WORLD BANK TECHNICAL PAPER NO. 488 WTP488 Work in progress for public discussion August 2000 Governance Impact on Private Investment Evidence from the International Patterns of Infrastructure Bond Risk Pricing Nina B. Bubnp s e Ninrl B. Bllbno-va~~~~~~~~~~~~~~~~~~~~~~~~~~~~v Recent World Bank Technical Papers No. 406 Milazzo, Subsidies in World Fisheries: A Reexamination No. 407 Wiens and Guadagni, Designing Rulesfor Demand-Driven Rural Investment Funds: The Latin American Experience No. 408 Donovan and Frank, Soil Fertility Management in Sub-Saharan Africa No. 409 Heggie and Vickers, Commercial Managemnent and Financing of Roads No. 410 Sayeg, Successful Conversion to Unleaded Gasoline in Thailand No. 411 Calvo, Optionsfor Managing and Financirg Rural Transport Infrastructure No. 413 Langford, Forster, and Malcolm, Toward a Financially Sustainable Irrigation System: Lessonsffrom the State of Victoria, Australia, 1984-1994 No. 414 Salman and Boisson de Chazoumes, International Watercourses: Enhancing Cooperation and Managing Conflict, Proceedings of a World Bank Seminar No. 415 Feitelson and Haddad, Identification of Joint Management Structuresfor Shared Aquifers: A Cooperative Palestinian-Israeli Effort No. 416 Miller and Reidinger, eds., Comprehensive River Basin Development: The Tennessee Valley Authority No. 417 Rutkowski, Welfare and the Labor Market in Poland: Social Policy during Economic Transition No. 418 Okidegbe and Associates, Agriculture Sector Programs: Sourcebook No. 420 Francis and others, Hard Lessons: Primary Schools, Community, and Social Capital in Nigeria No. 421 Gert Jan Bom, Robert Foster, Ebel Dijkstra, and Marja Tummers, Evaporative Air-Conditioning: Applications for Environmentally Friendly Cooling No. 422 Peter Quaak, Harrie Knoef, and Huber Stassen, Energyfrom Biomass: A Review of Combustion and Gasifica- tion Technologies No. 423 Energy Sector Unit, Europe and Central Asia Region, World Bank, Non-Payment in the Electricity Sector in Eastern Europe and the Former Soviet Union N!To. 424 Jaffee, ed., Southern African Agribusiness: Gaining through Regional Collaboration No. 425 Mohan, ed., Bibliography of Publications: Africa Region, 1993-98 No. 426 Rushbrook and Pugh, Solid Waste Landfills in Middle- and Lower-Income Countries: A Technical Guide to Planning, Design, and Operation No. 427 Marino and Kemper, Institutional Frameworks in Successful Water Alarkets: Brazil, Spain, and Colorado, USA No. 428 C. Mark Blackden and Chitra Bhanu, Gender, Growth, and Poverty Reduction: Special Program of Assistance for Africa, 1998 Status Report on Poverty in Sub-Saharan Africa No. 429 Gary McMahon, Jose Luis Evia, Alberto Pasc6-Font, and Jose Miguel Sanchez, An Environmental Study of Artisanal, Small, and Medium Mining in Bclivia, Chile, and Peru No. 430 Maria Dakolias, Couirt Performance around the World: A Comparative Perspective No. 431 Severin Kodderitzsch, Reforms in Albanian Agrictulture: Assessing a Sector in Transition No. 432 Luiz Gabriel Azevedo, Musa Asad, and Larry D. Simpson, Management of Water Resources: Bulk Water Pricing in Brazil No. 433 Malcolm Rowat and Jose Astigarraga, Laj'in American Insolvency Systems: A Comparative Assessment No. 434 Csaba Csaki and John Nash, eds., Regional and International Trade Policy: Lessons for the EU Accession in the Rural Sector-World Bank/FAO Workshop, ,rune 20-23, 1998 No. 435 lain Begg, EU Investment Grants Review No. 436 Roy Prosterman and Tim Hanstad, ed., Legal Impediments to Effective Rutral Land Relations in Eastern Europe and Central Asia: A Comparative Perspective No. 437 Csaba Csaki, Michel Dabatisse, and Oskar Honisch, Food and Agricultutre in the Czech Republic: From a "Velvet" Transition to the Challenges of EU Accession No. 438 George J. Borjas, Economic Research on the Determinants of Immigration: Lessonsfor the European Union No. 439 Mustapha Nabli, Financial Integration, Vulnerabilities to Crisis, and EU Accession in Five Central European Countries No. 440 Robert Bruce, loannis Kessides, and Lothar Kneifel, Overcoming Obstacles to Liberalization of the Telecom Sector in Estonia, Poland, the Czech Republic, Slovenia, and Hungary: An Overview of Key Policy Concerns and Potential Initiatives to Facilitate the Transition Process No. 441 Bartlomiej Kaminski, Hungary: Foreign Trade Issues in the Context of Accession to the EU (List continues on the inside back cover) WORLD BANK TECHNICAL PAPER NO. 488 Governance Impact on Private Investment Evidencefrom the International Patterns of Infrastructure Bond Risk Pricing Nina B. Bubnova The World Bank Washington, D.C. Copyright (© 2000 The International Bank for Reconstruction and Development/THE WORLD BANK 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. All rights reserved Manufactured in the United States of America First printing August 2000 12340403020100 Technical Papers are published to communicate the results of the Bank's work to the development community with the least possible delay. The typescript of this paper therefore has not been prepared in accordance with the procedures appropriate to formal printed texts, and the World Bank accepts no responsibility for errors. Some sources cited in this paper may be informal documents that are not readily available. The findings, interpretations, and conclusions expressed in this paper are entirely those of the author(s) and should not be attributed in any manner to the World Bank, to its affiliated organizations, or to members of its Board of Executive Directors or the countries they represent. The World Bank does not guarantee the accuracy of the data included in this publication and accepts no responsibility for any consequence of their use. The boundaries, colors, denominations, and other information shown on any map in this volume do not imply on the part of the World Bank Group any judgment on the legal status of any territory or the endorsement or acceptance of such boundaries. The material in this publication is copyrighted. The World Bank encourages dissemination of its work and will normally grant permission promptly. Permission to photocopy items for internal or personal use, for the internal or personal use of specific clients, or for educational classroom use, is granted by the World Bank, provided that the appropriate fee is paid directly to Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, U.S.A., telephone 978-750-8400, fax 978-750-4470. Please contact the Copyright Clearance Center before photocopying items. For permission to reprint individual articles or chapters, please fax your request with complete information to the Republication Department, Copyright Clearance Center, fax 978-750-4470. All other queries on rights and licenses should be addressed to the World Bank at the address above or faxed to 202-522-2422. ISBN: 0-8213-4818-3 ISSN: 0253-7494 Nina B. Bubnova is a consultant for the energy sector unit of the Europe and Central Asia Region of the World Bank. Library of Congress Cataloging-in-Publication Data has been applied for. Contents Foreword .......................... vii Acknowledgements .. ........................ viii Abstract ......................... ix Executive Summary .. ........................ xi. 1. Introduction ..........................1 2. Previous Research .......................... 3 3. Analytical Framework ................................ 6 Dependent Variables .......................................................... 7 Independent Variables . 8 4. Data ............................................. 10 5. What Drives Governance Risk Perception: Cluster Analysis of Political and Regulatory Risk Data ........................................ 14 Cluster Segregation of the Governance Risk Indicators ............................. 14 Country Clustering by the Level of Political Risk ................ .................. 21 6. Impact of Governance Risks on Investment Cost at Entry: Analysis of Cross-Sectional At-Issuance Spread Data .................. 24 Impact of Country (Macro)-Level Regulatory Risks ................................ 24 Model Application: Estimation of Infrastructure Bond Spreads in the Czech Republic and Hungary .3........ ...... ..4.. .. .................. 34 Industry-Level Regulatory Risks: The Cost of Regulatory Discretion .................. 36 7. True Market Perceptions of Governance Influence on Infrastructure Finance: Evidence from the Time-Series Bond Data .42 Risk Premium in the Secondary Market: Real Factors of Change ..................... 42 Regulatory Risk and Credit Rating History of Infrastructure Bonds ................... 46 8. Conclusions .......... 48 Methodology Appendix 1. The Mathematical Foundation of Biplot Analysis .......... 53 Methodology Appendix 2. OLS Spread Model for Cross-Sectional Data: Macro-Level Regulatory Risks Impact Measurement ................... 56 Methodology Appendix 3. Test for Universe Selection Bias-Truncated Regression .... 58 Methodology Appendix 4. Instrumental Variables and Two-Staged Least Squares Regression for Industry-Level Regulatory Risks Impact Measurement .................... 60 Methodology Appendix 5. Generalized Estimating Equations Model for Estimation of Secondary Market Bond Spread .62 Methodology Appendix 6. Multinomial Logit Technique for Estimation of the Probability of the Bond Credit Rating Change ...................... 64 iii Governance Impact on Private Investment Annexes A-G ........ 65 Notes ........ 74 References ..... 76 Figures Figure 1 Position of Regulatory Risks in Respect to Two Principal Components . 17 Figure 2 Bond Country of Issuance Position on Regulatory Risk Indicator Vectors . . .18 Figure 3 Regulatory Riskiness of Bond Issuer Countries .22 Tables Table 1 Research Data and Sources .13 Table 2 Composition of the Two Principal Components of the Regulatory Spread Data .16 Table 3 Clusters of Governance Risk Factors .19 Table 4 Correlation Coefficients of Main Regulatory Risk Indicator Clusters and the Log of Spread .21 Table 5 Country Classification by Regulatory Riskiness .22 Table 6 Spread Behavior in Three Regulatory Risk Country Groups .23 Table 7 Variables Selection for Step 2 OLS Regression Analysis of the Spread. 25 Table 8 Results of Log Spread Regression on Selected Independent Variables. 27 Table 9 Results of Spread Regression on Various Governance Risk Factors .32 Table 10 Coefficients of the End Result of Automated Stepwise Regression .33 Table 11 Registered Eastern European Bond Issues, 1990-98 (All Currencies) . 34 Table 12 Registered Western European Bond Issues, 1990-98 (U.S. Dollar Issues) . . .35 Table 13 Actual Average Spread Observations in Western and Central/Eastern European Countries, 1995-98 .35 Table 14 Model Estimation of Risk Spreads, W\orldwide and by Country, 1997 . 36 Table 15 Classification of Employed Power Sector Governance Framework Characteristics .38 Table 16 Correlation Coefficients of Power Industry Characteristics and the Log of Spread .38 Table 17 Instrumental Variable for Power Sector Regulatory Discretion .39 Table 18 Two-Stage Least Squares with Instrumental Variable Regression Results. 40 Table 19 Augmented Regression Test for Appropriateness of Instrumental Variable Approach .41 Table 20 Distribution of Industry Sectors of Bond Issuers .43 iv Contents Table 21 Distribution of Institutional Types of Bond Issuers ...................... 43 Table 22 Results of Historical Spread Regression ............................... 44 Table 23 Results of Historical Spread Regression, with Four Regulatory Risk Clusters ......................................................... 45 Table 24 Multinomial Logit Regression Results ................................. 47 Annex Tables Table A. 1 Correlation Analysis of Data Interrelationships-Correlation Clusters ..... 65 Table B. 1 OLS Regression of Log of Spread on Individual Regulatory Risk Factors . . .66 Table B.2 OLS Regression of Log of Spread on Individual Regulatory Risk Factors . . .67 Table C. 1 Definitions of Bond Issue Characteristics ............................. 68 Table D.1 Bond Issuer Characteristics Descriptive Statistics ....................... 69 Table E. 1 Macroeconomic Characteristics Descriptive Statistics .70 Table F. 1 Macroeconomic Variables Inner Correlation Coefficients .71 Table G. 1 Cluster Membership: Regulatory Risk Indicators .72 Table G.2 Cluster Membership: Countries by Level of Regulatory Riskiness .73 v Foreword The Energy Department of the World Bank's Europe and Central Asia Region has been under- taking analytical work on issues pertinent to the Bank-wide effort to confine corruption and strengthen governance frameworks in member countries in order to foster economic development, the provision of essential infrastructure services, and the expansion of private investment. This research is a step toward a better understanding of the nature of governance frame- works, the risks they place on private investment, and the barriers they raise to infrastructure development. The findings of this research are valuable both for public and private sector actors: assisting the former in the design of public policies aimed at the creation of attractive investment environments; and assisting the latter in making the appropriate strategic choic- es when entering infrastructure investment business transactions. Hossein Razavi Director Energy Department Europe and Central Asia Region vii Acknowledgements The author would like to express her gratitude to Messrs. Hossein Razavi, Director, and Laszlo Lovei, Lead Energy Specialist, Energy Department, ECA, for launching this research on the impact of governance in infrastructure finance and for their support during the project. Special thanks go to Ms.Jenifer Wishart and Messrs. Robert Bacon, Gregory Ingram, Eugene Gurenko, Philip Keefer, Michael Klein, Guy Pfefferman, Warrick Smith, Pablo Spiller, John Quigley, Oliver W\rilliamson, and seminar participants at UC Berkeley for their comments on previous versions. The author is grateful to Mr. Ilya Lipkovich, Virginia Polytechnical School, for his advice on the application of modern statistical theory to the analysis of the research. Funding for this research was generously provided by the grant of the Research Committee of the WAlorld Bank and by the Award of the Graduate Chancellor of the University of California, Berkeley. Nina B. Bubnova University of California, Berkeley viii Abstract During the last decade, infrastructure finance and provision graduated from traditional means to more innovative ones, primarily initiated by private companies and supported through their equity and debt. Capital markets increasingly became the main funding source for infrastructure projects worldwide, including investments in developing and transition coun- tries where infrastructure penetration still falls considerably short of needs. Infrastructure bonds served as the most popular method of oil, gas, electricity, telecommunications, and trans- port project financing in these countries throughout 1990-99, thereby substituting govern- ment funding. Thriving markets require not only an appropriately designed economic system, but also a secure political foundation that limits the ability of the state to confiscate wealth. This requires a presence of political institutions that credibly commit the state to honor economic and political rights. Investments in infrastructure are particularly susceptible to the risks of government interference. Constructive government noninvolvement-necessary regulation excepted-is likely to fail in societies that lack the institutions to serve as checks and balances on government action, as well as in those societies that do not possess a longstanding tradi- tion of law and order. These and other risks impinge on investor confidence, an effect exac- erbated by the recent global financial crisis. The need to boost private capital flow to infra- structure projects requires policy measures that will reduce investor perceptions of the risk of default due to adverse government action. This paper uses international cross-sectional and time-series infrastructure bond risk premium and credit rating history data from the past decade to examine the factors that influ- ence investor risk perceptions and that inflate the cost of borrowing for essential infrastruc- ture. The information thus generated about key governance risks is then analyzed for its pol- icy implications. ix Executive Summary Prior to the late 1980s, funds for long-term infrastructure investments in most emerging economies were provided only by or through government. Fiscal austerity measures and other factors have since caused most governments to cut their budgetary contributions to infra- structure, and private sector funding has emerged as a substitute for government funding: In accordance with the common practice of infrastructure finance in the industrialized world, bonds in particular have become a major conduit of private capital flow to infrastructure projects in emerging markets. The rapid growth of private capital investment has however been insufficient to satisfy market demand in these countries: an estimated 3.5 billion peo- ple worldwide do not have access to basic infrastructure services today.' One way to encourage greater private investment in infrastructure development is to reduce investor risk. In addition to macroeconomic, capital market, and firm-intrinsic fac- tors, the likelihood of infrastructure bond default is increased by nontransparent and unpre- dictable regulatory frameworks, weak systems of contract enforcement, and the lack of polit- ical stability. These factors are particularly significant because infrastructure investments are typically large, long-term, and irreversible, and because they are dependent on sales to domestic markets and on government regulatory intervention, which may include the setting of prices and quality standards. Such risks have not been taken into account in previous empir- ical analysis of private lending to infrastructure sectors. This research paper aims to contribute to our existing knowledge by providing a thor- ough examination of the effect that governance frameworks, both political and regulatory, have on investors' risk perceptions and on associated costs. It aims to achieve this by analyz- ing the rich cross-sectional data and time-series data generated by the universe of fixed-rate infrastructure bonds issued and traded during the past decade. The research considered such polity and governance aspects as the historical enforcement of the law, including the extent of regulatory discretion and the political constraints on that discretion; the presence of cor- ruption; the quality of bureaucracy; the ability of the system to enforce adherence to contracts; and the likelihood of expropriation and other political risks. It additionally identified those political and regulatory risks that most concern investors. Wthile most previous research has focused separately on the cases of the developing and industrialized worlds, this paper pro- vides a comparative analysis of developed and emerging infrastructure bond markets. This enables us to see how the factors that drive infrastructure finance in the two country groups differ, and thereby enables us to identify the policy implications of these differences. The methodological approach used is innovative in five areas: (1) it enhances the explanatory capacity of earlier credit risk models by combining firm-intrinsic and macro- economic factors and by adding the full spectrum of risks associated with the quality of gov- ernance and regulation in a country and sector of bond issuance; (2) it applies hierarchical xi Governance Impact on Private Investment cluster analysis to understand an array of current regulatory and political risk indicators and to discern the variables that drive the governance risk perception in a given country; (3) it complements analysis based on cross-sectional data by constructing a database of time-series yield data on infrastructure bonds traded in the secondary market and by examining the fac- tors that affect bond risk in real market conditions; (4) it investigates the role of industry struc- ture and regulation as applied to the power sector segment of infrastructure bonds; and (5) it compiles and analyzes the rating history of the entire bonds universe and uses the history of upgrade and downgrade events to investigate the impact of governance over the lifetime of an infrastructure bond. The research utilized the techniques of principal component analy- sis and cluster analysis; ordinary least squares analysis and two staged least squares with instrumental variables regression analysis; truncated and augmented regression analysis; and multinomial logit regression analysis. The research yielded five relaced, yet distinct conclusions. First, it confirmed the hypothesis that governance risk has a strong impact on the per- ceived default risk of a fixed income security, although the effects of the various political and regulatory risks on bond spreads are clearly not homogeneous. Cluster analysis segregated an array of 19 regulatory and political risk indicators into four risk groups: (1) political disorder; (2) corruption, red tape, and autocracy; (3) poor economic policy; and (4) fundamentalism. Empirical evidence suggests that what matters most to investors are the risks of cor- ruption, the poor quality of the bureaucracy, and the lack of institutional constraints on gov- ernmental action. As estimated by this research, governance-borne risks in emerging bond markets, calculated as payments that could have been avoided, cost investors about $677.7 mil- lion in 1997. (Note that dollar amounts in this paper are U.S. unless otherwise indicated.) These risks, classified here as corruption, red tape, and autocracy risks, are followed in importance by the traditional risks of political disorder caused by civil war, by the involvement in politics of the military, by political terrorism, by the lack of adequate law enforcement, and by nationalization policies. Investors appear to be least concerned by risks that stem from short- sighted economic policies and poor planning, as well as by so-called fundamentalism risks that are associated with the prevalence of organized religion in politics, with internal racial and nationality tensions, and with the risk of external conflict. In other words, investors are able through diversification to discount some incompe- tence in political and economic leadership and some erratic economic policies. Empirically, it appears that poor political leadership and economic mismanagement are now seen as diver- sifiable incompetence and no longer constitute a systemic threat to investors. But while investors can similarly make allowances for differences in the cultural and moral endowrnents of a country, they cannot discount the hazards that stem from systemic incentives to steal (cor- ruption) or to break commitments (poor contract enforceability), which reduce expected val- ues and require an offsetting spread. These findings on the hierarchical importance of the four risk clusters were reinforced by results of the time-series examination of bond credit risk structure in the secondary mar- ket, with the only difference being that all four governance risk clusters had a significant effect xii Executive Summary on the secondary market spread. Spreads in the aftermarket essentially measure the mone- tary compensation for the risk of bond default, as dictated by real market riskjudgement, as opposed to bond spreads at-issuance, which are determined by associated riskjudgement by analysts at the underwriting banks and at the institutional investment entities. Second, the research identified a clear separation of the sample universe of countries into three regulatory and political risk clusters, defined as those with stable, weak, and risky governance climates. These groups to a large extent correspond to what the development com- munity conventionally calls the "industrialized," "emerging," and "developing" worlds. Members of the "risky" governance cluster2-the least developed countries and the stumbling transition economies-generally have very poor governance and poor macroeconomic indi- cators, and have to pay high premiums to investors for taking on the risk. Countries from the .weak" cluster-the emerging economies3-have some of the fundamentals in place, such as appropriate fiscal and monetary policies and ownership structures, but face governance issues, particularly those that give rise to political and regulatory risks, as their toughest chal- lenge. The "stable" or industrialized countries4 are in the clear, with the risk premiums of bonds issued in these countries found to be unaffected by governance risks. Investors in the stable cluster countries can focus on real commercial project risks rather than on policy risks created by potentially adverse government actions. A noteworthy conclusion to be drawn from this finding is that economic reform remains a necessary but not a sufficient approach to development. As a case study, the research applied the model developed here to estimate the expect- ed risk return for infrastructure investors in two transition economies, the Czech Republic and Hungary, in 1997. The Czech Republic notably issued more infrastructure bonds than any other Central and Eastern European country; Hungary, in contrast, preferred other types of infrastructure financing. The model estimation indicated that the cost of borrowing in Hungary should be higher and in the Czech Republic lower than the actual average spread for emerging market countries in that year, in order to compensate potential investors for the political and regulatory risks that they would face. This may partly explain the difference in the two countries' choices of infrastructure finance sources. Third, the research found a substantial degree of variation in the impact of regulato- ry and political risks on the spread for bond issuers, according to their institutional and own- ership characteristics. There is clear evidence of asymmetry in the degree of exposure to adverse government action that bond issuers face, dependent on their ownership status. Specifically, private sector infrastructure projects were found to be more vulnerable to the effect of political and regulatory risks than were public projects. Simultaneously, public own- ership of an infrastructure project was found to set the influence of firm-intrinsic factors on the bond spread to zero, suggesting that questions of the economic efficiency and financial prudence of public companies are not taken into account when the risk of their infrastruc- ture projects is evaluated. In reality, the business weaknesses of public companies tend to be compensated for by the ability of governments to cross-subsidize those companies and to apply favorable tax breaks. xiii Governance Impact on Private Investment This research further studies the effects internationally of risks posed by industry struc- ture and regulation. By examining the risk premium structure of bonds issued for the financ- ing of power sector projects, the research established that there is a fine line between the ben- efits that accrue from the granting of regulatory independence and the hazards, as perceived by investors, that potentially stem from giving discretionary powers to the price regulator. Specifically, the research found that the absence of institutional constraints on the ability of regulators to set prices significantly increases the investment risk, as measured by the bond spread. Low levels of power sector privatization were similarly found to increase investor per- ceptions of risk. Finally, a study of the credit rating history of the sample universe of bonds enabled an examination of the factors underlying investment's creditworthiness stability and creditwor- thiness change. This indicated that stable governance frameworks are essential for a stable credit rating history, and that changes in firm-intrinsic characteristics are instrumental in caus- ing the upgrade or downgrade of a bond by the major credit rating agencies. xiv 1. Introduction Fueled by fiscal constraints and growing disenchantment with the performance of state-pro- vided infrastructure services, the trend of liberalizing and privatizing infrastructure services that began in the late 1970s and 1980s turned into a wave that swept the world in the 1990s (World Bank, 1999). Recognizing the ability of the private sector to provide cost-effective, high- quality solutions for the delivery of essential utility services, more and more governments turned to private sources for the financing and provision of telecommunications, energy, trans- port, and water services. Even the traditionally state-ownership-oriented developing countries experienced a growth in private activity in infrastructure. Investment in infrastructure pro- jects with private participation grew dramatically from about US$16 billion annually to US$120 billion5 in 1990-97; although it declined by about one-fifth to US$95 billion in 1998, as a repercussion of the financial crisis that began in Asia, the World Bank estimates that the private sector today contributes about 40 percent of the total infrastructure investment in devel- oping countries. Increasingly, it is private capital rather than public expenditures that has funded infra- structure projects and firms. In accordance with the common practice of infrastructure finance in the industrialized world, bonds in particular became a major conduit of private capital flow to infrastructure projects in emerging markets in the 1990s. Despite its rapid growth, however, the extent of private capital investment in these markets has been insuffi- cient to satisfy demand: an estimated 3.5 billion people worldwide still do not have access to basic infrastructure services.6 The flow of private investment is extremely sensitive to political hazards associated with the unpredictability of future government actions. This problem is particularly acute in infrastructure development, because infrastructure investments are typically large, long- term, and irreversible, and they are usually dependent on sales to domestic markets. Government involvement is also heavier in these sectors, as manifested in the regulation of entry prices and in quality and environmental standards. Infrastructure services are addi- tionally often essential to the consumer, but are frequently provided by monopolists. Together, these factors increase political sensitivity to the prices charged. Pressure from consumers to keep prices low makes it politically difficult for governments to maintain prices that cover costs. Regulatory risks, defined as those risks arising from the application and enforcement of reg- ulatory rules, both at the economy-wide and at the industry- or project-specific level, includ- ing rules contained in contracts with governments, in laws, and in other regulatory instru- ments, can deter investors, increase the cost of investment capital, increase required tariffs, and/or reduce the proceeds from privatization. In addition, infrastructure projects typical- ly require large sunk investments that may take 10 to 30 years to recoup. Over such long peri- ods of time investors are exposed to serious risks, in particular the risk that public authori- 1 Governance Impact on Private Investment ties will not honor their agreements on tariff policy and payments to investors. Once investors are committed to a project and able to pull out only by taking on a huge loss, the government may be tempted to exercise administrative expropriation through lowering prices or not rais- ing them as agreed. One consequence of the 1997 financial crisis in developing countries in fact has been the deterioration of demand for and supply of private infrastructure, with many high-level projects cancelled and growth forecasts for developing countries revised downward. Regulatory risks such as these are further supplemented by broader-based macro-level polit- ical risks associated with poor governance and inadequate enforcement of law and order. Such risks include the familiar political risks of expropriation and nationalization, the uncertain- ty that stems from weak systems of contract enforcement and the absence of an independent judiciary, corruption, and poor bureaucratic quality. These problems together discourage potential investors from investing in environments that lack credible government commitment. However, it is not always clear what investors are looking for and what-apart from sustainable macroeconomic and ownership structure reforms-should therefore be the preconditions to encourage private investment and the suc- cessful privatization of infrastructure services. This paper uses the yield spread as a measure of credit risk to investigate the effects of various aspects of governance-including the pres- ence of corruption, quality of the bureaucracy, contract enforceability, law and order tradi- tion, expropriation, and other risks-on the perception of bond default risk. Of these, it seeks to identify those factors that most concern investors. The paper also seeks to enhance the explanatory capacity of earlier credit risk models by considering the full spectrum of inde- pendent variables: firm- and bond-intrinsic characteristics, macroeconomic factors, and the range of risks associated with the quality of governance and regulation in a country and sec- tor of bond issue. The impact of regulatory risks is further investigated by examination of time- series data on spreads of those infrastructure bonds that were traded in the aftermarket. The paper is organized into eight sections. Section 2 discusses prior research in this area. Section 3 elaborates the conceptual framework of the proposed study and Section 4 describes the data employed. Section 5 describes the results of the regulatory risk data cluster analysis and examines what drives governance risk perception. Section 6 discusses the use of cross- sectional at-issuance risk premium data to assess the impact of (1) macro-level and (2) indus- try-level regulatory risks on finance cost at investor entry, and further applies the model thus developed to study the impact of macro-level regulatory risks on infrastructure finance in Hungary. Section 7 utilizes time-series data on the life of infrastructure bonds in the secondary market to (1) identify what risks are significant in "real" market conditions and to (2) assess how political and regulatory risks contribute to the likelihood of positive or negative changes in a bond's credit rating history. Section 8 concludes the paper. Detailed descriptions of the applied models and estimation techniques used here are presented in five Methodological Appendixes at the end of the paper. 2 2. Previous Research The literature devoted to credit risk measurement can be divided into three broad categories. The first looks exclusively at firm-intrinsic determinants of credit risk and premium, and includes works by Hoboken (1991), Altman and others (1997), and others. As described ear- lier, this approach is based on analysis of bond issuer business fundamentals, including finan- cial statements, profitability, liquidity, and capital structure. The second category looks at cap- ital markets and firm determinants, particularly the level and volatility of company share prices, and is represented by Beaver (1968), Merton (1973), and Fama and French (1993). The third category, and the one within which this paper primarily belongs, is dedicat- ed to the macroeconomic and institutional determinants of credit risk and investment flows. Theoretical discussion of the economic impact of political institutions has expanded dra- matically since North and Thomas (1973) first outlined a "transaction cost view of econom- ic history." Acceptance of the crucial role that sociopolitical structures play in reducing the cost of contracts and in bargaining, monitoring, and enforcement has achieved the status of conventional wisdom, not only among economic historians such as North (1990), but also among economic development theorists such as Borner, Brunetti, and Weder (1995), Olson (1996), and the World Bank (1997). Political forces have the potential to intervene and halt the development of a market system or to redistribute the wealth thus created. As noted by WAeingast (1993), not only do political forces hold the potential to destroy a fragile, nascent economic system; the fear that that power will be exercised also deters the economic activi- ty necessary for economic growth. The absence of secure protection of the reward for effort deters investment and hence economic development (North (1990), Olson (1996), Williamson (1991)). The fundamental policy question is therefore whatform of political and regulatory systems are required to ensure that a viable, stable private market economy is pro- duced by these systems? The answer offered by the school of new institutional economics envis- ages the creation of political institutions that enforce governments to credibly commit to non- interference with private property. Such institutions are instrumental to the successful pro- motion of economic growth and the facilitation and participation of private investment (Levyand Spiller (1994), North (1993), Wtilliamson (1991)). In addition to this conceptual research, work has also been done to understand on an empirical level the relationship between governance and regulation and private investment in infrastructure. A number of research works appeared in the 1980s, when the finance com- munity in the United States became interested in the impact of Public Utility Commission regulations on the risks associated with investments in U.S. utilities. Analysis of utility bond ratings and rankings assigned to the U.S. state utility commissions by U.S. investment advi- sory services identified several factors that were suggested to be instrumental in explaining the greater risk associated with investment in utilities firms (Chandrasekaran and Dukes 3 Governance Impact on Private Investment (1981), Davidson and Chandy (1983), and Navarro (1983)). These factors included the util- ity firm's beta, regulatory lag, interest rates, the rate of return on equity, the election or appoint- ment of a utility commissioner, the existence of an option to pass on the cost of fuiel adjust- ment to the customer, and an allowance in the rate base for construction work in progress. Valuable evidence of the investment incentives that state regulation can provide is also available in the literature of public finance, in the form of analysis of the effect of state fiscal institutions such as balanced budget rules and restrictions on state debt issuance on the yields of state general obligation bonds. Poterba and Rueben (1997) find that states with tighter anti- deficit rules and more restrictive provisions on state debt issuance pay lower interest rates on their bonds. Other authors, including Eichengreen (1992), Goldstein and W'oglom (1992), Bayoumi and others (1995), and Lowry and Alt (1997) also find a correlation between fiscal regulations in the U.S. states and the interest rate on the bonds issued by these states. This evidence suggests that bond market participants in industrialized countries consider fiscal reg- ulations when assessing the risk associated with state bond issues, which if true might havc important implications for emerging market economies. For example, in Ukraine, strict municipal expenditure rules encourage timely payments to their local utilities by those bud- getary customers that are subsidized from the municipal budget; this helps maintain a reli- able cash flow and assures timely payments to the potential investor. The existing literature of the developing countries tends to limit explanation of private capital flow to one or two sets of independent variables, such as macroeconomic factors (Hajivassiliou, 1989), exchange and interest rate risk (Masuoka, 1990), and macroeconom- ic and project-specific variables (Dailami and Leipziger, 1997). An important exception is the dissertation work by Henisz (1996), who developed a measure of political constraints that pro- tect investors from governmental discretion. Henisz forged an explicit link between this mea- sure and variations in cross-national growth rates and the volume of infrastructure investment. There have been a number of systematic studies of the cost of capital raised through bond issues in developing countries. A detailed analysis of credit risk premiums was done by Edwards (1986), who empirically studied the pricing of bonds and loans in international mar- kets. The focus of this work, however, is limited to the risk premiums of sovereign bonds and loans to developing countries, and to macroeconomic independent variables such as the ratios of debt to GNP, reserves to GNP, gross investment to GNP, and debt service to exports, as well as to growth indicators, maturity, and principal of the bond. W\hile Edwards' paper identifies the significant effect of debt ratios on risk premium, it takes no account of governance risk indicators. Cantor and Packer (1996) analyzed the determinants of spreads on sovereign bonds for 49 countries in 1995, relating spreads to macroeconomic factors, default history, and to Standard and Poor's and Moody's country credit ratings. The limitations of this study, as iden- tified in Eichengreen and Mody (1998), include the fact that it analyzes only sovereign bonds. Cline (1995) limits his study to cases of highly indebted countries and considers only four macroeconomic determinants of interest rate spreads. A 1997 follow-up study by Cline and Barnes uses more recent data and a longer list of explanatory variables, but otherwise is subject to the same limitations. Eichengreen and Portes (1989) analyze a larger sample of 4 Previous Research bonds issued in the 1920s; however, the debt pricing patterns could have changed over time. A similar concern attaches to Edwards' 1986 study of bond spreads on 167 issues for 13 coun- tries in 1976-80. Finally, Eichengreen and Mody (1998) studied the spreads on 863 bonds issued in developing countries from 1991 through 1995, relating those spreads to bond characteris- tics, country macroeconomic factors, and global economic conditions. As in other studies, the potential significance of political and regulatory risks was not considered. The study was addi- tionally limited to Latin America and East Asia, and excluded examination of Eastern European bond issues. The objective of this paper is to fill in these gaps and to help achieve an understand- ing of what governance features are instrumental for stable private investment and partici- pation in infrastructure. To evaluate the influence of the described risk factors on the cost of debt capital in infrastructure projects, the determinants of bond spreads, defined as a risk premium over a U.S Treasury benchmark rate, are analyzed. The research considers the entire universe of bond issues, covering more than 40 developing and industrialized countries, and the entire universe of potentially significant factors-macroeconomic, financial, political, and regulatory-that are hypothetically related to the risk of bond default. 5 3. Analytical Framework To evaluate the influence of the previously described risk factors on the cost of debt capital in infrastructure projects, the determinants of bond spreads (defined as a risk premium over a benchmark rate set by riskless U.S. Treasury bonds) are analyzed for the universe of fixed- rate infrastructure bonds issued during the last decade. A fixed-income security investment decision is based on the relative value of the bond. One of the main factors in assessing this value is the risk of bond default. Investors are compensated for such risk by a risk premium, or spread, on bond yield.7 The spread is a dependent variable, with investors requiring a high- er spread when investing in riskier assets in order to cover the expected loss from potential default. Credit analysts and researchers employ three main analytical methodologies to analyze credit risk: (1) the firm-intrinsic approach, (2) the firm-capital market approach, and (3) the firm-macroeconomic approach (as reviewed in Altman and Saunders, 1997). A firm-intrinsic credit model is based on specific information about the issuing firm's fundamentals, including its financial statements, profitability, liquidity, and capital structure. A composite effect of these indicators is used to measure the extent of corporate default vul- nerability. Although a firm-intrinsic model enables the direct comparison of an individual com- pany with other companies that may have compromised their creditors, it does not account for capital market effects or for the global and local economic and governance conditions that also impact the probability of bond issuer default. Firm-capital market models rely almost exclusively on capital market movement infor- mation to assess the financial vulnerability of issuer firms. The key information used by these models is the level and volatility of the company's stock price, which has been proven to be a helpful indicator of a firm's financial health. The resulting measure of a firm's creditwor- thiness nonetheless remains a rather volatile indicator. In contrast to the firm-capital model, which takes security valuation notions as its explanatory variables, the third approach employs macroeconomic factors. The firm-macro- economic methodology analyzes the impact of variables such as economic growth, inflation, interest rates, and capital market activity on a firm's ability to sustain its financial obligations. Each of these approaches to credit risk estimation provides only partial understanding of the problem. In some cases they do not account for firm and capital market characteris- tics; in others, they do not factor in macroeconomic indicators. None of them takes account of the governance characteristics of the country and sector of bond issuance. This implies that these models, neither separately nor in combination, are sufficient to understand the deter- minants of credit risk in emerging markets characterized by volatile political and macroeco- nomic conditions. This research develops a simple framework to understand the role that country gov- ernance and industry regulation play in the formulation of investor risk perceptions and in 6 Analytical Framework the determination of the cost of infrastructure bond financing. It seeks to find an answer to the following questions: 1. What governance, political, and regulatory risks impact the cost of private debt capital for infrastructure projects? Which of these risks do investors perceive as diversifiable and which as irreversible? 2. What is the aggregate weight of individual and combined regulatory risks and how does their impact compare with that of other factors in governing risk premiums on private bond debt finance for infrastructure? 3. Which countries do potential investors perceive as being of the greatest risk? What are the institutional characteristics that prevent these countries from being classified as investment grade? 4. Are bond at-issuance risk premiums driven by analyst risk perceptions or by true market conditions? Is the structure of the bond spread at-issuance systemically different from that in the aftermarket? 5. How do governance factors affect bond risk pricing over the lifetime of the bond? Do changes in political and regulatory conditions matter in bond downgrades? To address these questions, this research analyzes a rich set of cross-sectional and time- series data on bond spreads and on the corresponding macroeconomic, governance, and bond financial characteristics. It uses the techniques of cluster (principal component) analysis, sim- ple ordinary least squares, and two-stage least squares, and applies the random correlation effects model and ordered and multinomial logit regressions. Dependent Variables Bonds are appropriate as the subject of analysis as they have become since the 1990s a major conduit of private capital flow to infrastructure projects in emerging markets. The basic value of a bond rests on the investor's assessment of the relative attractiveness of the expected stream of future interest receipts and the prospect of eventual recovery of the principal at maturity. Investors consider the risks underlying the bond contract in terms of the issuing company's future ability to generate sufficient cash with which to repay both interest and principal. The collective judgement of investors about the issuing company's ability to do this will influence both the price level at which the bond is publicly traded and the particular risk category accord- ed to the bond. Investors expect to earn a higher spread when investing in riskier assets to compensate for taking the extra risk. To evaluate the influence of the above factors on the cost of debt capital in emerging economies, we must analyze the determinants of bond yield spreads, defined as a risk pre- mium over a benchmark rate (that is, of a U.S. Treasury bond of the same maturity). The bond spread is measured by taking the difference between the annual yield of the considered bond and the annual yield of a U.S. Treasury bond of the same maturity on the date of bond issuance. To understand bond risk pricing in real market conditions, time-series analysis uses secondary 7 Governance Impact on Private Investment market bond spreads, as determined by the difference between the annual yield of the con- sidered bond and the annual yield of a U.S. Treasury bond of the same maturity on the date of secondary market transaction. Independent Variables Throughout this analysis, the research will employ the following exogenous risk factors: * Macroeconomic risk indicators-that is., a country's economic creditworthiness and propen- sity to default factors. * Bond issue and issuer characteristics-that is., firm-intrinsic and/or bond-intrinsic charac- teristics accounting for the firm's and project's financial sustainability and the issuer's cred- itworthiness. * Regulatory and political risks-that is., economy-wide measures of governance stability and credibility. Macroeconomic factors increasing the probability of default have traditionally been included in the analysis of investment trends and of the determinants of the cost of debt cap- ital. In analyzing a country's business environment, investors assess the likelihood of change in economic conditions, which could impact the market demand and the price of project out- put; the cost of project inputs, including labor costs; and the foreign exchange rate. Earlier studies have found that high private sector investment rates are associated with high domes- tic production and demand growth and low inflation, low fiscal deficits, and low external indebt- edness. This research accounts for the influence of the following macroeconomic factors as determinants of the cost of debt capital for infrastructure projects: (1) real GDP per capita, adjusted for inflation; (2) annual GDP growth rate; (3) the volume of foreign direct invest- ment, as a percentage of GDP; and (4) the volume of stocks traded, as a percentage of GDP. High national levels of real GDP per capita, GDP growth, and current revenues are nor- mally signs of an overall stable macroeconomic environment, and they therefore enhance the creditworthiness of bond issuers in such a country. The volume of foreign direct investment as a percentage of GDP serves as an indicator of the attractiveness of the local economy to investment from abroad, and high levels of foreign direct investment can in turn potential- ly make bond issuance more likely and the spread lower. The volume of stocks traded as a percentage of GDP and the amount of private, nonguaranteed debt signify the level of finan- cial market development in the country of bond issuance. High levels among these indica- tors signify good conditions for the issue of fixed-income securities by evidencing the exis- tence of appropriate financial infrastructure. Finally, the value of government fiscal claims on the private sector variable is an indicator of the extent of state control of the private sec- tor; the higher this figure, the greater the spread on bonds, as investors seek compensation for the risk of having to make excessive payments to the government in the form of fees, taxes, and so on. 8 Analytical Framework Bond issue and issuer characteristics such as principal amount, maturity, and duration are among the key variables in the bond coupon valuation analysis and are therefore incor- porated in this research as explanatory factors of bond spreads. Longer periods to maturity will expose potential investors to a greater period of risk, and will result in a higher evalua- tion of the risk of bond default. It should therefore be expected that higher maturity would be associated with higher spreads on bonds. The investment of large principal amounts is nor- mally affordable only to established and prominent firms, to whom the risk of default may be negligible compared to that of small firms. It should therefore be expected that high prin- cipal amounts would have a negative effect on the bond spread. Cross-default provision (that is., the provision that if a company defaults on one of its debt obligations it will automatically default on all of its obligations) makes the risk of default higher and hence has an increasing effect on the spread. The presence of collateral on the bond, a high issuer Moody rating, and a better issuer profile (debt performance history) should be expected to have a reducing effect on the spread, as these factors serve as supporting evidence of the borrower's creditworthiness. Firm-intrinsic characteristics should be expected also to play an important role in the default risk valuation of a bond. This paper seeks to incorporate the effects of these charac- teristics by including institutional and industry sector dummies characterizing the bond issuer. Securities issued by public sector entities are usually considered to be safe investments due to the government's obligation to backup debt repayments in the event of the default of the issuer company. Default risk assessment should also be expected to differ according to industry sector. The degree of state regulatory oversight imposed on the power, oil, gas, and telecommunications sectors examined in this paper differs, for example. The power sector is heavily reliant on domestic consumers and price regulations, but the export-oriented oil and gas sectors are less vulnerable to domestic regulatory shocks. Overall, it is hoped that the inclusion here of institutional and industry sector dummies will absorb the variations caused by the firm-intrinsic characteristics of the bond issuer and will therefore render unbiased the coefficients of other variables of interest. Vagueness and instability of legislative and regulatory systems can delay projects and push costs above budget (Razavi, 1996). In order to test the impact of governance on the per- ception of infrastructure investment risk, this research compiled a listing of the governance risks attributable to a country's political, regulatory, and legal frameworks. These risks include the absence of (1) sustainable economic policy, (2) the legal and regulatory frameworks that would prevent corruption and ensure a high quality of bureaucracy, (3) contract enforcement mechanisms, (4) democratic political leadership, and (5) a system of checks and balances on political discretion; and the presence of (6) external and internal war risks and (7) risks of political expropriation and terrorism. It is expected that high levels of governance risk, encompassing an absence of democracy and lack of a stable legal framework; high political risk of confiscation, expropriation, and corruption; poor contract enforceability; and poor bureaucracy and unstable sector regulations would have a strong effect on the risk premium of a bond at its issuance and throughout its secondary market life. 9 4. Data Each of the three methodological approaches of this research (cross-sectional, time-series, and downgrade history analysis) is centered on a different dependent variable. In cross-sec- tional analysis, this variable is the initial offer spread on fixed-rate infrastructure' bonds issued in U.S. dollars during 1990-99, as reported in the Euromoney Bondware database, which main- tains information on all security issues in eurobonds, foreign bonds, warrants, international equities, and other capital markets. Selected according to this principle, 645 bonds were issued in 40 industrialized, emerging, and underdeveloped countries9 by central governments (1), public corporations (48), publicly financed entities (13), public utilities (100), local gov- ernments (5), private corporations (202), privatelyfinanced entities, such as banks (177), and private utilities (98). For the time-series analysis, this research compiled bond lifetime data on the semian- nual yields of those main universe bonds that were traded in the aftermarket. The dependent variable for this approach is the annual risk premium observations on these bonds. (The annu- al observation was chosen to match annual changes in macro and regulatory risk factors.) The spreads were calculated by taking the difference between these yields and the yields on U.S. Treasury bonds of identical maturity, obtained by interpolating the historical U.S. government yield curve. This resulted in 26,905 weekly spread observations for 244 bonds (excluding trans- port bond issues) from the initial research universe of 645 infrastructure bond issues. W\eekly information on yields of corporate bonds traded in the aftermarket was collected from the Bloomberg database. The spread was calculated by taking the difference between the yield to maturity of the bond and the yield to maturity of U.S. Treasury fixed income securities of the identical time to maturity. The third methodological approach applies the events of infrastructure bond upgrades or downgrades, as assessed by independent rating agencies Standard and Poors and Moody's, over the lifetime of the bond, as registered in the Euromoney Bondware database. The dependent variable for this third approach is the triple event of the upgrade, downgrade, or no change of the bond's credit rating. Information on independent variables was collected from a variety of sources, includ- ing the Euromoney Bondware database, which contains key fixed-income security charac- teristics."0 Bond issue characteristics of an explanatory nature, as defined by this research hypothesis, were drawn from this database (see table 17). Macroeconomic indicators characterizing the state of macroeconomic stability, com- mercial development, financial market development, and the extent of foreign investment were drawn from the World Bank Development Indicators database. This database contains quantitative socioeconomic information spanning five decades for all countries worldwide. The macroeconomic explanatory variables are listed in table 18. 10 Data One of the most important tasks of this research project was to collect reliable data by which to evaluate the "riskiness" of the political and regulatory environments in countries of bond issue. It is naturally difficult to measure the perception of political risk, but there are at least 14 sources worldwide, providing a variety of indexes, that attempt to assess quality of governance."1 The 1Iorld Bank describes the governance indicators in these indexes as falling into three categories: (1) objective measures describing the prevailing political and institu- tional arrangements-for example, the frequency of elections, organizational structure, and the size and cost of government; (2) participant surveys of interested parties, reflecting the views of citizens, entrepreneurs, foreign investors, public officials, and others on the quality of gov- ernance in their own or other countries; and (3) cross-country assessments, often entailing polls of experts who are asked to rate the quality and effectiveness of government institutions. Much of this governance-related research is publicly available. Some risk-rating ser- vices-for example, Political Risk Services (ICRG), Standard and Poor's, and the Business Risk Service-rely on panels of experts who rate countries (or institutions within countries) using a defined set of criteria. The use of common criteria provides some comparability across coun- tries and over time, although the ratings still depend on the experts' interpretation of the criteria and their subjective perception of each country. Other organizations-for example, the Davos W\'orld Competitiveness Report, Gallup, and the Political and Economic Risk Consultancy-rely on participant surveys. One advantage of these is that they reflect the opin- ions of many firms or citizens closely connected with the country they are assessing. Some sur- veys, however, rely on voluntary responses and ask vague questions that invite interpretation by the respondent, limiting their value for cross-country comparison. Although the results of expert polls and participant surveys are unavoidably subjective, they often provide the best information on the less visible aspects of governance. Objective data on the prevalence of corruption are, almost by definition, difficult to obtain, leaving few alternatives to the subjective indicator. Subjective perceptions of the quality of governance are in any case often as important as simple facts. A country may have sound institutions accord- ing to objective criteria, but the confidence of its residents or investors in those institutions is also required for its governance to be assessed as good. Analysis of the available indicators is difficult, as most of them are intercorrelated and not entirely exogenous. This paper employs cluster analysis to handle the array of regulato- ry and political risk indicators that are currently being produced. For purposes of this analysis, the sources of regulatory risk indicators were selected such that they provide a measurement of the entire universe of governance risks as described in the Conceptual Framework section of this paper and, to the extent possible, that they cover all regions and years of observation presented in the selected bond universe. Following this principle, regulatory and political risk indicators were collected for the bond universe that included measurements of (1) sustainable economic policy, (2) the existence of legal and reg- ulatory frameworks capable of preventing corruption and ensuring good bureaucracy quality, (3) contract enforceability, (4) democratic political leadership, (5) the presence of external and internal war risks, and (6) the risks of political expropriation and terrorism. These 11 Governance Impact on Private Investment data were collected from the Business Environment Risk Intelligence (BERI) Business Risk Service database, the Political Risk Service database International Country Risk Guide (ICRG), and the Transparency International Corruption Perceptions Index. The indicator for pres- ence of a system of checks and balances on political discretion was obtained from the dis- sertation work of W Henisz. Henisz developed the Political Constraints Index, which mea- sures constraints on political opportunism and which is based solely on the number of dejure veto points in a given polity, maintaining the strong and unrealistic assumption of uniform- ly distributed preferences. This variable also accounts for the extent of the development and independence of a country's judiciary, and operates on the assumption, widely accepted in the new institutional economics literature, that the benefits of constraints on executive dis- cretion on average outweigh the costs of lost flexibility through their ability to provide insur- ance against government regulatory and policy takings. Table 1 on the following page summarizes all variables and data sources. 12 Data Table 1 Research Data and Sources Variable Level/type of obseruation Source Dependent 1990-99 Spread (difference of the yield on this bond Fixed-rate US$ bonds issued Euromoney Bondware and the yield on a U.S. Treasury bond of for the power, oil, gas, telecom, comparable maturity) and transport sectors Aftermarket spread Historical spreads for above Bloomberg Up (down)grade history Binary and amount of change Euromoney Bondware Independent Country economic indicators Stocks traded, total value (% of GDP) Country World Bank Development Private nonguaranteed debt (% of ext. debt) Indicators Database ("SIMA") Gross foreign direct investment (% of GDP) Quantitative socioeconomic Current revenue, excluding grants (% of GDP) indicators GDP per capita, PPP (current international $) Inflation, consumer prices (annual %) Ratio of official to parallel exchange rate GDP growth (annual %) Private capital flows, total (% of GDP) Claims on private sector (% of GDP) Project characteristics Collaterized (yes/no) Bond Euromoney Bondware Callable (yes/no) database Principal of the bond Quantitative and descriptive Country of issuer bond issue characteristics Issuer of public sector dummy Issuer sector dummy (power, oil/gas, telecom, transport) Maturity Regulatory envzronment 1. Economic stability; political leadership; Country International Country Risk external and internal conflict risk; corruption; Indexes based on expert Guide military and religion in politics; law and order opinion poll, 1984-98 tradition; racial and nationality tensions; political terrorism; civil war; political party development; quality of the bureaucracy 2. Bureaucratic delays; contract enforceability; Indexes based on expert Business Environment Risk nationalization risk opinion poll, 1990-95 Intelligence country index database 3. Corruption Perceptions Index Expert survey, 1980-98 Transparency International 4. Political Constraints Index Analysis based on dejure veto points in a country's polity, 1990-99 5. Extent of government control; privatization Industry W. Henisz progress; structure of the power sector Indexes based on expert World Energy Council survey, 1999 6. Pricing discretion of electricity regulator Survey of energy sector Energy Sector Management restructuring in developing Assistance Program, World countries, 1999 Bank International survey of Cambridge University Press electricity regulation, 1996 13 5. What Drives Governance Risk Perception: Cluster Analysis of Political and Regulatory Risk Data Cluster Segregation of the Governance Risk Indicators The first objective of this research was the identification of those elements of government pol- itics and regulations that dictate investor perception of political risk. Identification of the prin- cipal components of the set of 19 regulatory risk indicators compiled from the three above sources provided a convenient way to address this question. These components may be defined as the linear combinations of the original variables that are not correlated and that partition the variation in those variables in such a way that the first component carries the largest portion of the variance, and that the next component is not correlated with the first and accounts for the next-largest share of variance. The BERI, ICRG, and Transparency International data are collected fronm expert poll assessments of essentially the same aspects of governance. It should therefore be expected that some indicators from these databases would be highly correlated. Principal component analysis enables an exploration of the rela- tions present in this set of data, combating possible multicollinearity problems and reducing the dimension of the data set. This form of analysis is considered highly efficient in variables clustering and variables reduction, as it clearly reveals that the variables that have similar prin- cipal component profiles essentially measure the same. Correlation Analysis Correlation analysis of the data set of 19 regulatory risk variables provided an initial insight into the internal relationships of the regulatory and political risk factors (see table 25). Specifically, correlation analysis identified four clusters of strong positive relationships in the set of regulatory explanatory variables. The first cluster consists of positively correlated indicators from the ICRG database mea- suring the extent of (1) organized religion in politics, (2) external conflict risk, and (3) racial and nationality tensions. This relationship suggests that in a country where organized reli- gion dominates politics, external conflict risk and racial and nationality tensions are also like- ly to occur. The mean correlation coefficient among variables in this cluster is 0.46, with the highest coefficient 0.62 (occurring between racial and nationality tensions and organized reli- gion in politics) and the lowest, 0.36 (between organized religion in politics and external con- flict risk). 14 What Drives Governance Risk Perception The second cluster of highly related regulatory risk variables comprises indicators measuring the extent of (1) political party development, (2) corruption in government, (3) quality of the bureaucracy, (4) communication and transport infrastructure quality (variables from the ICRG database), (5) bureaucratic delays, (6) contract enforceability (variables from the BERI database), and (7) the. Corruption Perceptions Index (Transparency International) of a given country. The mean correlation coefficient among variables in this cluster is 0.76, with the highest coefficient 0.95 (between bureaucratic delays and contract enforceability) and the lowest, 0.69 (between political party development and the Corruption Perceptions Index risks). The third group of variables, which stands out for the vividness of intercorrelation (mean coefficient 0.75), consists of the indicators reflecting (1) the presence of the military in pol- itics, (2) the tradition of law and order, (3) political terrorism, (4) the risk of civil war, (5) the political risk rating (variables from the ICRG database), and (6) the nationalization risk (variable from the BERI database). The highest positive relationship in this cluster is between law and order tradition and the political risk rating (0.87), with the lowest correlation occur- ring between the risks of political terrorism and of nationalization (0.42). The fourth correlation cluster consists of the variables measuring (1) the effectiveness of a country's political leadership and (2) its economic policies, measured through the eco- nomic expectations versus realities and economic failures risk indicators from the ICRG data- base. The average correlation coefficient among the variables in this cluster is 0.76. The close relationship between bureaucracy and corruption indicators should come as no surprise. Perhaps more thought-provoking is the link between corruption and poor bureaucracy and the broader spectrum of political risks that includes nationalization, poor contract enforceability, and the dangers of military coups, civil war, and political terrorism. This may be explained by the fact that corruption prevents democratic development (as mea- sured by the political party development indicator) and thereby results in poor infrastructure. One of the strongest correlations across the four clusters is in fact the one between law and order tradition-an indicator that measures the historical existence of legal institutions-and democratic sustainability and the absence of corruption. Principal Component Analysis Principal component analysis of the regulatory risk data reinforced the findings of the ini- tial correlation analysis. Principal Components 1 and 2 are linear combinations of regulato- ry risk factors, weighted as indicated (table 2). These components account for 63 and 11 per- cent of spread variation, respectively. The principal components were graphically represented using the binary plot technique (see Methodology Appendix 1). When the underlying variables are highly interrelated (as is the case with the regulatory variables, which are naturally related to each other as they mea- sure the same effect or slightly different aspects of the same effect), it often suffices to extract just the first few components to accurately describe the relationships among the underlying variables. Since the first components account for the largest portion of the variation in the 15 Governance Impact on Private Investment Table 2 Composition of the Two Principal Components of the Regulatory Spread Data Regulatory risk factor Component 1 weightings Component 2 weightings Economic expectations vs. realities 0.171 0.505 Economic planning failures 0.158 0.532 Politics 0.157 0.507 External conflict risk 0.103 0.048 Corruption in government 0.263 -0.157 Military in politics 0.241 -0.074 Organized religion in politics 0.187 -0.146 Law and order tradition 0.267 -0.067 Racial and nationality tensions 0.124 0.111 Political terrorism 0.212 0.087 Civil war risk 0.214 -0.034 Political party development 0.246 -0.157 Quality of the bureaucracy 0.261 -0.177 Political risk rating 0.294 0.075 Bureaucratic delays 0.283 -0.090 Contract enforceability 0.283 -0.081 Nationalization risk 0.249 0.049 Communication and transport infrastructure quality 0.265 -0.108 Corruption Perceptions Index (Transparency Int'l) 0.250 -0.191 Attributed Variance 63% 11% Source: Author's calculations. data, there is little left for the last components, which can therefore be dropped from the analy- sis without any loss of information. The accuracy of this representation is measured by the percentage of variation in the original data exhibited by the first principal components (63 and 11 percent in this case, as displayed in table 3). W\then the first two components are suf- ficient for adequate representation of the data, the principal coordinates of both observations and variables can be naturally used to construct a two-dimensional scatter plot of interrela- tionships between two groups of independent variables (in this instance, countries and their levels of regulatory risk) according to their values at each observation point (in this case, bond). This is known as a binary plot. The binary plot employed here uses only the first two components extracted from the data matrix of countries and regulatory risk indicators. Each country is thus described by two coordinates (xc,y,), obtained by plugging the values of the regulatory variables for the coun- try into the equations for the first and second principal components. Each regulatory risk indi- cator additionally receives two coordinates (x,,y,) given by its coefficients on these two prin- cipal component linear combinations. The binary plot technique essentially means plotting variables (various regulatory risk indicators) according to their coordinates (xr,y,) and obser- vations (countries of bond issue; xp,yc) on the same scatter plot. The intermediate step of binary plot construction is depicted in figure 1. Specifically, regulatory risks are positioned on the bipolar graph in accordance with their two principal component coordinates (x,yr). 16 What Drives Governance Risk Perception Figure 1 Position of Regulatory Risks in Respect to Two Principal Components PC 2 0.600 Econ. plan. failures 0.500 o t leadership Econorric expectations 0.400 0.300 0.200 IRacial and nationality tensions 0.100 , Political Nationaliation Political risk / _ - irnal conflict terro_. - -rating < _ = = = ~~~~~~~~~~~~~~~~~~~~~~~Military in 0.000Laan 0r.25 order 0.300 0.3i 0 -0.100 Infracrabc ~~~~~~~~~~ @~- Corruption in eneforceability -0.200 religion in . < ~Goverrunent -0.200 Politics Quality of the Corruption bureaucracy perceptions -0.300 index PC 1 Note: X-axis - regulatory risks principal component I values, Y-axis - regulatory risks principal component 2 values. Source: Author's calculations. The advantage of the binary plot technique is that it allows the researcher to examine the clusters of variables and observations within the underlying multidimensional space. The cosines between the vectors-variables reflect the underlying correlations; the projections of the observations onto these vectors are proportional to their standardized values. The four clusters of regulatory risk indicators visible in figure 1 signify that regulatory risks within the clusters possess similar (close) coefficients on two principal components and are closely related. As can be seen from the graph, three out of four clusters of independent regulatory variables have close coefficients. For the binary plot on figure 2, the regulatory risk indicator variables were centered and standardized for each country (that is, the mean was removed from each value and the result divided by the standard deviation) prior to extracting the principal components, there- by resulting in all vectors being of the same unit length. Note that if two variables were close- ly correlated, their corresponding vectors on the binary plot would be almost collinear. Similarly, if two countries shared similar values for all regulatory variables, they would also be situated closely on the binary plot. 17 Governance Impact on Private Investment Figure 2 Bond Country of Issuance Position on Regulatory Risk Indicator Vectors Economic planning failures Pg litiall leadership " * tba XY tensions~~~Ecoomi texpectation r lux Umys * chl a qat * ury nationality Political politic U tsa tcnsions terror risk mex ratinn Source:~~~~~~~~~~~~~~~ Author' Natlonalizatio A visual inspection of figure reveals that regulatory risk indicators form fourisk identified tru Law and s phi svk Civil War ~~~~~~~~~~~~~~~~~~order ie col Mind t brat Pe tract U ven Busreacr thCU pilr r hke a elP Orgvaized Tevw ra quality religion in a Conruption in t grc politics t psitv government Comsption Quality of perceptions bureau,cmay index Note: X-axis - regulatory risks principal component 1 values, V-axis - regulatory risks principal component 2 values. Source: Author's calculations. A visual inspection of figure 2 reveals that regulatory risk indicators form four distinct groups, three of which are quite closely related. These groups essentially replicate the groups identified through earlier correlation and principal component analysis. On this biplot, the intersection of the rays indicates the mean values for all variables, with each of the rays on the plot representing the positive half-axes of the underlying variable. The viewer can approx- imately reconstruct the regulatory riskiness of any one country, as measured by any of the indi- vidual risk indicators, by making a perpendicular projection from that country onto the line that contains the corresponding ray. (As depicted, the rays indicate the positive semi-axes of the variables. The negative extensions of these rays, although not drawn on the graph, may be easily made to accommodate the projection.) For example, by making projections to the ray of the appropriate governance risk, in may be established that Korea has high economic planning and political leadership ratings, but a relatively poor rating for external conflict risk. The United States, Canada, and Norway have excellent (high) ratings for immunity against virtually all political and regulatory risks, as their projections fall farthest away from the ori- gin in the positive direction. 18 What Drives Governance Risk Perception Hierarchical Cluster Analysis of Regulatory Risk Data Principal component analysis of the regulatory risk data and earlier correlation calculations may be further supplemented by hierarchical cluster analysis of the regulatory risk indicators data. The idea behind cluster analysis is to combine universe observations into groups (clus- ters) based on some criteria of similarity. The techniques of cluster analysis differ according to the way that these similarities are computed and accounted for when building the clusters. WAith hierarchical methods, the partition of observations into clusters is achieved sequentially. Each observation is initially considered a separate cluster, with the closest clusters then merged into bigger groups. This process can be graphically represented as a dendrogram (see Annex G). The distance between two clusters is defined by Wzard's formula, which measures the increase in intra-class distances after the two clusters are merged. Minimization of this dis- tance maximizes the reduction of inter-cluster variation, which is a goal of cluster analysis. Hierarchical cluster analysis yielded the following groupings of regulatory risk indi- cators (table 3). These groupings are fully consistent with the results of the correlation, prin- cipal component, and binary plot analyses: Challenges Posed by Subjective Governance Indicators A common concern about governance indicators, both political and regulatory, is that they can be inconsistent, unreliable, or affected by bias on the part of the observer. A country rated high by one agency or panel of experts may be rated low by another, even though both claim Table 3 Clusters of Governance Risk Factors Fundamentalism Economicpolicy Corruption and Political disorder risks sustainability risks autocracy risks risks Organized religion Political leadership Political party Law and order in politics development tradition Economic planning External conflict risk failures Corruption in Civil war risks government Racial and nationality Economic expectations Military in politics tensions vs. realities Quality of the bureaucracy Political terrorism Infrastructure quality Political risk rating Corruption Nationalization risk Perceptions Index Bureaucratic delays Contract enforceability Source: Author's calculations. 19 Governance Impact on Private Investment to measure the same attribute. Recent research by Kaufmann, Kraay, and Zoido-Lobat6n (1999) proposed that this problem might be resolved by aggregating indicators from sever- al sources into an average or composite index-a poll of polls. This is the approach that Transparency International uses for its Perceptions of Corruption Index. The researchers found that individual governance indicators may be imprecise, but that an aggregation of indi- cators from many different sources can improve the precision of cross-country comparisons of governance. Aggregate indicators are not so precise that small differences between coun- try scores can be given much significance, but they do enable clear identification of the best and worst performers with a reasonable degree of statistical confidence. As a general rule, the more sources used, the greater the precision of the assessments made. Most of the analysis made in this paper is based on aggregate estimates of regulatory risk factors. Specifically, two principal components and the averages across identified clusters were used to explain spread variation. The influence of individual risk ratings was also test- ed, yielding valuable information on the impact of each on the spread. These tests show that the aggregate indexes across these four groups have greater explanatory power than most of the individual indexes. Correlation of the Spread and Regulatory Cluster Averages Further correlation analysis of the relationship between the spread and the regulatory risk data showed that the regulatory risk characteristics within four identified regulatory risk clus- ters affect the spread differently. By taking the means of all indicators within a given risk clus- ter and correlating them with the log of spread, it was determined that the mean of indica- tors within the corruption and autocracy risk cluster (comprising political party development, corruption in government, the quality of the bureaucracy, communication infrastructure qual- ity and measurements, bureaucratic delays, and contract enforceability) account for most of the bond spread variation (correlation coefficient -0.67, with significance at 99 percent level) .12 The mean of indicators within the political disorder risks cluster (civil war risks, the military in politics, law and order tradition, political terrorism, and nationalization risk) is also highly correlated with the spread, with a coefficient of -0.504, significant at the 99 per- cent level. The regulatory risk variables from the two other risk clusters were found to have a weaker relationship or no relationship with the spread: the average of regulatory risk indi- cators from the fundamentalism risk cluster has no significant relationship with the spread, and the average of the sound economic policy sustainability risk indicator has a weak nega- tive relationship within the spread, having a correlation coefficient of -0.182, significant at 99 percent level. Interestingly, the average of BERI characteristics (bureaucratic delays, con- tract enforceability, and nationalization risk) appears to have the highest significant coeffi- cient for correlation with the log of spread, of -0.715 at 99 percent level. This may be explained by the weighted presence of representative characteristics from the corruption and autocracy risks cluster and the political disorder risks cluster in this grouping of regulatory risk indicators (table 4). This analysis shows that there are governance-related risks that investors can avoid and risks that they cannot. Specifically, investors can diversify away some political and economic 20 What Drives Govemance Risk Perception Table 4 Correlation Coefficients of Main Regulatory Risk Indicator Clusters and the Log of Spread Funda- mentalism Corruptionl Economic Polit*al Log risk autocracy policy disorder BERI Political spread cluster risk cluster risk cluster risk cluster average constraints Log spread 1.000 Fundamentalism risks average 0.012 1.000 Corruption and autocracy average -0.670*** 0.061 1.000 Economic policy sustainability average -0.182*** -0.075** 0.228*** 1.000 Political disorder risks average -0.504*** 0.300*** 0.735*** 0.306*** 1.000 BERI average -0.715*** 0.099*** 0.942*** 0.255*** 0.690*** 1.000 Political Constraints Index -0.371*** -0.007 0.547*** 0.062 0.406*** 0.479*** 1.000 Notes: *** - denotes significance at 1 percent level; ** - denotes significance at 5 percent level. Source: Author's calculations. leadership incompetence and some erratic economic policies. However, they cannot diver- sify away the effects of systemic incentives to steal (corruption) or to break commitments (poor contract enforceability), which together reduce expected values and make necessary an off- setting spread. These preliminary findings are examined in greater detail in the later stages of this study through regression analysis of bond credit structure. Country Clustering by the Level of Political Risk Another notable result of the principal component analysis is the clear segregation of coun- tries into three regulatory/political risk groups. Under binary plotting, if two variables are closely correlated their corresponding vectors on the biplot will be almost collinear. Likewise, if two countries have similar values for all regulatory variables, they will be situated closely on the biplot. Segregation of countries into clusters is done by measuring their distance from the average level of risk on all regulatory risk indicators; here, the intersection point of the regulatory risk vector rays. According to the theoretical foundation of the biplot, countries that are situated far on the positive side of the Yand X axes possess excellent governance char- acteristics and may be considered virtually free of political and regulatory problems. These countries may be termed stable. Countries that are situated around the beginning of the axis possess average levels in their regulatory risk indicators, and may be termed 'weak." Finally, countries that lie far on the negative side of the Y and X axes are relevant outliers with high levels of political and regulatory risk, and are termed risky. 21 Governance Impact on Private Investment Table 5 Country Classification by Regulatory Riskiness Stable Weak Risky Austria Argentina Brazil Canada Chile Colombia Denmark Costa Rica Dominican Republic France Czech Republic Ecuador Germany Greece India Ireland Hong Kong Israel Japan Korea, Republic of Philippines Luxembourg Malaysia Russia Netherlands Mexico Trinidad and Tobago New Zealand Poland Uruguay Norway Qatar Venezuela Sweden Slovak Republic United Kingdom Spain United States Thailand Source: Author's classification. The visual evidence provided by the biplot is supported by hierarchical cluster analy- sis of the country regulatory risk data. Cluster analysis yielded the following groupings of coun- tries, classified according to their regulatory riskiness (table 5). Figure 3 Regulatory Riskiness of Bond Issuer Countries Econo-ic planning filoure P Polidcal lcadership ',P Economic expectations + MtyS fS/lttx + chl] + qat ury ~~~~~~~~~~~~~~ailand + eor + tha Polipical teror f + ris ein A tto afi Corld Law and order A phi A col A -d + Risky cateao A bra + esp crganiee 2 vacaequality + grcc oiia 0ot oeon corptobraccc Percepin Index A rus Note: Symbols: cm countries from Stable category ; + - Weak category; A - Risky category. X-axis - principal component 1 values, Y-axis - principal component 2 values Source: Author's calculations. 22 What Drives Governance Risk Perception The following binary plot chart illustrates this finding using three different symbols to denote the group distribution of country observation points (figure 3). This segregation of the countries of the research universe according to risk perception is supported by measurement of the average spread on infrastructure bonds. The total spread average across all years of the countries identified here as stable is 136.57; for the weak coun- tries it is 231.72; and for the risky countries it is 305.34. Further, the standard deviation by year from the average spread for the entire universe of countries is in most cases higher than that for each of the three country groups (table 6).13 These results suggest that there is a clear separation of countries into three regulatory and political risk clusters. The least developed countries and countries struggling in transi- tion generally have poor governance indicators and have to pay high premiums to encour- age investment. Governance issues are similarly a major challenge for the emerging economies, which must act to defuse a wide range of political and regulatory risks. The developed coun- tries alone are in the clear. The reasons for and the nature of the three regulatory quality clus- ters-and their policy implications-are analyzed in greater detail later in this paper. Table 6 Spread Behavior in Three Regulatory Risk Country Groups Evaluation of regulatory environment Spread (based on biplot analysis) Year data Stable Weak Risky Total 1991 Average 55.43 282.50 465.75 216.62 Standard deviation 22.44 53.03 133.76 203.78 1992 Average 51.67 259.67 421.33 234.20 Standard deviation 17.23 96.93 138.54 189.24 1993 Average 63.30 248.91 289.25 170.90 Standard deviation 35.01 141.20 100.85 139.42 1994 Average 60.83 310.71 335.38 189.96 Standard deviation 52.09 139.85 89.85 160.42 1995 Average 121.00 197.33 335.83 174.88 Standard deviation 165.59 175.86 65.99 172.54 1996 Average 90.77 200.55 244.22 172.99 Standard deviation 169.09 149.56 107.76 159.40 1997 Average 141.41 182.29 259.97 184.09 Standard deviation 172.85 114.62 155.16 153.36 1998 Average 227.27 280.53 300.44 249.24 Standard deviation 223.29 168.77 186.85 205.88 Total universe average 136.57 231.72 305.34 199.05 Number of countries 212 159 83 459 Standard deviation 177.45 150.78 139.76 174.16 Source: Author's calculations. 23 6. Impact of Governance Risks on Investment Cost at Entry: Analysis of Cross-Sectional At-Issuance Spread Data Impact of Country (Macro)-Level Regulatory Risks Data Layout and Basic Model This paper uses an augmented multifactor logistic risk spread model to analyze the deter- minants of the yield spread on bonds. Similar models have been used by Edwards (1986), Angbazo and others (1996), and Eichengreen and Mody (1997) to study different aspects of the formation processes of bond and loan spreads. For the purposes of this study, yield spreads on bonds are determined as follows: Model 1 log (S,,)=F(X,1)+esi The dependent variable is defined as the logarithm of the spread Si, over a benchmark (U.S. Treasury) of bonds issued to finance infrastructure projects in country i in year t. X is the vector of independent variables, which will be used to measure how the credit risk premium is related to a set of explanatory variables. Log of spread is a traditional measure of the risk premium that bond issuers have to pay investors to undertake the risk of a principal and inter- est loss. (Detailed specifications of the model may be found in Methodology Appendix 2.) Model 1 accounts for the effects of the three types of risk variables-macroeconomic, political-regulatory, and firm or project-intrinsic-that according to research hypotheses affect the spread. Based on the analyses outlined in Section 3, the variables in table 7 were selected to test the research hypotheses through model A OLS regression fitting. The dummy identifying the sector-public or private-of the issuer was constructed as follows: (1) public issuer type to encompass public utility, public corporate, central govern- ment, and public finance (total 162 bonds); (2) private issuer type to encompass private util- ity, private corporate, private finance, and supranational (total 482 bonds). Three dummies 24 Impact of Governance Risks on Investment Cost at Entry were additionally constructed to identify the sector of bond issuer: power, telecom, and oil and gas. More on Choice of Independent Variables To prevent the problem of multicollinearity, variables that were insignificant in their rela- tionship with the spread or highly correlated among each other (or both) were manually dis- carded following the principle of R square maximization and significance level threshold. Among the macroeconomic characteristics, two variables were discarded as endogenous and highly correlated with GDP per capita: these were claims on the private sector (correlation coefficient 0.66) and current revenue (0.40). (Table 19 indicates correlations among the macro variables.) The indicator measuring the level of private nonguaranteed debt was also discarded as being insignificant in explaining the spread. Among the bond issue characteristics, two variables (cross default and collateral pro- visions) were discarded from the regression analysis as being not significantly related to the dependent variable. The Moody's issuer rating was additionally excluded for the reason that it is an expert index-that is, a variable constructed based on evaluation of fundamental coun- try and bond issue characteristics, both of which are already included in this regression. It is important to note that the regression of the log of spread on three variables-GDP per capi- ta, the Moody's issuer rating, and the regulatory risk rating (average BERI)-resulted in sig- nificant coefficients on all of these variables, however. This observation suggests that the rat- ing agency (in this case, Moody's) does not take into account the entire regulatory risk asso- ciated with the country and sector of bond issuance. Of the regulatory variables, the following were used in the OLS regression tests: (1) the two principal components identified in section 5. 1; (2) the averages for indicators within four governance risk clusters (that is, the corruption and autocracy risk cluster, the political dis- order risk cluster, the fundamentalism risk cluster, and the economic policy sustainability risk Table 7 Variables Selection for Step 2 OLS Regression Analysis of the Spread Macroeconomic Political/regulatory Financial/bond Real GDP per capita Average values for 4 governance Principal amount of the bond risk clusters (Fundamentalism, Gross foreign direct Economic Policy, Corruption and Public sector type of bond investment as percentage Autocracy, and Political Disorder) issuer (dummy) of GDP Volume of stocks traded BERI Average (bureaucratic Industry sector of bond as percentage of GDP delays, contract enforceability, issuer (dummy) nationalization risk GDP growth Two principal components Political constraints index Source: Author's selection. 25 Governance Impact on Private Investment clusters); (3) the average of three BERI indexes; and (4) the Political Constraints Index mea- suring the number of veto points in a given polity. Table 8 summarizes the results of the regression on macro-level regulatory risks. (Results for industry-level regulatory risks specific to the power sector are provided in a separate chap- ter.) Table 8 presents the results of the ordinary least squares regression of the infrastructure bond spreads and the financial, macroeconomic, political, and regulatory risk characteristics. Each of the equations includes independent variables measuring (1) real GDP per capita, adjusted for inflation; (2) the country's volume of stocks traded, as a percentage of GDP; (3) gross foreign direct investment as a percentage of GDP; (4) GDP growth rate; (5) the prin- cipal amount of the bond; (6) a dummy indicating if the issuer is a public or private sector institution; (7) dummies indicating if the bond was issued for the power, oil and gas, telecom- munications, or transport sector; and (8) a variable reflecting the extent of governance risk in the country, measured by the average corruption and autocracy cluster regulatory risk rat- ing, the average of BERI ratings, the average of the political disorder risk rating, the Political Constraints Index, and, in one instance, by two principal components of regulatory risk data. Results for the Entire Sample Universe The first three columns of table 8 present experiments to explain the spread structure through different combinations of governance risk indicators. The equation in column 1 shows the effect of using these control variables to explain the relative spreads for the sample uni- verse of infrastructure bonds. The first model is tested with uses the two principal components of regulatory risk data and the Political Constraints Index to account for the impact of reg- ulatory risks on bond spreads. The results indicate with a likelihood of not less than 99 per- cent that spreads would rise by 22.2, 21.2 and 20.5 basis points were bonds to be issued in the power, oil and gas, and telecommunications sectors, respectively. These dummy variables essen- tially absorb the effect of firm-intrinsic risks on the risk premium. There is a 99 percent prob- ability that the spreads would be lower by 14.2 basis points if the principal of the bond were one year longer. Unit increases in the volume of stocks traded annually, as a percentage of GDP, and in real GDP per capita decrease the bond spread by 21.4 and 26.3 basis points respec- tively, with a confidence level of 1 percent. Higher levels of FDI increase the risk perception of the bond by 8.9 points with 90 percent probability. This contradicts the intuitive assump- tion that this variable is a sign of a country's stability. Public sector ownership of a project reduces investor perceptions of a bond's default risk, reducing the spread by 22.2 basis points with 99 percent probability. This finding conformns to the general W\all Street convention that private sector bonds cannot be priced below the sovereign ceiling. Finally, the two prin- cipal components of regulator,y risk data and higher regulatory risk ratings (indicating safer governance conditions in the countrv of bond issue) decrease the risk premium by 38 and 11.6 basis points respectively, with a confidence level of 1 percent. This set of control variables can explain about 63.4 percent of the spread variance. The equation in column 2 comprises a slightly different list of explanatory variables. Specifically, the two principal components of regulatory and political risk data are replaced 26 Table 8 Results of Log Spread Regression on Selected Independent Variables Sample Entire.w/ Entire W/ Entire wl Stabl Weak Risky Public Private Power Oil/gas Telecom Variable name 2 reg. PCs 2 reg. clusters BERI countries countries countries issuers issuers sector seictor sector Principal -.142*** -.139*** -.126*** -u118 -.198*** -.239** -.127** -.075 -.165** -.229*** -.046 (.001) (.001) (.001) (.170) (.000) (.022) (.003) (.338) (.034) (.000) (.664) Real GDP per capita -.263*** -.270*** -.217*** -.099 -.108** -.463** -.221*** -.209* -.114 -.143 -.513 (.000) (.000) (.000) (.198) (.033) (.003) (.001) (.094) (.241) (.151) (.001) GDP growth .019 -.014 -.007 .007 .035 -.012 .018 -.012 -.082 .036 .016 (.605) (.698) (.846) (.928) (.392) (.907) (.608) (.873) (.155) (.579) (.835) Stock volume -.214*** -.227*** -.164*** .262** -.413*#* -.281** -.207*** -.073 -.166** -.159** -.230** (.000) (.000) (.001) (.016) (.000) (.008) (.000) (.607) (.016) (.050) (.043) Public sector -.222*** -.221*** -.214*** -.485*** -.028 -.054 n.a. n.a. -.118* -.220*** - 356*** (.000) (.000) (.000) (.000) (.546) (.674) (.061) (.001) (.000) Power sector .212*** .200*** .198*** .616*** -.019 .345** .054 .202** n.a. n.a. n.a. (.000) (.001) (.001) (.000) (.848) (.047) (.571) (.018) Oil and gas .205*** .174** .163** ,354*** -.094 .296** .001 .279** n.a. n.a. n.a. (.000) (.002) (.003) (.001) (.351) (.047) (.994) (.002) Telecom .200*** .182*** .190*** 393*** -.019 .327** -.019 .374*** n.a. ni.a. n.a. -D (.000) (.001) (.000) (.001) (.848) (.040) (.818) (.000) Foreign direct .089* .105** .014 .304** .053 .143 -.052 .182* -.053 -.219** .284* investment (.072) (.029) (.769) (.034) (.504) (.361) (.255) (.058) (.399) (.049) (.024) BERI average n.a. n.a. -.474*** .285 -.550*** -.312** -.492** -.512*** -.513*** -.471*** -.327** (.000) (.121) (.000) (.007) (.008) (.000) (.00) (.000) (.016) Regulatory PCI -.380*** n.a. n.a. n.a. n.a. n.a. n.a. na. n.a. in.a. n.a. (.000) RegulatoryPC2 -.116** n.a. n.a. in.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. (.004) Political disorder n.a. .000 ii.a. in.a. n.a. n.a. n.a. n.a. n.a. n.a. na. (.997) Auitocracy and ni.a. -.422 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. corruption (.000) Political constraints -.049 .009 n.a. -.155 in.a. n.a. n.a. n a. n.a. n.a. n.a. index (.152) (.857) (.022) Degree of freedom 304 304 304 130 107 65 196 107 109 88 63 R square .634 .645 .661 .378 .864 .561 .787 .526 .701 .742 .693 Note: Dependent variable - log of bond spread. na. Not available. *** - Denotes significance at I percent level; ** - denotes significance at 5 percent level; * - denotes siginificance at 10 level. Figures in parentheses are two tail p-values. Soure.: Author's calculationis. Governance Impact on Private Investment by the averages of the two regulatory risk clusters that were identified as having the highest significant correlation with the spread-that is, the autocracy and corruption and the polit- ical disorder risk clusters. This change increases the explanatory power of the model, bring- ing it to 64.5 percent of the variance. An increase of one unit in the country's average autoc- racy and corruption risk indicators is predicted to decrease the spread by about 42.2 basis points. W\hen measured through regression analysis, the political disorder risks are found to have no significant relationship with the spread-this lack of significance may be explained by the high correlation between the variables within the red tape and political disorder clus- ters. Other coefficients drawn from the remaining independent variables have the same signs and relatively the same magnitude as in column 1. Column 3 presents the results generated by a model including the governance risk indi- cator as measured by the average of three BERI ratings (contract enforceability, bureaucrat- ic delays, and nationalization risk).'4 This group of risks was chosen for this and subsequent expression of the spread due to its strongest relationship with the spread, as defined by the earlier correlation analysis. Indeed, inclusion of this variable in the regression analysis improves considerably the explanatory capacity of the model, as measured by an R square of 0.661. According to the estimation results for this model, an improvement in the average BERI rating in a country results in a decrease of the spread by 47.4 basis points. Coefficients drawn from the other independent variables have the same signs and relatively the same magnitude as in columns 1 and 2. Results of Three Country Groups The equations in columns 4 through 6 test whether or not the effect of the above factors on the spread is consistent when the universe of countries is subdivided into three groups according to their level of regulatory risk, as defined earlier (see table below). Countries Classification by Level of Governance Risk Stable Weak Risky Austria Argentina Brazil Canada Chile Colombia Denmark Costa Rica Dominican Republic France Czech Republic Ecuador Germany Greece India Ireland Hong Kong Israel Japan Korea, Republic of Philippines Luxembourg Malaysia Russia Netherlands Mexico Trinidad and Tobago New Zealand Poland Uruguay Norway Qatar Venezuela Sweden Slovak Republic UK Spain USA Thailand Source: Author's calculations. 28 Impact of Governance Risks on Investment Cost at Entry Stable Countries (Good Governance) For the group of stable countries (column 4), the explanatory power of the model was sig- nificantly lower than for the entire universe of bonds, explaining only about 37.8 percent of the dependent variable variation. This finding suggests that the model does not have suffi- cient elements to perfectly replicate the conditions of bond pricing decisions in politically and regulatorily stable countries. As suggested by earlier literature, the explanatory power of the regression could be improved by adding more firm-financial, industry, and capital-mar- ket status characteristics to the right hand side of the equation. The results shown in column 4 indicate that political risk, as might be expected, has no influence on the bond risk pricing decision in regulatory risk-free countries. The same is true of the GDP growth variable, and there is only a weak positive relationship between the spread and volume of stocks traded and FDI levels in these countries. The only macroeco- nomic variable that has a significant negative relationship on the spread is real GDP per capi- ta, with a unit increase in real GDP per capita shown as likely to decrease the spread by 41.0 basis points. Richer countries have it cheaper. With the general weakness of other macro- economic and governance factors, it is the financial and industry sector characteristics of bond issues that stand out as having a substantial effect on the spreads in these countries: the spread increases by 61.6 basis points if a bond is issued in the power sector and by 35.4 and 39.3 basis points if it is issued in the oil and gas and telecommunications sectors, respectively. The high coefficient on the power sector dummy is not casual-the power sector in developed coun- tries is often subject to heavy price, quality, and environmental regulation, which can bring finan- cial losses to the company and hence increases the probability of default. In addition, in some countries of this group the deregulation process adds to the uncertainties associated with bond payments. Bond issuance by a public sector entity reduces the risk spread by 51.5 basis points. Essentially, these results certify that in countries in the stable group, infrastructure invest- ments are free of political and regulatory uncertainty and the only factors affecting the lender's decision are the financials of the project (represented by the yield to maturity) and the industry factor (captured by one of the industry sector dummies). Another important observation is that in stable countries, it is energy/utility investments that bear the highest level of risk. Countries with Weak Governance Column 5 presents OLS regression evidence on the association between the above set of vari- ables and the spreads on bonds from the second group of countries-those characterized by weak governance. Here the R square is significantly higher than that for the entire universe and for the group of stable countries (approximately 86.4 percent of the spread variance), suggesting that the research model is best suited for estimation of bond pricing decisions for regulatorily weak countries. The results of this regression are thought provoking. The sig- nificant coefficients have the expected signs and include only the political/regulatory risk rating (BERI), financial (principal amount of the loan), and macroeconomic (GDP per capi- ta and volume of stocks traded) factors. Political risk carries most of the weight (a unit increase in average BERI rating decreases the spread by 55 basis points with 99 percent prob- 29 Governance Impact on Private Investment ability), followed by the volume of stocks traded as a percentage of GDP and real GDP per capita. A 1 percent increase in stocks volume decreases the spread on bonds by 41.3 basis points with 99 percent probability. Larger principal bond issues are also viewed as less risky, with 95 percent confidence. It is also noteworthy that public sector association of the bond issuer has no significant positive effect on the bond credit risk spread, unlike in the stable country group-suggesting that a guarantee made by a weak state is not considered a protection. Finally, sector dummies measuring industry and firm-intrinsic characteristics have no explana- tory power. The latter suggests that firm-intrinsic characteristics are less relevant to fixed- income securities pricing in emerging markets. Poor governance affects even the most-effi- cient businesses. Countries with Poor Governance Results for the risky countries are summarized in column 6. The explanatory power of this regression is smaller than that for weak countries, explaining about 56.1 percent of the spread variation. Bond pricing in these countries reflects their macroeconomic (real GDP per capita, volume of stocks traded, GDP growth), bond (principal amount), sector (power sec- tor dummy), and regulatory risk (average BERI) characteristics. The coefficient on the reg- ulatory risk variable is less precise and has a smaller effect on the spread, decreasing it by 31.2 basis points with 95 percent confidence. Macroeconomic characteristics, including real GDP per capita and the volume of stocks traded, and the principal amount of the bond are all neg- atively related with the spread and are significant at the 5 percent level. Industry and firm- intrinsic risk factors are positively related to the spread with 95 percent likelihood, increas- ing it by 34.5, 29.6, and 32.7 basis points for the power, oil and gas, and telecommunications sectors, respectively. To summarize, regression analysis of the spread within the three country clusters reveals that the types of risks that concern investors differ across these clusters. The least devel- oped countries and countries struggling with transition generally have very poor governance indicators and have to offer high premiums to encourage investment. Emerging economies, which may already have in place appropriate fiscal and other macroeconomic policies and ownership structures, also face governance issues as their toughest challenge, as they seek to remove a whole range of political and regulatory risks. Developed countries alone are in the clear, with investors concerned only about real commercial project risk rather than about pol- icy risk created by adverse actions of governments. Public Sector Entity Bonds The next step of the regression analysis assessed differences in the risk pricing of private and public sector infrastructure bonds. C(olumns 7 and 8 present the results of this assessment. For public sector projects, the model yielded a high explanatory power, as measured by R square of 0.787. Here, the firm- and sector-intrinsic characteristics measured by sector dum- mies have no influence over the risk spread, a fact that can be attributed to state backing of public sector infrastructure projects. The factors of greatest significance in the risk assessment of public sector bond issues are real GDP per capita, the volume of stocks traded, the prin- 30 Impact of Governance Risks on Investment Cost at Entry cipal amount, and the regulatory risk rating. The latter has the greatest coefficient and decreases the spread with one unit of rating increase by 49.2 basis points at the 1 percent con- fidence level. Political climate and a country's macroeconomic characteristics are of the great- est importance for investors in public sector-sponsored infrastructure projects. Private Sector Entity Bonds For private sector projects, the regression estimation model yielded results of a smaller explanatory power (R square 0.526). This might indicate that the model is less suited for expla- nation of privately issued bond spreads than it is for publicly issued ones, and it is likely that a model taking greater account of firm-intrinsic and capital market performance variables would provide a better explanation of the spreads of such bonds. In contrast to the case of public sector infrastructure investments, what matters most here are the political and regu- latory risk factors and the firm- and sector-intrinsic characteristics of bond issue. A unit increase in the regulatory risk rating decreases the risk spread by 51.2 basis points at the 1 percent confidence level. Macroeconomic characteristics, except for (weakly) real GDP per capita and the level of foreign direct investment, do not play a major role in the pricing deci- sion for these bonds as companies are expected to evaluate the macroeconomic climate and be able to compensate for that climate with the economic efficiency of their projects. Regulatory risk, in contrast, appears to be a crucial factor, and one that can impede the abil- ity of private sector infrastructure firms to adjust to macroeconomic conditions. Firm- and industry-intrinsic characteristics also play an important role in determining the risk premi- um structure of private sector infrastructure bonds. Sector Differences Columns 9 through 11 present the results of an estimating spread model for the four infra- structure sectors represented in the bond research universe. For column 9, the model is esti- mated on a subset of power sector bonds to identify the determinants of the lender's risk pre- mium pricing decision in the case of power sector projects. The regression yields a relative- ly high R square, explaining about 70.1 percent of the dependent variable's variation. The coefficients on independent variables are significant and have the expected signs for volume of stocks traded and public sector dummy. The coefficient on political risk indicator (aver- age BERI) is positive, is of high magnitude, and is highly significant, which suggests that a unit increase in the regulatory risk rating would reduce the spread by 51.3 basis points. The determinants for oil and gas and telecommunications bonds are explored in a sim- ilar fashion in columns 10 and 11. The explanatory power of these models is also fairly high, with R squares of 0.742 and 0.693, respectively. The models for these subgroups of infra- structure bonds provided results similar to those of the power sector, with real GDP per capi- ta additionally proving significant in the telecom sector and the principal bond amount sig- nificant in oil and gas. The public sector dummy has even greater significance in decreasing the risk of bond default in these sectors than in the power sector. Another important distinction between the power and the oil and gas and telecommunications sectors is that political and regulatory risk (as indicated by the coefficient on the average BERI) is less important for oil 31 Governance Impact on Private Investment and gas and telecom projects than for power projects. This finding supports the hypothesis that power sector investments are riskier from the political perspective, because the product is consumer-oriented and serves the entire voter base, and is therefore highly susceptible to government price control and other regulations. The results of this regression make it possible to estimate the cost of governance-borne risks in emerging bond markets. Substitution into the spread equation of the values obtained for the independent variables gives a proxy of the costs of regulatory and political risks to infra- structure bond issuers, calculated as spread payments that could have been avoided, of about $677.7 million in 1997.15 Impact of Country-Level Governance Risk Factors on the Spread An important goal of this research is to establish the impact of various political and regula- tory risks on the risk price of infrastructure bonds. To estimate the "pure" impact on bond risk premiums of the individual components of the ICRG, BERI, and Transparency International index groups-as opposed to the composite indexes (BERI average and prin- cipal components) used in the above regression analyses-the spread model was separately tested with each of the regulatory and political risk indexes. Tables B.1 and B.2 at the end of this paper present the full results of these regressions. Table 9 compares the effects of each of these factors on the spread. Table 9 Results of Spread Regression on Various Governance Risk Factors Riskfactor name Standardized coefficient P-value R-square Corruption and autocracy risks cluster average -.414*** .000 .645 Political disorder risks cluster average -.166** .002 .601 Economic policy sustainability risks cluster average .082* .063 .592 Fundamentalism risks cluster average -.046 .246 .589 Communication and transport infrastructure quality (BERI) -.536*** .000 .648 Bureaucratic delays (BERI) -.382*** .000 .631 Contract enforceability (BERI) -.301*** .000 .614 Corruption Perceptions Index (TI) -.269*** .000 .618 Political party development -.265*** .000 .649 Quality of the bureaucracy -.248*** .000 .614 Corruption in government -.245*** .000 .620 Nationalization risk -.240*** .000 .601 Law and order tradition -.173*** .006 .599 External conflict risk -.081** .032 .595 Political terrorism -.066 .151 .591 Racial and nationality tensions -.048 .263 .590 Economic expectations vs. realities .042 .340 .590 Civil war -.047 .370 .590 Military in politics .044 .395 .590 Political leadership -.027 .540 .589 Organized religion in politics -.011 .781 .589 Economic planning failures -.009 .844 .589 Note: * - denotes significance at 10 percent level; ** - denotes significance at 5 percent level; *** - dcnotes sigiiificaince at 1 percent level. Source: Author's calculations. 32 Impact of Governance Risks on Investment Cost at Entry Table 9 lists the individual political and regulatory risks in order of the magnitude of their impact. Transparent and efficient bureaucratic systems, the absence of corruption, and the presence of democracy are shown to be the most effective tools in reducing the risk per- ception of an investor in infrastructure bonds. These factors are followed in importance by the traditional political risks of nationalization, expropriation, and war. The regressions of the spread on individual regulatory risk indicators are fully consistent with the findings of the earlier principal component and correlation analysis. Test for Universe Selection Bias-Truncated Regression The bonds included in this universe were by default all subject to conditions of issuance- that is, they were all issued in an environment that the issuer deemed to be equal or above an acceptable safety threshold. No bonds have ever been issued in an environment deemed too risky. This raises the concern that the universe of bonds is biased. Information on defunct projects that were initiated and aborted, or in which the issue of bonds otherwise failed, would potentially serve to redress this bias, but consistent information on such bonds is unfortunately not available. The econometric technique of truncated regression was used to test the potential of a selectivity bias and to obtain a more accurate (unbiased) estimate of the regression coefficients (table 10). Truncated regression relies on the assumption that information is only available on successful projects for which Y< c, where Yis the spread and c is a threshold that may be interpreted as the maximum spread (serving as an investment environment risk indicator) allowable for a project to be financed through a bond issue. (Detailed specifications of the truncated regression test are outlined in Methodology Appendix 3.) Should the selection bias be strong, the truncated regression would result in coefficients of significantly larger mag- nitude. In this case, the truncated coefficients were found to be larger but not significantly so, suggesting that the sample bias is not large and that the previous analyses are therefore legitimate. Table 10 Coefficients of the End Result of Automated Stepwise Regression Nonstandardized coefficients of Nonstandardized coefficients of Variable truncated regression OLS regression Currency amount -.000258 (.172) -.000254 (.170) Real GDP per capita -.099*** (.000) -.092*** (.001) Stocks traded -.189*** (.000) -.178*** (.000) GDP growth -.007 (.643) -.006 (.692) BERI -.854*** (.000) -.846*** (.000) Power .607*** (.000) .581*** (.000) Oil/gas .503*** (.000) .428*** (.000) Telecom .711*** (.000) .682*** (.000) Issuer public -.627*** (.000) -.589*** (.000) R square .646 .642 Note: *** - denotes significance at 1 percent level. Figures in parentheses are p-values. Source: Author's calculations. 33 Governance Impact on Private Investment Model Application: Estimation of Infrastructure Bond Spreads in the Czech Republic and Hungary As a case study, the research applied the model developed here to estimate the expect- ed risk return for infrastructure investors in two transition economies, the Czech Republic and Hungary, in 1997. The Czech Republic notably issued more infrastructure bonds than any other Central and Eastern European country; Hungary, in contrast, preferred other types of infrastructure financing.16 Infrastructure Finance in Western and Eastern Europe European countries are increasingly using bonds to fund infrastructure development. Table 11 summarizes the information on all infrastructure bonds issued in the countries of Central and Eastern Europe. According to these data, 37 bonds were issued in the region during 1990-98 to finance power, oil and gas, telecommunications, and rail and transportation projects. The Czech Republic leads with 23 bonds issued, 13 of which were issued by private infrastructure providers. The volume of infrastructure bond financing during the same period was much greater in Wtestern European countries. For the purpose of providing a perspective for the Central and Eastern European bond issues, table 12 summarizes U.S. dollar bond issues in Wzestern Europe for 1990-98. Table 13 presents the average risk premiums spreads on infrastructure bonds issued dur- ing 1995-98 in selected countries throughout Europe. The spreads are listed in basis point units. Prediction of the spread as a measurement of the cost of investor risk perception might help explain the differences in the infrastructure finance approaches of the two lead- ing transition economies. Model Application to Estimate Risk Spreads on Czech and Hungarian Infrastructure Bonds The resulting estimate is useful for comparing risk levels in these two countries with those of other emerging economies that have issued infrastructure bonds. Table 11 Registered Eastern European Bond Issues, 1990-98 (All Currencies) Number of issues, by industry and issuer type Power Oil/gas Railroads Telecom Total Issuer Central Public Private Private Private Pub6ic Public Private Prinvate Private Country nationality governmrent utility corporole utility corporate utility corporate corporate finance utility Czech Rep 8 9 1 2 2 1 23 Estonia 1 1 Poland 1 1 Russia 1 2 1 4 Slovak Rep. 2 1 4 7 Total 2 9 9 3 5 4 2 1 1 1 37 Source: Euromoney Bondware. 34 Impact of Governance Risks on Investment Cost at Entry Table 12 Registered Western European Bond Issues, 1990-98 (U.S. Dollar Issues) Number of issues, by industry and issuer type Energy Financial utility corporate Oil/gas Telecom Public Private Private Public Private Private Private Private Private Private Country utility corporate utility finance finance coporate utility corporate finance utility Total Austria 2 2 3 7 Germany 1 Denmark 1 1 Spain 1 2 2 1 6 France 3 1 2 6 Greece I I Ireland 2 2 Luxembourg 1 1 Netherlands 1 5 1 4 1 12 Norway 5 4 2 11 Sweden I I Uniited Kingdom 4 12 2 14 9 27 2 70 Total 7 4 12 4 21 6 17 2 7 34 1 3 119 Source: Euromoney Bondware. Table 13 Actual Average Spread Observations in Western and Central/Eastern European Countries, 1995-98 Country Average spread of public issuers Average spread of pivate issuers Austria 54 105.5 Spain 54.5 71.7 France 19.25 33 Britain 70.1 Greece 145 Netherlands 20 50 Norway 9.5 102.3 Sweden 96 Poland 428 Slovak Republic 115 Source: Euromoney Bondware, author's calculations. Using the regression coefficients estimated by the spread model in Section 6, as well as actual governance and macroeconomic data, spread estimates were prepared for hypotheti- cal bonds with a principal of US$250 million issued by Hungarian private and public sector entities in the power, oil and gas, telecommunications, and transport sectors in 1997. Similar estimates were prepared for the Czech Republic, which led Eastern and Central Europe in the number of infrastructure bond issues, as the data on actual bond spreads for the Czech Republic are not available. The results were compared with actual average spreads across the industrialized, emerging, and developing country groups. The results of the model estimation for Hungary and the Czech Republic are shown in table 14. The aggregation of actual spread data for the three groups of countries and for the different infrastructure sectors is also shown. 35 Governance Impact on Private Investment Table 14 Model Estimation of Risk Spreads, Worldwide and by Country, 1997 Mean spread of public issuers Mean spread of private issuers Stable countries cluster 56.32 (power) 112 (power) (actual) 28 55 (telecom) 318.53 (telecom) Weak countries cluster 166 74 (power) 241.43 (power) (actual) 245 58 (telecom) 313.43 (telecom) Risky countries cluster 296.4 (power) 249.6 (power) (actual) 394.07 (telecom) 286.67 (telecom) Hungary 181.3 (power) 380 (power) (estimated) 200.34 (telecom) 383.8 (telecom) Czech Republic 130.78 (power) 272.9 (power) (estimated) 132.16 (telecom) 275.6 (telecom) Source: Euromoney Bondware, author's calculaiions. The estimates of risk spreads on the hypothetical power and telecommunications bond issues by public and private issuers in Hungary are higher than those for the Czech Republic. They are also higher than the actual mean spreads of bonds issued in emerging economies (the "weak" countries) and of the bonds issued in most of the developing ("risky") countries. They are more than three times higher than the spreads on bonds in the developed ("sta- ble") countries. With the exception of the estimated spread on Hungarian public telecom bonds, all estimates are greater than the actual mean spreads on bonds issued in emerging mar- ket countries. Importantly, the model evaluated private sector bond investments as very risky in Hungary, with estimated spreads on private telecommunications, power, and transport bonds being significantly higher than the average estimates of those for developing countries. By contrast, the Czech Republic, which has been financing infrastructure through bond issues, including private ones, has risk spread estimations within the bounds of those of the emerging economies, and lower than those of Hungary. In practice, the risk level depends on the specific circumstances of each investment, but there are also general factors-including political and regulatory factors-that determine the risk rating. These factors lift the spreads for Hungary, for which a significant portion of the spread estimate variation is attributed to governance risks. A deeper understanding of the nature of these risks in Hungary is important for the promotion of private infrastructure investment. Industry-Level Regulatory Risks: The Cost of Regulatory Discretion To attract private investment at reasonable cost, governments must make credible commit- ments to rules that safeguard property rights, provide for enforcement of contracts, and ensure transparent and stable regulation. There is no single regulatory design that can be applied to all countries towin the confidence of investors, butworks byNorth, Levy, and Spiller (1994), W7eingast (1995), and Henisz (1996) suggest that utility regulation is likely to have the great- 36 Impact of Governance Risks on Investment Cost at Entry est credibility in conjunction with political systems that restrain legislative and executive rent- seeking and opportunism. Such restraints, however, are also likely to bring about efficiency losses and higher transaction costs. The shape of regulatory institutions in a society is usual- ly derived from the "specific institutional endowment of the nation" (Levy and Spiller, 1994),comprising its legislative and executive institutions,judicial institutions, ideology, and customs and culture, as well as the country's administrative capabilities. This research has thus far concentrated on country-level governance risk factors. It will next address the impact on the spread of risks that stem from governmental regulation of par- ticular industries. Infrastructure activities are normally subject to detailed industry- or project-specific reg- ulation, tailored to the objectives of those activities. The risk to investors arises from uncer- tainty over how the government will exercise its regulatory authority over the investment in order to control entry, prices, profits, and other parameters. Depending on the design of the regulatory framework, this risk might encompass one or more of three actions: (1) the exer- cise of discretion reserved to the government by law or contract in a manner unfavorable to the investor; (2) the amendment of laws or other unilateral regulatory instruments in a man- ner unfavorable to the investor; and (3) the repudiation of a specific contractual commitment on regulatory issues. Empirically it is most plausible to attempt to discern the impact of industry-level regu- lations in the power sector, as this sector is characterized by distinguishable regulatory frame- work patterns that are common internationally. In contrast, regulation of the oil and gas and telecom sectors varies greatly from country to country, and hence is difficult to categorize for modeling purposes. The analysis of this section is based on a set of exogenous industry characteristics, cho- sen as independent variables, which are likely to explain the bond credit risk. These variables include the extent of privatization, industry structure type (vertical vs. unbundled), and the ability (discretion) of a regulator to set or change prices independently of other political actors. Ownership and structure data and information on the pricing discretion of the power sector regulators were compiled from various literature sources, including the 1999 World Energy Council survey of power sector restructuring around the world and the 1999 ESMAP Report on the Restructuring of the Electricity Industry in Developing Countries. Table 15 defines the dummy variables on which the data were compiled. Initial correlation analysis of the relationship between these variables and the spread on power bonds revealed that the predominance of government ownership in the sector, as measured by no or limited privatization, is viewed as risky by private investors and is signifi- cantly positively correlated with the spread on power bonds. Vertical integration of electricity utilities is negatively related to the spread, suggesting that unbundled utilities are perceived as riskier by bond investors. Without government backing, unbundled utilities do not enjoy monopolist rents and face commercial and increased political risks. To the private investor, vertically integrated monopolies appear to be a safer investment. 37 Governance Impact on Private Investment Table 15 Classification of Employed Power Sector Governance Framework Characteristics Variable Power sector characteristics Extent of privatization 1, if no privatization taken place or privatization process initiated 0, if privatization substantially complete or always in private ownership Industry structure 1, if vertically integrated industry 0, if unbundled industry or mixed structure Regulator pricing 1, if regulator can set/change prices at its own discretion (that is, does not have discretion to follow the process of approval by other government policy institutions) 0, otherwise Source: World Energy Council's survey and author's definitions. It is also notable that there is a significant negative relationship between the spread and the ability of a utility regulator to set prices at its own discretion. This correlation suggests that the absence of political constraints on regulatory discretion increases the credit risk and deters investors (table 16). This phenomenon is studied in greater detail later in this section. Instrumental Variable Approach to Regulatory Discretion Measurement The endogeneity of the variable that indicates whether or not the power regulator has the discretion to set prices independentlv of other political or regulatory agencies has not been addressed by earlier research in this area. Such discretion is a function of a variety of struc- tural and political factors. If included in the regression of the risk premium without correc- tion for endogeneity, there is a risk that this variable will cause multicollinearity effects with other independent variables. To correct for this problem, the regulatory discretion to set prices variable was instrumentally predicted using exogenous structural societal and political vari- ables. The estimation (prediction) was performed via instrumental variables regression (two- stage least squares regression), characterized by the system of two simultaneous equations with two endogenous variables, with regulatory pricing discretion and the spread on the left hand Table 16 Correlation Coefficients of Power Industry Characteristics and the Log of Spread Log No Vertical Regulatory Correlation spread privatization integration discretion Log spread 1.00 No privatization .176** 1.00 Vertical integration -.296** .071* 1.00 Regulatory discretion .287** -.041 -.314** 1.00 Note: * - denotes significance at 10 percent level; * - denotes significance at 5 percent level. Source: Author's calculations. 38 Impact of Governance Risks on Investment Cost at Entry side and exogenous macroeconomic and structural governance factor variables on the right. (Detailed specifications of the applied model are provided in Methodology Appendix 4.) The variables used for prediction of an instrument free of the endogenous character- istics of the original regulatory pricing discretion variable include the Political Constraints Index; the BERI average macro-level regulatory risk indicator "blended" from the regulato- ry risks of greatest concern to infrastructure bond investors; and macroeconomic stability fac- tors, including real GDP per capita, volume of stocks traded as a percentage of GDP, GDP growth, and the level of foreign direct investment. To create an instrumental variable that would predict the values of an endogenous vari- able of regulatory pricing discretion, this variable was regressed on the above exogenous polit- ical environment characteristics. The resultant regression is based on 415 data points and has a relatively small explanatory power, explaining only about 17.1 percent of the dependent variable's variance. Both governance-related independent variables are significant and have a different impact on the predicted variable (table 17). Regression results suggest a significant negative relationship at 95 percent level between the number of veto points (political constraints) in the political system of a country and the likelihood that the regulator has discretion to set prices for the power sector. The variable measuring the level of country's bureaucratic quality, the risk of nation- alization, the law and order tradition, and contract enforceability is also significantly nega- tively related with the variable of regulatory pricing discretion. (Higher values of this variable imply lower risk associated with the components of the variable.) This finding suggests that societies that have high political and regulatory stability are likely to be characterized also by the absence of absolute regulatory pricing discretion. Table 17 Instrumental Variable for Power Sector Regulatory Discretion Independent variable Nonstandardized coefficients BERI average -.185** (.012) Political Constraints Index -.345**(.043) GDP per capita -.012 (.389) Volume of stocks traded -.0002 (.991) GDP growth -.020** (.016) Foreign direct investment .041**(.002) Principal amount -.014 (.602) Issuer public -.076* (.090) Power sector structure (vertically integrated vs. unbundled/mixed) -.128** (.025) No privatization -.157** (.004) Degrees of freedom 415 R square .171 Note: Dependent variable - Power Regulator Price Discretion (binary). * - Denotes significance at 10 percent level; ** - denotes significance at 5 percent level. Figures in parentheses are p-values. Source: Author's calculations. 39 Governance Impact on Private Investment Foreign direct investment and the volume of stocks traded as a percentage of GDP dif- fer in their impact on the regulatory (discretion variable. Specifically, larger volumes of foreign direct investment indicate an increased likelihood that the price regulator has discretion to set prices. In contrast, the volume of stocks traded has a negative relationship with the dis- cretion variable. Estimation of the Spread Determinants, with the Regulatory Discretion Variable Corrected for Endogeneity The next step of this analysis estimated the regression model of the log of bond spread using the following independent variables: real GDP per capita, volume of stocks traded, foreign direct investment, principal amount of the bond, issuer public dummy, and the extent of power sector privatization and vertical integration (power sector structure). An instrumental vari- able estimating the discretion of the price regulator was also used (table 18).Regression was estimated on 308 data points with an R square of 0.637. According to the regression results, all independent variables, with the exception of pub- lic issuer status and the power sector structure, have a significant relationship with a depen- dent variable. The coefficients on significant variables are in most cases as expected. Specifically, for bonds issued in countries with higher GDP per capita levels, better-developed stock markets, and larger amounts of foreign direct investment, the spread is likely to be smaller. Spreads on bonds issued with larger principal amounts also tend to be lower. In countries where there has been no privatization of the power sector, bond risk premiums tend to be higher. The variable of interest-the predicted extent of the price regulator's discretion-has an increas- ing effect on the bond spread. This reflects a higher risk perception for bonds issued by power sector utilities, which are subject to unconstrained price regulation. Table 18 Two-Stage Least Squares with Instrumental Variable Regression Results Independent variable Nonstandardized coefficients Regulator pricing direction predicted (instrumental variable) 2.844*** (.000) Real GDP per capita -.067** (.011) Principal amount -.173** (.004) Volume of stocks traded -.131** (.004) Issuer public -.120 (.195) Foreign direct investment -.153*** (.000) GDP growth .049*** (.000) No privatization .352*** (.000) Power sector structure (vertically integrated or unbundled/mixed) .087 (.48 1) Degrees of freedom 308 R square .637 Note: Dependent variable - log of bond spread. ** - Denotes significance at 5 percent level; *** - denotes significance at 1 percent level. Figures in parentheses are p-values. Source: Author's calculations. 40 Impact of Governance Risks on Investment Cost at Entry Test of Applied Approach Prudence: Augmented Regression Test An augmented regression test was conducted to test whether or not there is sufficient corre- lation between the disturbances to warrant estimation of regulatory pricing discretion by instru- mental variables. The augmented regression was conducted by including the predicted val- ues of the endogenous right-hand side variable (regulatory pricing discretion) as a function of all exogenous variables, in a regression of the original model of the spread (see Methodology Appendix 4). The test checks whether the coefficient on the predicted endogenous variable is significant, and if so if itjustifies application of two-staged least squares regression for spread explanation through industry-level regulatory risks. Table 19 presents the results of the aug. mented regression test. The results of the test indicate that the coefficient on the predicted variable for regu- latory pricing discretion is significant at the 99 percent level. This allows us to conclude that the two-staged least squares with instrumental variable estimation approach was prudent. The empirical tests of the impact of regulatory discretion on the risk perception of investors in power utility bonds provide food for thought, rather than seek to establish gen- eral truisms. On the one hand, there is a strong significant positive relationship between the variable measuring regulatory discretion and the risk premiums of power project bonds. This finding would seem to suggest that investors perceive the independent price setting capability of a regulator as an additional political hazard that increases the risk of a project. On the other hand, expert independent regulators with discretionary powers are often perceived as a bet- ter alternative to patchwork regulation provided by or with the government. WAhile empiri- cal results can convey information about general trends, case-by-case qualitative research is necessary to reach a full understanding of the dynamics of interaction between private invest- ment and the regulatory regime in individual countries. Table 19 Augmented Regression Test for Appropriateness of Instrumental Variable Approach Independent vaiable Nonstandardized coefficients Regulator pricing discretion .327*** (.001) Regulator pricing discretion predicted (instrumental variable) 2.541*** (.000) Real GDP per capita -.066** (.011) Principal -.132** (.028) Volume of stocks traded -.135** (.002) Issuer public -.126 (.167) Foreign direct investment -.162*** (.000) GDP growth .050*** (.000) No privatization .370*** (.000) Power sector structure (vertically integrated vs. unbundled/mixed) .074 (.539) Degrees of freedom 308 R square .652 Note: Dependent variable - log of bond spread. ** - Denotes significance at 5 percent level; *** - denotes significance at 1 percent level. Figures in parentheses are p-values. Source: Author's calculations. 41 7. True Market Perceptions of Governance Influence on Infrastructure Finance: Evidence from the Time-Series Bond Data Risk Premium in the Secondary Market: Real Factors of Change Conceptual Framework Analysis of risk spreads on bonds traded in the aftermarket is important for understanding which factors are significant in bond risk pricing in real-time economic and market condi- tions. Analysis of aftermarket bond trading enables investigation from a different angle of the relationship between risk spread, governance, and other risk factors-that is, from the per- spective of the "perfect market"judgment. It also enables comparison of the perfect market formulation of credit risk and that of the credit risk analysts of the underwriting investment banks and investment funds who priced the bonds at their issue. When bonds are traded, spreads are determined by the market, so secondary market bond spreads should therefore represent the true composition of risk factors that affect the bond over its lifetime. This sec- tion of this research paper compares the analyst-perceived and market-induced determinants of bond risk premiums in order to identify whether or not there is a systematic difference between the two. Data Sources This analysis uses time-series data on those bonds from the main universe (that is, bonds issued in U.S. dollars in 1990-99, as reported in the Euromoney Bondware database) that were trad- ed in the aftermarket, as registered by the Bloomberg data vendor. The data universe encom- passes 244 bonds issued in the electricity, oil, gas, and telecommunications industries of 34 countries.'7 Industry sector representation among these bonds is relatively even (table 20). The institutional type of issuers varied. Private corporations accounted for the largest share of bonds issued (79), with private utilities (54) and public utilities (50) also account- ing for a large proportion. The full breakdown of issuers is shown in table 21. The dependent variable in this analysis is the secondary market indicative sale spread. Information on these spreads was compiled from time-series data on the semiannual yields of those main universe bonds that were traded in the aftermarket, as registered by the Bloomberg data vendor. The spreads were then calculated by taking the difference between these yields and the yields on U.S. Treasury bonds of the same maturity, as obtained through 42 True Market Perceptions of Governance Influence Table 20 Distribution of Industry Sectors of Bond Issuers Issuer industry Number of issues Percentage Electricity 81 33.33 Financial 26 10.70 Oil/gas 61 25.10 Telecom 75 30.86 Total 243 100.00 Source: Euromoney Bondware. Table 21 Distribution of Institutional Types of Bond Issuers Institutional type Number Percentage Central government 1 0.41 Public corporate 25 10.25 Public finance 4 1.64 Public utility 50 20.49 Subnational 5 2.05 Private corporate 79 32.38 Private financial 26 10.66 Private utility 54 22.13 Grand total 244 100.00 Source: Euromoney Bondware. interpolation of the historical U.S. government yield curve. This exercise resulted in 26,905 weekly spread observations for the 244 bonds. The relatively constant nature of the independent variables data (single observation for bond financial and yearly observations for macroeconomic and political/regulatory indica- tors) means that the time-series database could be constructed using only the annual secondary market spread observations. The dependent variable-the spread-was taken as reported in the third week ofJanuary for each year throughout the bond's life in the secondary market. To allow for delays in the information on macroeconomic, political, and regulatory risk fac- tors leaking into the market, the independent variables observations were matched to these annual spread observations using the indicators from December of the preceding year. Bond financial characteristics were used as at issue. The resulting database of historical spread obser- vations comprises 253 observations, with those bonds that traded for less than one year in the secondary market dropped from the sample universe. Model Specification: Generalized Estimating Equations Model The model chosen to test the determinants of the historical spreads uses the GEE (general- ized estimating equations) population-averaged model approach, which assumes that the spread data for each bond are intercorrelated over time. Specifically, in addition to the main 43 Governance Impact on Private Investment independent variables (such as real GDP per capita, etc.) the model also includes "nuisance" variables measuring covariations between spread observations over time. There are two approaches to estimation of these nuisance variables: AR(1), which reduces the set of nui- sance covariates to only one parameter; and the Robust approach, which considers as many nuisance covariates as there are different pairs of observations in the same cluster (see Methodology Appendix 5). The results obtained via these two approaches are presented in the third column of table 22. Summary Results of Model Estimation The results of both measurement approaches are very close, suggesting that there is little, if any, time factor estimation bias. A comparison of historical data analysis (table 22, columns 2 and 3) and cross-sectional data analysis (column 4) may be used to test for systematic dif- ference between bond risk pricing by analysts (at issuance) and by the market (in secondary market trading): The signs of the at-issuance and real-time coefficients are the same, with the exception of the principal amount, the higher size of which in the secondary market conditions is found to actually increase the risk of bond defaults. The impact weights of the coefficients on the spread, however, are distributed differently. The magnitudes of all of the secondary market spread coefficients, with the important exception of the governance risk rating, are lower than those of the at-issuance spread. For example, the macroeconomic indicator of greatest importance in at-issuance risk pricing- Table 22 Results of Historical Spread Regression Standardized coefficients AR(1) estimation model Robust estimation model At-issuance spread OLS model Independemt (time series) (time series) (cross-sectional) Power .165*** .183*** .256*** Oil/gas .117** .137** .203*** Telecom .108 .124 .245*** GDP growth -.005** -.011*** -.021 BERI -.731*** -.752*** -.517*** Real GDP per capita -.067** -.037** -.193*** Public issuer -.090*** -.079*** -.248*** Gross foreign investment -.109 -.104 -.005 Volume of stocks traded .036 .034 -.166*** Principal .099** .124** -.089** R square .653 .758 .651 Note: Dependent variable - log of bond spread. ** - Denotes significance at 5 percent level; *** - denotes significance at 1 percent level. Source: Author's calculations. 44 True Market Perceptions of Governance Influence real GDP per capita-is attributed a more modest weight by the secondary market evaluation of the risk of bond default. More important, in the view of the market, is the rate of economic growth in the coun- try: the significance of the coefficient on this variable has changed from less than 90 percent to 99 percent probability. The market additionally considers power sector projects to be the riskiest, with telecommunications projects carrying no additional risk, per se. In contrast to the situation at bond issuance, the volume of stocks traded has no influence whatsoever on the risk pricing of previously issued bonds. The issuance of bonds by public entities still appears to impress investors as a guarantee of safety from default. The governance risk factor has the greatest impact on the spread, signifying that investors in off-the-run fixed income securities are even more polity- and regulatory risk-averse than the analysts pricing bonds at issuance perceive them to be. Analysis of the interaction of the spread and mean indicators of four governance risk clusters-corruption and autocracy, fundamentalism, political disorder, and economic poli- cy sustainability-supports the finding of earlier cross-sectional analysis on the hierarchy of these clusters (table 23). Time-series examination of bond credit risk structure in the secondary market supports the earlier findings, with the only difference being that all four governance risk clusters have a significant effect on the secondary market spread. It is notable that when the four regula- tory risk variables are taken together in the regression analysis of historical spreads, their com- bined power seems to crowd out the influence of other independent variables. Table 23 Results of Historical Spread Regression, with Four Regulatory Risk Clusters Independent variable Nonstandardized coefficients and P-value Power .359 (.217) Oil/gas .382 (.216) Telecom .260 (.399) GDP growth -.012 (.826) Real GDP per capita -.073** (.050) Public issuer -.147 (.192) Foreign direct investment -.052 (.243) Volume of stocks traded .040 (.492) Corruption and autocracy risk cluster average -.996*** (.000) Political disorder risk cluster average -.344** (.027) Economic policy risk cluster average -.213*** (.000) Fundamentalism risk cluster average -.298** (.014) Principal .099** R square .653 Note: Dependent variable - log of historical bond spread. ** - Denotes significance at 5 percent level; *** - denotes significance at 1 percent level. Figures in parentheses are p-values. Source: Author's calculations. 45 Govenance Impact on Private Investment Regulatory Risk and Credit Rating History of Infrastructure Bonds The fourth approach of the regulatory risk impact analysis examines the link between the gov- ernance framework and bond credit rating history. This approach is centered on the deter- mination of if and how regulatory and political risk factors affect rating downgrades and upgrades by the major credit control agencies Standard and Poors and Moody's. Data and Estimation Approach This research processed the upgrade and downgrade information registered in. the Euromoney Bondware database by assigning numerical values to the letter grades issued by the two agencies and registering the date when the rating change occurred. In order to estimate the probability of rating change, binary codes were assigned to bonds using the Rating Change (RC) variable: that is, if RC < 0, then Binary Change = -1; if RC > 0, then Binary Change = 1; otherwise 0. Using the date of rating change, a time-series database was constructed by linking these events with corresponding indicators of macro- economic and political risk conditions. The database contains 549 observation points drawn from annual observations of the rating changes registered for 245 bonds in the Euromoney Bondware database. The purpose of this analysis is to determine how changes in the governance regime and macroeconomic and firm conditions affect the decision of the rating agencies to alter the cred- it rating of issued bonds, and therefore indirectly the risk price of infrastructure bonds, as measured by the spreads. The analysis used the multinomial logit technique for three out- comes, which enables estimation of the probability that an observation belongs to one of the three observation groups: (-1) downgrade; (0) no change; and (1) upgrade. (Detailed spec- ifications of the multinomial logit model can be found in Methodology Appendix 6.) Results of Model Estimation The results of the multinomial logit regression suggest that changes in governance conditions, stock market growth, gross foreign direct investment, real GDP per capita, GDP growth, and firm-intrinsic risk factors significantly affect the decision made by the major rating agencies to change a rating (table 24). Analysis of the multinomial regression results segregate the factors of credit rating sta- bility from the factors of credit rating change (both positive and negative). According to this analysis, GDP per capita, gross FDI, and the industry sector dummies are variables capable of inducing credit rating change. Specifically, an increase in GDP per capita has a positive significant influence (at 95 percent level) on the probability of a bond being upgraded, but no influence whatsoever on the probability of downgrade. Positive changes in the power and oil and gas sector firm-intrinsic conditions have a significant negative influence on the prob- ability of a bond being downgraded, with negative changes accordingly increasing the down- grade probability of the bond. GFDI has a positive significant impact on the probability of a 46 True Maket Perceptions of Governance Influence Table 24 Multinomial Logit Regression Results Nonstandardized coefficients and P-value Independent variables Probability to belong to (-1) Probability to belong to (1) BERI average (governance risk) -4.796* (.057) -7.347** (.008) Volume of stocks traded 2.073* (.053) 1.866* (.067) Foreign direct investment 1.879** (.039) -.013 (.992) Real GDP per capita -.115 (.813) 2.163** (.005) GDP growth -.343** (.007) -.513*** (.000) Power dummy -5.995** (.014) -.329 (.698) Oil and gas dummy -4.067* (.082) -.811 (.437) Issuer public dummy .258 (.779) .211 (.812) Principal amount -.001 (.645) -.001 (.651) Constant 2.262 (.678) -10.959 (.019) Summary statistics Pseudo R2: 0.371 Number of observations: 332 Note: * - denotes significance at 10 percent level; ** - denotes significance at 5 percent level; *** - denotes significance at 1 percent level. Figures in parentheses are p-values. Source: Author's calculations. bond being downgraded, a finding that might be explained by the increased competition to existing investments that accompanies a rise in FDI. The second set of significant independent variables-the BERI average, measuring polit- ical and regulatory risks, and GDP growth-serve as factors of credit rating stability. Specifically, both the greater rate of GDP growth and the higher level of governance stability have a neg- ative impact on both the probability of a bond being upgraded or downgraded. Essentially, this signifies that a large increase in political and regulatory stability in a country decreases the volatility of credit rating change; a decrease in the quality of governance is conversely like- ly to bring about instability in bond credit ratings. Changes in GDP growth affect a similar pattern of change on bond ratings. Finally, the volume of stocks traded variable acts as a mirror reflection of the regulato- ry and growth variables. An increase in the volume of stocks traded has a positive significant effect on the probability of a rating upgrade or downgrade. This can be attributed to the link between a rise in a country's trading profile and the amount of attention that rating agencies pay to its market. The findings of this section allow us to conclude that governance frameworks, and specif- ically the associated presence or absence of political and regulatory risks, are crucial factors at the initiation of private infrastructure bond financing. Once financing has been attracted, even major improvements in the quality of governance are unlikely to affect the credibility of an investment and the cost of borrowed funds. In summary, good governance is instrumental in attracting and initiating private investment in infrastructure. 47 8. Conclusions The purpose of this paper has been to investigate the nature of interaction between the qual- ity of governance and the risks facing private infrastructure investment. The following aspects of governance were examined: (1) democratic development; (2) efficiency of the bureaucracy; (3) enforcement of the law and of contracts; (4) the transparency of and corruption in.gov- ernment activities; (5) traditional political risks of war, coups, expropriation, and tendency toward dictatorship, as evidenced bv organized religion or the presence of the military in gov- ernment; and (6) industry-level regulation. Macroeconomic, bond, and industry sector char- acteristics were also included in the analysis, thereby expanding upon earlier models exam- ining credit risk. The main research findings are as follows: First, the research confirmed the hypothesis that governance risk has a substantial impact on the risk of default on a fixed-income security issued to finance infrastructure projects. Different political and regulatory risks were additionally found to have nonhomogeneous impacts on the bond spread: cluster analysis segregated an array of risk indicators into four groups, categorized as risks of (1) political disorder; (2) corruption, red tape, and autocra- cy; (3) poor economic policy; and (4) fundamentalism. Empirical evidence suggests that what matters most to investors are the risks of cor- ruption, poor bureaucratic quality (red tape), and the lack of institutional constraints pre- venting adverse governmental action (autocracy). Second in the hierarchy of importance are the traditional risks of political disorder, as manifested by civil war, military control of poli- tics, political terrorism, the absence of a tradition of law and order, and nationalization poli- cies. Investors appear to be least susceptible to risks that stem from shortsighted economic policies and poor planning, and to fundamentalism risks such as those manifested by the preva- lence of organized religion in politics, internal racial and nationality tensions, and the risk of external conflict. In other words, investors are prepared to diversify away a fair amount of incompetence in political and economic leadership, and also of inconsistency in economic policy. Empirically, it appears that poor political leadership and economic mismanagement are increasingly being seen as diversifiable incompetence and no longer as a systemic threat to investors. Investors are also prepared to deal with differences in the cultural and moral endowments of a coun- try; they are not, however, prepared to risk the hazards that stem from systemic incentives to steal (corruption) or to break commitments (poor contract enforceability). These latter fac- tors reduce expected values and require an offsetting spread. The findings on the importance hierarchy of the four risk clusters was reinforced by the results of a time-series examination of bond credit risk structure in the secondary mar- ket. The only major difference in these results was that all four governance risk clusters had a significant effect on the secondary market spread. Spreads in the aftermarket essentially mea- 48 Conclusions sure the monetary compensation for a bond's risk of default, as dictated by real market risk judgement, as opposed to bond spreads at-issuance, which are determined by the riskjudge- ment of analysts from underwriting banks and institutional investment entities. Second, the research identified a clear separation of the countries of the sample uni- verse into three regulatory and political risk clusters, characterized by stable, weak, and risky governance climates. To a large extent, these groups to a large extent mirror what the devel- opment community conventionally calls the 'worlds" of industrialized, emerging, and devel- oping economies. Members of the risky governance cluster'8-the least developed countries and those countries struggling in transition-generally have very poor governance as well as poor macroeconomic indicators, and consequently have to pay high premiums to investors for taking on the risk of investment in such an environment. Countries in the weak cluster'9-the emerging economies-were successful in setting some of the fiscal, monetary policy, and ownership structure fundamentals right. Governance issues nonetheless remain their toughest challenge, as they have to resolve the issues that give rise to a whole range of political and regulatory risks. According to the estimates of this research, the price that infrastructure bond issuers paid in 1997 to cover the risks of corruption, red tape, and poor law enforcement was extremely high, amounting to otherwise avoidable spread payments of $697.7 million. These are the funds that could have been saved by the sponsors of infrastructure projects in emerging markets if the governments of those markets had been able to offer risk-free governance. Although these are subjective estimates, and reflect the inherent subjectivity of governance risk ratings, it is clear that poor policies and weak gov- ernance frameworks are extremely costly and that they impede the development both of need- ed infrastructure and of capital markets. Other factors dictating bond spreads in these coun- tries include the extent of capital market development, as measured by the volume of stocks traded as a percentage of GDP, and the overall macroeconomic stability, as measured by real GDP per capita. Higher values of these variables have a decreasing effect on infrastructure bond spreads. As a case study, the research applied the model developed here to estimate the expect- ed risk return for infrastructure investors in two transition economies, the Czech Republic and Hungary, in 1997. The Czech Republic notably issued more infrastructure bonds than any other Central and Eastern European country; Hungary, in contrast, preferred other types of infrastructure financing. The model estimation indicated that the cost of borrowing in Hungary should be higher and in the Czech Republic lower than the actual average spread for emerging market countries in that year, in order to compensate potential investors for the political and regulatory risks that they would face. This may partly explain the difference in the two countries' choices of infrastructure finance sources. Members of the stable cluster of countries20-the industrialized economies-are in the clear. The risk premiums of bonds issued in the political and regulatory environments of these countries were found to include no component reflecting governance risk. Investors in these countries are free to concern themselves exclusively with real commercial project risk, rather than with policy risk created by the adverse actions of government. Risk spreads are 49 Governance Impact on Private Investment predominantly determined by firm-, project-, and industry sector-intrinsic characteristics, as measured in this analysis by industry sector dummies. Another important factor is the insti- tutional type of bond issuer. The association of the bond issue with the public sector has a strong decreasing effect on the bond spread, as investors perceive this to be a sign of state underwriting of the project in the event of bond default. Interestingly, public sector associa- tion has little effect in the case of emerging markets, where investors perceive government commitments to be unreliable. The conclusion that may be drawn from these findings is that economic reform remains a necessary but not a sufficient approach to development. Third, the research found a substantial degree of variation in the impact of regulato- ry and political risks on the spread for bond issuers according to their institutional and own- ership characteristics. There is clear evidence of asymmetry in the extent of exposure to adverse government action that are faced by the different ownership types of infrastructure bonds. Specifically, private sector infrastructure projects were found to be systemically more vulnerable to the effect of political and regulatory risks than were public sector projects, and more so in the emerging market countries. Fourth, this research studied on an international level the effects of risks posed by indus- try structure and regulation. Examination of the risk premium structure of bonds issued for the financing of power sector projects the research found that there is a fine line between the benefits bestowed by regulatory independence and the hazards, as perceived by investors, that potentially stem from the same granting of regulatory discretion-particularly in regard to pricing. Specifically, the research found that the absence of political constraints on a reg- ulator's ability to set prices significantly increases the investment risk, as measured by the bond spread. Low levels of power sector pnivatization were also found to increase investor risk per- ceptions. These findings highlight a need for the careful assessment of the pros and cons when regulatory reform is undertaken and of the extent of independence afforded to regulators. Fifth, the project investigated the structure of risk spread from a new angle-that is, from the perspective of a "perfect market"judgement. It investigated whether or not the per- fect market formulation of credit risk is different from that of the analysts who set bond pric- ing at issuance. The results of time-series data analysis revealed that market risk perceptions deviate slightly from those of credit risk analysts at bond issuance in two aspects only. While regulatory and political risks remain an important and significant factor in the investor's risk perception over the bond's lifetime in the secondary market, macroeconomic conditions diminish in importance, with the single exception of the rate of economic growth, which has a significant decreasing effect on the bond risk spread. Finally, study of the credit rating history of the bonds universe enabled discernment between the factors of an investment's creditworthiness stability and the factors of credit- worthiness change. Stable governance frameworks were found to be essential for credit rat- ing stability, but firm-intrinsic factors wielded the greatest influence over changes in credit agency ratings. This suggests that governance frameworks are crucial factors at the initiation of private infrastructure bond financing. Once financing has been attained, even major improvements in the quality of governance are unlikely to affect the credit rating of borrowed 50 Conclusions funds. Good governance may be construed therefore as instrumental primarily for attract- ing and initiating private investment in infrastructure, and for maintenance of stable rates. Overall, this research provided empirical evidence that international governance con- ditions are a significant factor determining the flow of private finance to and the cost of bor- rowing for infrastructure. It also points to the need for future work to include a deeper qual- itative analysis of governance frameworks and of their associated political and regulatory risks, ideally on a case-by-case basis and including examination of the ways in which changes might be introduced into governance frameworks to provide better conditions for private infra- structure financing. 51 Methodology Appendix 1. The Mathematical Foundation of Biplot Analysis2' 0. Fact: any nxm matrix Yof rank r can be factorized (non-uniquely!) as a product Y=GH' (1) into an n by rmatrix G and mxr matrix H 1. Let us first assume an unrealistic case, that the n by m data matrix Yis exactly of rank 2, then using (1) any data entry y,j can be represented without loss of information as an inner product of the corresponding two vectors g 'and h. from matrices G and H, that is y,j =gi'h (2) By this factorization we assign 2-dimensional vectors g,, g2 ,..., g. to the rows of the orig- inal data matrix Yand 2-dimensional vectors h,, h2 ...., h. for the m columns of Y. 2. Now we can plot this n + m vectors on the two-dimensional plane, giving the represen- tation of the mn elements of matrix Y by means of the inner products of the corre- sponding row effect and column effect vectors. Such a plot was originally referred to as biplot in Gabrilel (71) as it allows the row effects and column effects to be plotted jointly. 3. In terms of the graphical interpretation, the inner product of the two vectors g and h can be appraised visually by considering it as a product of the length of one of the vectors times the length of the other vector's projection onto it. This allows one to see easily which rows or columns in the matrix Yare proportional to which rows or columns (same directions); which entries are zero (right angles between rows and columns) and so forth. 4. Now consider an arbitrary n by m data matrix Yor rank r < min (n,m). The idea is first to construct a rank 2 approximation of any such matrix or rank r (denoted as Y(2)) and then to use the biplot of this rank 2 matrix as a visual representation of the original matrix Y. Obviously, the usefulness of this biplot would depend on the quality of this approximation, that is on the distance II Y- Y(J,I. 53 Governance Impact on Private Investment 5. To approximate any rectangular n by m matrix Yof rank r by n by m matrix of a lower rank we can use the singular vallue decomposition technique. That is, we can represent Yas Y= Ai Pi q, (3) i=1 where * At is the square root of the i-th eigenvalue of the squared matrix YT and ° and (1) cov(e,jle J2) ='(i2lj) °. Therefore, unlike in the case of ordinary linear regression with mutually independent normal observations, here we are likely to have a clustering of spread observations positive- ly correlated with one another. To provide a historical perspective, the vast literature of the subject has considered models to treat such panel data. The approaches discussed include the time-series with fixed effects, random effects models, and the GEE-generalized estimat- ing equations model first developed in Liang Kung-Yee and Scoitt L. Zeger.24 Liang and Zeger introduced a class of estimating equations that give consistent estimates of the regression para- meters and their variances under mild assumptions about the time dependence. In fact, they consider a much more general case than is needed here, as they allow for non-normal errors and various link functions that connect y's with linear combination of x's. In our case, the link function is linear and error distribution is normal. W7hat needs to be estimated are the nui- sance parameters associated with unknown covariates (( j2, j,). There are two ways to mea- sure these covariations. First, the parametrical application of the GEE population-averaged model approach assumes that the covariance on the same bond would depend on how far away in time they are-6(j2 , ji) = p lj2-i4. This AR( 1) model would reduce the set of nuisance covariates to only 62 Methodological Appendix 5 one (parameter), p (of course, we have to estimate a). At the other extreme, one can con- sider as many nuisance covariates as there are different pairs of observations in the same clus- ter, and estimate them from the data. The second estimation approach, which allows accom- modation of the time factor in the spread model, is the GEE population-averaged model with the Robust estimator of the variance-covariance matrix. For estimation, the xtgee routine in STATA software was used. Several types of correlation patterns were tried: exchangeable (all same), AR(1) (first order autocorrelation), and unstructured. The results of these routines were very similar, which is consistent with some theoretical and simulation results that state that the analysis in GEE is robust to the mild deviation in the independent errors. 63 Methodology Appendix 6. Multinomial Logit Technique for Estimation of the Probability of the Bond Credit Rating Change The multinomial logit technique allows us to estimate the probabilities that an observation belongs to any of the k groups-in our case, (-1) downgrade; (0) no change; and (1) upgrade. This research exploited the multinomial logit model for k = 3 outcomes (which is an extension of the usual logistic regression with k = 2 outcomes) to simultaneously estimate the set of equations: 1+ exp(z2)) + exp(zi )) (1) zj(j) =a( j)xj, j =2,3 i=I l....n where y's are the set of binary response variables: y,j = 1 if the bond i belongs to the group j, and Yij N = 0; otherwise xi is the vector of explanatory variables for the i-th bond, and n is total number of bonds. One of the k groups (here it is the middle, 0 or no change) has to be fixed as a reference group (or base category). This means that the coefficients of the corresponding equation must be all set to zero, so that we actually estimate the set of k-1 equations, which in the case of a regular logit model with two groups reduces to a single equation with coefficients Ia0,a1 ,a2, ... a(. It can be easilyshown then, that exp(z°) expresses the probability that the bond ibelongs to a group (j), relative to the probability that it belongs to the reference category (here the reference category is the no-change group), namely: exp(z'j)) = Pr(y, = j) / Pr(yj = 1). (2) 64 Table A.1 Correlation Analysis of Data Interrelationships-Correlation Clusters &,i,ul I21sd (Q-alily (sOI7ssP Milia.Y 1.1al CiGi7 Pa5i ,cotoI: p5spn. Potblt iral Exlctmaol ((ig nsiorl ."otiooslily C ...p. Pe,iti-ol oJ B.-,sI, Cs-t(,ocl Nail-o. ZnJr, P-r.p in s,d.', Palil. n,"n A,4. eoX/snlat. jail as1sd- .onflitl sfigios leooio.s go. Pan t b--y sls.lay.s ro,f sic 7is,k pnnily As/lex politirs lmdi,,, Ii-i si.slki s ing xpeclations 1.00 0.87 0.69 0.04 (.10 (0.21 (1.33 (0.23 (03!) 0.49) 0.48 0.5(i 0.47 (0.35 (1.35 0.38 (0.41 0.32 (0.57 Ec(.7,. Plani. Fail 0.87 1.00 0.71 (1.1!) 0.20) (0.25 ((.33 (.31 (0.32 (0.43 (0.45 (0.52 (.38 0.2Yi 0.3(0 0.31 0.28 ((.21 ().5(i Pos liti al (easIer 0.69 0.71 1.00 (0.2(6 0.0(5 0,22 (0.33 (0.34 0.25 (.3(6 ().38 0.40 (0.33 ).21 (.33 0.47 0.48 0.42 0.5!) E.iernal coflict ().04 (.1 ((.2(i 1.00 0.36 0.39 ().18 (0.44 ).I(i (.2(i (.24 (0.22 ().I(i ((.17 (0.23 0.24 (0.25 (0,25 (0.43 Organized religimi ().1(i ((.2(0 0.0() 0.36 1.00 0.62 ().51 (0.49) (0.48 (0.57 ((.41) 0.3( (0.48 ((.48 (0.52 (0.4!) (0.5(i (0.31 ((.(7 Racial adlt 7latissnality Irlsinsso (0.21 (1.25 (0.22 0.39 0.62 1.00 (0.24 ((.24 (0.12 ((.33 ().2(i ((.2(1 ((.21 (0.22 ((.33 (0.25 ((.41) ((.23 0.51 go mver nsrsst (0.33 (0.33 ().33 (1.18 ().51 (1.24 1.00 0.78 0.82 0.83 0.82 0.61 0.79 0.76 (0.7(i (0.78 (0.5(6 0.5(i ((.83 Po5Iitcal parly ((.23 (0.31 0.34 (0.44 (1.49) (0.24 0.78 1.00 0.76 0.79 0.77 0.65 0.74 0.69 (i() (6(ili (0.40 (0.53 (0.77 Qlitlily uII7 bureaucracy ((.39) (0.32 (0.25 ().1( ((.48 (0.12 0.82 0.76 1.00 0.85 0.87 0.73 0.84 0.77 ().66(i (0.78 0.44 (0.58 0.78 Hureacralii (Ireay. 0.41) (0.43 (0.3(i ((.2(6 (1.57 (0.33 0.83 0.79 0.85 1.00 0.95 0.81 0.92 0.86 ((.(i8 (0.77 0.54 (0.54 0.85 0isnlracl culisrc-surll ((.48 (0.45 (0.38 ((.24 (0.49) 0.2(i 0.82 0.77 0.87 0.95 1.00 0.87 0.91 0.84 ().(i!) ((.81 (0.52 (0.57 ((.84 Ilnrasiriclsr re dluality (0.47 (1.38 ((.33 ().1G 0.48 (1.21 0.79 0.74 0.84 0.92 0.91 0.78 1.00 0.79 (0((4 (0.73 (0.4G 0.50 (0.77 (Corruption Pvrcrpliol7 IBslex ((.35 ().2(i (0.21 ((.17 (0.48 (0.22 0.76 0.69 0.77 0.86 0.84 0.75 0.79 1.00 ().58 (0.7(0 (0.47 (1.57 (0,71 MiIilary in 1snlifirs (0.35 ((.3() (0.33 (0.23 (0.52 ((.33 ((.7(i ().(1i) (.(i(i ()(i8 0.(i) 0.51 ()(i4 (1.58 1.00 0.74 0.69 0.61 0.81 Nati,..n. risk (0.5(6 (0.52 ((.4(1 (0.22 ((.3(i 0.21) (.(6I (.(i5 (0.73 ((.81 (1.87 1.1(1) ((.78 (0.75 0.51 0.66 0.42 0.50 0.72 L.aw/lorder iradiin ((.38 ((.31 (0.47 (0.24 ((.4(1 ((.25 (1.78 ().66(i ((.78 (0.77 (1.81 ().1(i( ((.73 ((.71) 0.74 1.00 0.73 0.81 0.87 Pul1il. 0(rrur ((.41 (0.28 ((.48 ().25 ((.5(1 (0.49) ((.5(1 (0.4(0 (1.44 (0.54 (1.52 (1.42 ((.4(1 11.47 0.69 0.73 1.00 0.58 0.78 Civil war risk ((.32 ((.21 (1.42 0,25 (0.39 ((.23 ((.53 (0.53 ((.58 (0.54 (0.57 ().5() (.5() ((.57 0.61 0.81 0.58 1.00 0.72 P(Ifi risk rati71g (0.57 ().5(i ((.59) ((.43 (0((7 ().51 (1.83 0.77 (0.78 (1.85 ((.84 (0.72 (0.77 ((.71 0.81 0.87 0.78 0.72 1.00 Sons,:, Astt-n'sr al-nlatiots. Annex B Table B.1 OLS Regression of Log of Spread on Individual Regulatory Risk Factors Name of the kested regulatory risk Economic Organized Law expectations Etonomic Corruption Military religion and Racial! Variable vs. planning Political External in in in order nationality Political name realities failures leadership conflict government politics politics tradition tensions terrorism Foreign direct .125**s .130*** .134*** .134*## .123*** .134*s .129*** .130*** .112** .121*** invest. (.005) (.005) (.003) (.002) (.004) (003) (.004) (.003) (.016) (.007) Principal -.099## -.100 ** -.102** -.105*** -.108*-* -.103** -.100** -.100** -.099** -.096** (.016) (.015) (.014) (.OlO) (.006) (.012) (.015) (0.014) (.016) (.019) Real GDP -.418*** -.404*** -.398*** -.406*** -.328*** -.422*** -.402*** -.305*** -.397*** -.373** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) GDP growth .076 * .083 * .089** -.079** -.001 .092*1 .079* .055 .099** .099** (.076) (.056) (.044) (.060) (.974) (.038) (.064) (.201) (.029) (.027) Stocks -.406 *** -.389 *** -.385*** -.377*** -.302*** -.404*-* -.393*** -,343*** -.409*** -.394*** traded (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Regulatory .042 -.009 -.027 -.08l1** -.245*** .044 -.011 -.173*** -.048 -066 risk (.340) (.844) (.540) (.032) (.000) (.395) (.781) (.006) .263 (.151) Public sector -.223 *** -.216*** -.214*** -.214*** -.228*** -.215*** -.219*** -.240*** - 216*** -.226*** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Power sector .256*#* .2585*5 .263*** .262*** .278*** .265** .259*** .248*** .246*** .271*** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Oil or gas .252*t* .257*** .262*** .253*** .240*** .257*** .259*** .228*** .251*** .268*** sector (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Telecom .248* 2*** .252*** .254*** .242*** .251*** .248** .246-** .246#** .263*** sector (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Df 307 307 307 307 307 307 307 307 307 307 R Square .590 .589 .589 .595 .620 .590 .589 .599 .590 .591 Note: Dependent variable - log of historical bond spread; * - denotes significance at 10 percent level; ** - denotes significance at 5 per- cent level; #** - denotes significance at I percent leveL Figures in parentheses are p-values. Source: Author's calculations. 66 Annex B Table B.2 OLS Regression of Log of Spread on Individual Regulatory Risk Factors Name of the tested regulatory risk ICRG composite Communication/ Quality regulatory transport TI Corruption Political Variable of the risk Bureaucratic Contract Nationalization infra. Perceptions Civil party name bureaucracy rating delays enforceability risk qualty Index war development Foreign direct .028** .128*** .089** .108** .028 -.027 .156*** .133*** .1 21 *** invest. (.022) (.003) (.041) (.015) (.606) (.571) (.000) (.003) (.005) Principal -.098** -.095 ** -.095** -.100** -.101** -.112*** -.094** -.099** -.093#* (.014) (.019) (.016) (.013) (.013) (.004) (.017) (.016) (.019) Real GDP -.335*** -.292*** -.225*** -.261 *** -.291 *** -.258*** -.283*** -.381 *** -.231 *** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) GDP growth .003 .081 * .013 -.003 .083** -.076* .023 .083#* .057 (.947) (.051) (.770) (.947) (.050) (.098) (.595) (.050) (.161) Stocks traded -.282 *** -.389*** -.229*** -.260*** -.339*** -.099* -.286*** -.380*** -.386*** (.000) (.000) (.000) (.000) (.000) (.095) (.000) (.000) (.000) Regulatory .248*** -.182*** -.382*** -.301*** -.240*** -.536*** -.269*** -.047 265** risk (.000) (.003) (.000) (.000) (.000) (.000) (.000) (.370) (.000) Public sector -.246 *** -227*** -.234*** -.246*** -.230*** -.215*** -.251*** -.225*** -.246*** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Power sector .245*** .250*** .224*** .249*** .256*** .258*** .285*** .269*** .210*** (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.001) Oil or gas .228*** .249*** .196*** .224*** .235*** .214*** .245*** .262*** .189*** sector (.000) (.000) (.001) (.000) (.000) (.000) (.000) (.000) (.002) Telecom .229*** .246*** .221*** -.246*** .259*** .228* ** .256*** .256*** .213*** sector (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) Df 307 307 297 297 297 297 307 307 307 R Square .614 .601 .631 .614 .609 .648 .618 .590 Missed Note: Dependent variable - log of historical bond spread; * - deniotes significance at 10 percent level; ** - denotes significance at 5 per- cent level; *** - denotes significance at I percent level. Figures in parentheses are p-values. Source. Author's calculations. 67 Annex C Table C.1 Definitions of Bond Issue Characteristics Indicator Definition Yield to maturity (annual) The annual rate of return on a security if held to maturity Underwriting risk period The number of days (inclusive) between pricing date and offer date Same guarantor Guarantor and issuer are same countries Force majeure (yes/no) Indication as to whether or not a syndicate may be discharged from its obligations due to certain events and adverse material changes in international, political, and economic conditions Cross-default (guarantor) A covenant by the issuer (guarantor) that an event of default will be deemecd to have occurred in a financing if a default occurs in any of its other financing Sinking fund (yes/no) Indicates where there is a sinking fund in operation to provide for redemption of an issue at regular intervals Issuer profile (-1,0,1) Based on issuer's profile Collaterized (yes/no) Notion of whether or not a bond is secured by collateral (nature of the collateral is indicated in the report text) Launch Moody's (S&P) rating The initial Moody's (S&P) long-term debt rating assigned to a specific issue Currency amount (principal) The nominal value of the bond in the currency of issue Issuer nationality The nationality of the issuer Presence of guarantor Guaranteed or nonguaranteed bond Issuer Moody's (S&P) rating The current Moody's (S&P) long-term debt rating of the issuer Years to maturity (maturity) Number of years from issue until final maturity Guarantor industry sector The area of industry in which the guarantor of an issue operates Issuer type Further classification of the issuer-private or public, plus broad activity description (government, bank, utility, or corporation) Issuer industry sector A code field describing the principal industry in which issuer is involved (project sector dummy) Guarantor type Further classification of the guarantor, similar to the issuer type Source: Euromoney Bondware definitions. 68 Annex D Table DA Bond Issuer Characteristics Descriptive Statistics Variable N Minimum Maximum Mean Std. Deviation Yield to maturity 583 0.00 64.83 8.40 4.49 Underwriting risk 583 -8.00 7.00 0.29 0.81 period Same-company 141 0.00 1.00 0.41 0.49 guarantor Force majeure 204 0.00 1.00 0.96 0.19 Cross-default issuer 222 0.00 1.00 0.87 0.33 Sinking fund 582 0.00 1.00 0.02 0.12 Issuer profile 583 -1.00 1.00 -0.11 0.60 Collateral (yes/no) 582 0.00 1.00 0.10 0.30 Moody's launch 176 2.00 20.00 10.43 5.67 rating Standard and Poors 174 3.00 20.00 11.68 5.01 launch rating Principal 583 5.00 2500.00 277.93 270.70 Presence of 583 0.00 1.00 0.24 0.43 guarantor Issuer (S&P) 392 2.00 20.00 11.08 5.98 Issuer (Moody's) 439 2.00 20.00 10.21 5.52 Maturity 580 1.00 100.00 10.26 10.50 Spread at launch 459 2.00 778.00 199.05 174.16 Logspread 459 0.69 6.66 4.86 1.02 Source: Euromoney Bondware and author's calculations. 69 Annex E Table E.1 Macroeconomic Characteristics Descriptive Statistics Obsermations Standard Indicator number Minimum Maximum Mean deviation Stocks traded, total value (% of GDP) 429 0.00 239.42 28.51 31.98 Overall budget deficit, including grants (% of GDP) 355 -16.00 5.11 -1.29 2.79 External debt (% of GNP) 222 18.36 79.40 37.31 11.53 Private nonguaranteed debt (% of external debt) 222 0.00 56.68 18.93 13.35 PPG debt service (% of central government current revenue) 191 3.19 52.84 17.57 10.16 Total debt service (% of GNP) 222 1.55 12.93 5.29 2.65 Financing from abroad (% of GDP) 257 -2.56 11.41 0.53 1.36 Gross foreign direct investment (% of GDP, PPP) 447 0.00 11.55 2.52 2.21 Current revenue, excludinggrants (% of GDP) 357 11.16 50.03 23.12 9.01 Domestic financing, total (% of GDP) 257 -3.81 8.48 0.31 2.00 GDP per capita, PPP (current international $) 451 1120.00 30140.00 14581.93 7115.97 Imports of goods and services (% ofGDP) 412 6.08 143.12 25.38 19.00 Inflation, consumer prices (annual %) 447 -0.09 2075.90 47.53 240.58 Ratio of official to parallel exchange rate 174 0.00 1.11 0.96 0.11 GDP growth (annual %) 451 -4.02 10.63 3.67 2.99 GDFI, private sector (% of GDP) 359 7.38 35.00 18.07 5.82 Private capital flows, total (% of GDP) 428 -10.66 12.26 1.82 4.25 State-owned enterprises, economic activity (% of GDP) 117 0.40 13.05 3.94 3.17 Claims on private sector (% of GDP) 448 8.75 208.94 83.49 56.64 Public expenditure on social security and welfare (% of GDP) 300 0.60 24.50 8.14 4.42 School enrollment, secondary (% net) 209 17.50 98.00 76.54 24.56 General government consumption (% of GDP) 412 2.98 29.83 13.86 6.54 Source: World Bank Development Indicators Database ("SIMA") and author's calculations. 70 Annex F Table F.1 Macroeconomic Variables Inner Correlation Coefficients Foreign GDP Private Claims on Log Stocks Overall direct Current per GDP capital private Pearson Correlation spread traded budget invest revenues capita In]Jlation growth Jlows capital Log spread 1.00 -0.33 0.07 -0.31 -0.35 -0.65 0.25 0.19 0.44 -0.66 Stocks traded -0.33 1.00 0.10 0.34 0.16 0.34 -0.10 0.18 -0.21 0.45 Overall budget 0.07 0.10 1.00 -0.07 -0.16 -0.22 -0.27 0.43 0.29 -0.03 Foreign direct in. -0.31 0.34 -0.07 1.00 0.76 0.47 -0.18 -0.10 -0.47 0.18 Current revenues -0.35 0.16 -0.16 0.76 1.00 0.40 0.03 -0.21 -0.45 0.31 GDP per capita -0.65 0.34 -0.22 0.47 0.40 1.00 -0.23 -0.31 -0.52 0.66 Inflation 0.25 -0.10 -0.27 -0.18 0.03 -0.23 1.00 -0.02 0.15 -0.07 GDP growth 0.19 0.18 0.43 -0.10 -0.21 -0.31 -0.02 1.00 0.29 -0.25 Private capital flows 0.44 -0.21 0.29 -0.47 -0.45 -0.52 0.15 0.29 1.00 -0.47 Claims on private capital -0.66 0.45 -0.03 0.18 0.31 0.66 -0.07 -0.25 -0.47 1.00 Source: Author's calculations. 71 Annex G Table G.1 Cluster Membership: Regulatory Risk Indicators Regulatory risk indicator 4 Clusters 3 Clusters 2 Clusters Economic expectations vs. realities 1 1 1 Economic planning failures 1 1 1 Political leadership 1 1 1 External conflict risk 2 2 2 Corruption in government 3 3 2 Military in politics 4 3 2 Organized religion in politics 2 2 2 Law and order tradition 4 3 2 Racial and nationality tensions 2 2 2 Political terrorism 4 3 2 Civil war 4 3 2 Political party development 3 3 2 Quality of the bureaucracy 3 3 2 Bureaucratic delays 3 3 2 Contract enforceability 3 3 2 Nationalization risk 3 3 2 Infrastructure quality 3 3 2 Corruption Perceptions Index 3 3 2 Source: Author's calculations. 72 Annex G Table G.2 Cluster Membership: Countries by Level of Regulatory Riskiness Country 5 Clusters 4 Clusters 3 Clusters 1: Argentina (arg) 1 1 1 2:Austria (aus) 2 2 2 3: Brazil (bra) 1 1 1 4: Canada (can) 2 2 2 5: Chile (chl) 3 3 1 6: Columbia (col) 4 4 3 7: Costa Rica (cri) 1 1 1 8: Czech Republic (cze) 5 1 1 9: Denmark (dnk) 2 2 2 10: Ecuador (ecu) 4 4 3 11: Ireland (irl) 2 2 2 12: France (fra) 2 2 2 13: Germany (deu) 2 2 2 14: Greece (grc) 1 1 1 15: Hong Kong (hkg) 3 3 1 16: India (ind) 4 4 3 17: Israel (isr) 4 4 3 18:Japan (jpn) 2 2 2 19: Korea, Republic of (kor) 3 3 1 20: Luxembourg (lux) 2 2 2 21: Malaysia (mys) 3 3 1 22: Mexico (mex) 1 1 1 23: Netherlands (nld) 2 2 2 24: Norway (nor) 2 2 2 25: New Zealand (nzl) 2 2 2 26: Philippines (phl) 4 4 3 27: Poland (pol) 5 1 1 28: Qatar (qat) 3 3 1 29: Russia (rus) 4 4 3 30: Spain (esp) 5 1 1 31: Slovak Rep. (svk) 5 1 1 32: Sweden (swe) 2 2 2 33: Thailand (tha) 3 3 1 34: Trinidad and Tobago (tto) 4 4 3 35: United Kingdom (gbr) 2 2 2 36: Uruguay (ury) 1 3 3 37: USA (usa) 2 2 2 38: Venezuela (ven) 1 3 3 Source: Author's calculations. 73 Notes 1. World Bank Development Report 1999. 2. "Risky" country cluster: Brazil, Colombia, the Dominican Republic, Ecuador, India, Israel, the Philippines, Russia, Trinidad and Tobago, Uruguay, Venezuela. 3. "Weak" country cluster: Argentina, Chile, Costa Rica, Czech Republic, Greece, Hong Kong, the Republic of Korea, Malaysia, Mexico, Poland, Qatar, Slovak Republic, Spain, Israel, Thailand. 4. "Stable" country cluster: Austria, Canada, Denmark, France, Germany, Ireland,Japan, Luxembourg, the Netherlands, New Zealand, Norway, Sweden, the United Kingdom, the United States. 5. All dollar amounts are in constant 1998 U.S. dollars 6. World Bank Development Report 1999. 7. The risk premium can be measured by comparing the yield of the bond in question with the yield of a relatively riskless bond (e.g., a U.S. Treasury bond) of the same maturity. 8. Defined as a bond issued in the energy utility, oil and gas, transport, or telecommunications industrial sectors. 9. Bonds of the analysis universe were issued in the following 40 countries: Germany, Luxembourg, Ireland, Ecuador, British Virgin Islands, Denmark, Colombia, the Dominican Republic, Brazil, Poland, Russia, Argentina, the Republic of Korea, Mexico, the Philippines, India, Costa Rica, the Czech Republic, the United States, Qatar, the United Kingdom, Canada, the Netherlands, Hong Kong, Israel, Norway, Trinidad and Tobago, Slovak Republic, Greece, Cayman Islands, Venezuela, Malaysia, France, New Zealand, Chile, Japan, Austria, Sweden, Spain, and Thailand. 10. Amortization, principal, issuer name, type, sector and nationality, collateral information, coupon, guar- antor information, industry and nationality issuer type, industry and nationality, issue type and price, pre- mium amount, S&P and Moody's issue at launch, issuer and guarantor credit ratings, spread at launch, matu- rity, yield to maturity, annual yield to maturity, etc. 11. The Business Environment Risk Intelligence (BERI) Business Risk Service, the Wall Street Journal Central European Economic Review, S&P Country Risk Review, EBRD Transition Report, Economic Intelligence Country Risk Service and Country Forecast, Freedom House, Gallup International 50th Anniversary Survey, World Economic Forum Global Competitiveness Survey, Heritage Foundation Economic Freedom Index, Political Economic Risk Consultancy Asia Intelligence report, Political Risk Service International Country Risk Guide, Institute Management Development World Competitiveness Yearbook, Transparency International Corruption Perceptions Index, World Bank World Development Report 1997 Survey, World Bank Country Policy, and the Institutional Assessment Database. 12. Highervalues for governance risk indicators indicate greater safety from political and regulatory risks. 13. The exception concerns countries from the stable cluster in 1997-98. Deviation from trend at this time might be attributed to the economic crisis that began in the Republic of Korea and that affected virtually all countries by causing drastic fluctuations in risk premiums. 14. The average is taken due to considerations of correlation between these three variables and the desire to produce a weighted estimate of all factors. The same logic is applied to other use of average and mean statistics for estimation of aggregate cluster effects on the spread. 15. Measured by (0.0055 (interest attributed to 1 unit corruption and autocracy risk in weak cluster coun- tries) * $7,610 million (total principal amount of countries from the weak cluster in 1997) * 12.9 (sum of average levels of such risks in the weak cluster countries) + 0.00312 (interest attributed to 1 unit of above risk in risky cluster countries) * $3,650 million (total principal amount of countries from the risky cluster in 1997) * 12.1 (sum of average levels of such risk in the risky cluster countries). 16. Jokay, Kalman, and Kopanyi, 1998, "Municipal Infrastructure Financing in Hungary: Four Cases," Processed, Hungary Subnational Development Program, The World Bank, Washington, D.C. 17. Bonds of the analysis universe were issued in the following 34 countries: Argentina(31), Austria(2), 74 Notes Bermuda(2), Brazil(19), Canada(16), Chile(8), Colombia(3), Cayman Islands(4), the Czech Republic(l), Germany(l),DominicanRepublic(1), Ecuador(l), Spain(3), France(4), the United Kingdom(29), Greece(l), Hong Kong(1), India (2), Ireland(l),Japan(24), the Republic ofKorea(9), Mexico(9), Malaysia(11), the Netherlands (6), Norway(8),New Zealand(2),the Philippines(14),Poland (3), Russia( 1), Slovak Republic( 1), Sweden(1), Thailand(2), the United States(14), Venezuela(4). 18. "Risky" cluster members: Brazil, Colombia, the Dominican Republic, Ecuador, India, Israel, the Philippines, Russia, Trinidad and Tobago, Uruguay, Venezuela. 19. "Weak" cluster members: Argentina, Chile, Costa Rica, the Czech Republic, Greece, Hong Kong, the Republic of Korea, Malaysia, Mexico, Poland, Qatar, Slovak Republic, Spain, Israel, Thailand. 20. "Stable" cluster members: Austria, Canada, Denmark, France, Germany, Ireland,Japan, Luxembourg, the Netherlands, New Zealand, Norway, Sweden, the United Kingdom, the United States. 21. Based on Gabriel, KR, 1971, "The Biplot Graphical Display of Matrices with Application to Principal Component Analysis," Biometrika 58: 453-67. 22. See G.S. Maddala "Limited-dependent and qualitative variables in econometrics," 1983, p 167, p 176 23. This approach is implemented in E-Views software, which was utilized for this estimation. 24. 1986, "Longitudinal Data Using Generalized Linear Models," Biometrika 73(1): 13-22. 75 References Ahn, C.M., and H.E. Thompson. 1989. "An Analysis of Some Aspects of Regulatory Risk and the Required Rate of Return for Public Utilities." Journal of Regulatory Economics 1: 241-57. Alexander, I., C. Mayer, and H. MWeeds. 1996. 'Regulatory Structure and Risk at Infrastructure Firms: An International Comparison." WJorld Bank Policy Research Working Paper 1698. WA'ashington, D.C. Altman, E. and A. Saunders "Credit Risk Measurement: Developments over the Last 20 Years."Journal of Banking and Finance (Netherlands); 21:1721-42. December 1997. Appleyard, T., andJ. McLaren. 1997. "The Nature of Regulatory Risk." University of Newcastle upon Tyne. Discussion Paper 97/3, Accounting and Finance. Archer, S. 1981. "The Regulatory Effects on Cost of Capital in Electric Utilities." Public Utilities Fortnightly / February 26, 1981. Bayoumi, T., and B. Eichengreen. March 1995. "Restraining Yourself: the Implications of Fiscal Rules for Economic Stabilization." International Monetary Fund Staff Papers 42. Bergara, D., E. Mario, W Henisz, and P. Spiller. 1998. "Political Institutions and Electric Utility Investment: A Cross-Nation Analysis." California Management Review 40 (2): 18-35. Bernstein, P.L. 1996. "Against the Gods: The Remarkable History of Risk." New York: John MWiley and Sons. Black, F., and M. Scholes. 1973. "The Pricing of Options and Corporate Liabilities." Journal of Political Economy 81: 637-654 Borner, S., A. Brunetti, and B. Wteder. 1995. "Political Credibility and Economic Development." New York: St. Martins Press. Cantor, R., and F. Packer. October 1996. "Determinants and Impacts of Sovereign Credit Ratings." Federal Reserve Bank of New York. Economic Policy Review Paper No. 2. CaouetteJ.B., Altman, E., and P. Naravanan. 1998. "Managing Credit Risk: the Next Great Financial Challenge." New York: W\illey. Chandrasekaran, P.R., and W. Dukes. 1981. "Risk Variables Affecting the Rate of Return of Public Utilities." Public Utilities Fortnightly, February 26. Chermak,J.M. 1992. "Political Risk Analysis, Past and Present." Resources Policy (September): 167-78. Cline, 1W. 1995. "International Debt Reexamined." Institute of International Economics. February Cline, W., and K Barnes. December 1997. "Spreads and Risk in Emerging Markets Lending." Institute of International Finance, Inc. IIF Research Papers, Paper No. 97. Cooper, L. 1999. "The Dawn of a New Era." Risk Magazine: Credit Risk Special Report (October): 2-3 76 References Dailami, M., and M. Klein. 1998. "Government Support to Private Infrastructure Projects in Emerging Markets." Policy Research Wtorking Paper No. 1868. World Bank. WAashington, D.C. Davidson, W.N. III, and P.R. Chandy. 1983. "The Regulatory Environment for Public Utilities: Indications of the Importance of Political Process." Financial Analysis Journal (November-December). Dialynas, C., and D. Edington. 1992. "Bond Yield Spreads: A Postmodern View." Journal of Portfolio Management 19: 68-75. Domowitz, I., J. Glen, and A. Madhavan. 1998. "Country and Currency Risk Premia in an Emerging Market."Journal of Financial and Quantitative Analysis 33(2): 189-216. Edwards, S. 1986. "The Pricing of Bonds and Bank Loans in International Markets: An Empirical Analysis of Developing Countries' Borrowing." European Economic Review 30: 565-589. Eichengreen, B., and A. Mody. 1998. "MThat Explains Changing Spreads on Emerging Market Debt." RMC Discussion Paper Series, No. 123. Eichengreen, B., and R. Portes. February 1989. "Dealing with Debt: The 1930s and the 1980s." Centre for Economic Policy Research Discussion Paper Series, Paper No. 2867. Fama, F., and K. French. 1993. "Common Risk Factors in the Returns on Stocks and Bonds." Journal of Financial Economics 33: 3-56. Formby,J.P., B. Mishra, and P.D. Thirstle. 1995. "Public Utility Regulation and Bond Ratings." Public Choice 84: 119-36. Goldstein, M., and G. Woglom. 1991. "Market-Based Fiscal Discipline in Monetary Unions: Evidence from the U.S. Municipal Bond Market." IMF Working Paper ATP/91/89. September. Hajivassiliou, V. 1989. "Do the Secondary Markets Believe in Life after Debt?" Cowles Foundation for Research in Economics, Yale University. Discussion paper No. 911: 1-[40] (U.S.). May copies. Hakansson, N. 1999. "The Role of a Corporate Bond Market in an Economy-and in Avoiding Crises." IBERWTorking Paper,June. Institute of International Finance. 1995. "The Use and Availability of Political Risk insurance.' Wlashington, D.C. J.P. Morgan. 1996. "Risk Metrics: Technical Document." New York. Jaspersen, F. 1994. "Trends in Private Investment in Developing Countries." IFC Discussion Paper 28. Kaplow, L. 1989. "Government Relief for Risk Associated with Government Action." NBER W\torking Paper Series. Working paper no. 3006. Kaufmann D., A. Kraay, and P. Zoido-Lobat6n. 1997. "Aggregating Governance Indicators." W'orld Bank Policy Research Wlorking Paper 2195. Keefer, P., and D. Stasavage. 1998. "Wlhen Does Delegation Improve Credibility? Central Bank Independence and the Separation of Powers." Centre for the Study of African Economics. Working paper no. WAPS/98-18(U.K.): 1-22. August. 77 Governance Impact on Pivate Investment Knack, S., and P. Keefer. 1995. 'Institutions and Economic Performance: Cross-Country Tests Using Alternative Institutional Measures." Economics and Politics 7(3): 207-227. Kolbe, A.L., W'B. Tye, and S.C. Myers. 1993. "Regulatory Risk: Economic Principles and Applications to Natural Gas Pipelines and Other Industries." Boston: Kluwer. Levy, B., and P. Spiller. 1994. "The Institutional Foundations of Regulatory Commitment: A Comparative Analysis of Telecommunications Regulation."Journal ofLaw, Economics, and Organization 10(2): 201-246. Masuoka, T. 1990. "Asset and Liability Management in the Developing Countries. Modern Financial Techniques: a Primer." Policy, Research, and External Affairs WAorking Paper UPS 454. W'orld Bank. Washington, D.C. Merton, R. 1973. "An Inter-Temporal Capital Asset Pricing Model." Econometrica 41(5). Navarro, P. 1983. "How WAall Street Ranks the Public Utility Commissions." FinancialAnalysts Journal. November-December. North, D.C 1990. Institutions, Institutional Change, and Economic Performance. New York: Cambridge University Press. .1993. "Institutions and Credible Commitment."Journal of Institutional and Theoretical Economics 149: 11-23. North, D.C., and R.P. Thomas. 1973. "The Rise of the W'estern World: A New Economic History." Cambridge: Cambridge University Press. North, D.C., and B.R. W\eingast. 1989. "Constitutions and Commitment: The Evolution of Institutions Governing Public Choice in Seventeenth Century England." Journal of Economic History 49(4): 803-832. Olson, M. 1996. "Big Bills Left on the Sidewalk: W7hy Some Nations are Rich and Others Poor." Journal of Economic Perspectives 10 (2): 3-24. Poterba, J., and K Rueben. 1997. "State Fiscal Institutions and the U.S. Municipal Bond Market." NBER Working Paper Series No. 6237. October. Razavi, H. 1996. "Financing Energy Projects in Emerging Economies." PennWrell Publishing Company. Roger, N. 1999. Public Policy for Private Sector Note # 196. Private Sector Development Department. Washington, D.C.: The Wlorld Bank. Rose-Ackerman, S., and J. Rodden. 1999. "Contracting in Politically Risky Environments: International Business and the Reform of the State." Paper prepared for the Government of Italy and the World Bank Conference "Private Infrastructure for Development: Confronting Political and Regulatory Risks," Rome, September 8-10. Processed. Sachs,J., and A. Werner. 1995. "Economic Convergence and Economic Policies." National Bureau for Economic Research Wlorking Paper 5039. Scherer, F.M., and D. Ross. 1990. "Industrial Market Structure and Economic Performance." 3rd ed. Boston: Houghton Mifflin. Smith, WM 1997. CoveringPolitical and Regulatory Risks: Issues and OptionsforPrivate Infrastructure Arrangements. World Bank Latin American and Caribbean Studies. Wrashington, D.C.: The WAorld Bank. 78 References 1997. Utility Regulators-Roles and Responsibilities. The World Bank Group Private Sector in Infrastructure Series. Volume II. WraAhington, D.C.: The World Bank. Wleingast, B.R. 1993. "Constitutions as Governance Structures: the Political Foundations of Secure Markets." Journal of Institutional and Theoretical Economics 149: 286-311. -, 1995. "The Economic Role of Institutions: Market Preserving Federalism and Economic Development." Journal of Law, Economics and Organization 11: 1-31. Williamson, 0. 1991. "Economic Institutions: Spontaneous and Intentional Governance." Journal of Law, Economics and Organization 7: 159-187. . 1996. "The Mechanisms of Governance." New York: Oxford University Press. Wilson, T.C. 1997. "Calculating Risk Capital." In Carol Alexander, ed., The Handbook of Risk Management and Analysis. New York:John Wiley and Sons. WAorld Bank Latin American and Caribbean Studies. 1997. Dealing with Public Risk in Private Infrastructure. Washington, D.C.: The MWorld Bank. 79 Distributors of World Bank Group Publications Pices and credi terms vary hrom CZECH REPUBLIC INDIA Eulyoo Publishing Co., Ltd. PERU SWEDEN coiuntry tocounrory. Conlsultvour USIS, NIS Prodejna Allied Publishers Ltd. 46-1, Susonq-Doag Editorial Desarrollo SA Wennergren-W4iliams AB lcal distributor beHore placing an Haveikova 22 751 Mount Road Jongro-Gu Apartado 3824, Ica 24z OF. 106 P. 0. Box 1305 order. 130 00 Prague3 Madras- 600002 Seol ma1 S-171 25 Sona Tel: (420 2) 2423 1486 Tel: (91 44) 852-3938 Tel: (82 2) 734-3515 Tel: (51 14) 285380 Tel: (46 8) 705-97-50 ARGENTINA Fax: (420 2) 24231114 Fax: (91 44) 852-0649 Fax: (82 2) 732-9154 Fax: (51 14) 286628 Fax: (46 a) 27-00-71 World Publicabons SA URL: htp:l/www.nis.czt INDONESIA LEBANON PHIUPPINES E-mail: mail@ewwi.se Av Cordoba 1877 DENMARK Pt. Indira Umited Ubririe du Uban Intemational Booksource Center Inc. SWITZERLAND 1120 Ciudad de Buenos Aires SamfundsLiteratr Jalan Borobudur 20 P.O. Box 11-9232 1127-A Antipoo St. Barangay. Ubrairne Payot Service Institutionnel Tel. (54 11) 4815-8156 RosenoerssAlle 11 PO. Box 181 Beirut Venezuela C(tm)tes-de-Montbenon 30 Fax: (64 11) 4815-.16 DK-1970 Frederiksberg C Jakarta 10320 Tel: (961 9) 217 944 Makati City 1002 Lausanne E-mail: wpbooks@infovia.com.ar Tel (4535) 351942 Tel: (6221)390-4290 Fax: (961 9)217 434 Tel: (63 2) 896 6501, 6505. 6507 Tel: (41 21) 341-3229 AUSTRALiA, FIJI. PAPUA NEW Fax: (45 35) 357822 Fax: (62 21) 390-4209 E-mail: hsayegh@librairie-du- Fax: (63 2) 896 1741 Fax: (41 21) 341-3235 GUINEA, SOLOMON ISLANDS, URL: httpJ/ww.sl.cbs.dk IAN liban.rom.lb POLAND ADECO Van Diermen VANUATU, AND SAMOA ECUADOR Ketab Sara Co. Publishers URL: hop i/vw.librairie-du- tnteroTational Publishing Service EdRionsTechniques D.A. lnformafion Services Libri Mundi Khaied Eslamboli Aver.. 6th Street iiban.ctr.lb Ul. Piekna 31137 Ch. de Lacuee 41 648 Wrhlthtiorse Road Libreria Intemnacional Delafrooz Alley No. 0 MALAYSIA 00-677 Warzawa CH1807 Blonay Mltcham 3132 Victoria PO.Box17-01-3029 PO. Box 15746-733 Universty of Malaya Cooperative Tel:(482)628-6089 Tel (41,21) 943 2673 Tel: (61) 3 9210 77 J77 Juan Leon Mera 851 Tehran 15117 Bookshop. Limit Fax: (48 2) 621-7255 Fax: (41 21) 943 3605 Fax: (61)3 9210 778 Quito Tel: (9021)8717819:B716104 P.O. Box1127 E-mail: books%ips0ikp.atmcompil THAILND E-mal: stplervice0dadirectLcom.au Tel: (5932)6521-606: (6932)6544- Fau: 9821) 0712470 Jalan Parntai 8amu URL: Central Books Distribution URL: http:/twvvwdadirecLcom.au 185 E-mail: ketab-sara@neda.net.ir 59700 Kuala Lunmpr 3Stp6SlomRipcgaawpVipsleeport 3em Road AUSTRIA Fax: (693 2) 504-209 Kowkb Publishers Tel: (60 3) 756-50re PORTUGAL Bangok 10600 Gerold and Co. E-nmail. librimul0fibrimundi.connec p°.aX19s7ser5 Fa:(06 5-42 a,kk15 GVeoidb ari 26 E-mail: Ibbrmu2l@librnmundi.com.ec PO Bo 97-1 Fax: (60 3) 755-4424 Livraria Purtuqal Tel: (662) 2336930-9 Weihburnasse 26 E-mai:iibrmu2iibiouldicom.ec Tehrarn E-mail: unkoopitm.net.my Apartado268i, Rua DoCarm Fax (662) 237-8321 A-1011 ten CODEU Tel: (98 21) 258-3723 MEXICO a0 7-74 Tel: (43 1) 512-47-31-0 Ruiz de Castilla 763, Edir. Expocoior Fax: (98 21) 258-3723 INFOTEC 1200 Lisbon TRINIDAD h TOBAGO URL htp/wowtwFgeroId.co/atonflne Primoer iso, Of. #2 GoveLAN SuppliesAqenry Av, San FeRando No. 37 Tel: (1) 347-4982 Systematics Studies Ltd. Fax: 43 1512-7-3129 Pimerpiso Of.42 IELAN Col. Torielto Guerra Fax: (1) 347-0264 BANGLADESH Tel/Fax: (593 2) 507-353; 253-091 GOverment Sa uplies Agency 14050 Mexico, D.F. StL Augustino Sopping Center Micro (ndustdies Develt E-mail: )578608 ISAL MuSolesaeacS EastemT Main Read, St. Augustine isane Sa et TMi3 ciddie puatObetver Socie Yanan A) St RdTel: (52 5) 624-2080 Avduois 0 4EGYPT, ARAB REPUBLIC OF Dubrin 2 Fax: (52 5) 624-2022 ROLMNIA Treldad:paTobagoeue s House 6. Road 16 E-mail: infotec~rtna.net.mx Corrnpani De Librarri Bucuresti S.A. Tl 009606 Dhanxmondi PJAxea Al Abram llistribuoon Agency Tel: (353 1) 661-3111 RLhl:6an.mSt.pcairo2.seor3 Fax: (068) 646-9467 Al Galaa Street Fax: (353 1) 475-2670 UR~ht:It.o.xSr isain.2.sco E-mawil: tobeftrnitd8d.riet Dhaka 1209 Cairo Mundi-Prensa Mexioo S.A. do C.V Bucharest Tel: (8 s 02) 3 Tel: (2032) 68-6832 ISRAEL c9Ri P) 141 -CPALTel: (40 1) 313 964 UGANDA Fax. (300 2) 811108 FTx: (20 2) 678-6033 Yomot Interatuore Ltd. CanuretMemod 1C ona Fax: (40 1) 312 4000 Gustro Ltd. Fax: tZIL 2FIN D PBox 3PO ox 56065 Cuaurn@ s PO Box 9997, Madhvani Building BELGIUM The Middle East Observer 3 Yoanani Hasandiar Street 06500 Mexios, DOF. RUJSSIA FEDERATION Plot 16/4 JiOja Rd. mAv du Roi 202 4} ;((335S8 0a) 1,211444 Sherif Stmt Tel Awtv 61560 Tel: (52 5) 533-565 lsdatelstvo Ves Mi3r2 Jean De LahRiioy 4RA StE ala Cairo Tel: (972 3) 6285-397 Fax: (2 5) 514-6799 Eai Kachn Peulok TelSdnr6 41) 261 467 Tel: (322) 38-169 Tel: (20 2) 393-9732 Fax: (972 3) 6285-397 NEPAL Moscow 101 317 Fax: (6 61) 261 460 fax: (6i32) 74538-01 Fax: (2032) 393-9732 R.O.Y. nteraDional Everest Media Inteational Services T: (7600) 0177 59 Fax Fax: (632 ) 630-76084 GERANt Ltdtl Fotaa52E (7- 1096) 91792an 69L E-mil ga@witogsatda.omey FINLAND PO Box 12066 (P. Ltd ozirrarin0igiasniet.ru UNITED KINGDOM B-aZIL AGater ikp AviO 61 5d T 836443 Micrornfo Ltd. Publicacoes Tecnicas Inotemacini PAO.KBox 1r20Ab Tel: (92 64946 Kathraridu SINGAPORE; TAIWT 9, CHINA P.. Box 3, Omeg(a Park, Alton. URL: http:ll FIw renouh)onac 51 Box010 i Fax: (972 3) 640 946S Tel: (977 1) 416 026 YANMAN; BRUNEI Hampshie GU34 s A Ratdait Gomde2011 4410 k-melt: -verbg.de 3JAMAI9 Fax: (977 1) 224 431 Hemrisiere Publication Services Ham G5 J 29 5 04 Sa P aulo GmiDe 2 41JKaa - Road u64-03 England Tel: (6611)2691-7366 P Fa (358 0)121-4436 URL hU6pJwww.r2 y1noJiB NETHERLANDS 3l27 TO. Bx119N City T0el: (44 1420) 60684 Fx: (5511) 260-6964 E-mail: akailauSQ'StxckMann11 Palesnian AuthoritylMiddle East De LndeboonoInternationale Sing(ar 349316 Fax: (44 1420) 89889 E-mail: postmaster8Lt2 10tbr R-eaine information Services Publicae shb7.- Tel: ( 5) 741-5160 POE-mail: bankomicroinfo cook URL: http://wweruroLb FRANCE P.OB. 19502 Jerusalem P0O. Box1202, 74980 AE Haaksbegen Fox: (65) 742-93566 URL: http://www,mlcrintx.co uk CANADA Editions Es2a; DBJ Tl: (572k2) 8271219 Tel: (31 53) 574-0004 E-mail: ashrgatedoasaconnect.com The Stationery Oice Renoum Fx Ltd. 48. mue Gay Lussac Fax: (9722) 6271834 Fax: (31 53) 572-9296 SLOVENIA 61 Nine Elms Lar2 e 5369sCteRoadox 76006 Parrs ITALY, UBERIA E-mail: lindeboo doninendarSbscikpo n London SW 6389, Conangio Rload.4 Tel: (33-1) 65-42-73-00 Licosa Commissinda Sansoro SPA URL http:Owwxwor1donIine.n-i- G A PTel:(44 171) 873-8460 LOntario K1J ~Fax: (33-1) 43-29-91-67 Via 54uca Di Calaborina,a1 delbon DE-mai .o csta HaL2r w Tel: (613) 745-2666 W. nafSTA cesta L Fax 244 781913 Fax: (813) 745-7660 GERMANY Caselia Postale 562 UR65,p w 1r Qa i a t000 LMsatoadr C-meal: o0n V erlAg 60126 Fire9 Tt ( EBSCO Na Ltd. Tel: (38 61)13383 47: 132 12 30 o37ce.0 FRL er1d 285-2ren98uFaxoks.com P (5sore Fl 45 Tel:2: (306)641-45 Pv Mai B 3691 Fax: (3864611336030 VENEZUELA URTE Dp I erOIREHONGuKNo,kCHINcLmMA631 Fax: (39465) 6-7e Ma2i7 hr 99914 E-mail: repansetkjgvestnik.si Tecni-Ciencia Libros, NAfaTel: (40 226) 940020 E-mail: LicosatH ftbcc.it Auckland SOUTH AFRICA, BOTSWAN Centro Cuidad Comercial Tamanco CHIN B F Ni41 C2-2 Caracase etrl R.Bo 87 P o 03 B0 Ea ln China Finanicial & Economnic Fax: (49228) 217492 URL:http:wwwftbccitAicosa Tel: (6459) 524-0119 F2tide: Nel: (a2)9964 6036 3 Publishing House URBL. htIpdwww.uno-ver2ag.de JAMAICA Pbihr Fax: (649) 524-0067 E-m iversity Press Southern Fae:(50 2(9696S45036 CYPRaUF SingJie E-ma4: sal8s@acia .conh IaR: Publishers Lt. Africa Fax MALDIVES aeirJrg GHANA 298 Old Hiope Road, Kingston 6 Oasis Official Vasco Boulevard, Goodwood ZAMBIA Tel: (98 10) 6401-7365 EULBooks Services Tel: 976-927-2085 P0. Box 3627 P.O. Box 12110, N1 City 7463 University Bookshop, University o Faa: (86 10) 6401-7365 P.0. Box44 Fax: 876-977-0243 oeiltlgrton Cape Tovn Zaanbia China BDok Import Cente EoTUC E-mail: irpvftolisiuom Tel: (lam4)e499r151 Tp(27 21(585 4400 Great East Road Campus Chin Box imor Centr Accarutzrt uoaHz732,Pnb o-og aoeCl P.0. Boa 2825 Tel: 2321 778643 JAPAN Fax: (94 4) 499 1972 Fax: (27 21) 595 4430 P.O. 800 32379 :Fax 223 21 778088 Eastern Book Service E-mail: cai@sactrix.ne.rnz E-mail: oxfordeloup.coza Lusaka Be~~lng Fax: 223 21 779099 3-13 ~Honyo 3hoe,Bnyku URL: http://wwwoasis books,cx,.nzi For sebscription orders: Tel: (260 1) 252 676 Chinese Corporation for Promotion 3-13E NE0-cnhmome,lSusuptinBeric Fx:t20x-k5395 of Hlumanilties GREET ky"11XNGRAItraira usnpaSrie Fo(61236 520 Papasotirlou S,A. Te:IxU 3) )3818-086 ZIMBABWEres imte You Fang A4u Tong, 35, Sto a StrFa:(U 3) 38163-0864 Univeraity Prasx Limited P2O) 80041096 XuanNeiDa Jie, Three Crowns Buildina Jeri'cho CrJighannesburg2024 Academiad@cholloks. Be~~~~)rng ~~~196862 Athens Emi:odswt-ebs.c0jp Prieatee Mil Bag 5609 drnebr922 t Te:(91) 6729 Tel: (30 1) 394-1026 UJRL: lbadan Tel: (27 11)0880-1440 4 Conald Road. Graniteside Fax. (86610(660 72 494 Fax: (30 1) 394-8254 http://www.belkkoame.or,jp/-svt- Tel: (234 22) 41-1366 Fax; (27 11) 080-6248 80.O.0ox667 Fax. (86 10) 660 72 494 ebs Fax: (234 22) 41-2 ~~~~~~~~~~E-mail: isseris.co.za Harare COLOMBIA "AMT l a:(3 2 4-066 Tel: 263 4 765036 Infoenlace Lria. Caure Diffusixn KENYA SPAIN Carrera 65No.61-21 5, BRue Capois Africa Book Service (E,A.) Ltd. PAKISTAN Murdi-Prens Libros, S.A. Fax: 2634 761913 Apartado Aeren 34270 C.P. 257 Quaran House. Mfangarm Stret Mirza Book Agenc Caste8o 37 Niiaed t,D.C. Purl-au-Prince P.O. 800 46246 66. Shahrah-e-Quaid-e-Azam 28001 Madrid SaTael do7 ogota,79 Tel (509) 23 9260 Nairobi Lahore 546000 Tel: (34 91) 4 363700 Tel: (67 1) 260-2798 Faxi: (509) 23 4659 Tel: (254 2) 223 641 Tel: (82 42) 736 3661 Fax: (34 91) 6 763908 COTE D'IVOIRE HOBO KONG, CHINA,- MACAO Fax: (254 2) 330 272 Fax: (92 42) 576 3714 EMail: libreria@Pmundiprensa.es Center d'Editixn et de Diffuslon Asia 2000 Ltd. Lg R:ltpWw.udpes.or Africaines (CODA) Sales & Circulateon Department Lorta House yagloeTw Muns de-Prensa 391eon 04 B.P. 641 202 Seabird House Mezxn4n 1 6hw BaigaloeTw aMe dareCent.39 Abidjan 04 ~~~22-28 Wyndtham StreeL Central P.O. Box 66077 POBox 13033 08009 Barcelona349 Tel: (225)246610-246611 Hong8Kan,CIn Nairobi Fax: (225) 24 512 0 6 51 Tel: 8521)53h0-1a469 Tel: (254) 2-336983, 221426 Karachi-75350 Tel: (34 3)488-3402 Fax: (652) 2526-1107 Fmx: (254) 2-330854 5601654 Tel: (82 21) 4474807ax(3)4-65 CYPRUS E-mail: saiesraasia200O.com.hkt E-mail: Legacyeoform-net.com E-mail: ouppakeTheoffoce.net SRI LANKA, THE MALDIVES Center for Applied Research URL: htlp://wwwasia2x0O.com.hk KOREA, REPUBLIC OF Lake House Baolcshoip 6T 4ee~re Enoi HUNGARY Dayang Books Trading Co. Pak Book CrTpoaton 100, Sw Chotampalam Gardiner 6-igisteLEgm Euro Info Servioe Intemanatioal Division Aziz Chamibers 21. Queen's Read Mawatha P.O. Box 2006 Mari r uoaHaz 783-20, Panigba Bon-Doang, Lahore Colomnbo 2 Nicosia H-1 30 cdapest ~ Socho-ku Tel: (92 42) 636 3222: 636 0898 Tel: (94 1) 32105 Fax: (367 2) 66-20610 Tel: (36 1) 36060o 24, 35060o25 Seoul Fax: (02 42) 636 2328 Fmx:(94 1) 432104 Fax: (357 ) 66-2051Fax: (36 1) 350 00 32 Tel: (82 2) 636-0655 E-mail: pbciebrain.reet.pk E-mail: LHI-Lsri.lan6ta.net E-mail: euroinifor0mail.matamhu Fax: (82 2) 636-0026 E-mnail: searnapiocholliannret Recent World Bank Technical Papers (continued) No. 442 Bartlomiej Kaminski, The Role of Foreign Direct Investment and Trade Policy in Poland's Accession to the European Union No. 443 Luc Lecuit, John Elder, Christian Hurtado, Francois Rantrua, Kamal Siblini, and Maurizia Tovo, DeMIStifying MIS: Guidelinesfor Management Information Systems in Social Funds No. 444 Robert F. Townsend, Agricultural Incentives in Sub-Saharan Africa: Policy Challenges No. 445 Ian Hill, Forest Management in Nepal: Economics of Ecology No. 446 Gordon Hughes and Magda Lovei, Economic Reform and Environmental Performance in Transition Economies No. 447 R. Maria Saleth and Ariel Dinar, Evaluating Water Institutions and Water Sector Performance No. 449 Keith Oblitas and J. Raymond Peter in association with Gautam Pingle, Halla M. Qaddumi, and Jayantha Perera, Transferring Irrigation Management to Farmers in Andhra Pradesh, India No. 450 Andres Rigo Sureda and Waleed Haider Malik, eds., Judicial Challenges in the New Millennium: Proceedings of the Second Summit of the Ibero-American Supreme Courts No. 451 World Bank, Privatization of the Power and Natural Gas Industries in Hungary and Kazakhstan No. 452 Lev Freinkman, Daniel Treisman, and Stephen Titov, Subnational Budgeting in Russia: Preempting a Potential Crisis No. 453 Bartlomiej Kaminski and Michelle Riboud, Foreign Investment and Restructuring: The Evidence from Hungary No. 454 Gordon Hughes and Julia Bucknall, Poland: Complying with EU Environmental Legislature No. 455 Dale F. Gray, Assessment of Corporate Sector Value and Vulnerability: Links to Exchange Rate and Financial Crises No. 456 Salman M.A. Salman, ed., Groundwater: Legal and Policy Perspectives: Proceedings of a World Bank Seminar No. 457 Mary Canning, Peter Moock, and Timothy Heleniak, Reforming Education in the Regions of Russia No. 458 John Gray, Kazakhstan: A Review of Farm Restructuring No. 459 Zvi Lerman and Csaba Csaki, Ukraine: Review of Farm Restructuring Experiences No. 460 Gloria La Cava and Rafaella Y. Nanetti, Albania: Filling the Vulnerability Gap No. 461 Ayse Kudat, Stan Peabody, and Caglar Keyder, eds., Social Assessment and Agricultural Reform in Central Asia and Turkey No. 462 T. Rand, J. Haukohl, and U. Marxen, Municipal Solid Waste Incineration: Requirementsfor a Successful Project No. 463 Stephen Foster, John Chilton, Marcus Moench, Franklin Cardy, and Manuel Schiffler, Groundwater in Rural Development: Facing the Challenges of Supply and Resource Sustainability No. 465 Csaba Csaki and Zvi Lerman, eds., Structural Change in the Farming Sectors in Central and Eastern Europe: Lessonsfor EU Accession-Second World Bank/FAO Workshop, June 27-29, 1999 No. 466 Barbara Nunberg, Readyfor Europe: Public Administration Reform and European Union Accession in Central and Eastern Europe No. 467 Quentin T. Wodon with contributions from Robert Ayres, Matias Barenstein, Norman Hicks, Kihoon Lee, William Maloney, Pia Peeters, Corinne Siaens, and Shlomo Yitzhaki, Poverty and Policy in Latin America and the Caribbean No. 469 Laurian Unnevehr and Nancy Hirschhom, Food Safety Issues in the Developing World No. 470 Alberto Valdes, ed., Agricultural Support Policies in Transition Economies No. 471 Brian Pinto, Vladimir Drebentsov, and Alexander Morozov, Dismantling Russia's Nonpayments System: Creating Conditionsfor Growth No. 472 Jit B. S. Gill, A Diagnostic Frameworkfor Revenue Administration No. 473 Esen Ulgenerk and Leila Zlaoui, From Transition to Accession: Developing Stable and Competitive Financial Markets in Bulgaria No. 474 loannis N. Kessides, ed., Hungary: A Regulatory and Structural Review of Selected Infrastructure Sectors No. 475 Csaba Csaki, Zvi Lerman, and Sergey Sotnikov, Farm Sector Restructuring in Belarus: Progress and Constraints No. 481 Csaba Csaki, John Nash, Achim Fock, and Holger Kray, Food and Agriculture in Bulgaria: The Challenge of Preparingfor EU Accession THE WORLD BANK 1818 H Street, NAV. NNashington, D).C. 20433 lU.S.A. Iclcpllolne: 202-477-1234 :acsimile: 202-477-6391 'Iclcx: MCI 64145 WORLD)B3ANK WIA 248423 WORLDIIANK Internet: vw-vw.vworldbank.org E-mlail: hooks(a worldbank.org, ISBN 0-8213-4818-3