WPS4101 Political Accountability and Regulatory Performance in Infrastructure Industries: An Empirical Analysis* Farid Gasmi Université de Toulouse I (GREMAQ, IDEI) gasmi@cict.fr Paul Noumba The World Bank pnoumbaum@worldbank.org Laura Recuero Virto Université de Toulouse I (GREMAQ) recuerovirto@hotmail.com Abstract The aim of this paper is to empirically explore the relationship between the quality of political institutions and the performance of regulation, an issue that has recently occupied much of the policy debate on the effectiveness of infrastructure industry reforms. Taking the view that political accountability is a key factor that links political structures and regulatory processes, we investigate, for the case of telecommunications, its impact on the performance of regulation in two time-series-cross-sectional (TSCS) data sets on 29 developing countries and 23 developed countries covering the period 1985-1999. In addition to confirming some well documented results on the positive role of regulatory governance in infrastructure industries, this paper brings some empirical evidence on the impact of the quality of political institutions and their modes of functioning on regulatory performance. This first analysis of the data sets shows that the (positive) effect of political accountability on the performance of regulation is stronger in developing countries. An important policy implication of this finding is that future reforms in these countries should give due attention to the development of politically accountable systems. JEL-codes: L51, H11, L96, L97, C23 Key words: Infrastructure industries, regulatory performance, political accountability. World Bank Policy Research Working Paper 4101, December 2006 The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org. *We thank Jean-Paul Azam, Antonio Estache, Frannie Léautier, and Wilfried Sand-Zantman for useful comments and discussions on the topic studied in this paper. Part of this research was undertaken while Farid Gasmi and Laura Recuero Virto were visiting the World Bank Institute during the summer of 2006. These authors thank members of the Institute, in particular, Gabriela Chenet-Smith for their warm hospitality. 1 Introduction The last two decades have witnessed a worldwide wave of reforms that have significantly affected both the market structure and the institutions in the infrastructure industries including high-tech sectors such as telecommunica- tions or electricity and more traditional domains such as water or postal services. In developed countries, the main objective of these reforms has been to improve the functioning of industries traditionally organized as what has become to be recognized as ill-performing public or private monopolies. The fundamental policy task has then been to redesign the legal and regu- latory frameworks so as to generate "proper" economic incentives in those industries, namely, inventives for operators to enhance their offerings, in par- ticular, in terms of cost efficiency, quality of service, and tariffs. While the reforms conducted in developing countries have been grounded on similar principles, in practice they differed markedly in at least two re- spects. First, even though there was clearly room for improving the perfor- mance of infrastructure industries in developed countries, one should recog- nize that in these countries service was typically available whereas in develop- ing countries it was sometimes merely non existent. This was for instance the case in the telecommunications industry when networks were not developed in large parts of the developing countries' rural areas. Second, and most importantly, the task of institutional design was far more challenging in developing countries. Developed countries essentially needed to work on how to modernize an already existing institutional fabric and a complex system of functioning rules built over a long history of political and economic administration of market economies. It is safe to say that in most of the cases, although for different reasons, this crucial experience was just lacking in developing countries. Beyond the fact that these countries had to follow the industrialized world in the setting of new institutions to regulate the reformed industries, an uneasy task by itself given the scarcity of human capital, they could only expect to deal with a severe inadequacy of the old administrative functioning rules. This inadequacy would be certainly felt more at the global level of the functioning of political institutions than 2 at the local level of the governance of regulatory institutions. More recently, policy makers in developing countries have pushed further the process of reforms of their infrastructure industries. After a period of implementation of policies of liberalization and privatization of some seg- ments coupled with the creation of regulatory authorities, large efforts have been allocated to improve the efficiency of the working of these authorities. Degree of independence, capacity of human capital, and particularly quality of governance are the three policy items that have mobilized much of these efforts. On the research front, however, both theoretical work on the optimal design of regulatory institutions and empirical work on the measurement of regulatory institutions' performance suggest that these specific items should not be analyzed independently from more general factors related to the gov- ernance of the economy as a whole. The main purpose of this paper is to investigate the relative weight of these factors in regulatory performance by means of an econometric analysis of two data sets on the telecommunications industry in developing and developed countries. The determinants of regulatory performance have been discussed both in the theoretical and empirical streams of the literature on infrastructure industries regulation. For our purpose, we distinguish two approaches. A first approach, which is conceptual in nature and inspired by political sci- ence, argues that when thinking about regulatory performance the relevant game is to be found upstream at the (higher) level of politics (Spiller and Tommasi, 2003). Another more empirical approach emphasizes the impact of regulatory governance on performance (Cubbin and Stern, 2005b). Our general view is that indeed the relationship between political structures and regulatory processes has to be given due attention when assessing regulatory performance. Hence, our study might be viewed as an attempt to merge both of these approaches in order to feed in some empirical elements to the debate on the relationship between political and regulatory institutions that so far has mainly taken place at a conceptual level. Our empirical strategy consists in implementing a series of econometric tests with a special attention given to variables that capture political account- 3 ability, a concept that we consider as fundamental in the exercise of the link between political structures and regulatory processes. Hence, we regard the (political game) equilibrium level of political accountability as an important determinant of the regulatory process' performance. This leads us to set up, and illustrate with our data, a test having as the null hypothesis that, all things equal, more political accountability should enhance the performance of regulation. In addition to merely testing its significance, we attempt to give some empirical substance to the conjecture that political accountability has an even stronger effect in developing countries.2 The plan of the paper is as follows. The next section summarizes some of the main theoretical and empirical arguments recently put forward in the literature on the design of institutions and the evaluation of regulatory performance in infrastructure industries. This section is not meant to be exhaustive but rather to serve the purpose of arguing that there is a need to merge these two streams of the literature on regulatory institutions. Sec- tion 3 describes the basic econometric-theoretical ingredients that constitute the elements of the empirical methodology we use to analyze two data sets on 29 developing countries and 23 developed countries covering the period 1985-1999. In section 4, we discuss the results of a preliminary analysis of these data. Our objective there is to uncover some general properties of the data and attempt to establish a diagnostic of stationarity for the regulatory performance variables. The actual empirical analysis of the relationship between political ac- countability and regulatory performance is taken up in Section 5 in two-steps. First, in subsection 5.1 we investigate the existence of causality relationships. This causation analysis provides us with a set of variables that could be cred- ibly used as independent variables in regressions of regulatory performance on political accountability variables. The results of such regressions are dis- cussed in subsection 5.2. Section 6 summarizes our empirical findings and discusses some policy implications. A detailed description of the data used, their sources, and some complementary material are given in the appendix. 2From a normative analysis perspective, assuming that regulatory performance in- creases social welfare, such a finding would suggest that marginal social benefit of political accountability is higher in developing countries. 4 2 Design of institutions and regulatory performance: The need for an integrated approach Recent contributions to the theory of the design of institutions and empirical work concerned with the measurement of their performance have brought to daylight the issue of the performance of regulation.3 Two approaches have been followed to examine the determinants of regulatory performance and outcomes. A first approach is conceptual and analyzes the role of political structures and processes. A second approach, more empirical in nature, emphasizes the impact of the quality of regulatory governance. We briefly review the main arguments developed by these two approaches and point to the need to develop a unified analytical framework. This study is a first empirical effort exerted towards this direction. The first approach analyzes the relationship between political structures and processes and the conduct of regulation by emphasizing the need to open the black box of the organization and functioning of governments (see Estache and Martimort, 1999 and North, 2000).4 In their analysis of the link between politics and regulation in the US, McCubbins et al. (1987) argue that, by reducing the costs of monitoring and by sharpening sanctions, administrative procedures can give rise to an equilibrium in which compliance with the preferences of political agents is greater than it otherwise would be.5 This relationship is explored by Levy and Spiller (1994) in the telecommunications sector through an analysis of case studies. In particular, they evaluate the potential for political agents to manipulate the regulatory process. They find 3Laffont (2005) devotes two chapters of his book on regulation and economic develop- ment to the discussion of issues related to the design of proper institutions in developing countries. 4By putting the political game at the heart of the analysis, this approach fits in the New Institutional Economics paradigm founded on the precepts of transaction cost theory and positive political economy. This constitutes an important departure from the standard normative approach to public economics. 5Bottom-up "fire-alarm" monitoring through external agents who are affected by regu- latory agencies' policies is a good example of a method that can reduce the informational costs of following the activities of agencies (McCubbins and Schwartz, 1984). 5 that sector performance can be satisfactory under a wide range of regulatory procedures as long as arbitrary administrative moves can be restrained. The link between the political and regulatory spheres is further analyzed in Spiller and Tommasi (2003) through the impact that the characteristics of political environments have on the ability of political agents to reach in- tertemporal cooperation. They argue that long term political cooperation is more likely to lead to stable and flexible regulatory policies, i.e, to effective regulation, when the agents with decision power have strong intertemporal relationships, policy and political moves are widely observable, good enforce- ment technologies are available, political exchanges take place in arenas where the previous features are satisfied, and the short-run payoffs from noncooper- ation are not so high. For example, these authors argue that more inefficient regulatory rules, i.e, a rigid regulatory context, may in fact provide higher incentives for investment whereas granting discretion to the regulator may lead to arbitrary outcomes if institutional endowments are low. Heller and McCubbins (1996) argue that incentives for investing in in- frastructure industries are not credible within a given regulatory structure unless there is a political context that makes them sustainable. Regulatory predictability is a key feature for gaining credibility, and hence the impor- tance role of political institutions in enhancing this predictability. The higher the quality of the political and institutional environment, the more difficult it is to change regulatory structures and procedures. In particular, the more veto political players with effective authority there are, the easier it is to block policy changes. Let us now turn to an overview of the empirical approach that empha- sizes the role of regulatory governance. The fundamental belief that moti- vates much of this line of research that essentially deals with infrastructure industries is that good regulatory governance is a prerequisite to a proper functioning of the positive relationship between regulatory incentives and regulatory performance. This belief is based on the conjecture that "..reg- ulatory agencies with better governance should make fewer mistakes, have their mistakes identified and rectified better and more quickly, so that good 6 regulatory practice is more readily established and maintained." (Cubbin and Stern, 2000a) The basic empirical implications of these "theoretical" hypotheses is that, thanks to the structuring and the practice of regulation it entails (e.g., as an independent regulator that makes transparent regulatory decisions), bet- ter regulatory governance increases capacity and enhances productive and allocative efficiency. In the case of telecommunications, which is the sec- tor concerned by our study, these implications are typically tested in data collected on a set of developing countries observed during a given time pe- riod. Regulatory performance is measured by mainline penetration rates and mainlines per employee, and regulatory governance is captured in an index (see Gutierrez, 2003a) that aggregates a set of aspects related to the struc- turation and internal organization of regulation. The methodology applied to both telecommunications (Gutierrez, 2003b) and electricity (Cubbin and Stern, 2000a) yields a positive impact of regulatory governance on output.6 A typical contribution to this line of research starts from the global con- ceptual view that the "..institutional quality is the dominant determinant of variations in long-term growth performance."7 However, in its imple- mentation part often it only accounts for micro dimensions of institutional quality embodied in what is referred to as the quality of regulatory gover- nance. Our view is that this approach should substantially gain in richness by drawing lessons from the literature on the design of institutions discussed in the beginning of this section. Our goal then is to take a step towards a unified approach that, when evaluating regulatory performance, in addition to specifying variables of regulatory governance, explicitly incorporates vari- ables that link political and regulatory structures and processes. Hence, our study can be viewed as a first exploration of the relative merits of such an integrated empirical approach. In our empirical analysis, the variables through which the interface be- tween political and regulatory structures and processes is going to materialize 6For a survey of empirical studies on regulatory performance and regulatory governance in developing countries, see Cubbin and Stern (2005b). 7See Cubbin and Stern (2005a) and the citations thereof. 7 are variables that are used to proxy the concept of political accountability. Broadly speaking, this concept may bedefined as "..a proactive process by which public officials inform about and justify their plans of action, their behavior and results and are sanctioned accordingly."8 A key idea here is that limiting the use and sanctioning the abuse of political power should help disentangling regulatory processes from the opportunistic behavior of political agents.9 The elections mechanism should, in principle, ensure political account- ability since citizens select representatives who hold bureaucrats and mem- bers of the judiciary system accountable for their behavior. However, this property of elections is hard to satisfy since the electoral process suffers from important information asymmetries between elected politicians and citizens and lack of politicians ex post accountability. Hence, "marketization" poli- cies of some segments of infrastructure industries, including the privatization of government monopolies, liberalization, and the application of private man- agement principles to state-owned entreprises, have proved to be reforms that improve political agents' accountability in a much more targeted way. When analyzing regulatory performance, beyond giving full consideration to such pro-accountability reforms as the above marketization policies, the indepen- dence of the regulator, and other factors related to the sector's regulatory governance, we believe that it is also important to give due attention to other pro-accountability factors that are related to the governance of the economy as a whole. Our empirical study is a modest effort motivated by such a belief. 3 Econometric methodology Our empirical investigation of the impact of political accountability on reg- ulatory performance relies on a series of regressions. In each of these regres- 8See Ackerman (2005). 9As noted by Spiller and Tommasi (2003), opportunistic behavior of politicians can be expected in infrastructure industries because of the important economic stakes involved. Indeed, these industries are characterized by very large sunk costs, substantial economies of scale, and a wide domestic consumption. 8 sions, the dependent variable is a variable that measures regulatory perfor- mance and the independent variables that retain much of our attention are variables that are used to capture political accountability. In view of the framework discussed in the previous section that forms the conceptual foun- dations of this empirical study, these variables of political accountability are regrouped into variables of "local" accountability meant to reflect the quality of regulatory governance in the sector, and variables of "global" accountabil- ity meant to reflect the quality of political governance in the economy as a whole. Regulatory performance is measured by the level of output (mainline pen- etration or cellular subscription), efficiency (mainlines per employee), or price (fixed residential, cellular). Local accountability is captured in variables re- flecting the degree of political and financial independence of the regulator, the level of transparency of accounts and regulatory decisions, the clarity of the allocation of tasks among institutions, the nature of the legal environ- ment, and the degree of social participation in regulatory decisions.10 As to global accountability, it is captured in variables reflecting the quality of the institutional framework (government integrity, efficiency of bureaucracy, strength of courts and enforcement capacity, government's commitment ca- pacity, and currency risk) and the quality of the political process (strength of checks and balances).11 When estimating the relationship between political accountability and regulatory performance, we control for some other vari- ables that are deemed important such as the degree of privatization of the incumbent and the level of competition. We also account for endogeneity when it is appropriate to do so. 10Thus, unlike most analyses of the impact of the reforms in infrastructure sectors, this study accounts for a large set of regulatory governance dimensions. Exceptions are Gutierrez (2003b) and Holder and Stern (1999) who have constructed detailed indices of regulators' characteristics in Latin American countries for the telecom sector, and in Asian countries for the electricity sector, respectively. These dimensions have been emphasized in the literature (see, e.g., Estache and Martimort, 1999) as important for regulatory agencies to be sustainable. 11Both the empirical and theoretical literatures suggest that it is not so much the extent of democracy that is relevant to investors but rather the ability of the government to credibly commit to a policy regime. To capture the level of policy stability, we choose to use an index that indicates whether there is an "effective" number of checks and balances. 9 Given the type of our data which are time-series-cross-sectional (TSCS), we choose to apply the Differenced Generalized Method of Moments (DIF- GMM) to estimate the relationships of interest. Lagrange multiplier and Wald tests applied to the data support the presence of dynamics and fixed effects which suggested to us the use of this method developed by Arellano and Bond (1991) for analyzing panel data and applied by Beck and Katz (2004) to TSCS data.12 A typical relationship is specified as a dynamic equation given by log(yit) = 0 + 1 log(yit ) + xit + µi + -1 it (1) where i = 1,2,...,N, t = 1,2,...,T, yit is a one-dimensional dependent variable representing regulatory performance, 0 and 1 are scalar parame- ters, xit is a vector of regressors representing, among other things, politi- cal accountability, is the associated vector of parameters, µi captures a country-specific fixed effect, and itis a disturbance term.13 The indices i and t refer to the country and the year respectively. For both data sets used in the analysis T = 15.14 For the data set on developing countries, N = 29, and for that on developed countries, N = 23.15 The following (standard) assumptions are made: E(µi) = 0, E( ) = 0, E( µ ) = 0, E(yi it it i 1 it) = 0 (2) In this setting, estimation can be potentially plagued by endogeneity com- ing from a correlation of two types: a correlation between the regressors and the fixed effect term, on the one hand, and a correlation between the regres- sors and the disturbance term, on the other hand. In our context, one might expect a possible correlation between the extent of reforms captured by some 12We are well aware of the (not yet settled) debate on the statistical properties of various methods used to fix problems due to dynamics in TSCS data. However, we choose to use DIF-GMM because, as an instrumental variables (IV) estimation technique, this method privileges consistency. 13Taking logs allows to minimize heteroskedasticity and influential outliers problems. 14These data sets cover the period 1985-1999. 15The lists of countries are given in the appendix. 10 regressors and some country characteristics such as population density and wealth which are embodied in the fixed effect term. Moreover, the regressors used to capture the degree of privatization and competition are likely to be endogenous, in particular, in the early stages of the reforms. For example, licenses are typically granted conditional on the fulfillment of some perfor- mance targets based on penetration, quality, or some other dimensions of the industry, and are often associated with exclusivity periods.16 The endogeneity problem stemming from the correlation of the first type is taken care of by merely expressing equation (1) in first differences to obtain log(yit) = 1log(yit ) + xit + -1 it (3) where is the first difference operator. However, this transformation brings with it another endogeneity problem due to the contemporaneous correlation between log(yit ) and the error term -1 it-1. But, note that this correlation is of the same nature as the correlation of the second type mentioned above.17 The question therefore boils down to finding instruments which can be used in the estimation of equation (3). We follow a standard approach in which lagged values of the potentially endogenous regressors are taken as possible instruments and then appropriate lag lengths are selected by investigating whether the disturbance term is serially uncorrelated or follows a moving average process of some order q, MA(q). In the case of a serially uncorrelated disturbance term, we have E( it is) = 0 for t = s, and the variables y and x lagged two and more periods are valid instruments.18 If the disturbance term is a MA(1), we have 16Endogeneity might also be a concern when using variables to capture some aspects of the structuration of regulation (see Laffont, 2005 for a discussion of some important factors that influence the structuration of regulatory institutions). A good example is the variable that indicates whether or not there exists an independent regulator. Indeed, the decision to create, and the timing of the creation of an independent regulator can be influenced by pre-regulatory performance. For an empirical account of the endogeneity of regulatory policies, see Gasmi and Recuero Virto (2006), Gutierrez (2003), and Ros (1999, 2003), among others. 17In fact, this problem concerns any other predetermined variable. 18Indeed, it can be seen that for T 3, E( itlog(yit )) = 0 and E(xit ) = -t -t it 11 E( it it-l) = 0 for l 1 and E( it it-l) = 0 for l > 1, and the variables y and x lagged three and more periods are valid instruments. More generally, if the disturbance term follows a MA(q), the valid instruments are y and x lagged (2 + q) and more periods.19 Another technical issue that needs to be addressed is that of stationarity of the dependent variable. Indeed, lack of stationarity can have two conse- quences in our context. A first consequence is that any estimation method applied to such a dynamic system is likely to be inaccurate.20 A second consequence has to do with the application of DIF-GMM. The available in- struments for the equation in first differences are likely to be weak which would impoverish the finite-sample properties of the estimator.21 To address stationarity, we follow Blundell and Bond (1998) who find that when series are close to non stationarity, DIF-GMM underestimates the coefficients of an autoregressive process of order one (AR(1)). For each candidate dependent variable (the regulatory performance variables), we then estimate an AR(1) with both DIF-GMM and System GMM (SYS-GMM) where the latter uses, in addition to the moment conditions used in DIF- GMM, instruments in first differences for the equation in levels (log(yit)). The use of SYS-GMM requires the following additional assumptions: 0, i = 1,2,...,N; t = 3,...,T, t = 2,...,t - 1. 19In practice, we start by using as instruments for the equation in first differences the variables log(y) and x lagged two and more periods . If the disturbance term in first differences presents no second-order autocorrelation, we are facing a serially uncorrelated disturbance term in levels which therefore says that the instruments used are valid. If the disturbance term in first differences presents a second-order autocorrelation, this indicates that, in levels, this term follows a moving average process and that the dependent variables log(y) and x lagged two periods is endogenous and hence is not a valid instrument. We then repeat the procedure by using, as instruments for the equation in first differences, the variables log(y) and x lagged n times (n 3) and more until we find no second-order autocorrelation in the disturbance term in first differences. 20For example, Beck and Katz (2004) show that with a non stationary dependent vari- able, the dispersion of the value of the coefficient in an autoregressive process of order one found with different asymptotically equivalent methods often exceeds its standard errors. 21To illustrate this point, assume that the dependent variable follows the AR(1) process log(yit) = log(yit )+µi + -1 itwith 1, i.e., the dependent variable becomes increas- ingly non stationary. Then, the instrument log(yit ) is not correlated with the regressor -2 log(yit ) in (3). Indeed, log(yit ) = ( - 1)log(yit ) + µi + -1 -1 -2 it-1 µi + it-1. 12 E(log(yi )µi) = 0, E(xi µ ) = 0, i = 1,...,N 2 2 i (4) As shown by Arellano and Bover (1995), since the SYS-GMM approach is immune to the weak instrument problem in the case of close to non station- arity, we use it as a benchmark.22 It is then possible to use as instruments in the equation in levels, the endogenous variables {y,x} lagged one pe- riod when the disturbance is serially uncorrelated, and lagged (q+1) periods when it follows a MA(q).23 As indicated in the beginning of this section, our investigation of the role of political accountability relies on a set of regressions. While the estimation of the coefficients of these regressions allows us to assess the (quantitative) impact of the political accountability variables on the regulatory performance variables, asking first whether there exists a causal relationship between these variables will allow us to meaningfully interpret this impact. We therefore perform some causality tests by combining the DIF-GMM estimation tech- nique with a Granger-causality testing procedure developed in Holtz-Eakin et al. (1988) for panel data. These tests are based on the estimation of the equation m m log(yit) = klog(yit ) + -k kxit + xit + -k it (5) k=1 k=1 which we use to see whether the variable used to capture political account- ability, x, "Granger-causes" the variable used to measure regulatory perfor- mance, y. Following Holtz-Eakin et al. (1988), we initially set the lag length 22 The way we use SYS-GMM as a benchmark is as follows. When this method yields an AR(1) coefficient greater than or equal to one, i.e., when the dependent variable is a pure non stationary stochastic process, we take first differences and check stationarity again. When SYS-GMM yields close to unit root (the dependent variable is close to being non stationary) and DIF-GMM yields a substantially smaller coefficient, then again, we work with first differences. Otherwise, i.e., when SYS-GMM doesn't yield close to unit root or yields close to unit root but DIF-GMM doesn't underestimate the AR(1) coefficient, we directly work with levels as this doesn't weaken the statistical properties of the estimator. 23 Indeed, it can be seen that E(log(yit -1-q)(µi+ )) = 0 and E(xit it -1-q (µi+ )) = it 0, i = 1,2,...,N; t = 3 + q,...,T. 13 m equal to 3 and check whether this lag length is "acceptable" by means of a Wald test of the significance of 3 and 3. If such a lag length is accepted, we test the joint significance of 1, 2, and 3 and conclude on whether x does not cause the variable y. If the lag length is not accepted, we repeat the procedure using the next smaller lag length. In the case where no lag length is accepted, we conclude that no causality running from x to y exists.24 4 Preliminary empirical analysis The purpose of this section is twofold. First, we attempt to uncover some general properties of the raw data from an examination of their descriptive statistics.25 Second, we discuss the outcome of our investigation of the sta- tionarity of the regulatory performance variables. Tables A1-A6 given in the appendix exhibit the list of variables and their designation, standard sum- mary statistics, correlation coefficients for some variables of interest, and compounded annual rates of increase for the data on developing and devel- oped countries. From Tables A2 and A5, we see that the correlations between variables of regulatory performance and political accountability are generally stronger for developing countries than for developed countries. This is particularly the case when regulatory performance is measured by mainline penetration (ml), cellular subscription (cel), and mainlines per employee (eff), and political accountability is captured by the strength of checks and balances (checks). The same is true when regulatory performance is measured by mainlines per employee and political accountability is captured by the regulatory gover- nance index (reg), and when regulatory performance is measured by price of cellular (p cel) and political accountability is captured by the quality of 24Strictly speaking, these causality tests concern the transformed variables as shown in equation (5). The political accountability variable x will represent in turn the qual- ity of the regulatory governance, of the institutional environment, and of the political process. See the appendix for a precise definition of the variables used to capture political accountability. 25This step should only be taken as a first diagnosis of the data that will, at best, suggest some of their aspects to be examined with some details. 14 the institutional environment (institutional). We also observe that, in both samples, the regulatory performance variables tend to correlate more with the variables which reflect the quality of the institutional environment than those that reflect the quality of the political process or the regulatory gover- nance. Tables A3 and A6 reveal that, when measured by mainline penetration, cellular subscription, or mainlines per employee, regulatory performance has, on average, increased twice as much in developing countries than in devel- oped countries over the 1985-1999 period. This might be due to the fact that, in the early part of the period, unmet demand was more important in developing countries. When measured by the monthly subscription to the fixed service (which has increased in both types of countries) or the price of cellular (which has decreased in both types of countries) instead, regula- tory performance seems to have improved more in developed countries. The significantly higher increase of the monthly subscription to the fixed tele- phone service in developing countries might be due to the fact that policies of tariff rebalancing have been relatively more intense in these countries. As the significantly lower decrease of price of cellular in developing countries, it might reflect a relatively less effective competition in this segment of the market as compared to developed countries. To conclude this brief check up of the data, we note the evolution of the quality of the institutional environ- ment and the political process showing a higher improvement in developing countries. However, this might only reflect the fact that these countries were lagging behind on these two dimensions. We now discuss the outcome of our investigation of the stationarity of the regulatory performance variables which will be the dependent variables of our regressions. Tables A7 and A8 given in the appendix show the results of the estimation of an AR(1) with both the DIF-GMM and SYS-GMM methods applied to the variables in levels, and then with the DIF-GMM method applied to the variables in first differences in the cases where they are found to be non stationary in levels.26 These tables give the DIF-GMM and 26A time trend is included in all AR(1) estimations to allow for stationarity around a trend. 15 SYS-GMM (one-step robust) estimates of the AR(1) coefficient, the estimate of the time trend coefficient, Time, the first- and second-order autocorrelation coefficients of the residuals in first differences, m1 and m2, the value of the J statistic for testing the validity of instruments, the value of the Dif- Sargan statistic that allows us to test the validity of the additional SYS- GMM conditions, the value of the starting lag of the instruments, L, and the number of observations actually used.27 From these two tables, we see that in almost all the AR(1) estimations, second-order autocorrelation of the residuals in first differences (m2) is re- jected using as instruments the initial lag of two periods and more for the variables in levels and one period for the variable in first differences. This confirms then the validity of these instruments. The only exception is the mainline penetration series (in first differences) in the data set on developing countries. In this case, we find empirical evidence that the disturbance term in levels follows a MA(2). The valid instruments then are the variables in levels lagged four periods and more for the equation in first differences, and the variables in differences lagged three periods and more for the equation in levels. In fact, the J test never rejects the validity of the instruments.28 We also see that the Dif-Sargan test never rejects the additional moment conditions required to use SYS-GMM. From Table A7, we see that the SYS-GMM AR(1) coefficient is greater than or equal to one for the series mainline penetration (ml), cellular sub- scription (cel), and mainlines per employee (eff), and hence conclude that these series are non stationary. Stationary is achieved when taking their first differences as can be seen from the results of DIF-GMM applied to these first differences shown at the right of the table. We therefore use these first dif- 27In all the tables presented in this paper, we indicate the significance at the 10%, 5%, and 1% confidence level by the superscript , , and respectively. Even if two- step GMM is known to be asymptotically more efficient than one-step GMM, we omit the two-step GMM estimates as we find that their asymptotic standard errors tend to be abnormally small even when we make the finite sample correction proposed by Windmeijer (2000). In fact, Arellano and Bond (1991) show by means of simulations that this apparent gain in precision might come at the cost of a downward finite-sample bias. 28Let us mention that Blundell and Bond (1999) interpret a rejection with such a J test as possibly due to measurement errors. 16 ferences in the remainder of the analysis of the data on developing countries. We further see from this table that the estimates of the AR(1) coefficient obtained with DIF-GMM applied to the series monthly subscription to fixed and price of cellular are smaller than those obtained with SYS-GMM. We conclude that the instruments for the equation in first differences are weak and hence we also use these series in first differences. Concerning the data on the developed countries, we see from Table A8 that the estimates of the AR(1) coefficient obtained with DIF-GMM applied to the series mainlines per employee (eff), monthly subscription to fixed (p res), and price of cellular (p cel) are also smaller than those obtained when SYS-GMM is applied instead. We therefore conclude again that the instruments for the equation in first differences are weak and use these series in first differences as well.29 5 Analysis of the relationship between political accountability and regulatory performance 5.1 Causality results In this subsection we address the issue of the existence of causal relationships between the variables of political accountability and regulatory performance. Tables A9-A14 given in the appendix show the DIF-GMM estimation results on which we build our testing procedure asking whether the variables of lo- cal accountability, namely, the regulatory governance index (reg) and global accountability, namely, the institutional environment index (institutional) and the index of checks and balances (checks), Granger-cause the variables of regulatory performance, namely, mainline penetration (ml), cellular sub- scription (cel), mainlines per employee (eff), monthly subscription to fixed 29In fact, for the purpose of our empirical analysis that seeks to cross-examine the results found with the developing and developed countries data sets, we ultimately use the regulatory performance series in first differences. 17 (p res), and price of cellular (p cel).30 In addition to showing the estimated values of the parameters associated with the explanatory variables listed at the left and some items already de- scribed in section 4, namely, m1, m2, J, L, and Obs., Tables A9-A14 include two Wald statistics. A first Wald statistic, Lag length, allows us to test for the joint significance of the coefficients associated with the dependent and the explanatory variables with the highest lag length. A second Wald statistic, Causality, allows us to test the joint significance of the coefficients associ- ated with the lagged political accountability variables when the Lag length test accepts the significance of the appropriate coefficients. The choice of valid instruments is made by using information contained in these tables and following the procedure discussed in section 3.31 From the results in Tables A9-A11 obtained with the data on developing countries, we see that in all estimations there exists a certain lag length which is accepted. Then, when proceeding to examine Granger-causality, Table A9 shows that regulatory governance causes regulatory performance except when using the cellular subscription or mainlines per employee variables to measure regulatory performance.32 Table A10 shows that the institutional environment causes regulatory performance independently of which of the five variables is used to measure regulatory performance. Finally, we see from Table A11 that the political process causes regulatory performance except when the latter is measured by the variables mainlines per employee or price of cellular. Table 1 below summarizes these findings on the existence of causality relationships in the data on developing countries. While some causality relationships are also found in the data on developed countries, the empirical evidence is somewhat weaker than in the case of 30We also include in our estimations some additional control variables as needed and account for any possible endogeneity problem. The estimates shown in these tables are those of the parameters of equation (5). 31See also footnote (18). In all the estimations shown in these tables, the disturbance term in levels is serially uncorrelated, except for the cellular subscription series (see Table A9) where the disturbance term follows a MA(2), and for the price of cellular series (see Tables A9 and A13) where it follows a MA(1). 32See the Causality statistic which is not significant in those two cases. 18 Table 1 Causality relationships (developing countries) Variable local accountability global accountability reg institutional checks ml Yes Yes Yes cel No Yes Yes eff No Yes No p res Yes Yes Yes p cel Yes Yes No the data on developing countries. Indeed, from the results shown in Tables A12-A14, we see that there are some estimations where no lag length and hence no Granger-causality relationship is accepted. More specifically, when testing whether regulatory governance causes regulatory performance and the latter is measured by mainline penetration or price of cellular, no lag length is accepted (see Table A12). Hence, we conclude that regulatory governance does not cause regulatory performance in either of these two cases. In the same vein, these data on developing countries do not show causality relationships between the institutional environment and regulatory performance when the latter is measured by mainlines per employee or price of cellular (see Table A13) and between the political process and regulatory performance when the latter is measured by price of cellular (see Table A14). In instances where a certain lag length is accepted, we proceed to exam- ine Granger-causality. From Table A12, we see that regulatory governance causes regulatory performance when the latter is measured by cellular sub- scription or monthly subscription to fixed. From Table A13, we see that the institutional environment causes regulatory performance when the latter is measured by mainline penetration, cellular subscription, or monthly sub- scription to fixed service. Finally, we see from Table A14, that the data on developing countries show that the political process causes regulatory per- formance only when the latter is measured by cellular subscription. Figure 2 below summarizes our discussion of the existence of causality relationships in the data on developed countries. From Tables 1 and 2, it is fair to say that, overall, the results presented 19 Table 2 Causality relationships (developed countries) Variable local accountability global accountability reg institutional checks ml No Yes No cel Yes Yes Yes eff No No No p res Yes Yes No p cel No No No in Tables A9-A14 support the proposition that, in developing as well as in developed countries, there exists a causal relationship between political ac- countability and regulatory performance. This is particularly true when we examine political accountability through the quality of the institutional en- vironment. Another interesting feature of the results is that global account- ability variables seem to be in a stronger causal relationship with regulatory performance than local accountability variables, and this is even more so in developing countries. Even though the empirical evidence of such relation- ships is admittedly stronger in the data on developing countries, we feel that the importance of the issue from a policy point of view warrants a careful analysis of the quantitative aspects of these relationships, a task which is taken up next. 5.2 Regression estimation results The preliminary analysis of the data performed so far sets the ground for a scrutiny of the relationship between political accountability and regulatory performance in the data on both the developing and developed countries. Let us briefly recall the different steps and outcomes of this analysis. We have started with a quick inspection of simple correlation coefficients between the variables used as proxies for these two concepts (see section 4). This light- handed checkup of the data has led us to conclude that there are reasons to believe that such a relationship exists indeed and is generally stronger in developing countries. The next step then has been to search in the data for evidence of a causal relationship running from political accountability to 20 regulatory performance. We have tackled this task by means of Granger- causality tests. These tests have also shown that such a causal relationship exists, although we have found a stronger empirical support for this relation- ship in the developing countries data (see subsection 5.1). In addition to bringing empirical evidence on the causal relationship between political accountability and regulatory performance, the Granger- causality tests provided us with some further information on the dynamic structure of this relationship. The end-product of this testing procedure is a list of potential variables to be included as regressors when estimating the quantitative impact of political accountability on regulatory performance. In order to minimize the risk of estimation inaccuracy, a serious threat in the context of dynamic data analysis which is ours, we made sure that, if needed, the variables used to measure regulatory performance, the dependent variables, were transformed so as to make them stationary (see section 4). Tables 3 and 4 below report DIF-GMM estimations of regressions draw- ing some of their main political accountability regressors from the set of variables that have "passed" the causality test performed in the previous section.33 The content of these two tables is similar to that of Tables A9- A14 already discussed in the previous subsection. Two additional items are appended however. First, we indicate, next to the entry "Endogenous reforms," whether the variables privatization (priva), competition in fixed (comp fix), competition in cellular (comp cel), and regulatory governance index (reg) have been included in the regressions as endogenous regressors or merely as exogenous.34 Second, we provide the value of a Wald statis- tic for testing the joint significance of time-specific effects captured in Time dummies.35 33These variables are selected on the basis of the results detailed in Tables A9-A14 and summarized in Tables 1 and 2 given in the appendix and the previous subsection respectively. For notational simplicity, in Tables 3 and 4 we take the transformations log and as implicit. 34We have already alluded to this endogeneity problem in section 3 (see also footnote 15). The decision to include these variables as endogenous, and hence to instrument them, was made on the basis of goodness-of-fit. 35Testing for the presence of time-specific effects seems particularly relevant in our context since some important events have occurred during the period under study. These events include, among others, the 1995 "Tequila" crisis, the 1997 South-asian crisis, the 1998-1999 financial breakdown, and some events related to technological progress such as the introduction of digital systems. 21 From Table 3 concerning the developing countries data, we see that, for any of the five variables used to measure regulatory performance, namely, mainline penetration (ml), cellular subscription (cel), mainline per employee (eff), monthly subscription to fixed (p res), and price of cellular (p cel), there is at least one variable used to represent political accountability which significantly impacts it. Except when regulatory performance is measured by the monthly subscription to fixed, the sign of this impact is as can be expected, i.e., the higher the political accountability, the better the regula- tory performance as reflected in higher output (increase in mainline pene- tration and cellular subscription), higher efficiency (increase in mainlines per employee), and lower prices (decrease in price of cellular). The apparently counterintuitive case where we find that higher political accountability (less risk of expropriation for operators and stronger checks and balances) leads to a higher monthly subscription to fixed service might in fact only reflect the extent of tariff rebalancing that typically takes place in developing countries during the early stages of the reforms. When we distinguish local account- ability (regulatory governance) from global accountability, it is interesting to note that the latter is more often found to have a significant impact on regulatory performance. Nevertheless, in the cases when it is found to be sig- nificant, the effect of regulatory governance on regulatory performance has the expected sign, namely, a better regulatory governance leads to a higher output and a lower price. The least we can say about the results obtained with the developed coun- tries data set is that Table 4 which presents them does not convey the same messages. A general comment that should be made at the outset is that these results are poor compared with those obtained when the developing countries data set is used. Indeed, as can be seen from Table 4, some reason- able regressions could only be found when using either mainline penetration (ml), cellular subscription (cel), or monthly subscription to fixed (p res) to measure regulatory performance. As to the impact of political accountability on regulatory performance, the only sensible results that could be recovered from the data on developed countries is a positive effect of regulatory gov- ernance (reg) on cellular subscription (cel) and a decrease in the monthly subscription to the fixed service (p res) with a lowering of the currency risk 22 to operators (currency).36 We finally note that, for developing countries where typically the divisions of powers is well balanced, the quality of the political process as reflected in the strength of checks and balances (checks) turns out not to be significant in explaining regulatory performance.37 We note that the dummies used to capture time-specific effects were al- ways significant at the 10% or lower significance level which suggests that attention should be given to important political and economic events in a country when examining the performance of regulation. We also observe that the reforms variables were used as endogenous regressors in all the regressions except when regulatory performance was measured by cellular subscription in the data set on developing countries and by the monthly subscription to fixed in the data set on developed countries. This is consistent with the idea that reforms are increasingly performance-based. To summarize, the findings suggest there are reasons to believe that local political accountability (regulatory governance) is generally an important determinant of regulatory performance in both developing and developed countries. The story is not so clear when it comes to global accountability. In the data set on developing countries, we found that the quality of the political process and the institutional environment have a favorable on regu- latory performance in terms of output, price and efficiency. In contrast, with the data set on developed countries the quality of the political process has been found not to have a significant impact on regulatory performance and the institutional environment showed even a negative impact on regulatory performance as measured by output. Tables 5 and 6 below summarize our discussion of the results on the impact of political accountability on regula- tory performance. 36Two additional effects were found significant, but with unexpected signs, namely, a lower risk of expropriation to operators was found to decrease mainline penetration and cellular subscription. 37Note that this result is consistent with the implications of the simple correlation coefficients (see Table A5). 23 Table 3 DIF-GMM parameter estimates (developing countries) yit mlit celit effit yit -1 0.247 0.322 -0.139 regit-1 0.003 corruptionit -1 0.086 0.017 bureauit -1 -0.023 0.012 lawit-1 0.003 0.017 expropriit -1 0.031 0.013 currencyit -1 -0.002 -0.003 corruptionit -3 -0.011 bureauit -3 0.003 lawit-3 0.006 expropriit -3 0.020 currencyit -3 -0.004 checksit -1 0.024 checksit -2 0.003 checksit -3 -0.001 privait 0.066 0.133 0.187 comp fixit -0.004 0.018 -0.119 comp celit 0.022 0.146 0.051 m1 -3.15 -2.61 -3.31 m2 1.55 0.33 -1.46 J 3.87 13.81 10.57 Time dummies 3.03 8.20 1.83 Endogenous reforms Yes No Yes L 5 2 3 Obs. 295 318 316 yit p resit p celit yit -1 -0.294 -0.215 regit-1 -0.010 regit-2 -0.007 corruptionit -1 0.001 -0.005 bureauit -1 -0.024 -0.002 lawit-1 0.035 0.001 expropriit -1 0.056 -0.043 currencyit -1 -0.016 -0.025 checksit -1 0.017 privait 0.185 0.869 comp fixit -0.147 0.001 comp celit 0.047 0.046 m1 -2.74 -1.78 m2 -1.62 -0.80 J 15.51 4.56 Time dummies 15.21 2.01 Endogenous reforms Yes Yes L 2 2 Obs. 152 162 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 24 Table 4 DIF-GMM parameter estimates (developed countries) yit mlit celit p resit yit -1 0.063 0.424 -0.078 regit -2 0.012 0.002 corruptionit -1 0.036 -0.004 bureauit -1 0.069 0.067 lawit-1 0.049 0.038 expropriit -1 -0.069 0.143 currencyit -1 0.012 -0.025 corruptionit -2 0.006 bureauit -2 -0.009 lawit-2 -0.005 expropriit -2 -0.007 currencyit -2 -0.001 checksit-1 0.019 privait -0.014 0.033 -0.017 comp fixit 0.014 -0.043 -0.016 comp celit -0.004 0.000 -0.044 m1 -2.72 -3.45 -3.33 m2 0.22 -2.11 -0.96 J 2.52 2.52 4.18 Time dummies 4.06 5.00 43.94 Endogenous reforms Yes Yes No L 2 2 2 Obs. 276 253 182 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. Table 5 Impact of political accountability on regulatory performance (developing countries) Variables local accountability global accountability reg institutional checks ml + NS + cel NA + + eff NA + NA p res - + + p cel - - NA Note: NA ans NS stand for not applicable and not significant respectively. 25 Table 6 Impact of political accountability on regulatory performance (developed countries) Variable local accountability global accountability reg institutional checks ml NA - NA cel + - NS eff NA NA NA p res NS - NA p cel NA NA NA Note: NA and NS stand for not applicable and not significant respectively. 6 Conclusion The quality of political institutions has long been emphasized in both the academic and the institutional spheres as being a crucial determinant of economic performance. This paper is a first attempt to draw lessons from the recent conceptual literature concerned with the role of the economy-wide governance in the shaping of regulatory outcomes and feed them into the more empirical approach that directly examines the impact of sector-wide governance on regulatory performance. Our "integrated" empirical approach rests on the idea that political accountability is a key factor in the interface between political and regulatory structures. This approach is illustrated for the case of telecommunications in developing and developed countries by analyzing the impact of political accountability variables on regulatory performance variables in two time-series-cross-sectional data sets. In this paper we have used two sets of variables to capture political ac- countability, local accountability variables and global accountability vari- ables. Local accountability variables include most of the features related to "regulatory governance," namely, unbundling of regulation from policy making, autonomy and independence of the regulator, accountability of the regulator, clarity in the allocation of mandates and attributes among gov- ernment institutions, legal aspects, transparency of regulatory practices, and participation in the regulatory process. These variables were synthesized in a regulatory governance index. Global accountability variables include variables concerning corruption, bureaucracy, law and order, expropriation, 26 currency risk, and checks and balances. We have estimated the impact of these political accountability variables on regulatory performance when the latter is measured by mainline penetration, cellular subscription, mainlines per employee, monthly subscription to the fixed, or price of cellular. Our empirical analysis of the two samples has shown a relatively weak effect of political accountability on the performance of regulation in developed coun- tries and a clear cutting effect in the case of developing countries where we found that the higher the political accountability, the better the regulatory performance. What implications can one derive from such a finding? During the last two decades, many developing countries have created reg- ulatory agencies mostly relying on advice provided by international financial institutions (IFIs) and international lawyers to implement these regulatory models. New regulatory institutions were however not tailored or customized enough to fit the local cultural, political and social endowments. Our paper once again stresses this very important requirement for success in developing new institutions. Furthermore, the paper goes beyond most current analyses in the area by extending the focus of the analysis to what we have referred to as issues of "global accountability" which reflect the quality of political institutions. Recent contributions have deepened the understanding of regulatory effec- tiveness along two dimensions. The first dimension is regulatory governance, a concept which is a bit broader than what our definition in this paper en- compasses. The second is regulatory substance, a concept which is meant to capture the way regulation is actually performed. Brown et al (2006) have proposed a comprehensive evaluation process of the effectiveness of regula- tory institutions. If implemented, this process will highlight not only the structural weaknesses but also the deficiencies stemming from the surround- ing environment of regulation, in particular, the political environment. It is thus important to devise policy mitigation instruments that incor- porate both of these dimensions. Unfortunately, common practices during the last decade or so have shown that donors' interventions are centered on structural issues. The analysis conducted in this paper clearly advocates for 27 the definition of a set of instruments of effective intervention with the ob- jective of achieving political accountability improvements in the practice of regulation. Indeed, building regulatory institutions in developing countries should be part of a broader strategy of "good governance" and not only be considered, as it has been in the past years, as a sectoral matter. International donors, including the World Bank, the Department for In- ternational Development, and others have been strong and effective advocates for good governance since many years, but a sound policy for supporting the development of politically accountable systems in developing countries has yet to be designed. The general wisdom is that in order to promote good governance one has to support the development of demand and sup- ply institutions for governance. Supply side institutions involve structural mechanisms for establishing a set of institutions with the goal of promot- ing accountability, whereas demand side institutions are those that advocate for good governance. Assuming that good governance is promoted, political accountability improves and so does the performance of regulation.38 A further aspect that needs to be highlighted is that established regula- tory agencies need long term support so that they can significantly improve regulatory practice. As opposed to the short term approach relied upon during the past years, IFIs should define long term programs to support regulatory institutions newly established so the latter can build the human capital as well as develop the technical tools and instruments required by an efficient practice of regulation. In designing reforms in the 1990s, the World Bank has usually included capacity building components in its loans to provide such support to regu- lators. However, although the intention was encouraging, this approach had suffered from two caveats. First, the approach was a short term one in that these programs assumed that newly established regulatory agencies will be- come self sustaining in five years whereas their host environments did not really support the development of such institutions. Consequently, as soon as the World Bank loan is signed or closed, most regulators did not benefit 38A country such as Chile demonstrates to some extent such a cycle. 28 from the support of their governments, were side-lined at best, or captured. Second, the approach was lacking appropriate mitigation instruments to deal with the political environment. The issue is how do we make regulation po- litically acceptable or supported. Little was done to understand the political game while establishing regulatory agencies. As a result, donors did not have a clear understanding of the political requirements to make regulation acceptable in a given country. Instead, the debate centered on ring fenc- ing regulatory agencies from political interference forgetting that regulation, in institutional terms, is no more than a delegation of power from elected officials to bureaucrats. With this in mind, technical assistance programs typically involved train- ing programs (skills building, hands on) to build up the human capital base, on the one hand, and helping the board or management of regulatory agen- cies to establish and comply with approved procedures and regulations, on the other hand. In effect, international development partners could also rely on a stick and carrot approach to catalyze necessary changes at the political level. Doing this would imply working only with those countries which are committed to improving political accountability. To sum up, future reforms should not only devote attention to improving regulatory governance (structural requirements, regulatory substance), but should pay much more attention to understanding the political context within which regulatory institutions will be performing in. In developed countries, as our quantitative results have shown, political accountability is already well established and practiced through an effective use by the electorate of its votes as a sanction tool. The focus therefore in those countries is on regulatory governance. In developing countries, political accountability is at an early stage of development and hence this calls for additional means and resources from development partners to promote good governance which will in turn enhance the quality of regulation. 29 Appendix · Data on developing countries A first data set contains observations on the following list of 29 devel- oping countries during the period 1985-1999: India, Sri Lanka, Malaysia, Pakistan, Thailand, C^ote d'Ivoire, Ghana, Kenya, Malawi, Tanzania, Uganda, South Africa, Jordan, Morocco, Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, Guatemala, Honduras, Jamaica, Panama, Peru, El Salvador, and Venezuela. Information have been collected on variables regrouped in five cate- gories: Regulatory performance, local accountability, global account- ability, and other variables. The designation of these variables, their sources, and their precise definition are given below. Regulatory performance Variable Source(s) Output · Mainline penetration ITU · Cellular subscription ITU Efficiency · Mainlines per employee ITU Price · Monthly subscription to fixed ITU · Price of cellular ITU 30 Local accountability Variable Source(s) structuration of regulation · Separation of the regulator -Bortolotti et al. (2001), Fink et al. (2002), Gutierrez (2003a), Ros (2003). -ITU World Telecommunica- tions Regulatory database. -Clark et al. (2004). autonomy/independence · Regulator's budget -Clark et al. (2004). · Can members of the regulatory commis- -Idem sion be fired by the executive? · Can the minister/president veto the reg- -Idem ulator's decisions? · Has the minister/president written pol- -Idem icy guidelines during the past year? Accountability · Is accounting separation mandatory? -Idem · Can the operator appeal if it disagrees -Idem with regulator's decisions? · Can other parties appeal? -Idem clarity of allocation of tasks · Who is in charge of resolving (intercon- -Idem nection) disputes? · Who controls pricing? -Idem · Who controls the procedure of licence -Idem granting? · Who decides on the number of licenses -Idem to be granted? · Who controls the procedure of spectrum -Idem allocation? Legal aspects · What type of approval is required for -Idem private firms in order to operate? participation/transparency · Are regulatory meetings open to the -Idem public? · Are explanations of regulatory decisions -Idem published? Regulatory governance · Regulatory governance index -Index computed from values of the previous local account- ability variables. 31 Global accountability Variable Source(s) Institutionalization · Corruption -IRIS dataset by Steve Knack and Philip Keefer for the IRIS Center at the University of Mary- land (1982-1997). -International Country Risk Guide (ICRG) risk ratings (1997-1999). · Bureaucracy -Idem. · Law and order -Idem. · Expropriation -IRIS dataset by Steve Knack and Philip Keefer for the IRIS Center at the University of Mary- land (1982-1997). · Currency risk -Exchange Rate Stability, International Country Risk Guide (ICRG) risk ratings (1985-1999). · Institutional en- -Index computed from values of the previous in- vironment index stitutionalization variables. Quality of the political process · Checks and bal- -DPI2000 Database of Political Institutions ances 1975-2000, Philip Keefer (Development Re- search Group), The World Bank (2002). 32 Other variables Variable Source(s) · Privatization -Various authors (Ros, 1999, 2003, Bortolotti et al., 2001, McNary, 2001, Li and Xu, 2004, Fink et al., 2002). -ITU World Telecommunications Regulatory database. -Operators and regulators websites. -Clark et al. (2004). -Private Partcipation in Infrastructure (PPI) Project World Bank database. -IPANeT Privatization Transactions data- base (World Bank). · Competition in -Various authors (Ros, 1999, 2003, Bortolotti fixed et al., 2001, McNary, 2001, Li and Xu, 2004, Fink et al., 2002). -ITU World Telecommunications Regulatory database. -Operators and regulatory authorities web- sites. -Clark et al. (2004). -http://www.gsmworld.com. · Competition in cel- -Idem lular Regulatory performance - Output . Mainline penetration: Number of telephone lines per 100 in- habitants that connect the subscribers' terminal equipment to the Public Switched Telephone Network (PSTN). . Cellular subscription: Number of users of portable telephones subscribing to a mobile telephone service with access to the PSTN. - Efficiency . Mainlines per employee: Number of mainlines per employee in the fixed service activity. - Price . Monthly subscription to fixed: Recurring fixed charge (in 2000 US dollars) paid by residential subscribers to the PSTN. This 33 charge covers only the rental of the line, not that of the ter- minal. . Price of cellular: Price (in 2000 US dollars) paid for a 3-minute call during peak hours from a cellular telephone. For reasons of inter-country comparability, this price corresponds to that of a call placed with a pre-paid card. Local accountability - Structuration of regulation . Separation of the regulator: Dichotomous variable which takes on the value 1 if the regulatory agency is separated from and not directly controlled by a ministry or a utility, and 0 other- wise. - Autonomy/Independence . Regulator's budget: Trichotomous variable which takes on the value 1 if the regulatory agency is financed from licence fees or donors contributions, 0 if it is financed from the general budget of the government, and 0.5 if it is financed from both types of sources. . Can members of the regulatory commission be fired by the executive?: Dichotomous variable with value 1 if the answer to the question is "no," and 0 if the answer is "yes." . Can the minister/president veto the regulator's decisions?: Dichotomous variable which takes on the value 1 if the answer to the question is "no," and 0 if the answer is "yes." . Has the minister/president written policy guidelines during the past year?: Dichotomous variable with value 1 if the an- swer to the question is "no," and 0 if the answer is "yes." - Accountability . Is accounting separation mandatory?: Dichotomous variable which takes on the value 1 if the answer to the question is "yes," and 0 if the answer is "no." 34 . Can the operator appeal if it disagrees with regulator's deci- sions?: Dichotomous variable which takes on the value 1 if the answer to the question is "yes," and 0 if the answer is "no." . Can other parties appeal?: Dichotomous variable with value 1 if the answer to the question is "yes," and 0 if the answer is "no." - Clarity of allocation of tasks . Who is in charge of resolving (interconnection) disputes?: Di- chotomous variable with value 1 if the answer to the question is "the regulator," and 0 if the answer is "the ministry" or "nobody." . Who controls pricing? (this variable concerns pricing of fixed (local), domestic long distance, international, ISP, and mobile services): Trichotomous variable which takes on the value 1 if the answer is "the regulator," 0 if the answer is "the ministry" or "nobody," and 0.5 if the answer is "both the regulator and the ministry." . Who controls the procedure of license granting? (this con- cerns licenses for fixed (local), domestic long distance, inter- national, ISP, and mobile services): Trichotomous variable with value 1 if the answer to the question is "the regulator," 0 if the answer is "the ministry," and 0.5 if the answer is "both the regulator and the ministry." . Who decides on the number of licences to be granted?: Tri- chotomous variable with value 1 if the answer to the question is "the regulator," 0 if the answer is "the ministry" or "no- body," and 0.5 if the answer is "both the regulator and the ministry." . Who controls the procedure of spectrum allocation?: Dichoto- mous variable which takes on the value 1 if the answer to the question is "the regulator," and 0 if the answer is "the min- istry." 35 - Legal aspects . What type of approval is required for private firms in order to operate?: Trichotomous variable which takes on the value 1 if the answer to the question is "a formal approval," 0 if the answer is "no approval at all," and 0.5 if the answer is "just a notification." - Participation/Transparency . Are regulatory meetings open to the public?: Trichotomous variable with value 1 if the answer to the question is "yes, all of them," 0 if the answer is "not at all," and 0.5 if the answer is "yes, some of them." . Are explanations of regulatory decisions published?: Dichoto- mous variable which takes on the value 1 if the answer to the question is "yes," and 0 if the answer is "no." - Regulatory governance . Regulator governance index: Variable which takes on the value 0 when the value of the variable separation of the reg- ulator is 0, i.e., when regulation is directly exercised by a ministry or a utility. When the variable separation of the reg- ulator takes on the value 1, i.e., when there exists a separated regulatory agency, this regulatory governance index takes on a value between 1 and 15 computed as the sum of the values taken by the local accountability variables described above which are 0, 0.5, or 1. Higher values of this index reflect better regulatory governance. Global accountability - Institutionalization . Corruption: Variable with values ranging from 0 to 10 and meant to reflect the degree of corruption of the political sys- tem. The higher the value of the variable, the less corrupt the political system. The particular concern here is with actual or 36 potential corruption in the form of excessive patronage, nepo- tism, job reservations, favors for favors, secret party funding, and close ties between politicians and business. . Bureaucracy: Variable with values between 0 and 10 used to assess the quality of the bureaucracy. Higher points are at- tributed to countries where the bureaucracy has the strength and expertise to govern without drastic changes in policies or interruption in government services. . Law and order: Variable taking values between 0 and 10. The "Law" part of this variable is used to assess the strength and impartiality of the legal system (e.g., due to the existence of a strong judiciary system). The "Order" part gives an indication of the popular observance of the law (e.g., low crime rate or law not routinely ignored as with illegal strikes without effective sanctions). Higher values of this variable reflect a better judiciary system. . Expropriation: Variable with values in the range 0-10 meant to assess the risk of expropriation of private investments in terms of outright confiscation or forced nationalization. Higher values of this variable reflect less risk of this type for opera- tors. . Currency risk: Variable taking values between 0 and 10 which captures the risk of operators stemming from exchange rate fluctuations. Again, higher values of this variable reflect a lower risk of this type. . Institutional environment index: Variable whose value is found by summing the values taken by the five institutionalization variables described above. Hence, the values of this institu- tional environment index are in the range 0-50. Higher values of this index reflect a better overall institutional environment. - Quality of the political process . Checks and balances: Variable with values in the range 0- 18 meant to give some indication on the division of powers.39 39This variable "..counts the number of veto players in a political system, adjusting for 37 Higher values of this variable reflect more balanced division of powers and, accordingly, a better functioning of the political process. Other variables . Privatization: Variable giving the % of the incumbent's assets sold to private investors. . Competition in fixed: Dichotomous variable which takes on the value 0 if the local segment (fixed) is a monopoly, and 1 if there are two or more operators in this segment. . Competition in cellular: Variable which takes on the value 0 if no license for cellular (analogue and digital) has been issued, 1 if one licence has been issued, 2 if two licenses have been issued, and 3 if three or more licenses have been issued. · Data on developed countries A second data set concerns the following list of 23 developed countries during the same period 1985-1999: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Ger- many, Greece, Iceland, Ireland, Italy, Japan, Luxembourg, Nether- lands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom and United States. Given the many similarities between this data set and the developing countries data set described at length above, below we give information only on items that are different in the developed countries data set. whether these veto players are independent of each other, as determined by the level of electoral competitiveness in the system, their respective party affiliations, and the electoral rules." (Henisz and Zelner, 2002). It assumes constant returns to additional voters. 38 Local accountability Variable Source(s) structuration of regulation · Separation of the regulator -Trends in Telecommunica- tion Reform 1999: Conver- gence and Regulation. ITU. -ITU World Telecommunica- tions Regulatory database. autonomy/independence · Regulator's budget -Telecommunications Regula- tions: Institutional Structures and Responsibilities. OCDE 2005. · Overruling of the regulator's decisions -Idem clarity of allocation of tasks · Who resolves (interconnection) dis- -Idem putes? · Who authorizes interconnection -Idem charges? · Who controls pricing? -Idem · Who controls quality of service? -Idem · Who is responsible of the issuing of li- -Idem cences? · Who is in charge of allocating the spec- -Idem trum? Regulatory governance · Regulatory governance index -Index computed from values of the previous local account- ability variables. Other variables Variable Source(s) · Privatization -Various authors (Ros, 1999, McNary, 2001, Li and Xu, 2004). -Trends in Telecommunication Reform 1999: Con- vergence and Regulation. ITU. · Competition in -Various authors (Ros, 1999, McNary, 2001, Li and fixed Xu, 2004). -Trends in Telecommunication Reform 1999: Con- vergence and Regulation. ITU. · Competition in cel- -Idem lular 39 Local accountability - Structuration of regulation . Separation of the regulator: Dichotomous variable which takes on the value 1 if there exists a separated regulatory agency not directly controlled by a ministry or a utility, and 0 otherwise. - Autonomy/Independence . Regulator's budget: Trichotomous variable which takes on the value 1 if the regulatory agency is financed from licence fees or operators' contributions, 0 if it is financed from the general budget of the government, and 0.5 if it is financed from both types of sources. . Overruling of the regulator's decisions: Dichotomous variable which takes on the value 1 if there do not exist organizations other than the courts that can overrule the regulator's deci- sions, and 0 if such organizations exist. - Clarity of allocation of tasks . Who resolves (interconnection) disputes?: Dichotomous vari- able with value 1 if the answer to the question is "the regu- lator," "the competition authority," or "the courts," and 0 if the answer is "the ministry." . Who authorizes interconnection charges? (this concerns in- terconnection with the networks of operators with significant market power): Dichotomous variable which takes on the value 1 if the answer to the question is "the regulator" or "the competition authority," and 0 if the answer is "the min- istry" or "nobody." . Who controls pricing?: Dichotomous variable with value 1 if the answer to the question is "the regulator" or "the compe- tition authority," and 0 if the answer is "the ministry." . Who controls quality of service?: Dichotomous variable with value 1 if the answer to the question is "the regulator" or "the competition authority," and 0 if the answer is "the ministry" or "nobody." 40 . Who is responsible of the issuing of licences? (this concerns licenses for fixed and mobile services): Trichotomous variable with value 1 if the answer to the question is "the regulator for both types of licenses," 0 if the answer is "the ministry for both types of licenses," and 0.5 if the answer is "the regulator for one license and the ministry for the other." . Who is in charge of allocating the spectrum?: Trichotomous variable which takes on the value 1 if the answer to the ques- tion is "the regulator," 0 if the answer is "the ministry," and 0.5 if the answer is "both the regulator and the ministry." - Regulatory governance . Regulatory governance index: This variable takes on the value 0 when there is no separation between the regulator and the ministry or the utility. When such a separation exists, this variable takes on a value between 1 and 8 calculated as the sum of the values taken by the local accountability variables described above. Other variables . Privatization: Dichotomous variable which takes on the value 1 if the assets of the incumbent have been partly (or totally) sold to private investors, and 0 if the incumbent is State-owned. . Competition in fixed: Dichotomous variable with value equal to 1 if there is more than one operator in the local segment (fixed), and equal to 0 if this segment is a monopoly. . Competition in cellular: Dichotomous variable with value 1 if there is more than one operator in the cellular segment (analogue and digital), and 0 if this segment is a monopoly. 41 · Descriptive statistics Table A1 Summary statistics (developing countries) Variable Designation Obs. Mean Std. Dev. Min. Max. ml Mainline penetration 435 5.27 4.96 0.11 22.36 cel Cellular subscription 431 0.81 2.09 0 15.96 eff Mainlines per employee 424 68.87 58.85 7.78 371.16 p res Monthly subscription to fixed 256 5.71 4.23 0 21.29 p cel Price of cellular 324 0.37 0.53 0 2.24 reg Regulatory governance index 435 2.59 4.60 0 13.5 corruption Corruption 435 5.07 1.43 1.66 10 bureau Bureaucracy 420 4.84 1.86 1.66 10 law Law and order 435 4.98 2.06 0 10 expropri Expropriation 420 7.24 2.00 2 10 currency Currency risk 435 5.88 1.98 1 10 institutional Institutional environment index 435 27.60 7.10 8 41.16 checks Checks and balances 423 3.12 2.06 1 18 priva Privatization 435 0.16 0.32 0 1 comp fix Competition in fixed 435 0.09 0.29 0 1 comp cel Competition in cellular 435 1.05 1.10 0 3 Table A2 Correlation coefficients (developing countries) ml cel eff p res p cel institutional 0.41 0.65 0.42 0.23 0.60 checks 0.34 0.39 0.36 -0.01 0.30 reg 0.19 0.57 0.30 -0.06 0.61 42 Table A3 Compounded annual rates of increase (developing countries) Country global accountability regulatory performance institutional checks ml cel eff p res p cel India 2.57 10.40 14.18 118.61 13.76 -2.83 - Sri Lanka 3.62 3.71 14.65 40.02 14.07 5.70 -2.23 Malaysia -0.13 2.07 8.90 28.66 12.71 -7.86 -11.25 Pakistan 2.77 8.16 11.75 57.12 11.56 13.14 -5.53 Thailand 2.38 1.31 14.75 14.65 13.03 -8.03 -11.26 C^ote d'Ivoire 0.15 5.07 7.14 105.48 10.62 -12.26 - Ghana 5.42 8.17 7.68 78.50 13.40 -9.51 - Kenya 0.20 8.16 3.89 74.52 1.42 6.25 -34.55 Malawi 0.47 10.40 2.92 171.59 -4.78 -15.62 19.35 Tanzania 3.05 8.16 4.80 89.82 8.63 -4.46 63.99 Uganda 5.85 -9.50 2.91 129.95 6.82 44.94 - South Africa 0.13 5.07 4.55 72.68 10.54 -3.97 -9.59 Jordan 3.77 0 5.30 72.03 5.72 -2.08 -7.83 Morocco 3.73 0 11.84 85.92 7.44 10.86 -16.24 Argentina 4.18 6.76 5.82 74.78 14.10 6.49 1.04 Bolivia 10.49 12.18 6.18 169.71 2.76 -0.24 -17.66 Brazil -0.36 -1.58 7.60 81.47 9.98 54.21 -11.90 Chile 3.43 10.40 11.61 81.66 8.31 -7.48 - Colombia 1.23 1.60 7.62 60.50 5.67 -8.84 -4.66 Costa Rica 1.60 5.07 7.67 62.27 7.03 -0.18 -12.28 Dominican Rep. 3.52 12.18 11.60 64.18 8.29 76.53 - Ecuador 0.13 0 8.37 78.41 9.66 19.69 8.53 Guatemala 5.60 5.07 9.18 139.78 13.17 -2.15 -13.03 Honduras 3.79 0 10.47 32.64 11.71 -7.73 - Jamaica 4.20 0 13.22 137.08 12.17 19.81 0.40 Panama 2.29 2.07 5.50 89.33 4.84 -19.55 - Peru 4.46 -4.83 8.57 143.06 17.42 27.36 6.58 El Salvador 6.49 2.07 10.82 71.02 15.60 45.10 -5.95 Venezuela 1.89 5.07 3.03 85.97 6.67 22.61 -12.49 Average 3.00 4.04 8.36 85.97 9.39 8.27 -3.64 : Figures computed for the period 1995-1999. : Figures computed for the period 1993-1999. 43 Table A4 Summary statistics (developed countries) Variable Designation Obs. Mean Std. Dev. Min. Max. ml Mainline penetration 345 48.09 10.87 14.52 73.56 cel Cellular subscription 344 8.92 13.51 0 63.37 eff Mainlines per employee 345 168.59 57.53 43.48 358.76 p res Monthly subscription to fixed 252 13.87 4.70 5.60 26.27 p cel Price of cellular 192 1.37 0.86 0 4.95 reg Regulatory governance index 345 2.62 3.11 0 8 corruption Corruption 345 8.73 1.37 3.33 10 bureau Bureaucracy 345 9.30 1.33 4.5 10 law Law and order 345 9.42 1.11 5 10 expropri Expropriation 345 9.73 0.66 4.6 10 currency Currency risk 345 8.68 1.16 4 10 institutional Institutional environment index 345 45.88 3.99 25.26 50 checks Checks and balances 345 4.46 1.62 2 16 priva Privatization 345 0.38 0.48 0 1 comp fix Competition in fixed 345 0.23 0.42 0 1 comp cel Competition in cellular 345 0.33 0.47 0 1 Table A5 Correlation coefficients (developed countries) ml cel eff p res p cel institutional 0.63 0.24 0.22 0.28 0.01 checks 0.07 0.04 0.01 0.12 0.24 reg 0.43 0.55 0.05 0.01 -0.07 44 Table A6 Compounded annual rates of increase (developed countries) Country global accountability regulatory performance institutional checks ml cel eff p res p cel Australia 0.30 1.60 1.98 28.04 3.98 -1.94 -4.86 Austria 0.31 0 2.10 82.20 2.34 -1.49 -24.03 Belgium -0.50 -5.87 3.36 91.28 4.36 4.86 -11.73 Canada 0.29 0 2.16 26.64 4.30 5.11 -3.51 Denmark 0.29 0 2.31 33.17 1.50 -0.98 -12.39 Finland 0.29 -2.03 1.52 33.30 1.66 10.75 -16.69 France -0.47 1.60 2.36 100.73 2.63 6.23 9.97 Germany 0.72 -1.58 2.48 50.22 4.37 -5.63 -4.72 Greece 3.48 0 3.78 93.65 6.91 2.67 - Iceland 0.59 -2.03 3.25 52.23 5.02 3.31 - Ireland 0.33 0 6.24 78.79 7.24 -1.92 7.37 Italy 0.86 0 3.02 66.69 5.50 10.78 -9.26 Japan -0.55 0 1.94 48.11 6.63 -0.07 -25.33 Luxembourg 0.07 -1.58 4.03 64.84 3.31 10.25 -26.66 Netherlands 0.14 1.60 2.97 86.97 -0.19 2.08 -5.32 New Zealand -0.02 0 1.35 38.32 12.18 -0.76 17.35 Norway 0.25 0 1.83 28.53 0.01 0.74 -16.49 Portugal 2.40 0 7.93 92.03 9.15 -0.81 -14.54 Spain 1.10 2.07 3.81 98.37 4.97 -4.30 5.77 Sweden 0.37 2.07 1.13 26.55 4.33 2.31 -5.07 Switzerland -0.24 1.07 2.47 60.92 1.14 -1.43 -21.80 United King. 0.35 -2.03 3.07 46.92 4.53 2.67 -0.39 United States -0.10 0 2.41 24.85 1.74 -1.70 -38.31 Average 0.44 -0.17 2.94 59.19 4.26 1.77 -9.55 : Figures computed for the period 1995-1999. : Figures computed for the period 1993-1999. · Stationarity of regulatory performance series 45 Table A7 Stationarity tests of regulatory performance variables (developing countries) log(mlit) DIF-GMM SYS-GMM log(mlit) DIF-GMM log(mlit ) -1 0.785 1.024 log(mlit )-1 0.382 Time 0.018 0.003 Time 0.001 m1 -2.22 -2.62 m1 -3.45 m2 1.46 1.07 m2 2.10 J 22.84 27.34 J 26.99 Dif-Sargan 4.5 L 2 2 L 4 Obs. 377 406 Obs. 348 log(celit) DIF-GMM SYS-GMM log(celit) DIF-GMM log(celit ) -1 0.965 1.020 log(celit ) -1 0.308 Time 0.044 0.045 Time 0.024 m1 -0.72 -0.82 m1 -2.45 m2 0.31 0.22 m2 0.73 J 28.95 28.52 J 27.91 Dif-Sargan -0.43 L 2 2 L 2 Obs. 371 401 Obs. 342 log(effit) DIF-GMM SYS-GMM log(effit) DIF-GMM log(effit ) -1 0.751 1.015 log(effit )-1 0.023 Time 0.028 0.005 Time 0.004 m1 -2.57 -2.59 m1 -3.26 m2 -1.51 -1.48 m2 -1.16 J 27.90 27.40 J 26.55 Dif-Sargan -0.50 L 2 2 L 2 Obs. 359 391 Obs. 328 log(p resit) DIF-GMM SYS-GMM log(p resit ) -1 0.680 0.804 Time -0.009 0.003 m1 -2.39 -2.47 m2 0.34 0.34 J 26.30 25.20 Dif-Sargan -1.1 L 2 2 Obs. 190 220 log(p celit) DIF-GMM SYS-GMM log(p celit ) -1 0.566 0.955 Time -0.015 -0.001 m1 -1.26 -1.97 m2 -0.10 -0.46 J 19.03 19.03 Dif-Sargan 0 L 2 2 Obs. 217 262 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 46 Table A8 Stationarity tests of regulatory performance variables (developed countries) log(mlit) DIF-GMM SYS-GMM log(mlit ) -1 0.883 0.931 Time 0.002 0.001 m1 -1.58 -1.62 m2 0.27 0.22 J 18.85 18.74 Dif-Sargan -0.11 L 2 2 Obs. 299 322 log(celit) DIF-GMM SYS-GMM log(celit ) -1 0.924 0.973 Time 0.042 0.037 m1 0.48 0.38 m2 -2.24 -2.28 J 20.50 22.20 Dif-Sargan 1.7 L 2 2 Obs. 298 321 log(effit) DIF-GMM SYS-GMM log(effit ) -1 0.685 0.906 Time 0.012 0.001 m1 -1.76 -1.39 m2 0.98 0.78 J 21.98 22.22 Dif-Sargan 0.24 L 2 2 Obs. 299 322 log(p resit) DIF-GMM SYS-GMM log(p resit ) -1 0.518 0.842 Time 0.004 -0.003 m1 -3.10 -3.11 m2 -1.13 -1.17 J 21.73 22.41 Dif-Sargan 0.68 L 2 2 Obs. 205 228 log(p celit) DIF-GMM SYS-GMM log(p celit ) -1 0.660 0.807 Time -0.021 -0.020 m1 -2.15 -2.30 m2 -0.55 -0.68 J 18.40 19.27 Dif-Sargan 0.87 L 2 2 Obs. 123 152 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 47 · Causality relationships 48 Table A9 Causality tests for local accountability variables (developing countries) yit log(mlit) log(celit) log(effit) yit-1 0.249 0.278 -0.164 yit-2 0.039 -0.253 yit-3 -0.264 regit-1 0.002 0.001 -0.001 regit-2 -0.001 -0.004 regit-3 0.001 corruptionit 0.005 0.054 0.009 bureauit -0.001 -0.008 0.016 lawit -0.001 -0.001 0.001 expropriit 0.003 0.010 0.002 currencyit -0.004 0.011 0.009 checksit 0.004 -0.016 0.009 privait 0.075 0.346 0.205 comp fixit -0.025 -0.025 -0.101 comp celit 0.004 0.139 0.018 m1 -11.36 -2.24 -3.03 m2 1.44 0.88 -0.43 J 227.75 14.06 14.44 L 2 4 2 Obs. 268 318 275 Lag length 7.02 2.47 7.97 Causality 2.19 0.14 0.47 yit log(p resit) log(p celit) yit-1 -0.303 0.562 yyit -2 -0.295 -0.254 it-3 regit-1 -0.016 0.003 regit 0.003 -0.018 regit-2 -3 corruptionit -0.001 0.035 bureauit 0.036 -0.121 lawit 0.024 0.002 expropriit 0.059 -0.344 currencyit 0.021 -0.002 checksit 0.014 0.011 privait 0.055 1.138 comp fixit -0.169 1.105 comp celit -0.014 0.061 m1 -2.05 1.64 m2 -1.64 0.71 J 16.26 1.10 L 2 3 Obs. 123 124 Lag length 3.84 13.90 Causality 3.35 4.15 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 49 Table A10 Causality tests for institutionalization variables (developing countries) yit log(mlit) log(celit) log(effit) yit-1 0.257 0.232 -0.097 yit-2 0.067 yit-3 -0.248 regit 0.001 0.001 -0.003 institutionalit -1 0.002 0.014 0.005 institutionalit -2 0.002 institutionalit -3 -0.004 checksit 0.004 -0.004 0.010 privait 0.074 0.258 0.172 comp fixit -0.023 0.046 -0.114 comp celit 0.06 0.037 0.011 m1 -3.51 -2.41 -3.14 m2 1.06 0.79 -1.69 J 14.08 23.48 22.39 L 2 2 2 Obs. 278 330 316 Lag length 12.06 4.51 5.84 Causality 5.71 6.36 3.25 yit log(p resit) log(p celit) yit-1 -0.311 -0.055 yyit -2 -0.255 it-3 regit -0.009 0.005 institutionalit -1 0.013 -0.021 institutionalit -2 0.007 institutionalit -3 checksit 0.019 -0.001 privait 0.019 0.736 comp fixit -0.181 -0.016 comp celit -0.009 0.003 m1 -2.11 -1.89 m2 -1.56 -0.33 J 18.75 8.08 L 2 2 Obs. 124 174 Lag length 6.07 6.62 Causality 4.29 9.35 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 50 Table A11 Causality tests for quality of the political process variables (developing countries) yit log(mlit) log(celit) log(effit) yit -1 0.245 0.342 -0.212 yit -2 0.069 -0.273 yit -3 -0.215 regit -0.001 0.007 -0.003 corruptionit 0.04 0.056 0.007 bureauit 0.001 -0.012 0.018 lawit -0.001 0.006 0.001 expropriit 0.005 -0.013 -0.001 currencyit -0.003 0.012 0.009 checksit -1 0.002 0.017 0.007 checksit -2 0.003 0.009 checksit -3 -0.004 privait 0.074 0.160 0.199 comp fixit -0.026 0.039 -0.092 comp celit 0.010 0.093 0.019 m1 -3.39 -2.44 -2.87 m2 1.69 0.44 -0.50 J 8.58 18.08 10.12 L 2 2 2 Obs. 265 318 274 Lag length 9.85 4.87 7.89 Causality 6.33 5.42 1.57 yit log(p resit) log(p celit) yyyit it-1 -0.314 -0.149 -2 it-3 regit -0.013 0.014 corruptionit 0.004 0.003 bureauit 0.013 -0.046 lawit 0.008 0.039 expropriit 0.031 -0.166 currencyit 0.012 -0.011 checksit -1 0.016 0.002 checksit -2 checksit -3 privait 0.007 0.859 comp fixit -0.184 0.030 comp celit 0.014 -0.044 m1 -2.62 -1.38 m2 -1.43 -0.15 J 17.27 6.11 L 2 2 Obs. 150 160 Lag length 6.40 8.00 Causality 5.31 0.09 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 51 Table A12 Causality tests for local accountability variables (developed countries) yit log(mlit) log(celit) log(effit) yit-1 0.061 0.680 -0.521 yit-2 -0.307 -0.353 yit-3 -0.235 regit-1 -0.001 0.002 -0.009 regit-2 0.012 -0.001 regit-3 -0.009 corruptionit 0.006 -0.030 0.109 bureauit -0.014 0.003 -0.055 lawit -0.005 -0.007 -0.039 expropriit 0.002 -0.004 0.256 currencyit -0.001 0.001 -0.016 checksit -0.002 0.020 0.063 privait -0.011 0.050 -0.050 comp fixit 0.014 -0.005 0.127 comp celit -0.003 0.046 0.088 m1 -2.63 -3.86 -2.49 m2 0.85 0.97 -0.33 J 12.51 4.54 5.28 L 2 2 2 Obs. 276 252 230 Lag length 0.32 8.38 5.39 Causality 9.17 0.48 yit log(p resit) log(p celit) yit-1 -0.187 -0.024 yyit -2 -0.2456 it-3 regit-1 0.018 0.013 regit -0.020 regit-2 -3 corruptionit -0.011 0.021 bureauit 0.096 0.004 lawit 0.058 0.064 expropriit -0.191 -0.099 currencyit 0.038 -0.013 checksit 0.041 0.002 privait -0.164 -0.066 comp fixit -0.121 -0.099 comp celit 0.041 -0.029 m1 -3.16 -1.84 m2 1.84 -1.45 J 14.08 7.34 L 2 2 Obs. 159 99 Lag length 17.84 1.61 Causality 2.92 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 52 Table A13 Causality tests for institutionalization variables (developed countries) yit log(mlit) log(celit) log(effit) yit -1 0.101 0.615 -0.268 yit -2 0.032 -0.244 yit -3 -0.034 regit 0.001 0.015 -0.008 institutionalit -1 0.001 -0.017 0.006 institutionalit -2 -0.003 0.012 institutionalit -3 0.022 checksit -0.001 0.011 0.045 privait -0.017 0.051 -0.054 comp fixit 0.007 -0.014 -0.001 comp celit -0.007 0.033 0.013 m1 -2.53 -3.93 -2.01 m2 0.73 -0.13 -0.81 J 16.61 9.68 17.22 L 2 2 2 Obs. 253 229 276 Lag length 3.52 3.97 2.08 Causality 3.57 4.07 yit log(p resit) log(p celit) yyyit it-1 -0.080 -0.091 -2 it-3 regit 0.001 -0.003 institutionalit -1 -0.016 0.001 institutionalit -2 institutionalit -3 checksit 0.038 0.001 privait -0.055 -0.073 comp fixit 0.023 -0.079 comp celit 0.026 -0.060 m1 -3.46 -1.94 m2 -1.27 -1.68 J 18.49 11.32 L 2 3 Obs. 182 99 Lag length 2.73 0.42 Causality 4.91 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 53 Table A14 Causality tests for quality of the political process variables (developed countries) yit log(mlit) log(celit) log(effit) yit-1 -0.001 0.718 -0.453 yit-2 0.203 -0.280 -0.301 yit-3 -0.229 regit 0.002 0.013 -0.013 corruptionit 0.007 -0.033 0.097 bureauit -0.048 0.057 -0.019 lawit 0.001 0.022 -0.017 expropriit 0.025 0.096 0.122 currencyit -0.003 0.007 -0.014 checksit -1 -0.001 0.016 0.058 checksit -2 0.001 -0.013 0.025 checksit -3 -0.021 privait -0.007 0.027 -0.044 comp fixit -0.003 -0.032 0.065 comp celit -0.004 0.049 0.084 m1 -2.64 -3.82 -1.63 m2 0.73 0.22 -1.49 J 10.82 15.87 3.74 L 2 2 2 Obs. 253 252 230 Lag length 3.47 1.74 5.72 Causality 0.24 3.57 1.54 yit log(p resit) log(p celit) yit -0.153 -0.019 yyit -1 -2 it-3 regit -0.001 -0.003 corruptionit -0.030 0.017 bureauit 0.099 0.012 lawit 0.066 0.060 expropriit -0.112 -0.056 currencyit 0.034 0.005 checksit -1 0.031 -0.012 checksit -2 checksit -3 privait -0.135 -0.056 comp fixit -0.063 -0.051 comp celit -0.007 -0.006 m1 -3.68 -1.85 m2 -1.27 -1.43 J 13.22 4.38 L 2 2 Obs. 182 99 Lag length 4.14 1.14 Causality 2.04 Note: The starting lag for the instruments is L and (L - 1) for the equation in first differences and levels respectively. 54 References Ackerman, J.M., 2005, "Social accountability in the public sector: A conceptual discussion," Social Development Paper 82, The World Bank. Arellano, M. and S.R. Bond, 1991, "Some tests of specification for panel data: Monte Carlo evidence and an application to em- ployment equations," Review of Economic Studies, 58: 277- 97. ----------­ and O. Bover, 1995, "Another look at the instru- mental variable estimation of error-component models," Jour- nal of Econometrics, 68: 29-51. Banerjee, A. and A.J. Ros, 2000, "Telecommunications privati- zation and tariff rebalancing: Evidence from Latin America," Telecommunications Policy, 24: 233-252. Beck, N. and J. Katz, 2004, "Time-series-cross-section issues: Dy- namics," Paper presented at the 2004 Annual Meeting of the Society for Political Methodology, Stanford University. Blundell, R. and S. Bond, 1998, "Initial conditions and moment restrictions in dynamic panel data models," Journal of Econo- metrics, 87(1): 115-143. -----------------------, 1999, "GMM estimation with persis- tent panel data: An application to production functions," The Institute for Fiscal Studies Working Paper Series No. W99/4. Bortolotti, B., DSouza, J., Fantini, M., and W.L. Megginson, 2001, "Sources of performance in privatized firms: A clinical study of the global telecommunications industry," Working Paper Nota di Lavoro 26.2001, Fondazione Eni Enrico Mattei, Corso Magenta 63-I-20123, Milano, Italy. Brown, A.C., Stern, J., and B. Tenenbaum with D. Gencer, 2006, Handbook for evaluating infrastructure regulatory sys- tems. World Bank Publications. 55 Clark, G., Haggarty, L., Kaneshiro, R., Noll, G.R., Shirley, M., Wallsten, S., and L.C. Xu, 2004, "New tools for studying net- work industry reforms in developing countries: The telecom- munications and electricity regulation database," AEI-Brook- ings Joint Center Related Publication 04-05. Cubbin, J. and J. Stern, 2005a, "Regulatory effectiveness and the empirical impact of variations in regulatory governance: Elec- tricity industry capacity and efficiency in developing coun- tries," World Bank Policy Research Working Paper 3535. ------------------­, 2005b, "Regulatory effectiveness: The im- pact of regulation and regulatory governance arrangements on electricity industry outcomes," World Bank Policy Research Working Paper 3536. Estache, A. and D. Martimort, 1999, "Politics, transactions costs, and the design of regulatory institutions," World Bank Policy Research Paper 2073. Fink, C., Mattoo, A., and R. Rathindran, 2002, "An assessment of telecommunications reform in developing countries," World Bank Policy Research Working Paper 2909. Gasmi, F. and L. Recuero Virto, 2006, "Telecommunications de- ployment and reforms in developing countries: A simultaneous approach," Mimeo, IDEI, Universit´e de Toulouse. Gutierrez, L. H., 2003a, "Regulatory governance in the Latin American telecommunications sector," Utilities Policies, 11: 225-240. ---------------, 2003b, "The effect of endogenous regulation on telecommunications expansion and efficiency in Latin Amer- ica," Journal of Regulatory Economics, 23: 257-286. Heller, W.B. and M.D. McCubbins, 1996, "Politics, institutions, and outcomes: Electricity regulation in Argentina and Chile," Journal of Policy Reform 1: 357-387. Henisz, W.J. and B.A. Zelner, 2002, "Problems with existing 56 measures of country risk," Mimeo, Wharton School of Busi- ness. Holder, S. and J. Stern, 1999, "Regulatory governance: Criteria for assessing the performance of regulatory systems," Utilities Policies, 8(1): 35-50. Holtz-Eakin, D., Newey, W., and H.S. Rosen, 1988, "Estimating vector autoregressions with panel data," Econometrica, 56: 1371-95. Laffont, J.J., 2005, Regulation and development, Cambridge Uni- versity Press. Levy, B. and P.T. Spiller, 1994, "The institutional foundations of regulatory commitment: A comparative analysis of telecom- munications regulation," Journal of Law, Economics and Or- ganization, 10: 201-246. Li, W. and L.C. Xu, 2004, "The impact of privatization and com- petition in the telecommunications sector around the world," Journal of Law and Economics, Vol XLVII. McCubbins, M.D. and T. Schwartz, 1984, "Congressional over- sight overlooked: Police patrol vs fire alarms," American Jour- nal of Political Science, 28: 165-179. ---------------, Noll, R.G., and B.R. Weingast, 1987, "Admin- istrative procedures as instruments of political control," Jour- nal of Law, Economics and Organization, 3: 243-277. McNary, R., 2001, "The network penetration effects of telecom- munications privatization and competition," Mimeo, Stanford University. North D.C., 2000, "Institutions and the performance of economies over time," Paper presented at the Second Annual Global De- velopment Conference, Tokyo, Japan, December 10-13, 2000. Ros, A.J., 1999, "Does network and competition matter?: The effects of telecommunications reform on network expansion and efficiency," Journal of Regulatory Economics, 15: 62-92. 57 ---------, 2003, "The impact of the regulatory process and price cap regulation in Latin American telecommunications mar- kets," Review of Network Economics, 2(3). Spiller, P.T. and M. Tommasi, 2003, "The institutions of regula- tion: An application to public utilities," Handbook of Telecom- munications Economics, Vol 2. Windmeijer, F., 2000, "Efficiency comparisons for a system-GMM estimator in dynamic panel data models," in Heijmans, R.D.H., Pollock, D.S.G., and A. Satorra (eds), Innovations in Multi- variate Statistical Analysis. 58