debt from another source that carries an interest rate of 20 percent a year. The proposal would be swiftly dismissed by the international fi- nancial institutions (IFIs), perhaps with unflattering remarks about the policymaker's knowledge of the basic laws of economics. Yet this kind of adjustment program describes a part of the pack- age of many Latin American macroeconomic stabilization programs of the past two decades, often supported by the IMF or the World Bank. Instead of "debt from another source," we have cuts in spend- ing on maintenance and construction of public sector infrastructure, which is estimated almost universally to have a high rate of return. The World Bank (1994, p. 17) estimated rates of return to infrastruc- ture investment during 1983­92 ranging from 19 percent (telecom- munications) to 29 percent (highways). Gyamfi, Gutierrez, and Yepes (1992) estimated economic rates of return of more than 70 percent for operations and maintenance on roads in Latin America. The World Bank (1992, p. 57) estimated a rate of return of 117 percent for non- wage operations and maintenance in irrigation in the mid-1980s in Indonesia. Cutting spending on a project with a high rate of return is econom- ically equivalent to borrowing at that high rate of interest--both free up resources today in return for lost resources tomorrow. Many Latin American governments cut infrastructure spending in the era of macroeconomic reform--a line item in the adjustment program that set adjustment back rather than forward. 1 2 THE LIMITS OF STABILIZATION It is true that Latin American governments needed to adjust. The Latin American debt crisis began on August 18, 1982, when Mexican Finance Minister Jesus Silva Herzog announced that Mexico could no longer service its external debt to international commercial banks. It soon became apparent to everyone that most Latin American countries had excessive debt and needed to retrench severely if debt ratios were to be manageable. Because the excessive borrowing was caused largely by persistent budget deficits, fiscal adjustment became an unavoidable task.1 When fiscal adjustment was insufficient, what had been an external debt crisis became a high-inflation crisis because governments resorted to printing money to finance their deficits in the absence of foreign lending. Countries like Argentina, Brazil, and Peru experienced ex- treme inflation episodes in the 1980s or early 1990s, and had to un- dertake severe fiscal adjustment to bring inflation under control. In these and other Latin American economies, placing public finances on a sustainable course was an essential step to restore macroeconomic stability. At the same time, there was a welcome shift in ideology throughout Latin America during the last two decades away from the state-led, inward-looking development paradigm of the 1960s and 1970s toward more reliance on markets, free trade, and the private sector. Growing disenchantment with pervasive government intervention in the econ- omy--ranging from price and interest rate controls to direct state in- volvement in the production of numerous goods and services--opened the way to a new development model in which market forces played the leading role in the allocation of resources. The state withdrew from most production activities in favor of the private sector in a radical paradigm shift aiming to raise economic efficiency and long-term growth. There is no question that fiscal retrenchment was necessary and that Latin America's state-led model had been exhausted. What is ques- tionable is the extent to which public infrastructure spending bore the brunt of adjustment. This is by no means a new discovery; earlier analyses have already documented the fact that in developing countries infrastructure expenditures often suffer a disproportionate compres- sion in times of fiscal austerity.2 This book offers unambiguous evi- dence that the Latin American experience of the 1980s and 1990s con- formed to the same pattern. The compression of infrastructure spending is largely a consequence of the myopic use of the current budget deficit to GDP ratio as the sin- gle yardstick to assess fiscal performance. It could be avoided easily if economic analysts--including the IFIs--were to change their thinking INTRODUCTION 3 and evaluate adjustment in terms of the only budget constraint that matters economically, namely, the intertemporal budget constraint. As will be discussed later in more detail, this constraint says that the pres- ent value of all future government revenues must be sufficient to cover the existing stock of debt plus the present value of all future govern- ment spending. For this calculation, the present value of revenue and expenditure is evaluated at the interest rate the government pays on its marginal borrowing. Any project with a higher return than that inter- est rate should be undertaken, because it makes it easier to meet the in- tertemporal budget constraint regardless of the effect of the project on the current budget deficit. Many infrastructure maintenance and construction projects have such high rates of return that they satisfy this condition. Yet for a long time the IFIs and the international financial community continued to view infrastructure cuts as a valid means to fiscal adjustment. Various rationalizations have been offered for this approach. One is that some macroeconomic crises are caused by shortages of liquidity rather than the kind of insolvency the intertemporal budget constraint covers. Cut- ting infrastructure spending could free up some short-term funds to avoid such damaging expedients as printing money. This argument is suspect, however. The role of the IFIs is precisely to ease liquidity crises in a way that preserves long-run growth potential while avoiding short-run destabilization. But, to return to the argument in the first paragraph, recommending cuts in infrastructure spending says that the adjusting country should resolve the liquidity squeeze by taking out a loan at an interest rate of 30 to 70 percent. An adjustment program meant to resolve a liquidity crisis should not have to resort to such costly sources of financing. A second rationalization for cutting infrastructure spending is that even if infrastructure spending has high returns, these returns may ac- crue to the society rather than the government. If the macroeconomic problem is caused by an excessive budget deficit, then infrastructure cuts could improve the budget picture even though they worsen the economy's long-run potential. This argument is shortsighted in several ways. First, if infrastructure cuts lower growth, then this will have a negative fiscal impact (described below). Second, fiscal policy can be designed to capture a good share of the high social returns to infra- structure spending. A third argument in favor of infrastructure cuts during fiscal ad- justment is that the spending is often going to white elephants that do not have a high rate of return. This is clearly true in some cases.3 It would be naïve to believe that everything called infrastructure spending in the fiscal accounts is necessarily productive, or that such 4 THE LIMITS OF STABILIZATION spending should be the only--or even the main--indicator of public infrastructure performance. Also, governments and IFIs should pay close attention to the incentives facing the government bureaucracy to provide efficient infrastructure services. However, these caveats fail to justify an across-the-board cut in infrastructure spending during fis- cal adjustment. It would be far better to cut just the white elephants and to improve incentives for service delivery, while preserving the productive new construction and maintenance projects from fiscal austerity. Finally, the argument is often made that the private sector could take over many aspects of infrastructure provision, and so cutting public in- frastructure spending is not such a big deal. This argument accords well with Latin America's shift away from the state-led development model of the 1960s and 1970s. Private provision of infrastructure is a prom- ising area and private infrastructure provision will be examined at great length in this book. However, with few exceptions, private provision is still at a relatively early stage in most countries (with the telecommuni- cations sector as the main exception). For the most part, infrastructure is provided publicly almost everywhere, and has been throughout the history of the now-rich countries. Even where private infrastructure provision is viable, the transition from public to private ownership has to be thought out carefully. Opening infrastructure industries to private sector involvement can make a lot of sense, but to cut high-return public infrastructure spending and expect the private sector to fill the breach overnight is a leap of faith. The conclusion is that cutting high-return public infrastructure in- discriminately during fiscal adjustment does not make sense in either macroeconomic or microeconomic terms. It makes about as much sense as the satirical business principle: "We take a loss on every item, but we make it up on volume." This book provides the main facts on the pattern of infrastructure spending under macroeconomic adjustment in Latin America over the past two decades along with evidence on its rate of return. No claim is made that the infrastructure cuts were so pervasive as to make the en- tire adjustment package in each country a step backward. It would also be illusory to assert that all spending classified as infrastructure necessarily led to maintenance or creation of productive capital. The point needs to be stressed, however, that spending cuts in Latin Amer- ica included some high-return projects that never should have been cut in an exercise designed to move to public sector solvency. The opening up of infrastructure industries to private sector participation has had mixed results and so far has not resolved Latin America's infrastruc- ture problems. INTRODUCTION 5 This introductory chapter presents an analytical framework high- lighting the relevant concepts on the intertemporal budget constraint and the growth impact of infrastructure. Then this organizing framework is used to summarize the main findings of the other chapters in the book. The Intertemporal Budget Constraint and Fictional Adjustment Many authors have identified the government's intertemporal budget constraint as the ultimate constraint on the government's fiscal activi- ties (see, for example, Anand and van Wijnbergen 1989; Auerbach 1997; Blanchard and others 1990; Buiter 1990; Buiter and Patel 1992, 1997; and Easterly 1999). Schematically, the intertemporal budget constraint states:4 Present value of tax revenues net of transfers ­ Present value of public consumption Initial public debt D d. (1.1) + Present value of seigniorage T c­Initial public capital + Present value of excess return on public capital The left-hand side of this expression is the present value of the en- tire future stream of the government's augmented noninterest sur- pluses on current account. Such augmented surpluses consist of four ingredients: taxes net of transfers; public consumption (the difference between the taxes net of transfers and public consumption is the cur- rent primary surplus);5 revenues from money printing; and the differ- ence between the financial rate of return on public capital (net of de- preciation) and the discount rate, which--for lack of a better term--will be called excess return. The latter is positive if the cash rate of return on public assets is higher than the discount rate, and negative other- wise.6 Note that this refers only to the direct cash return on public cap- ital. There can also be indirect revenue effects if public capital affects other fiscal variables. Most important, such indirect effects may arise through the impact of public capital on private capital and output and thereby on future tax collection, whose present value is the top item on the left-hand side of the intertemporal budget constraint. This issue will be reexamined later. The right-hand side of expression 1.1 is government debt net of public assets. Here debt should include both explicit and implicit debt as well as contingent liabilities.7 Solvency requires that the present value of augmented current primary surpluses be no less than initial net debt. Intuitively, for the government to be solvent it has to run a 6 THE LIMITS OF STABILIZATION surplus large enough to cover not only the interest on its (net) debt, but also some payment toward the principal. In contrast with the intertemporal budget constraint, the conven- tional deficit identity highlights the current accumulation of public debt. Furthermore, the focus is on explicit debt and, typically, limited attention is paid to implicit and contingent liabilities. In light of the intertemporal budget constraint, it is easy to under- stand the many tricks countries play to lower the conventional budget deficit (or the rate of debt accumulation) while avoiding real, long- term fiscal adjustment. The tricks range from rearranging the time pro- file of revenues or expenditures without altering their present value, to lowering the rate of debt accumulation by reducing the rate of asset ac- cumulation, to replacing explicit liabilities or recorded expenditures with hidden liabilities kept off the books. For example, oil-producing countries with adjustment programs pumped oil out of the reserves in the ground faster than they did during periods without adjustment programs. They got more current revenue at the cost of making less oil revenue available for sale in the future, thus lowering the current deficit at the expense of the future deficit, and fail- ing to improve the long-run fiscal picture.8 Governments can also simply shift expenditures and revenues across time to meet today's cash deficit targets. Often they resort to the expedient of delaying payments to government workers or suppliers. These arrears lower this year's cash deficit and explicit public debt, while increasing next year's cash deficit and the implicit public debt.9 These tricks are not exclusive to developing countries. They are also used frequently by industrial nations. Consider the United States dur- ing the effort to contain deficits at the time of the Gramm-Rudman bill. The Congress in fiscal 1987 postponed a $3 billion payday for military personnel into the following fiscal year. The Defense Secre- tary, Caspar Weinberger, also stretched out procurement of new weapons systems to lower the current expenditure, even though the stretch-out increased per unit costs (see Kee 1987, p. 11). Governments can also shift taxes over time. There are many anec- dotes of developing countries getting advance payments of taxes to meet IMF program deficit targets (see Kopits and Craig 1998). In the same way, the U.S. Congress moved about $1 billion in excise tax col- lections forward to meet the Gramm-Rudman deficit ceiling in fiscal 1987 (as discussed in White and Wildavsky 1989, p. 514). Reducing the rate of asset accumulation--that is, public investment-- is another commonly used approach to deficit reduction. It has been am- ply used in developing-country adjustment programs, as documented, for example, by Servén and Solimano (1992) and, in Latin America, by INTRODUCTION 7 this volume. From the intertemporal budget constraint above, a reduc- tion in public investment will improve the solvency position of the pub- lic sector if the rate of return on public capital falls short of the discount rate. However, it is important to note that this test needs to compare the discount rate with the total return on public capital--that is, not only the direct cash return but also the indirect one accruing through the im- pact of public capital on future output and tax collection. Privatization of state assets is a more expeditious way to reach the same end. If the government uses the proceeds to retire public debt, privatization reduces simultaneously the stocks of public assets and public debt in the right-hand side of (1.1). The reduction need not be one-for-one, however, because the volume of debt that can be retired depends on the price at which the assets are sold. In general, the sale price will reflect the present value of the future returns accruing to the purchaser. If these are the same as the returns that would have accrued to the government, then privatization is unlikely to help solvency.10 In other words, solvency is strengthened only if the government manages to sell the assets at a price above the present value of the net future re- turns that it could have derived from holding them. Again, both developing and industrial countries have resorted to these means to achieve deficit or debt reduction. For example, in the United States, the Gramm-Rudman initiative gave impetus to the idea of selling off state assets and counting the proceeds as revenue, ficti- tiously lowering the deficit--but with uncertain effects on solvency. Congress had long stalled on privatization of the railway company, Conrail, until Gramm-Rudman came along. When Gramm-Rudman created incentives for getting privatization revenues to meet budget targets, the Congress suddenly hurried up and sold Conrail. Another sleight of hand is to reduce current expenditure today in re- turn for a contingent or off-budget liability. For example, the govern- ment can switch from granting subsidies to state enterprises to guaran- teeing bank loans made to them to cover their losses. This creates the appearance of a deficit reduction and a slowdown in the accumulation of explicit debt. When the enterprises eventually default on their debt, however, the government has to pay off the debt and so winds up pay- ing for state enterprise losses just as it had been when subsidies were ex- plicit. Egypt, for example, phased out budgetary support to state enter- prises in 1991, but allowed loss-making enterprises to continue to operate on bank overdrafts and foreign loans. The Egyptian govern- ment periodically had to cover for loan defaults by these enterprises.11 Even more creatively, governments can also shift subsidies to state enterprises off the books by having public financial institutions (whose balances are seldom included in government deficit definitions) provide 8 THE LIMITS OF STABILIZATION subsidized lending to the state-owned enterprises (SOEs). In Argentina, before 1990, the central bank gave a subsidized interest rate on loans to loss-making public enterprises, reducing their interest costs and their losses (see Mackenzie and Stella 1996). Off-budget liabilities have played a particularly important role in connection with the privatization of infrastructure assets. To raise the sale price, governments have often provided price or rate-of-return guarantees to the purchasers, or have guaranteed their borrowing. For example, to protect the private owners from demand uncertainty, the Colombian government offered a minimum revenue guarantee to some toll-road concessions in the 1990s. Similarly, the Spanish government provided exchange rate guarantees on the foreign loans to toll-road concessions in the 1970s. In private power generation projects in Pakistan, take or pay clauses were common to shelter investors from the risk that installed capacity could go unutilized. (All these examples are from Irwin and others 1997.) Guarantees can make sense in the context of infrastructure projects because these involve large sunk costs and are highly vulnerable to op- portunistic government behavior (for example, through expropriatory regulation), two features that make them especially risky. Yet the guar- antees represent contingent government liabilities that are seldom ac- counted for explicitly. They shift the risk from the private owners of the infrastructure assets to the government. When the guarantees are called, typically at times of recession, their fiscal impact can be signif- icant. The proper design and valuation of guarantees on infrastructure projects have been studied at length elsewhere, and will not be pursued here (see Irwin and others 1997, and Brixi and Schick 2001). This brief catalogue suggests the assorted tricks that at one time or another have passed for fiscal adjustment--while having in reality lit- tle effect on public solvency. Europe's recent experience with the Maastricht Treaty offers an excellent case study on the use of tricks to meet deficit and debt targets. Cheating was widespread during the run- up to the May 1998 selection of countries to join the European Mon- etary Union, which involved complying with the deficit and debt tar- gets set out in the Stability Pact of the Maastricht Treaty. For example, Greece, not then a member of the European Union but trying hard to become one, announced in 1998 plans to privatize 11 state enterprises and three to four state banks. Among the enter- prises were such potentially profitable companies as Hellenic Telecom- munications Organization, Hellenic Petroleum, Water Supply Co., and two subsidiaries of Olympic Airways. Revenue from Greek priva- tizations was expected to total 0.8­0.9 percent of gross domestic prod- uct (GDP) in 1998­99 (from Dow Jones Newswires March 15, 1998). INTRODUCTION 9 Belgium was even less subtle, selling $2.5 billion worth of gold re- serves on March 19, 1998, and using the proceeds to reduce public debt by 1 percent of GDP. Sales of mobile phone licenses also brought revenue that could be applied to lower both deficit and debt. France used a more intricate device. France Telecom made a one- time payment to the government of 0.5 percent of GDP in return for the government shouldering Telecom's pension liabilities, an increase in implicit government liabilities not transparently recorded as gov- ernment debt. The proceeds reduced the deficit according to a Euro- pean Commission ruling! This conjuring trick accounted for half of France's deficit reduction in 1997 (see Dow Jones Newswires March 25, 1998; Economist December 14, 1996; and European Commission 1998). One skeptic noted that "the French budget process suggests that interpretive flexibility is simply being shifted from the Maastricht criteria to national accounting practices" (Hildebrand 1996). Like France, Austria got a one-time payment from a state enterprise (the Postsparkasse) in return for assuming pension liabilities (Euro- pean Commission 1998). Like Belgium, other temporary Austrian rev- enues came from sales of mobile phone licenses. Austria used a further sleight of hand, reclassifying some state enterprises from government to corporate sector, such as Asfinag, with substantial debts (Dow Jones Newswires April 8, 1998). Italy was more transparent: it levied a one-time Eurotax to meet the Maastricht deficit target in 1997, but announced that 60 percent of the tax would be refunded in 1999 (Economist Intelligence Unit April 23, 1998). The budget plan also foresaw lower debt from the proceeds of privatizing the highway management network, Autostrade, and the airline, Alitalia. Even the conservative Germans engaged in some illusory fiscal ad- justments. They reclassified public hospitals into the corporate sector in 1997, taking their debts out of general government debt (European Commission 1998). They also delayed interest payments on the pub- lic debt to lower the 1997 deficit, accelerated sales of shares in Deutsche Telekom, and used central bank profits from reserve reval- uation to pay off debt inherited from East Germany. (See Economist Intelligence Unit April 23, 1998, and Dow Jones Newswires May 14, 1998.) Fiscal Adjustment and Infrastructure Spending Cutting productive infrastructure spending can be a similarly fictional type of fiscal adjustment. It may even be counterproductive, in the 10 THE LIMITS OF STABILIZATION sense of weakening the solvency position of the government rather than strengthening it. Infrastructure spending by the government enters the intertemporal budget constraint in three places. First, it is part of total government spending--both investment spending related to the acquisition of in- frastructure assets and recurrent spending for operations and mainte- nance. Second, it raises future public revenues (a level effect)--both di- rect revenues to the extent that infrastructure user charges exist, and indirect revenues to the extent that an increase in infrastructure leads to permanently higher output and tax collection. This can be thought of as increasing government assets that will yield positive revenues in the future. Third and most important, if infrastructure spending raises the rate of growth of the economy, it will affect the sustainability of a given primary surplus. To highlight these facts, it is convenient to rewrite the intertempo- ral budget constraint in a slightly different form (see Buiter 1990): Present value of current primary surplus/GDP D­ Present value of public investment/GDP T [Initial public debt/GDP]. (1.2) + Present value of seigniorage/GDP This formulation differs from the previous one in two ways. First, all flow revenues and expenditures related to public infrastructure cap- ital have been added to the left-hand side of the equation (for simplic- ity, noninfrastructure capital is ignored). As a result, the right-hand side now contains only the public debt stock. Second, all magnitudes have been expressed as ratios to GDP. As a consequence, the rate used to discount future revenues and expenditures to arrive at their present value is now a net discount rate, given by the difference between the original (gross) rate and the rate of growth of GDP. The assumption is that this net discount rate is positive.12 The government is solvent if the above inequality holds. In fact, the government net worth is the difference between the left-hand and right-hand sides of (1.2). If it is negative, the government is insolvent with the current fiscal policies, debt levels, and net discount rate-- including the prevailing growth rate. Restoring solvency then re- quires some combination of higher growth, fiscal adjustment, and debt relief. If one thinks of a long-run steady state in which fiscal revenues and expenditures remain constant relative to GDP,13 then it is easy to simplify the above expression to highlight the various effects of INTRODUCTION 11 infrastructure spending mentioned earlier. In a long-run equilibrium, the preceding expression can be rewritten: Current primary surplus/GDP C­ Public investment/GDP S + Seigniorage/GDP (1.3) [Public debt/GDP]. [Discount rate ­ GDP growth rate] This expression is familiar from the fiscal solvency literature (Blanchard and others 1990, Buiter 1990, Buiter and Patel 1997, and Cuddington 1997). With strict equality, it becomes a condition for sta- bilizing the ratio of debt to GDP, and can be viewed as defining the primary surplus (augmented by seigniorage) required to keep constant the debt ratio for a given net discount rate.14 Thus the government is solvent if it is able to run a (augmented) primary surplus at least as large as that required to keep constant the debt-to-GDP ratio. Cuts in infrastructure investment reduce the public investment/GDP ratio and, other things being equal, tend to enhance solvency. Changes in infrastructure spending affect also the current primary surplus, through their derived effects on revenues (for example, from user charges) and expenditures (for example, operations and maintenance). But, in addition to these conventional effects, changes in infrastructure spending can also have an important impact on the intertemporal budget balance--for a given primary surplus relative to GDP-- through their effect on the growth rate. A cut in infrastructure spend- ing that over time leads to reduced growth raises the net discount rate and therefore lowers, other things being equal, the value of the left- hand side of (1.3). Thus, it requires a permanent increase in the aug- mented primary surplus (or a decrease in the debt ratio) to restore gov- ernment net worth to its previous level.15 More generally, any adverse shock to economic growth (like infra- structure shortages) is a fiscal shock that tends to bring the current public sector stance away from solvency. Conversely, anything that in- creases growth makes a given primary surplus more likely to achieve solvency. It is well-known that growth plays a critical role for govern- ment solvency.16 Budget planners in the United States are sufficiently familiar with this result to rely on optimistic growth projections to make future budgets balance. Surprisingly enough, however, there has been little talk of the role of growth when designing fiscal adjustment packages in developing nations. Furthermore, the effect of growth on fiscal solvency is larger the greater the stock of initial debt (this can be easily seen by multiplying 12 THE LIMITS OF STABILIZATION both sides of [1.3] by the denominator of the left-hand side). The intuition here is that growth effects on net worth are larger the greater your initial debt, because higher debt forces you to run a higher pri- mary surplus to service it. This means that any growth effect of infra- structure cuts is more costly in a high-debt country than in a low-debt country. A corollary is that an additional percentage point of growth reduces the amount of fiscal adjustment needed for solvency more in a high-debt country than in a low-debt country. Under what conditions does public infrastructure spending have powerful growth effects? It is likely to have a more positive effect when public infrastructure spending strongly complements private capital. If some forms of private capital can easily substitute for public infra- structure capital (as will be examined in the chapters on private pro- vision of infrastructure services), then the growth effects of public in- frastructure cuts will be lessened. In the end, it is an empirical issue. Easterly and Rebelo (1993) found in a large sample of countries that public infrastructure spending (measured as public spending on trans- port and communication) raised growth significantly, but the aggre- gate of all public enterprise investment spending had a negative effect on private investment. This suggests that there are many forms of pub- lic investment that substitute for private capital, but public spending on transport and communication is not one of them, at least over the sample period and countries considered by Easterly and Rebelo. Servén (1998) found a similar result for India. More generally, is opening up of infrastructure activities to the pri- vate sector sure to yield sufficient private investment to offset the cuts in public infrastructure provision? There is no reason to expect that this private­public offset will occur automatically. On the one hand, the opening-up needs to take place in the presence of an enabling in- stitutional and regulatory framework capable of attracting private in- vestment of the necessary volume and efficiency. On the other hand, the private sector response may be far from uniform across infrastruc- ture industries because the complementarity/substitutability between public and private projects may well differ across industries. The em- pirical record reviewed in this book offers clear proof of the lack of uniformity across infrastructure industries. Overview of This Volume This book presents the results of recent research sponsored by the World Bank's Latin America and Caribbean Region on the macro- economic dimensions of infrastructure in the area. Drawing from the INTRODUCTION 13 experience of more than a decade of public sector retrenchment from infrastructure activities and their opening up to private sector involve- ment, the main objective of the research was to assess the conse- quences of this changed private­public partnership from the perspec- tive of growth, public finances, and the quantity and quality of infrastructure services. In this general context, the book covers three main themes. First, it documents in detail the major trends in infrastructure provision in Latin America, offering a comparative perspective on the evolution of infrastructure spending and infrastructure stocks and on the changing roles of the public and private sectors. Second, the book provides a rigorous implementation of the analytical framework outlined in the first part of this introduction to gauge the impact of these infrastruc- ture trends on growth and public finances in Latin America. Third, it takes a first look at the macroeconomic consequences of private sector involvement and examines how the private sector response across countries and sectors has been shaped by regulatory and other factors. A brief summary of the book's contents is as follows. Chapter 2 sets the stage for the analysis in subsequent chapters by laying out the main facts regarding the performance of Latin America's infrastructure sec- tors during the period of macroeconomic adjustment and fiscal aus- terity that spans much of the 1980s and 1990s. The chapter builds from a comprehensive cross-country data set on public and private in- frastructure expenditure, infrastructure stocks, and (to the extent that information permits) their quality. These data were assembled for this research and used throughout the book. A detailed review of this in- formation, using the successful economies of East Asia as a benchmark for comparison, reveals that over the 1980s and 1990s Latin America fell considerably behind in both infrastructure quantity and quality. This widening gap can be attributed to a large extent to the gener- alized decline in infrastructure investment relative to GDP across Latin America over the period under consideration: as the chapter docu- ments, infrastructure spending is a good predictor of subsequent growth in infrastructure assets. The decline in infrastructure spending was led by the contraction of public infrastructure investment, which in a few countries virtually collapsed. Much, although not all, of the public spending decline can be traced to fiscal adjustment. In several of the region's major countries the cut in public infrastructure spending amounted to half or more of the reduction in the budget deficit ac- complished in those years. Contrary perhaps to popular perception, there is little evidence that the downward trend in public infrastructure investment mirrored the increased involvement of the private sector in infrastructure provision. 14 THE LIMITS OF STABILIZATION Chapter 2 shows that in many cases higher private investment came along with higher public investment as well, suggesting that the public and private sectors often played complementary--rather than compet- ing--roles. In most countries, private infrastructure investment did rise significantly with the opening up of infrastructure industries to private participation, but did so unevenly. The private sector response was most vigorous in telecommunications, and much weaker in roads and water. Finally, the evidence is not yet conclusive on the impact of in- creased private participation on the efficiency and quality of infra- structure, although for telecommunications private sector involvement is clearly associated with an improvement in service quality indicators. Against this background, chapters 3 and 4 put to work the analyti- cal framework outlined earlier based on the intertemporal budget con- straint. Chapter 3 provides a careful assessment of the contribution of Latin America's infrastructure gap to her output gap vis-à-vis the suc- cessful economies of East Asia. Over the 1980s and 1990s, the output gap between the two regions widened dramatically. Several studies have identified a close cross-country association between output growth and infrastructure growth, and this raises the question of what was the con- tribution of the widening infrastructure gap to the output gap. To answer this question, chapter 3 uses an infrastructure- augmented production function and, to identify it empirically, lays out an econometric framework able to separate the exogenous component of infrastructure growth from the endogenous one resulting from the impact of growing income on the demand for infrastructure services. The results from implementing this approach on a large cross-country time-series data set reveal a significantly positive effect of the exoge- nous component of infrastructure stocks on output, which is shown to be robust to alternative econometric specifications and measures of in- frastructure stocks. These empirical estimates are then used to assess the contribution of Latin America's lagging infrastructure accumulation to its lagging growth performance. Although there is a considerable degree of diver- sity across countries in the region, on average Latin America's infra- structure slowdown relative to East Asia could account for as much as one-third of the widening output gap between both regions. The substantial growth impact of infrastructure in Latin America that these results indicate suggests that the fiscal retrenchment of the 1980s and 1990s, which entailed a considerable degree of infrastruc- ture compression and hence a potentially significant growth cost, rep- resented a highly inefficient way to adjust public finances, as implied by the analytical framework outlined above. Chapter 4 examines this question and provides a detailed quantitative assessment of the INTRODUCTION 15 efficiency of infrastructure cuts in enhancing public sector solvency in Latin America. To do this assessment, the chapter considers the three components of the link between public infrastructure spending and public sector solvency: the effect of infrastructure spending on stock accumulation, the contribution of stock accumulation to output growth, and the im- pact of growth on the primary deficit. These three ingredients are then combined to gauge the contribution of infrastructure cuts to public solvency--or, to put it differently, the extent to which the short-run fa- vorable effect of spending cuts on public finances is offset by declining long-term revenue collection capacity caused by reduced growth. This analysis yields some key insights. First, the GDP growth cost of reduced infrastructure asset accumulation resulting from lower pub- lic investment was substantial in Latin America, exceeding 1 percent a year in several countries. As a result, much of the supposedly favorable effect of the investment cuts on the public sector balance was offset by higher future deficits resulting from lowered output growth. However, this offset, as well as the magnitude of the growth cost, show consid- erable variation across countries, depending on their respective levels of public debt and the composition of the infrastructure investment contraction. Estimated offset coefficients for major Latin American countries range from a low of 20 percent to a high exceeding 80 per- cent. The conclusion from this analysis is that, by engaging in this kind of fiscal adjustment biased against infrastructure, some Latin Ameri- can governments may have figuratively shot themselves in the foot. As already noted, the private sector response to the opening up of Latin America's infrastructure industries has been characterized by considerable diversity across countries and infrastructure sectors. Chapters 5 and 6 shift the focus of analysis to the causes and conse- quences of this uneven change in the degree of private participation. Chapter 5 provides an empirical assessment of the impact of pri- vate entry on major macroeconomic aggregates--per capita GDP, pri- vate investment, and current and capital expenditures of the public sector. The analysis uncovers a contrast between private sector entry in utilities and in transport. For example, private participation in transport is associated with increases in current public spending and decreases in public investment. The opposite happens with private participation in utilities. Thus, in the former case private investment crowds out public investment but likely requires increased subsidies, whereas in the latter these conclusions are reversed. Contrary perhaps to common perception, the general implication is that private sector involvement does not have an obvious favorable effect on public finances, which should sound a cautionary note for those policymakers 16 THE LIMITS OF STABILIZATION looking to privatization of infrastructures as a remedy for their fiscal troubles. The opening up of infrastructure to private initiative in different Latin American countries has taken place under a wide variety of reg- ulatory frameworks. In some cases, the opening up preceded the establishment of regulations and regulatory bodies. As already noted, infrastructure projects often entail large sunk costs, which put them at high risk of expropriation through adverse changes in regulation. Hence a sound and credible regulatory framework can make a big dif- ference in lowering perceived expropriation risk and attracting private investment. Chapter 6 presents an assessment of the role that the regulatory framework has played in Latin America in this regard. Using data on private investment in infrastructure in nine major countries, and tak- ing advantage of the diversity of private entry experiences and regula- tory environments across countries and infrastructure sectors, the chapter provides an empirical evaluation of the impact of key aspects of the regulatory regime: the passage of formal legislation liberalizing the investment regime, the establishment of a regulatory body, the de- gree of autonomy of the regulator, the size of the regulatory body, the sharing of risk between investor and regulator, and so on. It should be stressed that this is the first systematic exploration of these issues from a macroeconomic perspective, and as such its findings should be taken as tentative and suggestive of directions for further research. The results show that the existence of a regulatory body by itself does not have much effect on private participation once the passage of liberalization legislation has been taken into account. Among the fea- tures of the regulatory framework relevant for investment, the chapter finds that systems in which regulators are appointed by the executive are associated with higher private sector involvement than if the selec- tion of the regulator goes through the legislative branch, a fact that may reflect the critical role of regulatory predictability and credibility. Furthermore, private investment is positively associated with the regu- lator not being funded solely by the government, which likely echoes the importance of regulatory independence from the perspective of pri- vate investors. Final Remarks The findings reported in this volume reveal a mixed record regarding Latin America's experience with the public sector's withdrawal from infrastructure activities and its opening up to private initiative. But a INTRODUCTION 17 clear message emerges that fiscal austerity centered on the sale of pub- lic assets and the compression of growth-enhancing expenditures--in the hope that the private sector will come to the rescue--is not a prom- ising way to place Latin America's public finances on a firm footing. Illusory fiscal adjustment has been a worldwide phenomenon in re- cent years. The summary by Eichengreen and Wyplosz (1998) of Eu- ropean countries' adjustment to the Maastricht criteria is apposite: "European governments have relied on one-off measures--central bank sales of gold, refundable euro taxes, appropriation for the gen- eral budget of public enterprise reserves, and sales of strategic petro- leum reserves--to meet the Maastricht fiscal criteria for 1997." Eu- rope has seen a backlash to this widespread cheating, and European policymakers are now finally grappling with such long-term issues as how to deal with their crushing pension obligations. Yet concerns re- main that the rules set forth in the Stability Pact may permanently re- duce the public sector's contribution to infrastructure capital accumu- lation, and various proposals have been advanced for some kind of "Golden Rule" or other similar provision to protect public investment in Europe (see Balassone and Franco 2000). The fact that "everyone is doing it" offers little consolation for Latin Americans against the consequences of cosmetic fiscal adjust- ment based on disinvestment in infrastructure capital. The restoration of both fiscal solvency and long-term growth will require a more far- sighted approach to fiscal adjustment that protects the growth- enhancing spending done by Latin American governments. Fortunately, there is increasing awareness among policymakers that an intertemporal perspective on budget deficits and fiscal adjustment measures is the only way to properly evaluate their effect on fiscal sol- vency. It is hoped that this volume will contribute to this trend. Notes 1. See, for example, Edwards (1995) for a comprehensive account of macroeconomic adjustment in Latin America during the 1980s and early 1990s. 2. For example, Hicks (1991) reviewed several fiscal contraction episodes in developing countries during the 1990s and found that infrastructure was the item suffering by far the largest spending cuts in relative terms. 3. For example, Balassone and Franco (2000) noted that Italy has consis- tently maintained one of the highest ratios of public investment to GDP among industrial countries, but its relative position in terms of infrastructure stocks has failed to improve. 4. There are several alternative ways to present the intertemporal budget constraint. The method followed here is that of Buiter (1990, chapter 5). 18 THE LIMITS OF STABILIZATION 5. This is not exactly the conventional primary surplus. The top item in the left-hand side of (1.1) excludes the direct cash revenues derived from pub- lic capital assets, which are instead included in the excess return item. Thus, the primary surplus measured here excludes such revenues. 6. Also, if public investment does not translate one-for-one into public cap- ital accumulation (because of inefficiency and waste in public procurement, for example), then an additional term with a negative sign would appear on the left- hand side of the solvency constraint. It would simply capture the present value of the divergence between cumulative investment expenditure and the capital stock. 7. Implicit liabilities are those involving a moral or expected obligation that is not established by law or contract. Contingent liabilities are those trig- gered by an event that may or may not occur. See Brixi 2003. 8. Easterly 1999. In general, such a procedure would enhance solvency only if the rate of increase of the price of oil falls short of the discount rate. 9. The 1986 Government Finance Statistics Manual (IMF 1986, p. 31) recommended cash rather than accrual accounting. Current practice uses a mixture of cash and accrual accounting. When arrears become a serious prob- lem, the conventional approach to deficits in developing countries will often show the deficits explicitly as a financing item for an accrual-based deficit tar- get. The 1996 GFS Manual (IMF 1996, p. 16) recommended accrual account- ing. However, arrears still can be used to temporarily meet a gross public debt target because they are not included in the gross public debt. 10. To put it differently, privatization is likely to enhance solvency if the net returns that the purchaser can derive from operating the asset exceed those that the government would have been able to obtain. This can be the case if the purchaser is able to extract monopoly profits that the government was not exploiting, or also if the purchaser can operate the asset more efficiently than the government. 11. The Egyptian example is from World Bank (1997, p. 84). 12. Otherwise the economy is dynamically inefficient. In any case, the sol- vency constraint is trivially satisfied if r < g. 13. This implies that revenues and expenditures grow at the same rate as GDP. The time series analysis in chapter 4 suggests that this is not an unreal- istic assumption for the long run. 14. This is simply accounting and it does not address the issue of whether the given public debt ratio is optimal. 15. A similar argument was provided by Buiter (1990, chapter 13), who showed rigorously that public investment cuts can be inflationary in the long run--in other words, they may force the government to increase its recourse to money printing to balance the fiscal accounts. 16. Easterly (2001) discussed this point at length. References Anand, Ritu, and S. van Wijnbergen. 1989. "Inflation and the Financing of Government Expenditure: An Introductory Analysis with an Application to Turkey." World Bank Economic Review 3 (January): 17­38. Washington, D.C. Auerbach, Alan. 1997. "Quantifying the Current U.S. Fiscal Imbalance." National Tax Journal 50 (3): 387­98. INTRODUCTION 19 Balassone, Fabrizio, and Daniele Franco. 2000. "Public Investment, the Stability Pact and the `Golden Rule.'" Fiscal Studies 21 (2): 207­29. Blanchard, Oliver, Jean-Claude Chouraqui, Robert Hagemann, and Nicola Sartor. 1990. "The Sustainability of Fiscal Policy: New Answers to an Old Question." OECD (Organisation of Economic Co-operation and Develop- ment) Economic Studies 15 (autumn). Brixi, Hana. 2003. "Government at Risk: Dealing with Contingent Liabil- ities and Other Fiscal Risk." In Anwar Shana, ed., Ensuring Accountability When There Is No Bottom Line. Volume 1 of Handbook on Public Sector Per- formance Reviews. Washington D.C.: World Bank. Brixi, Hana, and Allen Schick. 2001. Government Risk: Contingent Lia- bilities and Fiscal Risk. Oxford and New York: Oxford University Press. Buiter, Willem. 1990. Principles of Budgetary and Financial Policy. Cam- bridge, Mass.: MIT Press. Buiter, Willem, and Urjit Patel. 1992. "Debt, Deficits and Inflation: An Application to the Public Finances of India." Journal of Public Economics 47 (March): 171­205. ------. 1997. "Budgetary Aspects of Stabilization and Structural Adjust- ment in India: The Painful Road to a Sustainable Fiscal-Financial-Monetary Plan." In Mario I. Blejer and Teresa Ter-Minassian, eds., Macroeconomic Di- mensions of Public Finance: Essays in Honour of Vito Tanzi. Studies in the Modern World Economy 5. London and New York: Routledge. Cuddington, John. 1997. "Analyzing the Sustainability of Fiscal Deficits in Developing Countries." World Bank Policy Research Working Paper 1784. Washington, D.C. Easterly, William. 1999. "When Is Fiscal Adjustment an Illusion?" Eco- nomic Policy: A European Forum 14 (28): 55­76. ------. 2001. "Growth Implosions and Debt Explosions: Do Growth Slowdowns Explain Public Debt Crises?" International Monetary Fund Semi- nar Series (January 13): 1­38. Easterly, William, and Sergio Rebelo. 1993. "Fiscal Policy and Economic Growth: An Empirical Investigation." Journal of Monetary Economics 32 (3): 417­58. Edwards, Sebastian. 1995. Crisis and Reform in Latin America: From De- spair to Hope. August. Washington, D.C.: World Bank. Eichengreen, Barry, and Charles Wyplosz. 1998. "The Stability Pact: More Than a Minor Nuisance." Economic Policy: A European Forum 13 (26): 65­113. European Commission. 1998. Euro 1999: 25 March 1998: Report on Progress Towards Convergence and the Recommendation with a View to the Transition to the Third Stage of Economic and Monetary Union. Brussels. 20 THE LIMITS OF STABILIZATION Gyamfi, Peter, Luis Gutierrez, and Guillermo Yepes. 1992. Infrastructure Maintenance in LAC: The Costs of Neglect and Options for Improvement. Volume 4: The Road Sector. World Bank Latin America and the Caribbean Technical Department Regional Studies Program, Report 17 (June). Washing- ton, D.C. Hicks, Norman. 1991. "Expenditure Reduction in Developing Countries Revisited." Journal of International Development 3 (1): 29­37. Hildebrand, Philippe. 1996. "Maastricht Fudge Factor." The International Economy (United States) 10: 50­52. IMF (International Monetary Fund). 1986, 1996. Government Finance Statistics Manual: Annotated Outline. Washington D.C. Irwin, Timothy, Michael Klein, Guillermo Perry, and Mateen Thobani. 1997. Dealing with Public Risk in Private Infrastructure. World Bank Latin America and Caribbean Studies. Washington D.C. Kee, James. 1987. "President Reagan's FY88 Budget: The Deficit Drives the Debate." Public Budgeting and Finance 7 (2). Kopits, George, and Jon Craig. 1998. "Transparency in Government Op- erations." IMF Occasional Paper 162. Washington, D.C. Mackenzie, George, and Peter Stella. 1996. "Quasi-Fiscal Operations of Public Financial." IMF Occasional Paper 142. Washington, D.C. Servén, Luis. 1998. "Does Public Capital Crowd Out Private Capital? Evi- dence from India." World Bank Policy Research Working Paper 1613. Wash- ington, D.C. Servén, Luis, and Andrés Solimano. 1992. "Striving for Growth after Ad- justment: The Role of Capital Formation." World Bank Regional and Sectoral Studies. Washington, D.C. White, Joseph, and Aaron Wildavsky. 1989. The Deficits and the Public In- terest: The Search for Responsible Budgeting in the 1980s. Berkeley: Univer- sity of California Press. World Bank. 1992. Adjustment Lending and Mobilization of Private and Public Resources for Growth. Policy and Research Series 22. Washington, D.C. ------. 1994. World Development Report. Washington, D.C. ------. 1997. Egypt: Issues in Sustaining Economic Growth. Country Eco- nomic Memorandum (March). Washington, D.C. 2 Latin America's Infrastructure in the Era of Macroeconomic Crises César Calderón, William Easterly, and Luis Servén DID THE QUANTITY AND QUALITY OF Latin America's infrastructure suffer from the prolonged period of macroeconomic stabilization and fiscal austerity in the 1980s and 1990s? To address that question, this chapter provides a comprehensive overview of the evolution of Latin America's infrastructure stocks, quality, and spending over the past decades. The chapter does not attempt to answer the question posed in a formal econometric manner that specifies the counterfactual of what would have happened if Latin America had not entered a period of macroeco- nomic crises. Instead some illustrative facts are given that may be con- sistent with some answers to this question and not with others. A long-standing literature has noted that fiscal adjustment is often implemented through cuts in public investment, including infrastruc- ture. As Roubini and Sachs (1989) observed, "In periods of restrictive fiscal policies . . . capital expenditures are the first to be reduced (often drastically)." During fiscal adjustment, the 1988 World Development Report of the World Bank (p. 113) found that governments cut capital spending by far more (about 35 percent) than other public sector cate- gories like wages (which were cut by about 10 percent). Also, Hicks (1991) found that from 1970 to 1984, in countries with declining gov- ernment expenditure, governments cut capital expenditures by more than current expenditures (­27.8 percent and ­7.2 percent, respectively). Servén (1997) found that Latin American public investment fell 2.5 percent of gross domestic product (GDP) from the 1970s to the 1980s, when the region was adjusting. East Asia, which did not need to adjust in the 1980s, had an increase of 3.7 percent. The World Bank (1994) 21 22 THE LIMITS OF STABILIZATION found that when African countries lowered their budget deficits from 1981­86 to 1990­91, "most of the cuts were in capital spending" (p. 47). De Haan, Sturm, and Sikken (1996) found that public invest- ment is reduced during times of fiscal stringency in Organisation for Economic Co-operation and Development (OECD) countries. Easterly (1999) argued that governments that do not really want to adjust en- gage in the illusion of adjustment by cutting both public debt and pub- lic assets (infrastructure). This chapter begins by assessing trends in quantity and quality of in- frastructure using data on 19 Latin American and Caribbean countries, excluding the smaller Caribbean economies because their data avail- ability is more limited and to avoid influencing the region-wide statis- tics with too many observations from small island economies. The seven East Asian Miracle countries serve as a comparator group against which the performance of Latin America can be judged. The chapter then looks at trends in infrastructure spending for nine major Latin American economies on which country data are available. The discussion examines the extent to which fiscal deficit reductions and public infrastructure spending reductions have moved together. The next step is to investigate to what extent the changes in public in- frastructure spending were driven by the privatization of infrastructure and the increased private spending on infrastructure. Finally, panel data econometric analysis is used to link the time path of infrastruc- ture quantity and quality to the path of infrastructure spending. Information on the efficiency of infrastructure investment (that is, the unit cost of infrastructure stocks) and the quality of infrastructure stocks is notoriously scarce; therefore much of the analysis in this chap- ter relies on comparisons of infrastructure stocks and expenditures across countries and time periods. This raises a major caveat--that our infrastructure spending and infrastructure stock measures are noisy indicators of the accumulation and availability of infrastructure assets, respectively. Thus, more infrastructure investment and bigger stocks are not necessarily better because they could just reflect more waste of resources. It is important to keep in mind this fundamental limitation of the available data throughout the discussion in this chapter. Comparative Trends in Latin American Infrastructure Quantity and Quality The first step is to review the evolution of Latin America's infrastructure indicators.1 To place it in perspective, the experience of Latin America is compared with that of a set of successful developing countries that did not need to undergo macroeconomic adjustment for most of the 1980s LATIN AMERICA'S INFRASTRUCTURE 23 and 1990s--the East Asian Miracle economies (as given in World Bank 1994). These are Hong Kong (China), Indonesia, Republic of Korea, Malaysia, Singapore, Taiwan (China), and Thailand. Furthermore, an assessment of the progress of these two developing regions vis-à-vis the industrial economies of the OECD in terms of infrastructure indicators is also carried out.2 The East Asian economies were growing faster than Latin American economies, so in principle faster growth in infrastruc- ture could reflect demand as well as supply factors. Rigorous analysis of the infrastructure-growth nexus is deferred to chapter 3. The initial focus is on the comparative performance in infrastruc- ture stocks. Starting with telecommunications, figure 2.1a shows the evolution of main telephone lines per worker (that is, relative to the labor force) over the past two decades across the three regions under consideration. In each case the regional median is shown. The discrep- ancy is tremendous in the growth in phone lines per worker between Latin America and East Asia. In 1980, Latin America trailed East Asia by a relatively small margin--89 versus 132 main lines per 1,000 workers, with both regions far behind industrial economies. Since then, however, the number of phone lines has expanded much more rapidly in East Asia than in Latin America. As a result, by 1997 East Asia had more than twice as many phone lines per 1,000 workers as Latin America--500 versus 232, respectively. Figure 2.1a suggests an apparent stagnation in main phone lines in East Asia and industrial countries in the 1990s, but this turns out to be caused by the substitution of cell phones for land lines. The graph in- cluding cell phones (figure 2.1b) shows that in these two regions the expansion of total telephone lines has continued without interruption throughout the 1990s, making Latin America's lag relative to these two regions even greater than in the case of main phone lines. By 1997 the total number of phone lines per 1,000 workers was 718 in East Asia, compared with 289 in Latin America. Other measures of the availability of telephone services portray a similar picture. Figure 2.1c reports regional medians of local connec- tion capacity per worker. It confirms that a huge gap has opened be- tween East Asia and Latin America since 1980, with few signs of abat- ing in the 1990s. And the same pattern seems to emerge for newer telecommunications technologies. For example, figure 2.1d shows that in the late 1990s East Asia acquired a considerable lead over Latin America in the number of Internet hosts per worker. The regional indicators in the above figures conceal a wide range of variation across Latin American countries, however. Figure 2.1e shows that a few of them (Argentina, Chile, and Costa Rica) were roughly on par with East Asia in main phone lines per worker in 1997, with Uruguay even ahead of East Asia. At the other end of the 24 THE LIMITS OF STABILIZATION Figure 2.1 Comparative Performance in Infrastructure Stocks a. Telephone Main Lines, 1980­97 (medians by region) Main lines (per 1,000 workers) 1,200 1,000 800 600 400 200 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries b. Total Telephone Lines, 1980­97 (medians by region) Main lines, including cellular phones (per 1,000 workers) 1,600 1,400 1,200 1,000 800 600 400 200 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries c. Local Connection Capacity, 1980­97 (medians by region) Connection capacity of local exchanges (per 1,000 workers) 1,400 1,200 1,000 800 600 400 200 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries LATIN AMERICA'S INFRASTRUCTURE 25 Figure 2.1 (continued) d. Internet Hosts, 1994­2000 (medians by region) Internet hosts (per 1,000 workers) 120 100.2 100 80 60 47.1 40 16.1 19.7 20 7.5 0.0 0.7 0.1 1.2 0.7 4.8 1.6 0 1994 1995 1998 2000 Latin America East Asia 7 Industrial countries e. Telephone Lines per Worker, 1980 and 1997 (lines per 1,000) Uruguay East Asia median Argentina Chile Costa Rica Colombia Panama Jamaica Venezuela, R.B. de Mexico Latin America median Brazil Dominican Republic Peru Ecuador Bolivia El Salvador Paraguay Guatemala 1997 Honduras 1980 Nicaragua 0 100 200 300 400 500 600 spectrum, three smaller economies (Guatemala, Honduras, and Nicaragua) lagged far behind. Even some major economies such as Brazil and Mexico have also lost considerable ground over time: they lagged way behind East Asia in 1997, even though in 1980 they had more phone lines per worker than the East Asia median.3 26 THE LIMITS OF STABILIZATION Figure 2.2 Power Generating Capacity by Region, 1980­97 (medians by region) Electric generating capacity (mw per 1,000 workers) 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries Figure 2.2 shows the trends in electricity generating capacity per worker. Here too East Asia has acquired a sizable advantage over Latin America during the past two decades. In 1980, East Asia's power generating capacity per worker was only 70 percent of Latin Amer- ica's; in 1997, it had risen to 165 percent. As in telecommunications, there is considerable cross-country vari- ation in power generation capacity in Latin America. Figure 2.3 re- veals that three countries exceeded the East Asia median in 1997-- Argentina, Uruguay, and República Bolivariana de Venezuela. However, several major Latin American economies, such as Brazil, Colombia, Ecuador, and Peru, lagged far behind and have made little progress over the past two decades. Figure 2.4a shows the length of the road network per worker in Latin America and the East Asian newly industrialized countries (NICs). Obviously, roads can vary greatly in quality, so cross-country comparisons have to be made with great care.4 Here Latin America has remained ahead of East Asia throughout the period of analysis, although the gap between the two regions has narrowed considerably over time. Figure 2.4b presents similar information concerning overall transport routes, which include railways in addition to roads; the qual- itative pattern is the same as in the preceding figure. Finally, figure 2.4c offers a comparative perspective of paved roads. Here the pattern is somewhat different. In 1980 Latin America was way ahead of East Asia in the length of the paved road network, but by the second half of the 1990s East Asia had reached virtual parity, with both regions still far behind industrial economies. The country-specific detail in figure 2.5 shows that in all but one of the Latin American economies listed, the length of the road network LATIN AMERICA'S INFRASTRUCTURE 27 Figure 2.3 Power Generating Capacity per Worker by Country, 1980 and 1997 (megawatts per 1,000) Venezuela, R.B. de Uruguay Argentina East Asia median Mexico Chile Jamaica Costa Rica Panama Latin America median Brazil Ecuador Colombia Peru Dominican Republic El Salvador Bolivia Nicaragua 1997 Honduras 1980 Guatemala 0 0.50 1.00 1.50 2.00 2.50 relative to the number of workers has declined over the past two decades. The only exception is Uruguay, which experienced a signifi- cant expansion even though in 1980 it was already the country with the largest road stock per worker. The picture in paved roads is simi- lar--the majority of the region's countries witnessed a decline in their paved road stock per worker, in contrast with the expansion that took place in East Asia over the past two decades. Figure 2.6 shows the trends in safe water availability, in terms of the fraction of the total population with access to safe water in the two regions. The data are much more limited than for the earlier indicators and span only the years 1988 to 1993. Over that period, East Asia showed a steady improvement in access to clean water, whereas Latin America suffered a deterioration. As a result, by 1993 the initial advantage of Latin America over East Asia had been reversed. The next step is a review of indicators of infrastructure quality and excess demand. Unfortunately, the data are much sparser on these indicators and only three--telephone line waiting times, electrical power losses, and percentage of roads paved--offer a continuous time series over several decades. Figure 2.7a shows the waiting time for a telephone line, calculated as the number on the waiting list for main 28 THE LIMITS OF STABILIZATION Figure 2.4 Length Comparisons of Transportation Routes a. Total Roads, 1980­97 (medians by region) Total roads (km per 1,000 workers) 35 30 25 20 15 10 5 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries b. Roads plus Railways, 1980­97 (medians by region) Roads + Railways (km per 1,000 workers) 35 30 25 20 15 10 5 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries c. Paved Roads, 1980­97 (medians by region) Paved roads (km per 1,000 workers) 25 20 15 10 5 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries LATIN AMERICA'S INFRASTRUCTURE 29 Figure 2.5 Road Length per Worker by Country, 1980 and 1997 (km per 1,000) Uruguay Costa Rica Brazil Bolivia Paraguay Argentina Jamaica Chile Venezuela, R.B. de Latin America median Panama Nicaragua Ecuador Mexico Peru Honduras Colombia Dominican Republic El Salvador 1997 East Asia median 1980 Guatemala 0 5 10 15 20 25 30 35 40 lines divided by the change in main lines in that year. In the early 1980s the median waiting time was 3 years in Latin America, versus 1.5 years in East Asia. In the 1980s and 1990s the backlog declined steadily in East Asia (and disappeared in industrial countries). In contrast, the median waiting time rose sharply in Latin America over the 1980s, to Figure 2.6 Access to Clean Water, 1985­93 (medians by region) Population with access to safe water (percent) 100 80 60 40 20 0 1985 1988 1993 Latin America East Asia 7 Industrial countries 30 THE LIMITS OF STABILIZATION Figure 2.7 Infrastructure Quality and Excess Demand a. Waiting Time for Main Lines, 1980­98 (medians by region) Waiting years for main lines 6 5 4 3 2 1 0 1980-84 1985-89 1990-94 1995-98 Latin America East Asia 7 Industrial countries b. Telephone Faults, 1991­95 (medians by region) Telephone faults per 100 main lines 70 60 50 40 30 20 10 0 1991 1993 1995 Latin America East Asia 7 Industrial countries c. Unsuccessful Local Calls, 1991­95 (medians by region) Unsuccessful local calls (percent) 50 40 30 20 10 0 1991 1993 1995 Latin America East Asia 7 Industrial countries LATIN AMERICA'S INFRASTRUCTURE 31 decline later in the 1990s. However, by 1997 Latin America still had a median waiting time in excess of half a year, whereas in East Asia the typical country had virtually no main line waiting list after 1994. This provides an indication of excess demand for phone lines in Latin America in the era of macroeconomic crises in the 1980s and 1990s and suggests strongly that the lag relative to East Asia was caused mainly by supply constraints rather than by lower demand. Figures 2.7b and 2.7c report two measures of the performance of the phone network: the number of telephone faults per 100 lines and the per- centage of unsuccessful local calls. In both cases the country coverage of the information is severely limited and the regional comparisons have to be made with caution because the regional aggregates include only a few countries. Furthermore, the available data refer only to 1991­95. The percentage of unsuccessful local calls does not show much differ- ence between Latin America and East Asia. In turn, the data on telephone faults per main line show much poorer quality of service in Latin Amer- ica than in East Asia. Because data do not exist on earlier years, it is im- possible to say whether Latin America's worse telephone service quality relative to East Asia's was caused by the macroeconomic crises of the past two decades or if it already existed prior to them. In any case, the obvious conclusion is that Latin America lags behind East Asia not only in the quantity of telecommunication services but also in their quality. Regarding power, the percentage of transmission losses relative to to- tal output offers a crude measure of the efficiency of the power network. Figure 2.8 offers a cross-regional perspective on power losses. The figure shows a clear deterioration in the power system during the era of fiscal austerity in Latin America in the 1980s and 1990s, with an incipient reversion only after 1995. In contrast, East Asia had roughly constant Figure 2.8 Power Losses by Region, 1980­97 (medians by region) Power losses (percentage of output) 20 18 16 14 12 10 8 6 4 2 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries 32 THE LIMITS OF STABILIZATION electrical power losses. Thus, although Latin America's service quantity indicator (generating capacity per worker) shown above displayed a modest upward trend during the past two decades, the quality of that service deteriorated sharply. Among Latin American countries, figure 2.9 shows that only Paraguay and Costa Rica improved on the East Asia norm for power losses in 1997. All other countries show higher power losses, strikingly large in some cases (Dominican Republic, Honduras, and Nicaragua). Moreover, only four countries (Chile, El Salvador, Jamaica, and Paraguay) experienced an improvement between 1980 and 1997. Finally, a rough measure of the quality of the surface transportation network is given by the percentage of roads paved. This is shown in figure 2.10a, which reveals a sharp increase in the road quality thus measured in East Asia, with the percentage of roads paved rising from 60 to 75 percent between 1980 and 1990. In contrast, Latin America made virtually no progress along this dimension over the past two decades. The country-specific data (figure 2.10b) show a similarly bleak picture: all Latin American countries fall well short of the East Asia median, with Jamaica as the only country coming close to it. Figure 2.9 Power Losses by Country, 1980 and 1997 (percentage of power output) Paraguay Costa Rica East Asia median Chile Bolivia Jamaica Guatemala El Salvador Mexico Peru Latin America median Brazil Argentina Uruguay Venezuela, R.B. de Panama Colombia 1997 Ecuador 1980 Honduras Nicaragua Dominican Republic 0 5 10 15 20 25 30 LATIN AMERICA'S INFRASTRUCTURE 33 Figure 2.10 Comparisons of Surface Transportation Quality a. Percentage of Total Roads Paved, 1980­97 (medians by region) Paved roads 100 90 80 70 60 50 40 30 20 10 0 1980 1990 1995 1997 Latin America East Asia 7 Industrial countries b. Paved Road Length, 1980 and 1997 (percentage of total road length) East Asia median Jamaica Argentina Venezuela, R.B. de Panama Mexico Guatemala Honduras Ecuador Latin America median Costa Rica Chile Uruguay El Salvador Colombia Peru Nicaragua Paraguay 1997 Brazil 1980 Bolivia 0 10 20 30 40 50 60 70 80 Trends in Infrastructure Spending in Latin America The comparative evidence just reviewed suggests that Latin America fell behind East Asia along most dimensions of infrastructure quantity and quality over the 1980s and 1990s, although performance varies a 34 THE LIMITS OF STABILIZATION Figure 2.11 Total Investment in Infrastructure in Selected Latin American Countries, 1980­98 Percentage of GDP 9.0 8.0 7.0 6.0 5.0 4.0 3.0 2.0 1.0 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru great deal across Latin American countries. The next task is to assess whether these trends relate to the observed performance of infrastruc- ture spending in the region. This is done using infrastructure investment data from major Latin American economies over the past two decades.5 Figure 2.11 depicts the trajectory of total infrastructure investment as a ratio to GDP in six major Latin American countries since 1980. The figure reveals three salient facts. First, the volume of infrastructure investment varies considerably across the countries shown. In the late 1990s, it ranged from 1 percent of GDP in Mexico to more than 7 per- cent in Colombia. Second, in most countries infrastructure investment experienced a substantial decline around the mid-1980s, which was reversed only partially, if at all, in the late 1990s. Third, Colombia and Chile are exceptions to this rule; they witnessed an infrastructure investment expansion, particularly during the late 1990s. Investment performance varied also across infrastructure sectors. Figures 2.12a through 2.12d depict the trajectory of total investment, relative to GDP, in each of four important sectors--telecommunica- tions, power, transport, and water. Investment in telecommunications displayed an upward trend in several countries, with Brazil and Mex- ico being the main exceptions (figure 2.12a). In power (figure 2.12b), by contrast, most countries witnessed an investment decline, particu- larly sharp in Brazil, which had been the leading investor in the early 1980s, and in Peru, where investment levels dropped to virtually noth- ing in the early 1990s. The only exception was Colombia, which more than doubled its power investment in the late 1990s. In transport (figure 2.12c), investment also followed a declining trend after the mid-1980s, with Chile as the only country to display a sustained recovery at the end of the 1990s. In a few countries (Argentina, Brazil, LATIN AMERICA'S INFRASTRUCTURE 35 Figure 2.12 Investment in Infrastructure in Selected Latin American Countries, by Sector, 1980­98 a. Telecommunications Percentage of GDP 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru b. Power Percentage of GDP 6.0 5.0 4.0 3.0 2.0 1.0 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru c. Roads and Railways Percentage of GDP 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru 36 THE LIMITS OF STABILIZATION Figure 2.12 (continued) d. Water Percentage of GDP 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru and Peru), investment remained at extremely low levels throughout the 1990s. Finally, in water and sanitation (figure 2.12d) both investment levels and trends were diverse across countries: investment fell to very low values in Peru but rose to record highs in Colombia in the late 1990s. Behavior of Public Infrastructure Investment To what extent did this performance of total infrastructure investment reflect the performance of public investment? With the public sector as the main or, in many cases, the only investor, the answer is that total and public investment moved closely together in most countries, at least until the mid-1990s. Figure 2.13 depicts the time path of public infrastructure investment as a percentage of GDP. Except for the late 1990s, the graph is strikingly similar to that for total investment (figure 2.11 above). It shows that public infrastructure investment col- lapsed after the mid-1980s in five of the six countries considered. The exception once again was Colombia, which succeeded in maintaining roughly unchanged public investment levels throughout the period. How was public infrastructure spending affected by fiscal austerity in the 1980s and 1990s in Latin America? Part of this expenditure re- duction may have reflected increasing efficiency in spending as meas- ured by a reduction in the unit cost of new assets of given quality. But this is unlikely to account for the bulk of the spending cut. Instead, the contraction of infrastructure spending likely resulted from the fiscal retrenchment the region underwent. It is possible to measure how much the change in infrastructure spending accounts for the observed change in the public sector surplus in each country. This is done in table 2.1, which compares the contraction LATIN AMERICA'S INFRASTRUCTURE 37 Figure 2.13 Public Investment in Infrastructure in Selected Latin American Countries, 1980­98 Percentage of GDP 6.0 5.0 4.0 3.0 2.0 1.0 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru in public investment with the change in the public sector primary (or noninterest) surplus, with both measured between the early 1980s and the late 1990s. The table shows that total public investment fell in all countries listed except for Bolivia. Public infrastructure investment, in turn, fell in seven out of the nine countries in the table. It rose in Ecuador Table 2.1 The Contribution of Infrastructure Compression to Fiscal Adjustment, Average 1980­84 versus Average 1995­98 Reduction in Contribution of public Change in investment reduction investment/ primary to fiscal adjustment GDP surplus/ (percent) Total Infrastructure GDP Total Infrastructure Country [1] [2] [3] [1] [3] [2] [3] Argentina 3.97 2.85 5.31 74.7 53.8 Bolivia 0.91 3.10 6.15 n.a. 50.3 Brazil 2.80 3.08 1.77 158.1 174.3 Chile 0.94 1.41 2.39 39.2 58.8 Colombia 0.45 0.04 4.69 9.6 n.a. Ecuador 1.57 0.68 1.81 87.0 n.a. Mexico 6.09 1.98 6.28 97.0 31.5 Peru 4.10 1.51 3.11 132.0 48.6 Venezuela, 3.49 0.41 1.88 n.a. n.a. R.B. de n.a. Not applicable. Source: Authors' calculations using the sources described in appendix 2A. 38 THE LIMITS OF STABILIZATION and showed virtually no change in Colombia. Comparison of columns one and two in the table reveals that in Bolivia, Brazil, and Chile, pub- lic infrastructure investment fell by more than total public investment, implying that noninfrastructure capital spending actually rose. The third column shows that the primary fiscal surplus rose in eight of nine coun- tries considered (all except República Bolivariana de Venezuela). In some of them, the magnitude of the rise was considerable. Columns four and five calculate the contribution of investment to the fiscal correction. Public investment contraction contributed significantly (that is, half of the total correction or more) to the adjustment in five of the eight adjusting economies. Infrastructure investment compression did the same in five economies. This is all the more remarkable because in- frastructure investment is typically a relatively small component of total public spending. The role of infrastructure compression was particularly large in Brazil, where the cut in infrastructure investment was almost twice as big as the fiscal correction. República Bolivariana de Venezuela is an extreme case because it reduced total and infrastructure investment without improving its primary surplus, so that in effect the investment compression financed a reduction in public saving. At the other extreme, Colombia and Ecuador managed to improve their fiscal balances with- out cutting public infrastructure (or total) investment. It is important to keep in mind that the figures in table 2.1 very likely understate the contribution of public infrastructure compression to the fiscal adjustment. The reason is that in most cases recurrent in- frastructure expenditures on operation and maintenance (O&M) were cut along with investment, so that the total decline in infrastructure- related spending was larger than the spending cut. This accounting decomposition does not impute a causal role to fis- cal adjustment, or even establish a correlation between fiscal correc- tion and infrastructure cuts. The (pooled) full-sample correlation be- tween the primary surplus and the public infrastructure investment/ GDP ratio is ­0.195, with a standard error of 0.077. This suggests a significant negative relation between both variables, but ignores the role of country-specific factors. A simple way to take them into account is to regress public infrastructure investment on the primary surplus, controlling for country-specific effects and time trends. This is done in table 2.2, which shows a quantitatively small, but highly significant, negative association between the primary fiscal balance and public infrastructure investment.6 However, there are significant country-specific time trends in infrastructure spending--negative in all cases except Ecuador and Colombia--which suggest that factors other than primary deficit adjustment may have been at work in the observed decline of public infrastructure investment. LATIN AMERICA'S INFRASTRUCTURE 39 Table 2.2 Regression of Public Infrastructure Investment/GDP on the Primary Balance/GDP Variable Coefficient t-statistic Primary surplus/GDP 0.0661 3.97 ARG-trend 0.0019 11.48 BOL-trend 0.0017 5.85 BRA-trend 0.0022 13.31 CHL-trend 0.0011 3.59 COL-trend 0.0001 0.51 ECU-trend 0.0005 1.50 MEX-trend 0.0010 4.52 PER-trend 0.0010 5.12 VEN, R.B. de­trend 0.0005 5.12 Adjusted R2 0.842 Number of countries 9 Number of observations 170 Note: FE­SUR fixed effects­seemingly unrelated regressions; FE­SUR estimates, 1980­98. Source: Authors' calculations. As noted above, only Colombia and Ecuador escaped the general trend toward infrastructure investment compression. These are also the only two countries, among those for which data exist, where the composition of public investment did not shift against infrastructure over the period of analysis. Figure 2.14 illustrates the changes over time in the composition of public investment between infrastructure and noninfrastructure items. It is immediately apparent that public in- frastructure investment lost ground relative to noninfrastructure in- vestment in all but the two countries mentioned. In these two coun- tries, infrastructure investment accounted for roughly 50 percent of total public investment in the late 1990s, whereas in other countries (Argentina, Mexico, and República Bolivariana de Venezuela) it rep- resented less than one-fourth of the total. The decline in public infrastructure investment experienced by most countries was not evenly distributed across infrastructure sectors. Figure 2.15 (p. 43) breaks down public infrastructure investment into four major components--power, telecommunications, transport, and water. In Argentina, Brazil, Chile, and Peru, the sharpest investment decline occurred in the power sector. In other countries--Bolivia, Mexico, and República Bolivariana de Venezuela--the compression affected transport investment most severely. Also, in several countries public investment in telecom had practically disappeared by the end of the 1990s. 40 THE LIMITS OF STABILIZATION Figure 2.14 Public Investment in Infrastructure and Noninfrastructure, by Country a. Argentina, 1980­98 Percentage of GDP 7 6 5 4 3 2 1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure b. Bolivia, 1980­98 Percentage of GDP 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure c. Brazil, 1980­98 Percentage of GDP 9 8 7 6 5 4 3 2 1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure LATIN AMERICA'S INFRASTRUCTURE 41 Figure 2.14 (continued) d. Chile, 1980­98 Percentage of GDP 7 6 5 4 3 2 1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure e. Colombia, 1980­98 Percentage of GDP 10 9 8 7 6 5 4 3 2 1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure f. Ecuador, 1981­98 Percentage of GDP 14 12 10 8 6 4 2 0 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure 42 THE LIMITS OF STABILIZATION Figure 2.14 (continued) g. Mexico, 1980­98 Percentage of GDP 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure h. Peru, 1980­98 Percentage of GDP 10 9 8 7 6 5 4 3 2 1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure i. República Bolivariana de Venezuela, 1980­98 Percentage of GDP 18 16 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure LATIN AMERICA'S INFRASTRUCTURE 43 Figure 2.15 Public Investment in Infrastructure, by Sector and by Country a. Argentina, 1980­98 Percentage of GDP 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water b. Bolivia, 1980­98 Percentage of GDP 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water c. Brazil, 1980­98 Percentage of GDP 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water 44 THE LIMITS OF STABILIZATION Figure 2.15 (continued) d. Chile, 1980­98 Percentage of GDP 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water e. Colombia, 1980­98 Percentage of GDP 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water f. Ecuador, 1981­98 Percentage of GDP 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water LATIN AMERICA'S INFRASTRUCTURE 45 Figure 2.15 (continued) g. Mexico, 1980­98 Percentage of GDP 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water h. Peru, 1980­98 Percentage of GDP 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water i. República Bolivariana de Venezuela, 1980­98 Percentage of GDP 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water 46 THE LIMITS OF STABILIZATION In a few instances, particularly in the telecom sector, the declining public investment trend documented above reflected the increasingly important role of private infrastructure investment. But this was by no means a generalized phenomenon across countries and infrastructure sectors. The next section reviews the observed pattern of private investment. Did Private Investment Replace Public Investment? Many Latin American countries opened their infrastructure sectors to private enterprise in the late 1980s and early 1990s. Table 2.3 shows the approximate date of effective opening up in different infrastructure subsectors in the countries under analysis. The opening up took a variety of forms, ranging from privatization of public enterprises to management contracts and private concessions. Appendix 2B provides a full account of the reforms across countries and sectors (see also Estache, Foster, and Wodon 2001). The private sector response to this opening up showed considerable diversity across countries and sectors. More detailed analyses of the response are provided in chapters 5 and 6. Here a descriptive account is given of the patterns of private infrastructure investment; the next section presents some formal empirical experiments. The evolution of private infrastructure investment relative to GDP in six major Latin American countries is depicted in figure 2.16. In five of the six countries, private investment took off in the late 1980s or Table 2.3 Infrastructure Reform Dates Country Telecom Electricity Roads Rail Water Argentina 1990 1992 1990 1990 1993 Bolivia 1987 1995 n.a. 1996 1997 Brazil 1995 1984 1996 1996 1995 Chile 1986 1986 1994 n.a. 1997 Colombia 1994 1992 1993 1995­97 1993 Ecuador 1994 1996 n.a. n.a. n.a. Mexico 1990 1998 1989 1996 1993 Peru 1990 1994 n.a. n.a. n.a. Venezuela, 1991 1992 n.a. n.a. n.a. R.B. de n.a. Not applicable. Source: See appendix 2B. LATIN AMERICA'S INFRASTRUCTURE 47 Figure 2.16 Private Investment in Infrastructure in Selected Latin American Countries, 1980­98 Percentage of GDP 4.5 4.0 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0.0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Argentina Brazil Chile Colombia Mexico Peru early 1990s. The exception is Brazil, where infrastructure investment of the private sector has hovered at around 1 percent of GDP over the past two decades. Among the other countries, Chile exhibited the ear- liest rise in private investment, followed thereafter by an upward trend that was also apparent in Colombia. In contrast, in Argentina and especially Mexico private investment appears to have stagnated in the second half of the 1990s. Also, in most countries, with Chile and Colombia as the exceptions, the total volume of private infrastructure investment remained quantitatively modest, at 1.5 percent of GDP or less. In some countries, the rise in private infrastructure investment came along with an upward trend in overall private investment. Figure 2.17 shows that this trend occurred in Argentina, Chile, and Peru, and to a more limited extent, in Ecuador and Mexico. In other cases, however, the increase in private infrastructure investment was not matched by a parallel rise in other types of private investment. Examples of this lat- ter situation were Bolivia, where noninfrastructure investment appears to have declined, as well as Colombia and República Bolivariana de Venezuela, where overall private investment displayed abrupt fluctua- tions during the period. The rise in private infrastructure investment was uneven not only across countries but also across infrastructure sectors. Figure 2.18 (p. 51) depicts the time pattern of private investment by sector of destination in the nine countries under analysis. In a majority of countries--Argentina, Chile, Ecuador, Peru, and República Bolivariana de Venezuela--the telecommunications sector became the prime 48 THE LIMITS OF STABILIZATION Figure 2.17 Private Investment in Infrastructure and Noninfrastructure, by Country a. Argentina, 1980­98 Percentage of GDP 25 20 15 10 5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure b. Bolivia, 1980­98 Percentage of GDP 18 16 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure c. Brazil, 1980­98 Percentage of GDP 20 18 16 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure LATIN AMERICA'S INFRASTRUCTURE 49 Figure 2.17 (continued) d. Chile, 1980­98 Percentage of GDP 25 20 15 10 5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure e. Colombia, 1980­98 Percentage of GDP 20 18 16 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure f. Ecuador, 1981­98 Percentage of GDP 18 16 14 12 10 8 6 4 2 0 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure 50 THE LIMITS OF STABILIZATION Figure 2.17 (continued) g. Mexico, 1980­98 Percentage of GDP 25 20 15 10 5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure h. Peru, 1980­98 Percentage of GDP 30 25 20 15 10 5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure i. República Bolivariana de Venezuela, 1980­98 Percentage of GDP 14 12 10 8 6 4 2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Infrastructure Noninfrastructure LATIN AMERICA'S INFRASTRUCTURE 51 Figure 2.18 Private Investment in Infrastructure, by Sector and by Country a. Argentina, 1980­98 Percentage of GDP 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 19971998 Telecommunications Electricity Roads + railways Water b. Bolivia, 1980­98 Percentage of GDP 3.5 3.0 2.5 2.0 1.5 1.0 0.5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water c. Brazil, 1980­98 Percentage of GDP 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water 52 THE LIMITS OF STABILIZATION Figure 2.18 (continued) d. Chile, 1980­98 Percentage of GDP 2.0 1.8 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water e. Colombia, 1980­98 Percentage of GDP 2.5 2.0 1.5 1.0 0.5 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water f. Ecuador, 1981­98 Percentage of GDP 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water LATIN AMERICA'S INFRASTRUCTURE 53 Figure 2.18 (continued) g. Mexico, 1980­98 Percentage of GDP 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water h. Peru, 1980­98 Percentage of GDP 1.2 1.0 0.8 0.6 0.4 0.2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water i. República Bolivariana de Venezuela, 1980­98 Percentage of GDP 1.2 1.0 0.8 0.6 0.4 0.2 0 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 Telecommunications Electricity Roads + railways Water 54 THE LIMITS OF STABILIZATION destination of private infrastructure spending in the late 1990s. In con- trast, the power sector took this role in Bolivia and Colombia. In Brazil there were no significant changes in the sectoral allocation of private in- frastructure investment over the period under analysis. Finally, Mexico appears to have been the only country where the transport sector be- came a prime destination for private investment. How did these sectoral patterns match the reforms introduced by most countries to open up their infrastructure sectors? To assess the response of private investment across countries and sectors to the reforms, the concept of reform time is used. To do this, the path of pri- vate infrastructure spending in each sector is examined before and after the year of reform identified in table 2.3 above, which is shown as year 0 in the panels of figure 2.19.7 Figure 2.19a shows the path of private telecommunications spend- ing in the nine Latin American economies considered. Private investment in this sector surged in the wake of opening up to private initiative. The largest increases were in Argentina and Chile, where postreform pri- vate investment peaked at US$40­$60 per capita. These increases are impressive compared with the average prereform public spending, which was around US$12 per capita.8 Similarly, figure 2.19b shows the path of private electricity spend- ing before and after liberalization. In most countries, private spending in this sector rose sharply around the time of reform, although Ecuador is a conspicuous exception. In most cases, the increases fell short of the average prereform public spending per capita in the sector, which was around US$32. For roads (figure 2.19c), Chile and Mexico show strong private spending increases, whereas the results seem more modest in the other countries reforming this sector. As a consequence, only in these two countries did total per capita spending actually rise after the reform. Also, in railways (figure 2.19d) only Argentina displayed a sharp in- crease in private spending per capita after the reforms. Even then, how- ever, the increase was sufficient only to keep total spending roughly at its prereform level (around US$10). In the other countries, total spend- ing per capita declined. Finally, results are also uneven in the water sector (figure 2.19e). In Bolivia private water spending increased before liberalization, perhaps in anticipation of the reform. In Argentina, liberalization yielded sig- nificant increases in spending. In Chile, however, private spending in the water sector showed little change after reform. In spite of this di- versity in private sector response, total per capita spending in the wa- ter sector rose in most countries. But the main reason for this is that, unlike in other sectors, public investment per capita in the water sector LATIN AMERICA'S INFRASTRUCTURE 55 Figure 2.19 Private Investment Per Capita around the Date of Reform in Selected Country, by Sector a. Telecommunications US$ at 1995 constant prices, 3-year moving average 70 60 50 40 30 20 10 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Argentina Bolivia Brazil Chile Colombia Ecuador Mexico Peru Venezuela, R.B. de b. Electricity US$ at 1995 constant prices, 3-year moving average 50 45 40 35 30 25 20 15 10 5 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Argentina Bolivia Brazil Chile Colombia Ecuador Mexico Peru Venezuela, R.B. de c. Roads US$ at 1995 constant prices, 3-year moving average 40 35 30 25 20 15 10 5 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Argentina Brazil Chile Colombia Mexico 56 THE LIMITS OF STABILIZATION Figure 2.19 (continued) d. Railways US$ at 1995 constant prices, 3-year moving average 20 18 16 14 12 10 8 6 4 2 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Argentina Bolivia Brazil Mexico e. Water US$ at 1995 constant prices, 3-year moving average 10 9 8 7 6 5 4 3 2 1 0 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 Argentina Bolivia Brazil Chile Colombia Mexico did not decline but instead showed a rising trend in the majority of countries. In summary, these reform time graphs of private investment, as well as similar graphs constructed for public investment (not shown here to save space), do not seem to provide strong support for the popular perception that the reform and liberalization of infrastructure sectors led to a surge in private investment to replace declining public invest- ment. The above graphs suggest that such a perception might be roughly correct in the case of the telecommunications sector, but in the other sectors the picture is more mixed. This conclusion from the reform time graphs shown above is cor- roborated by the correlations shown in table 2.4. The overall correla- tion between public and private investment during the past two decades across the nine countries under study is statistically insignificant. Only two of the correlations by sector are significantly different from LATIN AMERICA'S INFRASTRUCTURE 57 Table 2.4 Correlation between Public and Private Infrastructure Investment, by Sector Total infrastructure investment 0.027 Power 0.010 Telecommunications 0.270** Roads and railways 0.383** Water 0.112 Note: Ratios to GDP, nine countries, 1980­98. ** Significant at 5 percent. Source: Authors' calculations. zero. They correspond to the telecom sector, in which public and pri- vate investment are negatively correlated, and the transport sector, where the correlation is strongly positive. A more formal test of the hypothesis that private infrastructure spending replaced public spending can be performed by noting that, if such a view were correct, one should see more of a reduction in public infrastructure spending in the countries and sectors where private in- frastructure spending increased the most. To verify this, public infra- structure investment is regressed on private infrastructure investment, with both expressed as percentage of GDP. Table 2.5 reports the results from three sets of regressions. The first two use aggregate private and public infrastructure investment and allow for country-specific constants and time trends, using a seemingly unrelated regressions (SUR) setup. The first specification reported in the table imposes a common coefficient on private investment for all nine countries. The result is surprising. The coefficient estimate equals 0.10 and is strongly significant, implying a positive statistical associa- tion between private infrastructure spending and public infrastructure spending, which suggests that the two are complements rather than substitutes. The country-specific trends (not reported to save space) are negative and significant in seven of the nine countries, with Colombia and Ecuador as the only exceptions. In other words, the data assign the reduction in public infrastructure spending to the preexisting trend rather than to the increase in private infrastructure spending. The second experiment in table 2.5 reports results from a less restric- tive empirical specification that permits the private investment/GDP ratio to carry different coefficients in each country. The estimated coefficients vary in sign and magnitude. Four are positive and five negative; their average equals 0.031. Four of the nine estimates are significant at the 5 percent level; three of these (Bolivia, Colombia, and Peru) are positive whereas the fourth (Brazil) is negative. These results reveal a considerable 58 THE LIMITS OF STABILIZATION Table 2.5 Regressions of Public Infrastructure Investment/GDP on Private Infrastructure Investment/GDP 1. Using aggregate investment by country (FE­SUR estimates with country-specific constants and trends) 1a. Pooled estimate 0.108** Number of significantly positive trends 1 Number of significantly negative trends 7 Adjusted R2 0.838 1b. Country-specific estimates (9 total): average 0.031 Number of significantly positive estimates 3 Number of significantly negative estimates 1 Number of significantly positive trends 0 Number of significantly negative trends 7 Adjusted R2 0.852 2. Using investment by country sector (FE estimates with country-sector-specific constants and trends) Country-sector-specific estimates (32 total): average 0.162 Number of significantly positive estimates 8 Number of significantly negative estimates 8 Number of significantly positive trends 2 Number of significantly negative trends 16 Adjusted R2 0.859 Note: FE­SUR fixed effects­seemingly unrelated regressions; nine countries, 1980­98. ** Significant at 5 percent. Source: Authors' calculations. degree of cross-country diversity in the relationship between private and public infrastructure investment. The pooling restrictions implicit in the earlier empirical specification (which assumed equal coefficients across countries) are clearly invalid: a Wald test of equality of coefficients across countries yields a p-value of less than 0.0001, unambiguously rejecting the null of equal coefficients. As for the country-specific time trends, most (seven out of nine) are significantly negative; the exceptions now are Chile and Ecuador, whose time trends are insignificant. In the last experiment reported in table 2.5, the regression of public on private investment is repeated but different regression coefficients for each country and sector are allowed, along with country- and sector-specific time trends and intercepts using a fixed effects specification. After drop- ping country sectors with missing data, this yields a total of 32 regression estimates of the impact of private on public infrastructure investment. Given the large number of parameters estimated, the table presents only a summary of the results. Again the coefficient estimates show a LATIN AMERICA'S INFRASTRUCTURE 59 wide dispersion. Seventeen are positive and 15 negative, and their mean equals 0.16. Of the 16 estimates significantly different from zero at the 5 percent level, 8 are positive and 8 negative. In spite of this diversity, the sectoral distribution of the estimates (whose individual values are not shown in the table) is suggestive. The 8 significantly negative esti- mates correspond to the power (4 estimates) and telecom and water (2 each) sectors in various countries. It is interesting that none of the transport sector offset coefficients is significantly negative. In turn, the 8 positive coefficients are found in transport (4), power (2), telecom (1), and water (1). This pattern of coefficient signs would suggest some re- placement of public by private investment in power, whereas in trans- port the relationship between public and private spending appears to be one of complementarity. For the country-sector-specific time trends, the vast majority (25 out of 32) remain negative. Among those statisti- cally significant, 16 are negative and only 2 are positive. The conclusion that can be drawn is that the observed decline in public infrastructure spending was not closely matched with those sectors and countries where private infrastructure spending surged. There is a great deal of diversity across countries and infrastructure sectors. In some individual cases private infrastructure spending in- creases did offset public infrastructure spending cuts. But in a large number of instances, the sectors where private spending increased the most were not those where public spending declined the most--or even where it declined at all. This suggests that private sector involvement did not lead to a generalized replacement of public spending with pri- vate spending. In principle, some degree of decline in aggregate spend- ing might be consistent with unchanged asset accumulation if the unit cost of assets is declining over time (as one would expect to be the case). However, the pervasiveness and magnitude of the observed spending decline suggests that the opening up to private activity was not a panacea for Latin America's infrastructure woes. Infrastructure Spending and the Quantity and Quality of Infrastructure The final step of the analysis concerns the link between infrastructure investment trends and the evolution of standard indicators of infra- structure stocks and their quality. The first task is to assess to what extent spending on infrastructure gets translated into actual quantity increases of infrastructure. It could be that public spending is misclassi- fied or is simply ineffective in creating new infrastructure. Pritchett (2000) reported many horror stories of public investment not translat- ing into effective increases in capital. 60 THE LIMITS OF STABILIZATION This section examines the effect of total infrastructure spending, public and private, in the respective sectors on the growth of the corre- sponding infrastructure stocks for the nine Latin American countries where data exist. This is done by estimating regressions with the growth in physical infrastructure stocks as the dependent variable and infra- structure investment (as a ratio to GDP) as the explanatory variable.9 Separate panel estimators for each of the infrastructure sectors under analysis are computed. In each case a dynamic specification is used to model the relationship between growth in physical infrastructure stocks and infrastructure investment, to capture lags in the capital accumula- tion process as well as inertia in investment decisions. Specifically, lags of both the dependent and independent variables are included in an autoregressive­distributive lag (ARDL) framework. The lag order of the ARDL is dictated by a compromise between the need to allow for time- to-build in the accumulation of stocks and the length of the available time series. For telecommunications, four lags proved sufficient. For power and transport (roads and railways), up to six lags were used. Al- though this specification might be insufficient given the long delays of- ten involved in the construction of power plants and railway routes, the short data samples available prevented use of longer lag specifications. Table 2.6 summarizes the empirical results of this procedure. The table reports a variety of empirical specifications with and without country and/or time effects, which respectively intend to capture country-specific and common factors affecting infrastructure accumu- lation. For transport routes, rather than fixed effects each country's to- tal land area (in logs) was used as an additional explanatory variable.10 In view of the generous parameterization of the estimated equations, to save space the table reports only the long-run impact of investment on the rate of accumulation of the asset in question. The first entry in table 2.6 reports pooled ordinary least-squares (OLS) estimates. For all three assets, the estimated long-run impact of total infrastructure investment on asset accumulation is positive and significant at the 5 percent level (8 percent in the case of transport). The long-run coefficient estimate reflects the percentage increase in the rate of asset accumulation associated with a permanent increase in in- vestment by 1 percent of GDP. Thus, for example, the coefficient at the top left corner in the table indicates that the rate of growth of phone lines per worker increases by 6.9 percent when telecom investment in- creases permanently by 1 percent of GDP. The explanatory power of the estimated equations varies across assets. It is highest for telecom, where the simple ARDL specification chosen accounts for more than three-fourths of the observed variation in asset growth rates, and lowest for power, where only 11 percent of the varia- tion is captured by the estimates. This echoes the concerns stated above LATIN AMERICA'S INFRASTRUCTURE 61 Table 2.6 Relationship between Physical Stocks and Investment Spending in Infrastructure Transport Telecom Transport total roads Estimation main lines Power total roads railways method (4,4)a (6,6)a (6,6)a (6,6)a I. Pooled OLS Total investment 6.89 1.97 4.43 4.07 (p-value) (0.00) (0.04) (0.08) (0.05) R2 0.77 0.11 0.38 0.36 II. Fixed effects Total investment 8.72 3.42 4.63 4.65 (p-value) (0.00) (0.03) (0.08) (0.01) ln area 0.04 0.05 (p-value) (0.05) (0.02) Fixed effects (0.03) (0.07) (p-value) R2 0.78 0.24 0.38 0.49 III. Fixed effects and time effects Total investment 7.99 6.38 5.22 6.00 (p-value) (0.00) (0.06) (0.07) (0.01) ln area 0.04 0.06 (p-value) (0.04) (0.00) Fixed effects (0.04) (0.08) (p-value) Time effects (0.00) (0.02) (0.01) (0.01) (p-value) R2 0.82 0.32 0.51 0.53 Note: Dependent variable is growth rate in physical infrastructure. The table re- ports the long-run elasticity of asset accumulation with respect to investment spending (as a ratio to GDP) derived from the ARDL estimates. For roads and transport routes, we use (log) land area rather than country fixed effects in specifications II and III. The sample includes annual data for 1970­98 on nine Latin American countries: Ar- gentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Peru, and República Boli- variana de Venezuela. OLS ordinary least-squares. a. Lag structure for ARDL model. Source: Authors' calculations. that asset accumulation may reflect investment performance with long and variable lags, perhaps longer than allowed for in our empirical spec- ifications because of the scarce number of observations available. Fur- thermore, for power the lag structure may vary considerably across coun- tries, depending on the kind of power generation added to the system, because different types of power plants involve different construction lags. 62 THE LIMITS OF STABILIZATION The next entry in table 2.6 adds country fixed effects to the telecom and power regressions and land area for roads. The long-run estimates do not change much for telecom and roads, with some increase in the ex- planatory power of their empirical equations. For power, however, the es- timated long-run effect of investment increases quite substantially, along with the R2. The land area variable as well as the country fixed effects are significant (although only at the 10 percent level in the case of power). Finally, the last entry in table 2.6 adds time dummies in the empirical specification, to control for omitted common factors driving asset accu- mulation across countries. This specification is also robust to the pres- ence of (common) trends in asset unit costs--for example, a declining cost per megawatt (MW) of power generation capacity. The set of time dummies is highly significant in all three equations. The long-run coef- ficient estimates for telecom and roads show relatively modest changes, although the fit of the respective equations improves noticeably, espe- cially in the case of roads whose R2 now exceeds 50 percent. For power, the estimated long-run effect becomes much bigger, and the fit of the equation improves substantially, with the R2 now exceeding 30 percent. In summary, the conclusion from these empirical experiments is that infrastructure investment is a robust predictor of subsequent changes in the physical infrastructure stock across countries and over time. The evidence is particularly strong in the cases of telecommuni- cations and transport routes. The simplicity of the empirical specifica- tions employed and the relatively short time span of the available data suggest that the link between investment and infrastructure accumula- tion is probably much stronger in reality than the above experiments reveal. This suggests that reductions in public, and hence total, infra- structure spending have negatively affected the quantity of infrastruc- ture available in Latin America over the past two decades. Although no systematic evidence was found that private investment had replaced declining public investment in infrastructure, it is never- theless possible that private spending might have translated into faster stock accumulation than public spending. The former might have shown greater efficiency than the latter by acquiring the same infrastructure stocks at a lower cost. In this scenario, the contribution of private in- vestment to stock accumulation should be greater than that of public investment. This possibility is investigated in table 2.7, which reports experi- ments similar to those performed in the preceding table but disaggre- gating total investment between its public and private components. If the latter is more efficient than the former, private spending should carry a significantly larger coefficient than public spending in the in- frastructure stock accumulation regressions. LATIN AMERICA'S INFRASTRUCTURE 63 Table 2.7 Relationship between Physical Stocks, Public and Private Investment Spending in Infrastructure Transport Telecom Transport total roads Estimation main lines Power total roads railways method (4,4,4)a (6,6,6) (6,6,6) (6,6,6)a I. Pooled OLS Private investment 3.82 0.61 14.61 14.57 (p-value) (0.00) (0.57) (0.04) (0.02) Public investment 1.07 0.93 0.24 0.85 (p-value) (0.15) (0.04) (0.06) (0.03) Equality tests (0.66) (0.83) (0.02) (0.15) (p-value) R2 0.78 0.12 0.44 0.44 II. Fixed effects Private investment 5.93 1.58 14.29 15.92 (p-value) (0.00) (0.07) (0.04) (0.06) Public investment 4.08 1.23 0.07 0.10 (p-value) (0.12) (0.08) (0.06) (0.03) Area (in logs) 0.01 0.01 (p-value) (0.06) (0.04) Fixed effects (p-value) (0.07) (0.08) Equality tests (p-value) (0.92) (0.88) (0.16) (0.07) R2 0.80 0.17 0.44 0.50 IV. Fixed effects and time effects Private investment 6.00 2.28 12.70 15.47 (p-value) (0.00) (0.07) (0.11) (0.01) Public investment 4.47 1.71 0.08 0.03 (p-value) (0.02) (0.09) (0.00) (0.04) Area (in logs) 0.01 0.01 (p-value) (0.03) (0.04) Fixed effects (p-value) (0.03) (0.19) Time effects (p-value) (0.01) (0.02) (0.05) (0.08) Equality tests (p-value) (0.95) (0.37) (0.03) (0.03) R2 0.83 0.40 0.57 0.60 Note: Dependent variable is growth rate in physical infrastructure. The table re- ports the long-run elasticity of asset accumulation with respect to investment spending (as a ratio to GDP) derived from the ARDL estimates. In the case of roads and trans- port routes, we use (log) land area rather than country fixed effects in specifications II and III. The sample includes annual data for 1970­98 on nine Latin American coun- tries: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Peru, and República Bolivariana de Venezuela. a. Lag structure for ARDL model. Source: Authors' calculations. 64 THE LIMITS OF STABILIZATION The empirical results in the table provide only limited evidence in favor of this hypothesis. For transport routes, the coefficient of private investment is consistently much larger than that of public investment, although Wald tests show that the difference between the two is sig- nificant only when both time effects and land area are simultaneously included in the regression. In contrast, in telecommunications and power the coefficient of private investment is in most cases somewhat larger than that of public investment, but the tests of equality yield no strong evidence against the null hypothesis that the "bang-per-buck" of private and public investment is the same, regardless of whether coun- try and time effects are included in the equation. However, it is impor- tant to stress that these results should be taken with caution, because the projects falling under private initiative could be systematically dif- ferent from those undertaken by the public sector. In summary, only limited evidence can be found that private invest- ment was more effective than public investment in expanding infrastructure asset stocks. But what about the quality of stocks? Did enhanced private participation lead to an improvement in the quality of infrastructure stocks? On the one hand, in those countries that privatized Table 2.8 Private Participation and Infrastructure Quality Coefficient on private Dependent sector Sample Total variable Method share t-statistic period obs. R2 Telephones Faults per 100 main lines SUR 52.89 3.2 1982­98 65 0.63 FE­SUR 45.90 2.7 1982­98 65 0.78 Percentage of unsuccessful calls SUR 18.75 1.7 1990­98 26 0.63 FE­SUR 8.34 0.8 1990­98 26 0.72 Years on waiting list for main lines SUR 0.16 5.2 1970­98 150 0.72 FE­SUR 0.20 7.1 1970­98 150 0.84 Electricity Power losses (percent of output) SUR 2.64 3.2 1971­98 204 0.94 FE­SUR 3.95 6.9 1971­98 204 0.98 Note: Regression of quality indicators on private sector investment share in nine selected Latin American countries. SUR seemingly unrelated regressions. FE­SUR fixed effects­ seemingly unrelated regressions. Source: Authors' calculations. LATIN AMERICA'S INFRASTRUCTURE 65 the infrastructure sector, some mixed evidence exists of quality improve- ments. Table 2.8 shows that all the telephone quality service indicators (telephone faults per line, percentage of unsuccessful local calls, and years spent on waiting list for phone service) get significantly better, the higher the share of the private sector in telecommunications spending. This re- sult holds regardless of whether one controls for country fixed effects. On the other hand, there is a perverse result in the electricity sector because power losses increase with increased private share of power spending. However, this again could reflect reverse causality because governments may have wanted to privatize inefficient enterprises in the power sector that were running high power losses. Also, it could reflect heterogeneity among public and private power projects, making their respective power losses not strictly comparable with each other. Figures 2.20a through 2.20d explore the same issue in a different way. They present scatter plots relating infrastructure quality indicators to the share of the private sector in total infrastructure investment, us- ing 10-year averages instead of the annual data underlying the regres- sions in table 2.8. This should make it easier to detect the changes in quality if these occur only gradually over time, as the new private sec- tor projects reach completion and become numerous enough to affect overall infrastructure quality in a significant way. Each point in the graphs represents one country-decade observation. Figure 2.20 Infrastructure Quality and the Private Share of Investment in Infrastructure a. Unsuccessful Local Calls versus Private Investment Unsuccessful local calls (percent) 60 y = -23.917x + 37.471 50 R2 = 0.1663 40 30 20 10 0 0 0.2 0.4 0.6 0.8 1.0 1.2 Private share of investment in telecoms 66 THE LIMITS OF STABILIZATION Figure 2.20 (continued) b. Telephone Faults versus Private Investment Telephone faults per main line 140 y = -75.865x + 100.13 120 R2 = 0.4065 100 80 60 40 20 0 0 0.2 0.4 0.6 0.8 1.0 1.2 Private share of investment in electricity c. Waiting Years per Main Line versus Private Investment Waiting years per main line 0.60 y = -0.1505x + 0.2799 R2 = 0.0869 0.50 0.40 0.30 0.20 0.10 0 0 0.2 0.4 0.6 0.8 1.0 1.2 Private share of investment in telecoms LATIN AMERICA'S INFRASTRUCTURE 67 Figure 2.20 (continued) d. Electricity Losses versus Private Investment Electricity production losses 25 20 15 10 5 y = -3.2834x + 18.417 R2 = 0.0663 0 0 0.2 0.4 0.6 0.8 1.0 1.2 Private share of investment in electricity The verdict from the figures is similar to that emerging from the re- gressions: the association is clear between improving quality of telecommunications service and private participation. In the case of the power sector, there is also some hint at declining power losses, although the evidence appears much weaker than for telecommunications. Conclusion The 1980s and 1990s saw a widening of the infrastructure gap be- tween Latin America and other successful developing economies like those in East Asia. A comparative review of a comprehensive set of in- frastructure quantity and quality indicators reveals that during that pe- riod Latin America fell behind along most dimensions analyzed. Latin American public infrastructure spending declined as a percent- age of GDP during the era of macroeconomic crises in the 1980s and 1990s. Part of this decline is associated with fiscal adjustment (reductions in budget deficits), but the magnitude of this association is small and the trend in infrastructure spending is still down even after controlling for budget balances. This suggests that some portion of the reduction in pub- lic expenditure was not driven by deficit reduction. Furthermore, there is 68 THE LIMITS OF STABILIZATION only limited evidence to support the common perception that privatiza- tion (specifically, private sector entry into infrastructure industries) ex- plains the observed downward trend in public infrastructure spending. Although this seems to be true in a few cases, there are at least as many (or even more) instances in which higher private infrastructure spending is associated with more public infrastructure spending. Private infrastructure spending did increase after the infrastructure sectors were opened up to private participation, but did so unevenly. Opening up to the private sector was most successful in telecommuni- cations and electricity, with water, roads, and railways showing more uneven results; there were some laggards even in telecommunications and power. The levels of private infrastructure spending were gener- ally below the prereform public infrastructure spending in each sector. Moreover, there was no universal tendency for public infrastructure spending to fall after liberalization. Infrastructure spending is a good predictor of subsequent growth in infrastructure stocks; it is a particularly robust predictor for telephone lines and transport routes but also for power generation capacity. If the quantity of infrastructure has an effect on output levels, as a grow- ing literature has argued (Canning 1999, and Röller and Waverman 2001) and the next chapter will assess, then fiscal retrenchment imple- mented through cuts in public infrastructure spending represents a my- opic and potentially self-defeating adjustment strategy, because it low- ers future output and thus the tax collection and debt-servicing capacity of the economy (Easterly 2001).11 The evidence also suggests that under this kind of fiscal austerity, Latin America's infrastructure lag behind East Asia is unlikely to get better soon. There is no clear evidence that private sector participation has raised the efficiency of infrastructure investment--as reflected by the transla- tion of spending into asset accumulation. There is, however, some evi- dence that the increased private sector share of infrastructure has had some positive effect on infrastructure quality. All of the telephone serv- ice quality indicators improve with an increased private sector share in telecommunications, although the evidence regarding the relationship between power sector efficiency and private participation is less strong. The conclusions from this chapter have to be taken with some cau- tion because little information is available on measures of infrastruc- ture quality and investment efficiency comparable across countries. Thus the conclusions have had to rely primarily on indicators of asset stocks and spending volumes. Nevertheless, the picture that emerges from our comprehensive review of the available data indicates strongly that Latin America's infrastructure sector performed poorly during the era of macroeconomic crises of the 1980s and 1990s. Privatization has so far been no panacea and a huge gap has opened relative to East LATIN AMERICA'S INFRASTRUCTURE 69 Asian NICs. As the next chapter will show, for Latin America to recover its long-run growth potential, increased attention to infrastructure pol- icy is well warranted. Appendix 2A. Infrastructure Database The data underlying the analysis in this chapter come from the infra- structure database assembled for this work. The database includes both physical indicators of quantity and quality of infrastructure endowments and measures of public and private infrastructure investment expendi- tures. Here we give a brief description of both components of the dataset. Physical Infrastructure Data Public Utilities Telephones and Telephone Main Lines. Following Canning (1998), we use the number of telephone sets and the number of main lines con- nected to local telephone exchanges as our measure of the provision of telephone services. Although both measures are highly correlated, Canning suggested that the number of telephone main lines is a better measure of the capacity of the telephone system. The variables taken from Canning's database are displayed in table 2A.1. We extend Canning's data with more recent figures taken from the Annual Reports of the International Telecommunications Union (ITU). Furthermore, ITU provides other indicators that could be used to meas- ure the quantity and quality of telephone services. A summary of the coverage and availability of those indicators is presented in table 2A.2. Regarding coverage across regions, we can summarize the time- series dimensions for some regions as presented in table 2A.3. From the table, it is clear that the quality indicators--faults per 100 main lines and waiting list for main lines--have more limited coverage, especially the former. Table 2A.1 Telephone Service Variables Variable Period Source Number of telephone Annual, 1960­95 ITU, AT&T, United lines Nations discontinued after 1995 Number of telephone Annual, 1960­98 ITU, AT&T, main lines United Nations Source: Canning 1998. 70 THE LIMITS OF STABILIZATION Table 2A.2 Summary of Coverage and Availability of Telephone Service Indicators Cross-section and Variables Frequency/period time dimensions I. System capacity Connection capacity of Annual, 1975­98; 1975­98: between local exchanges selected years: 1960, 55 and 175 countries; 1965, 1970 1960­75: 10 countries; time series (TS): mean of 12 observations per country (opc) and median of 11 opc II. Operation and access Main telephone lines Annual, 1975­97; 1975­97: between 158 in operation selected years: 1960, and 209 countries; 1965, 1970 1960, 1965, 1970: more than 100 Percentage of main Annual, 1985­97 1985­97: between 80 line equipment for and 109 countries; TS: direct international mean of 5 opc and dialing median of 4 opc Percentage of urban Annual, 1980­97; 1990­97: between 27 and main lines selected years: 1960, 98 countries; 1980­89: 1965, 1970, 1975 no more than 10; 1960­75: only 2 countries (HKG, SGP); TS: mean of 3 opc and median of 1 opc Percentage of Annual, 1975­97; 1975­97: between 54 residential main selected years: 1960, and 172 countries; lines 1965, 1970 1960­75: no more than 6 countries; TS: mean of 10 opc and median of 9 opc Number of local Annual, 1975­97; 1975­97: between 12 and telephone calls selected years: 1960, 52 countries; 1960, 1965, 1970 1965, 1970: more than 7; TS: mean of 3 opc and median of 0 opc Number of national Annual, 1975­97; 1975­97: between 15 long-distance selected years: 1960, and 87 countries; 1960, telephone calls 1965, 1970 1965, 1970: 7; TS: mean of 5 opc and median of 1 opc LATIN AMERICA'S INFRASTRUCTURE 71 Table 2A.2 (continued) Cross-section and Variables Frequency/period time dimensions Percentage of Annual, 1975­97; 1975­97: between 2 and households with a selected years: 1960, 36 countries; 1960, telephone (limited 1965, 1970 1965, 1970: only 1 coverage) (CAN); TS: mean of 1 opc and median of 0 opc III. Costs Residential monthly Annual, 1980­97 1990­97: between 119 telephone and 176 countries; subscription (US$) 1980­89: 23 countries; TS: mean of 6 and median of 7 opc Residential telephone Annual, 1980­97 1990­97: between 120 connection charge and 177 countries; (US$) 1980­89: 26 countries; TS: mean of 6 and median of 7 opc IV. Quality Percentage of Annual, 1980­97 1990­97: between 45 unsuccessful local and 98 countries; calls 1980­89: 6 countries; TS: mean and median of 2 opc, and a maximum of 14 (GBR) Telephone faults per Annual, 1980­97 1990­97: between 64 100 main lines and 127 countries; 1980­89: 22 countries; TS: mean and median of 4 opc Waiting list for main Annual, 1975­97; 1990­97: between 76 lines selected years: 1960, and 175 countries; 1965, 1970 1960­75: 55 countries; TS: mean of 14 and median of 13 opc Note: CAN Canada; GBR Great Britain; HKG Hong Kong (China); SGP Singapore. Source: ITU, World Telecommunications Development Report, various years. 72 THE LIMITS OF STABILIZATION Table 2A.3 Telecommunications Indicators: Time-Series Coverage by Region Faults Waiting Main Main Number per 100 list for Region/ lines lines in of Connection main main statistics (ML) operation telephones capacity lines lines I. Latin America and the Caribbean (42 countries) Average 27 23 23 12 3 12 Median 28 24 33 12 3 13 Min/max 0/38 14/26 0/36 0/23 0/14 0/24 II. East Asia and the Pacific (35 countries) Average 25 20 17 11 4 13 Median 26 24 20 9 4 13 Min/max 0/38 0/26 0/36 0/25 0/9 0/26 III. Western Europe (25 countries) Average 28 23 20 13 5 17 Median 38 26 25 16 5 20 Min/max 0/38 0/26 0/35 0/23 0/17 0/26 Source: ITU, World Telecommunications Development Report, various years. Energy. The measure of infrastructure in electricity, as taken from Canning (1998), is the electricity generating capacity (in kilowatts). We have annual observations for the 1950­95 period. The main sources for these data were the United Nations' Energy Statistics and Statistical Yearbook. In table 2A.4 we report other variables that could be used as proxies for energy. We extend these data using mainly the United Nations' Energy Statistics and Statistical Yearbook. In table 2A.5 we present some ba- sic information on the time-series coverage of the indicators for infra- structure in the energy sector. Sanitation and Sewerage. For this category we have found observa- tions only for selected years within the 1970­97 period. The main source is the World Bank's World Development Indicators. The vari- ables are the percentages of population with access to safe water and sanitation in urban and rural areas. We also report total access. The limited coverage of the series could be observed in table 2A.6, which summarizes the time dimension in some selected regions. Public Works Roads. Canning (1998) presented two indicators for the stock of in- frastructure in roads (table 2A.7). LATIN AMERICA'S INFRASTRUCTURE 73 Table 2A.4 Variables Used as Proxies for Energy Frequency/ Cross-section and time Variables period dimensions I. Output and consumption Electric power Annual, 1960­97 1971­97: 130 countries; consumption (in kwh 1960­70: 27 countries; or kwh per capita) Time series (TS): mean of 17 opc, median of 26 opc, with a maximum of 37 observations for 27 countries Electric production Annual, 1960­97 1971­97: 109­129 countries; 1960­70: 24 countries; TS: mean of 16 opc, median of 26 opc, with a maximum of 37 observations for 24 countries II. Quality Electric power Annual, 1960­97 1971­97: 100­129 countries; transmission and 1960­70: 24 countries; TS: distribution losses mean of 16 opc, median of (percent of output) 18 opc, with a maximum of 37 observations for 24 countries Source: World Bank, World Development Indicators, various years; United Nations, Statistical Yearbook CD-ROM, various issues; United Nations, The Energy Statistics Yearbook, various years. We extend these data using recent issues of the International Road Federation World Road Statistics (see table 2A.8). According to Canning (1998), the raw data on road length seem too unreliable to be useful; even using national sources it appears impossible to construct data that are consistent either across countries or over time. Canning used the percentage of the main paved and unpaved road network as a measure of quality. Other available indicators are limited in coverage and capture only the transportation impact of these roads. Irrigation. The main source for measures in this category is the World Bank's World Development Indicators (table 2A.9). The time-series dimension for these indicators in some important regions are summarized in table 2A.10. Other Transport Sectors: Railways. The only measure provided by Canning (1998) is the rail route length. The main data sources for the 74 THE LIMITS OF STABILIZATION Table 2A.5 Energy Indicators: Time-Series Coverage by Region Electricity Electricity power power Electricity Electricity transmission Electricity consumption power power and distribution Region/ generation (kwh per consumption production losses statistics capacity capita) (kwh) (kwh) (% output) I. Latin America and the Caribbean (42 countries) Average 27 15 15 14 14 Median 36 26 26 26 22 Min/max 0/36 14/27 0/37 0/26 0/26 II. East Asia and the Pacific (35 countries) Average 20 13 13 13 13 Median 24 0 0 0 0 Min/max 0/36 0/37 0/37 0/37 0/37 III. Western Europe (25 countries) Average 24 25 25 25 25 Median 36 37 37 37 37 Min/max 0/36 0/37 0/37 0/37 0/37 Source: United Nations, Energy Statistics and Statistical Yearbook, various years. Table 2A.6 Sanitation and Sewerage Indicators: Time-Series Coverage by Region Safe water (percentage of Sanitation (percentage of Region/ population with access) population with access) statistics Total Rural Urban Total Rural Urban I. Latin America and the Caribbean (42 countries) Average 3 3 3 2 2 2 Median 4 3 3 2 1 2 Min/max 0/7 0/7 0/7 0/5 0/5 0/5 II. East Asia and the Pacific (35 countries) Average 3 3 3 2 2 2 Median 3 3 3 2 1 2 Min/max 0/7 0/8 0/8 0/5 0/6 0/6 III. Western Europe (25 countries) Average 3 3 3 2 2 2 Median 3 2 2 2 1 1 Min/max 0/7 0/7 0/7 0/5 0/5 0/5 Source: World Bank, World Development Indicators, various years. LATIN AMERICA'S INFRASTRUCTURE 75 Table 2A.7 Indicators for Roads Variables Frequency/period Problems Total road length Annual, 1950­97 Frequent gaps and large (in km) changes. Differences in the definition of roads across countries and over time. Paved road length Annual, 1950­97 Large variations in quality. Data (in km) do not reflect the width of the road and do not account for the age of the road. Source: Canning 1998; International Road Federation, World Road Statistics, various issues; United Nations, Statistical Yearbook, various issues. Table 2A.8 Other Indicators for Roads Cross-section and Variables Frequency/period time dimensions Road traffic Annual, 1990­97 1990­97: between 13 and 60 (vehicles per km) countries; Time series (TS): mean of 2 opc, median of 0 opc, with a maximum of 9 observations for 8 countries Roads, goods Annual, 1990­97 1990­97: between 23 and 57 transported countries; TS: mean of 2 opc, (million of tons median of 0 opc, with a per km) maximum of 10 observations for 4 countries Source: World Bank, World Development Indicators, various years; International Road Federation, World Road Statistics, various years. Table 2A.9 Indicators for Irrigation Cross-section and Variables Frequency/period time dimensions Irrigated land Annual, 1960­97 1960­96: between 143 and 164 (hectares) countries; Time series (TS): mean of 25 opc, median of 36 opc, with a maximum of 36 observations for 134 countries Irrigated land Annual, 1960­97 1960­96: between 136 and 156 (percentage of countries; TS: mean of 23 crop land) opc, median of 36 opc, with a maximum of 36 observations for 122 countries Source: World Bank, World Development Indicators, various years. 76 THE LIMITS OF STABILIZATION Table 2A.10 Irrigation: Time-Series Coverage by Region Irrigation Region and statistics As percentage of crop land In hectares I. Latin America and the Caribbean (42 countries) Average 27 27 Median 36 36 Min/max 0/36 0/36 II. East Asia and the Pacific (35 countries) Average 18 18 Median 25 27 Min/max 0/36 0/36 III. Western Europe (25 countries) Average 16 20 Median 0 36 Min/max 0/36 0/36 Source: World Bank, World Development Indicators, various years. length of railway lines are Mitchell's International Historical Statistics (1992, 1993, 1995) until 1980 and The World Bank's Railways Data- base (available online) thereafter. Canning (1998) also used national sources to supplement these data.12 The World Bank has developed the Railways Database that com- prises data for 1980­97. From this database we have some variables that could be useful to measure capacity and quality of the railways: stock of main diesel locomotives; stock of main electric locomotives; passenger­kilometer (in millions); goods transported, freight ton­kilo- meter (in millions); goods transported, freight ton­kilometer per wagon (000); diesel locomotive availability (in percent); operating ra- tio with normalization; and operating ratio without normalization. Finally, we summarize the time-series information across countries for some selected regions (table 2A.11). Data on Investment in Infrastructure The sample covers nine Latin American countries at an annual fre- quency over the period 1970­98. Definition of Public Sector Table 2A.12 presents the definition of public sector used in the figures of public investment in infrastructure. LATIN AMERICA'S INFRASTRUCTURE 77 Table 2A.11 Roads and Railways: Time-Series Data for Selected Regions Roads Railways Route length Paved route length Route length Region/statistics (km) (km) (km) I. Latin America and the Caribbean (42 countries) Average 12 15 26 Median 9 12 34 Min/max 0/34 0/34 0/38 II. East Asia and the Pacific (35 countries) Average 14 13 20 Median 9 0 32 Min/max 0/36 0/36 0/38 III. Western Europe (25 countries) Average 21 17 24 Median 30 22 36 Min/max 0/36 0/36 0/38 Source: World Bank, Railways Database, various years. Definition of Transport Sector The definition of the transport sector varies somewhat across coun- tries, as shown in table 2A.13. Table 2A.12 Public Sector Definitions Used in the Figures of Public Investment in Infrastructure Country Telecom Power Transport Water Argentina GG SOE GG SOE GG SOE GG SOE Bolivia GG SOE GG SOE GG SOE GG SOE Brazil GG SOE GG SOE GG SOE -- Chile GG SOE GG SOE GG SOE -- Colombia GG SOE GG SOE GG SOE GG SOE Ecuador GG SOE GG SOE GG SOE GG SOE Mexico GG SOE GG SOE GG SOE GG SOE Peru GG SOE GG SOE GG SOE GG SOE Venezuela, GG SOE GG SOE GG SOE GG SOE R.B. de -- Not available. Note: GG denotes general government spending on infrastructure. SOE denotes state-owned enterprise spending on infrastructure. Source: National sources listed below. 78 THE LIMITS OF STABILIZATION Table 2A.13 Definition of the Transport Sector Country Transport Sector Argentina We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Bolivia We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Brazil We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Chile We have aggregate data only for investment in transport. This includes all categories (roads, railways, ports, and airports). There is no breakdown for any of these four categories. Colombia We have aggregate data only for investment in transport. This includes all categories (roads, railways, ports, and airports). There is no breakdown for any of these four categories. Ecuador We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Mexico We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Peru We have investment data for both roads and railways (separately). We do not have data on investment in ports and airports. Venezuela, We have investment data for both roads and railways R.B. de (separately). We do not have data on investment in ports and airports. Note: Aggregate data on investment in the transport sector includes spending on roads, railways, ports, and airports. However, we do not have the specific investment in each subsector. We lack data on railways for Chile and Colombia. Source: National sources listed below. National Sources of Information for the data on Infrastructure Investment To obtain data on infrastructure investment, we gathered information mostly from national sources. Here is the list of documents used, by country. LATIN AMERICA'S INFRASTRUCTURE 79 Argentina, 1970­98 Infrastructure--General Information: Fundación de Investigaciones Económicas Latinoamericanas. 1992. "Capital de Infraestructura en la Argentina: Gestión Pública, Privati- zación y Productividad." Buenos Aires. Secretaria de Hacienda. "Cuenta de Inversión 1994­97." Sub- Secretaria del Presupuesto. Ministerio de Economía, Buenos Aires. Telecommunications: Celani, Marcelo. 1998. "Determinantes de la Inversión en Telecomu- nicaciones en Argentina." CEPAL Serie Reformas Económicas No. 9. Santiago de Chile. Power: Adrián Romero, Carlos. 1998. "Regulación e Inversiones en el Sector Eléctrico Argentino." CEPAL Serie Reformas Económicas No. 5. San- tiago de Chile. Transport: Delgado, Ricardo. 1998. "Inversiones en Infraestructura Vial: La Expe- riencia Argentina." CEPAL Serie Reformas Económicas No. 6. Santiago de Chile. Bolivia, 1980­98 Infrastructure--General Information: Antelo, Eduardo. 2000. "Politicas de Estabilización y de Reformas Estructurales en Bolivia a partir de 1985." CEPAL Serie Reformas Económicas No. 62. Santiago de Chile. Barja Daza, Gover. 1999. "Las Reformas Estructurales Bolivianas y su Impacto sobre Inversiones." CEPAL Serie Reformas Económicas No. 42. Santiago de Chile. Instituto Nacional de Estadística. Varios números. "Bolivia en Cifras." World Bank. 1993. "Bolivia: Public Expenditure Review." Wash- ington, D.C. Telecommunications: Barja Daza, Gover. 1999. "Inversión y Productividad en la Industria Boliviana de Telecomunicaciones." CEPAL Serie Reformas Económi- cas No. 16. Santiago de Chile. Power: Barja Daza, Gover. 1999. "Inversion y Productividad en la Industria Boliviana de la Electricidad." CEPAL Serie Reformas Económicas No. 15. Santiago de Chile. 80 THE LIMITS OF STABILIZATION Brazil, 1970­98 Infrastructure--General Information: Cavalcanti Ferreira, Pedro. 1996. "Investimento em Infra-estrutura no Brasil: Fatos Estilizados e Relacoes de Longo Prazo." Pesquisa e Plane- jamento Economico 26 (2) (August). Cavalcanti Ferreira, Pedro, and Thomas Georges Malliagros. 1998. "Impactos Produtivos da Infra-estrutura no Brasil: 1950­95." Pes- quisa e Planejamento Economico 28 (2) (August). ------. 1999. "Investimentos, Fontes de Financiamiento e Evolu- cao do Setor de Infra-estrutura no Brasil: 1950­96." FGV EPGE Ensaios Economicos 346. Coes, Donald V. 1994. "Macroeconomic Crises, Policies and Growth in Brazil, 1964­90." World Bank Comparative Macroeco- nomic Studies. Washington, D.C. Rigolon, Francisco J. Z. 1998. "O Investimento em Infra-estrutura e a retomada do crescimento economico sustentado." Pesquisa e Plane- jamento Economico 28 (1) (April). Chile, 1980­98 Infrastructure--General Information: Ministerio de Obras Públicas, Transportes y Comunicaciones. 2000. "Inversión en Infraestructura: Rol sobre el Crecimiento, Desarrollo Económico y la Globalización." Santiago, Chile: Gobierno de Chile. Moguillansky, Graciela. 1999. "La Inversión en Chile: ¿El Fin de un Ciclo en Expansión?" Santiago, Chile: Fondo de Cultura Económica Chile S.A. Moguillansky, Graciela, and Ricardo Bielschowsky. 2000. "Inver- sión y Reformas Económicas en América Latina." Santiago, Chile: Fondo de Cultura Económica Chile S.A. Colombia, 1973­98 Infrastructure--General Information: Departamento Administrativo Nacional de Estadística (DANE). 1996. "Cuentas Nacionales: Gastos en FBKF por sector institucional según finalidad 1973­95." Bogotá. We should note that DANE data have been computed according to commitments and not on a cash flow basis. Additionally, depreciation of the existing stock has been considered. Transport: Ministry of Transport. 1995. "El Transporte en Cifras, 1970­94." Bogotá. Ministry of Transport. 1997. "El Transporte en Cifras, 1970­96." Bogotá. LATIN AMERICA'S INFRASTRUCTURE 81 Ecuador, 1981­98 Infrastructure--General Information: Banco Central del Ecuador. Varios números. Boletín Anuario. CEPAL/PNUD. 1993. "La Política Fiscal en Ecuador, 1985­91." Serie Política Fiscal No. 35. Santiago de Chile. World Bank. 1991. "Reformas del Sector Público para lograr el crecimiento en una época de decreciente producción petrolera." Wash- ington, D.C. ------. 1993. "Ecuador Public Expenditure Review: Changing the Role of the State." Washington, D.C. Mexico, 1970­98 Infrastructure--General Information: Banco de México. 1995. "La Encuesta de Acervos, Depreciación y Formación de Capital." México, DF. Presidencia del Gobierno. 1999. "IV Informe del Gobierno: México 1988­98." México, DF. ------. 2000. "V Informe del Gobierno: México 1989­99." México, DF. Secretaría de Hacienda de México. Various years. "Inversión Pública Federal por Entidad Federativa." México, DF. Peru, 1970­98 Infrastructure--General Information: Banco Central de Reserva del Perú. Varios números. "Memoria An- nual." Lima. CUANTO S.A. Varios números. "Perú en Números." Lima. Instituto Nacional de Estadística e informática (INEI). Varios números. "Anuario Estadístico." Lima. República Bolivariana de Venezuela, 1980­98 Infrastructure--General Information: Oficina Central de Estadística e Informática. Varios números. "An- uario Estadístico." Caracas. World Bank. 1992. "Venezuela: Decentralization and Fiscal Is- sues." Washington, D.C. (December). Telecommunications: Comisión Nacional de Telecomunicaciones (CONATEL). Website info: www.conatel.gov.ve/indicadores.htm. Appendix 2B. The Liberalization of Infrastructure Industries in Latin America This appendix provides a brief chronology of the opening up of Latin America's infrastructure sectors to private participation (tables 2B.1, 82 THE LIMITS OF STABILIZATION Table 2B.1 Infrastructure Reform Laws (year of enactment, by sector) Energy Transport Country Electricity Gas Roads Railways Telecom Water Argentina 1989 1989 1989 1989 1989, 1998 1989, 1992 Bolivia 1994 n.a. 1998a 1998a 1996 2000 Brazil 1995 n.a. 1993b 1994­95 1995 n.a. Chile 1985 1986 1990 1990 1985, 1994 1988­89 Colombia 1991 n.a. 1991 n.a. 1991 n.a. Ecuador 1994 n.a. n.a. n.a. pending n.a. Mexico n.a. n.a. 1989 1995 1989, 1996 n.a. Peru 1992 n.a. n.a. n.a. 1992, 1998 n.a. Venezuela, pending n.a. n.a. n.a. 1990, 1997 n.a. R.B. de n.a. Not applicable. a. A concession law appeared in 1998, although three concessions had been granted to the private sector after 1996, before the concession law was passed. b. In early April 2000, the government announced a new format for the toll-road concessions to come. Source: Country summaries in the appendix. Table 2B.2 Sale and/or Concession of Public Enterprises in Infrastructure Sectors (starting year) Energy Transport Country Electricity Gas Roads Railways Telecom Water Argentina 1992 1992 1990d 1990d 1990 1993d Bolivia 1995 .. .. 1996d 1996 1997d Brazil 1996 1997a 1996d 1996d 1996 .. Chile 1986 1986b 1993d 1995­97 1986 1993d Colombia 1992 1996b 1994d .. .. .. Ecuador pending .. .. .. .. .. Mexico .. 1995­97c 1989d 1996d 1990 1994d Peru 1994 .. .. .. 1994 .. Venezuela, pending .. .. .. 1991 .. R.B. de .. Negligible. a. Some partial divestitures were carried out. The bulk of gas generation and distribution is in public hands. b. Only one privatization was carried out (Promigas). c. Repsol was partially privatized in two stages, 1995 and 1997. d. O&M with major private capital expenditure (concessions). Source: Country summaries in the appendix. LATIN AMERICA'S INFRASTRUCTURE 83 Table 2B.3 Greenfield Projects in Infrastructure Sectors (starting year) Energy Transport Country Electricity Gas Roads Railways Telecom Water Argentina 1992 1996 .. .. 1990 .. Bolivia 1997 .. .. .. .. .. Brazil 1984 1998 .. 1996 1984 1995 Chile 1990 1995 .. .. 1986 1996 Colombia 1993 1994 .. .. 1994 1997 Ecuador 1996 .. .. .. 1994 .. Mexico 1998 1996 .. .. 1990 1993 Peru 1996 .. .. .. 1990 .. Venezuela, 1992 .. .. .. 1991 .. R.B. de .. Negligible. Source: Country summaries in the appendix. 2B.2, and 2B.3). The discussion focuses on the nine countries under consideration in the main text and draws from national sources.13 Overview For each country and sector examined, we highlight (a) the timing of the sale or the concession of public enterprises to the private sector; (b) the opening up to private greenfield projects, that is, investment primarily related to the acquisition of new assets; and (c) the passage of reform leg- islation, which may precede or follow private sector entry into old or new infrastructure projects. In some cases reform legislation is passed in two waves: the first one aims at allowing private entry, whereas the sec- ond establishes the regulatory framework in the liberalized sector. Drawing from the country summaries that follow, it is possible to construct a comparative timetable for each of these three reform dimen- sions. This is done in tables 2B.1, 2B.2, and 2B.3. On the basis of these tables, we construct table 2.3 in the main text, which for each sector and country takes as the relevant date the earliest one of the three dates in the tables above. Some specific issues should be kept in mind. First, in the telecommu- nications sector, there are typically two stages: privatization and liberal- ization (of access and/or tariffs), when the monopoly status disappears and competition is allowed. Second, in the power sector, privatization 84 THE LIMITS OF STABILIZATION and liberalization typically came together. Third, in the gas sector, whenever private participation is allowed, the private sector's main task is related to pipeline projects. Country Summaries Argentina Argentina started its privatization program in 1989 after the approval of the Ley de Reforma del Estado (No. 23696) under the Menem pres- idency. That law authorized privatization of public enterprises (PE). The comprehensive privatization program was launched jointly with an ambitious program of structural adjustment. During 1990­92, 20 public enterprises were fully or partially privatized. In the electricity sector, privatization of the three public enterprises, Servicios Eléctricos del Gran Buenos Aires (SEGBA), Hidroeléctrica Nor- Patagónica S.A. (HYDRONOR S.A.), and Agua y Energía Eléctrica de Argentina (AYEE) started in 1992. Although greenfield projects have been proposed since 1992, the bulk of them have taken place after 1995. The publicly owned companies in the gas and petrol sectors were also privatized. Gas del Estado was privatized in 1992. Only one greenfield project was proposed in 1996 with a total investment of US$350 million. The Empresa Nacional de Telecomunicaciones (ENTEL) started pri- vatizing in 1990 and was divided into four new private companies (Telecom S.A., Telefonica de Argentina, Telinter, and Startel). Like Mexico and the República Bolivariana de Venezuela, the publicly owned telecommunication companies were sold with a monopoly on basic service for a fixed period of exclusivity, but with requirements to expand and improve basic service. Greenfield projects were also pro- posed after 1990, although most of them took place after 1995 because of the monopoly structure of the sector after privatization. After 1990 the private sector was awarded toll concessions on most transited roads. The concessionaires of toll roads are responsible for maintenance, construction, and reconstruction operations. There are no greenfield projects in the sector. Concessions of railways (freight network, passenger, and commuter urban railroads) also started in 1990. There are no greenfield projects in the sector. Finally, water supply and sewerage services were decentralized to the provinces in 1980, but the central government retained control over serv- ices in the capital city. In 1993 a concession was granted for the opera- tion of water and sanitation services in Buenos Aires, through franchise bidding, to the private sector. Additionally, since 1995, some concessions LATIN AMERICA'S INFRASTRUCTURE 85 (Build-Rehabilitate-Operate-Transfer, BROT) have been awarded to the private sector to operate potable water supply and sewerage services in the provinces of Santa Fe, Cordoba, Corrientes, and Tucuman. There are no greenfield projects in the sector. Legal support for water and sanita- tion reforms comes from the Ley de Reforma del Estado in 1989, and the Decree 9999/92 to define the regulatory framework. Bolivia In the telecommunications sector, the government established a new legal and regulatory framework in 1996. The new law facilitated, among other things, the immediate entry by the private sector into such areas as leased lines, cellular phones, and data transmission. To capitalize the sector, ENTEL became a mixed corporation whose shares were owned by the government and by ENTEL workers. Al- though greenfield projects have existed since 1987, they were negligi- ble in number and investment volume relative to other countries until 1996, when they started to become more important and once the new telecommunication regulation was enacted. The general electricity law, approved in December 1994, mandated vertical deintegration of the sector. Currently, in the electricity sector, the state-owned, vertically integrated Empresa Nacional de Electrici- dad S.A. (ENDE) owns about 62 percent of the installed capacity and supplies around 56 percent of the generation sold at the bulk power level. The investor-owned, vertically integrated Corporación Boliviana de Energía Eléctrica (COBEE) owns some 19 percent of the total installed generating capacity. The privately owned Cooperativa Rural de Electrificacion (CRE) provides distribution services in Sta Cruz. Only one greenfield project was proposed in 1997 with an investment volume of US$97 million. Provision and distribution of gas is in public hands. Only one green- field project was proposed, in 1998, with an investment volume of US$2.2 billion. The sector has not been liberalized. Some railways concessions were provided to the private sector in 1996­97. No greenfield projects have been proposed. A law for con- cessions was enacted in 1998. Roads management, maintenance, construction, and reconstruction are under government control. No greenfield projects have been pro- posed. A concessions law was enacted in 1998. Finally, only a few concessions were granted to the private sector for water supply and sewerage services in 1997. No greenfield projects have been proposed. A new law has been presented recently (2000) to allow private participation in the sector. 86 THE LIMITS OF STABILIZATION Brazil Brazil started its privatization process under the Collor de Melo pres- idency in October 1991. The process began with a reduced number of public enterprises in the tradable goods sector (mining, manufactures). Currently, in the electricity sector, Centrais Electricas Brasileiras S.A. (Electrobras) is a federal utility holding company, with four re- gional integrated generating and transmission utility subsidiaries. The federal government owns, via Electrobras' newly created subsidiary Sistema Nacional de Transmissao de Energia Elétrico (SINTREL), the two high-voltage interconnected transmission systems. State govern- ment and municipalities own most of the distribution utilities. The state of Tocatins started to privatize its distribution utility in 1990; other state-owned (central and noncentral government) utilities were considered for privatization in 1995, with privatization beginning in 1996. Although greenfield projects started to appear in 1984, they were negligible in volume. The bulk of this type of project in the power sector appeared in 1996, together with the privatization process. Only some partial divestitures were carried out in the gas sector in 1997. The bulk of the generation and distribution of gas is in public hands. Only one important greenfield project (BROT type) was pro- posed in the sector in 1998 with a total investment volume of US$2.2 billion. The sector is not liberalized. In the telecommunications sector, full divestitures started in 1996. There have been greenfield projects since 1984, although of negligible size and number. Greenfield projects started to become significant in number and volume after 1997, together with the privatization process. As for roads, a Federal Road Concession Program for toll roads was created in 1993, with a first wave of concessions taking place in 1994­95. The second wave of concessions was prepared in 1996. However, state and municipal governments manage the bulk of the road network. In early April 2000, the government announced a new format for the toll-road concessions to come. There are no greenfield projects in the sector. Initially railways were under full control of the public sector through three public operators: Rede Ferroviaria Federal (RFFSA), Ferrovias Paulistas (FEPASA), and Companhia Vale Rio Doce (CVRD). However, poor performance resulted in pressures to priva- tize the sector. Concessions were granted in the rail sector in 1996­98. The bids (FEPASA, RFFSA) were for the operation and maintenance of each network for 30 years. Only one greenfield project was pro- posed in 1996 with a volume of US$1.26 billion. LATIN AMERICA'S INFRASTRUCTURE 87 Finally, no privatization program was carried out in the water and sewerage sectors. A negligible number of greenfield projects were pro- posed after 1995 and were generally modest in investment volume. The sector has not been liberalized. Chile Chile was a leader in privatization, having started in 1975. Two pri- vatization waves can be distinguished: the first during 1975­82 and the second during 1985­89. In 1990 the new democratic government modified the privatization process, announcing that the sale of con- trolling stakes to the private sector would be limited to a few, small public enterprises; in other cases only a minority participation would become available to private investors. The government also announced its willingness to allow private participation in public infrastructure projects (water and sewerage, roads, and railways). Concessions have been a main tool for promoting competition. Laws regulating the electric and telecommunications sectors in Chile guarantee all firms applying for a concession the right to receive it. Concessions have been provided to any private sector agent seeking them, even in industries or stages of production and distribution that are closer to being natural monopolies. The rationale is that the regulator, by increasing the number of producers, favors consumers by creating the conditions for more competition, but the result is that con- cessions frequently overlap. In the telecommunications sector, the Corporación de Fomento de la Producción (CORFO), a state-owned corporation, owned 89.5 per- cent of Compañía de Telecomunicaciones de Chile (CTC) and 99 per- cent of ENTEL until 1986. The privatization of these two public telecommunication enterprises started in 1986 and was completed in 1990. In 1994 competition for national long-distance service was fi- nally allowed and in 1997 seven firms joined CTC and ENTEL to com- pete in the domestic long-distance service market. Also, competition in the cellular mobile telephone industry increased. The second wave of privatization in the electricity sector ran from 1986 until 1990. The two public enterprises, Empresa Nacional de Energía S.A. (ENDESA) (generation, distribution) and Compañía Eléc- trica de Chile (CHILECTRA) (distribution) were privatized and split into different enterprises. Greenfield projects appeared in 1990 and started to be significant in volume (even if not in number) in 1994. Currently all power generation is in the hands of the private sector. In the gas sector no privatization as such was carried out. Genera- tion or exploitation is still in public hands, transportation of gas is 88 THE LIMITS OF STABILIZATION done by public enterprises or by entities with concessions, and gas dis- tribution is carried out by entities with concessions only since 1986, when the law for concessions was enacted. Greenfield projects in the sector are negligible. In the road and rail sector, railway privatization started in 1995 with partial divestiture of Ferrocarril del Pacífico S.A. (FEPASA) and it continued with the full divestiture of Ferrocarril Regional del Norte (FERRONOR) in 1997. Since 1993 the government has been approv- ing concessions to the private sector to manage the road network. There are no greenfield projects in the transport sector. Between 1988 and 1989, new legislation decentralized responsibil- ity for publicly owned water and sewerage services in Chile, by creat- ing autonomous regional service companies. The national government owns the majority of shares in these companies through its Develop- ment Corporation. A national regulatory agency, the Superintendence of Sanitary Services, was created to regulate both public and private water and sewerage services. Under Chilean law, all water service com- panies, whether public or private, are structured as stock corporations and operate by virtue of concessions granted by the Ministry of Public Works. Concessions, which are granted for an indefinite period of time, have been awarded since 1993. No greenfield projects are pres- ent in the sector. Colombia The Constitucion Politica of 1991 was established to put an end to the state monopoly in public services. After the constitution, a significant number of public enterprises were singled out for privatization, among them major mining, banking, and tourism enterprises. On the telecommunications side, Colombia chose to open the sec- tor to new competition instead of privatization. Several greenfield projects were presented after 1994. Reform of the power sector started in 1992 (including privatization of some public entities) and finished in 1998, having achieved a major degree of private participation in the sector. A number of greenfield projects appeared after 1993. In the gas industry only one privatization was carried out in 1996 (Promigas). Greenfield projects are negligible in the gas sector. Some concessions of highways were approved in 1994. However, railway management remains under public sector control. There are no greenfield projects in the transport sectors. Finally, water supply and sewerage services were not privatized and only one greenfield project was proposed, in 1997. LATIN AMERICA'S INFRASTRUCTURE 89 Ecuador In the telecommunications sector, the Instituto Ecuatoriano de Teleco- municaciones (IETEL) was owned entirely by the Ecuadorian state. IETEL also had the monopoly for local, long-distance, and interna- tional telephone service. IETEL had the authority to regulate the telecommunication sector until 1992, when a separate regulatory or- ganization was set up to perform this task--the Superintendencia de Telecomunicaciones (SUPTEL)--along with a new state-owned corpo- ration named Empresa Estatal de Telecomunicaciones (EMETEL). This corporation took over the assets of IETEL and was granted mo- nopoly status for the provision of local, long-distance, and interna- tional telephone services. In preparing EMETEL for privatization, in June 1997, the government divided the firm into two limited liability companies. After rescheduling the auction for both companies several times, when the final date arrived (November 20, 1997) none of the interested investors submitted a bid. Only a few small greenfield proj- ects were carried out after 1994 (two a year). In the electricity sector, the main entities are the state-owned and vertically integrated Instituto Ecuatoriano de Electrificación (INECEL); the investor-owned utility Empresa Eléctrica de Ecuador (EMELEC), which has been subject to INECEL's technical and finan- cial control since 1985; and several private and municipal entities. Leg- islation submitted to congress in 1994 proposed to restructure the sec- tor, advocating deregulation and competition. The proposed law would divest all government-owned assets in generation, transmission, and distribution after reorganizing INECEL and consolidating distri- bution utilities into four or five enterprises. All new investment would be undertaken by the private sector. However, the privatization process is still pending. Only one greenfield project was initiated in 1996 with a total investment cost of US$30 million. Mexico In 1989 a law allowing privatization of the telecommunication state- owned enterprise was enacted. Before the privatization in 1990, Tele- comunicaciones de México (TELMEX) was a 66-percent-state-owned corporation, with the rest in the hands of local private shareholders. In 1990, TELMEX was granted a monopoly on fixed line telephone serv- ices until August 1996. After 1996 other firms offered long-distance services, but TELMEX will maintain the monopoly for local fixed tele- phone services until 2026. Cellular telephones and value-added serv- ices were opened to competition immediately. Since privatization, 90 THE LIMITS OF STABILIZATION several firms have been awarded licenses for cellular and long-distance telephone service. Greenfield projects in the sector started to be signif- icant after 1996. In the electricity sector, the Comisión Federal de Electricidad (CFE) is a state-owned enterprise that currently owns and operates most gener- ating plants serving the public power system and provides all transmis- sion and distribution service except in Mexico City. Since 1992, private power generators in the form of independent power producers (IPPs), self-generators, co-generators, and power importers are allowed to par- ticipate in the sector. Greenfield projects are almost negligible (only one in 1998). The liberalization of the sector is, then, still pending. In the gas sector, Repsol was partially privatized in two stages, 1995 and 1997. Additionally, greenfield projects started to have some weight in 1997. Until the early 1990s, publicly owned Ferrocarriles Nacionales de Mexico (FNM) was the largest company in the railway sector. The process of reform took off with President Carlos Zedillo. In February 1995 the Mexican Congress approved an amendment to the constitu- tion opening opportunities for private sector investment. Privatization started in December 1996 with concessions of 50 years. These conces- sions allow bidders to operate, exploit, and build new lines. The second stage of the privatization process was the sale of the shares owned by the government in the concessionaire companies through a public bidding process open to private investors. By June 1999, the process of opening the main Mexican rail lines to private operators was almost finished and virtually the whole Mexican railroad system had been privatized. During 1987­94, the government awarded 52 concessions of toll roads to the private sector. In April 1997, the government announced a new plan in the road sector and, in late 1997, assumed all bank lia- bilities along with temporary ownership of 23 toll roads. In the water and sewerage sector, the Federal District Water Com- mission awarded general contracts for a 10-year period (with the pos- sibility of extension) in 1993. Other concessions were awarded to the private sector after 1994 and a small number of greenfield projects have been developed since 1993. Private operators are involved in dis- tribution and commercial activities, but not in production. Addition- ally, most private participation in the sector has been carried out through PTOs (Plantas de Tratamiento), so that full liberalization of the sector is still pending. Peru In the telecommunications sector, the Peruvian government sold 35 percent of its shares in ENTEL and Compañía Peruana de Teléfonos LATIN AMERICA'S INFRASTRUCTURE 91 (CPT) to Telefónica Internacional of Spain in 1994. Telefónica took over the operation of both firms and within a year the firms merged into a newly formed firm called Telefónica del Peru. At the time of pri- vatization the firms were licensed to provide local and long-distance telephone services in the whole Peruvian territory. The license granted a five-year monopoly in fixed and long-distance service (ending in 1999). Competition in public payphones, cellular (local), cable TV, and value-added service was allowed. Two firms, Telefonica and Tele2000­Bellsouth, provide cellular mobile telephone service. How- ever, competition in the cellular sector was allowed only in Lima until 1998 when Tele2000­Bellsouth won the concession for Band B for the rest of Peru. The number of greenfield projects in the telecommunica- tions sector is almost negligible, with three projects in 1990 (US$150 million) and one in 1995 (US$30 million). The reason for this is the five-year monopoly status in fixed and long-distance service given to Telefonica del Peru. Thus, competition in the sector is still very weak. The power sector underwent restructuring and initiated a major pri- vatization program in 1994, following enactment of a new Electric Concession Law in 1992. The law opened the sector to private partic- ipation in all areas; required the separation of generating, transmis- sion, and distribution functions and ownership; and aimed at complete divestiture of all state-owned sector enterprises. Currently, 62 percent of Peru's generation capacity and 75 percent of the country's distribu- tion system are in private hands. Although the number of greenfield projects in the sector has been negligible so far, additional competition is being promoted. No liberalization has taken place in the rest of the infrastructure sectors. República Bolivariana de Venezuela Before privatization, the Venezuelan state owned 100 percent of the assets in the telecommunications sector. Compañía Anónima Nacional de Teléfonos de Venezuela (CANTV) had a monopoly in local and long-distance service. Since 1988, it was also the sole provider of cel- lular phone services. In May 1991, a license for the provision of cellu- lar telephone service was awarded to Telcel Celular S.A., a consortium of Venezuelan investors and BellSouth. Thus, Telcel Celular S.A. started competing with CANTV in the cellular phone business nation- wide. In November 1991, CANTV was privatized and received a 35-year concession contract. The license granted a monopoly status for a period of nine years in local and long-distance (national and in- ternational) telephone services. That is the reason for the limited num- ber of greenfield projects in Venezuela since 1991 (four in 1998 and 92 THE LIMITS OF STABILIZATION only one project from 1991 to 1996). The sector was not liberalized until more recently (1997). In the electricity sector, there are five state-owned and seven investor- owned utilities (IOUs) The largest state-owned enterprises are Electrifi- cación de Caroni (EDELCA) and Compañía Anónima de Admin- istración y Fomento Eléctrico (CADAFE). Electricidad de Caracas (EdC) is the main IOU, supplying most of Caracas and holding part ownership in three other IOUs. Distribution systems were reorganized into regional enterprises before privatization. CADAFE is being reorganized into four regional distribution units, a separate transmission unit, and separate hydro and thermal generating units, with privatization expected for many of these units. There have been a few greenfield projects since 1992 (a total of five each year, stopping in 1996). Liberalization of the sector is, then, pending. No liberalization has taken place in the rest of the infrastructure sectors. Notes 1. The data are drawn from the infrastructure database assembled for this research. A summary description of the sources and coverage is given in appendix 2A. 2. OECD is defined here excluding the Republic of Korea and Mexico. 3. The country detail is similar in the case of total phone lines and local con- nection capacity, and therefore for the sake of brevity is not presented. 4. This is particularly so in the case of unpaved roads. An indicator prefer- able to road length, used in the text, would be their length in terms of lane­ kilometers equivalent, but such information is not widely available across countries and over time. However, even this improved metric cannot reflect the overall efficiency of the road network, which can vary greatly across countries. 5. The data sources are described in appendix 2A. 6. The small magnitude of the regression coefficient is somewhat puzzling. Allowing for lagged effects of the primary balance does not lead to significant changes. 7. To smooth some large jumps in the data series, we use a centered three- year moving average in the graphs. This means that they would not be greatly affected qualitatively if the reform date were to be shifted forward or back- ward by one or two years. 8. We calculate the prereform public spending per capita as the average over 1970­89 for all countries that have data on the sector, in 1995 dollars. 9. Other experiments using instead the log of real infrastructure spending, or its ratio to the lagged stock, as explanatory variable yield qualitatively sim- ilar results, so we do not report them here. In the cases of telecom and trans- port routes, we also experimented with alternative definitions of the depend- ent variable: total phone lines, rather than main lines, for telecommunications; and total and paved roads, rather than total roads plus railways, for transport. The results were virtually indistinguishable from those reported in the table. LATIN AMERICA'S INFRASTRUCTURE 93 10. The area variable typically carried a positive coefficient significant at the 5 percent level or better, so we opted for retaining this specification for the transport equation. We also experimented with population density as an ad- ditional variable, but it always turned insignificant in the regressions. Finally, we also estimated specifications including land area in the accumulation equa- tions for phone lines and power, but the estimated coefficient on the area vari- able was always very far from significance at conventional levels. 11. We return to these issues in chapters 3 and 4. 12. Canning suggested, however, that the data on the length of the line could present some problems. First, they do not take account of the number of tracks in the railway. Second, there are changes in the coverage because of the treat- ment of rail lines owned by companies for industrial use and not open to the public (for example, railways owned by the sugar industry in Latin America). 13. The material in this appendix is based on background work by Pilar Blanco. References American Telephone and Telegraph Company. Various years. The World's Telephones. New York. Calderón, César, William Easterly, and Luis Servén. 2002. "The Output Cost of Latin America's Infrastructure Gap." Central Bank of Chile Working Paper No. 186, October. Canning, David. 1998. "A Database of World Stocks of Infrastructure, 1950­95." World Bank Economic Review 12 (3): 529­47. ------. 1999. "The Contribution of Infrastructure to Aggregate Output." World Bank Policy Research Working Paper 2246. Washington, D.C. de Haan, Jakob, Jan Egbert Sturm, and Bernd Jan Sikken. 1996. "Govern- ment Capital Formation: Explaining the Decline." Weltwirtschaftliches Archiv 132 (1): 55­74. Easterly, William. 1999. "When Is Fiscal Adjustment an Illusion?" Eco- nomic Policy (April): 57­86. ------. 2001. "Growth Implosions and Debt Explosions: Do Growth Slow- downs Cause Public Debt Crises?" Contributions to Macroeconomics Vol. 1: No.1, Article 1. (http://www.bepress.com/bejm/contributions/vol1/iss1/art1) Estache, Antonio, Vivien Foster, and Questin Wodon. 2001. Accounting for Poverty in Infrastructure Reform: Learning from Latin America's Experi- ence. Washington, D.C.: World Bank. Hicks, Norman L. 1991. "Expenditure Reduction in Developing Countries Revisited." Journal of International Development 3 (1): 29­37. 94 THE LIMITS OF STABILIZATION International Road Federation. Various years. World Road Statistics. Geneva. ITU (International Telecommunications Union). 1996. World Telecommu- nications Indicators. Geneva. ------. Various years. World Telecommunications Development Report. Geneva. Mitchell, Brian R. 1992, 1993, 1995. International Historical Statistics. Grover Dictionaries Incorporated. Pritchett, Lant. 2000. "The Tyranny of Concepts: CUDIE (Cumulated, De- preciated, Investment Effort) Is Not Capital." Journal of Economic Growth 5 (4): 361­84. Röller, Lars-Hendrik, and Leonard Waverman. 2001. "Telecommunica- tions Infrastructure and Economic Development: A Simultaneous Approach." American Economic Review 91: 909­23. Roubini, Nouriel, and Jeffrey Sachs. 1989. "Government Spending and Budget Deficits in the Industrial Countries." Economic Policy 8: 99­132. Servén, Luis. 1997. "Uncertainty, Instability, and Irreversible Investment." World Bank Policy Research Working Paper 1722. Washington, D.C. United Nations. Various years. Statistical Yearbook, CD-ROM. New York. ------. Various years. The Energy Statistics Yearbook. UN Statistics Divi- sion, New York. World Bank. 1988. World Development Report. Washington, D.C., Oxford and New York: Oxford University Press. ------. 1994. Adjustment in Africa. Policy Research Report, Washington, D.C., Oxford and New York: Oxford University Press. ------. Various years. World Development Indicators. Washington, D.C. ------. Various years. Railways Database. Washington, D.C. Available online at www.worldbank.org/html/fdp/transport/rail/rdb.htm. 3 The Output Cost of Latin America's Infrastructure Gap César Calderón and Luis Servén CHAPTER 2 SHOWED THAT OVER THE last two decades Latin America lost substantial ground vis-à-vis other developing and industrial regions in the quality and quantity of infrastructure assets. Although there was considerable diversity across countries in the magnitude of this phe- nomenon, it affected virtually all infrastructure sectors in all of the region's countries. Table 3.1 provides a summary illustration of Latin America's infra- structure growth relative to that of the seven successful economies of East Asia (the "tigers").1 The table presents the change in the infra- structure gap over 1980­97--measured by East Asia's infrastructure stocks per worker relative to those of Latin America--using both re- gional averages and medians. The two sets of figures tell the same story. Latin America's infra- structure gap grew by a huge margin in the last two decades: 40 to 50 percent for road length, 50 to 60 percent for telecommunications (defined as the total number of main telephone lines), and as much as 90 to 100 percent in power generation capacity. The loss of ground was particularly marked in the 1980s for all three assets in the table. In the 1990s Latin America continued to fall behind at a rapid pace in power generation capacity, but its loss of ground in transport routes proceeded at a slower pace than in the previous decade and the gap in telecommunications infrastructure ceased to expand.2 The consequences of this loss of ground for growth and welfare in the region are a matter of concern. Lack of adequate infrastructure services results in lower productivity and higher production costs for private producers. Poor road and telecommunication networks 95 96 THE LIMITS OF STABILIZATION Table 3.1 The Widening Infrastructure Gap, Latin America versus East Asia (percentage change in relative infrastructure stocks per worker) Medians by region Simple averages by region Infrastructure asset 1980­97 1980­89 1990­97 1980­97 1980­89 1990­97 Main phone lines 63.58 45.86 ­14.01 47.61 42.52 2.98 Power generating capacity 101.21 50.03 40.66 91.14 45.61 39.56 Roads 43.98 21.34 10.09 52.53 36.11 13.14 Memo item: Change in relative GDP per worker 88.89 52.66 26.60 90.24 55.75 26.55 Note: Each cell in the table shows the percentage change in the stock of the respec- tive infrastructure asset in East Asia minus the same change in Latin America. Source: Authors' calculations. increase transport and, more generally, logistic costs, which--for Latin America--have been shown in comparative studies to exceed the international norm by wide margins (Guasch and Kogan 2001). The reduced profitability in turn discourages private investment. Through all these channels, the result is lower output growth. For later refer- ence, the bottom of table 3.1 also shows that the gap in gross domes- tic product (GDP) per worker (in terms of adjusted purchasing power parity [PPP]) between East Asia and Latin America grew by some 90 percent over 1980­97. As with infrastructure, Latin America's loss of ground was particularly marked in the 1980s. Figure 3.1 brings out graphically the association between infra- structure accumulation and growth performance. The figure plots the average growth rate of GDP per worker over the last four decades against the average rate of growth of infrastructure endow- ments--with the latter measured by the simple average of the growth rates of phone lines, kilometers of roads, and power gener- ation capacity, all in per-worker terms. Even with this crude meas- ure of infrastructure stocks, a strong positive correlation is apparent between infrastructure accumulation and growth performance. A simple cross-country regression of growth on infrastructure accu- mulation yields a highly significant positive regression coefficient and an R2 of 32 percent. Strong as this empirical association is, it need not reflect causation from infrastructure services to aggregate output. The observed corre- lation could reflect reverse causation from GDP to infrastructure demand, or the action of third factors affecting both GDP and THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 97 Figure 3.1 Infrastructure Accumulation and Growth, 1960­97 (country averages, percent) Growth in GDP per worker 7 6 5 4 3 2 1 0 -1 -2 y = 0.505x + 0.0006 R2 = 0.3253 -3 -2 0 2 4 6 8 10 Growth in infrastructure stocks per worker Rest LAC EAP7 Note: EAP7 seven countries in the East Asia and Pacific region; LAC Latin America and the Caribbean region; Rest all other countries. infrastructure stocks. Thus, the key question is: what was the role of Latin America's growing infrastructure gap in the widening of the out- put gap? The rest of this chapter is devoted to answering that question. Methodological Approach The empirical approach of this chapter is based on the estimation of an aggregate production function augmented with infrastructure cap- ital. The analysis is related closely to that in Canning 1999, and fol- lows a recent literature concerned with the contribution of infrastruc- ture to aggregate output (Canning and Bennathan 2000, Demetriades and Mamuneas 2000, Esfahani and Ramirez 2002, and Röller and Waverman 2001). For simplicity, the approach taken here follows the literature in adopting a Cobb­Douglas specification of the infrastructure- augmented production function:3 y k h z (1 )l (3.1) where y is aggregate value added (GDP), k is the physical noninfra- structure capital stock, l denotes labor, h is human capital, and z is a 98 THE LIMITS OF STABILIZATION measure of infrastructure capital. All the variables are expressed in logs, and constant returns to scale are assumed. It is important to note that (3.1) implicitly assumes that infrastruc- ture services are a fixed proportion of the infrastructure capital stock. Thus, other things being equal, larger stocks should be reflected in higher aggregate output. This approach is analogous to that conven- tionally used in standard production functions excluding infrastruc- ture, which assume that physical and human capital services are pro- portional to the respective stocks k and h. In principle, the parameter in (3.1) should capture the elasticity of output with respect to infrastructure for given values of the other inputs. However, this presumes that k includes noninfrastructure capital only. In reality, what this includes is data on the total capital stock, including both infrastructure and other physical assets. Thus, infrastructure capi- tal appears twice in the equations--as part of k, and separately as z. Hence, the parameter captures the extent to which the productivity of infrastructure exceeds (if 0) or falls short of ( 0) the productivity of noninfrastructure capital. See Canning 1999 for further discussion. The contribution of infrastructure capital to output can be found by noting that the measured capital stock is a weighted sum of infrastruc- ture and other physical assets, with weights given by their respective rel- ative prices. Thus, letting k denote noninfrastructure physical capital, one can write: K pz Z k k z (3.2) K pz Z K pz Z where uppercase letters denote the anti-logs of lowercase variables; pz is the relative price of infrastructure capital in terms of noninfra- structure capital; and the assumption is that the latter is approxi- mately equal to the price of overall capital, under the presumption that infrastructure assets are typically a small fraction of the total capital stock.4 Combining (3.1) and (3.2), the elasticity of output with respect to infrastructure can be expressed: 0y g ua K hz (3.3) 0z where pz Z u K (3.4) K pz Z THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 99 is the share of infrastructure in the overall physical capital stock. These expressions involve log-linear approximations around an arbitrary point (for example, the sample mean), and hence should be evaluated accordingly. In practice, because infrastructure stocks typically ac- count for relatively small portions of the overall capital stock, the dif- ference between z and the naïve estimate should be fairly modest. Finally, it is worth noting that (3.4) captures only the direct impact of infrastructure on output, leaving aside the possible indirect impact occurring through the effects of infrastructure on the accumulation of other productive inputs, of which the most important is noninfra- structure capital. To the extent that both types of capital are gross complements in production (as assumed here), an increase in infra- structure capital raises the profitability of noninfrastructure capital and, other things being equal, should lead over time to a higher K, which in turn should cause a further output expansion. By ignoring this indirect effect, the contribution of infrastructure to output over the long term is likely to be underestimated in the calculation below.5 Empirical Implementation For estimation purposes, equation 3.1 above is rewritten in terms of ratios to the labor force: yit lit ai bt a(kit lit) b(hit lit) g(zit lit) eit (3.5) Here the subscripts i and t are used to index countries and years, respectively; the terms ai, bt capture country-specific and time-specific productivity factors; and it is a random disturbance that will be assumed uncorrelated across countries and over time. The objective is to estimate the parameters of equation 3.5 using a large panel data set. Annual data for the period 1960­97 from 101 industrial and developing countries are used--or close to 4,000 obser- vations. In practice, some of the instrumental variable estimators employed below use up several lags of the variables to construct instruments, so that the effective data set comprises 101 countries and 3,232 observations. To ensure comparability across estimators, the data set is limited to this reduced sample even when employing simpler estimators using no lags.6 Sample coverage and data sources are described in detail in the appendix. The measures used for output (GDP) and physical capital per worker are based on suitably expanded versions of the Summers-Heston 100 THE LIMITS OF STABILIZATION data set (Summers and Heston 1991), whereas the (log) human capital stock is measured by the number of years of secondary schooling of the working-age population.7 Regarding infrastructure capital, the focus is primarily on the three standard indicators of infrastructure endowments used in table 3.1: elec- tricity generating capacity (in gigawatts), road length (in kilometers), and the number of main telephone lines. However, some experiments are also performed with alternative measures of infrastructure capital. Each of these variables is scaled dividing by the total labor force. Although these measures of infrastructure capital are admittedly crude--in partic- ular, they do not capture variations in the quality of infrastructure--they are chosen because of their broad availability across countries and over time, and their frequent use in the recent empirical growth literature. There is by now a considerable literature reporting empirical esti- mates of equations similar to (3.5) above (see Gramlich 1994 for an overview). In the present panel context, there are four main issues to take into consideration: cross-country heterogeneity, common factors, measurement error, and endogeneity. The first issue is the possible cross-country heterogeneity of the pro- duction technology. Imposing a common technology when in reality production functions vary across countries would lead to inconsistent estimates. To address this issue, country-specific effects ai are allowed for in the estimations below. Omission of fixed effects is known to lead to a large overstatement of the contribution of infrastructure to output (see, for example, Holtz-Eakin 1994, and Röller and Waverman 2001). A second specification issue concerns the possible existence of omit- ted common factors--such as the world business cycle--causing output to move together across countries. These common factors can result in cross-country residual correlation, which in turn leads to invalid infer- ences with the estimation methods to be used below. To eliminate the common factors, time-specific effects in the estimated regressions are allowed for; this is equivalent to a regression in which each variable en- ters as a deviation from its cross-sectional mean in the year in question. The third problem is measurement error, which is likely to be im- portant particularly for infrastructure stocks. There are two reasons for this. On the one hand, the quality of the stocks (for example, the condition and capacity of roads or the reliability of power plants) can vary greatly not only across countries but also within countries. It is unfortunate that data on the quality of infrastructure are not readily available for a large cross-country time-series data set such as the one under consideration.8 On the other hand, the timing of changes to the reported infrastructure stocks is to some extent arbitrary. For exam- ple, impassable roads or unusable portions of railway track may remain in the books for some time before being suddenly removed THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 101 from the reported stock figures, or new power plants may not become fully operational until some time after completion. Formally, these considerations imply that infrastructure may be measured with error, so that the time-varying disturbance it may include a measurement error correlated with the infrastructure variables. Standard estimates of (3.5) would therefore be subject to attenuation bias, most likely causing underestimation of the coefficients of the infrastructure stocks. Related to this is the problem of endogeneity, which may affect the infrastructure regressors in (3.5) and perhaps the physical and human capital stocks per worker. It can be argued that infrastructure stocks are jointly determined with output per worker, but the positive corre- lation of infrastructure stocks with output found in the data could merely reflect the fact that the income elasticity of infrastructure demand is positive. Arguably, similar considerations could be made for the physical and human capital stocks. In the univariate case, standard least-squares estimation in the pres- ence of reverse causation from output to infrastructure would lead to an upward bias in the infrastructure coefficient; in the multivariate case the situation is more complex and the direction of the bias cannot be established a priori--and even more complex in the presence of measurement error that may introduce attenuation bias. One way to address the two-way causality between infrastructure and output would be to develop a fully specified simultaneous model of infrastructure supply and demand. Unfortunately, this would pose stringent data requirements well beyond the scope of this research.9 An alternative, less-demanding way to tackle both measurement error and endogeneity is to use an instrumental-variable estimation approach. However, there are few exogenous instruments available with the broad time-series and cross-country coverage needed here. Demographic variables are perhaps the only obvious source of identi- fying information, because they are likely to affect the demand for infrastructure (as well as physical and human capital) services without being subject to reverse causation or correlated with the infrastructure measurement error. Thus, urban population and population density (both in logs) are used as outside instruments.10 These strictly exogenous instruments are complemented with ap- propriate internal (that is, weakly exogenous) instruments constructed along the lines of Griliches and Hausman (1986) and Arellano and Bond (1991), given by suitably lagged values of the explanatory vari- ables in (3.5). Specifically, the first differences of (3.5) are used to remove the country-specific effect:11 ¢(yit lit) ct a¢(kit lit) b¢(hit lit) g¢(zit lit) ¢eit (3.6) 102 THE LIMITS OF STABILIZATION where ct bt bt 1. Under appropriate assumptions about the serial correlation of it(the time-varying disturbance, possibly inclusive of measurement error), lagged levels of the variables on the right side be- come valid instruments. In particular, if itis serially uncorrelated and the regressors are weakly exogenous (that is, uncorrelated with future real- izations of itbut not with its current or past realizations), then the sec- ond and higher lags of the regressors become valid instruments in (3.6). More generally, if itfollows a moving average process of order q, then lags q 2 and higher of the regressors would become valid instruments. Validity of the instruments used in the estimation can be tested di- rectly through Sargan tests of orthogonality between the instruments and the error term, as well as indirectly through tests of first- and higher-order autocorrelation of the errors (see Arellano and Bond 1991). For example, if itis serially uncorrelated, then its first differ- ence included in (3.6) should display first- but no higher-order auto- correlation, in which case twice-lagged regressors are indeed valid instruments, as stated earlier. The above discussion characterizes the generalized method of mo- ments (GMM) estimator in first differences of Arellano and Bond (1991). However, under additional assumptions, a more efficient IV estimator may be available: the system GMM estimator of Blundell and Bond (1998), which combined estimation of (3.6) and (3.5) using lagged differences of the regressors as instruments in the level equation 3.5.12 The validity of these additional instruments can be checked through difference-Sargan tests of orthogonality between the extra instruments and the error term. Estimation Results Table 3.2 reports the sample correlations among the dependent and in- dependent variables. The figures below the main diagonal reflect the correlation among the levels of the variables, whereas those above the diagonal correspond to their first differences. Anticipating some of the experiments below, two alternative infrastructure measures are presented for transport routes, total roads, and total roads plus rail- ways (with the latter variable available only for a smaller country sample), and two measures as well for telecommunications--main lines and total lines, including cellular.13 In both levels and differences, real GDP per worker shows a signif- icant correlation with each of the infrastructure measures and with the physical and human capital stocks per worker. Among the infra- structure variables, the biggest correlation with GDP corresponds by in Total lines phone 0.14** 0.06** 0.11** 0.05** 0.03 0.03 0.97** n.a. iables var the Main phone lines 0.13** 0.06** 0.05** 0.07** 0.03 0.04** n.a. to 1.00** refer diagonal routes 0.05** 0.02 0.00 0.03 1.00** n.a. 0.25** 0.26** Transport main the below Roads 0.05** 0.02 0.01 0.03 n.a. 1.00** 0.27** 0.28** Values logs. in 0.07** 0.05** n.a. 0.21** 0.19** 0.60** 0.59** Electricity generating capacity 0.05**­ expressed 0.04** 0.01 n.a. 0.29** 0.02 0.02 0.30** 0.32** Secondary schooling schooling) (except and Physical capital 0.21** n.a. 0.34** 0.47** 0.20** 0.20** 0.55** 0.57** differences. worker first to per refer GDP n.a. 0.71** 0.39** 0.50** 0.23** 0.22** 0.58** 0.60** Correlations measured are diagonal percent. the 5 calculations.' at Sample lines lines applicable. schooling variables above generating routes capital Not Authors 3.2 All phone phone values capacity n.a. Note: **Significant Source: Table Variable GDP Physical Secondary Electricity Roads Transport Main Total levels; 103 104 THE LIMITS OF STABILIZATION far to the telecommunication measures. Unsurprisingly, the magnitude of the correlations is much bigger when the variables are expressed in levels than when they are expressed in differences. In turn, the infra- structure measures are also positively correlated with each other, again more so in terms of levels than in terms of differences. Finally, there seems to be little difference between the two alternative measures of transport routes (their correlation exceeds 0.99 in both levels and dif- ferences) and the two measures of telecommunications infrastructure (their correlation is 0.97 in differences and virtually 1.00 in levels). Before proceeding to GMM estimation, table 3.3 reports empirical results using simpler estimators for equation 3.5.14 The first two columns present ordinary least-squares (OLS) estimates on the cross- section (column one) and pooled sample (column two), neither of which is robust to heterogeneity, measurement error, or endogeneity of the regressors. The two sets of estimates are similar: in both cases a Table 3.3 Infrastructure-Augmented Production Function: Alternative Estimates 1 2 3 4 Cross-section Variable OLS Pooled OLS Within 2SLS Physical capital 0.472 0.387 0.245 0.414 (5.324) (7.685) (7.199) (7.644) Secondary schooling ­0.005 0.016 0.135 0.017 (0.123) (0.474) (2.758) (0.492) Electricity generating capacity 0.030 0.051 0.068 0.047 (0.512) (1.137) (2.294) (1.002) Roads ­0.055 ­0.046 0.026 ­0.049 (1.702) (1.473) (0.707) (1.586) Main phone lines 0.147 0.185 0.133 0.169 (2.433) (4.432) (4.544) (3.883) R2 0.954 0.939 0.987 0.939 1st-order autocorrelation (p-value) n.a. 0.000 0.341 0.000 2nd-order autocorrelation (p-value) n.a. 0.000 0.945 0.000 Number of observations 101 3,232 3,232 3,232 Number of countries 101 101 101 101 Note: Dependent variable is log GDP per worker. 2SLS two-stage least-squares. All variables are measured per worker and (except schooling) expressed in logs. t-statistics in brackets are heteroskedasticity-consistent. Source: Authors' calculations. THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 105 sizable output contribution of the capital stock and a significant effect of telecommunications infrastructure are found. The remaining coeffi- cients are insignificant, although that on transport routes approaches statistical significance with a counterintuitive negative sign. The pooled OLS results also show strong evidence of serial correlation of the residuals, a clear symptom of misspecification. Column three reports the within estimator, which controls for country-specific effects but not for endogeneity or measurement error. In the presence of the latter, the within transformation can lead to badly misleading estimates (see Griliches and Hausman 1986). In the present case, it can be seen that all the regressors carry positive coeffi- cients, all significantly different from zero except for that of transport routes. Among the infrastructure variables, telecommunications car- ries a much larger coefficient than the rest, similar to the OLS results.15 The estimators presented so far ignore the issues of measurement er- ror and endogeneity. Column four reports two-stage least-squares (2SLS) estimates of (3.5) using as instruments the current and first three lags of urban population and population density, plus the second lags of the explanatory variables.16 The estimates obtained in this manner are similar to the pooled OLS estimates and equally disappointing. Apart from the physical capital stock, only the telecommunications variable is significant. Moreover, a Sargan test rejects the validity of the instrument with a p-value of less than 0.001--an unsurprising outcome in view of the strong evidence of autocorrelation of the residuals shown in the table, which provides a clear indication of misspecification. Table 3.4 turns to GMM estimation using alternative specifications and instrument sets. Column one reports the base specification, using the difference-GMM estimator and the same instrument set as in the last column of table 3.3--twice-lagged levels of the explanatory vari- ables plus the current value and three lags of the exogenous demo- graphic variables. Comparison of these GMM estimates with the within estimates in table 3.3 shows that in every case the former are larger in magnitude than the latter, which hints at the possible presence of attenuation bias in the within estimates.17 Moreover, the GMM es- timates of the coefficients of all three infrastructure variables are all sta- tistically significant (although only at the 10 percent level in the case of power). They are also of roughly similar magnitude. Finally, the diag- nostic tests provide support for the selected specification--the Sargan test shows no evidence against the validity of the instruments and, as anticipated, the serial correlation tests hint at first-order but no higher- order serial correlation of the differenced-error term. Column two provides a robustness check by lagging the instruments one extra period--that is, using the third rather than the second lags of 4 differences 101 System 0.222 (7.867) 0.222 (5.520) 0.109 (2.970) 0.005­ (0.084) 0.147 (6.164) 0.000 0.002 0.555 0.778 3,232 worker. Levels per GDP log is 3 101 0.351 (7.903) 0.159 (3.443) 0.177 (2.468) 0.105 (1.241) 0.138 (3.168) 0.000 0.141 0.143 0.888 3,232 variable Demographics Dependent 3 logs. t 101 in 0.361 0.169 (3.815) 0.123 (2.148) 0.117 (2.195) 0.140 (3.236) 0.000 0.312 0.106 0.794 3,232 (11.034) Differences Levels expressed 2 schooling) t 12 101 0.363 0.148 (3.361) 0.112 (1.809) 0.119 (2.197) 0.151 (3.634) 0.000 0.319 0.111 0.793 3,232 (10.832) (except Levels and brackets. Estimates worker in -value)p( per GMM -value)p( -value)p( -statisticst measured capacity are significance calculations.' Alternative lines joint schooling generating -value)p( of autocorrelation autocorrelation observations countries variables Authors 3.4 capital of of All specification, test phone test Note: Source: Table Model instruments Physical Secondary Electricity Roads Main Wald Sargan 1st-order 2nd-order Number Number Heteroskedasticity-consistent 106 THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 107 the regressors as instruments (in addition to the demographic variables). The results are virtually identical to those in the preceding column, and the diagnostic tests continue to lend support to the specification. So far, lagged infrastructure and physical capital stocks have been used as instruments. One might object that these variables belong in the production function--if infrastructure assets take time to become productive--so that they do not provide identifying information. This assertion can be tested by dropping them and limiting the instrument set to the exogenous demographic variables. Thus in column three of table 3.4 only the current and first two lags of urban population and population density, as well as their squares, are included as instru- ments, along with the second lag of the schooling variables. Neverthe- less, the estimation results are similar to those in the preceding columns. The only exception is the coefficient on power generating ca- pacity, which becomes considerably larger than before. All other coef- ficients are virtually unchanged, although that on roads is now esti- mated with poor precision. Finally, in column four the system GMM estimator of Blundell and Bond (1998) is used, combining the levels equation 3.5 with the first- difference equation 3.6, and adding as instruments for the former the twice-lagged first differences of the same instruments used in column one. The parameter estimates that result are somewhat different from those obtained from the difference-GMM estimator. If anything, they are close to the within estimates in the previous table. However, the Sargan test clearly rejects the validity of the instruments, whereas the difference-Sargan test (not shown in the table) yields a p-value of less than 0.001 percent and thus provides an equally strong indication of misspecification. This suggests that the stationarity condition discussed earlier, required for the validity of the system GMM estima- tor, does not hold in our data. In view of these results, the remaining experiments are based on the difference-GMM estimator and retain the same set of instruments as in the base specification of column one in table 3.4. Using this as a starting point, table 3.5 presents experiments using alternative specifi- cations. Column one reproduces the initial specification for ease of comparison. In column two, roads plus railways are used, rather than roads alone, to summarize the transport network infrastructure. This leads to the loss of some 10 percent of the sample. The parameter es- timate on the combined transport variable is similar to that obtained earlier using roads only, although the point estimate is somewhat im- precise. As for the other parameters, the coefficient on power increases about 50 percent relative to its value in column one, whereas that on phone lines declines somewhat. However, these changes are modest 108 THE LIMITS OF STABILIZATION Table 3.5 First-Difference GMM Estimates of Alternative Specifications Specification Variable 1 2 3 4 Physical capital 0.363 0.365 0.363 0.365 (10.832) (12.642) (10.768) (10.718) Secondary schooling 0.148 0.119 0.139 0.153 (3.361) (2.780) (3.274) (2.792) Electricity generating capacity 0.112 0.174 0.118 0.123 (1.809) (2.642) (1.910) (2.000) Roads 0.119 0.117 0.119 (2.197) (2.180) (2.109) Roads railways 0.116 (1.646) Main phone lines 0.151 0.113 0.152 (3.634) (2.284) (2.832) Total phone lines 0.161 (3.507) Main phone lines squared ­0.009 (0.039) Wald test of joint significance (p-value) 0.000 0.000 0.000 0.000 Sargan test (p-value) 0.319 0.607 0.329 0.261 1st-order autocorrelation (p-value) 0.111 0.115 0.12 0.110 2nd-order autocorrelation (p-value) 0.793 0.536 0.793 0.789 Number of observations 3,232 2,941 3,232 3,232 Number of countries 101 92 101 101 Note: All variables are measured per worker and (except schooling) expressed in logs. Dependent variable is log GDP per worker. Heteroskedasticity-consistent t-statistics in parentheses. Source: Authors' calculations. relative to the respective standard errors. The other coefficients remain unchanged. Next, in column three main phone lines are replaced with total (main mobile) phone lines as the indicator of telecommunications infrastructure. This makes virtually no difference for any of the pa- rameter estimates, or for the diagnostic statistics, all of which are al- most identical to those in column one. Finally, in column four, nonlinear effects of telecommunications equipment are explored along the lines reported in Röller and THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 109 Waverman (2001); the authors found that the elasticity of output to telecommunications stocks increases with the level of the telecommu- nications stock. To explore this issue, the square of main phone lines per worker is added to the equation. Its estimated coefficient turns out to be negative, but wholly insignificant; the remaining coef- ficients show virtually no change. Thus the conclusion is that the data show little indication of nonlinear effects of telecommunications infrastructure.18 In all the specifications reported in table 3.5, the diagnostic statis- tics support the model. The Sargan tests show no evidence against the choice of instruments, and the serial correlation tests provide a mild suggestion of first- but no higher-order autocorrelation. The Output Cost As noted earlier, the empirical estimates reported so far do not cap- ture the total contribution of infrastructure to output because infra- structure stocks are already included in the overall capital stock. To identify that impact it is necessary to compute the elasticity of output with respect to infrastructure assets, as in equations 3.3 and 3.4. To compute the share of the different infrastructure stocks in the overall capital stock, data on the cost of infrastructure assets collected by Canning and Bennathan (2000) are used. There are some caveats, however. These costs are available only for a limited number of coun- tries, and do not necessarily correspond to assets of homogeneous quality. They also show a large degree of cross-country variation. For the purposes of this chapter, because the primary interest is the per- formance of Latin America, the capital stock shares are computed using the cost data available for countries in this region and the aver- age ratios of the relevant stocks over 1980­97; then the medians of the country-specific figures are taken as relevant regional value. Limited experiments with alternative ways to construct these shares led usually to roughly similar results; however, because many other procedures are possible, the results have to be taken as illustrative. They are re- ported in the middle column of table 3.6. According to the figures in the table, telecommunications infra- structure accounts for a little more than 1 percent of the overall capi- tal stock, whereas power and roads represent 14 percent and 16 per- cent, respectively. Using these shares for the calculation in equation 3.4, the elasticities reported in the third column of the table are ob- tained. As it turns out, the elasticities of the three infrastructure stocks are all of similar magnitude, with the largest corresponding to 110 THE LIMITS OF STABILIZATION Table 3.6 Elasticity of Output per Worker with Respect to Capital per Worker Regression Share of total Total Capital asset estimate capital stock elasticity Infrastructure capital Main phone lines 0.152 0.012 0.156 Power generating capacity 0.112 0.140 0.163 Roads 0.119 0.163 0.178 Noninfrastructure capital 0.363 0.685 0.249 Note: Capital stock shares are the medians of country values computed on the basis of cost data from Canning and Bennathan (2000) and from asset stock data for Latin America. Source: Authors' calculations. roads and the smallest to phone lines. The differences are very small, however--on the order of a few hundredths of a percent--and because of the uncertainties surrounding the underlying calculations, a com- mon value for all three is used below, which as a working hypothesis is placed at 0.16. This estimated elasticity can be used to provide a rough idea of the contribution of infrastructure stocks to the diverging performance of GDP per worker between Latin America and the East Asian tigers over the last two decades. More precisely, this is achieved calculating the portion of the change in the gap in GDP per worker between the two regions that can be attributed to the differential evolution of their respective infrastructure stocks--the infrastructure gap--that was portrayed in table 3.1 above. This is done in table 3.7, which shows the role of each infrastruc- ture asset in the widening GDP gap, as well as the combined role of all three vis-à-vis the other inputs--human capital and noninfrastructure physical capital. The table reports calculations using both unweighted means and regional medians. The estimated contributions of the infrastructure assets are sub- stantial. The top line in the table shows that all three assets combined account for about one-third of the widening GDP gap between East Asia and Latin America. In other words, the differential evolution of infrastructure assets in Latin America and East Asia widened the cross- regional gap in GDP per worker by some 30 percent over 1980­97. Of this total, the largest contribution (nearly half) corresponds to power generating capacity, whereas phone lines and roads combined had an impact of similar magnitude to that of power infrastructure on the GDP gap. This relative ranking of assets is unsurprising in view of THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 111 Table 3.7 The Infrastructure Gap and the Output Gap: Contribution of Various Inputs to the Change in Relative GDP per Worker, Latin America versus East Asia, 1980­97 Inputs Medians by region Simple averages by region 1. Infrastructure 33.40 30.61 Main phone lines 10.17 7.62 Power generating capacity 16.19 14.58 Roads 7.04 8.40 2. Noninfrastructure capital 30.28 29.86 3. Human capital 10.88 7.07 Sum 74.56 67.53 Actual change in GDP per worker 88.90 90.24 Residual 14.33 22.71 Note: The contribution of each input to the change in relative output is calculated multiplying the change in the input by the respective output elasticity estimate. The elasticities used are those in table 3.6. Source: Authors' calculations. their respective evolution depicted earlier in table 3.1, according to which power had the worst performance over the two decades under analysis. It is worth noting also that the results are similar whether re- gional medians or averages are employed in the calculation. The table also shows the contributions of the two conventional inputs--physical (noninfrastructure) and human capital. The slower accumulation of physical capital in Latin America relative to East Asia accounts for another 30 percent increase in the output gap--an amount similar to that attributable to infrastructure. Finally, the differential evolution of human capital across the two regions is responsible for up to another 10 percent increase in the output gap. The bottom line in the table shows that the estimated model tends to underpredict the change in the output gap between the two regions. Between 15 and 20 percent of the latter is left unexplained. Table 3.8 offers an individual-country perspective on the same phe- nomenon. For each country, the table reports the change in the infra- structure gap and the income gap (vis-à-vis the East Asia average) over 1980­97, as well as the contribution of the former to the latter. The first three columns of the table show that over the period in question nearly every country in Latin America lost ground relative to East Asia in all three infrastructure assets considered. The only exceptions were the [2]/[1] 36.17 34.76 37.56 66.00 39.00 44.36 49.39 30.65 50.07 31.90 20.69 32.34 44.31 50.68 32.75 34.48 15.67 33.66 in American change [2] Total 31.39 29.78 34.65 29.39 32.16 38.97 40.86 32.80 44.49 29.42 20.34 30.45 50.42 45.20 33.71 31.69 12.06 35.39 the Latin Relative output in 7.70 8.40 0.85 7.11 2.10 5.06 0.84 3.71 5.08 5.07 7.47 multiplying Selected 10.78 11.55 12.11 12.37 10.01 14.16 12.63 infrastructure relative Telecom of in Change versus calculated is the change 7.79 6.03 6.75 8.20 8.97 9.05 7.04 5.63 7.13 8.29 2.87 6.20 Roads 11.33 17.71 16.50 11.05 14.31 11.24 of Asia 3.7. the output Contribution to table East 7.47 relative Power 15.89 15.35 17.13 18.91 16.84 18.45 21.05 11.64 22.93 17.52 17.01 12.45 26.10 23.91 20.34 15.38 16.55 from in Contribution Worker, change estimate per [1] the 86.78 85.69 92.25 44.53 82.46 87.85 82.72 88.85 92.22 98.29 94.15 89.19 91.91 76.99 Output 107.00 113.80 102.92 105.13 to Gap: elasticity GDP asset worker output Output 5.67 5.58 per 51.36 56.01 71.84 47.42 76.99 14.01 80.76 33.71 24.72 82.48 66.73 94.40 33.86 33.83 49.79 84.21 Telecom the Relative infrastructure in stocks respective and changes each the of by Gap Roads 45.84 35.49 39.71 66.66 48.25 52.76 53.22 97.07 65.02 41.38 33.14 84.19 41.93 48.79 66.11 16.91 36.49 104.15 Change Relative median) the Infrastructure contribution to Asia Power 93.48 90.29 99.06 68.45 73.23 90.46 43.94 97.37 100.75 111.22 108.55 123.84 134.87 103.09 100.05 153.53 140.67 119.62 the East Infrastructure 97­ the Stocks de country, to calculations.' The 1980 Republic R.B. each (relative Authors 3.8 For Rica stock Salvador Note: Source: Table Infrastructure Countries, (percent) Country Argentina Bolivia Brazil Chile Colombia Costa Dominican Ecuador Guatemala Honduras Jamaica Mexico Nicaragua Panama Peru El Uruguay Venezuela, asset 112 THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 113 Chile and Jamaica in telecommunications and Uruguay in roads. Every country listed in the table also lost ground in power generation capac- ity per head, and the extent of the lag was particularly dramatic in the Dominican Republic, Guatemala, Nicaragua, and Panama. Except for Panama, these countries were also the least dynamic in the stock of roads, whereas Ecuador, Mexico, and Panama lost the most ground in telecommunications. The table also shows that the contribution of the infrastructure gap to the gap in income per worker--computed in the same way as in the preceding table--was positive for every country listed. In other words, in every country the widening infrastructure gap added to the income gap over the sample period. The output cost of lagging infrastructure was particularly large in Central America: in Guatemala, Nicaragua, and Panama the loss of ground in infrastructure assets widened the income gap by more than 40 percent relative to East Asia. At the other end, Jamaica and Uruguay were the least bad performers--that is, their loss of ground in infrastructure involved only a relatively modest cost in output per worker. Summary Over the last 20 years Latin America fell behind in infrastructure quan- tity and quality vis-à-vis other developing and industrial regions. Virtu- ally all countries and infrastructure sectors in the region were affected by this relative slowdown, which was particularly pronounced in the 1980s. The analysis in this chapter shows that this widening infrastructure gap can account for a considerable fraction--about one-third on aver- age--of the increase in Latin America's output gap relative to the suc- cessful East Asian economies over the 1980s and 1990s. Lagging telecommunication assets, power generation capacity, and road net- works all contributed to Latin America's loss of ground in output per worker. Although there is a fair degree of diversity across the region's economies in the magnitude of this effect, lagging infrastructure in every one of the countries analyzed added to the output lag vis-à-vis the East Asian tigers. These conclusions are based on empirical estimates of the contri- bution of infrastructure stocks to aggregate output computed over a large cross-country time-series data set, using an infrastructure- augmented production function specification. This framework yields positive and significant estimates of the output contributions of all three infrastructure assets considered, and of physical and human capital as well. 114 THE LIMITS OF STABILIZATION This analysis has to confront some difficulties, however, such as the potential endogeneity of infrastructure stocks and the fact that they are subject to measurement error because of heterogeneity in infrastruc- ture quality across countries and over time, among other things. To overcome these problems, instrumental variable estimators combining internal and external instruments are used. On the whole, the empiri- cal results support the approach taken. There is little evidence against the validity of the instruments, and the estimates do not change signif- icantly when alternative instrument sets are used or the instrument set is restricted to exogenous demographic variables only. This can be viewed as confirmation that the empirical estimates capture the effect of the exogenous component of infrastructure on output, and hence provide a valid basis for the chapter's inferences on the role of the in- frastructure slowdown in the slowdown of Latin America's growth over the period of analysis. Appendix Sample Coverage and Data To estimate the production functions presented in tables 3.2 to 3.5, an- nual data for 101 countries for 1960­97 (38 observations per country) were collected. Note that in the regression framework all figures are ex- pressed as magnitudes per worker. Output has been approximated by using the real GDP in 1990 PPP U.S. dollars from Summers and Heston (1991), complemented by the data from the Global Development Net- work Growth Database created by William Easterly at the World Bank. Analogously, data on domestic capital stock from Summers and Heston and Easterly were used. The labor input is proxied by the total labor force as reported by the World Bank's World Development Indicators. Regarding infrastructure stocks, physical indicators were used for the different infrastructure sectors. First, the number of telephone main lines served as a proxy for infrastructure in telecommunications. We comple- mented the data in Canning (1998) with recent figures from the Inter- national Telecommunications Union (ITU) annual reports. Second, the data on electricity generating capacity (in kilowatts) were taken from the United Nations Energy Statistics and Statistical Yearbook. Finally, data on road length (in kilometers) were used for the transportation sector. The data were obtained from the International Road Federation World Road Statistics. One caveat regarding these data, as noted by Canning (1999), is that they may exhibit significant variations in quality. In particular, they do not reflect the width of the roads or their condition. THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 115 Notes 1. Hong Kong (China), Indonesia, Republic of Korea, Malaysia, Taiwan (China), Thailand, and Singapore. 2. However, if one looks at total (main mobile) phone lines rather than just main lines, the relative performance of Latin America in the 1990s was worse than shown in the table--the gap with East Asia continued to expand in the 1990s, although at a slower pace than in the 1980s. 3. Canning and Bennathan (2000) and Demetriades and Mamuneas (2000) also presented estimates using translog specifications. 4. A similar procedure was followed by Canning and Bennathan (2000). 5. On this see Demetriades and Mamuneas (2000), who distinguished between the short run with noninfrastructure capital predetermined, and the long run, over which noninfrastructure capital adjusts to its optimal value. They also defined an intermediate run in which the capital stock partially adjusts to its equilibrium level. 6. However, empirical estimates using the entire sample are very similar to those using the reduced sample. 7. This accords with the finding of Barro and Sala-I-Martin (1995) that the growth contribution of secondary education is more significant than that of primary and higher education. For this study empirical specifications using broader definitions of human capital (inclusive of primary and/or tertiary schooling) yield more imprecise estimates of the contribution of human capital, and have only minimal effects on the coefficients of the physical capital and infrastructure variables. To save space, they are not reported. 8. Note that by including time and country effects in the empirical specifi- cation one can account for country-specific levels, as well as cross-country changes, in infrastructure quality--but not for country-specific quality changes. However, as noted by Esfahani and Ramirez (2002), using a panel data set similar to the one used in this study, standard infrastructure quality measures (such as power losses as percentage of power output or phone faults per telephone line) are strongly correlated with infrastructure quantity indica- tors. Hence the variation in the latter captures to a considerable extent the variation in the former as well. 9. In particular, one would need cross-country time-series data on the prices of infrastructure services, which are not available for a broad country sample such as the one considered here. The only example of such an approach in the recent literature is Röller and Waverman (2001), who devel- oped an empirical supply­demand model along the lines in the text but including only telecommunications infrastructure. The model is estimated using data for Organisation for Economic Co-operation and Development (OECD) economies. 10. These two variables are among the determinants of infrastructure de- mand in Esfahani and Ramirez (2002), and thus provide a source of identifi- cation in their empirical model. 11. Note that lagged levels of the variables on the right side are unlikely to provide valid instruments for the estimation of (3.5) because of the presence of time-invariant country-specific factors that may be correlated with the levels of the regressors at all lags. 116 THE LIMITS OF STABILIZATION 12. For lagged differences of a regressor x to provide a valid instrument for the levels equation, it is necessary that E[aixit] E[aixis] for all t and s. This is essentially a stationarity assumption (see Blundell and Bond 1998). 13. Railway data are unavailable for some 300 country-year observations. 14. Except for column one, all estimates reported in this and later tables include a full set of year dummies that were highly significant in all cases. 15. Various panel cointegration estimates were also computed, using the techniques of Kao and Chiang (2000) for nonstationary panels, with results very similar to the within estimates in table 3 (see also Baltagi 2000). These es- timates are subject to the same measurement error and simultaneity biases as the within estimator. They are not reported here to save space. 16. In anticipation of other experiments reported later, the instrument set includes also second lags of primary and tertiary schooling, total roads per worker, and total phone lines per worker. 17. The GMM estimates are not very different from those reported by Esfahani and Ramirez (2002), who found that the elasticities of output with respect to power generation capacity and telephone lines are, respectively, around 0.13­0.16 and 0.08­0.10. 18. It is also useful to compare these estimates with the results of Röller and Waverman (2001) for OECD countries. Their production function speci- fication ignores human capital and roads and power, does not impose constant returns, and employs a nonlinear transformation of the stock of phone lines. It can be shown that if the same transformation were used here, the resulting estimate of the elasticity of output with respect to phone lines would be very similar to that reported by Röller and Waverman. The elasticity with respect to physical capital, however, is much higher in their case (more than 0.50). References Arellano, Manuel, and Stephen Bond. 1991. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations." Review of Economic Studies 58: 277­97. Baltagi, Badi, ed. 2000. Nonstationary Panels, Panel Cointegration and Dynamic Panels. Amsterdam: Elsevier Science. Barro, Robert, and X. Sala-I-Martin. 1995. Economic Growth. New York: McGraw-Hill. Blundell, Richard, and Stephen Bond. 1998. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models." Journal of Economet- rics 87 (1): 115­43. Canning, David. 1999. "Infrastructure's Contribution to Aggregate Out- put." World Bank Policy Research Discussion Paper 2246. Washington, D.C. Canning, David, and Ezra Bennathan. 2000. "The Social Rate of Return on Infrastructure Investment." World Bank Policy Research Discussion Paper 2390. Washington, D.C. THE OUTPUT COST OF LATIN AMERICA'S INFRASTRUCTURE GAP 117 Demetriades, Panicos, and Theofanis Mamuneas. 2000. "Intertemporal Output and Employment Effects of Public Infrastructure Capital: Evidence from 12 OECD Economies." Economic Journal 110: 687­712. Esfahani, Hadi, and Maria Teresa Ramirez. 2002. "Institutions, Infra- structure and Economic Growth." Journal of Development Economics 70 (2): 443­77. Gramlich, Edward. 1994. "Infrastructure Investment: A Review Essay." Journal of Economic Literature 32: 1176­96. Griliches, Zui, and Jerry Hausman. 1986. "Errors in Variables in Panel Data." Journal of Econometrics 31: 93­118. Guasch, Jose Luis, and Joseph Kogan. 2001. "Inventories in Developing Countries: Levels and Determinants, a Red Flag on Competitiveness and Growth." World Bank Policy Research Discussion Paper 2552. Washington, D.C. Holtz-Eakin, Douglas. 1994. "Infrastructure in a Structural Model of Eco- nomic Growth." National Bureau of Economic Research Working Paper 4824. Cambridge, Mass. Kao, Chihwa, and Min-Hsien Chiang. 2000."On the Estimation and Infer- ence of a Cointegrated Regression in Panel Data." Advances in Econometrics 15. Röller, Lars-Hendrik, and Leonard Waverman. 2001. "Telecommunica- tions Infrastructure and Economic Development: A Simultaneous Approach." American Economic Review 91 (4): 909­23. Summers, Robert, and Alan Heston. 1991. "Penn World Table (Mark 5): An Expanded Set of International Comparisons, 1950­1988." Quarterly Jour- nal of Economics 106: 327­68. 4 Infrastructure Compression and Public Sector Solvency in Latin America César Calderón, William Easterly, and Luis Servén PUBLIC INVESTMENT AND INFRASTRUCTURE spending are often singled out for drastic cuts at times of fiscal retrenchment. Chapter 1 noted that this has been a common feature in episodes of fiscal adjustment in de- veloping countries, and chapter 2 highlighted its key role in Latin America's attempts to correct public sector imbalances over the last two decades. But the same phenomenon has been amply documented in in- dustrial economies. For example, out of a total of 32 episodes of sig- nificant budget consolidation in European Union (EU) countries over the period 1980­97, public investment fell relative to gross domestic product (GDP) in 25 cases, and in 23 of them investment fell by more than other primary spending. The fiscal targets of the Maastricht Treaty may have given new impetus to this practice. Eight EU countries that flunked the deficit criterion in 1992 had managed to meet it by 1997. All eight had lowered their public investment ratios, and seven of them had reduced investment more than other primary outlays.1 There are several reasons for this pattern of fiscal adjustment. It reflects, in part, a worldwide trend of increased reliance on markets and the private sector, along with reduced government involvement in pro- duction. In some cases it is also a reaction to the excessive expansion of public investment (including projects clearly identifiable as white ele- phants) during boom times. Also, there are admittedly compelling political economy reasons for this kind of adjustment--for example, 119 120 THE LIMITS OF STABILIZATION cutting investment in new roads or maintenance of existing ones is likely to entail much less political fallout than civil service downsizing. But fiscal adjustment centered on the compression of public infra- structure spending often reflects a flawed approach to the sustainabil- ity of public finances. This approach is concerned only with govern- ment liabilities and ignores the role of public sector assets--in other words, the flow of future public revenues (Buiter 1990). In such a framework, fiscal adjustment may mean little more than a parallel reduction in both liabilities (such as debt) and assets (such as infra- structure) of the public sector that leaves its net worth unaffected, or, even worse, reduces net worth if the rate of return on the assets (in- clusive of both their direct and indirect returns) exceeds the cost of the debt. Chapter 1 termed this kind of fiscal adjustment illusory. As chapter 2 documented, the period of fiscal austerity that most of Latin America underwent during the 1980s and 1990s was character- ized by a sharp contraction in infrastructure investment. In most cases recurrent infrastructure expenditures on operation and maintenance or O&M (for which cross-country data are unfortunately unavailable) were cut along with investment, so that the total decline in spending related to infrastructure was larger than the investment cut. But even ignoring this, the data cast doubt on the quality of the fiscal retrench- ment observed in several Latin American countries, given the adverse impact of persistent infrastructure compression on long-term growth documented in chapter 3. From the perspective of public sector solvency, the key issue is that fiscal adjustment biased against infrastructure accumulation can be largely self-defeating. As chapter 1 argued, the immediate effect of infrastructure spending cuts is to reduce the public deficit and, other things being equal, increase the public sector's net worth. But this is only the beginning of the story. Reduced infrastructure expen- ditures lead over time to a decline in infrastructure stock accumula- tion and, as shown in chapter 3, in output growth as well. This in turn implies a reduction in the economy's debt-servicing capacity, thereby weakening public sector solvency, as discussed in chapter 1. This adverse indirect impact on net worth via output growth can partly (or, under extreme conditions, even fully) offset the direct fa- vorable impact of infrastructure spending cuts, making the latter a very inefficient--even counterproductive--strategy to enhance pub- lic sector solvency. This chapter assesses quantitatively the growth cost of public infra- structure compression for major Latin American economies during the fiscal austerity period of the 1980s and 1990s, and examines the effects of infrastructure spending cuts on public sector net worth. INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 121 Thus, the chapter puts to work the analytical framework of chapter 1 using the empirical information on infrastructure and its contribution to growth presented in chapters 2 and 3, respectively. Two limitations of this analysis should be stated from the outset. First, the analysis intends to be illustrative rather than definitive. Its purpose is to provide an idea of the orders of magnitude of the factors shaping the solvency impact of infrastructure spending changes, and not to give the last word on their exact value. Second, because of the limited availability of infrastructure spending data, the analysis is lim- ited to the same nine Latin American countries that were the focus of much of chapter 2--Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Peru, and República Bolivariana de Venezuela.2 Framework The analytical approach follows Easterly (2001) and is based on a suit- ably modified version of the framework outlined in chapter 1. It is also closely related to the framework used by Buiter (1990, chapter 13), with the main difference that the focus here is on a growing economy rather than one approaching an equilibrium with constant output. The starting point is the public sector's budget identity describing the dynamics of public debt: · b(t) (r g)b(t) (t). Here b is the stock of public debt relative to GDP, r is the real interest rate, g is the rate of GDP growth (with both assumed constant for sim- plicity), and represents the augmented primary surplus of the public sector (that is, the noninterest budget surplus plus seigniorage revenues) as a ratio to GDP. From the dynamics of public debt it follows that t t b(t t) e( r g)t b(t) e( r g)(s t) s(s)ds. t Solvency means that the government cannot forever pay the interest on its outstanding debt simply by issuing more debt. Ultimately, the debt/GDP ratio will have to grow at a rate below the real interest rate minus the growth rate of real GDP. More precisely, what is required is lim e (r g)b(t t) 0; in other words, that the present dis- S counted value of the debt stock far into the future not be positive, with the discount rate given by the difference between the real interest rate 122 THE LIMITS OF STABILIZATION and the real growth rate.3 It is easy to see from the above expression that this is equivalent to requiring q (t) e (r g)(s t) (s)ds b(t) 0. (4.1) t In other words, net worth , defined as the present discounted value of the government's present and future stream of budget surpluses aug- mented for seigniorage, minus its stock of debt outstanding (all rela- tive to GDP), cannot be negative.4 This is just a restatement of expres- sion 1.2 in chapter 1. The augmented primary surplus can be further decomposed into seigniorage, infrastructure spending, and everything else. Take seigniorage revenue first. This can be expressed as h, where is the rate of growth of the stock of base money and h is the money stock/GDP ratio. Using this fact, the augmented primary surplus can be written (t) p(t) i(t) (t)h(t), where p is the primary surplus before infrastructure expenditures and i is the ratio of infrastructure spending to GDP. It is important to recognize that the noninfrastruc- ture primary surplus as a proportion of GDP could itself depend on the growth rate of the economy: other things equal, faster growth rates might imply larger surpluses (or smaller deficits) through a rising tax/GDP ratio or a declining expenditure/GDP ratio; hence, in principle, p p(g, .). In turn, the money/GDP ratio should depend basically on the nominal interest rate; that is, letting denote the inflation rate, h h(r ), with h 0. Consider a long-run equilibrium in which the money/GDP ratio, the noninfrastructure primary surplus, and the ratio of infrastructure spending to GDP all remain constant. For the money stock to remain constant relative to GDP, it must be the case that g , that is, the rate of money growth must equal the rate of growth of nominal GDP. In such conditions (4.1) can be further simplified to p(g, .) i (g p)h(r p) b. (4.2) r g Taking r, , and the initial debt/GDP ratio as given, the impact of a change in infrastructure spending on net worth is 1 0 dg d Tdi. (4.3) r g 0g di b b Direct effect Indirect effect INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 123 This expression highlights the two ingredients mentioned earlier: the direct effect via the infrastructure spending component of the pri- mary surplus and the indirect effect arising from the impact of infra- structure accumulation on growth. The direct effect is unambiguously negative, implying that it makes infrastructure spending and net worth move in opposite directions. In turn, the indirect effect via growth is likely to be positive. Inspection of (4.2) shows that the indirect effect works through three channels: first, by affecting the level of the non- infrastructure component of the primary deficit p; second, by chang- ing the ratio of seigniorage revenue to GDP; and, third, by altering the present value of a given stream of augmented primary deficits through the term 1/(r g), along the lines described by Easterly (2001) and already mentioned in chapter 1. This expression can be simplified further by noting that the growth impact of infrastructure spending can be expressed as the growth con- tribution of infrastructure stock accumulation (analyzed in chapter 3) times the impact of infrastructure spending on stock accumulation (examined in chapter 2): dg dg d¢z d¢z hz (4.4) di d¢z di di where z is the rate of growth of infrastructure stocks, and hz dg d z is the growth contribution of infrastructure stock accumulation. In turn, from (4.2), the impact of growth on net worth, holding the noninfrastructure primary surplus constant and near a point where net worth is small, can be written as 0 b h ` . (4.5) 0g 0 r g Thus, the impact of growth on net worth is positive and proportional to the initial stocks of debt and money. For the debt stock, this has already been emphasized by Easterly (2001). The intuition is that an additional percentage point of growth reduces the amount of fiscal adjustment needed for solvency more in a high-debt country than in a low-debt country, and more so the smaller the net discount factor r g. As for money, the argument is similar: higher growth allows larger seigniorage revenue collection, and the present value of those extra revenues is larger the greater the money/GDP ratio and the smaller the net discount factor. Putting all these pieces together, the effect of infrastructure spend- ing changes on net worth can be expressed as d 1 0p d z c 1 ab h b hz d (4.6) di r g 0g di 124 THE LIMITS OF STABILIZATION The term in square brackets in the right-hand side of this expression can be interpreted as the impact of infrastructure spending on the annuity value of public net worth relative to GDP, and the factor (r g)­1 serves to bring the annuity to present-value terms.5 This latter expression shows how the direct contribution of infra- structure spending cuts to raising net worth is offset by adverse growth effects, and identifies what factors determine the magnitude of such offset. Thus, the offset is larger if debt and money ratios to GDP are high, if infrastructure makes a large contribution to output, and if in- frastructure asset accumulation closely tracks infrastructure spending. Thus, the actual extent of this offset is an empirical matter. Assessing its magnitude requires data on debt and base money ratios and empirical counterparts for d z 0p z, di, and 0g. The first of these expressions pro- vides the link between infrastructure stock accumulation and growth; the second ties together infrastructure spending and stock accumula- tion; and the third captures the impact of growth on the noninfra- structure primary surplus. They are examined in turn. Empirical Implementation Take first the link between infrastructure stocks and growth. This was examined in chapter 3, which presented empirical estimates of zfor various infrastructure assets in an aggregate production function framework, using a large cross-country time-series data set and employ- ing a variety of econometric specifications. In the vast majority of cases, those estimates showed positive and significant contributions to aggregate output of all infrastructure assets considered. For the purposes of this chapter, the estimates of z reported in table 3.6 will be used. Consider next the link between public infrastructure spending and the time path of infrastructure stocks: d z di in (4.6) above. In theory, stock accumulation should track spending (especially investment) closely, but in reality variation in the quality and cost of assets across countries and over time can make the link much more tenuous.6 Perhaps as a result of this, there have been very few assessments of the spending-accumulation link, especially in a cross-country (not to men- tion multi-asset) framework. One rare exception is the recent work by Röller and Waverman (2001), who explored the effects of investment in telecommunications on phone line density in industrial countries. Chapter 2 above presented a preliminary quantification of the link between infrastructure spending and asset accumulation for the nine Latin American countries--Argentina, Bolivia, Brazil, Chile, Colombia, INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 125 Ecuador, Mexico, Peru, and República Bolivariana de Venezuela--for which the necessary information disaggregated by type of asset (in other words, transport networks [inclusive of roads and railways], power, and telecommunications) could be collected. Note the caveat that the analysis included only investment spending and not other rel- evant expenditures (such as O&M) that may also affect the evolution of the quantity and quality of stocks over time.7 In spite of this limitation, the regression results reported in chapter 2 using a variety of specifications reveal a highly significant association between infrastructure investment and the ensuing trajectory of infra- structure assets. Variation of the former across countries and over time accounts for a considerable portion of the observed variation in the latter, which is particularly high in the case of telecommunications and transport routes. Furthermore, the results are robust to the use of alternative definitions of the relevant asset stocks--total instead of main phone lines, or roads alone instead of roads plus railways. Thus, for the analysis in this chapter, the estimates of the long-run ef- fect of investment on asset accumulation derived from those regressions, and reported in table 2.6, are taken as the proper measure of d z di Specifically, the calculations below use the estimates obtained from the third specification in table 2.6.8 The final ingredient required for empirical implementation of the analytical framework outlined in the previous section is the response of the primary surplus, exclusive of infrastructure spending, to changes in the growth rate of GDP. On this point, the automatic stabilizer view of fiscal policy suggests that revenue and spending ratios should both be affected by changes in the economy's growth rate over the cycle-- the former positively and the latter negatively. However, the auto- matic stabilizer function of fiscal policy is known to be weak in de- veloping economies in general, and Latin America is no exception to this rule (Talvi and Vegh 2000). Furthermore, the present analysis is concerned more with long- than with short-term growth, and on this front theoretical predictions regarding the response of fiscal rev- enues and expenditures to changes in growth are much less clear. For these reasons, the assessment of the impact of growth on the noninfrastructure primary deficit offered below is based on regressions of public revenue and spending ratios on the growth rate of GDP using data for 1970­97 for a group of Latin American economies defined by data availability.9 The results are reported in tables 4.1 and 4.2. In each case, a number of panel estimates were computed, variously including or excluding country fixed effects and time dummies in the regression specification. These are intended to control, respectively, for unobserved country-specific factors and for common factors influencing 126 THE LIMITS OF STABILIZATION public revenue and expenditure across countries. Other experiments were also performed allowing for dynamics in the impact of growth on revenue and expenditure ratios, but they are not reported to save space.10 Table 4.1 presents estimation results for tax revenues and total pub- lic revenues as a ratio to GDP. In addition to growth, the regressions also include the tax reform index of Morley, Machado, and Pettinato (1999) as a determinant of public revenues. The regression sample is Table 4.1 Taxes and Growth: Panel Data Regression Analysis Specification and Dependent variable variable Tax revenues (% GDP) Total revenues (% GDP) I. OLS Output growth 0.009 0.005 (0.045) (0.068) Tax reform 0.0153 0.101 (0.009)** (0.029)** R2 0.051 0.062 II. Within-group estimator Output growth 0.066 0.059 (0.029)** (0.042) Tax reform 0.014 0.107 (0.008)** (0.032)** R2 0.239 0.167 III. OLS with time effects Output growth 0.007 ­0.008 (0.050) (0.074) Tax reform 0.017 0.092 (0.012) (0.028)** R2 0.031 0.078 IV. OLS with country and time effects Output growth 0.084 0.068 (0.031)** (0.044) Tax reform 0.017 0.082 (0.013) (0.037)** R2 0.342 0.271 Note: The sample covers the years 1970­97. The countries included are Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Guyana, Honduras, Haiti, Jamaica, Mexico, Nicaragua, Panama, Peru, Paraguay, El Salvador, Trinidad and Tobago, Uruguay, and Venezuela, R.B. de. The number of observations in each panel is 425. The figures in parentheses are standard errors. OLS ordinary least-squares. ** Significant at 5 percent. Source: Authors' calculations. INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 127 Table 4.2 Government Spending and Growth: Panel Data Regression Analysis Sample Specification and variable All countries Latin America I. OLS Output growth ­0.073 ­0.074 (0.054) (0.067) R2 0.021 0.027 II. Within-group estimator Output growth ­0.090 ­0.066 (0.035)** (0.045) R2 0.071 0.044 III. OLS with time effects Output growth ­0.049 ­0.033 (0.054) (0.075) R2 0.058 0.058 IV. OLS with country and time effects Output growth ­0.065 ­0.021 (0.034)** (0.047) R2 0.148 0.122 Note: Dependent variable is government spending as a ratio to GDP. The sample includes 60 countries over the 1960­97 period (1,620 observations), of which 20 countries are from Latin America (540 observations). The sample of countries is Argentina; Australia; Bangladesh; Belize; Benin; Bolivia; Brazil; Burundi; Chile; China; Colombia; Costa Rica; Côte d'Ivoire; Dominican Republic; Ecuador; Egypt; El Salvador; Ethiopia; Fiji; Gabon; Gambia, The; Great Britain; Greece; Guatemala; Haiti; Honduras; India; Indonesia; Japan; Kenya; Korea, Rep. of; Malawi; Malaysia; Malta; Mauritania; Mauritius; Mexico; Morocco; Nicaragua; Nigeria; Pakistan; Panama; Papua New Guinea; Paraguay; Peru; Philippines; Singapore; South Africa; Sri Lanka; Sudan; Sweden; Syria; Thailand; Tunisia; Turkey; Uruguay; United States; Venezuela, R.B. de; Zaire; and Zimbabwe. The figures in parentheses are standard errors. ** Significant at 5 percent. Source: Authors' calculations. limited to Latin America because the tax reform index is unavailable for other countries. For both tax and total revenues, the estimates reveal a positive effect of tax reforms, as measured by the reform index, on the revenue/GDP ratio. However, for tax revenues the impact is significant only when time dummies are excluded. As for GDP growth, which is 128 THE LIMITS OF STABILIZATION the variable of interest here, its effect is always positive for tax rev- enues and also for total revenues, except in the regression including only time effects. The impact of growth is generally insignificant, how- ever. The exception to this rule is provided by the regressions of tax revenues including fixed effects, which exhibit a positive and signifi- cant growth coefficient. For total revenues, the growth coefficient is never significant. For spending (table 4.2), results are shown for both Latin America and a broader country sample. The growth coefficient estimates are uniformly negative, as could be expected from the automatic stabilizer view of fiscal policy, but they are significant only for the broader sam- ple and only if fixed effects are included. For Latin America, the estimates are insignificant in every specification. On the whole, therefore, both the revenue and expenditure esti- mates in tables 4.1 and 4.2 provide little evidence of any major effects of growth on the noninfrastructure primary deficit. Thus, for practical purposes the calculations below shall take 0p 0g 0. The Impact of Infrastructure Spending on Public Sector Net Worth It is now possible to put together the different pieces developed in the preceding analysis and illustrate the impact of government infrastruc- ture spending on the public sector's net worth. To do this, it is convenient to focus on the effects of spending on the annuity value of net worth introduced earlier. From (4.6) and using 0p 0g 0, this can be expressed as da d z (r g)d 1 c(b h) hz d. (4.7) di 0i di The term in square brackets is the indirect effect via growth from (4.3) above. As already noted, it tends to offset the direct impact of infrastructure spending changes on the annuity value of public sector net worth, which is itself negative and equal to minus one. Using the empirical estimates just discussed, the extent of this offset is computed in table 4.3, which calculates the impact on the annuity value of public net worth of a permanent cut in spending on each of the three infrastructure assets considered--that is, the right side of (4.7). The calculation is presented for different values of the debt/GDP ratio. It is important to stress once again that these computations are INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 129 Table 4.3 Impact on the Annuity Value of Net Worth of a Cut in Infrastructure Investment by 1 Percent of GDP (percent of GDP) Initial Cut in investment in public Initial base debt/GDP money/GDP Power generation (percent) (percent) Telecommunications capacity Transport routes 0 0 1.00 1.00 1.00 10 10 0.78 0.79 0.79 30 10 0.56 0.58 0.57 50 10 0.35 0.38 0.36 70 10 0.13 0.17 0.15 Note: For each value of the debt/GDP ratio, the table shows the impact on annual- ized net worth, as percentage of GDP, of a decline in investment in each infrastructure asset by 1 percent of GDP. Source: Authors' calculations. based on a very simple framework and rely on first-order approxima- tions that admittedly may be very rough. Thus, the calculations should be viewed as illustrative. Subject to these caveats, the first row of the table shows that with a zero public debt stock and a zero base money stock (equivalently, ignoring seigniorage) an infrastructure spending cut translates one- for-one into increased net worth. The reason is that at zero debt and seigniorage, the reduced growth resulting from slower infrastructure expansion has no (first-order) effect on the economy's sustainable debt stock. Thus the growth slowdown is of no consequence for pub- lic solvency. As the debt stock rises, however, the table shows that a consider- able portion of the favorable impact of spending cuts on public sector net worth is offset by the solvency-weakening effect of reduced growth. For these illustrative calculations in the table, the stock of base money is set at 10 percent of GDP. When the public debt stock equals 10 percent of GDP, a cut in public telecommunications investment by 1 percent of GDP raises the annuity value of public net worth by only 0.78 percent of GDP--in other words, 22 percent of the spending cut is offset by future reduced growth. The offset is numerically similar regardless of which of the three assets consid- ered--transport routes, power, or telecom--is the object of the spend- ing cut. At higher levels of public indebtedness, the offset is much larger. For example, when the public debt/GDP ratio reaches 70 percent (and the base money stock still remains at 10 percent of GDP), the estimates 130 THE LIMITS OF STABILIZATION imply that a cut in infrastructure investment by 1 percent of GDP raises the annuity value of net worth by only a small amount--be- tween 0.13 and 0.17 percent of GDP, depending on the asset compo- sition of the spending cut.11 Given this assessment of the impact of public infrastructure invest- ment on public sector net worth, one may ask to what extent Latin America's public infrastructure compression of the 1980s and 1990s contributed to stronger public finances. The empirical estimates allow, again, an illustrative, if not conclusive, answer to this question. It is im- portant to note that the numerical illustration below assumes that changes in public infrastructure investment are translated one-for-one into changes in total infrastructure investment--in other words, private investment remains unaffected. This is obviously not what was ob- served in practice, and in this sense the experiments conducted here re- flect a partial equilibrium, before the adjustment of private investment. With this important qualification, table 4.4 provides a preliminary assessment of the solvency impact of the observed changes in public in- frastructure investment. The table repeats the generic calculations in table 4.3, but uses the actual debt and base money ratios and infra- structure investment changes observed in the nine Latin American countries under consideration between the early 1980s and late 1990s. The first column in the table reports the total change in public in- vestment in the three infrastructure assets under analysis over the pe- riod in question. All the countries listed, except Ecuador, witnessed an investment decline, most markedly Argentina, Bolivia, and Brazil.12 The second column calculates the impact of those spending cuts on annual GDP growth. The impact is computed by adding up the indi- vidual growth effects of the observed changes in public investment in each of the infrastructure assets considered, with the individual calcu- lations based on the parameter estimates of the output contribution of each asset. Again it is important to emphasize that these calculations assume that changes in public investment translate fully into changes in aggregate infrastructure investment. Hence they reflect the partial- equilibrium growth impact of public sector retrenchment, before changes in private infrastructure investment. The adverse growth impact obtained in this manner is considerable for Argentina, Bolivia, and Brazil, where the estimated GDP growth cost is 3 percent a year. It is also significant for Chile, Mexico, and Peru (around 1.5 to 2 percent a year). At the other end of the spectrum, the adverse growth impact is small in Colombia and República Boli- variana de Venezuela, which experienced only small investment cuts. Finally, the growth effect is positive in Ecuador, which increased pub- lic infrastructure spending over the period under consideration. of Offset ([2]/[1]) 28.45 24.65 coefficient (percent) ­ 91.85 25.54 48.45 21.40 54.84 48.91 54.57 Value 1 worth GDP) Annuity change of net [2] 98.­ 1.90 0.25 2.22 0.77 0.34 the annualized 0.25­ 0.99 0.67 0.31 1995 on Implied in public (percent and 84­ d 1980 Changes ratio money 4.00 6.14 2.30 3.19 4.60 3.42 3.79 3.99 3.55 GDP Base (percent) between to period. 94­ Investment 1980 c GDP the to ratio 23.48 83.00 22.25 42.55 21.28 50.67 42.21 48.74 21.03 telecommunications over (percent) Debt and Infrastructure averaged in b power, rate Actual 98­ ratios Implied 2.75­ 3.13­ 3.10­ 1.58­ 0.36­ 0.56 2.07­ 1.53­ 0.41­ routes, 3.6. of change growth (percent) 1995 and a 2.6 debt/GDP transport Effect in versus GDP. public GDP) tables public to 84­ in of in [1] 2.65­ 3.03­ 2.98­ 1.49­ 0.44­ 0.56 1.95­ 1.48­ 0.41­ investment investment external relative infrastructure 1980 Change (percent estimates and public the banks in de calculations.' using domestic Partial-Equilibrium Worth, outside R.B. change the Authors 4.4 Net of Actual a. Calculated.b Sum.c Currency.d Source: Table Public Country Argentina Bolivia Brazil Chile Colombia Ecuador Mexico Peru Venezuela, 131 132 THE LIMITS OF STABILIZATION The next two columns in the table show each country's debt and money ratios to GDP. Because in several of the countries considered bank reserves earned interest during at least part of the sample period, the money ratio reported corresponds only to currency outside banks, which always earns no interest and is therefore closer in spirit to the concept of base money employed in the analytical model above. Column five in table 4.4 presents the impact of these public spend- ing changes on the annuity value of public sector net worth. The sign of the impact is positive for all countries showing spending cuts, and negative for the only one showing an increase (Ecuador). However, the "bang-per-buck" varies considerably across countries. This is prima- rily because of their different levels of public indebtedness, and mar- ginally because of the different composition of the investment cuts by infrastructure asset (not shown in the table) observed in each country. Thus, in highly indebted Bolivia, a cut in public infrastructure spending by 3 percent of GDP raises the annualized net worth of the public sector by only 0.25 percent of GDP, whereas a similar spending cut in lower-debt Brazil yields a net worth increase of more than 2 per- cent of GDP. The last column in the table gives an idea of the efficiency of these infrastructure spending cuts as a device to raise public net worth. It reports the fraction of the spending cuts that was not reflected in an increase in annualized net worth--in other words, the overall offset coefficient. The offset is largest in Bolivia, where it exceeds 90 percent. In the other countries the offset coefficient ranges between 20 percent and 55 percent--that is, between 20 percent and 55 percent of the observed infrastructure investment cuts failed to improve the public sector's financial position. These offset coefficients strongly suggest that in most countries infrastructure investment cuts represent a very inefficient strategy for strengthening public finances. One important caveat to these calculations is that by equating cuts in public infrastructure investment with cuts in total infrastructure in- vestment--in other words, ignoring the private sector response to the public sector's retrenchment--the calculations lead to an overstate- ment of the growth reduction caused by public spending cuts and hence to an overstatement of the offset coefficients. It is true that in some countries--with Chile as the leading example--private infra- structure investment did expand considerably as public investment contracted, dampening (or even reversing) the adverse impact of pub- lic spending cuts on the accumulation of infrastructure assets. Hence, to the extent that the decline in total infrastructure investment was typ- ically less pronounced than the decline in public infrastructure invest- ment (or, to put it differently, that infrastructure asset accumulation INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 133 declined less than proportionately with the public investment contrac- tion), the calculations above provide an upper bound on the adverse growth implications and thus the inefficiency of public infrastructure investment cuts as a means of enhancing public solvency. However, the infrastructure investment data for Latin America do not support the simplistic view that public investment cuts are auto- matically offset by private investment rises. The evidence shows con- siderable diversity across countries and infrastructure sectors in the re- gion in terms of private sector response.13 In other words, public sector retrenchment per se does not lead to a private investment takeoff. Other ingredients, such as an appropriate regulatory and institutional environment, are necessary to encourage private sector involvement in infrastructure activities.14 In this sense, the above calculations under- score the dangers posed by public infrastructure compression for growth and public finances when those necessary ingredients, and thus the private sector response, are lacking. Summary Public infrastructure spending often takes a major hit at times of fiscal contraction. The experience of Latin America over the last two decades accords with this observation. In several of the region's large economies, infrastructure investment cuts accounted for half or more of the reduction in the primary deficit achieved between the early 1980s and the late 1990s. Moreover, this figure probably understates the total contribution of infrastructure spending cuts, given that infra- structure O&M expenditures likely fell in most countries along with investment. The analysis in this chapter has shown that fiscal adjustment through public infrastructure compression can be largely self-defeating in the long run, because of its adverse effect on growth and hence on the debt-servicing capacity of the public sector. The calculations re- ported here show that the growth cost of reduced infrastructure asset accumulation resulting from lower investment was substantial--in several countries, the estimated adverse impact on the long-run growth rate of GDP exceeds 1 percent a year. As a result, much of the sup- posedly favorable effect of the investment cuts on public finances was likely offset by higher future deficits resulting from reduced future out- put, although the magnitude of the growth cost and the fiscal offset varies considerably across countries, depending on their levels of pub- lic indebtedness and the asset composition of the infrastructure invest- ment contraction.15 134 THE LIMITS OF STABILIZATION The main implication of these results is not that infrastructure spending should never be cut under any circumstances. The lesson in- stead is that under realistic circumstances infrastructure compression may represent a highly inefficient way to achieve fiscal adjustment. Its consequences for future growth and public revenues should be care- fully considered and assessed against those of cuts in other spending items, when deciding on a course of action for fiscal retrenchment. Appendix 4A Testing for Unit Roots in Public Revenues and Public Expenditures As a preliminary step for the revenue and expenditure regressions in the text, we assess the time-series properties of the different measures of government revenues and spending, as well as real output. We apply panel unit root techniques developed by Im, Pesaran, and Shin (1995). They jointly tested the null hypothesis that every time series in the panel is nonstationary. The approach consists in running augmented Dickey-Fuller (ADF) unit root tests for each country, and averaging the t-values of the test statistics found. If the data from each country are statistically independent, then, under the null, the average t-value approximates the average of independent random draws from a distri- bution with known expected value and variance (that is, those for a non- stationary series). This provides a much more powerful test of the unit root hypothesis than the usual single time-series test (Im, Pesaran, and Shin 1995). Before carrying out the ADF regressions, we remove any common time effect. Hence, we regress the variable on a set of time dummies and take the residuals. This reduces the risk of correlation across coun- tries. In each case, the ADF regressions using those residuals are run with a constant, a deterministic trend, and five augmenting lags. (A) Government Revenues. We use data on government revenues and real GDP for the Latin American countries that have a complete data set for 1970­95, that is, 17 countries and 26 observations per country. From the results reported in table 4A.1, we cannot reject the existence of a unit root for all our variables in levels. However, we reject the unit root hypothesis for the first differences. Finally, we can also reject the unit root hypothesis when expressing the revenue measures as ratio to GDP. We use the latter specification in the regressions. INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 135 Table 4A.1 Panel Unit Root Tests, Government Revenues, Government Spending, and Real Output Levels First differences Variable Without trend With trend Without trend With trend A. Government revenues and real output, 1970­95 (annual) Real output (in logs) ­1.38 ­1.57 ­2.04** ­2.51** Tax revenue (in logs) ­1.14 ­1.70 ­2.03** ­2.51** Current revenue (in logs) ­1.35 ­2.04 ­2.10* ­2.50** Tax revenue/GDP ­1.74 ­2.59** ­2.36** ­2.79** Current revenue/GDP ­1.79 ­2.70** ­2.42** ­2.84** B. Government spending and real output, 1970­97 (annual) Real output (in logs) ­1.19 ­2.10 ­2.13** ­2.43* Government spending (in logs) ­1.40 ­2.23 ­2.14** ­2.45** Government spending/GDP ­1.71 ­2.41** ­2.43** ­2.50** Note: The table reports the t-bar (t¯NT) statistic, defined as the sample average of the t-statistics obtained from the ADF regressions of individual countries. Before per- forming the ADF regressions for individual countries, we remove the common time dummies from all variables. For the critical values of the t¯NT statistic, see table 4 in Im, Pesaran, and Shin (1995). * Significant at 10 percent. ** Significant at 5 percent. Source: Authors' calculations. (B) Government Spending. Using data on government spending and real output for a sample of 60 countries for 1970­97, we test the sta- tionarity of both series (in logs). In table 4A.1, we show that the series are nonstationary in levels and stationary in differences, that is, they are I(1) processes. We next express spending as a ratio to GDP, and find that we can reject the presence of a unit root. We use the latter specification in the regressions. Notes 1. The performance of public investment during episodes of fiscal adjustment in European countries is examined at length by Balassone and Franco (2000). 2. The sources of the data used in this chapter are listed in appendix 2A. Be- cause much of infrastructure spending is often done by lower levels of government or by public enterprises, it is important to base the analysis on infrastructure spending data for a broadly defined public sector. These data could be collected only for the countries listed in the text. An alternative would 136 THE LIMITS OF STABILIZATION be to use the IMF's Government Finance Statistics, as done, for example, by Jonakin and Stephens (1999), which offer much broader country coverage. However, that source covers only the central government of the countries con- cerned, and thus provides a very limited view of public infrastructure spending. 3. As already noted in chapter 1, if g r, so that the discount rate is neg- ative, the economy is dynamically inefficient and any debt stock, no matter how large, is consistent with solvency. 4. This is not the only possible definition of government net worth, but is a convenient one for the purposes of the discussion in this chapter. 5. Let the annuity value a (r g) . hen da d di 0 (r g)di, and the latter expression is just the term in square brackets in (4.6). 6. See Pritchett (2000) for a discussion of this point. 7. O&M data are notoriously difficult to obtain on a comprehensive or even comparable basis across countries. This data limitation is also shared by the study of Röller and Waverman (2001) cited in the text. 8. Observe that the regressions in question relate stock accumulation to to- tal investment, implicitly assuming that the contribution of public infrastruc- ture investment to the accumulation of infrastructure assets is identical to that of private investment. This assumption was tested in chapter 2, and the results of the tests were reported in table 2.7. Although in power and telecommuni- cations there is no evidence against the hypothesis that public and private in- vestment contribute equally to asset accumulation, for transport routes the re- sults suggest that the contributions of public and private investment do differ. For simplicity, this divergence is ignored here, and therefore the calculations below have to be taken with some caution. 9. Because GDP growth is a stationary variable, it is necessary to check first that the revenue and spending ratios are stationary as well--otherwise the regression just described will yield inconsistent parameter estimates. This is done following a three-stage procedure, described in detail in appendix 4A. The first stage verifies that revenues, expenditures, and GDP are I(1) variables, using the panel unit root test of Im, Pesaran, and Shin (1995). The test statis- tics cannot reject the null of a unit root for any of the three variables. Next, the same methods are used to test whether revenue and expenditure ratios to GDP contain unit roots. In all cases the presence of unit roots can be rejected once a deterministic trend is included. This allows the use of standard estima- tion methods as described in the text. 10. The regression results reported in the tables change very little with the addition of lags of the dependent and independent variables. 11. At even higher debt stocks, the offset could become more than full, and the spending cut would actually reduce the public sector's net worth. This sit- uation is similar to the one explored by Buiter (1990, chapter 13), in which public investment cuts lead to lower output and taxes in the long run and thus require higher inflation to balance the fiscal accounts via seigniorage. 12. Note that these figures differ somewhat from those reported in chapter 2. The reason is that table 4.4 considers only investment in trans- port routes, power, and telecommunications, whereas the data shown in chapter 2 include, in addition, investment in other items such as water and gas. 13. See chapter 2 for further details. 14. This is discussed in chapter 6. 15. The calculations presented here are subject to a number of caveats. They gloss over possible heterogeneity in the cost and/or quality of infrastructure INFRASTRUCTURE COMPRESSION AND PUBLIC SECTOR SOLVENCY 137 assets across countries and over time. They also reflect a partial-equilibrium view before any private sector investment response. For this reason, the results reported in this chapter have to be taken as a preliminary illustration rather than a definitive assessment. References Balassone, Fabrizio, and Danielle Franco. 2000. "Public investment, the Stability Pact and the `Golden Rule.'" Fiscal Studies 21 (2): 207­29. Buiter, Willem. 1990. Principles of Budgetary and Financial Policy. Cambridge, Mass.: MIT Press. Easterly, William. 1999. "When Is Fiscal Adjustment an Illusion?" Economic Policy 28: 57­86. ------. 2001. "Growth Implosions, Debt Explosions, and My Aunt Marilyn: Do Growth Slowdowns Cause Public Debt Crises?" World Bank Pol- icy Research Working Paper. Washington, D.C. Hicks, Norman L. 1991. "Expenditure Reductions in Developing Coun- tries Revisited." Journal of International Development 3 (1): 29­37. Im, Kyung So, M. Hashem Pesaran, and Yongcheol Shin. 1995. "Testing for Unit Roots in Heterogeneous Panels." Manuscript. University of Cambridge (revised version: December 1997). IMF (International Monetary Fund). Various years. Government Finance Statistics. Washington, D.C. Jonakin, Jon, and Mark Stephens. 1999. "The Impact of Adjustment and Stabilization Policies on Infrastructure Spending in Latin America." North American Journal of Economics and Finance 10: 293­308. Morley, Samuel A., Roberto Machado, and Stefano Pettinato. 1999. "Indexes of Structural Reform in Latin America." CEPAL Serie Reformas Economicas 12, January. Pritchett, Lant. 2000. "The Tyranny of Concepts: CUDIE (Cumulated, De- preciated Investment Effort) Is Not Capital." Journal of Economic Growth 5 (4): 361­84. Röller, L., and L. Waverman. 2001. "Telecommunications Infrastructure and Economic Development: A Simultaneous Approach." American Eco- nomic Review 91: 909­23. Talvi, Ernesto, and Carlos Vegh. 2000. "Tax Base Variability and Pro- cyclical Fiscal Policy." National Bureau of Economic Research Working Paper 7499. Cambridge, Mass. 5 Macroeconomic Effects of Private Sector Participation in Infrastructure Javier Campos, Antonio Estache, Noelia Martín, and Lourdes Trujillo THE RELEVANCE OF THE DESIGN OF institutions for the effectiveness of poli- cies is now well recognized by policymakers (see World Bank 2002b for a recent survey of the evidence). This design becomes particularly impor- tant where reforms have significantly changed the types and roles of play- ers. The restructuring of the infrastructure sector to increase competition and private sector participation in Latin America over the last decade provides a clear example of such an institutional change. Since the late 1980s, many Latin American countries have progressively opened their infrastructure sectors to private operators, seeking a remedy to structural deficits and hoping to foster investment and growth. The literature on the impact of these reforms can be classified into three broad types. The first type focuses mostly on the macroeconomic effects of a macroeconomic view of the reforms--the macro­macro group. The second examines the sector-specific effects of sector- specific reforms--the micro­micro group, and the third, the macro- economic effects of sector-specific reforms--the micro­macro group. The macro­macro category is by far the most populated. The field has been able to generate fairly detailed econometric analyses from the relatively good macroeconomic databases available (see McGillivray and Morrissey 1999 for a recent survey). The micro­micro group has generated fewer analytically strong studies, partially because detailed relevant data are not easy to obtain. 139 140 THE LIMITS OF STABILIZATION Most of the published articles have focused on Argentina and Chile, where enough time has gone by to generate reasonable time-series data (see Guasch 2001 for an overview). A much more modest literature has underpinned the third category--the micro­macro group that examines the macroeconomic effects of sector-specific reforms--even if these sectoral reforms have been key components of the overall macroeconomic restructuring agenda. Exceptions include the literature on the general equilibrium effects of reform or the literature on con- vergence (see De la Fuente 2000; or Estache, Foster, and Wodon 2002). This chapter contributes to the micro­macro literature by offering a first empirical assessment of the macroeconomic effects of increased private sector participation in the management and financing of the infrastructure sectors (PPI, or private participation in infrastructure) in Latin America.1 The chapter's main purpose is to provide empirical evidence of the effects on several key macroeconomic variables of the increased role of privatization, defined as the decision to rely on the pri- vate sector to implement projects.2 In this analysis some institutional factors are isolated and country effects are controlled for, recognizing that each country in the region may face different sources of risks. Given these objectives, the chapter suffers from at least two major drawbacks. First, the quality of the data available is a significantly re- strictive factor and limits the possibility of drawing very strong policy conclusions. This drawback is, however, also a source of strength be- cause it highlights the main direction for additional analytical work. Second, the chapter lacks an explicit theoretical model to justify a find- ing that increased PPI should have specified micro­macro effects.3 However, the goal here is not to test any specific theory but rather to provide, if possible, statistically significant evidence on the sign (or di- rection) and size of the effects of PPI on the most common macroeco- nomic indicators. In this chapter, general empirical relationships are specified be- tween each of the macroeconomic variables and different subsets of in- struments that summarize when and how private participation was in- troduced in each country and under which institutional environment. The results cannot be interpreted as causal associations; they represent correlations that can only hint at what the macroeconomic impact of privatization (if any) has been so far in Latin America. In spite of these limitations, useful results were obtained relying on standard econometric techniques. First, pooled data models were esti- mated ignoring country-specific effects. These models provide both initial values for the micro­macro effects and a benchmark for com- parisons. However, if unobserved individual heterogeneity (that is, country-specific effects) is relevant in the statistical relationships, its MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 141 omission yields biased estimates. To overcome this problem, panel data models that allow for an explicit testing of individual hetero- geneity were also estimated.4 The differentiation between these two types of model specifications yields evidence on the effects of the pri- vatization policies for the region as a whole as well as for average country-specific effects. The remaining sections of the chapter are organized as follows. In the next section, the methodology followed to draw a minimum set of robust policy implications on the effects of PPI is discussed. The sam- ple and the most relevant variables used are then described. The main results of the impact of private participation policies on each macro- economic variable are given. Finally, a summary of the main empirical implications is presented. Testing the Macroeconomic Effects of PPI There is no simple way to anticipate the overall macroeconomic effects of a policy opening infrastructure to the private sector in a particular country because many tradeoffs are at stake. The best that can be done with the kind of data available is to focus on reduced forms that net out structural positive and negative effects of the reforms on the key macroeconomic variables, which cannot be separated out in the usual way because of lack of data. From the viewpoint of private investment, for instance, many privatization policies are expected to bring about positive results in the medium term or in the long run if the overall efficiency of the economy is improved as a result of the policy changes. However, the long-run payoffs may be preceded by short-run costs if increased competition reduces margins and profits and thus hampers the investment capabilities of private investors. Because the sectors can be studied only at a very aggregate level and sector-specific reforms are difficult to pick up, only the accumulated effects of the policy changes year after year can be identified and these effects cannot be assigned to any specific policy change. That is why the focus of this chapter is on a limited concern that has not been stud- ied so far. The focus is on identifying the outcomes that genuinely can be attributed to the net effects of private sector participation in infra- structure projects. The size of the outcomes is also computed but is for now less interesting because it probably represents a large number of offsetting effects. With these limitations in mind, the chapter proposes a formal test of the consequences of infrastructure privatization on four selected macroeconomic variables: total gross domestic product (GDP) per 142 THE LIMITS OF STABILIZATION capita, private investment, public investment, and current public ex- penditures. The first dependent variable is measured in levels. By fo- cusing on per capita figures, it is possible to get a modest look at the impact on poverty through income levels. The other three are calcu- lated as a percentage of the GDP. Within infrastructure, the distinction is made between utilities (electricity, gas, water, sanitation, and telecommunications) and transportation (airports, ports, railways, and roads) to test for possible differences.5 Because the timing of the changes and the policy environment vary significantly across coun- tries, a time trend and variables that represent the institutional frame- work are controlled for simultaneously. In addition, the possible exis- tence of (unobservable) country effects is specifically taken into account and tested.6 Formally, the data are handled in two separate ways. First, all avail- able data are used, pooling together the whole information set into a single sample. In this pooled data case, where each country and year is treated as a separate observation (denoted by subscript i) and no indi- vidual heterogeneity is allowed, the following linear relationship for each of our four macroeconomic variables is specified and estimated:7 yi di xi i. (5.1) The term yi represents the dependent variable, is the intercept, di is a vector of dummies that accounts for private participation and its starting year, and xi is a vector of control variables that includes a time trend and others that characterize the country's institutional framework. Finally, iis a normally distributed error term, uncorre- lated with the regressors, and , , and the (vectors of) parameters to estimate. The variables included in x reflect the political and governance sit- uation, taking into account the degree of political stability of the coun- try (approximated by the degree of internal conflict) and the strength of the governance structure of the country. According to the specifi- cation of x i it is possible to derive separate models from expression 5.1. In the first one (Model 1), the macroeconomic variables are ex- plained by two dummy variables that reflect whether some form of private participation exists in utilities and in transport (they will be labeled DU and DT, respectively). The second model (Model 2) tests, in addition, for the effect of investment associated with a specific form of private sector participation on each one of the macroeconomic variables. Three types of privatization contracts associated with pri- vate investment are distinguished: divestures or sale of the assets (DIV), concessions (CONC), and greenfield projects (GP), which are MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 143 new investments such as new power generators or toll roads. Each of these variables is defined as the share of total investment from priva- tization associated with each contract type. The expected sign on these explanatory variables varies with the macroeconomic variable explained. If the predictions of the advocates of privatization are credible, one should expect a net positive effect of infrastructure privatization for GDP per capita and for domestic in- vestment as percentage of GDP, because these are some of the core macroeconomic promises of privatization. For the share of public investment in GDP and for the share of current expenditures, the a priori expectation would be a negative sign, because infrastructure privatiza- tion is expected to reduce the overall size of the public sector. In addi- tion, stronger institutions are likely to generate better macroeconomic performance. The second model tested makes use of the panel characteristics of the sample, where a number of individuals (21 countries, denoted by subscript j) are repeatedly observed through time (t 1985, . . . , 1994). In the panel data case it is now possible to study specifically whether there are country-specific effects not included in expression 5.1. For each of our macroeconomic variables, the linear relationship that is tested becomes yjt djt xjt j jt. (5.2) Both dependent and independent variables have time-variability, but Model 1 and Model 2 could be estimated again, using the same defi- nitions of x provided above. The most significant difference between (5.2) and the pooled data case is that a country-specific effect (labeled j) is explicitly accounted for, whereas the error term jtis again nor- mally distributed and uncorrelated with the regressors. It is precisely the nonappearance of the country-specific term that may bias the estimates in the pooled data case because of a standard omitted-variable problem (Amemiya 1985). Panel data models allow for a method to correct this problem, using either a fixed effects or a random effects approach. In the first case, the (unobserved) individual heterogeneity is represented as a parametric shift in expression 5.2. It is as if a new intercept, j j , time-invariant and particular to each country, were defined and the estimation by ordinary least- squares (OLS) would explicitly consider it. In the random effects case, the individual heterogeneity term is assumed to be part of the error term, ujt j jt . The error becomes autocorrelated, and the model must be estimated by generalized least-squares (GLS). 144 THE LIMITS OF STABILIZATION Unfortunately, both approaches do not always yield the same re- sult, as observed by Hausman (1978). However, if the effects of omit- ted variables can be appropriately summarized by a random variable and the (unobserved) individual effects may also represent the igno- rance of the investigator, it does not seem unreasonable to treat in one case the source of ignorance as fixed ( ) and in the other case as ran- j dom (ujt). It appears that one way to encompass the fixed effects (FE) and the random effects models is to assume from the outset that the ef- fects are random and use GLS to estimate them. The immediate check, summarized in the Hausman test, would be then to contrast whether the heteroskedasticity of the model allows a fixed effect approach.8 The Hausman test is used when there are two estimators of the pa- rameter vector (for example, GLS and FE). Under the null hypoth- esis (H0), individual effects are not correlated with the regressors, GLS is consistent and efficient, but FE is inefficient. Under the alternative H1, FE is consistent but GLS is inconsistent. This allows a routinely performed comparison between fixed effects and random effects esti- mates. A final important question regarding model specification is related to potential dynamic effects in our estimated relationships. Unless the economies behave in a hyper-rational way and manage to internalize instantaneously the effects of reform policies, the optimal lag for the dummies included in vector d should be different from zero. It is nat- ural to expect that privatization may not convey its full (positive or negative) consequences immediately. Instead, based on a simple look at the facts in the region, a reasonable lag of one or two years should be considered. These dynamic effects are investigated by estimating-- for each of our dependent variables, for each data case (pooled versus panel), and for each of our models (Models 1 and 2)--slightly differ- ent variations on (5.1) and (5.2), where the dummies have been lagged one and two periods. The results of all these estimations, reported be- low, permit analysis of the macroeconomic effects of privatizations, by type of process, considering the existence of country-specific effects, and taking into account short-run versus medium-run impacts. The Variables, the Data, and Their Limitations A sample of 21 Latin American and Caribbean countries, excluding only Belize, French Guyana, and Surinam among mainland states, was collected. In principle, this geographical dispersion offers enough vari- ety of infrastructure reform experiences and of income levels to yield useful policy conclusions. The time period covered stops in 1998, just MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 145 before the effects of the Asian crisis started to have a major impact on the financing of Latin America's infrastructure. The specific sample size for each macroeconomic variable consid- ered in this study varies across the models estimated because compa- rable data could not be obtained for all variables for all countries. The largest samples cover all of the 21 countries. The smallest focuses on only 16 countries. Because the overall sample tracks the changes in the role of the private sector in infrastructure for 14 years (from 1985 to 1998), the econometrics can make use of panel-of-data approaches as described above. Because there are several variables for which no in- formation was available for some years, the panel is unbalanced.9 The macroeconomic dependent variables--GDP per capita, total public investment, total private investment, and current public expen- ditures--are from the World Development Indicators produced by the World Bank (2002a) and are all expressed in 1995 U.S. dollars at con- stant prices.10 Table 5.1 summarizes the ranking of the countries covered by the sample, for the sample time average, and for each one of the macro- economic variables. At first glance the table shows the lack of consis- tency of countries in ranking, suggesting that there are enough differ- ences in behavior across variables to justify a separate analysis of each macroeconomic variable individually. The table also shows the main sources of imbalance in our data panel. The fiscal deficit is the least complete variable because values for El Salvador, Guatemala, Guyana, Honduras, and Jamaica are missing. The data quality issues already referred to start here. A measure- ment problem may exist in the definition of several of these macro- economic variables in relation to the concerns addressed here. According to the World Bank's World Development Indicators data- base, public expenditure, and public investment all refer to the central government alone. However, much infrastructure-related activity is usually developed by public enterprises that may finance themselves outside the central government's budget. These data are thus not picked up by this database as public investment and get picked up only as part of total investment by national accounts. The only way any change resulting from increased private participation can be identified is through the decline in transfers from the central government to the public enterprises, once these are replaced by private operators. This measure is imprecise, however, because public sector accounts are not very detailed at the sector level and hence the data could pro- duce imprecise results. For example, the data fail to capture much of the impact on recurrent public expenditures and on public investment in utilities privatization when current or capital expenditures on utilities (2) 8.29 6.38 7.84 9.51 7.86 8.23 8.63 12.34 16.00 10.44 12.56 16.49 10.45 10.79 19.00 11.81 14.85 24.01 17.66 15.22 13.43 Value expenditure 5 9 4 2 7 1 3 6 8 17 10 14 13 12 21 20 11 15 19 18 16 Current Ranking (2) 19.04 15.36 21.38 23.87 21.61 26.58 20.54 15.56 14.29 29.64 10.45 25.43 29.23 22.70 23.97 21.94 22.26 21.31 17.72 12.97 18.67 Value investment 6 3 1 4 2 7 5 9 8 Private 14 18 11 10 13 17 19 21 12 16 20 15 Ranking GDP. (2) of 1.49 7.86 2.32 5.02 7.82 4.95 4.64 3.65 2.67 4.59 7.61 3.23 3.54 4.49 3.62 2.51 3.76 9.60 15.91 12.40 Ranking Value investment percentage and in 4 7 5 8 9 1 6 (2) Public 20 19 13 17 10 ---- 2 3 16 15 11 14 18 12 Levels Ranking capita; per (1) US$ 863.76 671.65 440.21 697.03 476.69 in Variables: Value 7065.76 4269.94 3444.01 2182.56 2471.03 1518.33 1488.07 1398.81 1567.49 4102.54 2847.40 1793.86 2391.83 4247.32 4989.11 3510.81 (1) elaboration.' GDP authors 1 3 7 9 5 8 4 2 6 17 11 14 15 16 19 21 18 13 20 12 10 investment. and Ranking Macroeconomic 2002a, domestic de Bank gross Main Tobago R.B. and available. World 5.1 GDI Rica Not Salvador -- Note: Source: Table Country Argentina Bolivia Brazil Chile Colombia Costa Ecuador El Guatemala Guyana Haiti Honduras Jamaica Mexico Nicaragua Panama Paraguay Peru Trinidad Uruguay Venezuela, 146 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 147 prior to privatization were made by public enterprises rather than the government, because the balance sheets of public enterprises are seldom well integrated in the published government budgets. Furthermore, this effect may differ for different types of infrastructure (for example, telecommunications and power usually belonged to the realm of public enterprise whereas roads and ports were typically under the central government), and at different government levels. Fortunately, in Latin America--with the major exception of Brazil (in the case of roads, for example)--the most relevant privatization transactions in the region generally involved the central government. The second matter of concern with the variables is the specific defini- tion of privatization. A set of infrastructure privatization dummies (d in the econometric model) were relied on, constructed from the World Bank PPI database on private participation in infrastructure projects (World Bank, PPI Project Data Base, available at http://rru.worldbank.org.) The dummies are as follows: · DU: takes a value of 1 starting on the first year there is a private utility project in a specific country (for example, a private power gen- erator or a private cellular operator). · DT: takes a value of 1 starting on the first year there is a (signifi- cant) private operator of transport infrastructure in a specific country. Table 5.2 shows the first year in which each dummy takes the value of 1. The main problem with this variable is that it reflects the start of Table 5.2 First Year for Private Participation in Utilities and Transport Country Utilities Transport Country Utilities Transport Argentina 1990 1991 Honduras 1994 n.a. Bolivia 1987 1996 Jamaica 1990 n.a. Brazil 1985 1985 Mexico 1991 1991 Chile 1987 1995 Nicaragua 1993 n.a. Colombia 1991 1994 Panama 1996 1994 Costa Rica 1989 n.a. Paraguay 1992 n.a. Ecuador 1985 1985 Peru 1985 1985 El Salvador 1995 n.a. Trinidad and 1991 n.a. Guatemala 1994 1997 Tobago Guyana 1991 n.a. Uruguay 1992 1993 Haiti 1995 n.a. Venezuela, R.B. de 1985 1985 n.a. Not applicable. Note: Average. Dummies DU and DT take value 1 from the starting year onward. Source: World Bank 2002a, and authors' elaboration. 148 THE LIMITS OF STABILIZATION reliance on some form of project finance scheme rather than a major effort to restructure the sector and to rely systematically on private fi- nance and operation for most of the sector. The correlation between the variable constructed this way and a variable that would focus on major policy changes is strong but far from perfect. The decision was made to stick to this approach because project finance data are more closely related to the actual investment levels that are expected to in- fluence the levels of macroeconomic indicators, in particular for the public sector. In addition, the attempt was made to distinguish between contract types associated with each project. To do so, the following variables associated with the three types of infrastructure privatization were constructed: · DIV: the number of divestitures or asset sales contracts in each year because of infrastructure privatizations for each of the two broad subsectors for each country · GP: the number of greenfield project contracts in each year be- cause of infrastructure privatizations for each of the two broad sub- sectors for each country · CONC: the number of concessions contracts in the database in each of the two subsectors. Each contract type variable is multiplied by the relevant dummy to ensure that the contract type only kicks into the regression after the first privatization in utilities and transport has started. This is recog- nized by a DT and DU suffix attached below to each contract type in Table 5.3, which summarizes the results. Table 5.3 shows that the institutional explanatory variables used as regressors in the model specifications (5.1) and (5.2) above are the fol- lowing: two institutional variables (labeled by x in the model) have been obtained from the World Development Indicators. The index of political stability (D) is approximated by the inverse of the degree of violence and its impact on the ability of the government to govern. The countries are ranked on a scale of 1 to 12 with the lowest rating allo- cated to the most unstable countries (for example, countries during a civil war) and the highest rating to the stable countries. The quality of the political system of the country (F) is also approximated by a rank- ing on a scale of 1 to 6. A ranking of 1 is allocated to the most corrupt countries. A value of 6 is allocated when a country is perceived to be corruption-free.11 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 149 Table 5.3 Average Value of the Institutional Variables between 1985 and 1998 Political Political Country stability Corruption Country stability Corruption Argentina 9.9 3.4 Honduras 5.8 2.1 Bolivia 5.9 2.1 Jamaica 9.1 2.6 Brazil 8.9 3.6 Mexico 9.4 3.1 Chile 7.4 3.2 Nicaragua 5.3 4.7 Colombia 5.4 2.7 Panama 8.0 2.1 Costa Rica 9.3 4.9 Paraguay 9.4 1.2 Ecuador 9.8 3.1 Peru 5.1 3.0 El Salvador 4.7 2.5 Trinidad and 8.8 2.8 Guatemala 6.3 2.5 Tobago Guyana 7.7 1.7 Uruguay 8.3 3.0 Haiti 4.7 1.4 Venezuela, R.B. de 10.4 3.0 Note: Political stability is measured from 1 (low) to 12 (high). Corruption goes from 1 (bad) to 6 (clean). Source: World Bank 2002a. The Results LIMDEP v.7.0 econometric software was relied on to obtain OLS and GLS estimates of the linear specifications (5.1) and (5.2) described above. For each dependent variable (GDP per capita, private invest- ment, public investment, and public expenditure), tables are provided, first for Model 1 (where privatization dummies are separated into transportation and utilities) and then for Model 2 (where contract types for transportation and utilities are separately identified). Each table is divided into two main columns that allow an explicit compar- ison between the pooled data case (that is, not taking into account the presence of country-specific effects) and the panel data case. Finally, the results presented in each column distinguish between the situation where the privatization dummies are simultaneous (zero lag) or are lagged one or two periods to identify delays or adjustments in the macro­micro effects. All estimated coefficients are accompanied by the standard goodness of fit statistics (t-coefficients at 95 percent of confidence, adjusted R2 values, and the corresponding log-likelihood ratios).12 Panel data results (which specifically account for the pres- ence of individual heterogeneity) correspond to the random effects specification, except when the result of the Hausman test suggests that fixed effects could be more appropriate. 150 THE LIMITS OF STABILIZATION Finally, the ultimate comparison between pooled data estimates and panel data estimates (in other words, whether country-specific effects are relevant or not) can be carried out through a general specification test on the covariance properties of the panel residuals. There are dif- ferent tests for this purpose in the literature. The standard LM-test proposed by Breusch and Pagan (1979) was chosen because its calcu- lation is simpler. The LM statistics, whose null hypothesis in this case implies that individual effects are not relevant, are shown. Tests are carried out in the final rows of each table. Effects of Private Participation in Infrastructure on GDP per Capita Table 5.4 summarizes the estimates of Model 1 using GDP per capita as the dependent variable, both for the pooled data case and the panel data case. Because the comparison tests show that panel data estimates (using the fixed effects approach) are preferred to pooled data estimates, the results worth considering are those in the final columns. Moreover, although the goodness of fit measures are to be taken cautiously in panel estimations, the values of the adjusted R2 are relatively high. The preferred regression suggests that the trend matters strongly and that the institutional variables are highly signif- icant with the expected sign, even when lagged dummies are included in the regression. As for the main focus of this chapter, the coefficients on the PPI dummies, DU and DT, suggest that only PPI in transport infrastruc- ture seems to have a positive (and significant) effect on GDP per capita, both when considered unlagged and when a lag of one or two periods is included. These results are somewhat surprising, but imply that the effect of PPI on growth varies across infrastructure types in Latin America. The lagged dummies do not alter the signs or size of these ef- fects very much, suggesting that the impact of PPI in transport may be distributed over time. Table 5.5 summarizes the results for Model 2, in which the dummies are separated by type of PPI (divestures, DIV; greenfield projects, GP; and concessions, CONC). The estimation methodology is consistent with table 5.4, because the Hausman test suggests that fixed effects are preferable and the LM test does not reject the existence of country-specific effects. The institutional variables in the panel data estimations are, respectively, positive and negative for D and F, with the same inter- pretation as above. However, the disaggregated effects of PPI types show several new results. First, divestitures and greenfield projects have significant and positive effects for utilities (even when lagged one 2 -ratiot Prob. 0.0000 Prob. 0.0006 3.18398 2.50583 2.31178 0.57773 7.00114 0.9823 1715.89 lags. Lag =2 d.f. 1 d.f. 5 21.3374 23.7823 67.1605 30.2842 480.864 Coefficient Lag different effects) test 1 -ratiot 0.6350 3.01499 2.78471 3.00623 8.31114 test (fixed LM 386.69 21.34 represent 0.9821 1860.7 case Hausman Lag cells 82.446 504.02 17.9955 25.2048 3.21975 data Coefficient Blank Prob. 0.0000 Prob. 0.0029 Panel -ratiot 3.8653 0.5787 3.36999 3.18444 8.40341 confidence. 2016.3 =1 d.f. 1 d.f. 5 of 0.98041 Unlagged 106.00 29.439 18.7264 28.3979 489.852 Lag level Coefficient tests test test 18 percent 1) 2 -ratiot 0.54850 1.71985 6.42126 2.47938 1.95242 6.04391 LM 432.92 95 Hausman at 0.3025 2190.09 Lag Comparison (Model 530.48 205.508 58.1843 257.547 236.825 1465.98 calculated Coefficient Prob. 0.0000 Prob. 0.05167 are Capita case 1 -ratiot 0.51919 1.74169 6.73199 2.83498 2.25789 6.51564 d.f. 1 d.f. 5 -ratiost Per data 2368.4 Lag 0.31729 Unlagged capita. GDP 52.253 591.80 1453.3 Pooled 174.878 252.367 249.947 test per on Coefficient test GDP LM 492.75 10.99 PPI is 1.641 -ratiot 2.5325 Hausman of 0.52042 6.94567 3.20049 6.90419 0.3273 2546.68 variable computations.' Unlagged Effects 43.922 641.87 158.661 242.708 264.676 1433.55 Coefficient method Authors 5.4 Dependent pooled random 2 R vs. vs. Lr Note: Source: Table Variable Constant TIME D F DU DT DU-1 DT-1 DU-2 DT-2 Adj. Log Estimation Panel Fixed 151 2 -ratiot 0.1131 3.76959 1.82271 1.94686 2.77911 3.34042 1.04156 2.55595 0.18168 Lag 29.884 83.771 23.0818 17.6267 58.2651 18.0522 31.2365 663.142 6.28027 Coefficient effects) 1 -ratiot 1.8994 0.3913 3.41408 2.19548 2.05414 3.86373 0.61229 2.54702 0.54961 (fixed case Lag 17.097 18.8212 20.6748 61.1194 13.1279 36.4376 672.495 178.181 29.2965 data Coefficient Panel -ratiot 2.6597 3.34287 2.57865 2.47859 2.22143 4.03708 0.77787 0.04648 0.85045 Unlagged 38.75 9.805 16.954 25.503 23.5891 73.6671 12.9868 21.8197 722.373 Coefficient 2) 2 -ratiot 0.5937 2.52653 4.41238 6.27975 2.56155 3.95182 2.54982 0.80218 0.61293 1.13581 Lag (Model 841.8 323.96 128.067 228.998 220.828 118.364 102.304 128.024 841.789 1522.47 Coefficient Capita case 1 -ratiot 6.7498 2.23009 4.67099 2.94337 3.41141 2.49363 1.21337 0.61826 0.90207 1.50967 Per data Lag GDP 98.48 Pooled 678.623 118.527 229.748 235.477 106.818 183.028 867.062 2125.38 404.275 on Coefficient PPI -ratiot 1.8647 4.8668 0.5994 of 7.08405 3.44418 4.99402 2.51293 1.43128 0.43957 1.07172 Unlagged 836.1 Effects 109.31 99.394 209.51 476.72 165.07 523.698 225.598 260.585 128.391 Coefficient 5.5 Table Variable Constant TIME D F DIVDU GPDU CONCDU DIVDT GPDT CONCDT DIVDU-1 GPDU-1 CONCDU-1 DIVDT-1 GPDT-1 CONCDT-1 DIVDU-2 GPDU-2 CONCDU-2 DIVDT-2 GPDT-2 CONCDT-2 152 Prob. 0.0000 Prob. 0.0939 0.9821 1715.32 2 = d.f. 1 d.f. 9 Lag test test LM 419.37 14.89 0.9811 1866.01 Hausman Prob. 0.0000 Prob. 0.043 1 confidence. 0.9795 2020.16 = d.f. 1 d.f. 9 of Lag level tests test test percent LM 493.99 17.36 95 Hausman at 0.4111 2166.67 Comparison calculated Prob. 0.0000 Prob. 0.0495 are d.f. 1 d.f. 9 -ratiost 0.4169 2344.90 Unlagged capita. test per test GDP LM 566.59 16.95 is Hausman 0.4159 2523.87 variable computations.' method Authors Dependent pooled random 2 R vs. vs. Lr Note: Source: Adj. Log Estimation Panel Fixed 153 154 THE LIMITS OF STABILIZATION and two periods). Concessions, on the other hand, do not yield sig- nificant coefficients. For transport, only divestitures seem to have a relevant impact on GDP per capita. Divestitures are sometimes viewed as the strongest form of commitment to the private sector to take care of the delivery of the services. What this suggests, at least in a first analysis, is that only the strongest commitment to a private sec- tor role has an impact on GDP per capita. Effects of Infrastructure PPIs on Private Investment Table 5.6 shows the results from Model 1 using private investment (as directly reported by the World Development Indicators database [1999]) as the dependent variable. An interest rate variable, LR (the lending rate listed in IMF 2002), was added to ensure a better specifi- cation of the model for both the pooled data case and the panel data case. As in table 5.4 above, country-specific effects are relevant, accord- ing to Breusch and Pagan's LR test, but now the Hausman tests sug- gest that random effects, instead of fixed ones, are the preferable way to specify j . In general, it seems that this model is not as good in ex- plaining what happens to private investment. The trend continues to be a significant factor as is the degree of po- litical stability. The measure of corruption used does not perform well because it does not appear to have a statistically significant effect. Most interesting is that the PPI dummies (except for DT when lagged two periods) are never significant. Table 5.7 tells a very similar story. Again, panel data (with random effects) are preferable to pooled data, but the overall significance of the model is lower than for GDP per capita. As for our variable of concern, the emerging story is interesting. It suggests that greenfield projects can make a difference but do so with a negative sign, implying some crowding-out of other private investment projects. The results also show, somewhat expectedly, that concession contracts in transport have a positive lagged effect on private investment. As is well known by the specialists of investment promotion programs, good transport services are crucial to attract investment. These results confirm their experience. Because there was some concern about the quality of the dependent variable used, the models were also run by redefining private invest- ment as the difference between total investment and public investment. This analysis is carried out in table 5.8, where Model 1 (and Model 2, which was not reported here) has been reestimated using this new def- inition of the dependent variable. The estimates--once more, panel 2 -ratiot Prob. 0.0000 Prob. 0.361 3.80217 1.58399 0.64676 1.07466 0.17563 0.47977 7.94757 0.6741 484.37 Lag 2 0.1647 d.f. 1 d.f. 6 0.47517 0.29703 0.40788 0.00069 0.01851 15.5332 Coefficient effects) Lag 3.961 test 1 -ratiot 0.2771 8.14323 2.50088 0.80148 0.58817 0.14876 test 6.58 (random LM 344.24 543.15 Lag 0.66016 Hausman case 14.7674 0.43668 0.43663 0.44978 0.00022 0.13708 0.01072 data Coefficient Prob. 0.0000 Prob. 0.1553 Panel -ratiot 2.9258 8.81839 3.68471 1.23564 0.67633 0.81816 0.51042 confidence. 1 of 0.6789 590.903 d.f. 1 d.f. 6 Unlagged level 14.9078 0.36130 0.45569 0.61882 0.00014 0.69279 0.48218 Lag Coefficient tests test percent 1) test 95 9.34 2 -ratiot 2.20199 2.02643 1.55013 2.16141 0.50375 0.71746 6.27925 LM 353.10 at Hausman (Model 0.1311 585.539 Lag Comparison calculated 0.4425 0.36487 0.86932 0.00188 0.58515 0.04304 12.3933 Coefficient are Prob. 0.0000 Prob. 0.13 case -ratiost Investment 1 -ratiot 6.98812 2.28893 2.80695 1.67031 2.48127 0.62162­ 0.19090 d.f. 1 d.f. 6 data 647.51 Lag 0.15799 Unlagged Private investment. Pooled 12.0409 0.32733 0.54518 0.82739 0.00126 0.70641­ 0.01114 test on Coefficient test 9.86 private LM 337.56 PPI is -ratiot Hausman of 8.02367 2.41093 3.95373 1.37261 2.62139 0.66619 3.15398 0.2131 703.43 variable computations.' Unlagged Effects 11.861 0.0007 2.7784 0.29529 0.67260 0.60011 0.74026 Coefficient method Authors 5.6 Dependent pooled random 2 R vs. vs. Lr Note: Source: Table Variable Constant TIME D F LR DU DT DU-1 DT-1 DU-2 DT-2 Adj. Log Estimation Panel Fixed 155 3.23 -ratiot 7.9948 2.8282 5.47164 1.89907 1.00481 0.26587 0.71361 0.39855 1.99786 2.22362 1 Lag (random) 15.1976 0.53941 0.55977 1.12983 0.00037 0.04177 0.50265 0.35248 1.59119 18.8736 2.77924 Coefficient case data -ratiot 8.14836 5.68235 3.45748 1.79418 0.77739 0.48231 1.91108 0.25257 0.45052 0.49100 0.23612 Panel Unlagged 15.0516 0.48999 0.54424 0.96090 0.00016 0.05864 0.30279 0.10866 1.75513 1.52561 0.12912 Coefficient 2) -ratiot 7.17573 2.36962 2.80214 1.54584 2.41655 0.28148 0.36051 0.34985 0.51858 0.01056 0.0807 1 (Model Lag case 12.1531 0.32373 0.52336 0.77469 0.00123 0.04318 0.07583 0.25176 3.20066 0.14229 0.13469 Coefficient Investment data Private Pooled -ratiot 1.2658 7.89852 2.42325 3.49916 1.76644 2.57747 0.28951 0.27260 0.19009 0.34429 1.27965 on PPI Unlagged of 0.0007 0.0347 0.1222 0.9697 11.8566 0.28293 0.57704 0.78758 0.05293 2.06112 6.08447 Coefficient Effects 5.7 Table Variable Constant TIME D F LR DIVDU GPDU CONCDU DIVDT GPDT CONCDT DIVDU-1 GPDU-1 CONCDU-1 DIVDT-1 GPDT-1 CONCDT-1 156 Prob. 0.0000 Prob. 0.3079 0.6787 535.99 1 d.f. 1 d.f. 10 Lag test test LM 364.74 11.67 0.6785 588.73 Hausman confidence. of level tests percent Prob. 0.0000 Prob. 0.655 95 at 0.1453 646.96 Comparison d.f. 1 calculated d.f. 10 are Unlagged -ratiost test test 7.73 investment. LM 376.60 Hausman private 0.1757 706.64 is variable computations.' method pooled random Authors Dependent 2 R vs. vs. Lr Note: Source: Adj. Log Estimation Panel Fixed 157 2 -ratiot 7.2551 3.4953 2.0644 2.104­ 0.8560 0.167­ 0.084­ Prob. 0.0000 Prob. 0.3402 0.7365 Lag 407.24­ 2 1 6 12.717 Public 0.36512 0.31643 1.1380­ 0.0004 0.130­ 0.002­ d.f. d.f. Coefficient Lag Minus test (random) 1 -ratiot 5.11725 3.25328 2.59491 0.96630 2.34242­ 2.67919­ 0.67958 test 6.79 LM 341.30 case 0.720 456.39­ Hausman Lag data 8.18617 0.42688 0.46902 0.43905 0.0010­ 2.7763­ 0.03528 Investment Coefficient Panel Prob. 0.0000 Prob. 0.1188 -ratiot 7.70941 3.94077 2.77623 2.46355­ 0.55941 0.78091­ 1.35159 1 Domestic 0.720 497.00­ d.f. 1 d.f. 6 Unlagged 1.0974 11.9965 0.33188 0.37075 1.07941­ 0.00010 0.56833­ Lag Gross Coefficient tests test as test 2 -ratiot 4.64182 2.74017 1.88699 0.71691 2.01302­ 2.02563­ 0.72729 LM 372.38 10.14 investment. Hausman Defined 0.1008 Lag 522.14­ Comparison 8.56549 0.42191 0.38643 0.37264 0.0015­ 2.1426­ 0.0391 GDI-public Coefficient Prob. 0.0000 Prob. 0.22 as case Investment 1 -ratiot 7.60141 4.13288 2.91702 2.25479­ 0.19924 1.4864­ defined 0.22991­ d.f. 1 d.f. 6 data 0.1322 Lag 574.47­ Unlagged Private investment Pooled 12.0482 0.38012 0.41659 1.06257­ 6.10905 1.13005­ 0.00707­ test on Coefficient test 8.14 private LM 411.83 PPI is 1) -ratiot Hausman of 5.90699 3.33634 2.88952 0.83851 2.26562­ 2.24367­ 0.05994­ 0.1353 624.24­ variable computations.' Unlagged Effects (Model 8.52524 0.38858 0.48252 0.35229 0.0006­ 2.3802­ Coefficient 0.05089­ method Authors 5.8 Dependent pooled random 2 R vs. vs. Lr Note: Source: Table Investment Variable Constant TIME D F LR DU DT DU-1 DT-1 DU-2 DT-2 Adj. Log Estimation Panel Fixed 158 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 159 data estimates--score slightly better, particularly the institutional vari- ables (which exhibited the same signs as in tables 5.4 and 5.5) and the utilities PPI dummy, but again the overall significance of the model is not as good as had been hoped. The estimates do suggest, however, that a lagged crowding-out was taking place during the 1980s and 1990s as a result of the increased presence of private sector participa- tion in utilities. Effects of Infrastructure PPIs on Public Investment The estimates in table 5.9 summarize the effects of PPI policies on pub- lic investment. The overall statistical results are similar to those of pre- vious tables (and particularly, again, panel data are preferred and po- litical stability is the strongest institutional explanatory variable). The coefficients of the policy variables reveal several notable dif- ferences. First, the unlagged PPI dummies are significant and have the strongest statistical significance, but the impact of PPI is still strong with a one-year lag. Second, and much more interesting, the PPI in utilities and transport infrastructures has a different sign (posi- tive and negative, respectively). PPI in utilities complement or crowd in public investments, whereas PPI in transport substitutes for or crowds out public investment. What this may reflect is that reforms in the utilities sector are used by governments to raise matching resources from private operators for the sector, whereas for transport, private investments allow governments to reduce their commitments to the sector--in terms of expansion, at least. These results hold, however, only at the aggregate level because it is not possible to draw similar, if more subtle, conclusions from a disaggregation of contract types. Table 5.10, where Model 2 estimates are presented, suggests that disaggregating the PPI dummies by contract type (DIV, CONC, GP) not only reduces the overall significance of the panel data model but also eliminates the validity of individual coefficients in all cases. Effects of Infrastructure PPIs on Recurrent Public Expenditures The effects of PPI on recurrent public expenditures summarized in tables 5.11 and 5.12 (pages 164 and 166, respectively) follow a partic- ularly interesting pattern, especially when contrasted with the pattern seen for the effect of PPI on public investment. From an overall statis- tical viewpoint, the unlagged panel case with fixed effects provides the best results according to the values of the comparison tests; as usual, 2 -ratiot Prob. 0.0000 Prob. 0.6563 4.32134 0.60620­ 1.01417 0.49584 0.98976 1.33359­ 0.8888 Lag 244.885­ 2 1 6 5.20166 0.0266­ d.f. d.f. 0.06459 0.11744 0.31974 0.55866­ Coefficient effects) Lag Test test 1 -ratiot 4.1948 1.24162­ 1.61114 0.11457 2.41344 1.57607­ Ratio 4.15 (random 341.15 0.8560 Lag 300.381­ Hausman case 4.83884 0.0541­ Likelihood 0.10937 0.02644 0.85183 0.65454­ data Coefficient Prob. 0.0000 Prob. 0.7838 Panel -ratiot 4.20102 1.38683­ 2.27065 0.08736 2.75014 1.85432­ 1 confidence. 0.8314 347.08­ d.f. 1 d.f. 6 of Unlagged 4.44897 0.05740­ 0.14989 0.01922 0.97727 0.75619­ Lag level Coefficient tests test test 1) percent 3.20 2 -ratiot 3.38307 0.05000­ 0.8315­ 1.77114 2.40797 1.45039­ LM 395.18 95 Hausman at 0.0393 (Model Lag 435.03­ Comparison 4.06308 0.0049­ 0.11209­ 0.61026 1.73967 0.95966­ calculated Coefficient Prob. 0.0000 Prob. 0.7664 are case Investment 1 -ratiot 3.75475 0.7266­ 0.13596­ 1.44382 3.14047 2.14511­ -ratiost d.f. 1 d.f. 6 data 0.0603 Lag 486.19­ Unlagged Public 3.9936 Pooled 0.0624­ 0.01665­ 0.44173 2.23534 1.28836­ test investment. on Coefficient test LM 479.3 3.33 public PPI is -ratiot Hausman of 3.94788 0.98179­ 0.26252 1.64164 3.35006 2.74791­ 0.079 533.17­ variable computations.' Unlagged Effects 3.66962 Coefficient 0.07364­ 0.02823 0.44421 2.28888 1.50248­ method Authors 5.9 Dependent pooled random 2 R vs. vs. Lr Note: Source: Table Variable Constant TIME D F DU DT DU-1 DT-1 DU-2 DT-2 Adj. Log Estimation Panel Fixed 160 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 161 political stability matters. The time trend has been eliminated because there was a multicollinearity problem with the institutional variable. The coefficients of the policy variables reveal several new ele- ments. First, both dummies are statistically significant, which implies that there is an interaction between public expenditures and privati- zation. Second, the PPI dummies for utilities suggest that private in- vestment in telecoms, energy and water, and sanitation has a declin- ing impact over time on recurrent public expenditures (as seen in the declining t-ratios for the lagged variables), whereas the positive sign on the transport dummy suggests that as the private sector starts in- vesting in transport, recurrent public expenditures in the sector in- crease. Third, the longer the lag with which the investment is ac- counted for, the lower the impact of private participation in utilities on these public expenditures. However, the longer the lag for trans- port, the higher the impact. The fact that the PPI in utilities and transport infrastructures has a different sign (negative and positive, respectively) is quite a significant result. For transport, this reflects the common wisdom among practi- tioners that investments in the sector are only viable when the opera- tion of the services allowed by the investment is subsidized. In other words, there is a complementarity between recurrent public expendi- tures and private investment expenditures in transport. For utilities, the observation that PPI reduces recurrent public expenditures in util- ities may reflect the fact that PPI often leads to significant cost reduc- tions and that subsidy levels tend to decline once private operators take over operations. It may also suggest that during the 1990s, at least, public and private expenditures in the sector were substitutes. Table 5.12, however, suggests that this result does not hold for all types of private sector participation. For divestitures in the utilities sec- tor, it seems that when PPI takes place with that type of privatization contract, recurrent expenditures increase. Conclusion This chapter provides empirical evidence on the impact that private participation in infrastructure has had on key macroeconomic vari- ables in a sample of 21 Latin American countries during the 1985­98 period. The effects on GDP per capita, current public expenditures, public investment, and private investment were examined, controlling for country effects and institutional factors. The most interesting ini- tial conclusions focus on the sign of the average macro effects of these micro reforms as estimated from Model 1. Table 5.13 (page 168) -ratiot 3.85598 2.28561 0.33593 0.54395 1.06111 0.04515 0.15809 0.44201 1 0.102901 0.186214 Lag (random) 4.81907 0.08577 0.03976 0.07574 0.01017 0.24170 1.61612 Coefficient 0.004202 0.159847 0.091974 case data -ratiot 3.86902 0.04831 2.61672 0.00797 0.15074 0.76188 0.21648 0.15255 0.62068 Panel 0.046816 Unlagged 0.1971 4.38753 0.00185 0.00192 0.00826 0.05281 0.34932 0.14526 Coefficient 0.177649 0.008682 2) -ratiot 2.83953 1.02017 1.88307 0.91265 0.30543 0.14025 1.19824 1 0.435871 0.091964 0.769687 (Model Lag case 3.05074 0.08516 0.15569 0.50780 6.09788 1.18086 Coefficient 0.086508 0.049902 0.579349 0.011500 Investment data Public Pooled -ratiot 3.17881 2.12111 1.40861 0.29531 0.78691 0.11882 0.38211 0.90236 on 0.972953 0.833599 PPI of Unlagged 3.00518 0.10843 0.03527 0.30248 0.41943 1.08498 0.40700 Coefficient 0.070658 0.084315 0.579541 Effects 5.10 Table Variable Constant TIME D F DIVDU GPDU CONCDU DIVDT GPDT CONCDT DIVDU-1 GPDU-1 CONCDU-1 DIVDT-1 GPDT-1 CONCDT-1 162 Prob. 0.0000 Prob. 0.9306 0.5475 303.393 1 d.f. 1 d.f. 10 Lag test test 4.34 LM 413.20 0.5199 352.592 Hausman confidence. of level tests Prob. percent 0.0000 Prob. 0.8829 95 at 486.68 Comparison 0.03386 d.f. 1 d.f. 10 calculated are Unlagged -ratiost test test 5.12 investment. LM 490.14 Hausman public 0.0525 534.039 is variable computations.' method pooled random Authors Dependent 2 R vs. vs. Lr Note: Source: Adj. Log Estimation Panel Fixed 163 2 -ratiot Prob. 0.0000 Prob. 0.0055 4.79349 1.48251 1.03656 2.32731 0.6687 610.06 Lag 2 1 5 0.55101 0.55759 0.60165 2.22864 = d.f. d.f. Coefficient Lag effects) test 1 -ratiot 5.10504 1.23573 1.78406 1.86185 Test (fixed LM 386.69 16.48 0.6648 677.59 case Hausman Lag 1.7084 0.60036 0.46611 1.03453 data Coefficient Prob. 0.0000 Prob. 0.0019 Panel 1) -ratiot 4.72158 1.02608 3.45204 1.91227 1 confidence. 0.676 731.53 = d.f. 1 d.f. 5 of Unlagged (Model 0.53505 0.37483 1.90985 1.60745 Lag level Coefficient tests test Test percent 19 2 -ratiot 0.1333 9.28352 1.58207 5.18957 1.27273 LM 432.92 95 Hausman at 0.0933 746.58 Expenditures Lag Comparison 10.3882 0.21873 1.67279 1.04329 0.11565 calculated Coefficient Prob. 0.0000 Prob. 0.0004 are Public case 1 -ratiot -ratiost 9.89115 1.53208 5.34853 1.84674 0.39110 d.f. 1 d.f. 5 data 0.1040 821.45 Recurrent Lag Unlagged 1.5276 on Pooled 10.9635 0.21252 1.69406 0.33347 test expenditure. Coefficient Test PPI LM 492.75 22.12 public -ratiot Hausman is of 10.7462 1.34699 5.82027 2.88953 0.49698 0.128 886.74 variable computations.' Effects Unlagged 0.1788 2.3251 0.4033 11.3039 1.76807 Coefficient method Authors 5.11 Dependent pooled random 2 R vs. vs. Lr Note: Source: Table Variable Constant D F DU DT DU-1 DT-1 DU-2 DT-2 Adj. Log Estimation Panel Fixed 164 MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 165 summarizes the main results with respect to the statistically significant signs that could be identified. The first obvious fact to emerge from table 5.13 is that transport and utilities privatization should not be expected to have the same macroeconomic effects. Transport has a significant positive effect on per capita income; utilities have no observable effect. Second, PPI, at best, leaves private investment constant but in the case of utilities tends to crowd investment out, which is the opposite of the effect it has on public investment. Indeed, the third result to emerge is from a public sector perspective. Utilities investment leads to increases in public in- vestments but reduces recurrent expenditures. The opposite holds for transport. In other words, there is crowding in of public investment for PPI in utilities and crowding out for transport. Also, although private transport investments require a matching commitment to operational subsidies, the arrival of private utility operators reduces the burden of these operational subsidies. The results generated by Model 2 are in general less interesting. The disaggregation of PPI per contract type yielded few statistically signif- icant results. The most interesting ones are that divestitures, the strongest form of commitment to the private sector, have clear positive effects on GDP per capita. The second interesting result is that conces- sion contracts and greenfield projects in transport have significant pay- offs in future investments. Finally, divestitures in utilities and transport concessions tend to increase recurrent expenditures. These results, however limited, provide the first econometric evi- dence on the macro effects of micro reforms for the region in which PPI policies have been the most active. Much better data is needed to draw more specific and more robust policy conclusions. Much more ambi- tious econometric analysis is also needed. In particular, causality has not been tested and an optimal lag structure has not been identified be- cause of data limitations. As the PPI experience progresses and more and better data become available, it should be possible to refine these results. But for now, these results already provide enough reasons to be concerned about a good assessment of the macro effects and in particular the fiscal effects of private participation in infrastructure. The fact that the effects on GDP per capita are neutral at worst and most probably positive is good news, but privatization comes at a risk with respect to its effects on the public sector accounts. The revelation of this risk may be the main contribution of this chapter because it is inconsistent with the fis- cal gains expected by many policymakers as they engage in infrastruc- ture privatization programs. 2 -ratiot 0.6523 7.97494 5.19081 1.75473 1.25544 1.48609 0.90769 0.56524 1.14032 Lag 3.7521 13.4454 0.55700 0.62041 0.10595 0.15917 0.32673 1.82049 0.78765 Coefficient case 1 -ratiot 0.5856 6.18977 1.48095 2.02477 0.92870 1.22012 0.50511 0.945324 data Lag Panel 0.70928 0.56458 0.18662 0.10929 0.44321 2.00054 2.97191 0.52354 Coefficient -ratiot 0.7388 6.24888 0.84395 2.10343 0.36408 1.24778 0.53024 0.84723 Unlagged 0.69229 0.31248 0.16150 0.04272 0.44645 1.82768 1.97114 0.32307 Coefficient 2) 2 -ratiot 9.75749 2.27076 5.21548 0.43473 0.22158 0.64509 0.76645 1.13554 2.25521 (Model Lag 10.8341 0.29091 1.64259 0.04841 0.03067 0.37117 3.90711 10.1595 2.32176 Coefficient Expenditures case 1 -ratiot 0.4143 10.3069 2.56723 5.06543 0.27820 0.22135 0.58592 0.03476 0.96937 data Public Lag on Pooled 11.5392 0.33320 1.58883 0.03432 0.03225 0.24189 3.16602 0.31654 1.00619 Coefficient PPI of -ratiot 2.879 0.2677 11.0155 5.24564 0.95671 0.18406 0.11170 0.51358 1.16415 Effects Unlagged 0.3618 0.0648 2.8262 11.8751 1.59759 0.10125 0.02751 4.98035 0.16325 Coefficient 5.12 Table Variable Constant D F DIVDU GPDU CONCDU DIVDT GPDT CONCDT DIVDU-1 GPDU-1 CONCDU-1 DIVDT-1 GPDT-1 CONCDT-1 DIVDU-2 GPDU-2 CONCDU-2 DIVDT-2 GPDT-2 CONCDT-2 166 Prob. 0.0000 Prob. 0.1767 0.6724 606.39 2 d.f. 1 d.f. 9 agL test Test 12.7 LM 419.37 0.6683 673.94 Hausman Prob. 0.0000 Prob. 0.0255 1 confidence. 0.672 731.29 d.f. 1 d.f. 9 of Lag level tests test Test percent LM 493.99 18.96 95 Hausman at 0.1064 742.71 Comparison calculated Prob. 0.0000 Prob. 0.0113 are -ratiost d.f. 1 d.f. 9 0.0895 821.57 Unlagged test expenditure. Test LM 566.59 21.30 public Hausman is 0.091 890.77 variable computations.' method Authors Dependent pooled random 2 R vs. vs. Lr Note: Source: Adj. Log Estimation Panel Fixed 167 168 THE LIMITS OF STABILIZATION Table 5.13 Summary of Signs of Average Macroeconomic Effects of PPI Variable PPI in utilities PPI in transport GDP/capita Not significant Private investment Not significant Public investment Recurrent public expenditures Note: positive impact on the microeconomic variable; negative impact on the microeconomic variable. Notes 1. Siniscalco, Bortolotti, and Fantini (2001) provided a similar study for countries of the Organisation for Economic Co-operation and Development. 2. The recent literature on regulation theory explicitly acknowledges that the term private participation is much more general than privatization. The former encompasses many different forms that include divestures, concessions, management contracts, leases, and so on (see Laffont and Tirole 1998). 3. The new growth theory literature is the most likely place to find such a model (see, for example, Aghion, Caroli, and García-Peñalosa 1999). In addi- tion, political economy models may help explain under what circumstances pri- vatization policies can be a success or a failure (see Alesina and Perotti 1996). 4. See Chamberlain (1984) for a survey and examples on the use of macroeconomic panel data. 5. For a review on how different types of infrastructures affect macro- economic fundamentals, see, for example, Munnell 1992 or Gillen 1996. 6. As usual, a strong misspecification risk is always present in this sort of ad hoc model. However, because our idea is to isolate partial correlations among the privatization variables and the macroeconomic one, the use of the time trend and the institutional variables is the easiest way of minimizing that risk in this kind of heterogeneous sample. A lagged dependent variable (tried at preliminary stages of the work) would have done something similar, but at the cost of one degree of freedom and lower significance levels. 7. Nonlinear specifications were also discarded in preliminary estimations. 8. This argument has been widely discussed in the panel data literature. For example, Arellano (1993) insists on the fact that in the fixed effects model investigators make inferences conditional on the effects that are in the sample, whereas in the random effects model inferences are based on the population. But there is really no distinction in the nature of the effect: it is up to the in- vestigator to decide whether to make one type of inference or the other. 9. However, this can easily be handled in the econometrics (Greene 1995). 10. Although initially tried, the models on the effects on the public deficit were rejected because the variable was not sufficiently reliable. For the inter- ested reader, PPI in utilities tends to be associated with an immediate increase in the deficit whereas PPI in transport is associated with a delayed increase in the deficit. MACROECONOMIC EFFECTS OF PRIVATE SECTOR PARTICIPATION 169 11. This modeling strategy has been used before. See Fosu 2001, for example. 12. Goodness of fit measures in GLS models should be taken with caution. In particular, R2 has no clear interpretation in such context. References The word "processed" describes informally produced works that may not be available commonly through libraries. Aghion, Philippe, E. Caroli, and C. García-Peñalosa. 1999. "Inequality and Economic Growth: The Perspective of the New Growth Theories." Journal of Economic Literature 37: 1615­60. Alesina, Alberto, and R. Perotti. 1996. "Income Distribution, Political In- stability, and Investment." European Economic Review 40 (6): 1203­28. Amemiya, Takeshi. 1985. Advanced Econometrics. Oxford, U.K.: Blackwell. Arellano, Manuel. 1993. "On the Testing of Correlated Effects with Panel Data." Journal of Econometrics 59: 87­97. Breusch, T. S., and A. R. Pagan. 1979. "A Simple Test for Heteroskedas- ticity and Random Coefficient Variation." Econometrica 47: 1287­94. Chamberlain, Gary. 1984. "Panel Data." In Zvi Griliches and M. Intriligator, eds., Handbook of Econometrics, Vol. II. Amsterdam: North-Holland. De la Fuente, Angel. 2000. "Growth and Infrastructure: A Survey." Washington, D.C.: World Bank Institute. Processed. Estache, Antonio, V. Foster, and Q. Wodon. 2002. "Accounting for Poverty in Infrastructure Reform--Lessons from Latin America's Experi- ence." Development Studies. Washington, D.C.: World Bank Institute. Fosu, A. K. 2001. "Political Instability and Economic Growth in Develop- ing Economies: Some Specification Empirics." Economic Letters 70: 289­94. Gillen, D. W. 1996. "Transportation Infrastructure and Economic Devel- opment: A Review of Recent Literature." Logistics of Transportation Review 32: 39­62. Greene, William. 1995. LIMDEP v.7.0 User's Manual. New York: Econo- metric Software, Inc. Guasch, J. L. 2001. "The Impact on Performance and Renegotiation of Concession Design: Lessons from an Empirical Analysis of Ten Years of Ex- perience." Washington, D.C.: World Bank Institute, Studies in Development. Hausman, J. A. 1978. "Specification Tests in Econometrics." Economet- rica 46: 1251­71. 170 THE LIMITS OF STABILIZATION Kaufmann, Daniel, A. Kraay, and P. Zoido-Lobatón. 2000. "Governance Matters." Washington, D.C.: World Bank. Processed. IMF (International Monetary Fund). 2002. International Financial Statis- tics Yearbook. Washington, D.C. Laffont, J. J., and J. Tirole. 1998. A Theory of Incentives in Procurement and Regulation. Cambridge, Mass.: MIT Press. McGillivray, Mark, and O. Morrissey. 1999. Evaluating Economic Liber- alization. New York: St. Martin's Press. Munnell, Alicia. 1992. "Infrastructure Investment and Economic Growth." Journal of Economic Perspectives 6 (4): 189­98. Siniscalco, Domenico, B. Bortolotti, and M. Fantini. 2001. "Privatisation around the World: New Evidence from Panel Data." Fundazione Eni Enrico Matei Working Paper 77. Milan, Italy. World Bank. 2002a. World Development Indicators. Washington, D.C. ------. 2002b. World Development Report: Building Institutions for Mar- kets. Washington, D.C. 6 Regulation and Private Sector Participation in Infrastructure Sheoli Pargal THIS CHAPTER ASSESSES THE IMPORTANCE of the regulatory framework as a determinant of private sector investment in infrastructure, using re- cently compiled data on private and public sector investment in the water, power, telecommunications, roads, and railways sectors in nine large countries in Latin America.1 Controlling for standard determi- nants of investment, the impact of variables that represent different aspects of the prevailing regulatory regime on a country's ability to attract private investment in infrastructure is analyzed. During the last decades of the 20th century many countries in Latin America undertook public sector reform and introduced private partic- ipation in formerly state-dominated sections of their economies through management contracts, concessions, or outright privatizations. In the infrastructure sectors this was motivated by a desire to improve performance and increase efficiency in service provision, as well as by the fact that governments were constrained in increasing service cover- age or improving public utility performance by limited fiscal resources and a multitude of competing claims on these resources. But investment in infrastructure is characterized by large, up-front, usually sunk costs that lead to a high risk of expropriation, long gestation lags before rev- enues are generated, and revenues that are usually generated in local currency. These aspects lead to a need for both long-term commitment and long-term financing in local currencies. However, the limited depth of nascent capital markets is rarely able to generate funding of the ma- turity and volume necessary to finance private infrastructure invest- ment in Latin America. As a result, governments have made a concerted effort to attract foreign capital. 171 172 THE LIMITS OF STABILIZATION Analysts agree that an environment of macroeconomic and political stability and policy credibility and the existence of a sound regulatory framework are necessary for lowering the perceived risk of expropria- tion and thus for attracting private capital. In particular, the character of the entities entrusted with regulation determines confidence in the integrity of the system as a whole (see, for example, Kerf and others 1998). In this chapter, the amount of private investment attracted in each infrastructure sector in the countries studied is related to a set of independent variables that includes the characteristics of regulatory entities. This is a first attempt to test the assertion that the lack of in- dependent regulation can be a major hindrance to attracting private sector investment in infrastructure in developing countries. The study is of particular relevance for reforming countries because the Latin America and Caribbean region is, among the developing regions in the world, farthest along the road to deregulation of basic in- frastructure services. It faces second-generation issues of appropriate reg- ulation that others have yet to encounter. By characterizing regimes in terms of their ability to attract private investment in infrastructure, the analysis provides an empirical foundation for policy choices related to in- stitutional structure and regulatory frameworks. In accord with intu- ition, the results are consistent with the idea that government action to increase regulatory certainty and to minimize the perceived risk of ex- propriation through the establishment of independent regulatory bodies is a critical determinant of the volume of private investment flows. This chapter is organized as follows. The next sections provide background on the broad experience of the countries being studied and the approach taken here to assess the quality of the regulatory environment. The data are then discussed and, in the final sections, the estimation strategy and empirical findings of the analysis are described and conclusions presented. Private Investment in Infrastructure in Latin America The study covers Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Mexico, Peru, and República Bolivariana de Venezuela between, roughly, 1980 and 1998. Average annual public and private invest- ment levels by sector and country before and after the passage of leg- islation permitting private entry are shown in appendix table 6A.5. The figures in the chapter display the evolution of private and public investment by sector and country over the period studied. Almost all the countries in the sample had passed reform legislation by the mid-1990s in the telecommunications, power, and roads sectors. REGULATION AND PRIVATE SECTOR PARTICIPATION 173 The year in which legislation was passed in these countries varies widely, with Argentina and Chile being the earliest movers--the first infrastructure regulatory agency in Chile (Subtel or Subsecretaría de Telecomunicaciones) was established in 1977. Liberalization (through the introduction of competition and private sector participation) has also been deepest and most wide ranging in Argentina and Chile. For example, these are the only countries to enact legislation opening the water sector to private investment. A point of interest is that the passage of legislation permitting pri- vate entry has not always been necessary for the private sector to invest in different sectors in this region. In Bolivia, for instance, three railway concessions were granted in 1996 although the relevant legislation was enacted only in 1998. In general, however, enacting legislation for- malizes the sector liberalization and makes it less likely that the open- ing up will be reversed. The striking increase in average annual investment in telecommuni- cations and electricity and the large jump in the share of private sector investment in these sectors and in the roads sector following liberaliza- tion are illustrated in figures 6.1 and 6.2. In telecommunications and power the average annual share of public spending in total investment spending after liberalization declined to significantly less than 50 per- cent almost across the board. Prior to the opening of the sector, private investment in telecommunications was negligible or actually zero. Fol- lowing liberalization, private investment increased and public invest- ment declined, so that by 1998 private investment exceeded public spending in the sector in almost all the countries studied. Most countries in the sample granted initial exclusivity periods to privatized telecom- munications firms or set limits on entry into the sector.2 In general, the liberalization of access and tariffs came later. Similarly, between 1980 and 1998, public investment in the power sector generally declined whereas private investment increased, ultimately exceeding public spending in that sector. Liberalization in the power sector was usually accompanied by the restructuring of vertically integrated utilities. Although legislation permitting private entry into the roads sector had been approved in five of the nine countries by 1993, the role of the state continued to be substantial even after 1993. With the striking exception of Mexico, where an ambitious toll-road program was launched in the late 1980s, annual public investment in roads far sur- passed private investment in most Latin American countries. Private investment in roads increased slowly--the share of the private sector in total investment after liberalization typically remained below 50 per- cent. Public investment in the water sector also greatly exceeded private investment throughout the period, reflecting the limited liberalization 174 THE LIMITS OF STABILIZATION Figure 6.1 Annual Average Share of Private Investment in Total Infrastructure Investment in Selected Latin American Countries, by Sector a. Telecommunications Percent 100 90 80 70 60 Before liberalization 50 After liberalization 40 30 20 10 0 Peru Argentina Bolivia Brazil Chile de ColombiaEcuador Mexico Venezuela, R.B. b. Electricity Percent 100 90 80 70 60 Before liberalization 50 After liberalization 40 30 20 10 0 Peru Argentina Bolivia Brazil Chile de ColombiaEcuador Mexico Venezuela, R.B. c. Roads Percent 100 90 80 70 60 Before liberalization 50 After liberalization 40 30 20 10 0 Chile Peru gentina Bolivia Brazil lombia uador Mexico ezuela,de REGULATION AND PRIVATE SECTOR PARTICIPATION 175 Figure 6.2 Average Annual Investment in Infrastructure in Se- lected Latin American Countries, by Sector a. Telecommunications US$ million (1992) 5,000 4,500 4,000 3,500 3,000 Before liberalization 2,500 After liberalization 2,000 1,500 1,000 500 0 Peru Argentina Bolivia Brazil Chile de Colombia Ecuador Mexico Venezuela, R.B. b. Electricity US$ million (1992) 6,000 5,000 4,000 Before liberalization 3,000 After liberalization 2,000 1,000 0 Peru Argentina Bolivia Brazil Chile de Columbia Ecuador Mexico Venezuela, R.B. c. Roads US$ million (1992) 2,000 1,800 1,600 1,400 1,200 Before liberalization 1,000 After liberalization 800 600 400 200 0 Peru Argentina Bolivia Brazil Chile de Colombia Ecuador Mexico Venezuela, R.B. 176 THE LIMITS OF STABILIZATION of the sector in Latin America. This was true for private investment in Chile as well. However, driven largely by the concession contract for Buenos Aires, private investment in water in Argentina grew to exceed public spending during the 1990s. Private investment in Bolivia's wa- ter sector also rose to substantially exceed public spending, particu- larly after 1991--even in the absence of enabling legislation. A possible explanation for the difference in the average share of pri- vate investment in roads and water compared with power and telecommunications may lie in the natural monopoly elements of the road and water sectors. The low traffic density of rural and secondary road networks means that they are less amenable to cost-covering tar- iffs and thus less attractive to private concessions. As a result the need for public spending on roads is likely to continue. A similar argument relating to the cost of provision and scale economies has been applied to rural water systems. It is also often po- litically difficult to auction off the responsibility to ensure road access or the responsibility for basic water service to the private sector be- cause of the perception that these are core infrastructure services that the state should provide. Finally, water and power are the sectors in which the need for tariff adjustment is usually most pressing. This makes the political economy of private entry especially relevant in these sectors. Particularly in the case of water, where the general pub- lic (and not a small group) is often affected, anecdotal evidence points to the political difficulty of raising tariffs to cover costs. In fact, such tariff increases led to the failure of several concession contracts in the late 1990s. Analytical Approach Although there is a large body of empirical literature on the determi- nants of investment, including investment in infrastructure, this litera- ture has focused mainly on testing traditional economic theories of in- vestment behavior (see Everhart and Sumlinksi 2001 for a recent overview) rather than on assessing the contribution of the regulatory framework to the investment environment. Recent empirical work, however, has demonstrated the critical role of the institutional environment in determining the magnitude of investment flows. For instance, the option approach to investment reviewed by Servén (1996) underlined the deterrent effect of uncertainty on private investment, especially when investment is sunk. Investor perceptions about the probability of reform reversal are often a key REGULATION AND PRIVATE SECTOR PARTICIPATION 177 determinant of their willingness to invest. Lack of sustainability and credibility of reform can thus be a self-fulfilling expectation leaving countries in a low-level investment equilibrium. The general lesson from this analytical and empirical literature is that the stability and predictability of the incentive framework may be even more important than the level of investment incentives in determining the level of in- vestor confidence. Econometric work by Wallsten (2001a, 2002) on telecommunica- tions reform in developing countries is relevant to the analysis de- scribed in this chapter. Wallsten (2001a) found that country-level telecommunications performance is positively related to regulation, as measured by a dummy indicating whether the country had established a separate telecommunications agency not directly under the control of a ministry. That study used a fixed effects approach to explore the im- pact of privatization, competition, and regulation on telecommunica- tions performance in Africa and Latin America between 1984 and 1997. Wallsten (2002) showed that countries that established separate regulatory authorities prior to privatization saw increased telecommu- nications investment compared with countries that did not and that in- vestors are willing to pay more for telecommunications firms in such countries. These findings are consistent with the hypothesis that investors re- quire a risk premium to invest where regulatory rules remain unclear. The analysis conducted for this book contributes to the literature on the role of regulation in private sector development. Detailed data on meas- ures of regulatory independence were used in addition to data on both private and public investment in five sectors across the major economies of Latin America. This level of analysis allowed a more thorough as- sessment of the importance of independent regulatory institutions on the climate for private investment. The effectiveness of regulatory institutions depends on the structure and process of regulation, key aspects of which are the independence, competence, and clarity of mandate of the regulatory agency; the transparency and openness of the regulatory process; and the existence of formal oversight and timely judicial review.3 Smith (1997a, 1997b), discussing the desirable attributes of utility regulators, considered in- dependence from the regulated firm, customers, and political authori- ties essential. He underlined the important tradeoff between the need to limit regulatory discretion (as, for example, through regulation by contract) to reduce the risk of expropriation and the need to retain the flexibility to respond to new environmental and market conditions (for example, in rapidly changing sectors like telecommunications). The 178 THE LIMITS OF STABILIZATION openness and transparency of the regulatory process lessen the proba- bility of capture by different interest groups. An important additional consideration is the country's stability and reputation for respecting private property rights--which can go a long way in assuaging in- vestor concerns and thus allowing the regulator to retain substantial discretion without significantly increasing the cost of capital. But the ultimate accountability of the regulator is critical. In this chapter the regulatory environment in the countries and pe- riods under study are described in terms of the following four dimen- sions: · The passage of legislation that permits private investment in sec- tors traditionally reserved for the public sector, and the existence of a regulatory body. The passage of enabling legislation is particularly im- portant because Latin American regulatory frameworks are rooted in civil law. · The autonomy of the regulator. Autonomy or independence is captured by its attributes--the location of the regulatory body outside the government; a separate source of funding (that is, independent of the vagaries of annual budgetary appropriations); and popular support, involving both the legislative and executive branches in the appoint- ment process. Lacking data, it was impossible to assess the importance of aspects of independence such as security and length of tenure of reg- ulators (with staggered terms that are not coincident with the electoral cycle). · The size of the regulatory agency, with a larger body limiting the probability of capture by different interest groups (as well as the gov- ernment). Whether the prospect of being able to capture the agency would make a smaller agency more attractive to private investors is an empirical question. A larger size would allow for a range of profes- sional expertise and diversity of opinion (see, for example, Smith 1997a, 1997b, and 1997c), both critical to the competence of the agency. Commentators have argued, however, that a smaller agency could be more efficient in decisionmaking and more predictable, and that individual regulators might be more accountable than those in large commissions, which would make a smaller agency more attrac- tive from the investor point of view. All these factors would suggest a positive relationship between size and private investment flows. · The degree of risk borne by the investor as measured by whether the tariff regime is rate of return or price cap. Rate-of-return tariff reg- ulation limits the risk taken by the investor vis-à-vis a price-cap regime and thus might be positively related to private investment in infra- structure. Also, Alexander and Irwin (1996) have presented evidence REGULATION AND PRIVATE SECTOR PARTICIPATION 179 that price-cap regulation, by subjecting firms to greater risk, increases the cost of capital. Data Data sources are described in table 6.1. Macroeconomic data are taken from the World Bank's World Development Indicators database and the International Monetary Fund's International Financial Statistics (IFS). Investment data by sector were obtained from the database de- scribed in chapter 2 (appendix 2A). Data on regulatory variables (see table 6A.6) were obtained from Guasch (2001). A physical measure of the infrastructure capital stock each year by sector is used as a control variable. This consists of the following: for roads, total road length and paved road length; for railroads, total length of the rail network; for telecommunications, the number of tele- phone main lines; for energy, the electric generating capacity in kilo- watts; and for water, the growth in the percentage of the population with access to clean water. Pritchett (2000) has pointed out that stan- dard expenditure-based units of capital, particularly public capital, are often inaccurate in what they measure. Especially when it comes to the public sector and in countries where the government is a large investor, the divergence between investment effort and public sector capital stock is very high.4 This divergence renders suspect analyses that equate public spending on infrastructure with the value of infrastructure capital. With that caveat in mind, physical measures of capital stock were chosen for use as controls, even though they are not comparable across sectors. The regressions include a dummy that takes on the value 1 in years following the passage of legislation permitting private investment in Table 6.1 Data Sources Variable Data source Real gross domestic product (GDP) WB World Development Indicators Investment deflator WB World Development Indicators GDP deflator WB World Development Indicators Interest rate IMF IFS Real public investment See appendix 2A, chapter 2 Real private investment See appendix 2A, chapter 2 Regulatory variables Guasch 2001 Physical capital stock See appendix 2A, chapter 2 180 THE LIMITS OF STABILIZATION utilities (because these sectors were often considered the prerogative of the state). Even though private entry had begun prior to the passage of relevant legislation or the setting-up of formal legal and judicial frameworks for private participation in some countries, the passage of legislation (rather than the earliest private entry into the sector in each country) is used as the measure of liberalization because there is greater certainty implied by the existence of a formal legal basis for pri- vate investment. Summary descriptive statistics for the entire data set and for the set of variables measuring regulatory structure are included in appendix tables 6A.1 and 6A.2. Because, for the most part, liberalization and the development of regulatory frameworks started only in the 1990s, there are substantially fewer observations on the regulatory variables. Appendix table 6A.3 is the correlation matrix for the complete data set and indicates how different determinants of private invest- ment flows in infrastructure hang together. Public investment and private investment are significantly negatively correlated, support- ing the idea that they are overall substitutes. Private investment is also significantly positively correlated with the dummy for the pas- sage of reform legislation and with the existence of a regulatory body. The passage of reform legislation and the existence of a regu- latory body are highly positively correlated but not perfectly so-- reform legislation had been passed in only 40 percent of sector- country combinations prior to the establishment of a regulatory authority. Appendix table 6A.4 is the correlation matrix for the set of variables measuring aspects of the regulatory regime. The corre- lation between private investment levels and the passage of legisla- tion opening the sector to private investment is significantly more positive in this subset of the data. Estimation In the two basic models examined in this chapter, fixed effects regres- sions are used to explore the relationship between different groups of independent variables and private infrastructure investment. In all models the dependent variable is the log of real private sector invest- ment by country, year, and infrastructure subsector. The first model examines the determinants of private infrastruc- ture investment using a dummy for whether a regulatory body ex- isted that year and a dummy for whether enabling legislation had been passed by that year as the only indicators of the regulatory en- vironment. This model is also estimated for the four major sectors REGULATION AND PRIVATE SECTOR PARTICIPATION 181 separately--telecommunications, roads, electricity, and water--to capture sector-specific idiosyncrasies. The second model is estimated for the years during which a regulatory body exists. This permits the inclusion of characteristics of the regulatory regime as explanatory variables in the analysis and provides an opportunity to assess the impact of the type of regulatory regime on private investment flows to infrastructure. The reduced form equation being estimated, for each country i, sec- tor j, and year t is Ipijt f (Igijt, GDPit , rijt, pijt, Kijt , Rij Dijt) 1 1 , where Ipijt private sector investment, Igijt public sector investment, GDPit 1 gross domestic product lagged, rijt real rate of interest, pijt price of investment goods, Kijt 1 previous period physical cap- ital stock in the sector, Rij regulatory regime, and Dijt dummy for whether a reform law had been passed. An agnostic stance is taken about whether public sector investment is complementary to or a substitute for private investment in the sec- tor, noting the lack of consensus on this issue in the literature. Lagged capital stock would be expected to be negatively related to investment based on standard accelerator theories as well as on marginal pro- ductivity and cost of capital arguments. The sign on lagged GDP is expected to be positive because higher income should lead to greater capacity to invest, and more investment should also lead to an in- crease in incomes over time. The real rate of interest is included to capture the impact of the cost of financing on investment decisions, and a measure of overall investment goods prices is included to account for the actual cost of investment.5 Both these price variables reflect the opportunity cost of capital and are expected to have a neg- ative relationship with the amount of private investment attracted to a particular sector. The existence of a regulatory body and the passage of reform legis- lation would both be expected to be positively related to the volume of private investment flows because both represent government commit- ment to constraints on its own power. This results in less scope for dis- cretionary or arbitrary action (for example, against investor interests) and thus would imply a more certain business environment. The no- tion that investors require a risk premium to invest when regulatory rules remain unclear is supported by Wallsten (2002), who found greater investor willingness to pay for telecommunications firms in countries that have established regulatory authorities. 182 THE LIMITS OF STABILIZATION In the second set of regressions the dummy for the passage of reform legislation is augmented by the following regulatory indicators: whether regulatory decisions involve ministerial participation or the regulatory body is part of a ministry, whether the appointment of regulators involves both the legislature and the executive branch or only the executive branch, the size of the regulatory body, whether the regulatory body is funded solely by the government, and whether the tariff regime is rate of return. Most of these variables capture the degree of autonomy of the regulator from the executive branch and the susceptibility of the regulatory regime to government control or sub- version by capture (for example, by regulated entities). One would expect that a regulatory body being housed in a ministry or any ministerial involvement in decisionmaking would be negatively related to private sector confidence because an arm's-length relation- ship between the regulator and the government is generally desired. Likewise, regulators who are appointed by the executive branch of the government and those who are entirely dependent on the government for funding are unlikely to be independent. Low autonomy is expected to be negatively related to investor confidence and private investment flows. The expected sign on the coefficient of agency size is ambigu- ous. To the extent that a larger agency is likely to be more balanced and competent and less likely to be captured, a positive relationship between the size of the regulatory body and private investment might be expected, with the caveat that ease of capture might be attractive in some governance contexts. Empirical Findings Exploring the determinants of private investment over time uses the log of annual private sector investment (in millions of 1992 U.S. dollars) in each sector and country (for example, telecommunica- tions in Argentina in 1995) as the dependent variable. Economy- wide control variables are the log of public investment in the sector and country each year, real GDP lagged one year, one-period-lagged physical capital stock in the sector in the country, the real price of investment goods, the real rate of interest, whether a law permitting private entry has been passed, and whether a regulatory body for the sector exists in the country. A time trend and sector dummies (omitted sectors are railways and gas) are included and the data are pooled over all years, sectors, and countries covered. The estimation accounts for the panel structure of the data by putting in country fixed effects. REGULATION AND PRIVATE SECTOR PARTICIPATION 183 The results of regressions are reported in which both the investment price and the real rate of interest are included, although including the real rate of interest leads to the loss of some 200 observations. This is because comparable real interest rate data were not available for some countries during the earliest years covered in this analysis.6 Determinants of Private Sector Investment The results of the base regressions are presented in Models 1 and 2 in table 6.2. Model 1 is largely consistent with expectations. The overall relationship of private investment and public investment is one of substitutability. As might be expected, private investment is positively Table 6.2 Country Fixed Effects Estimation Model 1 Model 2 Control variable Coefficient t-statistic Coefficient t-statistic Log of real public investment ­0.1201513 ­2.033** ­0.3324242 ­3.109** (ln_pub) Log of lagged real GDP (ln_lgdp) 4.607274 6.695** 6.830853 2.893** Trend ­0.2715046 5.483** ­0.3513241 2.157** Lagged capital stock (lag_k) 2.94e­07 2.488** ­1.80e­07 ­0.482 Regulatory body in place (Rbexist) 0.2573572 0.497 Real rate of interest (Rroi) ­0.0118967 ­3.775** ­0.0131696 ­0.420 Investment price (Invprice) ­0.131716 ­0.112 0.7038148 0.145 Dummy: passage of legislation 3.640372 7.026** 6.087679 5.717** opening the sector (Dreform) Dummy: telecommunications 0.5296759 0.819 6.806443 3.200** Dummy: roads 0.6889568 1.258 1.643749 1.179 Dummy: water ­2.344596 ­4.135** 1.604541 0.982 Dummy: electricity 0.9701421 1.789* 1.725853 1.181 Regulatory body inside the ministry 4.018742 2.858** (Rbminis) Dummy: appointment of regulator ­5.512656 ­2.370** approved by legislature (Drbelec) Number of members of regulatory 0.2788061 1.556 commission (Rbnum) Dummy: regulator's budget solely ­5.512654 ­3.489** from government (Rbudgov) Dummy: rate-of-return legislation ­0.9717241 ­0.472 (D_ror) Constant ­55.26021 ­6.700** ­86.95147 ­3.107** Number of observations 693 183 R2 within 0.5122 0.6469 Note: The dependent variable is the log of real private investment. * Significant at 10 percent. ** Significant at 5 percent. Source: Data set used for the analysis (appendix 2A). 184 THE LIMITS OF STABILIZATION related to past period real GDP, indicating that richer economies generate greater private investment flows. A 1 percent increase in pre- vious period real GDP is associated with an increase of 4.6 percent in private investment levels. Lagged real GDP was used as the explana- tory variable to reflect the potential causal relationship between GDP and private investment.7 Investment volume is negatively related to the real rate of interest and the price of investment goods but significantly positively related to whether legislation enabling private entry has been passed--the mere act of passing legislation liberalizing private entry into a sector increases private investment by 3.6 percent. The dummy for passage of such leg- islation absorbs a fair amount of the effect of having a regulatory body in place, and indicates that in many cases the legal basis for private entry is probably more important than the actual institutional frame- work governing private sector participation.8 Using a dummy to cap- ture the opening of the sector is limiting in that it does not capture crit- ical elements of the post-opening and post-privatization competitive environment, which would affect incentives to invest. Whether a firm facing competition is likely to invest more or less than a monopoly is an empirical question. In several cases Latin Amer- ican state-owned infrastructure firms were privatized as monopolies or granted exclusivity periods of varying lengths, as in the telecommuni- cations sector. As shown by Wallsten (2001b) in his study of telecom- munications privatization in developing countries, granting a monop- oly concession seriously reduces investment by the privatized firm relative to firms that face competition. Unfortunately such data were not available for most of the sectors and countries studied.9 The estimation includes previous period capital stock because the coefficient on it is highly significant, although it is a physical measure that varies by sector and is not easy to interpret. The coefficient on capital stock is positive, which is contrary to what theory would sug- gest. However, this appears to be an artifact of aggregation because it is uniformly negative in the regressions disaggregated by sector.10 Controlling for other factors, the water sector received significantly less private investment than other sectors, whereas private investment in power was higher than in the other sectors. Characteristics of the Regulatory System Given that a regulatory body exists, what aspects of the regulatory structure are critical to attracting private investment in infrastructure? Model 2 in table 6.2 presents the results of the fixed effects regression REGULATION AND PRIVATE SECTOR PARTICIPATION 185 restricted to the years after a regulatory body had been established. The passage of legislation opening the sector is still a significant and positive determinant of the volume of private investment flows, and public investment is clearly being replaced by private investment. Al- though the sign and significance of the other economy-wide variables are unchanged, the real rate of interest and the investment price index are no longer significantly different from zero. Also, after controlling for regulatory factors, the telecommunications sector attracts signifi- cantly more private investment than do the others. For the regulatory variables, some results require further explo- ration. For instance, private investment volumes are significantly pos- itively related to the regulatory body being located inside a ministry or to ministerial involvement in decisionmaking.11 In addition, systems in which regulators are appointed by the executive branch are associated with greater private investment than if the selection of the regulatory body goes through both the legislature and the executive branch. On the face of it, both these aspects of the regulatory structure should mil- itate against private investor interest because they imply lower con- straints on the government's power to expropriate the value of an in- vestment. On more reflection, however, it seems that these results may underline the critical need of investors for regulatory predictability and credibility. For instance, investors are likely to expect that decisions made by a regulator housed in a ministry will not be overturned. In ad- dition, a regulatory body appointed by the executive may be consid- ered stronger by virtue of having the full power of the executive branch behind it, and be perceived to speak with a clearer voice than a regu- latory body whose appointees have to go through approval by the leg- islature. A natural question is whether these considerations are specific to Latin America. They may result from the historical existence of gener- ally strong executive branches on the continent. Such arrangements may thus increase investor certainty about government intentions and could result in more private investment than otherwise would occur. It is particularly interesting that, consistent with intuition, private investment is positively associated with the regulator not being funded solely by the government. This is an important element of regulatory independence. In addition, the number of commissioners or regulators is positively related to investment volumes (significant at 12 percent), reflecting the possibly greater independence, broader expertise, and lower likelihood of capture of a larger commission.12 Although not significant, investment volumes are negatively related to rate-of-return tariff regulation that limits both downside and upside risk. Alternative tariff regulatory mechanisms such as price caps would provide the 186 THE LIMITS OF STABILIZATION opportunity to earn a higher rate than the fixed rate of return, which would compensate for the greater risk of investment in sectors like infrastructure. Sector-Specific Results Table 6.3 presents the results of the base regression for the telecommu- nications, roads, electricity, and water sectors separately. Multicollinear- ity among the indicators of regulatory structure and lack of variation of these variables over time within each country led to the decision not to run sector-specific models with regulatory variables included. Public and private investment in telecommunications and power are strong substitutes. A somewhat weaker but still substitutable relationship is observed in the water sector. In roads the relationship is complementary but not significant. A complementary relationship between public and private investment in the roads sector would indeed be expected because of the difficulty of obtaining private financing for nonprimary highways.13 In all cases higher lagged GDP is associated with higher private investment volumes and, in contrast to the find- ings of Models 1 and 2, lagged capital stock is negative and significant in all the sectoral regressions apart from the one for water. The passage of legislation opening the sector to private entry is sig- nificantly positively related to private investment volumes in telecom- munications and roads but less so in the power sector and virtually not at all in the water sector. The result on power is somewhat surprising because liberalizations of telecommunications and power have been deeper and wider than those of the other sectors, with a decline in the importance of government investment going hand in hand with the in- crease in private participation. The regulatory regime might also have been expected to be more critical for power than for telecommunica- tions given the greater contestability of the latter. As in the regression on pooled data (Model 1), the existence of a regulatory body is not a significant determinant of private investment in any sector after con- trolling for the passage of enabling legislation. This leads to the question of whether the water sector differs in some crucial respect from other infrastructure sectors. As mentioned earlier, the natural monopoly aspects of water distribution and trans- mission are stronger than in other utilities. Also, opening up the water sector to private investment, which may require an increase in tariffs to cover costs, tends to be politically more difficult than liberalization in nonessential sectors.14 Investors may thus expect greater scrutiny for water sector investments. -statt 1.55­ 4.003** 2.852** 0.000 0.292 0.798 3.493**­ 1.476 3.974**­ Water 109 1.0194­ 6.4141 0.3059 0.0018 0.3486 1.7566 0.0205­ 1.8979 0.5274 Coefficient 76.1823­ -statt 2.500**­ 3.435** 4.703** 4.101**­ 0.202 0.783 1.620­ 1.780* 2.912**­ Power