University of Southern California Dissertations and Theses

Essays on regulation of public utilities and the provision of public goods

(USC Thesis Other)

Essays on regulation of public utilities and the provision of public goods

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content ESSAYS ON REGULATION OF PUBLIC UTILITIES AND THE PROVISION OF PUBLIC GOODS by Sunita Surana A Dissertation Presented to the FACULTY OF THE GRADUATE SCHOOL UNIVERSITY OF SOUTHERN CALIFORNIA In Partial Fulfillment of the Requirements for the Degree DOCTOR OF PHILOSOPHY (ECONOMICS) August 2005 Copyright 2005 Sunita Surana Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. UMI Number: 3196900 INFORMATION TO USERS The quality of this reproduction is dependent upon the quality of the copy submitted. Broken or indistinct print, colored or poor quality illustrations and photographs, print bleed-through, substandard margins, and improper alignment can adversely affect reproduction. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if unauthorized copyright material had to be removed, a note will indicate the deletion. ® UMI UMI Microform 3196900 Copyright 2006 by ProQuest Information and Learning Company. All rights reserved. This microform edition is protected against unauthorized copying under Title 17, United States Code. ProQuest Information and Learning Company 300 North Zeeb Road P.O. Box 1346 Ann Arbor, Ml 48106-1346 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D ed ication To m y parents and Bharat Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Acknowledgem ents It gives me great pleasure to acknowledge the love, support, and encouragement of many people who have made this endeavor possible. First and foremost, I would like to thank my advisor, Professor Isabelle Perrigne, who has helped me in countless number of ways. I thank her for initiating and sustaining my interest in Empirical Industrial Organization. Much of the work in this dissertation has been accomplished under her guidance. I gratefully acknowl edge her support and encouragement. The second chapter of the thesis is based on joint work with her. I am greatly indebted to her for the numerous hours she has spent in reading my work with such exquisite attention, and preparing me for the job market. It is a debt I cannot adequately repay. When Professor Perrigne left for Pennsylvania State University, I continued my work under the guidance of Professor Guofu Tan. I most gratefully acknowledge his generosity and immense support during some crucial and difficult times. Without his support, it would not have been possible for me to complete this dissertation. I also take this opportunity to thank Professor Geert Ridder for always being there to answer all my questions. He offered valuable criticism and indispensable advice at various stages of my graduate studies. He has been a constant source of inspiration and it was a privilege to have him on my committee. iii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. I am enormously grateful to Professor Joel M. Guttman for his unflagging support and encouragement. Chapter 4 of this dissertation is based on joint work with him. It has been a great privilege to work with him and learn so much during the process. I would also like to thank Professor Juan Carrillo, the late Professor Jean-Jacques Laffont, and Professor Quang Vuong for serving on my qualifying examination com mittee. I shall ever remain grateful to Professor Laffont for first introducing me to the study of regulation and spending several hours discussing my research. I am also very grateful to Professor Peter Gordon for his very helpful suggestions and com ments and for serving as the outside member on both my qualifying examination and final defense committees. My very special thanks go to Professor Jeffrey Nugent. He has helped me in so many different ways that I have almost lost count. I thank him for believing in me when no one else did. He made a world of difference. I would also like to take this opportunity to extend a sense of profound gratitude to Professor Sudipto Dasgupta. His encouragement has left an indelible mark on me. For their support and help throughout my doctoral study at USC, I thank all the faculty members, staff and students at the department of Economics. It has been a privilege to be a part of this team. I especially thank Professor Caroline Betts, Professor Isabelle Brocas, and Professor Juan Carrillo for their active support in my job search. I would also like to thank Philippe Gagnepain and Marc Ivaldi for providing iv Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the data analyzed in Chapters 2 and 3 and Roper Center for the data analyzed in Chapter 4. I owe a long-standing debt of thanks to my grand parents and parents. I am deeply appreciative of their unwavering support, infinite love, and constant prayers. I thank my parents for the sacrifices they have made to see me fulfill my dream. Words cannot express the gratitude I feel for everything that they have done for me. I cannot thank my father enough for never letting me quit. Special thanks go to my sisters, Vinita and Tripti, and my brother, Vikram, for the many ways in which they have enriched my life. For unfailingly supporting me, I would also like to thank my parents-in-law, my brother-in-law, Deepak, and my sister-in-law, R.eshma. Above everyone else, I would like to thank my husband, Bharat. W ithout him, this endeavor would not have been possible. I thank him for those endless hours of discussions, for stopping his own work so that mine could be done, for treating all my worries like his very own, for his tireless effort in answering all my queries ... be it incentive theory, Mathematica coding, or preparing for presentations. I cannot justify in words what he has had to put up with. I thank him for making my life beautiful. This is for him. v Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table of Contents Dedication ii Acknowledgements iii List of Tables viii List of Figures x Abstract xi 1 Introduction 1 1.1 Introduction 1 1.2 Empirical Tests of Contract Theory 3 1.3 Recent Trends in Public Transportation 7 1.4 Regulation and Politics 8 1.5 Quality Matters 12 1.6 Voluntary Provision of Public Goods: An Economic Puzzle 14 2 Politics and Regulation: The Case of Public Transit 18 2.1 Introduction 18 2.2 Politics and Regulation: Some Evidence from Public Transit 21 2.2.1 Political Data 21 2.2.2 Public Transportation Data 30 2.3 Economic Model and the Estimation Method 44 2.3.1 Discussion of Various Approaches 45 2.3.2 The Model 49 2.3.3 Estimation Method 58 2.4 Empirical Results 69 2.5 Conclusion 77 3 Incentive Regulatory Contracts with Endogenous Quality Choice: An Empirical Analysis 79 3.1 Introduction 79 3.2 Overview of Industry, Measures of Quality and Data 86 vi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.2.1 French Urban Transportation System 86 3.2.2 Verifiable Vs Non verifiable Quality 89 3.2.3 The Data 91 3.3 The Model 98 3.3.1 The Framework 98 3.3.2 The Optimal Regulatory Allocation 102 3.4 Empirical Methodology 107 3.4.1 Empirical Specification 107 3.4.2 Assumptions on the Error Terms 114 3.4.3 Identification and Estimation 115 3.4.4 Results 122 3.5 Conclusion 127 4 Public Good Contributions as Signals: Reputation, Trust, and the Evolution of Reciprocal Preferences 130 4.1 Introduction 130 4.2 Assumptions 140 4.2.1 Market Transaction Trust Games 140 4.2.2 Voluntary Provision of a Public Good 145 4.3 Analysis of the Game: Exogenous Population Mixture 148 4.3.1 The Game without Public Goods 148 4.3.2 The Game with Public Goods 154 4.4 Endogenizing the Proportion of Altruists 162 4.5 Evidence from U.S. Survey Data on Trust 168 4.6 Concluding Remarks 175 References 178 Appendices 186 A Appendices for Chapter 3 186 A.l Appendix A .l 186 A.2 Appendix A.2 187 A.3 Appendix A.3 187 B Appendices for Chapter 4 195 B.l Appendix B.l 195 B.2 Appendix B.2 197 v ii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Tables 2.1 1983 and 1989 Election Results 24 2.2 Unemployment Rate and Election Results 27 2.3 Participation Rate and Election Results 29 2.4 Hedonic Price Model for Public Transportation 39 2.5 Demand Parameters 71 2.6 Cost and Effort Disutility Parameters 72 2.7 Parameters for the Weight Function 74 2.8 Summary Statistics for Consumer Surplus Weights 75 2.9 Cost of Public Funds and Efficiency Parameters 76 3.1 Summary Statistics 92 3.2 Summary Statistics for Cost per Rider and Busfare 93 3.3 Summery Statistics of Measures of Quality and Cost Efficiency 94 3.4 Regression Results of log(seatkilometers per unit surface area) and log(price) on a set of exogenous variables 97 3.5 Regression Results of log(s) on Second Order Polynomial of logarithms of Exogenous Variables 119 3.6 Estimates of Demand Parameters 122 3.7 Estimates of Parameters of Cost and the Disutility of Effort 124 viii Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.8 Estimates of Shadow Cost of Public Funds and the Parameters of the ‘Type’ Distribution 125 4.1 Summary Statistics for Dependent Variables 171 4.2 Summary Statistics for Explanatory Variables 171 4.3 Ordered Probit Regressions of Trust on Selected Variables 173 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. List of Figures 2.1 Price Densities for Left Wing and Right Wing Cities 33 2.2 Price Densities and the Change from Left to Right 33 2.3 Price Densities and the Change from Right to Left 34 2.4 Price Densities and Presence of Environmentalists 34 2.5 Price Densities of Left Wing Cities with or without 35 Environmentalists 2.6 Price Densities of Right Wing Cities with or without 35 Environmentalists 3.1 Trend of Key Economic Variables Over Time 95 3.2 Price versus Seatkilometers per unit Surfare Area 126 4.1 Market Transaction Trust Game 141 x Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Abstract This dissertation consists of four chapters. The first chapter provides a general introduction. The second chapter analyzes the impact of politics on regulatory con tracts in the public transportation industry. While adopting an incentive regulation framework, we take into account such political factors by considering a non benev olent regulator. We estimate the parameters of the demand and cost functions as well as the shadow cost of public funds, the density of firms’ types and the weight associated with consumer surplus as a function of political factors and financial constraints faced by cities. Empirical results, using data from French bus transit contracts, show that cities with leftist majorities and with environmentalists in their councils tend to put larger weights on consumer surplus leading them to lower the price for public transportation. As expected, the level of debt per capita restraints the cities in their redistributive policies. In the third chapter we study incentive regulatory contracts that set some aspect of product quality as well as price. We first extend the Laffont and Tirole (1990) model to incorporate stochastic demand and cost, find the optimal contract in the presence of asymmetric information between the regulator and the regulated firm, and suggest how it can be implemented. We then derive the econometric model and xi Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. estimate the consumer’s demand function, the firm’s quality adjusted cost function, the distribution of the firms’ type, and the shadow cost of public funds again using data from French urban transportation. The fourth chapter develops a theory of the evolution of preferences for reciprocity, trust, and the voluntary provision of public goods in a society composed exclusively of rational, Bayesian optimizers. We study the evolution of a community consisting of “opportunists,” who simply maximize material payoffs, and “reciprocators,” who prefer joint cooperation to exploiting their opponents. In equilibrium, opportunists contribute to the provision of the public good in order to maintain reputations for being reciprocators, so as to be trusted in their private market transactions. The need to provide a public good enhances the evolutionary stability of the reciprocator type. The model is tested using U.S. survey data on trust. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 1 Introduction 1.1 Introduction As defined by Stone (1982), regulation is “a state-imposed limitation on the discretion that may be exercised by individuals or organizations, which is supported by the threat of sanction.” Regulation of business activities has been practiced for many centuries and can be traced back to the Middle Ages when merchant, craft, and professional guilds dominated economic activity. Interest in the study of regulation has substantially increased over the last three decades. One of the most important driving forces of this interest is the wave of regulatory reforms that swept across the globe during this time period. Since the 1970s a great many regulatory policies have been “reexamined with an eye toward major reform and often complete deregulation” (Noll and Owen, 1983). Even though the United States and many other developed countries embarked on the process of deregulating their industries in the early 1970s, to this date the process is far from over. It is a widely debated and discussed topic among economists, policy makers, regulators and industry representatives. Proponents of regulation have offered various rationales, the most prominent among them being correction of 1 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. “market failure” and the advancement of “public interest.” Commonly cited sources of market failure are barriers to entry, externalities, and imperfect information which make it difficult for perfect competition to be sustained and hence the justification for regulation. In many instances, regulation is also used as a justification for facilitating the transition of industries from regulated environments to competition. Opponents are quick to add that deregulation improves productivity and efficiency, and more often than not leads to lower prices. The debate is far from over. The next two chapters of this dissertation focus on different issues in regulation of public transit services in the presence of information asymmetries between the regulator and the regulated firm. Since the early 1980s economists have applied the principal-agent theory to analyze problems of regulation.1 In the second chapter, we adapt Laffont and Tirole (1986) to take into account political factors influencing regulatory contracts by considering a non benevolent government. The third chapter extends Laffont and Tirole (1990) to incorporate stochastic demand and cost and studies the optimal contract when both price and quality are regulated. Using data on French public transportation industry and municipal elections, the second chapter highlights the role of politics in policy formation and the third chapter is devoted to the study of regulation of quality, again using data on French public transit. The fourth chapter explores the evolution of preferences for reciprocity, trust and the Tor some of the earliest works, see Baron and Myerson (1982), Sappington (1982, 1983), and Laffont and Tirole (1986). 2 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. voluntary provision of public goods. The rest of the chapter is organized as follows: the next section provides a brief overview of some of the recent developments in empirical contract theory. Section 1.3 traces the trends in public transportation. Section 1.4 discusses the influence of politics over public policy and summarizes the main findings of Chapter 2. Section 1.5 emphasizes the importance of modeling regulation of quality. The economic puzzle of voluntary contribution by rational agents and our explanation of the same are discussed in section 1.6. 1.2 Empirical tests of contract theory Since the early 1970s there has been a tremendous surge of interest in the theo retical analysis of optimal contracts. However, until late 1980s not much theoretical prediction had been put to empirical test. Economists have since then undertaken a number of very interesting studies that test contract theory predictions in diverse fields such as sharecropping, insurance markets and managerial pay. Some of the noteworthy contributions in this field are Ackerberg and Botticini (1999), Chiap- pori and Salanie (1997, 2000), Cawley and Philipson (1999), to name a few. While studying the link between the form of contracts and observed behavior of agents, it is crucial to consider two factors that could create selection biases on the parameters of interest. First, contracts could induce the corresponding behavior through their 3 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. underlying incentive structure (moral hazard), and second, differences in observed behavior could be explained due to some unobserved heterogeneity across agents (adverse selection).2 One way commonly used to address selection problems is to ensure that the allocation of people to contracts is exogenous. Ausubel (1999) uses data on responses to large-scale randomized trials in preapproved solicitations by bank credit cards to empirically ‘document the existence of adverse selection in credit markets as a real-world phenomenon.’ His findings are, (i) on an average, respondents have shorter credit histories and inferior credit ratings, (ii) comparing customers from different offers, respondents to inferior offers are found to have worse observable credit-risks than those who respond to superior offers, and (iii) even after controlling for observable characteristics of respondents to inferior offers, the respondents are found more likely to default. These findings are in support of the presence of adverse selection in the credit card market. A very widely cited random experiment is the Rand Health Insurance Experi ment whereby participating families were randomly assigned to one of 14 different insurance plans, that involved different coinsurance rates and upper limits on an nual out-of-pocket expenses borne by them. One of the main findings of the study was that the largest decrease in the use of outpatient services was observed between families with free plans and those with plans involving a 25% copayment rate. See 2See Chiappori and Salanie (2003) for a thorough discussion of these issues. 4 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chiappori and Salanie (2003) for a comprehensive summary of recent work testing contract theory. Structural empirical estimation of models of regulation has also not kept pace with theoretical findings. One reason for this lag is the difficulty in estimating such models, given that the dimensions of private information are not observed by either the regulator or by the econometrician. The informational asymmetry between the regulator and the firm may be of two forms: hidden action that the regulated firm may take, for instance cost reducing effort, and hidden information that the firm may have on its inherent characteristic or ‘type.’ In the presence of informational asymmetries, the design of optimal regulatory contracts entails providing firms with incentives to truthfully reveal their ‘type’ through the appropriate choice of con tracts. While this is achieved through price distortion (beyond the Ramsey price) in the Baron and Myerson (1982) model, it is achieved through effort distortion in the Laffont and Tirole (1986) model. Wolak (1994) was among the first to estimate the parameters of a regulated firm’s production function while explicitly modeling the role of informational asymmetry in the regulatory process. He estimates the pa rameters of the production function for urban water delivery in California under two behavioral models of regulator-utility interaction, one with symmetric information and the other with asymmetric information between the two contracting parties. The econometrician, however, cannot observe the private information parameter in either of the two cases. Using a non nested hypothesis testing procedure he con- 5 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. eludes that the model with asymmetric information provides a better description of the observed levels of costs and output. His work differs from previous studies in that he explicitly models the impact of the utility’s private information on the cost function (model closer in spirit to Baron and Myerson (1982)), and among other things recovers an estimate of the distribution of this private information. Unlike Wolak, Gagnepain and Ivaldi (2002) assume that the actual regulatory contracts in place are not optimal and restrict the contracts to being either fixed price or cost plus type of contracts. Using a principal-agent framework, they analyze such regulatory contracts in the French public transportation industry. The authors also perform a comparison of current and optimal regulatory outcomes by simulating optimal contracts based on their parameter estimates. The choice of optimal regu latory contracts is endogenized by Perrigne (2002). Much of the work in the next two chapters of this dissertation builds upon the framework developed in Perrigne (2002). From the economic model (which models the regulator and the regulated firm in a principal-agent framework in the spirit of Laffont and Tirole (1986)), she derives the econometric model making very few assumptions on the errors of the model. Her model not only accounts for the unobserved ‘type’ of the firm but also the hidden effort exerted by the manager. Using the identifying assumptions, the model is estimated using generalized method of moments and method of simulated moments to recover the parameters of the model. The methodology adopted in the next two chapters very closely follows this approach. 6 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.3 Recent Trends in Public Transportation Transit rider ship is on the decline in most of the developed nations. In the US, according to the census statistic (2000), only 5 percent of the nation’s 128.6 mil lion workers use public transportation to get to work. Decline in transit ridership between 1980 and 1993 has also been reported for UK, Italy, Norway and West Germany; whereas ridership has risen in some other countries like France, Canada and Austria (See Pucher, 1995). Researchers have attributed rising incomes and suburbanization of jobs as major reasons for the long-term decline in public trans portation ridership in most developed nations. In an attem pt to arrest the declining use of public transit, policy makers throughout North America, Western Europe and Australia have subsidized the use of public transportation. Public transit, thus, is used as an instrument to achieve redistribution of wealth through subsidization of the service. Through regulation of this sector the government can also achieve balanced urban development by ensuring a well coordinated network of mass trans portation. Another justification often offered for government regulation of public transit is to ensure that the elderly, the physically disabled, and the low income citizen are provided sufficient urban mobility. 7 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1.4 Regulation and Politics The impact of politics on policy outcomes has long been of interest to researchers. As opposed to the “public interest” theories of regulation (that assume regulation to be established largely in response to public interest related objectives), many pro ponents of “private interest” theories have modeled regulators as selfish politicians who seek survival, i.e., reelection, by choosing regulating policies that serve special interest groups who may offer them their support during elections. Noteworthy among these are the “capture theory” of Bernstein (1955) and Huntington (1952), the “cartel by design” hypothesis of Kolko (1965), Stigler’s (1971) characterization of regulation as an economic good with demanders and suppliers, and theories in which political agents influence policy decisions in favor of their constituents (Mc- Cubbins, Noll and Weingast, 1989) to name a few. Becker (1983) constructs a model of political competition among pressure groups. According to him, competi tion among these groups determines the equilibrium structure of taxes, subsidies and other political favors that are eventually realized. Becker shows that the political equilibrium depends on the efficiency of each group in exerting pressure, the number of members in the group, and the deadweight cost of taxes and subsidies. He argues that groups compete for political influence by spending time, money and resources to produce political pressure in order to achieve their respective goals. Grossman and Helpman (1996) analyze a model of competition between two political parties 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. that have fixed positions on some issues (party ideology), but vary their positions on others (pliable policies) to attract votes, campaign contributions and support from interest groups. To attract votes, parties announce policy platforms and engage in political advertising. An equilibrium consists of a pair of platforms and a set of con tribution schedules such that no group or party has any incentive to deviate given the anticipated behavior of others. The equilibrium platforms and contributions determine the election outcome which in turn determines which platforms will be pursued. They show that the party that caters more to the special interest groups is expected to win a majority of seats. Noll (1971) discusses two explanations of regulatory failures: the first view is that regulatory failures are consequences of ‘errors by incompetence’; the second is that regulators pursue objectives that are not always in public interest, i.e., their mistakes are ‘errors by design’. More recently, considerable evidence has been found that highlights the influence of the public and consumer interests on regulatory pol icy. Various empirical studies show that special interest groups indeed wield consid erable influence over public policy. They pursue their quest for political advantage by a number of different means: gather media attention, generate and contribute to public debates that support their positions, mobilize or cajole impressionable voters, contribute to political parties and to individual candidates’ campaigns. Cropper, Evans, Berardi, Dulca-Suares and Portney (1992) show the importance of interven tion by special interest groups in Environmental Protection Agency’s decision to 9 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cancel or continue the registrations of cancer-causing pesticides. They show that support and pressure by environmental advocacy groups increases the probability of cancellations, whereas active involvement by pesticide manufacturers and farm ers who use them substantially lowers the probability of cancellation. In a similar vein, Ando (1999) evaluates the ability of interest groups to influence the timing of decisions to add specific species to the endangered species list by exerting pressure on the Fish and Wildlife Service. She finds that public support or opposition can significantly hasten or slow the progress of candidate species through the various stages of the listing process. Goldberg and Maggi (1999) conduct a structural es timation of the Grossman-Helpman (1994) model that addresses political economy of trade protection and as predicted by the model, the authors find that, among other factors, trade protection is also explained by whether or not the industry is politically organized. In the words of Winston and Crandall (1994), “To be sure, regulatory policy has often been at odds with consumer welfare. The reason for that, we argue, is that politicians draw on a portfolio of policies both to get reelected and to pursue their own ideological interests.” In similar spirit, to empirically investigate the impact of politics on policy decisions, in Chapter 2 we consider a non benevolent regulator, who driven by his ideology and office-seeking behavior maximizes a weighted sum of consumers and producers surplus.3 Chapter 2 is based on joint work with Professor 3In the words of Winston (2000), “It is no secret that policymakers-appropriately-respond more 10 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Isabelle Perrigne. A preliminary analysis of data from the public transportation in dustry in France suggests that design of regulatory contracts is influenced by factors such as the political party in power, the degree of presence of environmentalists in the city council, and the financial constraints faced by the city. In light of these findings we consider the weight to be a function of whether the party in majority is left or right wing, the proportion of environmentalists in the city council, and the financial constraints of the city as measured by the debt per capita. We use data on regulatory contracts in the French public transportation industry for 57 cities from 1985 to 1993 to conduct our empirical investigations. The data is compiled from the responses to an annual survey conducted by the Centre d’Etudes sur les Reseaux, les Transports, l’Urbanisme et les Constructions Publiques (CERTU), a national institution dedicated to the study of urban transports and networks. In addition, we collected data on French municipal elections for the years 1983 and 1989 to complete the data required for our analysis. Using a structural model we find that cities with left wing majorities and with environmentalists in the councils tend to put a higher weight on the consumer surplus, and thereby charge lower price for public transportation. We also find that cities with high debts per capita tend to charge higher bus fares given that they have limited resources for subsidization of public transit. The sizeable impact of politics on policy formation is an institutional reality and an ever growing field of research is studying this phenomenon. Wood and to political forces than to market forces.” 11 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Waterman (1991) have noted: “evidence for active political control is so strong that controversy should now end over whether political control occurs. Future research should turn toward exploring the determinants of political control.” Our paper is an attem pt in this direction. 1.5 Quality M atters Structural empirical models have generally been concerned with price regulation alone. However, regulation of utilities in most instances is not limited to controlling prices alone. The goal of the regulator is to ensure that consumers get reasonable services at affordable prices and hence they have adopted an integrated approach toward price and quality regulation. In the third chapter of this dissertation we study regulatory contracts in the public transit industry in France when both price and features of quality are set by the regulator. We first extend the Laffont and Tirole (1990) model to incorporate stochastic demand and cost, find the optimal contract in the presence of asymmetric information between the regulator and the regulated firm, and suggest how it can be implemented. We then, following Perrigne (2002), derive the econometric model and estimate the consumer’s demand function, the firm’s quality adjusted cost function, the distribution of the firms’ type, and the shadow cost of public funds. Many researchers have made use of the standard microeconomic duality theory 12 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in estimating the cost function in the urban transportation industry. Some papers that one might peruse with benefit include Viton (1981), Berechman and Guiliano (1984), and Obeng (1984, 1994). We consider a quality adjusted cost function when the regulated firm has more information about its efficiency and the effort that it exerts than the regulator. We use seatkilometers (obtained by multiplying the num ber of seats available by the distance over which the respective seats travel) per unit surface area as a composite measure of quality. This measure accounts for three crucial dimensions of quality: first, the passenger capacity offered which reflects the comfort of the service enjoyed by passengers; second, the network coverage across the city which reflects the convenience of service; and third, the frequency at which bus service is made available to its users which reflects the availability of public transit service. Regulating seatkilometers per unit surface area is equivalent to reg ulating the number of seatkilometers covered since the area of the city is constant. As cost incurred in the provision of service is determined by seatkilometers and not seatkilometers per unit area, we include seatkilometers as the measure of supply related quality in cost estimation. This has interesting implications for the estima tion of the cost function. In particular, we find significantly lower cost elasticity of demand-related output than that found in Perrigne (2002) and Perrigne and Surana (2004). The reason is that when the cost function is not adjusted for quality, any increase in cost due to higher demand that is attributable to greater provision of service is not accounted for in the cost function. This leads to overestimation of the 13 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. cost elasticity with respect to demand. One caveat common to the analyses done in the next two chapters is that we have restricted our model to being static in nature. Two considerations favor our econometric approach: (i) the model used is explicitly static, and (ii) the data set used is primarily cross sectional in nature with many more cities than years. The reason for restricting attention to a static model is that, as shown by Laffont and Tirole (1988), in a dynamic setting the extremely complex nature of equilibria makes it extremely difficult to characterize incentive schemes. One difficulty is the so-called “ratchet effect,” i.e., if a firm reveals its type truthfully and produces at a low cost then the regulator, inferring that the firm is efficient, may offer a very demanding contract in subsequent periods, thereby thwarting any incentive an efficient firm may have to reveal its type truthfully. Stated in a nutshell, the dynamics of incentive regulation can become very complicated and has not been considered in our work. 1.6 Voluntary Provision of Public Goods: An Eco nomic Puzzle According to the proponents of the free-rider theory (Olson (1965), Olson and Zeckhauser (1966)), the “logic of collective action” would lead to the non provision of public goods on a voluntary basis due to the nonexcludability nature of public 14 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. goods. In the words of van de Kragt, Orbell and Dawes (1983), “rational and selfish individuals will recognize the opportunity to ‘free ride’ on the contributions of others which public goods offer and, secure in the knowledge that they can share in the good once it is provided, will withhold or severely curtail their own contributions.” However, as noted by Andreoni (1995), “millions of people give to public goods ... and they generally contribute sizable sums.” A long series of public good experi ments also indicate that real world agents typically contribute to the provision of non excludable public goods. Some of the first laboratory experiments illustrating that people make voluntary contributions to public goods were conducted by Dawes, McTavish, and Shaklee (1977). Providing experimental subjects with an opportu nity to make contributions toward public goods (in a setting in which contribution was a dominated strategy), the authors showed that voluntary contributions were indeed significant. They further showed that with added ability of communication among participants doubled the contribution rates to more than 70 percent. Since then a plethora of experiments have reinforced the same result. See Guth and Tietz (1990) and Ledyard (1995). The fourth chapter of this dissertation, entitled “Public Good Contributions as Signals: Reputation, Trusts, and the Evolution of Reciprocal Preferences,” offers an explanation for the existence of such “reciprocator” types. More precisely, we study the evolution of preferences for reciprocity, trust and the voluntary provision of public goods in a community consisting of two types of agents: ‘opportunists’ 15 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. who simply maximize their material payoffs, and ‘reciprocators’ who prefer joint cooperation to exploiting their opponents. Chapter 4 is based on joint work with Professor Guttman. Though the primary objective of the paper is to explain the voluntary provision of public goods, we introduce bilateral private transactions in order to model agents’ private incentive to contribute to public goods. Contributions toward public goods serve as signals of the contributor’s trustworthiness. By free riding agents may lose reputations for trustworthiness that are valuable in private good transactions.4 We show that these signals, particularly when they are costly, can effectively separate the more trustworthy from the less trustworthy agents. This may induce the “opportunistic” types to mimic their behavior in order to preserve reputations for trustworthiness. In each stage of his or her career an agent decides on two things: (i) whether to honor trust in a bilateral market transaction and (ii) whether to contribute to the provision of a non excludable public good. While individual interactions are one shot, it is assumed that agents know the history of play of all others in the community. If an agent, however, moves to another community, he starts afresh. We assume that a random exogenous variable determines whether or not an agent moves. We further assume that an evolutionary mechanism selects for the type that is relatively successful and show that when the cost of contributing is sufficiently large, 4In the words of Miller (1997), “it is possible to assume that people are concerned (among other things) with acceptance and status as defined by a group of relevant others. Social goals do not make people any less goal-oriented; they do not necessarily make people any less selfish.” 16 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reciprocators survive in an evolutionary equilibrium. We also find that trust is higher in communities where population mobility is low. We test these two hypotheses using data from three sources. The 2000 Social Capital Benchmark Survey conducted in 41 U.S. communities by the Saguaro Seminar at the John F. Kennedy School of Government at Harvard University constitutes the basic data set. We use U.S. Census data to obtain the mobility index which is defined as the proportion of people who have lived in the same house for the last five years as a measure of (low) mobility. We argue that contribution toward public goods tends to be costly when the environment is harsh, and hence we use the average minimum temperature in the month of January as a measure of how costly it is to contribute. This data is collected from www.weatherbase.com. The empirical results support the theoretical predictions. We find trust to be higher in places where the environment is harsh and we also find trust to be relatively higher in stable communities where the population mobility is low. 17 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 2 Politics and Regulation: The Case of Public Transit 2.1 Introduction It is almost a truism that politics influences the economic policies undertaken by public enterprises.5 Interdependencies between political agents and policy outcomes is very common. Examples are not hard to come by. Shleifer and Vishny (1994) while formulating a model of bargaining between politicians and managers discuss some such examples. In this chapter we look at regulatory contracts in the French transportation industry. The question we seek to answer is whether politicians affect regulatory contracts. More specifically, does the political composition of city council affect the prices (busfare) that consumers are charged? One of the most important driving forces of this interest is the gap between developments in theoretical litera ture and empirical testing of such models. Noteworthy among theoretical works are Stigler (1971), Becker (1983), Laffont (1996), Laffont and Tirole (1991), to name a few. But structural empirical research has not kept pace with theoretical findings. Our study is an attem pt to somewhat bridge this gap. 5Laffont and Tirole (1991) aptly remark, “A major task of economics and political science is to explain the pattern of government intervention in industries.” 18 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The impact of politics on policy issues has been widely debated and discussed. Researchers in varied fields like economics, political science, and sociology have con tributed to this debate. Stigler (1971), one of the pioneers in this field, showed that the regulatory policies could be captured by small business industries. Peltzman (1976) and Becker (1983, 1985) built upon this work. But these studies did not take into account the informational asymmetries between the regulator and the regulated industry. This lacuna was filled by the work of Laffont and Tirole (1991) and Laffont (1996). Laffont and Tirole (1991) study the politics of government decision-making in a three tiered model (the politician being the principal, the regulatory agency be ing the supervisor, and the firm being the agent). Among various other findings the authors show that ‘the organizational response to the possibility of agency politics is to reduce the stakes that interest groups have in regulatory policies.’ In another study Laffont (1996) shows how the cost reimbursement rules and the pricing rules in a regulatory environment depend on the political majority and whether the reg ulated firm is public or private. Helpman and Grossman (1996) highlight the role of special interest groups (SIGs) while studying competition between political parties for seats in legislature. Apart from organizing public debates and gathering media attention, SIGs make campaign contributions to parties that reflect their opinions in public policy. Grossman and Helpman show that each party is induced to behave as if it were maximizing a weighted sum of the aggregate welfares of the different groups. 19 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In recent years a number of empirical studies, as discussed in Chapter 1, have reinforced the hypothesis that politics does influence public policy. Using data on French public transportation industry for the years 1985-1993 we find that public policy tends to move in response to election outcomes. Our results show that left-right partisan divide matters in policy outcome of the French transportation system. More specifically, we find busfare to be significantly higher with a right majority vis-a-vis a left majority. The period under study witnessed the steady rise and assimilation of environmentalists into French politics. Their aim is to ensure that policies and investments undertaken in the transportation sector reflect conservation of energy, and the protection of environmental quality. With their ability to focus media attention on the impact of transportation policy and reform on the environment, and to generate and contribute to the public debate on transportation issues, the environmentalists too have had a significant impact on public policy. We find busfare to be lower in those cities which have “Green” representation in their city councils. It is not surprising, however, that the impact of left-right divide or the presence of environmentalists in the city council is subject to financial constraints of the city. We find that, ceteris paribus, cities with higher debt per capita are likely to have higher busfare and the opposite is true for cities with lower debt per capita. The remainder of the chapter is organized as follows. In the next section we describe the regulatory contracts in the French transportation industry and provide 20 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. detailed information on the data. Section 2.3 describes the model. In the same section we also discuss the identification and the estimation of the model. The fourth section provides the results and the fifth section concludes. 2.2 Politics and Regulation: Some Evidence from Public Transit This section presents data from two different sources. A first data set contains detailed information on regulatory contracts for 57 French cities from 1985 to 1993. To assess the role played by politics on these regulatory contracts, we collected data on city council elections and the composition of city councils. 2.2.1 Political D ata The 1985-1993 period has experienced one election at the city level, which took place in March 1989. Since the data on regulatory contracts covers years before 1989, we also collected data on the previous election, which took place in March 1983. City councils in France are renewed every six years through democratic elections. The size of the city council is proportional to the population in every city. In our data, the number of its members varies from 37 to 101 according to the population 21 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. size. Several parties enter into the electoral competition, each proposing a list of candidates. The election takes place in one or two rounds. The final allocation of seats in the city council among the various lists in competition corresponds to a mix of majority and proportional rules. If a list of candidates obtains the absolute majority in the first round, there is no second round of election for the city in question. The winning list then obtains 50% of the seats, while the remaining 50% of the seats are allocated through a proportional rule among all the lists in competition. This allocation rule gives a comfortable majority to the winner of the election.5 The new city council elects the new mayor of the city, who is in general at the very top of the winning list. Given the strong majority of the winning list, there is no conflict for the election of the new mayor. For the election held in 1983, 37 cities out of the 57 cities in our sample had a single round of election, while there were 27 in the 1989 election. When none of the lists obtains an absolute majority in the first round, a second round is organized a week later. The second round is organized as follows: only the lists, which obtain a minimum of 10% of votes in the first round, are allowed to participate in the second round. It is, however, possible for the lists which obtain between 5% and 10% of 5Let us consider a simple example with three lists of candidates in competition. The winning list (party A) obtains 60% of the expressed votes in the first round , while party B and party C each obtain 20%. According to the allocation rule adopted in France for 30 seats to be allocated, party A will have 50% of the 30 seats plus an additional 60% of the remaining 15 seats, which makes a total of 24 (= 15 + 9). Both parties B and C will each obtain 20% of the remaining 15 seats, i.e., 3 seats for party B and 3 seats for party C. Party A will govern the city with 80% of the seats of the City Council. Despite having each obtained 20% of the votes, party B and party C will obtain only 10% of the seats in the city council. 22 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. votes in the first round to merge with a list, which obtains more than 10% in the first round. This rule favors various political alliances between the two rounds of election giving hope to parties with a low popularity to have at least a representative in the city council. The final allocation of seats follows the rule that we have described.6 When combining both elections, we find more than 70 different list names, which is quite large, many of them participating only in a small number of cities. A large majority of lists are, however, affiliated to a political party. We can roughly consider seven political tendencies in France from the left to the right wing. Parties associated with a communist ideology are considered as extreme left, while parties with a socialist ideology are considered as leftist parties. Between the left and the right wing, there are several parties in the ‘center.’ Each of these political parties call themselves as either ‘left centre’ or ‘right centre’. Since there are no major differences between center left and left and between center right and right, and given the strong likelihood that these parties will form alliances with left and right parties respectively, we have labeled these parties as either left or right. They usually obtain a low proportion of votes. Parties with a republican or democrat ideology are considered as the right wing, while parties based on a nationalist ideology are labeled as extreme right.7 Lastly, the past 25 years have experienced the emergence of new 6Let us consider a simple example. In the first round of the election, several lists are in competition. Only 4 have obtained at least 10% of the votes. The results of the second round give 40% to party A, while parties B, C and D obtain 20% each. For 30 seats to be allocated, party A will obtain 21 seats (= 15 + 6) or 70% of the council while other parties will obtain 3 seats each or 10% of the council. 7The diversity of political parties in France is larger than in the US. It may look surprising to 23 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. parties in Prance with environmental interests. They tend to act as special interest groups lobbying against polluting industries and car pollution. We consider them as a special interest group as their arguments refer less to ideology when compared with other parties. For instance, they make a great use of scientific studies on pollution and planet warming and their consequences on health. They favor free public transportation to induce more car drivers to switch to a more environmentally friendly mode of transportation.8 In some cities, some of their representatives have signed agreements with the left wing after the first round. After their relative success in the first round of the 1989 election, they have maintained their lists at the second round.9 The following table provides the results at both elections for extreme left, left, right, extreme right and environmentalist parties. These figures provide the average percentage of votes at the final round of the election as well as the percentage of seats obtained in the city council.1 0 Table 2.1: 1983 and 1989 Election Results Party votes 1983 city council votes 1989 city council Extreme Left 14.33 10.21 10.41 8.71 Left 30.32 36.37 39.48 35.86 Right 53.91 52.50 45.11 45.13 Extreme Right 0.81 0.11 2.63 5.15 Environmentalist 0.36 0.60 2.23 4.51 some readers that both republican and democrat are considered as the right wing. Various right wing parties in Prance use either the term republican or democrat in their affiliation. 8In the same spirit, they favor the development of bike paths and pedestrian areas in cities. 9Data provide detailed information on the political affiliation of elected council members. 10The sum of the figures in Table 1 does not sum up to 100% because of some independent lists. 24 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Between 1983 and 1989, we observe some changes for the leftist and the rightist parties. The percentage of seats obtained on average by the left wing was stable though the 1989 election was more favorable in terms of proportion of votes. The right wing experienced a significant decrease in popularity both in terms of average proportion of votes obtained and representation in city councils. In 1983, the right wing won in 32 cities, while the left wing and the extremist left won in 20 and 5 cities, respectively. In 1989, the right wing won in 26 cities, while the left wing and the extreme left won in 28 and 3 cities, respectively. Fourteen cities experienced a major switch in their majority, 4 cities switched from a leftist to a rightist majority, while 10 cities switched from a rightist to a leftist majority. The few cities with extreme left majorities are mainly located in the north of France, where traditional industries such as steel and mining have declined in the seventies. Consequently, these cities are characterized by a high unemployment rate and experience a difficult economic transition. Table 2.1 shows that the representation of extreme left and extreme right parties are not negligible, the former has had a decline in popularity while the latter has experienced an increase in popularity between 1983 and 1989. In 1983, 13 cities in our sample had some council members affiliated with the extreme right party and 56 cities had some extreme left members. In 1989, there were 29 cities with extreme right council members and 45 cities with extreme left council members. The most striking change between these two elections, however, is the emergence 25 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of the environmentalists.1 1 In 1983, only 8 cities in our sample had at least one environmentalist in their respective city councils, while the environmentalists had representatives in 27 cities after the 1989 election. We note that the environmental ists have usually maintained their position while gaining seats in other cities.1 2 We also note that environmentalists are more likely to obtain seats in cities with leftist majorities than in cities with rightist majorities. In 1989, only 7 cities out of the 26 rightist cities had at least one environmentalist in their city councils while there were 19 out of the 28 leftist cities. In 1983, the results were quite different with 6 cities out of the 32 rightist cities and 2 cities out of the 20 leftist cities. Generally, their representation is quite small as they obtain between 1 to 5 seats but their influence has been significant in policy decisions, especially in terms of public transportation as shown in the next subsection. To conclude this section, we discuss the two actors in these elections: the politi cians and the voters, who have different objectives in these elections. First, we discuss the characteristics of the electors of the various political parties. Given the 1 1 We consider it as a major change for the following reasons. It was the first election in which the environmentalists became important actors in French politics and this trend has been confirmed by the following elections. Moreover, it is not unusual in France that election results fluctuate so much for parties within a short period of time. Such a fluctuation happened a few years ago between the presidential election and the congress election within a short period of time. The extreme rightist candidate obtained about 20% of the votes at the first round of the presidential ele c tio n , w h ile v e ry few re p re s e n ta tiv e s o f h is p a r ty w ere elected a t th e co n g ress d u e to a n o b v io u s lack of popular support. 1 2 An interesting question is to understand whether this vote has been motivated by high levels of pollution. A first look at cities with environmentalists in their city councils suggests that there is no particular pattern between pollution or the presence of polluting industries and the environmentalist vote. We, however, do not have data on pollution to assess empirically such a statement. 26 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. well known split between the left and right wings, the left being in favor of redis tributive policies while the right in favor of more liberal policies, we have computed the correlation coefficient between the percentage of votes and the unemployment rate for every city. The unemployment rate in our sample varies from 7% to 23% with an average value of 15%. The results are given in the table below when pooling the data from both elections. Table 2.2: Unemployment Rate and Election Results Party Correlation Coefficient Extreme Left 0.1926 Left -0.1051 Right -0.1088 Extreme Right 0.0950 Green -0.1383 Not surprisingly, there is a strong positive correlation between the unemployment rate and the extreme leftist vote. This fact is observed in north and east of France, an area which has experienced a decline of traditional industries. In these industries, a big proportion of the labor force is unionized and unions in France are very close to extreme leftist parties. Regarding the extreme right vote, it seems that there is a correlation with immigration such as in South of France, though in some cases the extreme right wing has a significant presence, in areas such as Brittany, where the immigration has always been low. The vote for extremist parties are usually based on strong ideological factors. The correlation is negative for the other parties. Regarding other population characteristics, we do not have more data to draw some 27 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. conclusions and have to rely on other studies. It seems that young generations are more likely to vote for the left wing, while older generations are more conservative with a rightist vote. Urban population is more likely to vote for the left wing, while rural population is more likely to vote for the right wing. Individuals’ wealth and social status is usually correlated with political choice such as low revenue and working class voters favoring the left wing, while high revenue and professional class voters favoring the right wing.1 3 For each city at both elections, data provide the total number of registered voters and the number of actual voters. We can use this information to compute the participation rate at these elections. In 1983, the participation rate was on average 72.55%, with values ranging from 60% to 81%, while in 1989 it decreased to an average of 64.83% with values ranging from 53% to 78%. The following table provides the correlation coefficients between the participation rate and the percentage of votes obtained by different parties. On the one hand, we observe a strong correlation for extreme left parties indicating a strong mobilization of voters. On the other hand, the correlation is significantly negative for extreme right parties and environmentalists. This result is somewhat difficult to interpret suggesting swing or opportunist voters who can change their votes at the next election.1 4 13It is a very rough representation of the electorate in France and of their political preferences. We can find many examples contradicting this tendency. In general, leftist parties favor redistri bution which benefits low income households, while rightist parties favor tax cuts which benefits high income households. 14It may also indicate voters disillusioned by the standard leftist and rightist parties who have 28 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2.3: Participation Rate and Election Results Party Correlation Coefficient Extreme Left Left Right Extreme Right Green 0.19 -0.04 0.13 -0.36 -0.42 Lastly, it would be interesting to analyze the motivation of politicians. There are two assumptions in the political economy literature. Namely, politicians are either rent seekers or partisans. The former refers to politicians who are self inter ested. Politicians can be interested by some rents associated to the position or by just maintaining their positions as long as possible. We call these politicians rent seekers or office seekers. A good indicator of self interested politicians could be the proportion of cities for which the same party with the same leader is reelected in 1989. Out of the 57 cities in our sample, 36 mayors elected in 1983 were reelected in 1989, which represents about 63%. For the remaining 21 cities, we find that, for 12 of them, the mayor in 1983 ran for the 1989 election at the head of a list and, for 5 of them, the mayor elected at the 1989 election, was participating at the 1983 election at the head of a list. Therefore, for 84% of the cities in our sample, we find that the same political leaders have participated at both elections. For the remaining 16%, it is, however, possible that the political leader at one election was participating at the other election taking the second or third position on the list, preferred either to vote for something different such as environmentalists or to abstain. 29 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The data provide the identity of the list head only. Many of the mayors elected in large cities have later held offices at the national level and in the government sug gesting again their office-seeking motivation.1 5 Nonetheless, the empirical analysis of the following section will show that their behavior responds to both ideological and office-seeking motivations. 2.2.2 Public Transportation D ata The data provide detailed information on the operating costs including labor, energy and material costs, the number of buses as well as the number of employees. Fixed costs such as capital are not included in the data as these costs are not born by the firm. In particular, buses are provided by the city. Detailed information on production can be found such as the network size, the number of seat kilometers, the average speed and the number of passengers. In addition, the revenues from the bus fares and subsidies are also provided. All the data have been deflated and expressed in constant 1985 FF. See Perrigne (2002) for a detailed description of the data. In this paper, we are interested on whether the political composition of the city council and the political affiliation of the mayor have influenced the regulatory contract. Since the contract is characterized by the pair, price and transfer, we assess loThe data provide some interesting examples of political dynasties such as in Toulouse, where two generations of the Baudis family have been mayors holding the city hall for more than 40 years. Another example includes Bordeaux with Chaban-Delmas, who has ruled the city hall for more than 30 years. 30 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. any relationship between the price and the political composition of the city council. Moreover, the political composition of the city council may have also influenced the quality of the service offered to passengers. Since public transportation is a highly subsidized activity (the fare covers on average only 48% of operating costs), it can be considered as a redistributive toll among the population. In this sense, we can expect that cities with a leftist ma jority will subsidize more public transportation than cities with a rightist majority by offering a lower price to commuters. In the same spirit, we can also expect that cities with a leftist majority will offer a better service with an extended coverage to render public transportation accessible to all. Given the strong arguments used by environmentalists during their campaign on free transportation and pollution, we can expect them to lobby on these issues by reducing the price of public trans portation. Given that their representation in city councils is small (see table 2.1), it is unclear whether they can really influence the city policies in terms of public transportation. To analyze the right/left opposition, we have spilt the data set into two data sets according to the affiliation of the majority in the city council. To simplify the analysis and because of the low number of cities with an extreme left ist majority, we have considered these cities as leftist wing.1 6 Cities with a leftist majority have an average fare equal to 2.25 French Francs (FF), while cities with a rightist majority have an average fare equal to 2.44FF, confirming our expectations. 16The results are not dramatically different. 31 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.1 displays both price densities. Figure 2.1 shows a larger range of values for right wing cities with a larger mode than for left wing cities.1 7 Since some cities have experienced a political change between the two elections, we have drawn similar graphs while comparing cities with the same majority in 1983 and 1989 and cities experiencing a change in the majority between 1983 and 1989. Though we have a small number of observations for assessing these changes, we observe a clear change in pricing policies. For instance, we compare price densities for cities with a leftist majority at both elections with cities switching from a leftist to a rightist majority. See Figure 2.2. The density for the former cities is unimodal with a large variance, while the density for the latter is bimodal with a small variance. The former is slightly at the left of the latter. The comparison of these densities suggests that only one city experiencing a switch has maintained its low prices while the other cities have (gradually) changed their prices for public transportation while increasing them. Figure 2.3 displays the price density for cities with a rightist majority at both elections and for cities with a rightist majority in 1983 switching to a leftist majority in 1989. The contrast is not as striking in this case but we can observe a smaller mode for the latter than for the former with a second mode at a larger value indicating that cities experiencing a change in majority have slightly 17The data do not provide detailed information on the pricing policy. About 80% of cities in France offer special fares to some groups of the population such as for young, elderly, unemployed and low revenue people. Leftist and rightist cities may differ in their pricing policies through these discounts given to subgroups of the population for redistribution purposes. 32 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.1: Price Densities for Left Wing and Right Wing Cities o R ight □ Left .8 - T T T T 8 2 4 6 B usfare Figure 2.2: Price Densities and the Change from Left to Right o LEFT TO R IG H T □ LEFT T O LEFT 1 .5 0 T “ 6 2 4 0 B usfare Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.3: Price Densities and the Change from Right to Left o R IG H T T O LEFT □ RIG H T TO RIGHT .8 T T T T " 8 2 4 6 B u sfare Figure 2.4: Price Densities and Presence of Environmentalists o G re e n □ No g re e n .8 .6 .4 .2 0 0 2 4 b u sfa re 6 8 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 2.5: Price Densities of Left Wing Cities With or Without Environmentalists o N o G re e n □ W ith G reen .8 .6 .4 .2 0 "T~ 0 2 4 b u sfa re Figure 2.6: Price Densities of Right Wing Cities With or Without Environmentalists o N o G re e n □ W ith G re e n 1.5 1 5 0 0 2 4 6 8 b u sfa re Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and gradually lowered the price for public transportation. We have performed a similar analysis for various quality indicators such as the network coverage measured by the ratio of network size to the area of the city and the number of seats offered divided by the size of the population. The observed differences are minimal with a slightly larger mode for the left wing cities for the network coverage. Similarly, the number of available seats per capita is slightly larger for cities with a leftist majority. These results suggest that the quality of service does not differ much according to the left/right majority of the city council.1 8 As mentioned earlier, the environmentalists are strong advocate of public trans portation in the hope that its development will reduce the level of car pollution in cities. It would be interesting to assess whether the presence of their representatives in city councils has an impact on the policy for public transportation.1 9 For this purpose, we conduct a similar study relying on price densities as well as for other various quality indicators. When considering the full sample of 57 cities, we find that the average price of public transportation for cities with no environmentalist in their city councils is equal to 2.41FF, while this average price drops to 2.15FF when there is at least one environmentalist in the city council. This implies that on 18The graphs are available upon request. 19Following the 1983 election, 8 cities had ’green’ representative(s) in the city council, while there are 27 cities with a ’green’ representation following the 1989 election. This provides enough observations to assess such a change. Since the election is held in Spring 1989, we consider that some changes, if any, occured only starting in 1990. 36 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. average passengers pay about 12% more for public transportation in cities with no environmentalist in the city council. Figure 2.4 displays both price densities. To analyze this effect further, we consider separately cities with a leftist or right ist majority. For leftist cities, the average bus fare is equal to 2.32FF when there is no environmentalist council member, while it is equal to 2.12FF when there are some. We have observed some alliances between the environmentalists and the left wing. Not surprisingly, the leftist politicians have then to reconsider their environmental policies when governing the cities. This fact could partly explain this observed dif ference. For cities with a rightist majority, we observe a more striking difference with an average bus fare equal to 2.5IFF when there is no environmentalist in the city council and equal to 2.19FF when there is at least one environmentalist in the city council. This represents a difference of about 15%. Figures 2.5 and 2.6 display the price densities for leftist and rightist cities, respectively. The range of values is clearly larger for the rightist cities with no environmentalists as displayed by Figure 2.6. Figure 2.5 displays a larger mode for the leftist cities with no environmentalists with a larger range of values as well. To our knowledge, the right wing has never had any particular agreement with the environmentalists between the two rounds of election. Therefore, the above ar gument of political alliance is not valid. We have then to find other explanations to rationalize such differences. The French election system offers a strong majority to the winning list to govern the city. In this respect, the winning list has free hands 37 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to govern according to their own aspirations. Empirical evidence would suggest that politicians’ main motives are to hold office and to be reelected. Politicians have then to maximize the probability of being reelected in a dynamic perspective. When the electors express some concerns for environment through their votes to environmentalists, it is in the politician’s best interest not to ignore this signal and to apply some environmental policies to capture these votes at the next election. In the political economy literature, these voters are called the swing voters as they are mobile across parties because of their low concern for ideology. If we view the environmentalists as a special interest group, their voters will vote for the party rep resenting their interests. By implementing an environmental policy such as offering public transportation at a lower price, leftists or rightist politicians may hope to gain some of these votes at the next election.2 0 To conclude this section, we have performed a hedonic price analysis for bus fare, while relating bus fare or price to a set of various quality indicators and political factors keeping in mind the previous evidence. We also introduce some indica tors for the cities’ financial constraints. As discussed previously, politicians are self-interested and would like to please their electorate with the objective of being reelected. They can do so by reducing the price for public transportation. When 20When looking at quality indicators, we observe that the extent of the network is significantly larger for leftist cities with environmentalists, namely 247 versus 172. For rightist cities, we observe a larger extent of the network when there is at least one environmentalist, namely 166 versus 140. It is unclear how the environmentalists have influenced the quality of public transit. 38 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. considering politicians motivated by ideology, they could reduce the price for public transportation while considering it as a redistributive tool. Politicians are, however, refrained in their policies by financial constraints. Though all the cities in our sam ple have some debts, cities cannot run into systematic deficits every year. We can expect that cities with already high debts will be constrained in their redistributive Table 2.4: Hedonic Price Model for Public Transportation Variable Fixed Effects Random Effects Constant -1.9018 — -1.9017 (0.000) (0.028) Network Extent 0.1179 0.2691 0.1978 (0.000) (0.000) (0.000) Area 0.0929 — 0.784 (0.000) (0.072) Seats per Capita 0.4109 0.2645 0.3479 (0.000) (0.037) (0.000) Debt per Capita 0.0970 — 0.0849 (0.000) (0.157) Unemployment Rate 0.3111 — 0.3144 (0.000) (0.034) Leftist Majority -0.0767 -0.0392 -0.0396 (0.001) (0.186) (0.145) Environmentalist -1.4880 -0.9448 -1.2362 (0.002) (0.045) (0.007) Time Trend 0.0254 0.0174 0.0277 (0.000) (0.000) (0.000) R 2 0.344 0.309 0.772 objectives or policies in general. Moreover, a tax on firms has been gradually imple mented in Prance starting in 1971 to partly subsidize public transportation. When the economic activity is low, the city will have to pay a larger proportion of subsi dies to public transportation since the taxes levied on firms will be less. Therefore, Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the city’s debt and economic activity could constrain the politicians in their public transportation pricing policies. We regress the logarithm of prices on a constant, the logarithm of the city area, the logarithm of the network size divided by the area of the city, the logarithm of the number of seats available divided by the population size or number of seats per capita, the logarithm of the amount of city debt divided by the population size or debt per capita, the logarithm of the unemployment rate, a dummy for a leftist majority, the proportion of seats held by environmentalists in the city council as well as a time trend.2 1 All the results are displayed in Table 2.4, while the p-values are given between parentheses. All the coefficients in the model that do not take into account the panel structure of the data are significant and meet our expectations. The quality of the service offered to commuters as measured by the network extent and the number of seats per capita contribute to increase the price of public transportation. An increase in the quality would raise demand and that would lead to more buses being operated thereby raising costs and being reflected in an increase in price. We note, however, that an increase in 1% of either the extent of the network or the number of seats per capita increases the price by less than 1%, namely 0.12% and 0.41%, respectively. A 2 1 Note that the data on the debt per capita axe unavailable for the 1985-1993 period. We then collect data for 1994 and 1995 and average over the two years. Note that these observations are related to the post 1989 election. We have then checked whether the debt is larger for leftist cities after 1989 than for rightist cities as we could expect. We have not found any particular pattern. 40 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. larger city area is associated with a higher price. It is well known that sprawl cities are on average more costly to operate than compact cities. This cost increase is again reflected in the price. We have introduced two variables in the hedonic price equation to assess the financial constraints of the city. Cities with a high debt per capita have to be cautious in their expenses to control further development of debt. Thus these cities are financially constrained in subsidizing public transit. A 1% increase in the debt per capita increases the price by 0.1%. This result suggests that cities put an important priority on their public transit increasing by little the bus fare when the level of debt increases. The unemployment rate has a more dramatic effect probably because of the tax levied on firms to subsidize public transit. A city experiencing a large unemployment rate is likely to receive less taxes from firms, implying a larger proportion of the transportation costs to be subsidized by the city leading the city to increase bus fares. An increase by 1% of the unemployment rate increases the bus fare by 0.32%. Given that there is a positive correlation between unemployment rate and the debt per capita, we can expect this effect to be even stronger for cities experiencing both, namely a large debt and an important unemployment rate. We also consider two political variables, viz., dummy for left majority and the presence of ’green’ in the city council. The two political variables are significant. A dummy has been considered for the left versus right majority instead of the proportion of the council seats held by leftist politicians. In cities with a rightist majority, 77.92% of council seats are held by rightist politicians, while in cities with a leftist majority 41 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 60.85% of council seats are held by leftist politicians. In view of these figures, it is not as much the proportion of seats held by leftist politicians which matters in the policy decisions taken by the city council but rather the nature of the majority in the city council. In contrast, the proportion of seats held by environmentalist is a variable of interest as it represents their popular support. A larger proportion of environmentalists may have a greater influence on the policy decisions because it represents a larger proportion of swing voters that the majority could attract in the next election.2 2 Everything else being equal, a switch from a rightist to a leftist majority would decrease the price by about 7.3%. Similarly, an average number of environmentalists in the city council (4.80%) would decrease the price by 6.89% with the same majority. The time trend indicates a price increasing over time due to other factors. To take into account the panel structure of the data, we have also considered a model with fixed effects and a model with random effects. The magnitude of the coefficients is somewhat different but the coefficients remain significant except for the dummy for the left majority. This could be explained by the fact that a large proportion of cities, about 74%, have not experienced a change in the majority over the period. We also note that the debt variable becomes less significant in the random effects model. Since the data on debt provides a single observation for each 22Conditional on the presence of environmentalist in the city council, they have obtained 2.63% of council seats after the 1983 election while this number increases up to 4.80% after the 1989 election. 42 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. city in the sample, it may have been captured in the random effect. For the fixed effect model, we have tested the equality of the fixed effects. This hypothesis is strongly rejected with a p-value equal to 0.000. It is unclear whether the city effects are correlated with the other exogenous variables in the model such as the quality variables and the political variables though our intuition will favor correlation. We have then performed a Hausman test. The p-value is equal to 0.0715, which does not provide a clear response since we can consider the fixed effect model at 10% and the random effect model at 5%. Since the 5% level is the most widely used, we consider that there is no correlation between the city effects and the exogenous variables of the model given in Table 2.4. The variable debt per capita could raise the possibility of a simultaneity problem in the model. We note that the amount to be subsidized computed as the total operating costs minus the revenue from bus fares is still a small proportion of the total budget of the city, namely 6.65% on average, excluding a simultaneity problem.2 3 Moreover, in view of regulatory models, the error term in the hedonic price equation includes the unobserved firm’s type, which can be potentially correlated with the exogenous variables in the hedonic price model. This term of unobserved heterogeneity is captured by the city effect. Since the random effect model is preferred over the fixed effect model, such correlation is negligible. 23Note that a tax on firms is used to subsidize public transit. We can then expect that public transit represents less than 6.65% of the cities’ budget. 43 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We can then consider that the variables in the hedonic price model do not suffer from endogeneity. This section has provided strong empirical evidence that the political composi tion of the city council influences the price of public transit and therefore its regu lation. The next sections present a model incorporating this political dimension in the regulatory contract as well as its estimation on the data. 2.3 Economic M odel and the Estim ation M ethod We need to consider an incentive regulatory contract model, while incorporating a political dimension which is in agreement with the empirical evidence displayed previously. Moreover, this model has to be structurally “estimable.” We exclude regulatory capture models as in Laffont and Tirole (1991) since the environmentalists directly participate in the election. In addition, these models involve three parties instead of two (three-tier models) and have been derived for two types of firms. As such, their estimation would require additional data and the strong assumption that firms reduce to good (efficient) and bad (inefficient) firms.2 4 We first discuss other possibilities that we found in the literature. 24The extension of such models to a continuum of types seems to be extremely difficult to conduct and is beyond the scope of this paper. 44 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 2.3.1 Discussion of Various Approaches Laffont (1996) and Laffont (2000) propose some extensions of the Laffont and Tirole (1986) model while taking into account a political dimension to industrial policy. Two types of consumers are considered in the population, namely type 1 and type 2. A proportion 1 — a of type 2 consumers enjoy the good more than the proportion a of type 1 consumers. Type 2 consumers could be considered as those using public transit and type 1 consumers those who do not use public transit. Note that, with the exception of Paris which is not included in the data, 1— a is quite large in the case of public transit.2 5 When considering a simple random majority model, type 1 and type 2 consumers, whoever gets the majority at the election, will delegate to a politician the design of the regulatory contract. Majority 1 representing type 1 consumers will maximize the social welfare for the proportion of type 1 consumers only ignoring the remaining 1 — a proportion of the population.2 6 Similarly, majority 2 representing type 2 consumers will maximize the social welfare for the proportion of type 2 consumers only, while taking into account that these consumers enjoy the good more by considering a weight ( 3 larger than one for the consumer surplus. To simplify, we consider the case of public ownership of the firm. The results are as follows. When majority 1, (say) right wing, wins the election, the solution to 25We have not found exact data on the proportion of the urban population using public transit. At least one person uses public transit in 60% of French households. 26While not taking the bus, type 1 consumers still derive some utility from public transit but not as much as type 2 consumers. We can consider that they benefit from public transit as it decreases the level of traffic congestion. 45 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the problem gives an equation similar to the Ramsey pricing, while the contract induces less effort from the firm relative to the Laffont and Tirole (1986) model. An important assumption in the model is that the proportion a needs to be larger than 1/(1 + A), A representing the shadow cost of public funds. In developing countries this cost is estimated at 0.3. This would require a to be larger than 0.77, which is a very strong majority. Similarly, when majority 2, (say) left wing, wins the election, the solution to the problem gives a distorted level of effort, which is in this case similar to the one in the Laffont and Tirole model (1986), while the price tends to be lower than the standard Ramsey pricing because of the larger weight ( 3 on the consumer surplus. Here again, an important assumption is to have 1 — a larger than 1/(1 + A), implying a strong majority at the election. Relative to the social optimum, majority 1 leads to a smaller amount of public transit than is socially optimal, while majority 2 leads to a larger amount of public transit than is socially optimal. This makes sense since politicians in this model care for their electorate only, which is supposed to enjoy public transit more or less according to their types 2 or 1, respectively.2 7 The model can be further extended while considering private ownership, while assuming that the firm belongs (say) to the right (majority 1). 27Laffont (2000) considers an incomplete contract framework, in which the constitution has the choice to delegate or not the design of the regulatory contract to the politicians based on the unknown value for the cost of public funds. If the economic conditions are unstable or the variance of A is large, it is better to delegate to the politician. The problem is different in our data since the design of the contract is delegated to a transportation authority under the control of the politicians. 46 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. As expected, the provision of public transit is smaller than under public ownership because the firm’s rent is undervalued by majority 1. Despite the fact that this model provides an interesting framework, taking into account the political dimension, it does not fit into the analysis of our data for the following reasons. First, it is not so simple to divide the population into 2 types of consumers and to assume that these consumers have a uniform vote. As men tioned previously, the proportion of transit commuters is probably less than 50% of the population. Moreover, to our knowledge, transit commuters are not well organized to constitute a special interest group with representatives.2 8 Moreover, transit commuters represent a very heterogenous population. Though we have not found a statistical study on who are the transit commuters in France, various docu ments suggest the following users. Given the quasi absence of school buses in French cities, many middle school and high school students rely on public transit. Accord ing to data provided by the Groupement des Autorites Responsables des Transports (GART), there are 1.9 million students taking the bus every day in France. The car ownership rate in France is about 79.4% and only 28.5% of French households have 2 cars or more, leaving a non negligible proportion of the population relying on other transportation modes than the car. Moreover, only 77% of the population over 18 28A typical example of this lack of organization occurs when public transit is on strike. Transit commuters may lose days of work because of these strikes organized by a small group of unionized workers. Transit commuters have never lobbied to restrict laws on strikes for public transit bearing many of the consequences of such transportation strikes. This absence of lobbying is probably due to their dispersion. 47 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. years has a driving license (89% of men and 67% of women). This would suggest that sections of population with low revenue such as immigrants and working class households and with mobility problems such as elderly are the main users of public transit in addition to students.2 9 Given the heterogeneity of public transit consumers, it seems inappropriate to assume a unique vote in our case. Students and immigrants do not vote in general, while working class households are usually leftist partisans. In contrast, elderly people tend to be more conservative with a rightist vote. Though we could roughly predict that transit users usually favor a leftist ideology, the split is not as sharp. Second, these models require a very strong majority for the politicians to ignore a part of the population. Our data show a single city with such a strong majority (at least 77% of the votes for a A equal to 0.3).3 0 Parties govern with a comfortable majority given the mixed majority/proportional electoral rule in Prance. For cities with a rightist majority, the average proportion of seats in the city council is 77.92%. Thus, not all the cities in our sample would satisfy the assumption of having the majority with a least 77% of the seats in the city council. Third, the data show strong evidence of the influence of environmentalists in the choice of the price for public transportation. The models that we have briefly discussed ignore this possibility as the politicians only care for their majority. Lastly, the ownership of firms operating 29 A study by the GART also shows that two thirds of the public transit users are women. 30Cannes in South of France had 81.26% of the votes obtained by the right wing in the 1989 election. 48 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. public transit cannot be easily classified as private versus public. Three companies, namely Keolis, Connex and Transdev, shared 74% of the networks in Prance in 2001, representing 90% of the market share in terms of passengers. Two of these companies are semiprivate and are affiliated with public firms. In addition to the difficulty of defining the public versus private ownership, the data do not provide information on the identity of the operator. Moreover, it seems difficult to consider that these companies belong to the left or right majority. While finding inspiration from this literature, we propose an alternative model. 2.3.2 The M odel We adopt a Principal-Agent framework to model the incentive regulatory con tracts between the regulator and the operator with a political economy dimension. As discussed in Perrigne (2002), a model of incentive regulation with ex post ob servability of the costs as first studied by Laffont and Tirole (1986) seems to be a good approximation of the reality. Our data suggest that contracts are nonetheless influenced by the political composition of the city council. Therefore, we need to relax the widely accepted assumption of a benevolent utilitarian maximizer. We will come back to this assumption later when we describe the model. The basic idea is to assume that the firm in charge of operating the bus service has hidden information on its efficiency and takes hidden cost reducing actions called 49 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. effort. These lead to the so called problems of adverse selection and moral hazard, respectively. Both affect the firm’s cost in the sense that a higher efficiency and effort level tend to decrease the costs. Both are assumed to be unknown to the regulator. The regulator does not know the firm’s efficiency and effort level in the sense that even when observing ex post the firm’s cost, he cannot disentangle these effects from the random shock in the cost. Through the contract proposed by the regulator to the firm, which consists of a price and a cost reimbursement rule based on ex post observed costs, the firm will be induced to exert appropriate effort level without lying about its efficiency. This leads to the well known efficiency-rent trade off, where the regulator has to give enough incentives to the firm in terms of profit to reach efficiency while extracting some firm’s rent. The firm knows its efficiency or type denoted by 9, which is assumed to be drawn from a distribution F(-) defined on [9,0] with a density /(•) > 0, where 9 denotes the most efficient firm and 9 the least efficient one. The regulatory authority offers a contract to the firm based on the expected demand and cost, which are both subject to some external shocks.3 1 This contract is defined as the pair (p,t), where p denotes the price that the firm will be authorized to charge the consumers and 31The Laffont and Tirole (1986) model assumes a constant marginal cost with an additive random shock, while the demand is known with certainty. A fixed demand is unrealistic in the case of public transit, whose demand can fluctuate from one year to another. French data show that the demand for public transportation has increased over the period of study, while public transit demand has experienced a slight decrease over the recent years. It seems difficult to predict with certainty these fluctuations justifying a random demand. 50 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. t , the transfer given to the firm. This transfer is a cost reimbursement rule as a function of the ex post observed realized cost and the announced type. After the demand and cost are realized, the transfer is paid to the firm based on its actual (observed) cost. Because public transportation is a private good, the firm faces a demand, de noted by y(p), subject to some random shock denoted e^. To simplify we adopt a multiplicative random shock and denote y(p) the deterministic part of the de mand, namely the expectation of the demand with respect to ej is equal to yip)- The revenue generated from the bus fares is equal to R(p) = py(p). This demand generates a gross consumer surplus denoted S(p), while S(p) — R(p) denotes the net consumer surplus and can be computed as f£° y{p)dp following Assumption 1 in Perrigne (2002). The firm’s effort is denoted by e. Effort is costly to the firm and therefore causes disutility due to effort, ip{e). Usual assumptions are made such as ip'(-) > 0, > 0,Ve > 0 and 'ipi0) = 0.3 2 The firm’s cost denoted by C(y,9 — e) is subject to a multiplicative random shock ec. The cost is increasing in 0 and decreasing in e. The firm’s utility is defined as U = t — ip(e), where t is the net transfer to the firm, which is a function of the firm’s type 9 and the realized cost C(y,9 — e), namely t{9, C{y,9 — e)). The gross transfer is by definition equal to t + C(y , 9 — e). Lastly, giving a subsidy to the firm requires additional taxes which introduce a distortion in 32We could add > 0 to avoid stochastic mechanisms. 51 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the economy. Thus the shadow cost of public funds denoted by A needs to be taken into consideration. Though we will estimate A , in western countries where the tax collection system is efficient, A is estimated at O.3.3 3 The firm’s optimization problem is assumed to be independent of political con siderations as, to our knowledge, the firms operating public transportation do not change over time according to the political majority. Though some firms are semipri vate, we still consider that their optimization problem is to maximize their utility U (9, e) = t(y(9),9 — e) — ip(e) with respect to the two decision variables (9, e). Note that y becomes a function of 9 through the price decided by the regulator. The firm may lie about its efficiency parameter 9. Let 9 denote its true efficiency and 9 its announced efficiency. Thus, a firm with 9 efficiency, announcing 9 efficiency, has a utility U(9,9, e) = t{9,C(y{9,9 — e))) — ip{e). Because the demand and the cost are subject to some random shocks unknown at the time of the design of the contract, the firm maximizes the expected value of its utility, namely E[U(6,9, e)], where E[] denotes the expectation of the term between brackets with respect to the random shocks (€d,ec). Expectation has to be taken with respect to both random shocks since ea affects y. Maximizing E[U(9,9, e)] with respect to e for a fixed value of 9 gives an effort as a function of the announced 9 conditional on the true 9, i.e. e{9\9). Thus the 33This value is widely accepted among economists. It is expected to be larger in developing economies. See Ballard, Shoven and Whalley (1985) for an estimate of A from US macroeconomic data. 52 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. firm’s maximization problem can be written as maxg E[U(9,9)\, where U(0,6) = t(9,C(y(9),9 — e{Q\0))) — xp(e(9\9)). The solution of this program gives 9 — 9(9). Using truth-telling, U(9,9) = U(9,9). To simplify notations, we denote the latter by U(9). Maximizing E[U(9)\ gives the following first-order condition after some algebra E[C7'(#)1 = -V-'(e). (1) Equation (1) provides the incentive compatibility (I C ) constraint to the regulator’s maximization problem. We need to pay particular attention to the regulator’s maximization problem. It is more complex than for the case of a public good as considered usually in the theoretical literature. As a m atter of fact, it has to take into account various factors such as the net consumer surplus S(p) — py(p), the revenue evaluated at the cost of public funds (1 + A)py(p), the cost for the gross transfer to the firm evaluated at the cost of public funds (1 + A)(t + C(y,9 — e)) in addition to the firm’s profit or utility U. We have discussed in detail that politicians are self-interested in the sense that they maximize the probability of being reelected to maintain their position. Thus they have to please their electorate while trying to convince more electors to vote in their favor in the next election. Moreover, they can be motivated by partisan ideologies, which can be dictated by their parties such as redistributive 53 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. policies for leftist parties. For both reasons, the politician is not a benevolent utilitarian maximizer. Given the heterogeneity among bus commuters, it seems, however, difficult to target successfully a particular group of the population to gain their votes. This problem has been acknowledged in the political economy literature for the provision of public goods. Thus, the politicians cannot consider only a part of the population in their maximization problem. We propose the following solution to the Principal’s maximization program. The politician will maximize the various components of the social welfare, while putting a different weight on the consumer surplus according to his/her political affiliation and composition of the city council. In particular, we expect that cities with a leftist majority and cities with environmentalists in their city council will put a larger weight on consumer surplus than other cities. The impact on the contract will be studied further. We consider the following “social welfare” for self-interested politicians S W = /3(S(p) - py(p)) + (1 + A)py(p) - (1 + A)(t + C(y , 9 - e)) + U, where (3 is a weight on the consumer surplus. The functional form for /3 will be discussed in the section on the estimation method. It should be noted that this weight can take a large range of values depending on how much politicians care 54 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. about consumers. Using t = U + 4>(e), the social welfare function can be written equivalently as S W = PS(p) + (1 + A - P)py(p) - (1 + A M e ) + C(y, 9 - e)] - XU, where U = U(9). The regulator’s objective is to maximize the above welfare function under the firm’s incentive compatibility and individual rationality constraints. The latter states that the firm should expect a positive profit to accept the contract. Be cause the regulator decides the contract to offer that he offers to the firm at the beginning of the period and because the demand and the cost are both subject to random shocks unknown to both parties ex ante, everything needs to be taken in expectation with respect to these two random shocks. This gives the following maximization problem for the regulator or politician max f E[SW}f(6)d9, (2) P > e’U J q Subject to E[U'(6)] = - ^ '( e ) (IC ), E[U(6)] > 0 (IR ), where S W is given by (2). Because the firm’s efficiency or type 9 and therefore its effort level e(9) are unobserved, the regulator maximizes an expected social wel- 55 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. fare with respect to 6. Because U (•) is decreasing in 9, the individual rationality constraint is satisfied if and only if E[U(6)\ = 0. It is then commonly replaced by E[U(9)\ = 0, i.e. the least efficient firm will have a zero profit, since S W is decreas ing in U. The firm will then reveal the truth about its efficiency level 6 and will exert appropriate level of effort. Such a maximization problem can be solved using the Pontryagin principle. Let Eed[y(p)\ — y(p). The Hamiltonian is H = j ( l + A)py(p) + P J y{p)dp- (1 + A)^(e) - (1 + X)E[C(y,9 - e)\ - x E m e ) ) } m + v m - W ) , where /j(-) is the co-state variable and E[IJ(•)] is the state variable. After some basic algebra, the first-order conditions are as follows HP = {(1 + A)py'{p) + (1 + A - 0)y{p) (3) — (1 + A)E p ^ “ ] } /( 0 ) = O , He = { - ( 1 + - (1 + A ) « Z | t f l l | m _ ^(0)V,"(c) = 0, W -H sium = V (« ) = /*'(«). (5) where Hx denotes partial differentiation of the Hamiltonian H with respect to X and y'(p) denotes the expectation of the derivative of y(p) with respect to p. 56 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. By integrating (6) and using the transversality condition p,(6) = 0, it gives ji{6) — AF(6). Replacing ji{0) by AF(6) and rearranging terms, (4) and (5) become a cost function with a constant marginal cost and an additive random shock such as C(y, 6 — e) = (0 — e)y+ec, (7) gives the Ramsey pricing rule for (3 — 1, where the price depends on the cost of public funds A , the marginal cost and the demand elasticity. Such a specification is in general chosen by theorists giving the numerator of the left-hand side of (7) as p — dC/dy (see for instance Laffont and Tirole (1986)). Because we consider a more general functional form, (7) does not have the same interpretation. Moreover, because of /?, the price distortion takes a different form. While taking the simple example of a constant marginal cost, we observe that the price will be closer to the marginal cost for/3 > 1 since (1+A — /3)/(l+A ) < A/(l+A) assuming a constant price elasticity. In contrast, when (3 < 1, the price will be larger than the Ramsey price. Equation (8) provides the effort level. The optimal effort level is distorted relative to the first best as in the Laffont and Tirole (1986) model providing similar incentives - ( 1 + A - P ) y ( p ) 1 (6) P 1 + A p y'{p) ’ ^ (e) = < 9 E W - e ) ] (7) Equation (7) provides the pricing rule adopted by the regulator. When considering 57 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. to the firm. Equations (7) and (8) will be the basis of the econometric model to estimate incentive regulatory contracts. 2.3.3 Estim ation M ethod The econometric model is defined from (7) and (8). Since the estimation problem is similar to the one studied in Perrigne (2002), we will rely on the method used therein to estimate the model. As a matter of fact, (7) differs due to the introduction of the consumer weight (3. To define the econometric model, we need first to parameterize the demand, the cost and the disutility of effort. The demand function for public transportation takes the following simple form yip) = exp(d0)Z^pd2 exp(ed), where Zd denotes a vector of exogenous variables such as the city’s characteristics and indicators for quality, is a random shock. Using this specification, d2 (d2 < 0) is interpreted as the price elasticity. As in Wolak (1994), we can assume that this demand is defined only for p < pmax and takes a zero value for any price above this maximum price.3 4 34Assuming the existence of a maximum price allows a well-defined net consumer surplus for this constant elasticity demand function for any d2 < 0 . 58 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Given the complexity of the model, we choose a rather simple representation of the technology using a Cobb-Douglas production function. Under this assumption, the variable cost function is expressed as C(y, 9 - e ) = exp(/30) exp{(3L(9 - e))jPLLj^ y {jp fy exp(ec), where Pl and pM denote the price for labor and material, respectively, and ec is a random shock. We assume that the firm’s type and effort affect labor efficiency. As in Wolak (1994), capital costs are excluded and C represents the operating costs.3 5 We impose homogeneity of degree one in price by setting (3L+(3M = 1. We could also introduce in (10) some exogenous variables Zc associated with a vector of coefficients /3Z. Regarding the disutility of effort ip(e), we choose the following exponential form 'ip(e) — exp(ci'e) — 1, where a is a parameter to be estimated. To satisfy the assumption of the model, a needs to be strictly positive. The random shocks ed and ec are unobserved ex ante by both the regulator and the firm. These random shocks will be interpreted as 35Working with operating costs can be easily justified. The main capital (buses) is provided by the City or the local transportation authority and does not show up in the firm’s accounting report. Moreover, an accurate measure of capital costs is usually difficult to obtain. When adding the labor, energy, maintenance and other material costs, we obtain about 90% of the costs reported by the firms in the sample. 59 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the error terms in the econometric model that we will define in the next section. We make the following assumption on ed and ec. Assumption A: The random shocks ed and ec satisfy E[exp(ed)\Z] = 1 and E[exp(ec)\Z] — 1, where Z is a vector of exogenous variables. Moreover, the firm’s type 9 is independent of ed conditional on Z. The first part of assumption A is quite natural following the multiplicative random shocks in (9) and (10).3 6 Note that we do not make any other assumptions on the distribution of the random shocks beyond their first conditional moments. Moreover, we do not make any independence assumption between these two random shocks as they can be correlated.3 7 The second part of Assumption A is in agreement with the definition of the firm’s type as made by theorists in the sense that 9 is idiosyncratic to the firm and known before any realization of the demand. Moreover, it is crucial to identify the model as discussed later. Assumption A will be used later in the estimation method to specify moments defining GMM estimators. It remains to discuss the parameterization of the firms’ type density /(•). We consider a Gamma density, i.e. f(9 ; r, 7) = 7 (79)r~1 exp(— 70)/T (r), where r G W+ and I \r ) = / 0 ° ° xr 1 exp(—x)dx as the gamma density allows enough flexibility.3 8 36The expected net consumer surplus E[S(p) — p y {'[))] becomes f£°y(p)dp. 37This assumption precludes maximum likelihood estimation, which requires a full parametric specification for the joint density of the random shocks as in Wolak (1994). We could have con sidered a log-log form of (9) with an expected random shock ed equal to zero. In this case, yip) would be interpreted as the expected logarithm of the demand, which would complicate (7) and (8 ) and therefore the derivation of the econometric modelling. 38 Given that we have no information on the type density, it would have been interesting to 60 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We observe N = 57 x 9 = 513 contracts and index by i the ith contract be tween a firm and its corresponding city. Rewriting the price equation (7) using the above parameterization and the logarithm, the econometric model is defined by the following three equations yt = exp(d0)Z^p^2 exp(edi), (D) Ci = exp{Po)PLiPl m Lyl i y exP( P l ^ ~ e(°P)) exp(eci), (C) log pi = f30 + log Ec + log (3y - log ^1 + 1 \ X + + P l 1o§PLi + (! - Pl) 1o§PMi + {Py - 1) log y(Pi) + PL(6i - e(6i)), (P ) where is the weight for the consumer surplus and e(6,j) is the solution of the system of equations (7) and (8) at a given value of $i defining the optimal effort level, for i = 1,... , N. The term Ec arises because the cost needs to be considered in expectation with respect to e(i and ec, namely Ec = E[exp(/3yed + ec)]- In two of these equations, the effort e(0j) appears. Both e(8i) and B .L are unobserved. consider a model where /(•) is left unspecified or non parameterized. This would avoid any mis- specification issue and would reinforce any policy analysis. This would lead to a semiparametric model. Additional information would be needed to identify nonparametrically the type distribu tion. Such an issue is left for future research. 61 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Another equation could be used for the price. In particular, solving (7) and (8) in (p, e) given d,t gives the regulator’s price schedule p(Qi) and the firm’s optimal effort e(&i). Elementary algebra gives logPi = a- , , / 1 R ,— - (a +(3L) \ o g f l + ]- ^ a + hl + d2(—&Py + O L + (3l ) ^ \ 1 1 + A — 1 + A d2 + a 0 L\ o g ^ + a\ogpMi + (a(p - 1) - PL)(d0 + di logZdi) PMi + / J i ( Q9i + i ° g ( i + I T A 7 i ^ ) ) } ’ ( n ei = - o 7 - o ',-----— —A K ' - f 3 d 2\ o g [ l + l + X o l + (3^ + d2(y —oi(3y + e x . + (3l ) y y 1 + A d +Al(1 + ^2) l°g + (1 + ^2) log PMi + (3y(do + di log Zdi) PMi + /3t(i+m - ( 1 - - i))iog(1 + 1 “a a^ .’ ^ ) )) w for i = 1,... ,N, where K and K' are constant terms. Namely, K = a(/30+log Ec) + (a + PL)log(3y + PL{loga - log^L) and K' = (1 + d2)/30 + \ogpL(l - d2(f3y - 1)) + (1 + d2) log Ec + log a(—1 + d2((3y - 1)) + (3yd2 log (3y. From an econometric point of view, the random shock ed in (D) is interpreted as an error term. Similarly, the term di — eidp in (P) and the last term in (P’) as a function of di can be interpreted as an error term with a nonzero mean. The term di — e(di) in (C) is part of the error term exp(0L(di — e(dt))) exp(eC I). As a matter 62 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of fact, this term can be interpreted as a term of unobserved firm’s heterogeneity in (C). The difficulty in estimating such a model arises from the fact that we observe neither the 9i, nor the effort level e(0, ;) and that we cannot disentangle the part arising from the firm’s type 6t and the one from the random shock eci from the error term exp(f3L(6i — e(0,))) e x p ^ ) . Moreover, the density /(•; r, 7 ) of the firms’ types needs to be estimated whose parameters (r, 7 ) appear in e(8l) and in p(0t) as in (P > ) 3 9 In addition to the parameters (do, di,d2,P0,0L, Py, A , a, r, 7 ), we need to estimate the weight (3i for each observation in our sample. The reduced form analysis that we have previously conducted suggests that bus fares are a function of political factors. These political factors are taken into account in the model through the weight (5. We then propose to parameterize the weights /?, as a linear function of exogenous variables, namely Pi = Vo + v'i ZPi, where Zpi is a vector of variables reflecting the political composition of the city 39Wolak (1994) introduces an additional “econometrician” error term in the price equation. By parameterizing the joint distribution of all the error terms including 9, the model is identified. Such an “econometrician” error term is not necessary. Moreover, our method does not require to parameterize the distribution of the error terms e,j and ec as only a first moment restriction is used to identify and estimate the model. 63 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. council and the city’s financial constraints. Thus, the parameters to be estimated are (d0, di, d2, (30, (3L, f3y, r)0,r}1, A , a, r, 7 ), while the observables are {Ci ,pLi,P a h,yi,Pi,Zdi,ZPi,i = 1,... ,N}. We briefly discuss the identification of the model as the estimation procedure closely follows the identification. Using Assumption A which provides a first conditional moment for exp (6*), the parameters do, d\ and d2 are identified from (D) as well as the error terms e*, i = 1,... , N. Instead of using (P) to identify the parameters (0L, (3 , rj0, rq), we propose to use (P’) because (P’) will also allow us to identify a. In particular, using Assumption A, (P’) allows us to identify the parameters a, (3L and (3y. Note that (.P') (as well as (P)) allows us to identify only r;0/ (1 + A) and r/1 /( 1 + A ).40 As /3l , (3y, d2, r/0/ ( l + A ) and r/x/ (1 + A ) are identified, (P) allows to recover the error terms up to a constant, namely @0 + log Ec + (3L{6i — e(&i)), i — 1,... , N . Using these recovered error terms and Assumption A which provides a first conditional moment for exp(eci), (C) allows us to identify the error terms eci, i = 1,... , N. The recovered error terms (e*, eci,i — I ,... ,N) and (3y allow to compute Ec Al It remains to identify (30 and A as well as the parameters (r, 7 ) of the density /(•). Several identifying strategies can be used. A first strategy will consist in using the recovered values for {30 + 0 L (Oi — e(9i)) from (P), which is also equal to a function of parameters /30, A,r, 7 using (E). As the distribution of the left-hand side is known 40Note that the constant term in (P’) is not identified because we use an orthogonality condition. 41Note that we could have used this step to identify the parameters r/0/(1 + A) and 7 7 / ( 1 + A ) instead of identifying the latter from {P'). 64 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (from (P)), the above parameters are identified. Similarly, a second strategy could consist in using (P'), which provides a function of parameters /30, A , r, 7 . Using a similar argument, the parameters are identified. The estimation method closely follows the identification discussed previously. It is a multi-step procedure, where equations are estimated in a specific order allowing at each step to recover information, which will be used in the following step.4 2 The method is, however, simple to implement despite the complexity of the model. The first step consists in estimating the demand equation defined by (D) using Assumption A. In particular, the conditional moment E[exp(edi)\Zi] = 1 is used to define the unconditional moment P[$(Zj)(exp(e*) — 1)] = 0, where $(•) is a vector function of exogenous variables or instruments Z. This defines a Nonlinear GMM estimator based on the following moment E [ * ( Z j){y.exp ( -d 0)Z,-*p-'i ! - l } ] = 0. (8) The vector of instruments <&(Zi) could be chosen optimally following Chamberlain (1987). In the application, we choose an identity function for < & (•). We perform a two-step nonlinear GMM, where the matrix of weights is computed optimally to 42This multi-step procedure is certainly not the most efficient as estimating the full model jointly will bring some efficiency gains. Nonetheless, estimating the full model could introduce inconsistency of parameter estimates due to a misspecification of /(•)• As the estimation of /(•) is performed in the final step involving {(30,r, 7 ), the estimates for (d0. d \ , d?, fly) d° not suffer from such a potential inconsistency. 65 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. reduce the variance of the estimated parameters. See Hansen (1982) and Hayashi (2000) for GMM estimation. We then obtain estimated parameters (do,di,d2) and estimated residuals edi, i = 1,... , N. The second step consists in estimating the price equation (P’), whose error term (say) rji is a function of namely % = (3L(a6i+^(0i))/(a+P L + d2(-a/3y+a+(3L)). The moments of ql are unknown. We are using instead Assumption A, which implies that this error term is independent of edi conditional on Zt. This independence can be expressed as a conditional covariance equal to zero. This defines a Nonlinear GMM estimator based on the following moment E log p i , ----- —y ^ log PL a + /3l + d 2(—aPy + a + PL) pMi log pMi V a OL + P i + d 2 ( — O L p y + OL + “ + .log (l + i - J L . - T 7^ ! rT T Zpj o: + PL + d2(—aP + ol + PL) y d2 ^ (1 + A ) 0 ^ 2(1 + A ) a 0 y - a - p L OL + P i + d ^ — O i P + OL + P i ) (d0 + d1 log Zdi) > (exp(ed < ) - 1) = 0, (9) where d0 + di log Zdl, d2 and edi can be replaced by their estimates obtained from the first step, i.e. do+di log Zdi, d2 and edi, respectively.43 The estimated coefficients give a unique solution for the estimated coefficients (a, PL, py, r/0/ ( l + A), pl/ (1 + A)). 43As in the first step, we choose the identity function for the vector of instruments and perform a two-step nonlinear GMM estimator using an optimal weight matrix. 66 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The third step consists in estimating the cost equation (C), which involves (0* — e(6i)). Note that this term of unknown value appears in (P) as well. Using the previous estimates, (P) provides an estimate for this term up to a constant, namely P0 + log Ec + PL(di — e(0j)), i = 1,... , N. From Assumption A, we can exploit the first moment condition on the error term ec j, namely E[exp(eci)\Zi] = 1. Using the estimates from the previous two steps and the terms 0, — e(0,) estimated up to a constant, the first moment condition on exp(ecl) provides an estimate for the error terms ec * , i — 1,... , N. Note that we can test whether the random shocks on demand and cost are independent. For instance, we can compute a simple correlation coefficient using the e^s and eC is to assess the degree of correlation between the two. The term E c is by definition E[exp(/3ye( jA + eci)]. A simple method for estimating Ec consists in averaging the exponential of the estimated terms PyC dt + ec j. The estimated E c is denoted by Ec. It remains to estimate the constant term of the cost function /30, the cost of public funds A and the parameters of the firms’ type density (r, 7 ). This is the purpose of the fourth and final step of our estimation method. These parameters appear in (P’). The three previous steps provide estimates for do, d\ , e ? 2>a , PL, Py, Ec, as well as r]0/ ( l + A ) and r]1/( 1 + A), which could be used in (P’). Thus, we obtain log pi = ^(0*; P0, A , r, 7 ), i — 1,... ,N. As r e 1N+ , we can develop for each value 67 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of r = 1,2,... a Method of Moments estimator based on the first three moments for estimating (/?0, A , 7 ). This gives E [log = E[^(^;/30,A,r,7 )], E[logpj]2 = E['E2(0i; P0, A , r, 7 )], E flogpi]3 = E [ ^ 3( ^ ; / ? 0, A , r , 7 )]. The function A,r, 7 ) does not lead, however, to some moments that are computationally tractable. We propose instead to estimate the first moment f ty(0;/3o,A,r,j)f(0;r, 7)dd, the second moment f \I/2(#;/30, A,r, 7)/( 0;r, and the third moment J T3(0; j30, A , r, 7)/(#; r, 7)dd by simulated moments. Using the importance sampling method, the first moment will be estimated by 1 V 1 y lil(gi.;A)Ar.7) , , e ' (10) where 6 * ^ , is a simulated value for d-i drawn from a density g(-). Likewise, the second and third moments can be estimated by their simulated counterpart. W ithout loss of generality, we can choose g(-) to be an exponential density with parameter equal to 1. This defines a Method of Simulated Moments (MSM) estimator. See Gourieroux and Monfort (1996) for the asymptotic properties of MSM estimators. In particular, the estimator provides consistent estimates for S fixed. We can perform a MSM 68 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. estimator for each value of r = 1 ,2 ,_ __ For each of r, we obtain a triplet of estimates (/30,A, 7 ). We can find different criteria to choose among these different triplets. We can look at the best adjustment for higher moments or compare the value function. 2.4 Empirical Results As detailed in the previous section, the model consists in estimating the parameters of the demand, the cost, the weights for consumer surplus, the disutility of effort and the firms’ types density as well as the cost of public funds. The first step consists in estimating the demand equation (D) involving the variables y , the price p and a vector of exogenous variables Zj. The demand y is defined as the number of passengers. The number of passengers offers an important variability with a variation coefficient larger than 1 and is strongly correlated with the size of the population as expected. Thus, the size of the population is a natural choice for an exogenous variable.44 The demand for public transit can also be affected by the quality of the service offered. The quality of public transit is multidimensional and many aspects are difficult to measure such as the cleanliness of buses . Following our previous reduced form analysis, we consider two indices of quality, the extent 44We could expect that the demand for public transit will vary with the age distribution of the population and with the average revenue. Data are either incomplete or do not offer enough variability across cities to be included in our empirical analysis. 69 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of the network and the number of seats offered per capita. Both variables should have a positive impact on demand as higher values of both make public transit more attractive to commuters. Both variables offer enough variability to be entertained in the estimation with coefficients of variation equal to 1.3 and 0.41, respectively. Since speed indicates city characteristics, many empirical studies on demand for public transportation also include the speed, measured as the average speed of buses. Though this variable offers small variability in our data, we have included it in the model. The results are given in Table 2.5.4 5 The price variable is expected to be endogenous in a demand equation. In our model, the price is determined as a function of the firm’s type, i.e. pi = p(0.j). Following Assumption A, the firm’s type is independent of the error term q . A s such, the price variable should not be endogenous. For security, we perform the estimation with various instruments for the price variable. In particular, we choose three instruments, which are the unemployment rate, the city area and the debt per capita. The Hansen test statistic for the validity of the instruments is equal to 5.439 distributed as a Chi square with 2 degrees of freedom, which leads us to consider these instruments as acceptable. 45Table 5 does not include a time trend as its coefficient is insignificant with a negative sign. 70 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2.5: Demand Parameters Variable Coefficient t-value Constant (d0) 0.939 19.70 Population 1.094 61.39 Network Extent 0.109 7.65 Seat per Capita 0.767 14.57 Speed -0.165 -2.85 Price (d2) -0.359 -3.621 The price elasticity is found to be equal to — 0.359 resulting in a relatively inelas tic demand to price variation. The magnitude of this coefficient is in the [— 0.2, — 0.5] range as surveyed by Oum, Waters and Young (1992). The coefficient for population is positive and strongly significant. We can interpret this coefficient as follows. If we divide the demand equation by the population, we would find a coefficient equal to 0.094. When the population increases by 1%, the demand per capita would increase by about 0.09%. An increase in traffic congestion due to an increase in the popula tion size would increase the demand for public transportation per capita, inducing more people to choose the public transit as their mode of transportation. The extent of the network and the number of seats per capita have both significantly positive coefficients. The demand is more responsive to an increase in the offered capacity per capita than for an increase in the extent of the network. By increasing the extent of the network, the transportation authority probably offers bus service to remote areas or suburbs, where, given the distance involved, commuters tend to prefer to use their own car explaining this relatively low demand elasticity with respect to network extent. Regarding the speed variable, a negative value may be related to 71 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the nature of the city. A low average speed may indicate a compact city, where the demand for public transit is usually high. In contrast, a large average speed may indicate a sprawl city, where the demand for public transit is usually large. Such an argument could explain the obtained negative coefficient for the speed variable. The second and third steps involve the estimation of the price and cost equations. These steps allow us to recover important coefficients such as the parameters of the cost function j3y and j3L, the parameter for effort disutility a as well as the parameters y to recover the weights (3^ The following table provides the former estimates.4 6 Table 2.6: Cost and Effort Disutility Parameters Variable Coefficient Py 0.917 P l 0.752 a 0.496 Detailed information on the price for labor and material inputs can be found in Perrigne (2002). The coefficient (3y is slightly smaller than 1 indicating slightly decreasing returns to scale. Given that y measures the number of passengers and not the offered capacity, we could have expected a smaller value as an increase of 1% in the number of passengers is expected to increase the cost by a smaller 46Note that we do not provide any t-values for the estimated coefficients in Tabies 6 and following. As discussed previously, the multi-step estimator is not the most efficient one and corrections for standard errors should be computed. An alternative method to compute standard errors would be to conduct bootstrap. 72 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. amount than 1%.47 The estimated coefficient 0 L implies a coefficient for material price equal to 0.248 as homogeneity of degree one in input prices has been imposed. The noticeable result of Table 2.6 is the parameter for the disutility of effort. This parameter has been estimated without imposing any constraint. Its positive value is in agreement with theory. The third step allows us to recover the error terms ec in the cost equation. Using the recovered error terms a( i from the demand equation, we can test whether these error terms are independent. We find a correlation coefficient equal to 0.1497 indicating a low correlation. We can think about weather conditions, which can affect both demand and cost.4 8 The second step also provides estimates for the parameters of the weight func tion.4 9 In view of the reduced form analysis conducted in the data section, we have chosen three variables, which will determine the weight on the consumer surplus, namely a dummy for a leftist majority, the proportion of environmentalists in the city council and the amount of debt per capita. The results are given in Table 2.7.5 0 47A similar result is obtained in Perrigne (2002). This result can be explained by the strong correlation between the number of passengers and the capacity offered. 48Since error terms capture omitted variables, it is likely that these variables are not the same for the demand and cost. 49The second step allows us to recover the values of the coefficients r]0 and r]1 up to a multi plicative term, which is a function of the cost of public funds A . The results displayed in Table 7 have been obtained by dividing the estimated coefficients by 1 + A with A has been obtained in the final and fourth step of our estimation. 50W e u se th e p r o p o r tio n o f e n v iro n m e n ta lists , i.e. ta k in g a v a lu e b e tw e e n 0 a n d 1. F o r th e d e b t p e r c a p ita , w e c o n sid e r a v a lu e e x p re sse d in 1 0 ,0 0 0 F F . 73 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 2.7: Parameters for the Weight Function Variable Coefficient Constant Term (rj0) Leftist Majority Proportion of Environmentalists Debt per Capita 0.9792 0.3439 1.2913 -0.0747 The constant term tells us that a city with a rightist majority, no environmen talist in its city council and no debt will consider a weight for the consumer surplus slightly less than one, which almost corresponds to the case of the benevolent util itarian maximizer. If this rightist city incurs some debt, the weight on consumer surplus will decrease with the amount of debt per capita to attain values smaller than one. Similarly, a city with a leftist majority, no environmentalist and no debt will consider a weight equal to 1.3231. This weight will increase with the presence of environmentalists. These results are interesting as we could have expected more partisan views from politicians. For instance, the difference between a leftist and a rightist majority is just equal to 0.3439, which may look still quite small, everything else being equal. Subsidizing public transportation can be viewed as redistributing revenue among the population. Though a rightist majority does not care as much, both parties care about redistribution and consumers when deciding the pricing for public transit. The coefficient for environmentalists confirms their successful influence on politicians when deciding the pricing of public transit as displayed in the reduced form analysis. The city financial constraint is a non negligible factor in determining the weight. The amount of debt per capita is on average 4,632FF. 74 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. For a city with an average debt per capita, the weight will decrease by the amount of 0.0346 everything else being equal, suggesting a relatively low impact of finan cial constraints in their decisions of subsidizing public transit. The following table displays some summary statistics for the weights. Table 2.8: Summary Statistics for Consumer Surplus Weights Mean STD Minimum Maximum N Weight (Full Sample) 1.1300 0.1825 0.8862 1.4216 514 Left with No Green 1.2903 0.0135 1.2448 1.3175 163 Left with Green 1.3507 0.0352 1.2790 1.4216 90 Right with No Green 0.9428 0.0180 0.8862 0.9734 202 Right with Green 0.9891 0.0278 0.9555 1.0451 58 Cities with left majorities tend to put bigger weights on consumer surplus than cities with right majorities. This tendency is accentuated with the presence of environmentalists. Though the values are relatively close to one, we observe some variability across cities with a maximum value equal to 1.4216, which is about 60% more than for the city with the minimum value at 0.8862. It would be interesting to conduct some counterfactual simulations. One of the major political change between the 1983 and the 1989 elections has been the introduction of environmentalists in city councils. The data show that 11 cities with a leftist majority experienced such a change.5 1 Everything else being equal, the estimation results of the model predict that these cities would decrease their price for public transit on average by 20.36% for an average value of debt per capita and for 6.2% of environmentalists in the city 51Note that 24 cities out of 57 did not experience any political change regarding the nature of the council majority and the presence of environmentalists in the city council. 75 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. council at the new election.5 2 We observe a decrease by less than 10% in the data. Note that our simulation considers everything else being equal while the quality of the public transit has improved over the period explaining this difference. We have also simulated the impact of an increase by 1,000FF of the debt per capita. While considering a leftist city with no environmentalist and an average debt per capita, we find that the price for public transit would increase by 2.37% everything else being equal. If the city has (say) 5% of environmentalists in the city council, the increase would be only by 1.43%. Conducting a similar exercise for a city with a rightist majority and no environmentalist, the model predicts an increase by 15.04% of public transit price, while the price would increase by only 7.51% with (say) 5% of environmentalists in the city council. Such a difference can be explained by the different weight on consumer surplus according to these different scenario. The final step of our estimation procedure provides an estimate for the cost of public funds A , the constant term in the cost function (30 and the parameters of firms’ type density r and 7 . The results are given in the following table. Table 2.9: Cost of Public Funds and Efficiency Parameters Parameter Estimate Constant term (0O ) -2.000 Cost of Public Funds 0.360 r 1 7 1.130 52We observe that these 11 cities had on average 6.2% of environmentalists in their city council after the 1989 election. 76 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We find that r = 1 gives the best adjustment for the final step of our estimator.5 3 This corresponds to a density with an important concentration of relatively good types. The estimated value of 7 gives an expected value for firms’ types equal to 0.885 and a variance equal to 0.783. This gives an expected value for 0 — e of the order of 0.5, indicating that firms’ labor is on average half efficient since we have assumed that firms’ effort and type affect labor efficiency. When considering a simple example of an employee working on an average 40 hours per week, his efficient labor is equivalent on average to only 20 hours of work. The cost of public funds is slightly larger than the value estimated for developed countries. Given that public transit is subsidized through a tax on firms, we can expect that raising additional taxes to subsidize public transit is quite costly to the local economy. In this respect, a value larger than 0.3 is not really surprising. 2.5 Conclusion The objective to this chapter is to study the role of politics in regulation. The data, spanning nine years from 1985 through 1993, provides information on regu latory contracts in the transportation industry in 57 French cities. We introduce a non-benevolent regulator to analyze incentive regulatory contracts. We specify an econometric model and estimate it in several steps. We find that regulatory policies 53This value has been chosen as it provides the lowest value for the value function. 77 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. are indeed affected by the political majority and by the presence of ‘Green’ in the city council. In this respect our results contribute to the long standing debate on the role of politics in regulation. We also find evidence suggesting that cities with lower debt per capita have more generous redistributive policies. Moreover, we estimate the shadow cost of public funds. One shortcoming of this analysis, however, is that we introduce the politician’s objective function by assumption rather than deriving it from a model of politics. This is one direction for future research. 78 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 3 Incentive Regulatory Contracts w ith Endogenous Quality Choice: An Empirical Analysis 3.1 Introduction Theory has made great advances in the study of contractual relationships between the regulator and the regulated firm. The study of normative aspects of natural monopoly pricing date back to the work of Dupuit (1844). However, until the early 1980s research in this area widely neglected informational asymmetries and strategic behavior in the process of regulation.5 4 This gap was bridged by the path breaking works of Baron and Myerson (1982), and Laffont and Tirole (1986). Since the work of Baron and Myerson (1982), theoretical analyses of regulatory contracts have witnessed a tremendous growth.5 5 Though there has been a surge of interest in the theoretical study of regulatory contracts, structural estimation of models of regulation has been very limited. 54Informational asymmetry between the regulator and the firm may be of two forms: hidden action that the regulated firm may take, for instance cost reducing effort, and hidden information that the firm may have on its inherent characteristic or ‘type.’ 55For a survey on the development of the “new economics of regulation” see Laffont (1994) and Laffont and Tirole (1993). 79 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The seminal work of Laffont and Tirole (referred to as LT henceforth) provides a framework and a set of analytical tools that facilitates the analysis of regulatory policies which rely on the ex post observation of cost. Laffont and Tirole (1990) extend the basic model to cover pricing and incentives in a multiproduct firm, a special case of which is to include verifiable dimensions of quality as ‘quantities of fictitious outputs.’ The contribution of this paper is twofold. First we extend the LT (1990) model to incorporate stochastic demand and cost functions when the terms of contract specify the price to be charged, the (verifiable) quality of service to be provided, and the financial agreement between the contracting parties.5 6 We then find the resultant optimal contract using general specifications for demand and cost, and specify the transfer function that implements the optimal contract. Second, following Perrigne (2002), we derive the econometric model, and identify and estimate it using data from the French urban transportation industry for the years 1985-1993. We recover estimates of the consumer’s demand function, the firm’s cost function, the distribution of the firms’ ‘type,’ and the shadow cost of public funds. W ith empirical validation, the theoretical model acquires additional credence in the study of social regulatory tools. Though many studies have dealt with theoretical analyses of regulatory con tracts, empirical work has by far neglected the role of informational asymmetries 56 Since the industry we study is more closely represented by the Laffont and Tirole model, in that the two contracting parties agree on a cost sharing rule, we adapt this model to study regulatory contracts between the regulator and the firm. 80 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. when estimating cost functions of regulated firms. Many economists have made use of the standard microeconomic duality theory and estimated cost functions and a system of factor demand functions.5 7 There is, however, a yawning gap between the progress made in the theoretical literature and the empirical findings.5 8 Wolak (1994) was among the first to incorporate the role of informational asymmetries in his estimation of the production function of California water utilities. His model is closer in spirit to the Baron-Myerson (1982) model. He estimates both a model with symmetric information and with asymmetric information and finds that the model with private information provides a better description of the observed levels of cost and output. In a somewhat different vein, assuming that the offered contracts are not optimal, Dalen and Gomez-Lobo (1997) outline the estimation of cost functions in regulated industries characterized by asymmetric information. They show that not accounting for the moral hazard parameter would lead to downward bias in the estimates. Gagnepain and Ivaldi (2002) propose a methodology for estimating in formational asymmetries in the French urban transportation industry for the years 1985-1993. They, however, do not endogenize the choice of the type of contracts which is taken as given and restricted to being either fixed price or cost plus con 57Examples of such work abound. In the urban transportation literature some examples are Viton (1981), Berechman and Guiliano (1984), and Obeng (1984,1994) to name a few. For an exhaustive survey on the early development in the economics of transportation see Winston (1985). 58During the last decade there has been substantial empirical work testing the predictions of contract theory in various fields ranging from sharecropping to managerial pay. See Chiappori and Salanie (2003) for a comprehensive survey of recent work. 81 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. tracts. Perrigne (2002), and Perrigne and Surana (2004), endogenizing the choice of contracts, estimate incentive regulatory models where the regulator offers a menu of contracts which consists of the price to be charged and the transfer that the firm receives. It has, however, been widely recognized and acknowledged that regulation of utilities in many instances is not concerned with prices alone. The objective of the social planner is to provide consumers with ‘reasonable’ services at affordable prices. Hence, a primary concern of the regulatory authority is to ensure that the regulation of prices does not manifest itself in the reduction of quality levels. This underpins the rationale for the requirement of minimum standards of service. As Kahn (1970) has aptly stated, “Price really has no meaning except in terms of an assumed quality of service; price is a ratio, with money in the numerator and some phys ical unit of given or assumed quantity and quality in the denominator. Price regulation alone is economically meaningless. Moreover, the na ture of our dependence on public utility services is typically such that customers may correctly be more interested in the denominator than in the numerator - in the reliability, continuity, and safety of the service than in the price they have to pay.” Governments have, thus, adopted an integrated approach towards price and qual ity regulation. Besides playing a leading role in establishing the price structure, the regulator also assumes the responsibility of designing the service features. He, thus, serves as a ‘moderator’ between the consumers and the utility provider in deciding the optimal combination of prices and quality. Unlike previous empirical studies (which account for asymmetric information in the regulatory process), we endoge- 82 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. nize the choice of quality by the regulator allowing him to control service features such as network, coverage, seat capacity offered, and the frequency of service by directly regulating their provision.5 9 Not contracting upon ‘quality’ could lead the operator to severely cut corners on these dimensions of quality whenever marginal gain from serving is less than marginal cost incurred. To avoid such behavior it becomes imperative for the regulator not just to regulate price but also regulate the quality of service. We assume that it is costly to provide ‘quality’ and hence specify a quality- adjusted cost function, i.e., in addition to including a demand-related measure of output, i.e., number of passenger trips in the estimation of cost, we also include a supply-related measure of quality.6 0 We use seatkilometers per unit surface area as a composite measure of quality.6 1 Discussion on the choice of the measure of quality is relegated to the section on data. It must, however, be mentioned here that regulating seatkilometers per unit surface area is equivalent to regulating the number of seatkilometers covered since the area of the city is constant. As cost incurred in the provision of service is determined by seatkilometers and not seatkilometers per 59Closer study of contracts between the Local Transportation Authority and the transit operator in Prance reveals that the terms of contract not only specify the price that is to be charged but also the level of service that must be provided. For instance, the contracts specify the network area that has to be covered and the number of buses that must be operated at specified frequency. See Revolle (1999). 60The reason for doing so is that the use of inputs is systematically related to the supply-related measure of quality and not the demand-related measure of output. 6 1 Seatkilometers is obtained by multiplying the number of seats available by the distance over which the respective seats travel. 83 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. unit area, we include seatkilometers as the measure of supply related quality in cost estimation. This has interesting implications for the estimation of the cost function. In particular, we find significantly lower cost elasticity of demand-related output than that found in Perrigne (2002) and Perrigne and Surana (2004). The intuition behind this is that higher demand is generally accompanied with a more extensive provision of service. Increase in cost is due to both higher demand and greater service coverage and not accounting for seatkilometers in the cost function leads to overestimation of the cost elasticity with respect to demand. Researchers and policy makers often estimate cost functions to draw inferences about economic efficiency and the structure of technology. Moreover, cost functions serve as a guidance tool for regulatory boards which design contracts based on cost benchmarks, and for managers of the operating firms in deciding the budgetary allocation of resources. For regulated industries that are heavily subsidized (like the urban transportation industry that we study), collecting taxes to subsidize the industry is costly, and hence it is crucial to correctly study the cost structure of the industry. Misspecification of the estimated model could lead to undesirable policy designs. While the regulator can control verifiable aspects of quality by directly contract ing upon them, intangible service features such as cleanliness of buses and bus stops, delays in the transportation system, etc. are not amenable to direct control. We hope to pursue the case of non verifiable quality in future research. 84 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Public transit is a powerful instrument of social reform. Hence, government regulation of this sector is justified on many counts. First, through direct control the government can achieve balanced urban development by ensuring a well coordinated network of mass transportation. Second, it can ensure the implementation of socially desired levels of ‘essential’ service by regulating availability and continuity. Third, effective use of public transportation can play a vital role in reducing congestion and protecting the environment. And last but not the least, an important objective of regulating the transportation system is to achieve redistribution of wealth through subsidizing the service.6 2 The rest of the paper is organized as follows: the next section provides a brief overview of the French urban transportation industry, documents the evolution of the regulatory structure in that sector, the classification of different types of quality, and discusses the source of data. Section 3.3 discusses the theoretical model for the case of verifiable quality. Section 3.3.1 describes the basic framework, while section 3.3.2 solves for the optimal regulatory contracts. The empirical methodology for the estimation of the model with verifiable quality is dealt with in section 3.4. Section 3.4.1 discusses the empirical specification. The stochastic assumptions are outlined in section 3.4.2, while sections 3.4.3 and 3.4.4 respectively discuss the estimation technique and the results. Section 3.5 concludes. The appendices describe the 62See Small and Gomez-Ibanez (1999) for a survey on issues related to regulation of urban transportation. 85 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. French transportation history in greater detail, present the summary statistics of the data, show the implementation of the optimal contracts through a menu of linear schemes, and provide proofs of the results in the text. 3.2 Overview of Industry, Measures of Quality, and Data 3.2.1 French Urban Transportation System To better appreciate the analysis and the timeframe during which the data is used, a brief mention of the French transportation history is in order. On December 30, 1982 the “Loi d’Orientation des Transports Interieurs (LOTI)” or the “General Law on Ground and River Transport” was passed in France. Significant changes were in the offing and the highly centralized public transportation system was to receive the much deserved restructuring. The law laid down the fundamental principles and procedures for the development of urban transportation. It marked the beginning of the decentralization of the transportation system. Until the early 1980s France was known for her “centralizing tradition.” Decisions were taken in Paris and were implemented throughout the country by State appointed officials. The same being true of the transportation sector, the reform of 1982 marked the dawn of a new era. 86 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The responsibility of regulating the transportation system was devolved to lo calized administration. Each local administrative unit created its own Local Trans portation Authority (LTA) which entrusted a private, semi-public, or in a few in stances a public firm with the management and provision of public transportation.6 3 More precisely, the service could be provided in three basic ways: the local authority could provide the service itself through a specially created local public company, the provision of service could be delegated to a private company, or to a semi-public company (in which the public sector was required to own more than 50% of the stock but no more than 80%).64,65 France, thus, created a model of public service delegation with the goal of ensuring greater efficiency in the public transportation system without jeopardizing its social and economic roles: it gave an increasingly large share to the private sector within a contractual framework while preserving the principle of controlled operation.6 6 The local authority and the company in charge of providing the service entered into a formal contract which comprehen sively outlined the roles of the conceding authority and of the operator respectively. 63The LTAs together formed a national association which is called “Groupement des Autorites Responsables des Transports” (GART). 64The physical limit of the LTA is called Perimetre des Transports Urbains or the Urban Trans port Perimeter. Within this limit the service has to be provided under the authority of the LTA. 65In most instances the service was provided by a private company. Paris was and still is an exception, where the LTA (called STP- ‘Syndicat des Transports Parisiens’) and the two major operators, RATP and SNCF, are all controlled by the central government. 66The principle of controlled management was to shield the economy from the adverse effects that could accompany unreined competition in this sector. The negative consequences of unre stricted market openness have been experienced by some towns in Algeria, in Morocco, and in Latin America (for instance in Santiago in Chile). These towns witnessed extreme proliferation of minibuses on profitable lines causing traffic jams and pollution at the expense of remote areas which went neglected. See Barbieux (1999). 87 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The statement of the mission, the fare structure, the level of quality to be supplied (description of schedules, routes, and lines), the terms of financing, the commercial and technical risks, the description of the general monitoring procedure, the length of the contracts and a clause specifying the conditions for terminating or renegoti ating its terms before the end of term of the contract, etc. were explicitly stated in the contract. The length of the contract differed from city to city but it usually lasted about 5 years. However, the contract was subject to annual changes and to the addition of clauses.6 7 The city remained the owner of the infrastructure and the rolling stock; only the provision of service was franchised. The financial commitment of the LTA in almost all cases required it to give the operator a lumpsum transfer and a fraction of the cost overruns based on some average of previous years (incurred in the operation of the bus service).6 8 It is important to mention here that the financial terms of the contract varied from city to city. It is our attem pt here to model reality as closely as possible.6 9 67Every year the operator in question was required to prepare a report accounting for its overall activity and the quality of service provided. Necessary amendments could be made to the contract for the subsequent years if it was deemed fit by the LTA. 68A good account of the financial arrangement between the LTA and the operator can be found in Revolle (1999). 69Our model does not provide a perfect representation of the actual regulatory process in place, but it does provide (as stated by Wolak (1994)) “a reasonable description of the prices and quan tities observed which provides, at best, a stylized model of how these magnitudes are arrived at.” 88 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The years following the passage of the General Law on Ground and River Trans port witnessed the emergence of three large companies as operators, viz., VIA-GTI, CGEA, and Transdev. This was the result of vesting the LTA with the power of choosing its own operator.7 0 This was true until the passage of the “Sapin Law” in 1993. Contracts are now given out after a formal bidding process among the prospective operators. After the enactment of the Sapin Law, Local Transportation Authorities have been found entering into contracts with smaller local companies and with non-French companies in addition to the three large corporations men tioned above.7 1 The idea of auctioning contracts when several firms are potential candidates for realizing a project was put forward by Demsetz (1968); it now finds application in many countries including France. 3.2.2 Verifiable Vs. Non Verifiable Quality Due to its multi dimensional nature, we must exercise caution when defining quality. Quality of service is a reflection of the passenger’s perception of transit performance. As noted by Danaher, Nowlin, Parkinson and Ryus (1999), “transit quality of service - the overall measured or perceived performance of transit service from the passenger’s point of view - is important to all communities. Transit quality 70There were few other smaller companies such as CARIANE, and VERNEY. They, however, operated in a very limited number of cities. 71See Appendix A.1.1. Duthion, Vincent and Ziv (1999) provide a comprehensive summary of the effects of Sapin Law on competition in the French transportation industry. 89 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of service measures reflect two important aspects of transit service: (1) the degree to which transit service is available to given locations, and (2) the comfort and convenience of service provided to passengers.” It is thus evident that any good measure of quality must account for comfort, convenience, and availability of service. We are guided by these features in our choice of measure of quality as will be discussed in the next subsection. Quality can be classified into two broad categories, verifiable and non-verifiable. It is said to be verifiable when its level can be ex ante (costlessly) included in the contract and can be ascertained ex post by a court; it is non-verifiable when ex post its level cannot be costlessly ascertained. In the transportation industry examples of verifiable quality could include factors such as the hours of operation, specific route structures, etc., while examples of non verifiable quality could include the cleanliness of buses and bus stops, the courtesy of the driver towards the passengers, etc.7 2 Another distinction must be made based on whether the consumer can observe the quality before or after he makes the purchase. When quality is ascertained by consumers before a purchase is made (in this case riding the bus), it is commonly known as a search good.7 3 In the context of public transportation such as bus service 72Laffont and Tirole (1991) consider the case of non verifiable quality and analyze the circum stances under which quality concerns call for low powered incentive schemes. 73 An important and interesting classification of goods based on quality was first done by Nelson (1970). He distinguished betwen ‘search’ goods and ‘experience’ goods. A good is referred to as a ‘search’ good when quality can be ascertained before the purchase is made and it is an ‘experience’ good when the quality is learnt only after the purchase is made. To this classification Darby and Karny (1973) added a third category, namely, the ‘credence’ good, where the nature of quality is rarely learned. 90 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. and railways, quality is generally a search good. See Laffont and Tirole (1991) for classification of goods into search or experience goods and Danaher, Nowlin, Parkinson and Ryus (1999), referred to DNPR henceforth, for definitions of measures of quality of service. 3.2.3 The D ata The data encompassing regulatory contracts for 57 cities in Prance during the years 1985-1993 is compiled from the responses to an annual survey conducted by the Centre d’Etudes sur les Reseaux, les Transports, l’Urbanisme et les Construc tions Publiques (CERTU), a national institution dedicated to the study of urban transports and networks. On the financial side, the data set provides information on operating costs, wages paid to labor, material cost, number of employees, the number of bus drivers, revenue generated from the sale of tickets, and the number of tickets sold. It also provides information on service features such as network size, the number of buses operated and their total seat capacities, and the num ber of seatkilometers covered. Data on city characteristics such as the population size, debt per capita, surface area, and unemployment rate are also available. The summary statistics are presented in Table 3.1. 91 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3.1: Summary Statistics Variable Mean Standard Deviation Total number of passengers (103) 22819.03 25096.84 Busfare (FF) 2.34 0.85 Speed (kilometers per hour) 16.62 2.64 Population (103) 230.64 184.92 Seatkilometers (103) 571256.04 565584.73 Total cost (103 FF) 108106.09 123505.28 Price of labor (103 FF) 156.32 19.19 Price of material (103 FF) 77.26 17.48 Revenue from sale of tickets (103 FF) 50208.41 59942.01 Number of buses 149.71 134.67 Unemployment rate 0.15 0.03 Debt per capita (FF) 4631.60 2126.40 It was assessed in 1990 that the total expenses of the public transport sector in Prance were about 7 billion USD of which about 30% was paid by the users of the service, 32% by private companies (through the “Versement Transport” or the transportation tax)74, 14% by the national government through grants and subsidies, 14% by local governments (through raising loans and self-financing, i.e., use of the LTA’s own capital), and the remaining 10% came from other sources (like advertising and other commercial products). See Duthion et al (1999). In our data we find the average cost per rider to be 4.96 FF while average busfare to be only 2.34 FF implying that the deficits must be covered by subsidy for the transportation service to be financially viable. These figures are expressed in constant 1985 French Franc. Summary statistics are shown in Table 3.2. On average we find 48% of the operating 74More details about this tax can be found in Appendix A. 1.1. 92 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 3.2: Summary Statistics for Cost per Rider and Busfare* Variable Mean Std. Dev. Min. Max. Cost per rider 4.96 1.23 2.91 10.22 Busfare 2.34 0.85 1.00 7.62 Sale/ Cost 0.48 0.13 0.22 1.08 *Note: All values are expressed in constant 1985 FF. costs to be covered by revenue generated from busfares. For all but one city we find the total cost to be higher than the total revenue from busfares. We thus find ample evidence suggesting subsidization of the urban bus service. Given the multi faceted nature of quality, it is very important to carefully choose our measure of quality. As discussed in the previous subsection, important con stituents of a good measure of quality are: passenger capacity, service coverage and transit availability. Passenger C apacity: The passenger or person capacity for a transit route can be defined as “the maximum number of people that can be carried past a given location during a given time period under specified operating conditions without unreasonable delay, hazard, or restriction, and with reasonable certainty,” - (DNPR). Higher capacity implies comfortable rides for passengers as they do not have to stand for long periods of time and can also use their time productively. Moreover, crowded buses slow down transit operation as riders take more time to get on and off the bus. 93 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Service C overage and Transit A vailability: Unlike personal automobiles which have near universal access to locations and which can be used at any desired time, transit service is limited to specific areas and specific hours of operation. Bigger network coverage implies a higher probability of riders finding available service near the origin and destination of their trips thereby enhancing the convenience of rides. Longer hours of service allow customers with greater flexibility in selecting travel times and higher frequency of transit availability implies shorter wait times thereby making rides more convenient. Given the limited availability of data on quality features of bus service, we are restricted to only two measures, namely, seatkilometers per unit surface area of the city and seats per capita. We use the former as our composite measure of quality. Table 3.3: Summary Statistics of Measures of Quality and Cost Efficiency Variable Mean Std. Dev. Min. Max. Skm/Sq. km. surface area (105) 391.17 508.70 13.63 4174.76 Seats per capita 0.06 0.02 0.01 0.12 Cost / seatkilometers 0.19 0.04 0.11 0.30 Unlike seats per capita, it not only accounts for the passenger capacity offered (for comfortable rides), but also the network coverage across the city (for convenience), and the frequency at which buses are made available or transit availability (for shorter wait times). A change in any combination of these alters the seatkilometers 94 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. per unit surface area.7 5 Summary statistics of measures of quality and cost efficiency (as measured by cost per seatkilometer) are presented in Table 3.3. It is interesting to observe if the data exhibit any trend over time. We find that average operating costs per city have gone up by 11.10% over the 9 years Figure 3.1: Trend of Key Economic Variables Over Time p L i b O a V a O 120 115 110 105 Year S' < v a a O .195 .19 .185 .18 Year 0 ) o .0 3 u C O -s a ^ a H D < V % O h a CO 44 42 40 38 36 Year a 03 065 - O s -i < D a a 3 O CO Year 75An alternative interpretation of our measure of quality could be to treat it as verifi able/regulated output. The crucial point to note is that it is controlled by the government and and hence is endogenously determined as a part of the contract. 95 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Figure 3.1: Trend of Key Economic Variables Over Time continued 240 230 2 2 0 210 Year 2.5 2.4 2.3 2.2 2.1 Year ^ .52 " < P 1985 1990 1995 Year despite gains in efficiency with average cost per seatkilometer falling by 8.21%. This has been mainly due to an increase in the quality of service with average seatkilometers per unit surface area rising by 19.46% over the nine years. There has also been an increasing trend in the average provision of seats per capita with an increase of 18.97%. Demand has risen by 11.47% over the years thereby raising costs. It is also interesting to note that the fraction of cost covered by the sale of 96 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. tickets has fluctuated between 0.44 to 0.51 over the nine years of our study. Average busfare has fluctuated between 2.2 FF and 2.5 FF. The trends of these variables over time are shown in Figure 3.1. As a preliminary analysis, we regress logarithm of seatkilometer per unit surface area on exogenous factors such as logarithm of debt per capita, logarithm of the unemployment rate, logarithm of population, logarithm of speed, and time trend.7 6 We also regress logarithm of price on the same set of exogenous variables. The results are presented in Table 3.4.7 7 Table 3.4: Regression Results of log (seatkilometers per unit surface area) and log (price) on a Set of Exogenous Variables* Skm per unit surface area Price Debt per capita 0.28 (0.07) 0.07 (0.03) Unemployment rate -0.06 (0.15) 0.47 (0.05) Population 0.74 (0.06) -0.02 (0.02) Speed -0.04 (0.22) 0.30 (0.09) Time trend 0.02 (0.01) 0.01 (0.01) Constant -0.70 (0.91) 0.41 (0.38) R2 0.29 0.16 *Note: Figures in the parentheses are std. errors of the corresponding coefficients We find that cities that are financially constrained tend to have higher prices. A 1% increase in debt per capita raises prices by 0.07%. Given that public transit is a basic necessity, the quality of service does not suffer in the wake of an increase 76Data on debt per capita are unavailable for the years 1985-1993. We thus use the average over the years 1994 and 1995. 77Regressions with fixed and random effects qualitatively give similar results. Results of these regressions and the Hausman specification test are available upon request. 97 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in debt per capita. High unemployment is suggestive of low economic activity and this leads to a reduction in the tax revenue raised through V.T., and hence planners are forced to raise prices. For a 1% increase in the unemployment rate, prices rise by nearly 0.5%; the effect on quality is insignificant. Cities with bigger population tend to have better service and somewhat lower prices. Populated cities tend to be congested and hence public authorities encourage the use of public transportation by offering better quality of service and lower prices. High speeds indicate a ‘sprawl’ city and given that provision of bus service is more expensive in such cities, we find higher prices being charged to cover costs; the effect on quality is insignificant.7 8 3.3 The M odel 3.3.1 The Framework The regulator entrusts the operator with the provision of bus service. The regu lated firm or the operator has private information about its efficiency which affects its cost. It also makes cost reducing effort unobservable by the regulator. More 78We also added some political variables (such as a dummy for a rightist majority and the proportion of environmentalists in the city council) to these regressions. Data on the political variables are collected from results of the municipal elections. City councils in Prance are renewed every six years through democratic elections. During the period of our study one election was held in 1989. For the years before 1989 we collected data on the previous election which was held in 1983. We find prices to be higher in cities with rightist majorities while the presence of environmentalists on the city council tends to lower prices. Perrigne and Surana (2004) show that cities with leftist majorities and with environmentalists in the city council put higher weight on consumers’ surplus thereby leading them to charge lower prices. It is interesting, however, to note that while politics has a sizeable impact on prices, its effect on quality is insignificant. 98 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. precisely, the firm’s ex post realized cost is given by C — C(q, s, (3 — e,ec), where, P is the firm’s technological parameter or its ‘type,’ e is its cost-reducing effort, q is the quantity of output, s denotes verifiable quality of service, and ec is a random shock on cost. All else being equal, cost is increasing in /3, q, and s and decreasing in e. We further specify the cost function as C = C (f3 — e,q, s) exp (ec) . The regulator does not know either the firm’s type or its effort in the sense that both cannot be disentangled from the random shock on cost even though the cost can be observed ex post by the regulator. The firm’s type, (3, is assumed to be continuously distributed on [/?,/?] with cumulative distribution function F (•) and strictly positive density / (•), where (3 denotes the most efficient firm while /3 denotes the least efficient. The regulator observes the ex post realized cost and knows the distribution of /?. By exerting effort the firm incurs a disutility (expressed in monetary terms) of ip (e). We assume, as is customary in the literature, ip' > 0, ip" > 0, ip'" > 0, and satisfies ip (0) = 0.7 9 We assume a stochastic demand given by q = q (p, s) exp (£d), where, p is the price and £(i is a random shock on demand. The deterministic part of demand, <?(•), is the expectation of demand with respect to £ < * . The accounting convention adopted is the same as in LT (1990), i.e., the ex post observed cost is reimbursed by the regulator to the firm and the revenue, denoted by R (q, s), generated from the sale of tickets is paid directly to the regulator. 79 Assuming ip> 0 excludes stochastic mechanisms. 99 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In addition to cost reimbursement, the regulator compensates the firm by a net monetary transfer, t. We consider contracts of the form [(p, s, t) : p G R +, s € E R+, t € R]. The benev olent regulator’s objective is to determine contracts which maximize social welfare. From the revelation principle we know that the regulator can restrict attention to a truthful direct revelation mechanism, which induces the firm to report its type P 0 ) , s 0 ) , t { p , C ) where truthfully. A direct revelation mechanism takes the form is the reported type and C is the ex post realized cost. In other words, the regula tor announces a menu of price, quality, and transfer schedules and the firm chooses the appropriate contract. The transfer specifies a cost reimbursement rule and is a function of the ex post observed realized cost. The random shocks, and ec > are realized ex post, i.e., after the contract is signed. Upon accepting the terms of the contract the firm is required to satisfy the demand, q = q(p0), s 0 ) ) exp (c( j ) . We consider the case of full commitment by the regulator, i.e., contracts are duplicated in the case of repeated interactions with the same firm. The timing of the game is as follows: In the first period the firm learns about its efficiency, (3. In the second period the regulator offers the contract p($), s(f3),t(/3, C) to the firm. In the third period the firm accepts or rejects the contract.8 0 In the fourth period demand and cost are realized and in the fifth period 80We assume that rejection of contract gives the regulator a large negative payoff. This ensures that there is no ‘shutdown’ of any type. 100 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the contract is executed. The regulated firm’s objective is to maximize its expected utility given by (11) Throughout the paper, expectations are taken with respect to ec and q . The gross payment made by the regulator to the firm is t+ C . Any amount of this payment that is not collected as revenue from the sale of tickets is raised through distortionary taxation which inflicts a disutility of $(1+A) on taxpayers for every $1 raised in taxes, where A is the shadow cost of public funds. The expected utility of a firm of type (3 that reports $ and puts effort e is given The incentive compatibility (IC) constraint of the firm, which ensures that the firm of type (3 chooses the appropriate contract is given by, by Eip((3,{3,e) = E t ( p ,C ( q 0 ,£ d ) ,s 0 ) ,0 - e,ec)) - ip(e) . (12) )) max Eip (f3, (3, e) > max Eip{(3, ft, e) for all (3,{3 e [/?,/?]. 101 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The firm maximizes its utility by choosing the effort level and its reported type. Let us define e(ft\ft) as the effort that a firm of type (3 exerts when it reports its type as ft, i.e., e(ft\ft) = axgmaxEip(ft,ft, e). (13) e Using (13), the IC constraint can be written as, E(p(/3,/3,e(/3\{3)) > E<p(/3,ft,e0\/3)), for all ft, ft € [ft_, ,d] . Besides satisfying (IC), the regulator must ensure that the firm supplies the service, for which the firm’s expected utility from accepting the contract must be at least as large as its outside opportunity level of utility, i.e., E (U ) > 0. The reservation utility is assumed to be independent of type and is normalized to zero. This defines the individual rationality (IR) or the participation constraint. 3.3.2 The optim al regulatory allocation The utilitarian regulator’s objective is to maximize the social welfare function, which is a sum of the consumers’ welfare and the producer’s surplus. The social value attached with the production of output q with quality s is given by S (q, s). Therefore, consumers’ net welfare is given by S (q, s) — pq — (1 + A ) (t + C — pq) , where (t + C — pq) is raised through distortionary taxation. The producer’s welfare is given by the firm’s rent, U (ft). The regulator maximizes the social welfare subject to the individual rationality and the incentive compatibility constraints. 102 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Truth-telling requires that Eip((3,f3,e(j31/?)) is maximized at [} = (3, the first- order condition for which is given by Ecp((3, ft, e(j3\(3)) = 0 . Let U ((3) = ip (/3, (3) denote firm /?’s rent. Using the first-order condition for truth-telling and (13) we get the incentive compatibility condition as (14) The derivation of the incentive compatibility condition is shown in Appendix A.2. It shows that truth-telling requires leaving the regulated firm a rent which decreases with 0 at the rate -i//(e (/3)), where e (/?) is the effort that the regulator wishes to implement. From (14), U(-) is a decreasing function of 0, so the IR inequalities can be replaced by E [U (/3)] > 0. In addition, since social welfare is decreasing in U(-), the IR can be simply written as E [U (/?)] = 0. In addition to transfer, price, and quality, we treat e (0) as the regulator’s choice variable. Since, however, the regulator cannot observe effort, it is necessary to make sure that the firm has proper incentives to exert the optimal e (/?). In Appendix A.3 we show how through the design of an appropriate transfer function the regulator achieves this goal. The regulator’s maximization problem can now be specified as: 103 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. , , w T f , uni / & ( S W ) ] f ( 0 ) d 0 P subject to : E 0 ( 0 ) = -i> '(e (P )) E [U (3)] = 0 (15) (16) (17) where, S W = S ( q ( p ( 0 ) , s ( 0), c d ) , s (0)) - p (0) q (p (0) , s ( 0 ) , Ed ) + (1 + A ) p ( 0 ) q ( p ( 0 ) , s ( 0), e d) - (1 + A ) B { i ( 0 , C ( q ( p ( 0 ) , s ( 0 ) , e i ) , s ( 0 ) , 0 - e , e c)) + C ( q ( p ( f i ) , s ( 0 ) ,e d) , s ( 0 ) , 0 - e,ec)} + U . Note that the consumers’ surplus can be written as f q (p ,s (0 ), e.i) dp, where V Pmax is the price above which demand is zero. Using (11) to substitute out t(-) from SW, the Hamiltonian for the principal’s problem can be written as, H = E P m a x J {P) , £d) dp + (1 + A ) p (p) q (p (/3) ,s ( p ) , ed) - (1 + A ) {C (q (p (p) ,s(p ), ed) , s { P ) , P - e , £ c) + ^ (e (/?))} - XU + t*{P) (e (P))) i f(P ) where p (f3) is the Pontryagin multiplier. Taking p(P) ,s ( p ) , and e {(3) as the control 104 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. variables and E [U (/?)] as the state variable, the first-order conditions are given by Hp = E [-q (p ,s ,e d) + (1 + A ) {q (p ,s,sd) + p 9q{P gpSd)} — C l + a C (g ’S’f f ~ e ’£c) d(l( P ’S < Ed) 1 __ A V ' / d q d p J 1 P m a x Hs = E[§-s f q (p, s, £d) d p + { 1 + A ) p d^p’a'£^ (1 + A){ d s d C { q , s , P - e , e c ) d q ( p , s , e d ) _|_ d C ( q , s , p - e , £ c ) } l _ d s d s T = o, (18) H. = £ [ { - (1 + A) + j (e)}f (P) - i," (e) p (0)] = 0, —He(u) = A/ (P) = fl(0) . (19) (20) (21) Integrating (21) and using the transversality condition, fi(ff) = 0, we get p {(3) = AF ((3).81 P rop osition 1 The solution (p* ((3), s* (/3) , e* (/?)) of the regulator’s optimization problem satisfies the following conditions: „ l d C ( g , s , 0 - e , e c ) 9 g ( p , s , e d ) d q Op 9q(p,s) A g ( p ,s ) 1 1+A p 9q(p,s) dp • V (= m = (e) \ P m a x (1 + A) p - ^ + - § - s f q(p,s,ed) d p - (1 + X)E d C ( q ,s ,f 3 —e ,£ c ) 0 q ( p ,s ,£ d ) d q d s (22) (23) (24) 0C (q,s,(3-e,E c) d s = 0 . 81 T h e b o u n d a ry /? = (3 is u n c o n s tra in e d a n d h en ce th e tr a n s v e rs a lity c o n d itio n a t ft = ft is g iv en b y M /3) = 0. 105 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Equation (22) is the modified Ramsey pricing equation when demand and cost are random. Equation (23) illustrates the classic rent-efficiency trade-off whereby the effort of all but the most efficient type is distorted downwards away from the first best in order to limit the rent of the efficient types.82,83 The intuition is that the higher the effort levels, the higher the rents of asymmetric information (see IC). Rents are costly for the regulator due to the non negative shadow cost of public funds and hence effort levels are distorted away from the first best levels in an attempt to decrease rents. Equation (24) shows that the social marginal benefit from the provision of an additional unit of quality must equal its social marginal cost. We will later find the conditions under which the second-order condition for the (IC) is satisfied and verify that the optimal e {(3) can indeed be implemented through a proper choice of transfer. See Appendix A.3. We will use the equations (22), (23), and (24) together with the demand and cost equations to identify and estimate the model. The estimation technique consists of a multi-step GMM procedure and method of simulated moments. P P . 82From the incentive compatibility condition f ip' (e (5)) d,6 = — J U (6)d6 = U ((3), i.e., the rent _ P P of type (3 depends on the effort of types ((3,(3\. The least efficient type receives zero rent (from IR). 83In the case of symmetric information the first best (FB) level of effort could be obtained, i.e., ip'{eFB) = — and the regulator would leave no rent for any firm, i.e., t = ip(eFB) or U{fi) = 0 for any [3. 106 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 3.4 Empirical M ethodology 3.4.1 Empirical Specification The econometric model follows from the first-order conditions of the optimization problem. In addition, since we adopt a parametric approach to estimation, we spec ify functional forms for the demand and cost functions.8 4 We assume the following functional form for demand: exp (d0) zd dP d2So3 exp (ed) if p < pm ax 0 if p > pmax, where zd represents a vector of exogenous variables that affect demand, d\ is a parameter vector associated with zd, p is the price of a bus ride, d> 2 is the price elasticity of demand, s0 is the number of seatkilometers per unit surface area of the city, and d3 is demand elasticity with respect to quality.85,86 Total number of passenger rides represents demand.8 7 The price is defined as the average busfare per rider, i.e., it is obtained by dividing the total revenue generated from busfares 84Perris:ne and Vuong (2004) study the nonparametric identification of the Laffont and Tirole (1986) model. 85We do not consider price of an alternative mode of transport as the bus service had no close substitutes during the period of our study (1985-1993). This is true to date with a few exceptions in cities which have a well developed subway system. 86We assume a maximum price so that consumer surplus is a well defined notion. 87 “Generally, the appropriate definition, which incorporates the spatial nature of transportation, is the movement of a commodity or passenger from a specific origin to a specific destination over a particular time period; in other words, a commodity or passenger trip”-Winston (1985). Ideally we would have liked to account for passenger miles but data are unavailable. 107 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. by the total number of passengers. It must be mentioned here that busfare is not uniform in France (as in most countries) for all passengers as discounts are generally given to students, the elderly, and the disabled. The vector of exogenous variables Zd includes speed and population size. Since demand is strongly correlated with the population size, the choice of population as a determinant of demand is quite natural. Speed, on the other hand, may serve as a measure of whether the city is “compact” or “sprawl.” We hypothesize that demand is also determined by the quality of service provided. As a measure of quality that affects demand, we use seatkilometers per unit surface area of the city (denoted by sq = s/a), where s and a respectively denote seatkilometers and the surface area of the city. This measure of quality reflects not only the network coverage but is also indicative of the frequency of service availability and the comfort of the ride through provision of more seats. Thus, we would expect demand to be positively related with this measure of quality. We assume providing higher quality of service entails higher costs. While we conjecture that demand is influenced by s/a, we assume seatkilometers (denoted by s) affect cost. The correlation coefficient between operating cost and seatkilometers is 0.97, while it is 0.51 between the operating cost and seatkilometers per unit surface area. As expected, cost is much more responsive to variations in seatkilometers. Regulating seatkilometers per unit surface area is equivalent to regulating the total number of seatkilometers since the area of the city or conurbation is constant. To this end, we solve the regulator’s problem such that he controls the price and the 108 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. total number of seatkilometers provided. This is reflected in the solution of the optimization problem. See (26), (27), and (28). As mentioned before, there is two dimensional asymmetric information in the model. The first stems from the firm’s unknown ‘type’ and the second is caused by the cost-reducing ‘effort’ that the manager of the firm exerts.8 8 Following Gagnepain and Ivaldi (2002), we assume that the informational asymmetry regarding the ‘type’ of the firm is captured by differences in labor efficiency that affect cost.8 9 This choice rests on the fact that labor costs constitute about 60% of operating costs.9 0 The assumption is justified on the grounds of the firm being better informed than the regulator about the efficiency of the workers that it hires. Moreover, the firms can expend effort in monitoring and in training its workers to improve their efficiency. The effort e embodies the hidden action that the operator takes. We define the relationship between efficient and observed labor as L a — LE exp (j3 — e). where La denotes observed or actual labor, LE denotes efficient labor, and exp (ft — e) measures the informational asymmetries (we assume {3 — e > 0). It must be borne 88Since the development of frontier models for the study of efficiency there have been a number of studies estimating the efficiency of bus transportation systems. Among others, see Viton (1997), and Fazioli et al (1993). Most of the studies show that there are inefficiencies in the sector of public transportation. Assuming that contracts are determined exogenously, Dalen and Gomez- Lobo (2003) show that higher (lower) powered contracts lead to lower (higher) cost inefficiencies. 89The efficiency of the bus drivers determines how fuel-effective the buses are, how much wear and tear the capital undergoes and so on and so forth. Labor efficiency, thus, significantly affects the firm’s operating costs. Lan and Kuo (2003) find that aberrant behavior on the bus drivers’ part causes extra fuel consumption and mechanical abrading. "This assumption blends well with the theory since the manager of the firm usually does not have control over labor efficiency. 109 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in mind that labor share in total cost is given by PiLa = PiLe exp (j3 — e) = PiL e (where pi = pi exp(/3 — e)). We thus define the operating cost function of the firm as: C = exp (a0) qaysa'p%lp?1 exp [at (/? - e)] exp (ec) , (25) where pt is the wage rate, and prn is the price of material.9 1 The price of labor, pi, is obtained by dividing the total labor cost by the number of full-time employees working for the operator. Since fuel, maintenance, and repairs, the main constituents of material costs, are mainly used up by buses, the price of material, pm, is obtained by dividing the total material cost by the number of buses. We impose homogeneity of degree one in input prices (cc/ + a m = 1). The rolling stock and the infrastructure are owned by the LTA and hence do not enter the operating cost function. Since the cost function accounts for the quality supplied, we can distinguish between cost elasticity with respect to a demand related measure of output (total number of passengers), and cost elasticity with respect to a supply related measure of service feature (number of seatkilometers). Incorporating seatkilometers into the cost function has important implications for the estimate of cost elasticity with re 91 To keep the estimation tractable we assume a rather simple form for the cost function. Consid ering a more flexible form such as the translog function complicates the first order conditions of the optimization problem and hence the estimation procedure. The assumed cost function suffices for the purpose of estimation while taking into account informational asymmetries and incorporating the fact that provision of quality is costly, the two main features of our study. 110 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. spect to output. This will be discussed at length in the section on results. We use the number of riders as our demand related measure of output since it reflects “the economic motive for providing the service, namely the carrying of passengers” (Berechman and Giuliano (1985)). In the literature, what we have called supply related measure of service has often been called the “intermediate output” while the passengers have been called the “final output.” See Berechman (1993). It is worth emphasizing that our measure of quality can alternatively be interpreted as regulated output. The important distinction is that unlike demand, it is determined endogenously in the system with its level being enforceably included into the con tract. We specify the disutility of effort as ip (e) = exp (ae) — 1, with a > 0. Since the gamma family of distributions allows enough flexibility, it is chosen to represent the firms’ type, i.e., / (/3; r, 7 ) = 7 (7/3)r_1 exp (— 7/3) /Y (r), where r G N + is the shape parameter, and I /7 is the scale parameter. Substituting these parametric forms in (22), (23), and (24), after algebra we get, 1 6 = A K + a t (1 + d2) (log pi - log Pm) + (1 + d2) log pm + a y (d0 + di log zd - d3 log a) + {(1 + d2) + a yd3} log s + a t (1 + d2) /3 {1 - d2 (a, - l)}log ( l + j ( D ) (26) 111 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. logp = A Pi K ' + a ( o n log b logp,, Pm + {a (ay - 1) - ai} (d0 + 4 log zd - d3 log a) + {a (as + ocyd3 - 4 ) - atd3} log s + a i ^ + log^l+ (1 + A)/(/3)J (27) [a, + {ay - 1)4] log s = - [log [(1 + A ) (as + ayd3) exp (a0)] + (ay - l ) d 0 + log Ec] - on logpt - am logpn {Oiy - 1) 4 log zd + (ay - 1) 4 log a 4 1 d. + log (1 + A) 4 4 + 1 ayd2 log p - a i {(3 - e ), p d2+ l + 4 + 1 3 „<i2+l i^max (28) where, K = (1 + 4 ) a Q + {(1 + 4 ) — 0*3/4} log o * i + (1 + 4 ) log Ec + { 4 (a„ - 1) - 1} log a + a „ 4 {log a y - log ( l + ^(ITa) ) } , K ' = a 0a + a log Ec + {a + a/) log - (a + on) log ( l + d2(i+A)) + ai (log o l log ck/) , A = Oi + Oil + 4 (o * — C K Q ! ^ + Q !/) , Ec = E [exp + £c)] • 112 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. It should be mentioned that we do not obtain a closed form solution of the endogenous variables, p, e, and s, in terms of the exogenous variables. We have data on 513 contracts and index by V the iXh contract between the Local Transportation Authority (LTA) and the bus operator. The econometric model is defined by the following four equations. These equations are used to identify the parameters of the model. Q i = S exp (d0) 4 lp f 2 4 3 exp (edi) if p < pm ax 0 if p > Pmax j (D) Ci = exp (a0) Qi'J^isPmiPu exp [a, (ft - eft exp (eci) , ( C ) log Pi = log a y + a 0 — log I 1 + A + log Ec + a t (Pi — e*) + ai logpH d,2 (1 + A) + a m log Pmi + (ay - 1) log qi + a s log su (P ) log Pi = K' + a (a t log ^ + logp™^) + {a (ay - 1) - a ft (d0 + ft log zdi) + { - a (ayd3 - d3) + a/dft log + {a (as + a yd3 - d3) - a z d3} log s* +Cp ( a f t + log ( 1 + A a F(/3. (1 -ha) m. i f ) ) ] i n 113 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for i = 1, N. Equation (P ) is obtained by substituting for C * and in (22). Equation (P') is obtained by solving the first-order conditions of the optimization problem. 3.4.2 Assum ptions on the Error Terms In order to identify and estimate the model we make assumptions on the error terms. There are three error terms in the model: the ex ante unknown random shocks on demand and cost, e(i and ec respectively, and the regulated firm’s ‘type’ P. Following Perrigne (2002), we make the following assumptions, where Z is a vector of exogenous variables. (Al). The random shock on demand, Ed, satisfies E [exp (cd) \Z] — 1. (A2). The random shock on cost, ec, satisfies E [exp (ec) \Z] = 1. (A3). Conditional on Z, firm’s type, /3, is independent of Ed- The first two assumptions follow naturally from the specification of the demand and the cost functions. We do not make any assumption on the distributions of the errors beyond their respective first conditional moments.9 2 Assumption (A3) is consistent with the model in that the firm’s type is its inherent characteristic and is known to it before the realization of the shock on demand and hence is independent of it. 92 Another possibility could have been to log linearize the cost and demand equations and assume the respective conditional means of the errors to be zero. However, this would complicate the solution to the optimization problem and hence the estimation. 114 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The difficulty in estimating this model is twofold. First, the ‘type’ of the firm that is considered as a term of firm’s unobserved heterogeneity is not observed, and second, the firm’s effort is also not observed. To this end, we need to consider an estimation strategy which addresses these issues. Perrigne (2002) develops such an estimation technique. Our estimation extends this method while endogenizing the choice of quality by the regulator. 3.4.3 Identification and Estim ation The parameters of the model are identified from equations (D ), (C ), (P ), and (P'). We use (Al) to identify do,di,d2, and d$ from (D ). Using (A3) and (Al), equation (P1 ) is used to identify a ,a i,a m,a y, and a s. Quality is endogenous in (P'). However, identification follows given the fact that quality is a control variable of the regulator, and hence does not depend on the random shock on demand which is not known ex ante (it does depend on the first moment of Ed though). Using (P ) to eliminate the unobserved (/3{ — e*) term from (C), assumption (A2) allows us to identify A, with a , cq, a m, a y, and a s being already identified. In the final step we identify the constant term «o, and the parameters of the gamma distribution (r, 7 ). Then, (P1 ) can be written as logp = T {(5^ a 0. r, 7 ). Since the distribution of logp is known, we can identify oq, r- and 7 . 115 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We now outline the estimation procedure. Equations (D ), (C), and (P ) specify a system of simultaneous equations that will be estimated sequentially. We use (.Al) to estimate the parameters of the demand equation. In particular, we use the condi tional moment condition, E [exp ( e *:) \Z,} = 1, to specify the unconditional moment condition, E [g (Zi) (exp ( e * ;) — 1)] = 0, where g (Zi) denotes a vector function of exogenous instruments Zi.93 For simplicity we use the identity function for g (•). Hence, the unconditional moment condition can be written as, E [Zf exp (~d0) z ^ 'p '^ S o * - l}] = 0. (29) The above orthogonality condition suggests nonlinear generalized method of mo ments (GMM) as an estimation strategy.9 4 We use two-step GMM to get estimates of do, di, d,2, and d^. In addition, this step allows us to obtain the estimated residuals £di‘ Typically, demand and supply equations are simultaneously estimated. How ever, in regulated environments it has been argued by economists that independent estimation of the demand function is justified given that prices are determined by a regulatory process and given that contracts are generally signed for long terms. 93Following Chamberlain (1987) the vector function of instruments g {Zi) could be chosen optimally. 94An advantage of the GMM estimation technique over the alternative ML estimation is that GMM is based only on the moment conditions and does not presuppose the knowledge of the full density of the error term, i.e., the form of the likelihood function. See Hansen (1982) for GMM estimation. 116 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Therefore, prices cannot be regarded as set by a simultaneous system of demand and supply equations.9 5 Any possible simultaneity between demand and the service factors is avoided here due to two reasons. First, the service factors can be treated as recursively determined, that is the network coverage is not flexible in relation to the quantity demanded at least in the short run. Necessary adjustments in service parameters are only possible with a lag. Second, the long term nature of the con tracts, which also specify service features, and the fixed costs involved in adjusting the service levels avoid any possible simultaneity and potential identification prob lems. Moreover, given that prices and quality are set by the regulator, they are independent of ed which is unknown ex ante, thereby circumventing any possible endogeneity. The second step involves estimating the price equation. The error in (P') is denoted by 7 ] (fa) = (af3t + log ^1 + (P + 5^ Told ) ) ' Since we do not make any assumptions on the moments of rj (/?), we use (A3) and (Al) to derive the following conditional moment conditions, E log Pi - fa-/ log - f log pmi - a{ay A 1 } ^ (di log zdi) a(aydz—d2,)—aidz A log di - a(a3+Qvdf dz) aid3 log Si | (exp (edi) - 1) \Z{ = 0 , 95Balestra (1967) puts forth a similar agrument for the natural gas industry justifying an inde pendent estimation of demand. 117 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. where the parameters of the demand equation and £di can be replaced by the esti mates obtained from the first step. As before, we specify unconditional moments and use generalized method of moments as the estimation strategy. The unconditional moments are given by, E Zi | log Pi - fail log ^7 - f log Pmi ~ a(a?jA 1 ) ai {dl log zdi) _ dz ^ i a y- i y - a ^ l o g ^ _ a(as+ avd^ d s ) - a ld3 ^ ^ | ( e x p ^ ^ = 0 . The vector Z is chosen such that it is independent of £d. The parameters can be consistently estimated given assumptions (Al) and (A3).9 6 Using (26) and (27) to substitute for log p and e in (28), we find that E (log s\Z) is a complex function of the exogenous variables. Hence the linear projection involving log s as the dependent variable will contain not only log pi, logpm, log z( u , and log a as explanatory variables; other functions of the exogenous variables will appear as well. This has important implications for the choice of instruments. In addition to using linear terms of the logarithm of exogenous variables we can use higher order terms of the logarithm of exogenous variables as instruments as w ell. These nonlinear terms are uncorrelated with ed but are partially correlated with log s and hence can serve as instruments.9 7 We regress log s on a second-order polynomial (with a full set of interactions) of the 96log s(f3) does not depend on the realized value of ed' , only on the first moment of ed. 97See Wooldridge (2002) for a discussion on simultaneous equations models and the choice of instruments. 118 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. logarithms of exogenous variables. The results are presented in Table 3.5. We find Table 3.5: Regression Results of Ins on Second Order Polynomial of Logarithms of Exogenous Variables In s Coefficient P-value In Pi 0.790 0.706 In pm 0.968 0.306 Inzdi 3.225 0.000 In zd2 5.659 0.000 In a -2.136 0.000 (In p;)2 -0.605 0.091 (In Pm)2 -0.068 0.776 (ln^di)2 -0.075 0.076 (In zd2)2 -2.655 0.000 (In a)2 -0.146 0.000 In p/In pm 1.522 0.020 In pi In zdl -0.656 0.061 In pi In zd2 -0.458 0.628 In pi In a -0.123 0.676 In pm In Zdi -0.379 0.014 In pm In Z d2 0.654 0.199 In Pm In a 0.262 0.039 In zd i In zd 2 -1.044 0.000 In zd 2 In a 0.190 0.212 In zd i In a 0.165 0.000 constant 2.516 0.092 R 2 0.892 that log s is partially correlated with (log z^ )2 , (log a)2 , log pi log pm, logpm log zdx, log pm log a, log Zd\ log Zd2, and log zdi log a, where zdx is population and zd2 is speed. Regressing log s on a third-order polynomial of logarithms of exogenous variables does not significantly improve R2 over the second-order polynomial. Estimating the above moment conditions by GMM, we obtain estimates for a y, a s, an, and a. We use the standard J statistic to test for overidentifying restrictions. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The estimate of the shadow cost of public funds, A , is obtained from the third step of the estimation procedure. Log-linearizing the cost equation we get, log Ci = a 0 + a y log qi + a s log s{ + an log pH + a m log pmi + an (& - e*) + eci. Using (P ) to recover (/^ — e,:) and (A2), we derive the following unconditional moment condition, The only parameters remaining to be estimated are «o, 7 , and r. This is accom plished in the fourth and last step of the estimation procedure. Substituting the parameters estimated from the previous steps into equation (P'), we obtain logp where £ = , , ayEx- — . Once again we use GMM to estimate the above moment 1 + (1 + A )d 2 conditions and obtain an estimate for £. We also obtain the estimated residuals ec . , ;, i = 1, ..., N. Next we compute Ec by averaging the exponential of the estimated terms (£di&y + id) > and obtain the estimate of the shadow cost of public funds, A , from £ by substituting a y, o ? 2, and Ec. 120 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. as a function of q-q, r, and 7 , i.e. log p = T ((3p a 0, r , 7 ). For each value of r the first two moments are given by, E(logpi)2 = E [^ 2 (/3i; a 0,7)] • Since (3 is unobserved and since no closed form solution exists for these moments, we use the method of simulated moments.9 8 Furthermore, we use the technique function, h (•), whose density is non-zero over the support of (5. The two moments can thus be written as, bution with parameter 1. The method of moments, used to estimate tt0 and 7 , is a technique for constructing estimators of the parameters that is based on matching the sample moments with the corresponding distribution moments. For each value of r we obtain a pair (bo> 7 ). Different criteria can be used to find the best pair. We 98See McFadden (1989). of importance sampling. We take random draws of f3 from an arbitrary density where, (3is is a simulated value for We choose h (•) to be an exponential distri- 121 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. can choose the pair that minimizes the distance between the moments of simulated data and moments of true data or look at best adjustment for higher moments. 3.4.4 Results This section presents the estimates of the demand and cost parameters, the shadow cost of public funds, and the parameters of the firm’s type distribution. Estimation results of the demand equation are presented in Table 3.6. The coefficients have the expected sign and all coefficients but the speed coefficient are significant. Table 3.6. Estimates of Demand Parameters Variable Coefficient Std. Error Constant (d0) 0.530 0.070 Price -0.602 0.054 Extent of Network 0.110 0.040 Size of Population 1.211 0.030 Average Speed 0.141 0.091 We find the demand for public transportation to be relatively inelastic to price variation with a price elasticity of -0.602. An increase in the busfare does not induce many customers to revise choices in favor of alternative modes of transportation due to high transfer costs involved in making the switch. As expected, the measure of quality positively affects demand. For a 1% increase in seatkilometers per unit area of the city, the demand goes up by 0.11%. By offering a better network coverage, the city provides more frequent service or access to remote areas or longer service hours or any combination thereof. This increases the convenience of travel thereby 122 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. inducing a larger fraction of the population to avail of the bus service. We find that a 1% increase in population increases demand for public transportation by more than 1 %. Alternatively it can be interpreted as a 1% increase in population causes an increase of 0.21% in demand per capita. This suggests that with increasing congestion a larger section of the population prefers to use public transit. Intuitively, the effect of speed on demand can go in either direction. On the one hand, higher speed means people reach their destinations faster and hence we would expect the demand to increase. On the other hand, higher speed indicates a “sprawl” city and hence we would expect the demand to be lower. In our case the coefficient of speed is positive but insignificant at the 5% significance level. As mentioned above, the second step of estimation strategy provides estimates for a y,a s,ai, and a. As discussed, we use (log(speed))2, (log(area))2, log(p;) *log(pm), log (pm) * log (popu), logipm) * log (area), log (popu) * log (speed), and log (popu) * log (area) as instruments for quality. We use Hansen’ s J statistic to test for overi dentifying restrictions. The number of degrees of freedom of the J test equals the number of restrictions to be tested. The overidentifying restrictions are not rejected at 5% significance level. Results are presented in Table 3.7. All the coefficients have the expected sign." The cost elasticity with respect to wage is 0.48.1 0 0 Since we "Standard errors (not shown in the table) computed without taking into account the multistep nature of the estimation suggest that the coefficients are significant at the 5% significance level. The same is true of the results presented in Table 6. 100The labor share in total cost is on average 59%. 123 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. impose homogeneity of degree one in input prices, this implies a cost elasticity with respect to material price of 0.52. The estimates of cost elasticities with respect to Table 3.7: Estimates of Parameters of Cost and the Disutility of Effort Variable Coefficient O il 0.479 O L y 0.627 Q L S 0.912 a 5.010 J statistic 0.0004 < X o.95(6) passenger trips and number of seatkilometers offer interesting insights. All else being equal, one percent change in demand changes total costs by 0.63 percent. This estimate is significantly lower than that found in Perrigne (2002) and Perrigne and Surana (2004). The reason is that when the cost function is not adjusted for quality, any increase in cost due to higher demand that is attributable to greater provision of service is not accounted for in the cost function. This could result in an over estimation of cost elasticity with respect to demand. The increase in cost due to higher demand could mainly be caused by an increased expenditure on maintenance and fuel consumption. We find the cost elasticity with respect to seatkilometers to be 0.91, i.e. increasing the network coverage of the service increases the cost a little less than proportionately. Many studies using supply related measures of output have found constant returns to scale in the bus transportation industry.1 0 1 The value of a is positive as expected. 1 0 1 See Berechman and Giuliano (1985) for a review on economies of scale in the bus transportation sector. Also see Nelson (1972) and Wabe and Coles (1975). 124 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. In the third step we estimate the shadow cost of public funds. Using the esti mated values of E c, a y, and ri2 we get an estimate of A as 0.298, i.e., distortionary taxation inflicts disutility of 1.298 FF on taxpayers in order to collect 1 FF in taxes. The marginal cost of public funds in the U.S. has been estimated by a number of studies to be between 0.09 and 0.56 per dollar of tax revenue.1 0 2 Thus, our estimate of 0.298 for the shadow cost of public funds seems plausible. Next, estimates for a 0, and 7 are obtained for different values of r. We tried out r = 1,...,5. The best match between the simulated sample moments and the empirical moments is provided for r = 3. The results are presented in Table 3.8. The mean of the firms’ type is equal to 0.502 and the standard deviation is equal to 0.290. The density of the firms’ type is positively skewed with the mean being higher than the mode. Using (26) we calculate the expected value of exp(/3 — e) as 1.474, i.e., on average actual or observed labor is about 1.5 times the effective labor. Table 3.8: Estimates of Shadow Cost of Public Funds and the Parameters of the ‘Type’ Distribution Variable Coefficient A 0.298 O L 0 1.231 7 5.974 To assess the impact of change in seatkilometers on the average busfare, using {P') we find that increasing average seatkilometers by 107 (an increase of 1.75% 102See Browning (1976), Stuart (1984), and Ballard, Shoven and Whalley (1985). The marginal cost of public funds have been calculated for other countries as well. See Campbell (1975) for the estimation of marginal excess burben of Canadian commodity taxes, and Hansson and Stuart (1985) for Sweden. 125 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. from the average) and keeping all other variables at their sample averages, busfare Figure 3.2: Price versus Seatkilometers per unit surface area 10 C D CJ 5 0 seatkilometers /area increases by 2.98%. Carrying out a similar exercise for an increase in seatkilome ters per unit surface area we find that increasing average seatkilometers per unit surface area by 106, which corresponds to an increase of 2.56% from average value, and keeping other variables at sample averages results in an increase in busfare by 4.67%.1 0 3 This is indicative of the fact that governments have limited budgets that can be allocated to the transportation sector and hence an increase in quality will 103The impact of increase in seatkilometers or seatkilometers per unit surface area applies to situations where an external factor not considered in the model (and that also does not have an impact on the equilibrium relationship given by (P1)) changes. An example of such a factor would be an increase in concern for quality. 126 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. be met by an increase in price. Using (P'), busfare is plotted against seatkilometers per unit surface area in the figure above keeping all other variables constant and at their sample average values. Thus, all else being equal, we find that higher quality of service will lead to higher prices. 3.5 Conclusion This paper studies regulatory contracts when there are informational asymmetries between the regulator and the firm, with a particular emphasis on the case where not only price but also quality of service is explicitly included in the contract. We extend the Laffont and Tirole (1990) model to incorporate stochastic demand and cost, derive the optimal incentive contracts with endogenous quality choice, and study its implementation. Following Perrigne (2002), we assume parametric functional forms for demand for transportation, and quality adjusted operating cost for the regulated firm, derive the econometric model, and estimate it using generalized method of moments and method of simulated moments. Data is drawn from the annual survey conducted by CERTU, a French national institution devoted to the study of urban transportation, in 57 French cities for the years 1985-1993. To keep the framework simple, we have assumed “full commitment” on the regulator’s part, where he commits not to exploit the information that he gathers about the firm’s 127 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. type at the end of the first period of the contract. W ithout such an assumption, solution of the optimal contract and its estimation can become extremely complex.1 0 4 As expected, we find demand to be relatively price inelastic and to be significantly affected by the quality of service offered. The cost elasticity with respect to network coverage is close to unity and the cost elasticities with respect to input prices are close to the average factor shares in total cost. Assuming a service quality adjusted cost function we find significantly lower cost elasticity of demand-related output than that found in Perrigne (2002) and Perrigne and Surana (2004). Increase in cost is attributable to both higher demand and greater service provision and not accounting for quality in the cost function leads to overestimation of the cost elasticity with respect to demand. Furthermore, we estimate the distribution of the firms’ type and also get an estimate for the shadow cost of public funds. We find on average actual or observed labor to be about 1.5 times the effective or efficient labor. In this paper we have dealt with the case of verifiable quality. However, in many instances it is difficult to measure and quantify quality. Examples of such measures of quality are fewer delays in the transit service, cleanliness of buses, etc. These dimensions of quality cannot be enforceably specified in contracts as they cannot be verified ex post. In such cases sales can be used as a reflection of the quality of service provided. In another paper, using stochastic demand and cost functions, we 104Laffont and Tirole (1988) study the dynamics of incentive contracts without commitment and analyse the ratchet effect where the firm does not reveal its true type in the first period fearing that the regulator may extract all rent in future periods once he learns the firm’s true type. 128 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. show that through the design of an appropriate contract the regulator can induce the firm to produce the optimal level of non verifiable quality by rewarding or penalizing it based on a quality index. Estimating such a model is an interesting area of future research that we hope to pursue. 129 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Chapter 4 Public G ood Contributions as Signals: R eputation, Trust and the Evolution of Reciprocal Preferences 4.1 Introduction The voluntary provision of non-excludable public goods is one of the most promi nent puzzles of modern economics. A long series of public good experiments indi cate that real-world agents typically contribute to the provision of non-excludable public goods at levels exceeding those predicted by conventional economic models. Similarly, experimental work on the ultimatum game shows that individuals be have non-selfishly in situations where their generosity can be reciprocated.1 0 5 Some economists [e.g., Smith (2003)] find these results troubling enough to call for a fun damental reconstruction of microeconomic theory. In recent years, theorists have tried to interpret these observations by introducing the assumption that some or all economic agents have preferences for reciprocity, leading them to contribute to the provision of public goods if they expect others to do the same.1 0 6 Explaining behavior by introducing preferences that are consistent 105See Ledyard (1995) for a survey of these experimental results. See Giith and Tietz (1990) for a survey on the ultimatum game. 106See, e.g., Bolton and Ockenfels (2000). 130 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. with such behavior, however, is a solution of the last resort. Economists would like to be able to explain the existence of such “reciprocator types” rather than simply introduce them by assumption. As Smith (2003, p. 467) stated in his Nobel Lecture, “Technically, the issue is how most productively to model agent ‘types’ by extending game theory so that types are an integral part of its p red ictive content, rather than merely imported as an ex post technical explanation of experimental results.” This paper proposes an evolutionary model that predicts the persistence of re ciprocator types, in order to develop a more satisfactory explanation of why fully rational individuals voluntarily provide themselves with public goods at levels far exceeding those predicted by the conventional, static Cournot-Nash contributions mechanism. We also test our theory and find striking support for it, using a large, micro data set from the United States on trust. Our model not only explains non-opportunistic behavior in experimental data and the voluntary provision of public goods, but also an interesting geographical pattern in the level of trust as measured by survey data. A wealth of evidence [see, e.g., Uslander (2002)] suggests that the countries of Europe, and the states in the United States, which have the harshest climates—i.e. the Scandinavian countries and the Upper Midwestern states—display levels of trust far surpassing those of countries and states with more benign climates. While various ad hoc explanations have been offered to explain these facts, we offer an explanation that is rooted in the same theory we develop to explain trust and voluntary collective action in general. 131 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Our theory stresses the idea that highly observable public good contributions, such as a Midwesterner shoveling snow from the public sidewalk in front of his home, serve as signals of the contributor’s trustworthiness.1 0 7 These signals—particularly when they are highly costly—can separate the more trustworthy from the less trust worthy agents. When the environment is periodically harsh,1 0 8 for example in a climate that has very cold winters, human interdependence increases, and mutual help tends to be more costly and more visible than in milder environments. Shov eling snow, helping to start a neighbor’s car, or helping to push it out of a snow embankment, are more difficult when the temperature is — 20 degrees and the wind is blowing than if the temperature is above zero and there is no wind. Our theoretical model suggests that a subpopulation consisting of what we call “reciprocator types” is more likely to be evolutionarily stable in such environments than in milder en vironments. Through a reputational mechanism, a “nucleus” of reciprocator types, whose existence is fostered in harsh natural environments, may induce the remain ing, “opportunistic” types to mimic their behavior in order to maintain reputations 107See Katz and Rosenberg (forthcoming) for evidence that volunteering is widely perceived to be correlated with “pro-social” personality traits that are valued in the workplace. They quote, for example, a guide to young job-seekers which advises, “ Volunteering experience is very important to anyone looking for a job ... It shows that you are someone who cares about your community and that you are willing to spend your own time to help others.” 108We emphasize that Northern climates are periodically harsh, in order to contrast them to some tropical communities that occasionally suffer natural calamities like earthquakes and hurricanes, but do not endure such events on a regular basis. Our model would be less applicable to such communities, in which the probability of being called upon to contribute to the provision of public goods is presumably lower. 132 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. for trustworthiness. Thus we expect unusually high levels of trust and collective action in such environments. The starting point of our theory is the pioneering contribution of Kreps et al. (1982). Kreps et al. proved that if one of two rational players assigns a small probability to the proposition that his or her opponent is an irrational “tit-for-tat” (TFT) type, or alternatively, if both players assign a small probability to the propo sition that their opponent has a taste for reciprocity (preferring joint cooperation to exploiting his or her opponent), then, in a finitely repeated Prisoner’s Dilemma (PD) game, cooperation is an equilibrium outcome for at least some of the stages of the game. The assumed “seed” of uncertainty leads the player (incorrectly) sus pected of being an irrational TFT type or being a reciprocator to justify this doubt, by cultivating a reputation for being the suspected type. This carefully preserved reputation induces the opponent to cooperate, at least in the initial stages of the game. Kreps et al. simply introduced the players’ prior beliefs (i.e., the uncertainty of one or both players regarding the opponent’s type) by assumption. This assumption seems arbitrary, in that it requires that one or both players have slightly mistaken priors about their opponents. A natural solution to this problem would be that the two players are drawn from a larger population in which players of the suspected reciprocator type exist. Rational players, who know the population proportions of the various types, use these proportions to develop their prior beliefs regarding their 133 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. opponents’ type. But the question then arises, why would we expect reciprocator types to exist if their lifetime payoffs are lower than those of opportunists? A popular saying has it that “nice guys finish last.” This paper shows that this saying is not always correct—particularly where public good contributions can serve as signals to separate the “nice guys” from those who are not so nice. The present paper develops an “indirect” evolutionary model which provides the conditions under which part of a population of fully rational individuals will, in evo lutionary equilibrium, consist of “reciprocators” in the sense of Kreps et al. (1982) in their “Model 2” (which assumes two-sided uncertainty regarding the opponent’s preferences). We develop and test a theory in which reciprocators, or altruists as we rename them ,1 0 9 survive in a competitive, evolutionary environment where such players must compete with “opportunistic types” who prefer not to honor trust granted them. The presence of altruists in the population induces opportunists, as well, to honor trust, in order to preserve reputations for being altruists. These reputations are required in order to have trading partners in bilateral market trans actions. i°9We use the term “altruist” rather than “reciprocator” for two reasons. First, we assume that altruists always contribute to the provision of the public good, provided only that their net payoffs are non-negative. The fact that these contributions are independent of other agents’ contributions makes the term “altruist” more appropriate. Second, altruism is the commonly used term in the socio-biological literature for any behavior that benefits the player’s fellow agents at the expense of the player’s own fitness. 134 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. While our primary purpose is to explain the voluntary provision of public goods by rational agents, we introduce these bilateral transactions in order to model agents’ private incentive to contribute to the provision of public goods. As Olson [(1983), p. 15] observed, public goods are usually provided in a larger social context in which agents repeatedly transact with each other in the supply of private goods. By free-riding, agents may lose reputations for trustworthiness that are valuable in their private-good transactions.1 1 0 We formally model this insight in the context of a repeated game. Standard evolutionary models are generally regarded as differing fundamentally from conventional economic theory, in that individuals are not usually assumed to be rational. Instead, individuals are assumed to play pre-programmed strategies such as simply cooperating or defecting all the time, or (more realistically) rewarding cooperation and punishing defection by some strategy such as TFT. The early results of Axelrod (1984), which popularized the idea that TFT is a winning strategy in an evolutionary environment, thus explaining cooperation, have been shown to be rather fragile. For example, a pure-defect “mutant” can easily invade an all-TFT population if the PD interactions are finitely repeated. If, on the other hand, PD 110Tullock (1985) developed a similar idea, suggesting that businessmen contribute to charities and other public good activities in order to maintain reputations for trustworthiness in their business transactions. A related strand of literature studies gift-giving as signaling [Camerer (1988); see also Kranton (1996), and Carmichael and MacLeod (1997)]. 135 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. interactions are infinitely repeated, then there is no pure strategy that is evolution- arily stable [Boyd and Lorberbaum (1987)]. We propose to use a somewhat less conventional type of evolutionary model in order to endogenize the prior beliefs of rational players regarding their opponents’ type. We employ what has been called the indirect evolutionary approach, in which all players are assumed to be rational, and the evolutionary mechanism determines the population mixture of players with differing preferences.1 1 1 In other words, agent types are defined not by the agents’ strategies but by their preferences. Agents choose strategies to maximize expected payoff, just as in standard economic theory. Thus, unlike standard evolutionary models in the tradition of Axelrod (1984), we do not depart from the traditional assumption of maximizing behavior. Readers who are used to regarding the evolutionary and the conventional eco nomic paradigms as diametrically opposed may wonder what is gained by this marriage of the two approaches. The answer is simply that by combining these approaches, we can explain why players rationally expect their opponents to be cooperative or altruistic, rather than simply introducing these expectations by as sumption, as in standard models employing the idea of reputation. Thus we obtain an empirically testable theory of rational cooperation, rather than one that can predict any outcome by the introduction of appropriate prior beliefs. 111Early contributions to this literature include Prank (1987, 1988), Giith and Yaari (1992), and Giith (1995). 136 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. A basic result of the literature [e.g., Giith (1995)] is that unless agents observe some signal of their opponents’ type, preferences that do not reflect the “objective” or materialistic payoffs of the agent—for example, the altruistic type’s preferences of the present model—eventually will be driven to extinction. Altruistic types can survive only if agents exhibit a signal correlated, however weakly, with their type.1 1 2 The question then arises [see Samuelson (2001)]: why cannot opportunistic types also “emit,” a signal that they are altruistic types, thus making the signal uninfor mative? The present model provides an answer to this question: if a public good is provided voluntarily in parallel to the repeated market trust game, contributions to the provision of the public good can signal that the contributor is an altruist. If these contributions are sufficiently costly, opportunists will not find it optimal to contribute, despite the resulting damage to their reputations. Thus the coupling of the market trust game with the public good contribution game generates a “comple mentarity” between the two games. On the one hand, the public good game enables the altruists to survive, despite their material disadvantage in the repeated market trust game. On the other hand, as noted earlier, agents’ incentive to maintain their reputations in the repeated market trust game is what makes it optimal for them to contribute to the provision of the public good. 112Models that incorporate such signals include Frank (1987), Giith (1995), and Guttman (2003). Frank (1988) presents evidence that such signals are, in fact, emitted by humans. 137 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. When applying the theory of repeated games, the analyst must decide whether to model the game as finitely or infinitely repeated. The present model assumes that the relevant games are infinitely repeated, with a positive probability that each current stage is the last stage of the agent’s career in his or her community. Since we assume that information of agents’ past behavior is known only to members of the community in which he or she is located, an (exogenous) move to another community is equivalent to the (unexpected) end of the game for the agent in question. This is because the agent’s current-period cheating in the repeated market trust game is, by assumption, only imperfectly observable even in his or her own community, so that if the probability of moving is high enough, it becomes optimal to cheat. If the probability of moving is low enough, agents behave as they would in a standard infinitely repeated game with a low discount rate. We prefer the infinite-horizon assumption to the alternative assumption of a finitely repeated game with a commonly known endpoint, since we view the latter assumption as considerably less plausible than the former assumption.1 1 3 In order to provide a rationale for obtaining only a small number of equilibria, however, we follow Tirole (1996) in assuming the existence of only three (behaviorally distinct) player types: (a) “altruistic” types, who are similar to Tirole’s “honest” types, (b) permanent opportunistic types, similar to Tirole’s opportunistic types, and (c) 113For models using the fin ite-horizon assumption that are in other respects somewhat similar to the present model, see Guttman (2001a, 2001b, 2003). 138 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. transient opportunistic types, similar to Tirole’s “dishonest” types.1 1 4 Thus we avoid the unreality of the finite-horizon assumption without having the model lead to a large number of equilibria as in the Folk Theorem, but at the cost of introducing some added complexity. The remainder of this paper is organized as follows. Section 4.2 introduces the two player types, “opportunist” and “altruist,” and sets out the other assumptions of the model. Section 4.3 shows how, for a given prior belief regarding the opponent’s type, one obtains a Perfect Bayesian Equilibrium (PBE) in which rational players may voluntarily provide themselves with a non-excludable public good. Section 4.4 endogenizes these prior beliefs in an indirect evolutionary framework, by comparing the equilibrium expected payoffs of the altruists and opportunists. It is shown that the existence of a public good that needs to be supplied voluntarily by the community, enhances the evolutionary stability of the altruistic type. Section 4.5 tests the basic predictions of the model using U.S. data on trust. Section 4.6 offers concluding observations. 114Tirole [(1996), p. 6] similarly mentions the possibility of characterizing dishonest types as “transient agents who [therefore] do not care about their reputation.” If (as in the present model) there is no direct punishment of cheating, he assumes that “[hjonest agents have a strong distaste for and never engage in corrupt practices.” Unlike Tirole, we endogenize the proportions of the three types, and therefore can explain the existence of “honest” (equivalently, altruistic) types, rather than simply assuming the existence of such agents, as is standard in models of reputation. 139 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.2 Assum ptions 4.2.1 Market transaction trust games In each period of their “careers,” agents are randomly matched to play two-stage “trust games,” which can be viewed as market transactions, in which the costs of using the legal system to enforce the terms of the contract are prohibitive. In this game, the first mover (the buyer) decides whether to buy a good from the second mover (the seller). The decision of the buyer to buy will be called “trusting” the seller. If the buyer decides not to trust (abstains from buying), the game ends, and the payoffs of both players are zero. If the buyer trusts, then the seller can honor this trust by producing a good of an agreed-upon level of quality, entailing an expenditure of effort which carries a cost of e. If the seller honors the trust given him or her, both players benefit from this trade. The buyer’s net payoff (consumer’s surplus) is unity, and the seller’s net payoff is 1 — e. Alternatively, the seller can fail to honor the trust by expending a smaller amount of effort. For simplicity of exposition, the cost to the seller of this smaller amount of effort is zero (only the high effort level is irksome), but the product produced will have a probability of 9 of being defective. The seller’s effort level is his or her private information. If a defective good is in fact produced, the buyer’s net payoff is — a < 0. In this case, the seller still receives the payment transferred by the buyer, which he or she values at 1. 140 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. An opportunistic seller’s optimal move in a one-shot game of this type, if trusted, is to exert the lower level of effort, thus receiving a payoff of 1 rather than 1 — e. The buyer’s expected payoff of purchasing from an agent who exerts the low level of effort (cheats) is 1 — 9 — ad, since there still is a probability of 1 — 9 that the good will be high-quality, giving a payoff of unity to the buyer. We assume that 6 > 1/(1 + a), implying that 1 — 0 — aO < 0. It is therefore optimal for the buyer not to trust. Thus, in a one-shot game, no trust is granted, and the payoffs of both players are zero. Figure 4.1 shows the one-shot version of the market transaction game. Figure 4.1: Market Transaction Trust Game Buyer Trust Not trust Seller Buyer: 0 Seller: 0 Honor trust Cheat Buyer: 1 - 0 - a 0 Buyer: 1 Seller: 1 Seller: 1-e In the empirical literature on trust [see, e.g., Alesina and La Ferrara (2002)], one of the variables that has received particular attention is population mobility. 141 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Given that information on agents’ past behavior is often quite complete within a community but does not “flow” with nearly the same ease from one community to the next, agents in a geographically immobile community will have relatively long time horizons. Information on cheating in past periods will have consequences for a cheating agent as long as he or she remains in the community. If, however, the agent moves to another community, he or she can (to a first approximation) start “afresh” building a reputation for trustworthiness. Obviously, if moving were costless, cheating agents would move each time period. In reality, however, there are various monetary and non-monetary costs of moving, which make moves suboptimal except where there are particularly attractive gains from moving. We model the effect of this variable by introducing a random, exogenous variable that determines whether an agent stays in his or her community from one period to the next. We assume that agents have infinite time horizons, but in each time period there is a probability 1 — 8 that the agent will exogenously leave the community.1 1 5 It should be emphasized that the act of leaving the community is not a decision variable of the agent; it is “forced” on the agent by exogenous circumstances (e.g., by an unexpected opportunity in another community that is “too good to refuse”). 115The event that the agent “leaves” the community can be interpreted either as a move by the agent to another community that is “forced” by exogenous circumstances (this is the interpretation emphasized in the text) or as the death of the agent. The latter interpretation can be used in order to explain why agents have infinite horizons, but have finite lifespans (we speak of generations of agents in the evolutionary analysis of Section 4, so we need agents who have finite lifespans). We assume, however, that the probability of death is very small relative to the probability of moving, so that our analysis focuses only on the “mobility” interpretation. 142 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Given the exogeneity of the moves that an agent makes over the course of his or her career, the cost of moving plays no role in the analysis and is therefore ignored. We further assume, for simplicity, that an agent learns of the exogenous opportunity that makes moving optimal only very close to the end of his or her last period in the community. Thus the agent’s optimal actions in his or her last period in the community are the same as in previous periods. The probability that an agent will remain in the community in the next time period was denoted above, generically, as 8 . We now introduce a partition of the players into two types, “permanent” and “transient.” Permanent players have a 8 , denoted < 5 i, sufficiently large that it may be optimal for them to honor trust in an infinitely repeated market transactions game. Transient agents have a smaller 8 , denoted < $ 2, such that it is always optimal for them to cheat. The critical 8 satisfying 61 > 8 > 82 will be specified below. A player’s type is his or her private information, but the proportion of permanent players among all agents, p, is known to all agents. We just assumed above a partition of the players into two types, permanent and transient. We now introduce another partition of players along a different dimension, so that there will be four player types in all. The second partition of player types is between “opportunist” and “altruist” players. Opportunistic players maximize their expected material payoffs, i.e., the payoffs shown in Figure 4.1. Altruists “have a conscience” (i.e., cheating gives them a large, negative psychic payoff that makes 143 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. it suboptimal to cheat) and thus always honor trust given them.1 1 6 Therefore, if the buyer knew that the seller was an altruist, he or she would trust the seller, and both players would receive positive payoffs. Here again, however, we assume that the player’s type is private information. Agents know only the proportion of altruists, a. In a given generation of players, the proportion of altruists a is fixed. Agents of a given generation have children, and the proportions of the types change in the younger generation as a function of the relative undiscounted average material payoffs of the various types. We leave unspecified the exact mechanism by which these types change, but the underlying idea is that the higher the agent’s material payoff, the more children he or she is likely to raise, or the more likely that the children will adopt the preferences of the parent,1 1 7 and these children will be raised with equal probability over the agent’s lifetime (not career in the community). Time is discrete, and indexed by t = 0,1,..., oo.1 1 8 The length of a time period is defined n6Since permanent and transient altruists behave identically, we actually have, for the purposes of the analysis, only three player types. As noted in the Introduction, our reputational model is inspired by Tirole (1996) who also assumes three types of players, for the purpose of specifying beliefs off the equilibrium path. 117See Boyd and Richerson (1985) who develop the idea of “cultural evolution,” which assumes that children’s preferences are adapted to those of the more materially successful agents of the previous generation. 118Note that even though agents are assumed to have infinite time horizons, there is no multiplicity of equilibria in this model, unlike “Folk Theorem”-type results that are usually obtained in infinitely repeated games. This is due to the limited number of agent types, which enables a natural, clearcut (but indirect or implicit) punishment mechanism for opportunistic behavior (via the reputational mechanism), without having to assume arbitrary punishment strategies like trigger strategies. We should also note that while the signaling role of public good contributions could be modeled as a one-shot game, we need to model a repeated game in order to explain the evolutionary stability of the altruistic type. 144 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. as the time required for the quality of the product sold by each seller to become common knowledge to the entire community. In each of these periods, each player is randomly matched with two other agents. W ith one of these agents, the player in question plays the role of a buyer in a trust game as outlined above, and with the other agent, he or she plays the role of a seller. We assume that the number of agents in the community, jV, is large enough that the probability of being matched twice with the same seller is negligible. Agents do not know how many stages have elapsed since their fellow members entered the community. However, we assume that if an agent ever sold a low-quality product, this is known from the next period to all members of the community, for the rest of the agent’s career in the community. If the agent (exogenously) moves to another community, the members of the agent’s new community have no information of the agent’s history of play. 4.2.2 Voluntary provision of a public good In addition to playing this trust game, agents decide whether to contribute to the provision of a public good, the benefits of which are enjoyed by all members of the community. Contributions are discrete: either the agent contributes or he or she does not ( “free rides”). The public good, however, is assumed to be “seasonal” in the sense that contributions are not needed in each time period. Instead, once in every T time periods, the public good is provided, if at all. The quantity produced 145 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. of the public good is a linear function of the sum of the contributions. Let / denote the fraction of the members of the community who contribute, and A ; be a positive constant. The cost of contributing is c > k/N, and each agent’s private benefit of the public good equals kf. In order to make it possible for agents to contribute to the provision of the public good, the per-period cost of contributing is bounded by the agent’s net payoff of selling the private good (which is his or her income available to be contributed). If the agent sells a unit of the private good (without cheating) in the trust game he or she plays, his or her payoff will be 1 — e. To make it possible for such any agent to finance his or her contributions to the provision of the public good from payoffs of selling the high-quality private good, we assume that the average, per-period cost of contributing c/T is no higher than 1 — e.1 1 9 If, in equilibrium, no agents are trusted, the net income available to be contributed is zero.1 2 0 If an agent contributes, he or she increases / by 1/N. His or her net payoff of contributing is thus k/N — c < 0. Therefore, in terms of material payoff, it is optimal (in a one-shot game) not to contribute, but rather to take a free ride on the 119We assume that agents’ net payoffs from selling the private good are storeable for T — 1 periods, in order to allow them to save resources for contributing when contributions are needed. 120The agent’ s net payoff of buying the private good is assumed to be a non-saleable consumer’ s surplus, and thus is not useable for contributing to the provision of the public good. To eliminate the possibility of a lack of resources available to purchase the private good, induced by an overly high cost of the public good, we assume, for simplicity, that the agent has an exogenous endowment which can only be spent on buying the private good. If no market transactions take place due to a lack of trust, this endowment is used to produce consumables within the agent’s household. The payoff from this household production is normalized to zero. 146 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. contributions of the other agents. Opportunists care only about their material pay offs, and therefore would optimally free-ride in a one-shot game. Altruists, however, receive a “warm glow” from contributing, which—taken together with the material return of k/N —exceeds the cost c, making it optimal for them to contribute. An agent’s contribution, or lack of one, is known to all members of the community at the end of the relevant time period, and if an agent ever free-rides, this is recalled by all members of the community during the remaining stages of the agent’s career. In order for provision of the public good to be Pareto optimal, we further assume that c < k. Thus the total cost to the community of full provision of the public good ( / — 1), cN, will be less than the sum of the private benefits, kN. We assume that opportunists maximize the discounted sum of their payoffs over the course of his or her career in the community.1 2 1 The solution concept to be em ployed is the Perfect Bayesian Equilibrium (PBE), and we further restrict attention to equilibria that are symmetric and stationary. In the remainder of the paper, the term “equilibrium” means a symmetric, stationary PBE. 1 2 1 Discounting, in this model, derives from the fact that an agent has a probability of 6 < 1 of remaining in the community in his or her next time period. 147 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.3 Analysis of the game: exogenous population mixture In this Section, the proportion of altruists a is treated as an exogenous parameter. In the following Section, this variable will be endogenized by embedding the model in an indirect evolutionary process. We begin by analyzing the game without public good contributions. In Section 3.2, these contributions will be introduced into the game. 4.3.1 The game without public goods Consider an opportunistic agent who is deciding whether to cheat (not honor trust given him or her, by exerting the low level of effort) for the first time in his or her career. We first observe: P rop osition 1. I f a seller ever sells a defective product, he or she will not be trusted, in equilibrium, in all fu rth er stages of his or her career in the com m unity. Proof. There are two types of equilibrium to consider, under the stationar- ity restriction imposed in the previous Section: (a) permanent opportunistic types honor trust given them in all stages of their “careers” in the community; and (b) permanent opportunistic types cheat in all stages of their careers, as do transient opportunists (by definition). In case (a), if a player ever sells a low-quality product (which occurs with probability 6 if he or she cheats), then he or she is revealed to be 148 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. a transient opportunistic type. The expected payoff of buying from such an agent, given that he or she always cheats, is 1 — 6 — ad, which is negative, by assumption. case (b). If a player ever sells a low-quality product, he or she is revealed not to be an altruist. Again, the expected payoff of buying from such an agent, given that he or she always cheats, is 1 — d — ad, which is negative, so that, again, it will not be optimal to trust such an agent. ■ Let us calculate the discounted expected payoff to an opportunistic agent, acting as a seller, of honoring trust at each stage, assuming that the agent is trusted by all buyers with whom the agent is matched. The agent receives a stage payoff of 1 — e. We thus obtain the following expression for the discounted expected payoff of this strategy for the agent’s career in the community: where i = 1 for the permanent type and i = 2 for the transient type. If, on the other hand, the agent cheats (exerts the low level of effort) in all stages of his or her career, then his or her expected payoff will be 1 at each stage t, provided that he or she was not detected as having cheated in the past periods, which has Therefore it will not be optimal to optimal to trust such an agent. Now consider Eir(honor) = — e) (30) 149 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. probability (1 — 0)*.1 2 2 If, on the other hand, he or she was detected as having cheated in the past, his or her payoff is zero in all subsequent stages (by Proposition 1). Thus the discounted expected payoff of this strategy is O O - Eir(cheat) = ^ ^ ^ = l _ w i _ gy (31) t=o *' ' Comparing (30) and (31), we find that it is optimal to honor trust if and only if™ Denote the r.h.s. of (32) as 8. In accordance with our definition of “permanent” and “transient” types, we specify that 8\ > 8 while < 5 2 < 8. Until now, we have assumed that the buyer trusts the seller. Trusting the seller will be rational if altruists and, perhaps, permanent opportunists, are a sufficiently large proportion of all sellers. There are two possibilities to consider, which we will call Type 1 beliefs and Type 2 beliefs, respectively. 122Note that at each stage t, there were t previous stages, since the first stage is indexed 0. 123When honoring trust yields the same expected payoff as cheating, we will assume that the seller does not cheat. 150 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • Type 1 beliefs: Only altruists are expected to honor trust. In this case, the expected payoff of trusting is E ir(tru st) = a + (1 — a) (1 — 6 — ad) = 1 — 9(1 + a) + a 9 (l + a). E n (tru st) will be non-negative if and only if (33) Since we have assumed that 9(1 + a) > 1 (in order for it to be irrational to trust a seller who is known to be opportunistic), the r.h.s. of (33) is positive. • Type 2 beliefs: perm anent opportunists as well as altruists are expected to honor trust. Then the expected payoff of trusting is E n (tru st) = a + (1 — a )p + (1 — a;)(l — p )(l — 9 — a9) = 1 - (1 —p )9 (l + a) + a ( l - p )9 (l + a), which will be non-negative if and only if " - 1 (l-p)0(l + a)' (34) 151 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Note that the r.h.s. of (34) is less than the r.h.s. of (33), since p > 0. If p > 1 — 1/[6>(1 + a)], the r.h.s. of (34) will be zero or negative. In this case, even if a seller is known to be an opportunist, he or she will be trusted, provided that buyers have Type 2 beliefs—i.e., permanent opportunists are expected to honor trust. We now define a parameter, a min, which is the minimum proportion of altruists in the population that will induce buyers to trust sellers in equilibrium, depending on their beliefs, as follows: 1 1 — — ------- if buyers have Type 1 beliefs 9(1 + a) C ^m in — ^ 1 — -------— -------- if buyers have Type 2 beliefs (1 — p )9 (l + a) Provided that a > a min, the equilibrium in which buyers trust sellers, and both altruistic and permanent opportunistic sellers honor trust, is the unique, stationary PBE in the game without public good contributions. To see this, consider the only alternative stationary PBE—in which all opportunistic sellers do not honor trust. Since a > a min, sellers would nevertheless be trusted. Given that sellers are trusted, each permanent opportunistic seller would find it optimal to deviate and honor trust, since (30) is at least equal to (31) for such a seller. Therefore such an equilibrium would not exist. To summarize, P ro p osition 2. In the absence of the opportunity of making contributions to the provision of the public good, buyers will trust sellers in the unique, stationary 152 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. PBE of the game, if and only if a > ctmin. In this equilibrium, both altruists and permanent opportunists honor trust. For the purposes of the evolutionary analysis of Section 4, we must specify the conditions under which agents will rationally have Type 1 or Type 2 beliefs. Consider three cases: • If a < 1 — 1/[(1 — p)6( 1 + a)], sellers will not be trusted even if buyers have Type 2 beliefs. Therefore an opportunistic seller (even a permanent type) will not honor trust, even if the buyer with whom he or she is matched were to deviate and trust the seller.1 2 4 Given that permanent opportunists will not honor trust even if the buyer deviates from the equilibrium and honors trust, we conclude that Type 1 beliefs are rational when a < 1 — 1/[(1 — p )6 {l + a)\. • If a > 1 — 1/[0(1 + a)], sellers will be trusted even if buyers have Type 1 beliefs. Given that all buyers trust the sellers with whom they are matched, permanent opportunistic sellers will rationally honor trust. Thus we conclude that Type 2 beliefs are rational when a > 1 — 1/[0(1 + a)]. • If 1 — 1/[(1 — p)6( 1 + a)] < a < 1 — 1/[0(1 + a)], there are two equilibria: (1) If buyers have Type 1 beliefs, sellers will not be trusted, and therefore even 124The reason is that a permanent opportunistic seller, not expecting to be trusted by other buyers, will not be able to use the above comparison of equations (39) and (40) to “ justify” honoring trust for reputational reasons, since the seller is randomly rematched in each stage of the game with a different buyer, and equations (39) and (40) assume that all buyers trust sellers. Thus the current matching is, in effect, a one-shot game, in which cheating is optimal. 153 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. a permanent opportunistic seller will not honor trust even if the buyer with whom he or she is matched deviates and trusts. In this equilibrium, therefore, Type 1 beliefs are rational, and trust does not emerge. (2) If buyers have Type 2 beliefs, sellers will be trusted, and therefore permanent opportunistic sellers will honor trust. Therefore, Type 2 beliefs are rational in this equilibrium, and trust emerges. There is no obvious selection of which equilibrium will occur. For the sake of concreteness, we will assume that equilibrium (1), the “no-trust” equilibrium, will occur in this case. Recall that Type 2 beliefs are rational when a > 1 — 1/[0(1 + a)], and that, provided that this condition holds, if in addition p > 1 — 1/[0(1 + a)], even known opportunists will be trusted. We thus obtain P ro p o sitio n 3. In the absence of the opportunity of making contributions to the provision of the public good, if both a and p are at least equal to 1 — 1/[0(1 + a)], then all agents will be trusted. 4.3.2 The game with public goods We now introduce a public good, which is produced under the conditions specified in Section 2.2. Recall that altruists always contribute to the provision of a public good, if their net payoff from selling the private good covers the cost of contributing. 154 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. It may be optimal for opportunists to contribute as well, in order to disguise the fact that they are opportunists, and thus acquire the trust of buyers. Recall that the cost of contributing to the provision of the public good (net of the private material benefit of contributing) is c — {k/N), and this contribution is made only once every T periods. For simplicity, let us assume that T = 2, and let c = c — {k/N). Thus an agent’s “contribution stream,” if he or she decides to contribute, would look like a series of pulses, taking the level of c every other period, and zero in the remaining periods. To simplify our computations, we smooth out each two-period segment of this stream by performing a transformation which leaves the discounted present value constant (where discounting is due to the agent’s uncertainty that he or she will still be in the community and selling in the next time period). Denote the present-value-equivalent, per-period net contribution, for agents who honor trust, as chonoTii. Thus, c/lo n o r. (1 + 6 ,t) = c, and therefore c Q l o n o r ’i = T+Ji Similarly, for agents who do not honor trust, and therefore only have a probability of (1 — 0)8i of still selling in the next period, c Cc h e a t,i = x + ^ _ g y 155 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. If an opportunist both contributes and honors trust in all stages, his or her stage payoff will be 1 — C honor,i — e. Therefore the strategy of contributing and honoring trust in private-good transactions will have an expected discounted expected value of 0 0 E n (contribute A honor)i = E < 5 -[l - (e + chonorJ \ (35) t=o 1 (c T C honor^i) 1 — e c l - S i “ (i + $0(i-fc) Suppose instead that the opportunist cheats in each stage t but contributes to the provision of the public good. Then the opportunist’s expected stage payoff will be 1 — ccfieat,i> if the opportunist’s cheating in previous stages is still undetected, which has a probability (1 — 6)*. Thus the opportunist’s discounted expected payoff of cheating but contributing is O O E n(cont. A cheat)i = y ^ S -(l - 6 > )* (1 - ccfaeaM ) (36) t= o 1 Cchea.t,i l - S i ( l - 0 ) 1 c i - Si(i - e) ~ [i + s<(i - 0)][i - Si(i - e)\ 156 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Let A E n i = E n(contribute A honor)i — E n(contribute A cheat)i. As c increases, AEiXi decreases linearly: d(AEm) dc «?(2 - 0)6 < 0. [1 + <5,(1 - 0)][1 - < 5,(1 - 0)](1 + <5,)(1 - < 5 ,) Thus there is a maximum c, above which it is optimal to cheat. Comparing (35) and (36), this maximum c is Note that transient opportunists, even if they contribute to the provision of the public good, will nevertheless cheat in their private-good transactions. This can be verified by noting that the absence of a public good is equivalent to c = 0, and we have d(AEni)/dc < 0. Thus, if it is optimal to cheat when c = 0, it is certainly optimal to cheat when c > 0. In order for it to be rational for a (permanent) opportunist to contribute to the provision of the public good and honor trust in market transactions, however, it is not sufficient that c < c. It is also necessary that either (35) or (36) be non-negative, since the opportunist also has the option of neither selling nor contributing to the provision of the public good, and receiving a payoff of zero. Since E n (contribute A cheat) and E n (contribute A honor) decrease as c increases, there are levels of c large A {6.0 . e[l - < 5,(1 - 0)]}(1 + <5,)[1 + <5,(1 - 0)] «?«(2 - 6) 157 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. enough to make Eir (contribute A cheat) and E n (contribute A honor) negative. The critical c, above which E n(contribute A cheat)i < 0, is c(cheat)i = 1 + < 5 « (1 — 0). (37) Similarly, we can calculate the maximum c, above which contributing and honoring trust yields a negative expected payoff to permanent opportunists (as noted above, this expected payoff is always negative for transient opportunists). Setting the r.h.s. of (35) equal to zero, we obtain, for this maximum c,1 2 5 c(honor)i = (1 — e)(l + < 5 i). Let C m iry: = m m [c(cheat)i,c(honor)i\ and cm a X )i = m ax[c(cheat)i, c(honor)i\. We now consider three cases, for a given type i of opportunists (1 for permanent and 2 for transient): C ase 1: c > cm ax,,;. In this case, no opportunist of type i will optimally con tribute, even if contributing is the only way of acquiring trust. If this holds for both types of opportunist, then all contributors, in equilibrium, will be reciprocators, and thus only reciprocators will be trusted, for any a. Let P i (a ltru ist) denote the poste rior probability that a buyer attaches to the seller being an altruist (after observing 125Here we use 1 for the subscript, since, as noted above, transient opportunists will never honor trust in their private-good transactions, even in the presence of a public good. 158 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. whether the seller contributes to the provision of the public good). In this case, P r(a ltru ist \contribute) = 1, so we have Pr (a ltru ist \contribute) > a m ;„, implying that (by Proposition 2) contributors will be trusted, regardless of the proportion of altruists in the population, a. Case 2: c G (cm ;nj,, cm ax/;]. Setting aside the borderline case c(cheat) = c(honor), we must consider two subcases: • Case 2a: c(cheaf), < c < c(honor). In this subcase, only the altruists and permanent opportunists contribute in equilibrium, and (if c > c(cheat)i) both types honor trust in their market transactions. If c > c(c/ieaf)2, transient opportunists do not contribute in equilibrium, since the strategy of cheating and contributing yields a negative expected payoff. Thus contributors will be trusted, and non-contributors will not be trusted. • Case 2b: c(cheat)i > c > c(honor). In this subcase, only altruists contribute and honor trust in their market transactions, while opportunists of type i con tribute to the provision of the public good and cheat in their market transac tions. In this subcase, then, only Type 1 beliefs are rational. When a > a min, this would be the unique pure strategy equilibrium, since a is high enough to support trust of a randomly drawn agent. When a < a rain, there is a mixed strategy equilibrium, made possible by the added information provided by the observation that the seller has or has 159 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. not contributed to the provision of the public good in previous stages of the game. In such an equilibrium, sellers are trusted with probability less than unity. To see why, note that the optimal decision of a buyer, whether or not to trust the seller, is conditional on the observation that the seller has or has not contributed to the provision of the public good in previous stages of the game. Consider first the possibility that the buyers do not trust sellers even if they have contributed to the provision of the public good. Opportunists, in this case, clearly have no incentive to contribute. An individual seller who contributed, under this assumption, would be identified as an altruist, and therefore should be trusted. Thus a buyer’s strategy of not trusting agents, even if they contribute to the provision of the public good, would be subopti- mal since it ignores the information provided by the observation that a seller contributes. Now suppose that buyers trust sellers who contribute, and do not trust other sellers (since a < a m in). Given that c < c(cheat)i, oppor tunists of type i would contribute in order to be (mis)identified as altruists, and thereby be trusted. But this makes the strategy “trust (with probability 1) sellers who contribute” suboptimal, as well, since the buyers’ belief that sellers who contribute are altruists is inconsistent with opportunistic sellers’ optimal strategies. We conclude that no pure strategy equilibrium exists in this subcase when a < o;m in. In the mixed strategy equilibrium, to be detailed in Appendix A, both permanent and transient opportunists will contribute to 160 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the provision of the public good, but with a probability of less than unity. Buyers will trust sellers with probability less than unity. C ase 3: c < cm iiv;. In this case as in Case 2b, opportunists of type i will rationally contribute to the provision of the public good. As in Case 2b, if a > o min, we have a pure strategy equilibrium in which all opportunists of type i contribute and all contributors are trusted. If, on the other hand, a < o min, there are two mixed strategy equilibria. Details on these equilibria are provided in Appendix A. Equation (37) implies that c(cheat)i > c(cheaf)2, since 6 1 > < S 2. The relative sizes of c(cheat)i and c(honor)i , on the other hand, cannot be determined without specifying the underlying parameters, Si, 9, and e. The following proposition high lights the cases that yield clearcut equilibria, and which are of particular relevance to the analysis of Section 4. P ro p o sitio n 4. (a) If c > c(cheat)i and also c > c(honor) i, then only altruists will contribute to the provision of the public good in equilibrium, and only altruists will be trusted, for any a. (b) If c(cheat)i < c < c(honor)i, only altruists and permanent opportunists will contribute to the provision of the public good in equilib rium, and both will honor trust in their market transactions. Transient opportunists will not contribute and therefore will not be trusted. 161 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.4 Endogenizing the proportion of altruists We now endogenize the proportion of altruists by assuming an evolutionary mechanism which selects for the type that is relatively successful (has a higher average material payoff). In particular, we assume that: A rv — = <f>(Eity - E n 0), (38) where A a /A g is the change in a from one generation to the next, E n r is the average (undiscounted) per-period material payoffs of the altruists, E t t 0 is the average per- period material payoff of the opportunists. We assume that <//(•) > 0 and 0(0) = 0. We assume that this evolutionary process does not operate on p, the proportion of permanent types in the community. This assumption reflects the fact that the (locational) permanency of an agent is determined by a host of variables that are not in the expressions for Eirr and E t t 0. Thus p continues to be treated as an exogenous variable. As buyers, all agents face the same population mixture and behave identically. Thus it is only as sellers that their payoffs can differ, due to their differing strategies depending on their type. The evolutionary analysis is based on the expected payoffs of the various types, based on Propositions 3 and 4. 162 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Since the expected payoffs in (38) are undiscounted average payoffs, we calculate the weighted average payoff of an opportunist who either cheats or honors trust over his or her career in the community. The weights used in calculating these payoffs are the probabilities that a given stage t will be reached in the agent’s career, i.e., 6l, divided by the sum of the weights, ^ = 1/(1 — < 5 ). The net per-period cost of contributing to the public good, C honor or ccheat as appropriate, is equated simply to c/2 for the purpose of this computation, since the discount ing used in calculating c^onor and ccheat , as they are defined above, is irrelevant to computing the undiscounted average payoffs used in (38). Since altruists and perma nent opportunists behave identically if permanent opportunists honor trust, we have E n r — E t t 0(honor). If, on the other hand, (permanent or transient) opportunists cheat in their market transactions, we have O O [Eirr - ETT0(cheat)\i = ( l - e - 0 - (1 - <5/ ^ $(1 - 0)* ( l - (39) © (l e 2) (1 Si) W 1-!) i - <5/1 - e ) t=o 1 1 - <5/1 - 6) — e. When i = 1, the sign of (39) is ambiguous, but when i = 2, (39) is clearly negative, since (32) implies that e > <520/(1 — 5(1 — 0)]- Therefore, the average 163 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. undiscounted payoff of transient opportunists, if they are trusted, is larger than that of their (transient) altruist counterparts. Equation (39) can also be used to assess E n r — Eirffcheat) in the absence of a public good, simply by setting c = 0. Then (39) becomes 6 8 [Enr - E n 0(cheat)]2 |c=o = ^ ^ ^ - e < 0. Therefore, the italicized conclusion of the previous paragraph holds in the absence of a public good, as well: transient opportunists, if trusted, receive higher expected payoffs than their (transient) altruist counterparts. Before stating the evolutionary implications of the model when a public good is provided by the community, let us first ask what would happen if no public good were provided. In this case, Proposition 2 implies that if a < a m in, no sellers will be trusted. Thus the expected payoffs of the two types (altruist and opportunist) will both be zero, and by (38), a will be constant over time. If, in contrast, a > a m jn, permanent opportunists and altruists receive equal expected payoffs, since they be have identically. However, as we noted in the previous paragraph, transient oppor tunists receive higher expected payoffs than their (transient) altruist counterparts. T h u s , t h e o p p o r t u n i s t s a s a g r o u p r e c e i v e h ig h e r a v e r a g e p a y o f f s t h a n t h e a l t r u i s t s . Therefore a will decline over time, by (38), when a > cnm in- 164 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Let us define a critical point as a point a on the segment [0,1] at which the average, undiscounted expected payoffs of the altruists and opportunists are equal. Let us further define an evolutionary equilibrium region (EER) as a subsegment of contiguous critical points on the segment [0, 1], such that for points in the neigh borhood of this region, the difference in average expected undiscounted payoffs will induce a to move into that region, by (38).1 2 6 We thus obtain P ro p o sitio n 5. In the absence of a public good, there is an EER at a E [0, «„,;„). In this evolutionary equilibrium region, no agents are trusted. Now consider the evolutionary dynamics in the presence of a public good. Let us define an evolutionary equilibrium point (EEP) [Martinez Coll and Hirshleifer (1991)] as a point a G [0,1] such that, for any other point a in the neighborhood of n, the evolutionary selection dynamic (38) will lead the proportion of altruists to move in the direction of a over time. Given that transient opportunists (if they are trusted) have higher expected material payoffs than their altruistic counterparts, we can conclude that in the game with public goods, if transient opportunists are not “screened out” by c being larger than c(cheat)2, the relative fitness of the altruists and the opportunists will depend on the relative sizes of the transient and permanent subpopulations. If, on the other hand, c > c(cheat)2, transient opportunists will not contribute to the provision of the public good and therefore will not be trusted. If, in addition, c > c(cheat) 1, 126See Martinez Coll and Hirshleifer (1991) for a discussion of these concepts. 165 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Proposition 4 states that either (a) only altruists will contribute and only they will be trusted, or (b) if c < c(honor)i, permanent opportunists will contribute as well and will honor trust. In case (a), all opportunists will receive zero payoffs since they will not be trusted, and therefore the altruists will have higher material payoffs than the opportunists. In case (b), the same result obtains, since altruists and permanent opportunists behave identically and therefore receive identical payoffs, but the average fitness of all opportunists as a group will be smaller than that of the altruists, because transient opportunists will receive zero payoffs. We conclude that, if c > c(cheat)i ,1 2 7 then a will increase from one generation to the next. We thus obtain P ro p o sitio n 6 . If c > c(cheat)i, there is an EEP at a = 1. Proposition 6 implies that if c is sufficiently large, altruists will survive in evolu tionary equilibrium. Recall that Proposition 4 states that, in the absence of a public good, the proportion of altruists will be too low to support trust, in evolutionary equilibrium. Combining these results, we find theoretical support for the conjecture stated in the Introduction, that when a public good must be voluntarily supplied and the cost of contributing is relatively high, the evolutionary stability of altruists is enhanced, and the level of trust will be relatively high. Since opportunists maximize their expected payoffs, it may be surprising that al truists can receive higher average payoffs than opportunists. The reason underlying 1 2 7 Recall that c(cheat)\ > c{cheat)2. 166 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. this result is that the payoffs relevant to the evolutionary stability of a given type are undiscounted expected payoffs, while opportunists maximize their discounted expected payoffs. The probability that an agent will reach stage t in his or her career in a given community, 5*, enters into the calculation of both undiscounted and discounted expected payoffs (since the undiscounted payoffs must still “weight” high-probability events relatively heavily). But the undiscounted expected payoffs take account of the fact that agents, when moving to another community, start “afresh” and (given that they play stationary strategies) receive the same expected stage payoff as they received in the first period in which they were in their previous community. When opportunists decide, once in each two periods, whether to con tribute to the provision of the public good, they discount the benefit of contributing due to the probability 1 — Si that they will leave the community at the end of the current period,1 2 8 thus failing to reap the full benefit of the investment in their reputation. But such a move, if it occurs, does not affect their undiscounted stage payoffs, which are simply 1 — (c/2) if they cheat in their market transactions and 1 — (c/2) — e if they honor trust. Thus (discounted) expected-payoff-maximizing agents free-ride (and reveal their type) too often in terms of their fitness—their lifetime undiscounted expected payoffs. 128Additionally, if they cheat, there is a probability 8 that their current cheating will be detected before the next period and therefore they will not be trusted (by Proposition 1). 167 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 4.5 Evidence from U.S. survey data on trust Using U.S. survey data on trust, we tested the following two hypotheses deriving from the model: • H y p o th esis 1. Trust will be relatively high where the environment is relatively harsh. As we saw above, Proposition 6 implies that, for a sufficiently large c, the cost of contributing to the provision of the public good, then an all altruist population is evolutionarily stable. As we argued in the Introduction, c is high where the environment is periodically harsh. Mutual help will be more costly, in particular, in relatively cold climates. Thus we measure c by using the average minimum temperature in the month of January in the locality surveyed. The higher is this average minimum temperature, the lower is c. • H y p o th esis 2. Trust will be relatively high where population mobility is rel atively low. In terms of the model, where the population is less mobile, the proportion of permanent agents, p, is relatively high. Proposition 3 states that if both a and p are at least equal to 1 — 1 /[9(1 + a)], then all agents will be trusted. Our measures of trust and other covariates are from the 2000 Social Capital Benchmark Survey, conducted in 41 U.S. communities by the Saguaro Seminar at 168 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. the John F. Kennedy School of Government at Harvard University. Community organizations collected the data. These organizations decided what specific area(s) were to be surveyed and how many interviews were to be conducted. Usually the sur vey area was a county or a set of contiguous counties; in some cases the community sample was a municipality while in some others it was an entire state. The mobility index is obtained by merging the relevant part of the 1990 Census data with the Social Capital Benchmark Survey data. We use the proportion of people who have lived in the same house for the last five years as our measure of (low) mobility. As a measure of the harshness of climate of a particular region we use average low temperature for the month of January. This is extracted from www.weatherbase.- com and is merged with the Survey data. Weatherbase is a source for finding monthly weather records and averages for more than 16,439 cities worldwide. Various measures of trust are available from the Social Capital Benchmark Sur vey. Questions asked range from “How much can you trust people in your neighbor hood?” to whether most people can be trusted, whether policemen, people where one shops, or members of the respondent’s ethnic group can be trusted. We se lected two different measures of trust: (a) trust neighbors, which measures how much a person trusts people in his or her neighborhood, and (b) trustown, which denotes the trust of own ethnic group. These measures of trust were selected in order to focus on trust of individuals who are similar to the respondent in terms of 169 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ethnic background and location of their homes, since ethnic heterogeneity has been shown to be negatively correlated with trust [Alesina and La Ferrara (2002)]. By using trustown (trust of individuals in the agent’s own ethnic group) as a dependent variable, we make it unnecessary to include ethnic heterogeneity as an explanatory variable.1 2 9 As control variables we create seven education dummies, four age-bracket dum mies, six income dummies, a gender dummy (male=l), and ethnicity dummies (Afro- American, Hispanic, Asian). (See Appendix B for definitions of all variables.) It has been found in the empirical literature on trust that experiences of discrimination lead to lower levels of trust; hence we also control for “experience of discrimina tion.” Other variables that we use as controls are the number of close friends that the respondent has, a measure of the respondent’s happiness, a measure of interest in politics and national affairs (political interest), an index of income homogeneity, the population density in the respondent’s zipcode, the mean income of the respon dent’s community, and finally a measure of the respondent’s satisfaction with his or her current financial situation (econ status). For all the variables the observations with missing data were dropped. Of the original data set, there are a total of 19,588 remaining observations. Tables 4.1 and 4.2 present the summary statistics of the variables. 129We do not use a measure of ethnic heterogeneity in our regressions, because this variable is highly correlated with one of our main variables of interest (the climatic variable), creating a multicollinearity problem. 170 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.1 Summary Statistics for Dependent Variables Dependent variable No. of obs. Mean Std. deviation trust neighbors 19588 2.28 0.85 trust own ethnic group 17811 2.20 0.66 Table 4.2. Summary Statistics for Explanatory Variables Explanatory variable Mean Std. deviation low mobility 0.51 0.06 avg. temperature Jan. 22.9 11.82 high school 0.24 0.43 college 0.22 0.41 associate degree 0.11 0.31 grad training 0.04 0.19 bachelors 0.18 0.38 grad degree 0.14 0.35 20K < income < 30K 0.14 0.35 30K < income < 50K 0.26 0.44 50K < income < 75K 0.21 0.41 75K < income < 100K 0.12 0.32 income >100K 0.13 0.33 30 < age < 50 0.46 0.50 50 < age < 70 0.23 0.42 age > 70 0.08 0.27 male 0.42 0.49 experienced discrimination 0.42 0.86 Afro-American 0.12 0.32 Hispanic 0.07 0.26 Asian 0.02 0.15 friends 3.34 1.05 econstat 1.09 0.64 happy 2.34 0.59 political interest 2.92 0.95 income homogeneity 0.19 0.01 mean community income 3.18 0.37 population density 3.22 5.50 Since both measures of trust that we use are ordered dependent variables, we use ordered probit as the estimation procedure. In order to check whether the results are Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Trust = < robust to different specifications of trust, we run the same regressions with different specifications of trust as the dependent variable. The trust measures that we use are defined as follows: 0 if trust not at all 1 if trust “a little” 2 if trust “some” 3 if trust “a lot” Let the threshold levels corresponding to the different categories of trust be Ki, k.2 , and A s in standard ordered probit analysis, application of maximum likelihood yields estimates of the threshold parameters and the coefficient vector of the covariates. In a well behaved model we would expect to see < k 2 < K 3.1 3 0 Table 4.3 presents our results. Regarding our two hypotheses, we find: • Trust is higher when the climate is more harsh (colder), supporting Hypothesis 1. • Trust is higher when population mobility is lower, supporting Hypothesis 2, when the dependent variable is trust neighbors. The mobility index is, how ever, insignificant for trust of own ethnic group. 130This ordering was obtained in our regressions, for both dependent variables. In order to save space, we omit the estimates of these parameters in Table 4.3. 172 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Table 4.3: Ordered Probit Regressions of Trust on Selected Variables trust of: neighbors own ethnic group community level variables avg. temperature (Jan.) -0.004* (0.00) -0.01* (0.00) (low) mobility 0.36* (0.15) -0.05 (-0.31) income homogeneity 3.01* (0.78) 3.36* (0.82) population density -0.02* (0.00) 1 o o t—1 * (0.00) mean community income 0.11* (0.02) 0.04 (0.03) individual level variables high school 0.23* (0.04) 0.22* (0.04) college 0.26* (0.04) 0.21* (0.04) assoc, degree 0.32* (0.04) 0.26* (0.04) grad training 0.31* (0.06) 0.25* (0.06) bachelors 0.43* (0.04) 0.31* (0.04) grad degree 0.46* (0.04) 0.30* (0.04) 20K < income < 30K 0.12* (0.03) -0.02 (0.03) 30K < income < 50K 0.22* (0.03) -0.01 (0.03) 50K < income < 75K 0.33* (0.03) 0.02 (0.03) 75K < income < 100K 0.40* (0.04) -0.01 (0.04) income > 100K 0.40* (0.04) -0.08* (0.04) 30 < age < 50 0.34* (0.02) 0.13* (0.02) 50 < age < 70 0.57* (0.03) 0.28* (0.03) age > 70 0.78* (0.04) 0.53* (0.04) male -0.08* (0.02) -0.02 (0.02) experienced discrim. -0.14* (0.01) 1 O CO * (0.01) Afro-American -0.56* (0.03) -0.36* (0.03) Hispanic -0.44* (0.03) -0.20* (0.04) Asian -0.25* (0.05) -0.02 (0.06) friends 0.09* (0.01) 0.09* (0.01) econ status 0.13* (0.01) 0.11* (0.02) happy 0.28* (0.02) 0.31* (0.02) political interest 0.12* (0.01) 0.09* (0.01) Pseudo R 2 0.13 0.08 Log-likelihood -18851.27 -15766.27 Note: * denotes significant at 95%. Standard errors in parentheses. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. We also estimated the same regressions with random effects and obtained qual itatively similar results.1 3 1 As mentioned above, we controlled for income homogeneity of the different com munities. We used the Herfindahl index (i.e. the sum of the squares of the propor tions of people in the different income categories). The index is increasing in income homogeneity. We find higher levels of trust in more homogenous societies. Alesina and La Ferrara (2002) obtain similar results. We find that trust increases with the level of education. Trust is also higher for higher income respondents, when the dependent variable is trust neighbors. These results are consistent with Putnam ’s (2000) observation, “In virtually all societies ‘have-nots’ are less trusting than ‘haves,’ probably because haves are treated by others with more honesty and respect.” Similarly, when the dependent variable is trust neighbors, respondents located in higher income communities (irrespective of their own incomes) are more trusting. This can be understood in terms of the “reciprocity of trust” phenomenon observed by Glaeser, et al. (2000). Perhaps for the same reason, we also find in all the regressions that minority group members are less trusting than Whites. The coefficients of “happy,” number of close friends, and satisfaction with economic status are also positive and significant. Similarly, 131In other regressions, we included a dummy variable indicating that the respondent intended to stay in his or her community, and a dummy for home ownership. Inclusion of these variables (both of which received positive, significant coefficients) did not appreciably change the results for the remaining variables. In the regressions reported here, we dropped these variables due to a concern that they are endogenous to trust. 174 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. people who have had experiences of discrimination are significantly less trusting. This result supports the reciprocity argument of Alesina and La Ferrara (2002), where they suggest that “one trusts others if he is used to being treated fairly by his fellow men.” We find that trust increases with age. This is a well-known cohort effect: older cohorts exhibit more trust than newer cohorts; i.e., trust has been decreasing over time in the U.S. [see Alesina and La Ferrara (2002); Putnam (2000)]. We also find that trust is also higher for people with an active interest in politics. Finally, respondents in relatively urban communities (as measured by relatively high popu lation density) display less trust, presumably a consequence of the more anonymous interpersonal relations that are dominant in urban communities. 4.6 Concluding remarks In this paper, we have developed a theory of the voluntary provision of public goods that combines the insights of two, distinct traditions: (a) the conventional game-theoretic literature of repeated games with incomplete information, in which players develop reputations which stem initially from (assumed) uncertainty of one player regarding the other player’s type, and (b) the indirect evolutionary literature, in which player types are defined not by wired-in strategies, but rather by their preferences, and agents choose strategies to maximize their expected payoffs, as 175 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. in standard economic theory. The preferences we seek to explain by using the indirect evolutionary approach are preferences for reciprocity, which have recently been recognized as crucial to the explanation of the voluntary upholding of joint agreements and the private provision of public goods. Our analysis illustrates the complementarity of these two theoretical traditions. The evolutionary stability of agents with non-opportunistic preferences is easier to understand when the interaction of agents is modeled as a repeated game in which private-good and public-good decisions are made simultaneously. Conversely, the repeated-game reputational mechanism does not need to rely on ad hoc assumptions regarding players’ prior beliefs when these beliefs are endogenized by making them correspond to the population proportions of the various types, generated by an evolutionary process. Our theory utilizes the fact that the voluntary provision of public goods usually takes place in a broader social context, in which players also buy and sell private goods. Contributing to the provision of public goods, in such contexts, serves as a signal of the trustworthiness of the contributor, a signal that is important in obtaining trust in private good interactions. Thus, in our model, opportunistic agents contribute to the provision of public goods in order to preserve reputations as altruistic agents. But these reputations could not be established if there were no true altruists in the population, whose existence creates uncertainty by agents as to their partner’s type. Our model explains why these true altruists can persist in the 176 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. population, despite their vulnerability to exploitation by opportunists. This stability of a subpopulation of altruists derives from their rationality—which distinguishes them from the TFT types of traditional evolutionary models. A large body of laboratory experimental studies [e.g., Glaeser, et al. (2000)] as well as field experiments by social psychologists [e.g. Bryan and Test (1967)] support the hypothesis that agents indeed have tastes for reciprocity. In further accord with these experimental results [Andreoni and Miller (1993)], our model predicts an equilibrium mixture of the two types (altruist and opportunist), rather than a monomorphic population. Our empirical tests complement these experiments, by using non-experimental data from a large, micro data set on trust in the U.S. As the model leads us to expect, trust is higher in harsher climates, even when a host of other explanatory variables are held constant. Additionally, trust of neighbors is higher in relatively stable communities, in which the reputational mechanism can be expected to operate, when the dependent variable is trust of the respondent’s neighbors. 177 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. References A c k e r b e r g , D ., an d M. B o t t i c i n i (2002): “Endogenous Matching and the Empirical Determinants of Contract Form,” Journal of Political Economy, 110, 564-591. A .J.C . V a n d e K r a g t , J. O., a n d R. D a w e s (1983): “The Minimal Contribut ing Set as a Solution to Public Goods Problems,” The American Political Science Review, 77, 112-122. A le s in a , A ., and E . L. F e r r a r a (2002): “Who Trusts Others?,” Journal of Public Economics, 85, 207-234. A n d o , A. (1999): “Waiting to Be Protected under the Endangered Species Act: The Political Economy of Regulatory Delay,” Journal of Law and Economics, 42, 29-60. A n d r e o n i, J. (1995): “C ooperation in Public-G oods Experim ents: Kindness or Confusion?,” The American Economic Review, 85, 891-904. A n d r e o n i, J., and J. M i l l e r (1993): “Rational Cooperation in the Finitely Repeated Prisoner’s Dilemma: Experimental Evidence,” Economic Journal, 103, 570-585. A u s u b e l, L. (1999): “Adverse Selection in the Credit Card Market,” mimeo, Uni- veristy of Maryland. A x e l r o d , R. (1984): The Evolution of Cooperation. Basic Books, New York. B a l e s t r a , P. (1967): The Demand fo r Natural Gas in the United States; a D y namic Approach fo r the Residential and Commercial Market. North-Holland Pub lishing Company. B a l l a r d , C. L ., J. B . S h o v e n , a n d J. W h a l l e y (1985): “General Equilibrium Computations of the Marginal Welfare Costs of Taxes in the United States,” American Economic Review, 75, 128-138. B a r b ib u x , C. (1999): “Contractualisation - Competition - Regulation: Three Guidelines for Improved Efficiency in Urban Public Transport,” Sixth Interna tional Conference on Competition and Ownership in Land Passenger Transport. B a r o n , D., a n d R. M y e r s o n (1982): “Regulating a Monopolist with Unknown Costs,” Econometrica, 50, 911-930. 178 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. B e c k e r , G. (1983): “A Theory of Competition Among Pressure Groups for Polit ical Influence,” Quarterly Journal of Economics, 98, 371-400. ---------- (1985): “Public Policies, Pressure Groups, and Dead Weight Costs,” Jour nal of Public Economics, 28, 329-347. B e r e c h m a n , J. (1993): Public Transit Economics and Deregulation Policy. Ams terdam; New York: North Holland. B e r e c h m a n , J., an d G. G iu lia n o (1985): “Economies of Scale in Bus Transit: A Review of Concepts and Evidence,” Transportation, 12, 313-332. B e r e c h m a n , J ., and G . G u ilia n o (1984): “Analysis of the Cost Structure of an Urban Bus Transit Property,” Transportation Research, 18B, 273-287. B e r n s t e in , M. (1955): Regulating Business by Independent Commission. Prince ton University Press. B o l t o n , G., a n d A. O c k e n f e l s (2000): “ERC: A Theory of Equity, Reciprocity, and Competition,” American Economic Review, 90, 166-193. B o y d , R ., an d P. R ic h e r s o n (1985): Culture and the Evolutionary Process. University of Chicago Press, Chicago. B r o w n in g , E. K. (1976): “The Marginal Cost of Public Funds,” Journal of Po litical Economy, 84, 283-298. B r y a n , J ., a n d M. T e s t (1967): “Models and Helping: Naturalistic Studies in Aiding Behavior,” Journal of Personality and Social Psychology, 6, 400-407. CAMERER, C . (1988): “Gifts as Economic Signals and Social Symbols,” American Journal of Sociology, 94, S180-S214. C a m p b e ll, H. (1975): “Deadweight Loss and Commodity Taxation in Canada,” Canadian Journal of Economics, 8, 441-446. C a r m ic h a e l, H., and W. M a c L e o d (1997): “Gift Giving and the Evolution of Cooperation,” International Economic Review, 38, 485-509. C a w le y , J., an d T. P h ilip s o n (1999): “An Empirical Examination of Informa tion Barriers to Trade in Insurance,” American Economic Review, 89, 827-846. C h a m b e r la in , G. (1987): “Asymptotic Efficiency in Estimation with Conditional Moment Restrictions,” Journal of Econometrics, 34, 305-334. 179 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. C h ia p p o r i, P ., and B. S a l a n i e (1997): “Empirical Contract Theory : The Case of Insurance Data,” European Economic Review, 41, 943-950. ---------- (2000): “Testing for Asymmetric Information on Insurance Markets,” Jour nal of Political Economy, 108, 56-78. C h ia p p o r i, P. A., and B. S a l a n i e (2003): “Testing Contract Theory: A Survey of Some Recent Work,” in Advances in Economics and Econometrics - Theory and Applications, ed. by M. Dewatripont, L. Hansen, and P. Turnovsky, pp. 115-149. Econometric Society Monographs, Cambridge University Press. C o l l , J. M ., and J. H i r s h l e i f e r (1991): “The Limits of Reciprocity,” Rationality and Society, 3, 35-64. C r o p p e r , M., W . E v a n s, S. B e r a r d i, M. D u lc a - S o a r e s , and P. P o r t n e y (1992): “The Determinants of Pesticide Regulation: A Statistical Analysis of EPA Decision Making,” Journal of Political Economy, 100, 175-197. D a le n , D. M., and A. G o m e z -L o b o (1997): “Estimating Cost Functions in Regulated Industries Characterized by Asymmetric Information.,” European Eco nomic Review, 41, 935-942. ---------- (2003): “Yardsticks on the Road: Regulatory Contracts and Cost Efficiency in the Norwegian Bus Industry,” Transportation, 30, 371-386. D a n a h e r , A., L. N o w lin , T. P a r k in s o n , and P. R y u s (1999): “Transit Ca pacity and Quality of Service Manual,” Prepared by Kittelson and Associates, Inc. for Transit Cooperative Research Program, Transportation Research Board. D a r b y , M., and E. K a r n i (1973): “Free Competition and the Optimal Amount of Fraud,” Journal of Law and Economics, 16, 67-88. D a w e s, R., J. M c T a v ish , and H. S h a k le e (1977): “Behavior, Communica tion, and Assumptions About Other Peoples Behavior in a Commons Dilemma Situation,” Journal of Personality and Social Psychology, 35, 1-11. D e m s e t z , H. (1968): “Why Regulate Utillities,” Journal of Law and Economics, 11, 55-65. D upuit, J. (1844): “On the Measurement of the Utility of Public Works,” An- nales des Ponts et Chaussees, VII(2nd Series), Translated by R.H. Barback in the International Economic Papers, No. 2. London: Macmillan Co., 1952. 180 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. D u t h io n , B., P. V in c e n t , and J. C. Ziv (1999): “The ’ ’Sapin” Legislation and its Effects on Tendering and Competition in Public Transport in Prance,” CERTU and CNAM. F a z io li, R ., M. F ilip p in i, and P. P r io n i (1993): “Cost-Structure and Efficiency of Local Public Transport: The Case of Emilia Romagna Bus Companies,” Inter national Journal of Transport Economics, 20, 305-324. F r a n k , R. (1987): “If Homo Economicus Could Choose his Own Utility Function, Would He Want One with a Conscience?,” American Economic Review, 77, 593- 604. ------------ (1988): Passions Within Reason: The Strategic Role of the Emotions. Norton, New York. G a g n e p a in , P ., an d M. I v a ld i (2002): “Incentive Regulatory Policies: The Case of Public Transit Systems in France,” Rand Journal of Economics, 33, 605-629. G l a e s e r , E., D. L a ib s o n , J. S c h e in k m a n , and C. S o u t t e r (2000): “Mea suring Trust,” Quarterly Journal of Economics, 95, 811-846. G o ld b e r g , P ., and G . M a g g i (1999): “Protection for Sale: An Empirical Inves tigation,” The American Economic Review, 89, 1135-1155. G r o s s m a n , G ., and E. H e lp m a n (1994): “Protection for Sale,” American Eco nomic Review, 84, 833-850. G r o s s m a n , G ., and E. H e lp m a n (1996): “Electoral Competition and Special Interest Politics,” The Review of Economic Studies, 63, 265-286. G u th , W . (1995): “Incomplete Information About Reciprocal Incentives: An Evo lutionary Approach to Explaining Cooperative Behavior,” Internationl Journal of Game Theory, 24, 323-344. G u th , W ., and R. T i e t z (1990): “Ultimatum Bargaining Behavior - a Survey and Comparison of Experimental Results,” Journal of Economic Psychology, 11, 417-449. G u th , W ., and M. Y a a r i (1992): “An Evolutionary Approach to Explaining Reciprocal Behavior in a Simple Strategic Game,” in Explaining Process and Change: Approaches to Evolutionary Economics, ed. by U. W itt. University of Michigan Press, Ann Arbor, MI. G u tt m a n , J. (2001a): “Families, Markets, and Self-Enforcing Reciprocity Norms,” Annales dE conom ie et de Statistique, 63-64, 89-110. 181 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. ---------- (2001b): “Self-Enforcing Reciprocity Norms and Intergenerational Trans fers: Theory and Evidence,” Journal of Public Economics, 81, 117-151. ---------- (2003): “Repeated Interaction and the Evolution of Preferences for Reci procity,” Economic Journal, 113, 631-656. H a n s e n , L. (1982): “Large Sample Properties of Generalized Method of Moments Estimators,” Econometrica, 50, 1029-1054. H a n s s o n , I., and C. S t u a r t (1985): “Tax Revenue and the Marginal Cost of Public Funds in Sweden,” Journal of Public Economics, 27, 331-353. HARSANYI, J . (1973): “Games with Randomly Disturbed Payoffs: A New Rationale for Mixed Strategy Equilibrium Points,” International Journal of Game Theory, 2, 1-23. H u n t in g t o n , S. (1952): The Marasmus of the ICC. Yale Law Journal. J r ., M. O. (1983): “The Political Economy of Comparative Growth Rates,” in The Political Economy of Growth, ed. by D. Mueller. Yale University Press, New Haven, CT. K a h n , A. E. (1970): The Economics of Regulation, vol. I. John Wiley. ---------- (1971): The Economics of Regulation, vol. II. John Wiley. K a tz , E., a n d J. R o s e n b e r g (forthcoming): “An Economic Interpretation of Institutional Volunteering,” European Journal of Political Economy. K o lk o , G. (1965): Railroads and Regulation, 1877-1918. New York: Norton. K r a n t o n , R. (1996): “The Formation of Cooperative Relationships,” Journal of Law, 12, 214-233. K r e p s , D., P. M ilg r o m , J. R o b e r t s , a n d R. W ils o n (1982): “Rational Co operation in the Finitely Repeated Prisoners’ Dilemma,” Journal of Economic Theory, 27, 245-252. L a f f o n t , J.-J. (1994): “The New Economics of Regulation Ten Years After,” Econometrica, 62, 507-537. ----------- (1996): “Industrial Policy and Politics,” International Journal of Industrial Organization, 14, 1-27. ----------- (2004): Regulation and Development. Cambridge University Press. 182 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. L a f f o n t , J.-J., and J. TlROLE (1986): “Using Cost Observation to Regulate Firms,” Journal of Political Economy, 94, 614-641. ---------- (1988): “The Dynamics of Incentive Contracts,” Econometrica, 56, 1153— 1175. ---------- (1990): “The Regulation of Multiproduct Firms. I: Theory,” Journal of Public Economics, 43, 1-36. ---------- (1991a): “The Politics of Government Decision-Making; A Theory of Reg ulatory Capture,” Quarterly Journal of Economics, 106, 1089-1127. ---------- (1991b): “Provision of Quality and Power of Incentive Schemes in Reg ulated Industries,” in Equilibrium Theory and Applications (Proceedings of the Sixth International Syposium in Economic Theory and Econometrics), ed. by J. Gabszewicz, and A. Mas-Colell. Cambrigde University Press. ---------- (1993): A Theory of Incentives in Procurement and Regulation. The MIT Press. LAN, L. W ., an d A. P . K u o (2003): “Modeling Bus Drivers Aberrant Behav iors and the Influences on Fuel and Maintenance Costs,” Journal of Advanced Transportation, 38, 93-113. L e d y a r d , J. (1995): “Public Goods: A Survey of Experimental Research,” in Handbook of Experimental Economics, ed. by J. Kagel, and R. Roth. Princeton University Press, Princeton, NJ. M . M c C u b b in s, R. N., and B. W e in g a s t (1989): “Structure and Process as Solutions to the Politician’s Principal-Agency Problem,” Virginia Law Review, 74, 421-482. M c F a d d e n , D. (1989): “A Method of Simulated Moments for Estimation of Dis crete Response Models W ithout Numerical Integration,” Econometrica, 57, 995- 1026. M i l l e r , G. (1997): “The Impact of Economics on Contemporary Political Science,” Journal of Economic Literature, 35, 1173-1204. M.L. C r o p p e r , W .N . E v a n s, S. B. M. D.-S., and P. P o r t n e y (1992): “The Determinants of Pesticide Regulation: A Statistical Analysis of EPA Decision Making,” Journal of Political Economy, 100, 175-197. N e ls o n , G. R. (1972): “An Econometric Model of Urban Bus Transit Operations,” Paper P-863, Institute for Defense Analysis, Arlington, VA. 183 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. NELSON, P. (1970): “Information and Consumer Behaviour,” Journal of Political Economy, 78, 311-329. N o l l , R . (1971): “The Nature and Causes of Regulatory Failure,” Administrative Law Review, 23, 424-437. N o l l , R., and B. O w e n (1983): The Political Economy of Deregulation: Interest Groups in the Regulatory Process. American Enterprise Institute for Public Policy Research, Washington and London. O b e n g , K . (1984): “The Economics of Bus Transit Operations,” Logistics and Transport Review, 20, 45-65. ---------- (1994): “Specification of a Long Run Transit Cost Function to Estimate Input Bias from Subsidies,” Transportation Planning and Technology, 18, 187-98. O ls o n , M . (1965): The Logic of Collective Action: Public Goods and the Theory of Groups. Harvard University Press, Cambridge. O ls o n , M., and R. Z e c k h a u s e r (1966): “An Economic Theory of Alliances,” Review of Economicsand Statistics, 48, 266-279. P e lt z m a n , S. (1976): “Toward a More General Theory of Regulation,” Journal of Law and Economics, 19, 211-240. P e r r i g n e , I. (2002): “Incentive Regulatory Contracts in Public Transportation: An Empirical Study,” Working Paper, Pennsylvania State University. P e r r i g n e , I., and S. S u r a n a (2004): “Politics and Regulation: The Case of Public Transit,” In progress. P e r r i g n e , I., and Q . V u o n g (2004): “Econometrics of Incentive Regulation: Nonparametric Identification,” In progress. PUCHER, J. (1995a): “Urban Passenger Transport in the United States and Europe: A Comparative Analysis of Public Policies,” Transport Reviews, 15, 99-117, Part 1: Travel Behavior, Urban Development, and Automobile Use. ---------- (1995b): “Urban Passenger Transport in the United States and Europe: A Comparative Analysis of Public Policies,” Transport Reviews, 15, 211-227, Part 2: Public Transport, Overall Cost Comparisons and Recommendations. P u tn a m , R . (2000): Bowling Alone: The Collapse and Revival of American Com munity. Simon and Schuster, New York. 184 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. R e v o l l e , P. V. (1999): “Public-Private Partnership in Building Transport Lines with Right of Way in Prance: A New Start?,” CERTU. R o s s e t t i , M ., a n d B. E v e r s o l e (1993): Joum ey-To-W ork Trends in the United States and its M ajor Metropolitan Areas 1960-1990. US Department of Trans portation, Cambridge, MA. S a p p i n g t o n , D. (1982): “Optimal Regulation of Research and Development under Imperfect Information,” The Bell Journal of Economics, 13, 354-368. ----------- (1983): “Optimal Regulation of a Multiproduct Monopoly with Unknown Technological Capabilities,” The Bell Journal of Economics, 14, 453-463. S h l e i f e r , A., a n d R. VlSHNY (1994): “Politicians and Firms,” Quarterly Journal of Economics, 109, 995-1025. S m a l l , K., a n d J. G o m e z -I b a n e z (1999): “Urban Transportation,” in Handbook of Regional and Urban Economics, ed. by E. S. Mills, and P. Cheshire. North Holland. S m ith , V. (2003): “Constructivist and Ecological Rationality in Economics,” American Economic Review, 93, 465-508. S t i g l e r , G. (1971): “The Theory of Economic Regulation,” The Bell Journal of Economics and Management Science, 2, 3-21. S t o n e , A. (1982): Regulation and its Alternatives. Washington, D.C.: Congres sional Quarterly Press. S t u a r t , C. E . (1984): “Welfare Costs Per Dollar of Additional Tax Revenue in the United States,” American Economic Review, 74, 352-362. T.H . O u m , W . W ., a n d J. Y o n g (1985): “Concepts of Price Elasticities of Transport Demand and Recent Empirical Estimates: An Interpretative Survey,” Journal of Transport Economics and Policy, 26, 139-154. T i r o l e , J. (1996): “A Theory of Collective Reputations (with Applications to the Persistence of Corruption and Firm Quality),” Review of Economic Studies, 63, 1- 22 . T u l l o c k , G. (1985): “Adam Smith and the Prisoners’ Dilemma,” Quarterly Jour nal of Economics, 100, 1073-1081. USLANDER, E. (2002): The Moral Foundations of Trust. Cambridge University Press, Cambridge. 185 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. VlTO N, P. (1981): “A Translog Cost Function for Urban Transit,” Journal of Industrial Economics, 29(3), 287-304. ---------- (1997): “Technical Efficiency in Multi-Mode Bus Transit: A Production Frontier Analysis,” Transportation Reserach, 3IB, 23-39. W a b e , J. S., a n d O. B. C o l e s (1975): “The Short and Long-Run Cost of Bus Transport in Urban Areas,” Journal of Transport Economics and Policy, 9, 127- M O. W i n s t o n , C . (1985): “Conceptual Developments in the Economics of Transporta tion: An Interpretive Survey,” Journal of Economic Literature, 23, 57-94. ---------- (2000): “Government Failure in Urban Transportation,” AEI-Brookings Joint Center fo r Regulatory Studies. W i n s t o n , C ., a n d R. C r a n d a l l (1994): “Explaining Regulatory Policy,” Brook ings Papers on Economic Activity. Microeconomics. WOLAK, F. (1994): “An Econometric Analysis of the Asymmetric Information, Regulatory-Utility Interaction,” Annales d ’ Economie et de Statistique, 34, 13-69. W o o d , B., a n d R. W a t e r m a n (1991): “He Dynamics of Political Control of the Bureaucracy,” American Political Science Review, 85, 801-828. W o o l d r i d g e , J. M. (2002): Econometric Analysis of Cross Section and Panel Data. The MIT Press. 186 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Appendices A Appendices for Chapter 3 A .l Appendix A .l The Versement Transport: A new transportation tax, viz. the “Versement Trans port (V.T.)” was instituted in Paris in 1971 whereby all companies employing nine or more people and located within the corresponding urban area were required to pay a percentage of salaries paid as V.T. In 1973 the V.T. was extended to all cities with 300,000 or more inhabitants and later in 1974 to all cities with 100,000 inhab itants or more. The rate of V.T. depends on the size of the conurbation and the magnitude of the investments in public transportation. This rate, however, can not exceed 1.75% or 1.8% in some special cases. The period 1970-1982 marked the turning point in the history of French urban transportation as it received increasing attention from the planners and culminated in the passage of LOTI. The Sapin Law: Another watershed in the French urban transportation service was the passage of the “Sapin Law” on January 29, 1993. This law was passed in response to the dissatisfaction in the system due to the monopoly power of the big operators. The most significant consequence of the Sapin Law, apart from proce dural changes, was to introduce competition amongst the prospective operators as the contracts were now given through a formal bidding procedure. The competition in some instances also crossed national boundaries. For the first time, in 1999, a 187 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. non French (Spanish) operator entered the French market by winning the contract. CERTU has conducted many studies and the general conclusion arrived at is that the Sapin Law has not had a revolutionary impact. The three big corporations still occupy a large share of the market. The seeds of change, however, have been sown with an increase in competition in the franchising of the operation of public service. A .2 Appendix A .2 Derivation of I.C: The firm maximizes E(p(f3,f3,e) = E [ t 0 , C ( q 0 ,£ d ) , s 0 ) , P — e , e c)) — ip(e)\ with respect to its two control variables, e and (3. The incentive compatibility condition is derived in two steps. First, as already mentioned, for any report /3 that a firm of type (3 makes let the effort be denoted by e(/5|/3), where e(/?|/3) is the solution of E [— t 2Cs — ip'] — 0. Given this effort level, truth- telling requires that E<p(f3, p 3, e0\f3)) is maximized at (3 = /?, the first-order condi tion for which is given by E [ti + t 2 (C\qi + C2s — C^e) — ip'e] = 0. By definition, E U ((3) = E [ti + t 2 (CiQi + C2s + Cz — Cse) — ip'e]. Combining all of the above gives the incentive compatibility condition, viz., E[U ((3)] = — ip' (e (/?)). A .3 Appendix A .3 Implementation through a menu of linear contracts: Equations (22), (23), and (24) determine the optimal levels of price, effort, and quality, p* (0 ), e* (f3), and s* (/3). 188 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The firm’s expected utility is then determined from (14) and (17) as J ip1 (e* (6)) d6 P P and the expected transfer of the firm is given by f ip' (e* (5)) dS + Tp(e*((])). We must P still check the second-order condition of the firm’s IC and find a transfer function that implements the optimal contract. We first discuss the implementation of the optimal contract through a menu of linear schemes for our specified cost function given in (25). The implementation of the optimal contract can be extended to a more generalized cost function.1 3 2 Assume that the planner gives the firm the following transfer function: t * 0 , C ) = r*0 ) - K * 0 ) { C - C * 0 ) } (40) where, C * 0 ) = ,E[log{exp (ao) q * 0 ) o‘vs * 0 ) o‘3p f lp ^ 1 exp[at0 - e*0))\ * exp (ay£d + £c)}]> C = log {exp (a0) q * 0 ) ays * 0 ) asP?lP T e^P [a/ (P ~ e)] exp (ayed + ec)}, r ' 0 ) = tP(e*0)) + J 4>' (e* (< 5)) d6, and K * 0 ) = P The financial compensation scheme is such that the firm receives a lumpsum transfer of t* 0 ) , and if the ex post observed C differs from C*, the regulator and the firm share the difference in the fraction (1 — K * ,K * ). The term K * 0 ) denotes the power of the incentive scheme. Given ijj" (•) > 0, the higher the power of the 132For a cost function multiplicatively separable in {(3 — e) and (q,s,ec), i.e., C = G (j3 — e) H (q,s,e c) , Laffont and Tirole (1990) have specified conditions under which the opti mal regulatory policy can be implemented through a menu of linear contracts. 189 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. scheme, the more effort the firm puts in. The firm of type 0 solves max E[t*(0) — P ,e K *(0){logC — C*(0)} — if(e(0\0))]. Maximization with respect to e gives e(0\0) = e*(0). Maximization with respect to 0, and using e(0\0) = e*(0), and (14) we find that the firm truthfully reveals its type, i.e., 0 = (3. P ro p o sitio n 2 Given cost C, the contract (p*(0),s*(0),t*(0,C)] induces the regu lated firm to reveal its type truthfully and to exert the optimal effort e* (0), where p*(0) and s*(0) are the optimal price and quality of the regulator’s problem and t*(0,C) is given by (40). The proof is given below. We also find conditions under which the second-order condition for incentive compatibility is satisfied. The firm maximizes expected utility C(0, A e) = M e*(,8)) + j V (e* M ) ~ 0) ~ - e)} - if (e)] with respect to its two choice variables, its report of type (0) and the effort that it exerts (e). We first find firm’s optimal effort when it reports 0, i.e. e(0\0), for any given report 0 and show that given its report it exerts the optimal effort, i.e., e(0\0) = e*(0). Then we optimize with respect to 0 taking e*(0) as given and show that a firm of type 0 will find it optimal to report its true type. Taking first-order condition with respect to e we get if'{e*{0)) — i j j ( e ) , that is, e = e*(0). Second-order condition with respect to e yields —if" (e) < 0 (since disutility of effort increases at an increasing rate) which implies that ( ( 0 , 0,e) is 190 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. concave in e. Hence, given its report (3 the firm in order to maximize its utility always exerts the optimal effort for the report that it makes, viz. e*0). For any report /3, the firm’s utility, when it puts in effort e * 0 ), will be ( 0 , (3) — P « . ip(e*0)) + f ip' (e* (5)) d6 + ip '(e * 0 ))0 — (3) — ip{e*0)). We now show that report- 3 ing its type truthfully is the optimal choice for the firm. Differentiating ( 0 0 ) with respect to @ we get, J | = ip " (e * 0 ))e * 0 )0 — f3), which is zero at f3 — (3. Hence, the first-order condition is satisfied at the truthful report.1 3 3 In order to get the second-order condition for the incentive compatibility we take the second derivative of ( 0 0 ) with respect to ft which yields ip"'(e*0))(e*0))20 — (3) + ip"(e*0)) ’e’* 0 ) 0 — (3) + 0 ( e * 0 ) ) e * 0 ) . The required condition for the SOC to be non positive at $ = (3 is e * 0 ) < 0 . In the next few steps we will find the conditions under which e * 0 ) < 0 holds. L em m a 1 If {I + A ) ECzz < ip" (e) then the solution e* (/?) of the relaxed problem is decreasing in (3. P roof. Using Pontryagin’s maximum principle we get if p* 0 ), s* 0 ), e* 0 ) solve the problem (15)— >(17) and EU* is the associated optimal path then there exists a function p 0 ) such that for all (3 G [/3, /3]: 133The first order condition will also be satisifed at a fi ^ P for which e*(0) = 0. However, as long as e* 0 ) < 0 for all (3, we have a global maximum at ,6 (from we can see that under the assumption e*(/3) < 0 the firm’s utility is either constant or increasing for reports [3 < [3 and either constant or decreasing for reports 0 > (3 ). In Lemma 1 we find the condition under which e * 0 ) < 0 . 191 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. (i) H (E U ' (P ), V' (0), s' (0 ) , e* (0 ) , fi (0 ) , p) > H (E U * (P), p, s, e, p (P), /?) for all non-negative p, s, e; (ii) p(P) = — dH (- EU ,p (iii) /i (/5) = 0 and that there is no restriction on /j, (j3). From (21), p(P) is given by p.(0) = XF (P). Hence, we have H (EU* (p) , p* (p) , s* (P) , e* (P) , AF (p) , 0) > H (EU* (p) ,p* (P) , s* (p) , e, AF (P) , P) (41) for all e and all p G [/3, /3]. Now suppose that e* (P) is not decreasing in the range [/3,/3], i.e., suppose there exists a Pt > P2 with e* (Px) > e* (P2). Condition (41) implies that H (e* (Px) , PP) > H (e* (P2) , Pt ) , where, H (e, P) = H (EU* (P) ,p* (P) , s* (P) , e, A F (P) , /?). This, in turn, implies1 3 4 SKPi) ^ + (1 + A) E [C (q, s, P1 — e2, £c) + V ' (e2)] + i 1 ' (e2) > 0 or, (1 + A ) E [C (q, s, P1 - e * 2,e c) - C (q, a, P 1 - e*, ec)] (43) + (1 + A ) (e2) - i> (e^)] > & (el) - <// (e * 2)) A 134We have used ej to denote e* (/3j) and e2 to denote e* (/32). 192 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. Similarly, we have H (e* ((32) ,(32) > H (e* (f^ ) , /?2), which implies (1 + A) E [C (q, », h - e l ec) + < / > (e$)] - < l> ' (e'2) + ( l + X)E{C{q, s, 02 - e l , e c) + *P (e»] + V -' (e() > 0 (44) or, (1 + A ) E [C (q, a, (d2 - e * 2, ec) - C (q, a, (32 - ej, ec)] + (1 + A) [ip (e2) - ip (ej)] < pip’ (ej) - ip' (e * 2)) Ajf&j (45) Conditions (43) and (45) together imply (1 + A ) \ e [C (q, s, - e2, £c) - C (q, a, /?x - ej, ec)] - £ [C (g ,s ,/?2 - e£,ec) - C { q , s , p 2 - ej,ec)]| > W (e*) - ip' (e2))A m i fW i) F(P*) /(/3 2 ) (46) Now (1 + A) C 33 (g, — e, ec) < ip" (e) A ^ for all (3 and e implies (1 + A) / / £ C 3 3 (q , s , ( 3 - e , ec) ded/3 < f f ip" (e) A ^ ( j f § ) ded(3 P2 e2 P2 e 2 193 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. = ► (1 + A ) / E[C3 (q,s,0 - e * 2,£c) - C 3 ( q , s , 0 - e \ , e c)]d0 @2 </[V '(e;)-V ’, W)] ( 0 ) i 0 => (1 +A) ^E[C(q,s,p1 - e '2,e c) -C(q,s, 01 - e;,£„)] — E[C (q, s, 0 2 - e * 2,e c) - C (q, s, 0 2 ~ ej, £ , < W (ej) - $ ( e S ) ) A (47) m A fW i) m i m ) Equation (47) contradicts with (46), which implies that if (1 4- A) EC33 < V (e) A fp F(P) m then for any 0 X > 0 2 we cannot have e* (0X ) > e* (02). This proves that e* (0) < 0. 194 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. B Appendices for Chapter 4 B .l Appendix B .l The mixed strategy equilibria in Cases 2b and 3: To further analyze the mixed strategy equilibria in Cases 2b and 3, let Pr(trust) denote the probability that buyers will trust sellers who contribute in the mixed strategy equilibrium. Note that the opportunist must contribute to the provision of the public good—if he or she contributes at all—in each stage in order not to reveal his or her type.1 3 5 Note further that, in a mixed strategy equilibrium, the expected payoffs of the pure strategies that comprise the equilibrium mixed strategy must be equal. There are three (stationary) pure strategies available to the agent: (a) to contribute and honor trust in market transactions, (b) to contribute and cheat in market transactions, and (c) to free-ride and thus forgo market transactions entirely (giving a zero expected payoff). The expected discounted expected payoff of contributing and honoring trust (which can be optimal only for permanent opportunists, as noted above) becomes t-, , •, \ Pr (trust) (1 — ex) c Eit(contribute A honor) = --------- — -------------- ^ — frj’ ( ) 135That is, the mixing of strategies takes place through some opportunists contributing in each stage, while other opportunists (of the same type) not contributing at all. See Harsanyi [1973] for this interpretation of mixed strategy equilibria. 195 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. The critical P r(tru st) which makes the right hand side of equation (Al) zero is given by . c c Pr [trust) = — -----rrri------ r = ZTi--------r • (1 + < 5)(1 — ei) c(honor) Similarly, the expected discounted expected payoff of contributing and cheating (which applies to both types of opportunist) becomes 7 -, / -7 7 \ Pr (tru st) c E 7r(am tribute A cheat) = j _ 6(1 _ - (1 + *(1 _ 9)) (1 _ {(1 _ which becomes equal to zero at --------------------- Q Q Pr (tru st) = ---- -T--- -r — Z7-,----7 T - 1 + < 5 (1 — 0) c(cheat) In Case 2b, given that c > c(honor), we find that Pr (trust) > 1. This means that there is no mixed strategy equilibrium in which (permanent) opportunists contribute and honor trust. Thus the only mixed strategy equilibrium has both permanent and transient opportunists contributing and cheating (or free-riding), and being trusted with probability Pr (trust). In Case 3, we have c < c(honor). In this case, there exist two mixed strategy equilibria. In one equilibrium, opportunists mix between cooperating and cheating, on the one hand, and free-riding on the other, and are trusted with probability Pr (tru st). In the other equilibrium, opportunists mix 196 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. between cooperating and honoring trust, on the one hand, and free-riding on the other, and are trusted with probability Pr (tru st). B.2 Appendix B.2 D efinitions of variables • low temp: Average low temperature for the month of January in Fahrenheit Source: www.weatherbase.com. Weatherbase is a source for finding monthly weather records and averages for more than 16,439 cities worldwide. Their weather information is collected from a variety of public domain sources, including the Na tional Climatic Data Center. • (low) m obility: Proportion of people who have lived in the same house for the past five years ^(persons aged 5+ in 1985 lived in the same house)/(all persons aged 5+) Source: Census data (1990). The source for all the following variables is: The 2000 Social Capital Benchmark Survey, a research study undertaken by the Saguaro Seminar at the John F. Kennedy School of Government. The purpose of the Social Capital Benchmark Survey, con ducted in 41 U.S. communities, is to measure various manifestations of social capital 197 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. as well as its suspected correlates. The survey was conducted in the United States by telephone using random-digit-dialing (RDD) during July-November (2000), ex cept for West Oakland, California Survey (which was conducted between December, 2000 and February, 2001). Forty two different sponsoring organizations collected the data. • trustown: trust of own ethnic group 0. trust not at all 1. trust only a little 2. trust some 3. trust a lot • trust neighbors: trust neighbors 0. not at all 1. a little 2. some 3. a lot • high school: dummy for high school completed • college: dummy for some college education completed • assoc, degree: dummy for associate degree or specialized technical training completed 198 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. grad training: some graduate training completed • bachelors: bachelor’s degree completed • grad degree: graduate or professional degree com pleted. • 20K < income < 30K: dum m y for household incom e over $20,000 but less than $30,000. • 30K < income < 50K: dum m y for household incom e over $30,000 but less than $50,000. • 50K < income < 75K: dum m y for household incom e over $50,000 but less than $75,000. • 75K < income < 100K: dum m y for household incom e over $75,000 but less than $100,000. • income > 100K: dum m y for household incom e $100,000 or more. • 30 < age < 50: dum m y for 30 < age < 50. • 50 < age < 70: dum m y for 50 < age < 70 • age > 70: dum m y for age > 70 male: dum m y for m ale 1 if male 0 if female 199 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. • experienced discrim ination: experience of discrimination: 0. never 1. rarely 2. som etim es 3. often 4. very often • Afro-American: dum m y for Afro-American • Hispanic: dum m y for Hispanic • Asian: dum m y for Asian • friends: number of close friends 1. no close friends 2. 1-2 close friends 3. 3-5 close friends 4. 6-10 close friends 5. more than 10 close friends • econ status: satisfaction w ith current financial situation 0. not at all satisfied 200 Reproduced with permission of the copyright owner. Further reproduction prohibited without permission. 1. som ew hat satisfied 2. very satisfied • happy: Q uestion asked is “how happy are you?” 0. not happy at all 1. not very happy 2. happy 3. very happy. • political interest: interest in politics and national affairs 1. not at all interested 2. only slightly interested 3. som ew hat interested 4. very interested • mean community income: m ean incom e in the respondent’s com m unity • population density: Population density in zipcode, 1997, in thousands • income homogeneity: See text. Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Individual heterogeneity and program evaluation

PDF

Essays on exchange rates, prices, and corporate behavior during the Asian currency crisis

PDF

Essays on organizational forms and performance in California hospitals

PDF

Insurance mechanisms, forest clearance, and the effect of government policies in rural economies

PDF

Health interventions, health status, and labor market outcomes in Indonesia: Three essays

PDF

Intellectual property rights, quality control and incentives

PDF

Essays on auctions

PDF

Asymmetry and risk aversion in first -price sealed bid auctions: Identification, estimation, and applications

PDF

Compliance study of second-generation antipsychotics on patients with schizophrenia

PDF

Essays on technological evolution and financial returns to innovation

PDF

An empirical analysis of foreign aid and the provision of international public goods

PDF

Essays on agrarian institutions

PDF

Essays on consumption behavior, economic growth and public policy

PDF

Essays on contracting in the construction industry

PDF

Cloning and characterization of Grb4: A novel adapter protein interacting with BCR -Abl

PDF

By sea: The port nexus in the global commodity network (the case of the West Coast ports)

PDF

How financial statements in Mexican firms reflect changes in the financial/economic environment from 1978 to 1996

PDF

Dual labor markets, public debt management, and exchange rate movements

PDF

Analysis of the role of government in Taiwan's industrialization and economic development

PDF

Income inequality and economic growth: A theoretical and empirical analysis

Asset Metadata

Creator Surana, Sunita (author)

Core Title Essays on regulation of public utilities and the provision of public goods

School Graduate School

Degree Doctor of Philosophy

Degree Program Economics

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag economics, general,OAI-PMH Harvest,transportation

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Perrigne, Isabelle (committee chair), Tan, Guofu (committee chair), Ridder, Geert (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c16-458898

Unique identifier UC11341004

Identifier 3196900.pdf (filename),usctheses-c16-458898 (legacy record id)

Legacy Identifier 3196900.pdf

Dmrecord 458898

Document Type Dissertation

Rights Surana, Sunita

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA