WPS6559 Policy Research Working Paper 6559 Do Infrastructure Reforms Reduce the Effect of Corruption? Theory and Evidence from Latin America and the Caribbean Liam Wren-Lewis The World Bank Development Economics Vice Presidency Partnerships, Capacity Building Unit August 2013 Policy Research Working Paper 6559 Abstract This paper investigates the interaction between association is reduced when an independent regulatory corruption and governance at the sector level. A simple agency is present. These results survive a range of model illustrates how both an increase in regulatory robustness checks, including instrumenting for regulatory autonomy and privatization may influence the effect of governance, controlling for a large range of observables, corruption. The interaction is analyzed empirically using and using several different corruption measures. The a fixed-effects estimator on a panel of 153 electricity association between corruption and productivity also distribution firms across 18 countries in Latin America appears weaker for privately owned firms compared to and the Caribbean from 1995–2007. Greater corruption publicly owned firms, though this result is somewhat less is associated with lower firm labor productivity, but this robust. This paper is a product of the Partnerships, Capacity Building Unit, Development Economics Vice Presidency. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at liam.wren-lewis@parisschoolofeconomics.eu. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Do infrastructure reforms reduce the effect of corruption? Theory and evidence from Latin America and the Caribbean Liam Wren-Lewis e Libre de Bruxelles∗ Paris School of Economics and ECARES, Universit´ JEL: D73, L33, L51, O18 Keywords: Corruption, Infrastructure, Governance Sector Boards: EM, PSM ∗ Research Fellow, Paris School of Economics, 48 Boulevard Jourdan, Paris, France and As- sociate Fellow, ECARES, Universit´ e Libre de Bruxelles, Brussels, Belgium. Email: liam.wren- lewis@parisschoolofeconomics.eu The author would like to thank Luis Andr` es, Antonio Estache, Luis Guti´ errez and Mart´ın Rossi for provid- ing access to data and Steve Bond for the helpful comments and suggestions made throughout the course of this work. I have also bene�ted from discussions, comments and suggestions from Luis Andr` es, Daniel Clarke, Simon Cowan, Antonio Estache, Maitreesh Ghatak, Clotilde Giner, Clare Leaver, Elisabeth Sadoulet, Stephane Saussier, Francesc Trillas, Bruno Versailles, John Vickers, participants at various seminars and conferences and three anonymous referees. A supplemental appendix to this article is available at http://wber.oxfordjournals.org/. 1 Corruption is a major problem that can reduce growth and worsen productivity (Bardhan 1997; Lambsdorff 2005). One type of activity that is particularly vulnerable to corruption is o 2006; Es- the operation of network infrastructure (Bergara, Henisz, and Spiller 1998; Dal B´ tache and Trujillo 2009). As a result, practitioners and researchers have become increasingly interested in ways to reduce the impact of corruption on infrastructure performance (Estache and Wren-Lewis 2011). However, little evidence exists on whether the major sectoral reforms that have been implemented have had a signi�cant influence on the effects of corruption. Two important components of reform have been privatization and the creation of independent reg- ulatory agencies (IRAs). This paper analyzes the impact of these reforms with regard to their interaction with national corruption levels. A simple model demonstrates how ownership and regulatory autonomy may influence the effects of corrupt behavior. The resulting proposi- tions are analyzed empirically by considering the productivity of electricity distribution �rms in Latin America and the Caribbean (LAC) over the period 1995–2007. A number of previous studies have focused on the effect of corruption on infrastructure per- o and Rossi (2007), who formance. The study closest to the work of this paper is that of Dal B´ �nd that corruption is associated with inef�ciency among electricity distribution �rms in Latin America. However, these authors do not focus on how this association interacts with regulation and privatization, partly due to a lack of data on regulatory governance.1 Another set of papers have used recently collected data on regulatory governance to consider the impact of regulatory errez (2003a), Montoya and Trillas (2007), Cubbin and reforms in more detail, including Guti´ Stern (2006), and Zhang, Parker, and Kirkpatrick (2008). In particular, Andres, Azumendi, and Guasch (2008) produce evidence that better regulatory governance and privatization increase 1 Other studies include Guasch and Straub (2009), who examine the effect of corruption on renegotiation, and Estache, Goicoechea, and Trujillo (2009), who consider the impact of corruption on country-level measures of access, affordability, and quality. Clarke and Xu (2004) take a different approach by considering the effect of reforms on petty bribery to utility �rms. 2 the productivity of electricity distribution �rms in LAC.2 However, they do not consider the role of corruption. The main contribution of this paper is its evaluation of how the impacts of regulatory au- tonomy and privatization are related to corruption. The question is empirically interesting because theoretically, the interaction of corruption with these reforms is ambiguous (Boehm 2009; Martimort and Straub 2009). Overall, the empirical analysis suggests that regulation by an IRA signi�cantly reduces the association between corruption and inef�ciency. Indeed, variations in countries’ corruption levels appear to explain a substantial proportion of the heterogeneity in the effects of both regulatory independence and privatization. The analysis uses annual �rm-level data on 153 electricity distribution �rms across 18 coun- tries along with detailed measures of their respective IRA’s governance. To control for time- invariant omitted variables, the regressions use a �rm �xed effects model. The main results are robust to a range of permutations, allowing for �rm-speci�c corruption effects and including a large range of additional control variables. Moreover, the results are robust to instrumenting for the reform variables and corruption. The negative association of corruption and its interaction with independent regulation also remain when two alternative corruption measures are used, one of which is based on �rm surveys and the other on observed corruption in Brazil. In con- trast, the interaction between corruption and private ownership appears somewhat less robust, with the relevant coef�cient losing signi�cance when other corruption measures are used. The paper proceeds as follows. In the following section, a simple model illustrates how corruption, privatization, and the regulatory structure may interact in their effect on labor pro- ductivity. In this model, regulatory autonomy decreases the ability of corruptible politicians 2 For surveys of the empirical literature on privatization in developing countries, see Parker and Kirkpatrick (2005); Megginson and Sutter (2006); Boubakri, Cosset, and Guedhami (2008). 3 to interfere in the regulatory process. Privatization works through a different mechanism: it reduces the proportion of corrupt proceeds that are used to over-employ. Section II describes the data and outlines the empirical methodology used, which is based on estimating a labor demand function. The results are analyzed in Section III, both graphically and econometri- cally. Details of several robustness checks are provided in Section IV, including the addition of a large range of control variables, instrumentation for key variables, and the use of alternative corruption measures. The section also explores the effect of the relevant variables on measures of quality and prices. Finally, Section V concludes and suggests policy lessons. I Theoretical Model To provide a framework for the empirical analysis, a simple model is constructed that out- lines a potential mechanism through which corruption may interact with regulatory autonomy and ownership in its impact on productivity. The model is intended to provide one potential framework for thinking about the link between corruption, labor productivity, regulatory inde- pendence, and privatization. It is not meant to suggest that this is the only way in which these variables interact; in reality, the relationship is likely to be multifaceted. The model is a static game among four actors: society, a regulated �rm, and two interme- diaries whose role is to supervise and regulate the �rm on behalf of society - a politician and a regulator. In this model, corruption is de�ned as occurring when one or more of the interme- diaries collude with the �rm in a way that is detrimental to society. Such corruption is made possible by an asymmetry of information between society and the other actors regarding the �rm’s costs.3 3 The model therefore follows a tradition of modeling corruption in a three-tier hierarchy, as in Spiller (1990) and Laffont and Tirole (1991). Within this literature, other models also consider a situation with multiple su- pervisors (see, for example, Laffont and Martimort (1999)), and Estache and Wren-Lewis (2009) argue that it is important to consider the multiplicity of actors involved in the regulatory process. 4 The model focuses on the amount of labor employed by the �rm, L, to produce a required level of output, Q. The focus on labor is reasonable because capital inputs are closely related to both the number of connections and the geographical area of distribution and are therefore treated in the literature on electricity distribution as exogenous in the short run (Neuberg 1977; o and Rossi 2007). The required output level Q Kumbhakar and Hjalmarsson 1998; Dal B´ can also be treated as exogenous because it represents the obligation to provide electricity as o demanded to a given set of consumers, which is the mandate of the �rms in the sample (Dal B´ and Rossi 2007). Therefore, I consider productivity to be labor productivity in both the model and the subsequent empirical analysis because this is the variable most likely to be under the �rms’ direct control. To produce the output Q, the �rm must employ at least L(Q) people. From the point of view of society, there is some uncertainty over the number of people the �rm needs to employ. With probability ν , the �rm only needs to employ L(Q) workers to produce Q, but with probability 1 − ν, the �rm must employ (1 + γ )L(Q) workers. Assuming an exogenous wage rate w, the �rm’s total costs are thus either wL(Q) or w(1 + γ )L(Q). The �rm receives revenue from society to cover its costs; this revenue may arise fromcharging consumers or receiving a transfer from the government. Society wishes to minimize the revenue that the �rm receives on the condition that the �rm has enough revenue to cover its costs. Society has two agents in charge of deciding the �rm’s revenue, a regulator and a politician, both of whom are aware of the �rm’s true costs. It is assumed that society can strongly punish agents who set a �rm’s revenue above (1 + γ )wL(Q) because such behavior is against society’s interests in all situations. The range between wL(Q) and (1+ γ )wL(Q) therefore represents the discretion given to the supervisors. It is assumed that, to a certain extent, the politician is moti- vated to act in the interests of society but may also be motivated by promises of personal gain. In particular, I consider a situation in which the �rm can make an offer to the politician that 5 involves sharing part of any excess revenue that the �rm receives.4 The politician may either be honest or dishonest, with the probability of dishonesty being φ(c), where c is a measure of the overall level of corruption in the country. It is assumed that the probability that the politician is dishonest is increasing in the national corruption level in the country, i.e., φ (c) > 0. If the politician is honest, then he or she will never accept a bribe. If the politician is dishonest, then he or she would rather accept the bribe and attempt to set the �rm’s revenue at the maximum that he or she can get away with. We assume that information on bribes paid and the �rm’s true costs are unveri�able “soft� information and hence cannot be credibly transmitted to society. If the regulator is not independent of the politician (e.g., the regulator is located in a gov- ernment ministry), then we assume that the politician has de facto control of the �rm’s revenue and that the regulator does not play a signi�cant role. However, if the regulator is independent, then with probability α, the regulator will block any attempt by the politician to set the revenue above the �rm’s true costs.5 Consider, for instance, a case in which the �rm is over-paying for some input and is purchasing this input from a company owned by the politician. If the regula- tor is within the politician’s ministry, then the regulator is more likely to bow to the politician’s will to allow the purchase than if the regulator is in an independent organization. The �rm’s payoff function is such that it wishes to maximize its revenue. If its revenue is greater than its costs, the excess will be split among any bribed agents and parties within the �rm. A proportion 0 < π < 1 of the excess revenue that the �rm receives is distributed among various actors as “pro�ts�, bribes, or greater income for employees. The remaining proportion 1 − π is spent on employing a greater number of workers than necessary. This re- mainder represents the part of the pie given to employees whose wages are relatively inflexible 4 For the purposes of the model, it is not necessary to specify the size of this share. We simply require that the share needed to persuade the politician is suf�ciently small that there is always enough money to bribe. 5 We assume that the regulator cannot overrule the politician in the other direction - that is, set revenue high when the politician is not corrupted. In a more general model where this is possible, the creation of an independent regulator would have ambiguous effects with regard to corruption. 6 and who would otherwise be unemployed and essentially provides the link between the �rm receiving excess revenue (through corruption) and over-employment. To keep the model gen- eral, I abstract from the exact mechanism through which this split occurs. One possibility is that managers who successfully pursue corrupt activities expend less effort on ensuring ef�- o and Rossi (2007). Alternatively, if the the workers are cient employment levels, as in Dal B´ aware of the �rm’s true costs, they may need to receive a share of the corruption proceeds to be persuaded not to blow the whistle. The timing of the game is then as follows: 1. The �rm’s true labor needs and the supervisor’s corruptibility is made known to all actors except society. 2. The �rm may offer a (revenue-contingent) bribe to the politician. 3. The �rm’s revenue is set. 4. Any excess revenue received by the �rm is shared among the relevant parties. Let us now solve the model for the expected level of labor employed in equilibrium. With probability 1 − ν , the �rm is high cost, and hence the labor employed will be (1 + γ )L(Q). In contrast, with probability ν , the �rm is low cost. In this case, we need to evaluate the probability that the �rm is allowed to receive excess revenue. If the politician is honest, then he or she will not accept a bribe, and the �rm will receive no more revenue than necessary. In contrast, if the politician is dishonest, then he or she will accept a bribe and attempt to set revenue at the higher level. If the regulator is not independent, then the politician will succeed with certainty, whereas if the regulator is independent, the politician will succeed only with probability 1 − α. Hence, conditional on the �rm being low cost, the overall probability of the �rm being allowed 7 excess revenue is therefore φ(c)(1 − α1IRA ), where 1IRA is an indicator function that takes the value 1 when the regulator is independent and zero otherwise. Given this situation, the expected amount of labor employed is L = L(Q) [1 + (1 − ν )γ + νγ (1 − π )φ(c)(1 − α1IRA )] . (1) In this equation, we can see that the “excess� labor employed is proportional to the average revenue received corruptly. Taking logs and then approximating gives the following equation: ln(L) ≈ ln(L(Q)) + (1 − ν )γ + (1 − π )φ(c)(1 − α1IRA )γ. (2) The effect of a change in corruption levels on the log of labor employed can therefore be gathered by differentiating this equation: dln(L) ≈ (1 − π )φ (c)(1 − α1IRA )γ. (3) dc Having derived a relationship between corruption and employment, let us now consider the role of policy reforms.6 Privatization typically involves transferring �rm ownership from the state to an organization that is focused on maximizing pro�ts. The change in ownership is likely to create an extra outlet for the �rm’s excess revenue: owners’ pro�ts. In the context of the model, privatization can therefore be viewed as an increase in π .7 6 For the purposes of this model, we abstract from the reasons why these reforms are implemented. Because corrupt politicians may lose out from the reforms, one potential explanation is that these reforms are a response to pressure from outside actors, including citizens and international bodies. See Section IV for a brief discussion of such motivations. 7 This effect is similar to the effects of privatization in Shleifer and Vishny (1994), where privatization decreases the relative influence of those pushing for excess labor compared to pro�t-motivated managers. Bai et al. (2000) also discusses why state-owned �rms may be required to use their funds to over-employ. 8 Independent regulatory agencies are independent in the sense that they are not part of a government ministry or subject to direct executive control. Therefore, they are viewed as less sensitive to the will of political elites (Andres, Azumendi, and Guasch 2008). Their role is to implement regulatory policy, which may include setting tariffs, publishing information on �rms’ performance, and enforcing agreed standards of quality and supply. In the model above, it is therefore reasonable to assume that this reform creates the potential for the regulator to “block� corrupt political behavior.8 Modeling the two reforms in this way leads to the following proposition: Proposition 1. For a given output Q, labor employed by the �rm is increasing in the national level of corruption. Moreover, • (a) This effect is greater if the �rm is public rather than private. • (b) This effect is smaller if the regulator is independent. A rise in the national corruption level increases the probability that the �rm will be able to bribe the politician. This situation then increases employment because part of the gains that the �rm makes from this corruption will be shared with labor through excess employment. Privatization reduces the effect of corruption in this model because fewer of the corrupt gains are distributed to workers through excess employment. Note, however, that privatiza- tion does not reduce the amount that the �rm receives as a result of corrupt behavior; hence, consumers and taxpayers do not bene�t directly from this increase in productivity. An IRA’s 8 An alternative model could consider the idea that a regulator’s independence might directly impact the way in which rents are shared within the �rm in a way similar to privatization. For example, more transparent regulation may reduce the rents that workers obtain from holding more information than owners or politicians. In such a model, greater independence could then reduce the impact of corruption on labor productivity without an effect on revenue. However, such a model is not investigated here because revenue setting is generally considered to be a regulator’s primary role. 9 creation reduces the politician’s power and diminishes the effect of the politician’s corruption on productivity. The interaction between corruption and IRA creation is therefore negative. Overall, the model provides a simple framework for considering the potential relationship among corruption, labor productivity, and infrastructure reforms. An important point is that the model does not generate welfare implications. Because any change in productivity also results in a redistribution of revenues, reforms may simultaneously increase productivity and decrease welfare. II Data and Empirical Strategy In this section, we move to the empirical analysis. After describing the data sources used, we describe the econometric methodology, followed by a discussion of the identifying assumptions and the manner in which we should interpret our results. Data The empirical analysis is based on the electricity distribution sector in countries in Latin Amer- ica and the Caribbean from 1995–2007. The electricity distribution sector has many of the properties that are typical of network infrastructure, including close government regulation and limited direct competition. Moreover, the period includes a number of important reforms as well as substantial variation in the level of corruption, both within and between countries. Data on �rm performance are from the World Bank Latin American and Caribbean Electric- ity Distribution Benchmarking Database. This database contains data on 249 utilities across 25 countries between 1995–2007. Overall, the �rms represent 88% of all electricity connections in the region. Each utility �rm operates in only one country. The main analysis uses data on the total number of employees, the total number of connections, total electricity sold (in GWh), 10 Table 1: Summary statistics Panel A: Firm characteristics Mean Std. Dev. Minimum Maximum Employees 1,337 3,479 12 40,970 Connections 668,958 1,771,628 2,499 23,265,575 Electricity (GWh) 3,619 11,201 3 140,283 Interruption frequency (No. per year) 35 61 0 533 Interruption duration (hrs per year) 33 61 0 705 % of electricity lost 16 10 2 72 Avg. residential tariff ($) 84 30 11 177 Avg. industrial tariff ($) 75 25 9 147 Panel B: Corruption index Argentina 0.22 0.41 -0.29 0.71 Bolivia 0.14 0.51 -0.29 0.71 Brazil 0.01 0.61 -1.29 0.88 Chile -0.97 0.64 -1.79 0.21 Colombia 0.30 0.55 -0.29 1.21 Costa Rica -0.99 1.42 -2.29 1.21 Dominican Republic 0.19 0.90 -1.29 0.71 Ecuador -0.23 0.43 -0.70 1.09 El Salvador -0.51 0.67 -1.29 0.21 Guatemala -1.23 0.10 -1.29 -1.12 Haiti 1.38 0.35 0.80 1.71 Honduras 0.61 0.21 0.21 0.80 Jamaica 1.21 0.00 1.21 1.21 Mexico 0.32 0.51 -0.70 0.71 Nicaragua -0.25 0.68 -1.29 0.21 Panama 0.71 0.00 0.71 0.71 Peru -0.07 0.43 -0.95 0.71 Uruguay -0.29 0.00 -0.29 -0.29 Full sample 0.00 0.82 -2.29 1.71 Panel C: Regulators and �rms IRA No. of Firms, by ownership Start year ERGI Private Public Changed Argentina 1996a 0.64a 2 2 3 Bolivia 1996 0.84 1 0 6 Brazil 2000a 0.71a 10 4 21 Chile 1990 0.56 23 0 0 Colombia 1994 0.76 0 16 4 Costa Rica 1996 0.74 0 8 0 Dominican Republic 1998 0.75 0 0 2 Ecuador 1999 0.61 0 19 1 El Salvador 1997 0.82 1 0 4 Guatemala 1996 0.79 1 0 0 Haiti 1983 0.37 0 1 0 Honduras 1995 0.56 0 1 0 Jamaica 1997 0.72 1 0 0 Mexico 1995 0.72 0 2 0 Nicaragua 1985 0.75 0 0 2 Panama 1996 0.63 0 0 3 Peru 1996 11 0.84 2 7 7 Uruguay 2000 0.73 1 0 0 Overall median/total 1997 0.72 41 61 53 Source: World Bank; International Country Risk Guide a For Argentina/Brazil, regulatory statistics given are the median of the province/state regulators. and whether the �rm is privately managed. Summary statistics of the �rms’ characteristics are provided in Panel A of Table 1. Data on corruption are from the International Country Risk Guide, which contains annual country-level data. I use this dataset because it is speci�cally designed to allow for compar- isons between years and countries and contains observations for the entire period for which there are data on �rms’ performance. The ICRG corruption index is designed to capture the likelihood that government of�cials will demand special payments and the extent to which il- legal payments are expected throughout government tiers as ranked by panels of international experts. The ICRG index ranges globally between 6 (highly clean) and 0 (highly corrupt). To clarify the results, I reverse the ordering of the data such that greater values represent higher levels of corruption, and I transform the data such that the mean level of corruption in the total sample is 0. A positive value therefore represents an environment in which corruption is above the sample average, whereas a negative value represents a level of corruption below the sample average. Panel B of Table 1 provides summary statistics of the variables by country. Data on regulatory governance are from Andres et al. (2007) and include information on national electricity regulators in more than 20 countries as well as information on provincial regulators for certain states in Brazil and provinces in Argentina. The data are compiled from a survey containing more than 50 different questions to produce indices of various aspects of regulatory governance, including accountability, autonomy, and transparency. These include questions such as whether the regulator is �nanced directly by the government, whether minutes are available publicly, and the way in which the head of the agency is appointed. I primarily use the Electricity Regulatory Governance Index (ERGI) constructed by Andres et al. (2007), where a rating of 0 represents the worst possible measure of governance and 1 represents the best. For Argentina and Brazil, I use data on the provincial and state regulatory agencies because the regulation of electricity distribution �rms is conducted at this level. Henceforth, 12 the term “province� means the area for which the regulatory agency is responsible (the country, state, or province, as appropriate). Panel C of Table 1 provides summary statistics of the regulatory governance index (ERGI) and when the agencies were created. The data are cross- sectional, but because they include the year in which each regulatory agency was created, I transform the data into a panel by giving zero values to all variables in each year before the agency’s creation.9 In total, these three data sources combine to create a database of 153 �rms across 18 coun- tries with a total of 1,359 observations (this is the largest possible intersection of the three datasets). Panel C of Table 1 shows the number of �rms of each type in each country. Of the 153 �rms, 53 change ownership over the period (all but three from public to private), whereas 66 begin in the sample without a regulator and then become regulated. Econometric Methodology Kumbhakar and Hjalmarsson (1998) note that although productivity in electricity generation is mainly determined by technology, productivity in distribution is largely driven by management and ef�cient labor use. Moreover, because electricity distribution is highly regulated, decisions on technology and capital are likely to be outside of the �rm’s control, whereas the �rm typ- ically has control over labor. I therefore focus on labor productivity. Electricity distribution �rms in the sample have the obligation to meet demand. We can therefore consider the amount of electricity sold to �nal customers and the number of �nal customers served to be exogenous outputs. o and Rossi (2007) in estimating a parametric labor requirement function. I follow Dal B´ 9 I am therefore implicitly assuming that regulatory governance remains constant during the reign of the agency and that it is unrelated to the quality of regulation prior to the creation of the agency. This is obviously a strong assumption, but if it has any effect on my results, it is likely to bias them toward insigni�cance. Therefore, it should not be of great concern when interpreting the results. 13 In particular, I use a translog functional form because it provides a good second-order approx- imation to a broad class of functions. Included in this function are the number of electricity connections the �rm serves and the amount of electricity it sells. This equation for a panel of i = 1, , N �rms producing in c = 1, , C countries and observed over t = 1, , T periods may therefore be speci�ed as follows: i,t i,t i,t 2 i,t 2 i,t i,t li,t = αi + ψt + ω1 y1 + ω 2 y2 + ω3 y1 + ω4 y 2 + ω 5 y1 y2 (4) +β1 Corc,t + β2 Corc,t ∗ P rii,t + β3 Corc,t ∗ IRAi,t + β4 P rii,t + β5 IRAi,t + i,t , where l, y1 , and y2 are the natural logarithms of labor, sales, and customers, Cor is the level of national corruption, P ri is a dummy variable for private ownership, IRA is a dummy variable indicating whether the �rm is regulated by an IRA, the ω and β terms represent the coef�cients on the variables, and is the random error term. Because each �rm is present in only one country, there is a unique mapping from �rm i to country c. Firm �xed effects are included to control for any time-invariant unobservables, represented as αi in the equation above.10 Because each �rm is present in only one country/province, this approach controls for any time-invariant unobservables at the country/province level. Hence, the corruption and regulation terms are estimated based only on changes in these variables within countries/provinces over time. To account for time effects in a flexible way, I include year �xed effects ψt . The year �xed effects measure the productivity impact of sector-level shifts over time, such as secular technology trends, international macroeconomic fluctuations, 10 In addition to electricity produced and connections, Dal B´ o and Rossi (2007) also include the service area as an exogenous output and transformer capacity and the length of the distribution lines as exogenous capital variables. Unfortunately, the �rst two of these variables are not available in the extended dataset used here, and including the latter reduces my sample by more than half. However, because these variables vary little over time, they are likely to be controlled for by using �rm �xed effects. I test for this by performing the regressions with the length of distribution lines included in the translog function and �nd the variable to be insigni�cant, with no signi�cant changes in my results. Moreover, using the dataset from Dal B´ o and Rossi (2007), I �nd that their results are insensitive to the removal of the service area and transformer capacity from the labor demand function. 14 or energy price shocks. Standard errors are clustered on country-year combinations to address the concern that the shocks affecting �rms in a given country in the same year may be corre- lated. Overall, the estimated equation resembles Equation (2) from the theoretical model. Propo- sition 1 predicts that β1 > 0 and β2 < 0, with |β1 | > |β2 | and β3 > −β1 . Although the model also suggests that there should be an interaction between ownership and regulation, unfortu- nately, this cannot be analyzed empirically in this study because there are very few observations of unregulated private �rms in the sample. Discussion The identi�cation strategy described above exploits variation over time in corruption and in- frastructure reforms. This variation may be driven by a range of factors, including changes within government, the number of years since the government has been elected, and policy makers learning about best practices.11 Because we cannot precisely identify the causes of such variation, we must be cautious when drawing inferences from the results. In particular, factors that lead to variations in corruption and reform may also affect �rms’ productivity in other ways. To the extent that such alternative channels are observable, we can rule them out by adding extra control variables. I undertake such robustness checks in Section IV. However, a particular concern may be that countries carry out the observed reforms as part of a package of changes designed to improve performance or reduce corruption, and other (unobserved) parts of the package may thus have a signi�cant impact. If this is the case, we cannot identify how many of the resulting productivity changes are the results of the measured reforms and how many of these changes relate to other parts of the reform package. Thus, when interpreting the 11 Spiller and Tommasi (2005), for instance, argue that regulatory independence is more likely to occur when there is a greater division of powers and less preference homogeneity across government. 15 results, it is important to bear in mind that it is not necessarily the particular aspect of reform measured that is driving changes in productivity, and it should not be assumed that the reform in question would have exactly the same impacts elsewhere (Joskow 2005). Nonetheless, it is useful for guiding both policy and future research to understand whether signi�cant changes in productivity have accompanied these sector reforms. If the aspects of reform measured here explain signi�cant productivity changes better than other observables, this brings us closer to understanding how policy can affect �rm performance. A second concern that may arise with the outlined methodology is the risk of reverse causal- ity, which would arise if changes in �rms’ productivity affected our independent variables. In the case of corruption, reverse causality seems unlikely to be a concern. Sector-speci�c shocks will generally not affect country-level corruption, whereas economy-wide shocks are likely to be captured by the use of control variables such as GDP per capita in Section IV. However, reverse causality could be a problem in the case of sectoral reforms because under-performing �rms might encourage such reforms. However, this is likely to be more of an issue over the longer term because reform decisions are unlikely to be strongly influenced by poor perfor- mance over a small number of years. This issue is therefore unlikely to bias our results given the use of �rm �xed effects. However, to check for reverse causality, an instrumental variable approach is used in Section IV. Finally, it should be noted that in the model above, corruption and infrastructure reforms are treated as independent of one another. In reality, the two may well be correlated - for example, corrupt governments may reform less, or corruption may be reduced through a broad package of regulatory reforms. For our empirical methodology, however, this is not a serious problem because both variables appear simultaneously on the right-hand-side of our regression. 16 III Empirical Results Before beginning the econometric analysis using the methodology described above, we present the data graphically to consider the links among inef�ciency, corruption, and regulatory gover- nance. Graphical Analysis For this subsection, inef�ciency is measured by regressing the log of employees on the translog function described in the �rst line of Equation (4) and storing the residuals.12 This creates a measure of “excess labor�, which gives us an idea of how productive the �rm is in any year compared to the average of all �rms over the whole period. Figure 1 plots excess labor against corruption separately for �rms in four different environ- ments. The upper two panels consider observations of �rms operating under either no IRA or a “bad� IRA (i.e., below-median ERGI score), whereas the lower two panels consider observa- tions of �rms operating under “good� IRAs (i.e., an above-median ERGI score). These pairs are then divided into publicly operated �rms on the left panel and privately operated �rms on the right. Both reforms appear to affect the relationship between corruption and inef�ciency (Figure 1). The upper left panel, where �rms are publicly owned and not regulated by an above-average IRA, shows the clearest positive relationship between corruption and inef�ciency. The upper right and lower left panels show that the relationship between corruption and inef�ciency is weaker for �rms that are either privately operated or regulated by an IRA with above-median governance. Moreover, there appears to be no clear relationship between corruption and inef- 12 Firm and year �xed effects are not included in this graphical analysis. 17 Figure 1: Inef�ciency and Corruption by Ownership and IRA Governance No / Bad IRA, Public No / Bad IRA, Private 2 1 0 Excess Labour −1 Good IRA, Public Good IRA, Private 2 1 0 −1 −2 −1 0 1 −2 −1 0 1 Corruption Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Note: Excess labor is the residual when labor is regressed on the translog function of �rm outputs. An IRA is categorized as “bad� or “good� depending on whether its ERGI score is above or below median. The points plotted are averages across �rms for a given country-year after observations have been divided according to their ownership and regulation. �ciency when both of these reforms are undertaken, as shown in the lower right panel. I now investigate these results more formally using an econometric analysis. Econometric Analysis The results of the econometric analysis outlined in Section II are presented in Table III. The coef�cients on the translog function and year dummies are not reported to save space. It is worth noting that coef�cients on the terms in the translog function are reasonable, suggesting 18 that �rms have increasing returns to scale.13 The coef�cients on the translog function are very similar if the sample is split into private and public �rms, supporting the assumption that the translog function is relatively unaffected by ownership. Table 2: Baseline Regression (1) (2) (3) (4) (5) (6) (7) Corruption 0.21*** 0.21*** 0.21*** 0.34*** 0.17** 0.20*** 0.35*** (0.033) (0.028) (0.028) (.017) (0.072) (0.026) (0.039) Corruption × private - - - -0.26*** - -0.054** -0.10** 0.092*** 0.084*** 0.084*** 0.076*** (0.025) (0.024) (0.023) (0.057) (0.022) (0.021) (0.049) Corruption × IRA -0.14*** -0.14*** -0.14*** -0.099 -0.15*** -0.25*** (0.032) (0.027) (0.027) (0.074) (0.023) (0.037) Private dummy -0.29*** -0.26*** -0.26*** -0.27*** -0.26*** -0.25*** -0.43*** (0.036) (0.035) (0.035) (0.036) (0.041) (0.026) (0.042) IRA dummy -0.021 -0.022 (0.037) (0.23) Bad IRA dummy 0.12*** 0.12*** 0.12*** 0.14*** -0.072* (0.037) (0.036) (0.041) (0.038) (0.041) Good IRA dummy -0.11*** -0.11*** -0.13*** -0.19*** -0.14*** -0.27*** (0.038) (0.038) (0.040) (0.037) (0.031) (0.035) Corruption × bad IRA -0.14*** (0.037) Corruption × good IRA -0.14*** (0.027) Firm dummies Yes Yes Yes Yes Yes Yes Country dummies Yes Corruption * �rm dum- Yes mies IRA * country dummies Yes Private * country dum- Yes mies Observations 1359 1359 1359 1359 1359 1359 1359 Number of �rms 153 153 153 153 153 153 153 Adjusted R2 0.35 0.37 0.37 0.45 0.39 0.41 Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Note: The dependent variable is ln(labor employed). In all cases, we are estimating a translog labor require- ment function with year dummies and �rm �xed effects. To save space, the technological parameters of the translog function are not shown. Country-year clustered standard errors in parentheses. Coef�cients shown in italics are the mean effects across �rms/countries, with standard errors calculated accordingly. *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 13 The coef�cients suggest that if both output measures were to double, then the increase in labor required would be 59%. This is very close to the value obtained using the data from Dal B´o and Rossi (2007), which suggest that a doubling of outputs requires a 62% increase in employees. 19 Column (1) of Table III explores the association between corruption and productivity. We can see that the coef�cient on the corruption term is positive and strongly signi�cant, which suggests that higher corruption levels accompany a greater number of workers employed for a given function of outputs. However, we also see that corruption interacts signi�cantly with both the private ownership dummy and the dummy indicating the presence of an IRA. In both cases, the coef�cient is negative and of a smaller magnitude than the coef�cient of corruption. This �nding suggests that the negative relationship between corruption and productivity is sig- ni�cantly reduced if the �rm is either privately owned or regulated by an IRA. These results are consistent with Proposition 1 in the theoretical model above.14 It is also informative to consider the terms that do not involve corruption. The coef�cient on the IRA dummy in column (1) is insigni�cant. Given that corruption is scaled such that its mean sample value is zero, this �nding suggests that the creation of a regulatory agency has no effect on productivity if corruption is at the average sample level. The signi�cant coef�cient on the private dummy suggests that private �rms are more productive than public ones at average corruption levels. Column (2) introduces a measure of regulatory governance. The “Bad IRA� dummy indi- cates the presence of an IRA in the bottom 30% of regulators scored on the ERGI, whereas the “Good IRA� dummy represents the presence of an IRA that has an ERGI score in the top 70% of regulators.15 It is interesting to note that the coef�cient on the “Bad IRA� dummy is signi�cantly positive, whereas that on the “Good IRA� dummy is signi�cantly negative, sug- 14 The model also suggests that there may be an interaction among corruption, ownership, and regulatory au- tonomy. However, the introduction of this extra term is found to be insigni�cant. This result is likely due to the lack of �rms in the sample that are privately owned and not regulated by an IRA. 15 The 30/70 split was chosen because it maximizes the difference between the coef�cients on the two terms. The difference between the two coef�cients, however, remains strongly signi�cant for a range of other splits. I have also tried entering the ERGI directly, which returns similar results. I have explored breaking down the ERGI into different governance components, but no particular component is more successful in explaining �rm performance than the ERGI measure. Similarly, no governance component consistently improves upon the IRA dummy when interacting with corruption. See Wren-Lewis (2010, pp. 207-213) for more details. 20 gesting that governance is important when considering the link between productivity and IRA creation.16 In contrast, the coef�cients on the two terms interacted with corruption are almost identical, suggesting that both types of regulators are equally good at mitigating the effect of corruption. Column (3) displays the results of the regression using a simpler speci�cation in which these two coef�cients are imposed to be equal. This column forms the baseline regression for the future robustness checks. To explore the results further, columns (4)-(6) include a range of dummy variables that I interact with the variables of interest. In column (4), I allow for �rms to react differently to corruption by interacting corruption with time-invariant �rm dummies. The lack of reduction in the size or signi�cance of the Corruption × private or Corruption × IRA coef�cients shows that these results are at least partly driven by �rms that change ownership or regulation over the period. Column (5) allows for the effect of IRA creation to vary across countries. The Cor- ruption × IRA term becomes insigni�cant here, which suggests that a large part of this result is driven by differences in corruption between countries. In other words, the time variation in corruption is insuf�cient to give signi�cance to this coef�cient, although the coef�cient does not change signi�cantly. This �nding is unsurprising given that corruption is only measured at the country level and that the variance of measured corruption across countries is greater than the variance within countries. Finally, column (6) allows for the effect of privatization to vary across countries, with the coef�cient shown in italics here representing the average effect of privatization across countries. Although the coef�cient on the Corruption × private term falls, it is still statistically signi�cant at the 5% level. This �nding suggests that the privatization re- sult is only partially driven by differences in the effect of privatization across countries, with a 16 One factor in this result may be the greater commitment ability of well-governed regulators, as detailed by Levy and Spiller (1994) 21 signi�cant portion of this result stemming from a difference in the reaction of �rms to temporal changes in corruption levels. Finally, column (7) shows the results of the regression conducted with country dummies rather than �rm dummies. Here, we see that the results remain broadly similar, although the coef�cients on the reform variables are more negative. This result suggests that within countries as well as across time, �rms that are privately owned and/or better regulated employ less labor. Let us consider the size of the various effects by studying the coef�cients on the variables in column (3). Focusing on the coef�cient on corruption, the value of .21 suggests that an increase in the measured corruption of one standard deviation (.82) is associated with a 19% increase in the amount of labor employed for a given amount of outputs. However, this �nding assumes that the �rm is publicly owned and not subject to regulation by an IRA. If the �rm is private, then this association is reduced by approximately 40%. Alternatively, if the �rm is public but subject to regulation by an IRA, then the association is reduced by approximately two-thirds. o and The average effect across all �rms is therefore slightly smaller than that found by Dal B´ Rossi (2007), which is consistent with their sample containing a smaller proportion of private �rms and �rms regulated by an IRA. The importance of governance is also substantial: �rms regulated by a “Good IRA� rather than a “Bad IRA� have 25% fewer employees. Overall, three main conclusions arise from this econometric analysis. First, corruption ap- pears to be signi�cantly negatively associated with labor productivity. Second, this association is reduced if the �rm is either privately owned or if there exists an Independent Regulatory Agency. Third, �rms operating under an IRA with a higher level of regulatory governance operate more productively. 22 IV Robustness Checks In this section, we consider whether the results are robust to changes in the assumptions or methodology. The section begins by introducing extra control variables and then considers an instrumental variables approach. We also investigate the use of alternative measures of corruption, potential impacts on price and quality, and the precise timing of the impact of IRA introduction. Extra Control Variables One concern with the above results may be that the variables included are correlated with other omitted variables that affect �rm productivity. To check for this problem, other variables can be introduced into the equation, and we can observe whether the coef�cients on the original variables are affected. Because the baseline regression includes ownership and IRA dummies linearly and inter- acted with corruption, to test for omitted variable bias in these coef�cients, I include a range of control variables along with a term interacting each control variable with corruption. These control variables include a number of aspects of the regulatory environment, including the power of the incentive scheme and whether the electricity sector has been vertically disinte- grated. I then consider a number of country-level variables, such as GDP per capita, national wage levels, and labor regulation. A selection of these variables and their sources are provided in the Appendix. Key coef�cients of the regressions are reported in Columns (1) to (4) of Table 3, with the full results reported in the online appendix. Columns (1) and (2) of Table 3 report the coef�cients on the control variable and the control variable interacted with corruption. We can see that several of these additional variables and their interactions are signi�cant when in- troduced. However, from columns (3) and (4) of Table 3, we can see that the Corruption × IRA 23 and Corruption × private terms always remain signi�cant with similar coef�cients. Therefore, we can conclude that these interaction terms do not proxy for any other country-level vari- able. Moreover, the difference between the “Bad IRA� dummy and the “Good IRA� dummy always remains signi�cant. Hence, I conclude that regulatory governance does not proxy for an alternative country-level variable. To test whether corruption is proxying for an alternative variable, I include each control variable and its interaction with both private ownership and the IRA dummy. The results for a selection of the control variables used are reported in columns (5) to (9) of Table 3. Again, sev- eral of the variables and their interactions are signi�cant, but in general, the signi�cance of our terms of interest is not affected. On one occasion in Table 3, we can see that the coef�cient on the Corruption × private term becomes insigni�cant. This coef�cient becomes negative when the control variable of government surplus/de�cit over GDP is included in the regression along with an interaction with private ownership, which may be the result of the sample size being reduced substantially (from 1359 observations to 302), but it may also make us slightly more cautious about the interaction between private ownership and corruption. A plausible hypothe- sis is that less �scally responsible governments (i.e., ones with a higher de�cit) are associated with inef�cient public companies and that privatization has a stronger impact on productivity in these countries than in others. Because these governments are also more frequently corrupt, it appears that we cannot distinguish between whether corruption or the government’s de�cit is a better variable to explain the heterogeneous impact of privatization. However, we can say with some con�dence that there is no evidence that an alternative control variable would explain heterogeneity in the impact of IRA introduction better than corruption. Overall, therefore, I can conclude that the results are unlikely to be driven by omitted variable bias. 24 Table 3: Additional control variables Control X interacted with corruption Control X interacted with reforms X Cor × X Cor × IRA Cor × Pri X X × IRA X × Pri Cor × IRA Cor× pri (1) (2) (3) (4) (5) (6) (7) (8) (9) Price cap -0.83*** -0.0045 -0.094*** -0.078*** -0.096*** 0.055 -0.088*** -0.082*** Vert. disintegr. 0.77** -0.017 -0.15*** -0.076*** 0.13** -0.076 -0.0060 -0.095*** -0.082*** ln(GDP per capita) 0.40* 0.097** -0.14*** -0.095*** 0.58** -0.016 -0.23* -0.13*** -0.078*** Compensation/GDP 0.0062 -0.010** -0.14*** -0.086*** -0.060*** -0.054*** -0.013 -0.14*** -0.074*** Fuel exports/GDP 0.0014 -0.00013 -0.15*** -0.076*** -0.0032 -0.0066*** -0.0051** -0.13*** -0.063*** Urbanisation -0.0035 -0.0031** -0.15*** -0.12*** -0.0026 -0.0086*** -0.00089 -0.085* -0.079*** Exports/GDP -0.0054*** -0.00046 -0.16*** -0.086*** -0.00083*** 0.0039 -0.0062*** -0.14*** -0.068*** Shadow Economy 0.0068 0.00066 -0.17*** -0.081*** -0.0022 0.0082 -0.0071 -0.20*** -0.083*** Party’s yr in of�ce -0.0044*** 0.0020 -0.15*** -0.100*** -0.027*** 0.022** -0.0083** -0.18*** -0.098*** Political color -0.035*** 0.015** -0.11*** -0.086*** 0.027 0.048 -0.0051* -0.14*** -0.078*** Seperation of powers -0.095*** 0.068** -0.15*** -0.10*** -0.012 -0.18*** 0.13** -0.13*** -0.084*** Leg. election -0.0030 0.057*** -0.15*** -0.087*** -0.00097 -0.023 0.025 -0.14*** -0.085*** WB projects -0.0031 0.014*** -0.14*** -0.12*** 0.0012 -0.0047 0.0011 -0.14*** -0.10*** IMF agreement -0.018 0.031 -0.12*** -0.084*** -0.084* 0.044 0.058 -0.13*** -0.085*** Leg effectiveness 0.0096 -0.13*** -0.099*** 0.085 0.0070* -0.11*** -0.10*** General strikes -0.0049 0.046** -0.080** -0.11** -0.0044 -0.039 0.046 -0.13*** -0.10*** 25 Workers rights -0.013 -0.019 -0.014*** -0.11*** -0.12** 0.088* 0.13*** -0.15*** -0.095*** Government defecit 0.028*** 0.011* -0.20*** -0.15*** 0.082* -0.068 0.038** -0.17*** -0.072 Accountability -0.0053 0.021 -0.018*** -0.075** 0.25 -0.23 -0.065 -0.19*** -0.065** Pol. stability -0.046 -0.0034 -0.019*** -0.071** -0.090 0.0072 0.066 -0.19*** -0.074** Reg. quality -0.057 0.0055 -0.018*** -0.071** -0.28* 0.12 0.19*** -0.17*** -0.056** Rule of law -0.011 -0.0069 -0.018*** -0.065** 0.28 -0.39** 0.17 -0.17*** -0.063** Judic. Ind. 0.00018 -0.0039 -0.017*** -0.055** 0.026 -0.024 -0.0036 -0.18*** -0.057*** Poperty rights -0.023* 0.011* -0.017*** -0.066*** -0.046 0.018 0.0057 -0.16*** -0.055** Credit reg. -0.029 0.073 -0.017*** -0.058** -0.068*** 0.066* -0.013 -0.16*** -0.060** Labor reg. 0.0090 0.035 -0.017*** -0.055** 0.070 -0.035 -0.053 -0.17*** -0.064*** Business reg. -0.019 0.00021 -0.017*** -0.053** -0.065** 0.015 0.052*3 -0.16*** -0.041** Bank deposits -.46** -0.11 -0.013*** -0.066** -1.23*** 0.55** 0.19 -0.16*** -0.085*** Emplmnt. elasticity -.044*** -0.010 -0.014*** -0.080*** -0.080 0.034 0.011 -0.13*** -0.078*** Unemployment -0.0082 0.0050 -0.013*** -0.093*** -0.018** 0.015** -0.0046 -0.15*** -0.090*** Aid/GDP -2.39 -1.41** -0.014*** -0.079*** -6.20*** 4.15*** -1.57 -0.15*** -0.083*** Inflation -0.00023 0.0018 -0.013*** -0.078*** 0.0027** -0.0057*** -0.0021 -0.17*** -0.094*** Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Notes: Columns (1)-(4) report the results from the regression where the relevant control variable is interacted with corruption, and columns (5)-(9) report the results from the regression where the relevant control variable is interacted with the reform variables. The regression used is as in column (3) of Table 2. The columns report the coef�cients on the relevant variables, with X representing the control variable. See also notes to Table 2 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level Instrumental Variables One way to control for the potential endogeneity of the key explanatory variables is to use an instrumental variable approach. Therefore, this subsection tests the robustness of the results to instrumenting for the potentially endogenous variables: ownership, regulatory governance, and corruption. Instruments are chosen on the basis of being strong correlates of the variables in question and being most likely to meet the exclusion restriction (of not having an alternative channel through which they affect labor productivity). Although the available instruments are imperfect, they may provide some reassurance that the results are not driven by endogeneity problems. Although sectoral reforms were undertaken as a response to sector-speci�c concerns in some cases, in most instances, privatization and independent regulation were part of a wave of reforms that took place across multiple regulated sectors. Henisz, Zelner, and Guillen (2005) show that the factors that push countries toward reform include the number of reforms un- dertaken by trade partners and competitors as well as pressure from international donors. In the case of Brazil, for example, Prado (2012) describes how a combination of political factors and international trends led the government to undertake similar reforms across various reg- ulated industries. Given that a large amount of the variation in the timing of reforms comes from factors that are not speci�c to the electricity sector, we can instrument for our reforms using indicators of reform in other sectors. In this way, we can remove any bias in our esti- mates that result from sector-speci�c endogeneity, such as electricity reform prompted by �rm performance or omitted variables at the sector level. To instrument for regulatory governance, I use measures of regulatory governance in two other sectors: telecoms and water.17 Firm performance in the electricity sector is unlikely to 17 errez (2003b). For water, For the telecoms sector, I use an index of regulatory governance constructed by Guti´ I use a simple dummy that indicates whether an IRA exists, which I take from Estache and Goicoechea (2005) 26 influence regulatory governance in other sectors, and regulation of these sectors is unlikely to affect the electricity distribution sector through any means other than the effect through regu- lation. It is dif�cult to �nd a suitable instrument for �rm ownership because this is a �rm-level variable and because other available �rm-level variables are likely to affect productivity directly or to be affected by productivity. Therefore, I mustuse an instrument that is measured at the province or country level. I use dummy variables indicating whether private participation exists in various other infrastructure sectors in the province.18 This approach provides an indication of a province’s tendency to privatize network infrastructure, which is generally unaffected by the performance of the electricity distribution sector. Because these private projects occur out- side the electricity sector, they are unlikely to affect electricity distribution productivity through other channels. o and Rossi (2007) �nd no evidence In their study of corruption on labor productivity, Dal B´ that corruption should be treated as an endogenous variable, arguing that reverse causality is unlikely to be a problem due to the relatively small size of the electricity sector in the over- all economy. Nonetheless, I also instrument for corruption because we cannot rule out any potential endogeneity. The instrument used is a measure of freedom of the press, with the as- sumption that countries with greater press freedom have lower corruption.19 For the exclusion restriction to hold, we must also assume that press freedom does not affect electricity perfor- mance other than through corruption. Such an assumption may be reasonable if we believe that the press may report on poor electricity performance that results from corruption but that the press is unlikely to detail poor performance resulting from other factors that are typically more 18 This is constructed from the World Bank’s PPI Project Database. The four sectors are water, gas, railways, and sea ports. 19 Evidence for this assumption is provided, for example, by Chowdhury (2004). Dal B´ o and Rossi (2007) use imports over GDP as an instrument for corruption. This is not a strong instrument for the full sample that I use, but when I restrict my sample to the period that they analyze (before 2002), I �nd that the instrument provides signi�cant results similar to those reported in Table IV. 27 technically complex. Estimation is undertaken using the two-stage least squares estimator, with each of the in- struments and their appropriate interaction terms included in the �rst-stage regression. The F-statistics from the �rst-stage regressions suggest that the instruments are fairly strong, except for private ownership.20 Whereas the F-stat for the �rst-stage regressions of the other terms range from 12.14 to 27.13, the �rst-stage regression for the ownership dummy has an F-stat of 9.31, and the �rst-stage regression for the ownership dummy multiplied by corruption has an F-stat of 1.61. This is perhaps unsurprising given the dif�culties of instrumenting for inter- action terms between two variables and the fact that ownership varies at the �rm level, unlike any of the instruments. Nonetheless, it may suggest that we should be more cautious when interpreting the coef�cients related to private ownership. The results are presented in column (1) of Table IV. From this, we can see that the coef- �cients keep the same sign as in the baseline equation, and the terms previously found to be signi�cantly different from zero generally remain so. Two exceptions are the coef�cient on the “bad regulator� dummy and the coef�cient on private ownership, which are now both in- signi�cantly different from zero. However, we still reject the hypothesis that the coef�cients on the two regulator dummy variables are the same. The coef�cients of the corruption terms are slightly larger in magnitude than in the baseline regression, but they are less precisely estimated and therefore not signi�cantly different. The lower rows of Table IV present the results of tests of the validity of the assumptions used. These results include the p-value of the Sargan-Hansen test of instrument validity, which suggests that, on the assumption that at least one of the instruments is valid, there are no grounds to reject the assumption that the other instruments are valid. To test whether corrup- tion, regulatory governance, and ownership can be treated as exogenous, the difference of two 20 The full results from the �rst-stage regressions can be found in the online appendix 28 Table 4: Robustness checks Corruption measure: IV estimation 0/1 WBES Brazil (1) (2) (3) (4) Corruption 0.43** 0.20*** 0.085*** (0.18) (0.059) (0.037) Corruption × private -0.32* -0.13*** -0.021 0.067 (0.17) (0.032) (0.045) (0.36) Corruption × IRA -0.24* -0.10* -0.16*** -0.85** (0.13) (0.059) (0.033) (0.28) Private dummy -0.19 -0.23*** -0.29*** -0.31 (0.040) (0.038) (0.048) (0.28) Bad IRA dummy 0.015 0.17*** 0.22*** 0.78** (0.16) (0.035) (0.053) (0.27) Good IRA dummy -0.25* -0.070** 0.066* 0.64** (0.14) (0.035) (0.033) (0.23) Observations 1359 1359 1229 343 Number of �rms 153 153 141 35 R2 0.17 0.38 0.39 0.65 Endogeneity test p-value 0.21 Hansen J exog. test p-value 0.68 Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Notes: See notes to Table 2. *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level Sargan-Hansen statistics is calculated, one for the equation in which the possibly endogenous variables are instrumented and one for a speci�cation in which these possibly endogenous vari- ables are added to the instrument set.21 The p-value resulting from the associated test is 0.21; hence, we cannot reject the null hypothesis that these variables can be treated as exogenous. Alternatively, one can test for the endogeneity of the variables by running a Hausman test com- paring the baseline regression with the IV regression, and the obtained results cannot reject the null hypothesis of non-systematic differences in the coef�cients. I therefore conclude that it is reasonable to treat all of these variables as exogenous, as in the main analysis. 21 For more details of this test, see Baum, Schaffer, and Stillman (2003). 29 Alternative Corruption Measures As stated earlier, I have used the corruption index produced by ICRG because it was designed to be comparable over time within countries as well as between countries. However, there may be concerns that the results are driven by the peculiarities of this index. Therefore, I perform the baseline regression using three alternative measures of corruption, the results of which are presented in Table IV. First, the ICRG index is replaced with an indicator variable that takes a value of one if the ICRG corruption index is above the sample median and a value of zero if it is below. This indicator is largely insensitive to extreme values or small annual changes in the index. From column (2), we can see that the results do not change signi�cantly. Hence, we can be con�dent that the results are not driven by extreme values in the corruption index. Second, I use a measure of how signi�cant corruption is as an obstacle to doing business for �rms based on World Bank Enterprise Surveys. The measure differs from the ICRG index in that it is sourced from �rms rather than experts and that it provides an indication of corruption costs rather than corruption frequency.22 Moreover, because I have access to the �rm-level data, I can construct a measure of corruption for some states and provinces within Brazil and Argentina. Therefore, I do not use country-level indicators for these two countries. However, there are only one or two waves of this survey per country; therefore, I have to �ll in missing values.23 The measure is also scaled to produce comparable coef�cients to the measure of corruption used in the main analysis. The regression using this data is presented in column (3). We can see that all coef�cients remain similar to those obtained when the ICRG index is used. Although the Corruption × private term loses signi�cance, we cannot reject the hypothesis that 22 See Dethier, Hirn, and Straub (2008) for a survey of the literature using �rm survey data on corruption. 23 For locations for which I have at least one observation, I use linear interpolation to �ll in missing values between observations and take the value of the observation closest in time otherwise. 30 the ratio β2 /β1 is as previously found, and these coef�cients are as de�ned in Equation (4). Third, I use a measure of observed corruption constructed from data on the federal audit- ing of Brazilian municipalities used by Ferraz and Finan (2011). The corruption measure is the fraction of audited municipalities in which corruption was detected, varying by state, with municipalities weighted by their population. This variable measures the occurrence of corrup- tion rather than corruption perceptions. This distinction is important if corruption perceptions systematically differ from true corruption levels, as suggested by Olken (2009). The results are presented in column (4), from which we can see that the Corruption × IRA term is again signi�cant and negative. Because the measure of corruption is invariant over time, the �xed effects model used here cannot estimate the effect of corruption. However, using a random ef- fects model (not reported here) obtains a positive coef�cient that is signi�cant at the 5% level, with the Corruption × IRA term remaining negative and signi�cant. Overall, we can see that the previously identi�ed negative association between corruption and productivity and the mitigating effect of an IRA are robust to using these three alternative corruption measures. Quality and Prices One concern with the previous analysis is that the dependent variable, “excess labor�, may not have been “excess� but instead may have been employed to raise the quality of outputs. More- over, it would be useful to know whether the results identi�ed above extend beyond changes in labor employed to changes in consumer prices. Table IV presents the results of regressing other �rm-level variables on corruption, ownership, and regulation.24 Columns (1) and (2) use two measures of quality, the frequency and duration of interruptions in the power supply, whereas 24 Unlike the prior analysis, I do not regress the dependent variables on a function of �rm outputs. Doing so does not affect the results signi�cantly. 31 Column (3) shows the percentage of electricity lost through distribution as the dependent vari- able. In columns (4) and (5), two price measures are given: the tariffs faced (in $) by residential and industrial consumers, respectively. These variables were not included in the main analysis because they are not always observed. Summary statistics of all of these variables can be found in Table 1. Table 5: Quality and prices Dependent variable: Interruption Interruption % of elec. Residential Industrial Frequency Duration Lost Tariff Tariff (1) (2) (3) (4) (5) Corruption -5.92 -7.34 0.048 14.5 40.4* (8.00) (11.6) (1.04) (12.9) (22.8) Corruption × Private 1.68 1.37 1.10** -1.37 -7.15* (4.79) (7.87) (0.50) (3.19) (3.86) Corruption × IRA 1.31 1.31 -0.51 -13.2 -35.6 (4.46) (6.53) (0.98) (12.8) (23.2) Private dummy -11.6* -16.3* -1.27* -6.73*** -2.87 (6.51) (9.54) (0.72) (2.26) (4.06) Bad IRA dummy -15.0*** -21.1*** 6.32*** 15.0** 7.92 (4.13) (7.11) (1.04) (6.42) (14.9) Good IRA dummy -21.8* -20.6** -0.035 -18.3*** -11.8 (11.3) (9.52) (1.58) (4.23) (8.75) Observations 776 809 1211 979 571 Number of �rms 118 119 147 130 78 Adjusted R2 0.052 0.032 0.11 0.35 0.22 Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Note: In all cases we are estimating with year dummies and �rm �xed effects. Country-year clustered standard errors are in parentheses. *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level The results presented in columns (1)-(3) help alleviate any concern that previously noted variations in labor employed may reflect variations in quality.25 The coef�cients involving cor- ruption in these regressions are generally insigni�cant, suggesting that the corruption-related 25 I also enter each quality variable into the baseline regression and �nd each to be insigni�cant, with no signif- icant changes to the other coef�cients of interest. 32 results found previously are not driven by changes in quality. Moreover, the reforms of interest generally appear to be signi�cantly positively associated with quality. The coef�cients in columns (4) and (5) suggest that the previously noted changes in produc- tivity may correlate with changes in consumer prices. Though generally not highly signi�cant, each coef�cient is of the same sign as in the baseline regression. This �nding suggests that consumers may be reaping some of the gains of productivity improvements noted previously and gives some weight to the assumption in the model that productivity effects are driven by revenue control methods of regulation. The fact that these coef�cients are not as signi�cant as those in the productivity regression suggests that, in addition to consumers, other parties are gaining from the increase in labor productivity. This other party may be the government, because frequently marginal payments to utility �rms come from government subsidies. How- ever, this �nding may suggest that other actors, such as politicians or �rm owners, are gaining at the expense of workers, potentially producing a negative effect on social welfare. Dynamic Effects The econometric methodology above is based ona static model. However, we may be con- cerned that the underlying processes are much more dynamic and that failing to account for this dynamism may bias our results. For instance, if regulatory independence were influenced by corruption in the previous year (controlling for current corruption) and the effects of corrup- tion on productivity were cumulative, then the omission of lagged corruption would bias the coef�cient on the independent regulator term. To ensure that the results are not driven by such dynamic effects, I perform a number of further robustness checks. One particular concern may be that the above results relating to reforms are driven by some other trend in the way that corruption affects infrastructure productivity. To test for this 33 possibility, I perform the baseline regression including country-speci�c trends and allowing for the effect of corruption to vary over time. In both of these cases, there are no signi�cant changes to the main results. Another test is to consider how the impact of corruption on productivity changes according to when the IRA was created. The results of such an analysis can be seen in Figure 2, which plots the coef�cient on the corruption term interacted with indicator variables that depend on the number of years since an IRA was created. Figure 2: Coef�cient on Corruption, by Year since IRA Creation .4 .3 Corruption coefficient .2 .1 0 −4 −2 0 2 4 6 Years since IRA creation Source: Author’s analysis based on data from the World Bank and the International Country Risk Guide Note: The points represent the estimated coef�cient on the corruption term according by the number of years since the IRA was created, with the lines representing the 95% con�dence interval. With the exception of these terms, the regression was as it was performed in column (1) of Table III. The plot for -3 corresponds to the effect of corruption when the regulator will be set up in three or more years time, whereas the plot for 5 corresponds to the effect of corruption when a regulator was set up �ve or more years ago. 34 The way in which the corruption coef�cient changes over time is more consistent with a discontinuous shift at the point of IRA creation than a general trend over time (Figure 2). Al- though it appears that there may be some “expectation� effect in the year prior to IRA creation, it is only after the regulator is created that 0.21 (the estimated coef�cient on the corruption term in the baseline regression) is no longer in the con�dence interval.26 In addition to the above, the regressions were performed with lagged terms of each of the dependent variables included, with no signi�cant changes to the results. I also performed the regression clustering standard errors at the �rm level, which allows for the error term to be correlated within these clusters. Because the coef�cients of interest remain signi�cant, we can conclude that the results are probably not driven by the use of a static model rather than a dynamic one. Other Robustness Checks I have attempted other permutations of the baseline equation, including the following: • Dropping each year and country individually from the sample. • Replacing the variable MWh sold with MWh sold + losses to reflect varying amounts of electricity lost. • Using a Cobb-Douglas function rather than the translog used above as well as simply using Connections/Employees as a dependent variable. • Using a random effects estimator rather than a �xed effects estimator. • Including the length of the distribution network in the translog function. 26 Wren-Lewis (2010, pp.204-207) considers in more detail how the effect of IRA creation and privatization changes over time. In terms of privatization, productivity does appear to increase prior to the change in ownership, but the interaction with corruption is insigni�cant. 35 • Weighting by �rm size and splitting the sample into �rms that are small (i.e., below the median amount of electricity sold) and �rms that are large (i.e., above the median amount of electricity sold). In each of these permutations, the Corruption, Corruption × IRA, and Corruption × private terms remain signi�cant with the expected signs. V Conclusion This paper analyzes the relationship between corruption and regulated �rms’ productivity and the way in which this relationship interacts with policy reforms. The paper sets out a potential channel through which corruption increases labor employed and analyzes how privatization and regulatory autonomy may interact with this mechanism. This approach provides a framework for an empirical investigation of the effect of corruption on electricity distribution �rms in Latin America. The econometric analysis shows that corruption at the national level is negatively associ- ated with �rm productivity. This result adds to the increasing evidence that corruption can be detrimental to the performance of utilities. The association between corruption and productivity is smaller for private �rms than for public ones. This �nding suggests that privatization may be a way to reduce the potentially negative effects of corruption. However, the analysis suggests caution when making this pre- diction because the signi�cance of this result disappears when I include an interaction between ownership and the size of the government de�cit or use corruption measures from alternative sources. A more robust �nding is that the introduction of an Independent Regulatory Agency sub- stantially reduces the negative association between corruption and productivity. This result 36 survives controlling for �rm-speci�c corruption effects and introducing a large range of con- trol variables. Moreover, the result is robust to instrumenting for corruption and sector reforms and still holds when using alternative corruption measures based on �rm surveys and observed corruption. Such a result is possibly surprising given that the theoretical literature has generally focused on capture of the regulatory agent as the mechanism through which corruption impacts on regulated monopolies (Estache and Wren-Lewis 2011). However, there is reason to be cautious when interpreting the implications of these results. There is not strong evidence that electricity consumers are paying directly for the lower produc- tivity associated with corruption or that that they bene�t from the reduced effect of corruption produced by reforms. Although this may simply reflect a lack of suf�cient data on consumer prices and government subsidies, it remains possible that a sizable portion of the gains are cap- tured by �rm owners or other actors. More data are needed to be able to draw strong normative conclusions about the impacts of the reforms. Overall, the results emphasize the need to consider institutional weaknesses when devel- oping appropriate sectoral policies. The �ndings suggest that the implementation of policies considered international “best practices� has different effects depending on the level of cor- ruption in the country. Identifying precisely why this occurs and which aspects of regulatory governance are important in addressing corruption will require further research, through fur- ther study of individual countries and through the development of cross-country data on how regulatory governance varies over time. Notes 1 Other studies include Guasch and Straub (2009), who examine the effect of corruption on 37 renegotiation, and Estache, Goicoechea, and Trujillo (2009), who consider the impact of cor- ruption on country-level measures of access, affordability, and quality. Clarke and Xu (2004) take a different approach by considering the effect of reforms on petty bribery to utility �rms. 2 For surveys of the empirical literature on privatization in developing countries, see Parker and Kirkpatrick (2005); Megginson and Sutter (2006); Boubakri, Cosset, and Guedhami (2008). 3 The model therefore follows a tradition of modeling corruption in a three-tier hierarchy, as in Spiller (1990) and Laffont and Tirole (1991). Within this literature, other models also consider a situation with multiple supervisors (see, for example, Laffont and Martimort (1999)), and Estache and Wren-Lewis (2009) argue that it is important to consider the multiplicity of actors involved in the regulatory process. 4 For the purposes of the model, it is not necessary to specify the size of this share. We simply require that the share needed to persuade the politician is suf�ciently small that there is always enough money to bribe. 5 We assume that the regulator cannot overrule the politician in the other direction - that is, set revenue high when the politician is not corrupted. In a more general model where this is possible, the creation of an independent regulator would have ambiguous effects with regard to corruption. 6 For the purposes of this model, we abstract from the reasons why these reforms are imple- mented. Because corrupt politicians may lose out from the reforms, one potential explanation is that these reforms are a response to pressure from outside actors, including citizens and international bodies. See Section IV for a brief discussion of such motivations. 7 This effect is similar to the effects of privatization in Shleifer and Vishny (1994), where 38 privatization decreases the relative influence of those pushing for excess labor compared to pro�t-motivated managers. Bai et al. (2000) also discusses why state-owned �rms may be required to use their funds to over-employ. 8 An alternative model could consider the idea that a regulator’s independence might directly impact the way in which rents are shared within the �rm in a way similar to privatization. For example, more transparent regulation may reduce the rents that workers obtain from holding more information than owners or politicians. In such a model, greater independence could then reduce the impact of corruption on labor productivity without an effect on revenue. However, such a model is not investigated here because revenue setting is generally considered to be a regulator’s primary role. 9 I am therefore implicitly assuming that regulatory governance remains constant during the reign of the agency and that it is unrelated to the quality of regulation prior to the creation of the agency. This is obviously a strong assumption, but if it has any effect on my results, it is likely to bias them toward insigni�cance. Therefore, it should not be of great concern when interpreting the results. 10 o and Rossi (2007) also include In addition to electricity produced and connections, Dal B´ the service area as an exogenous output and transformer capacity and the length of the distri- bution lines as exogenous capital variables. Unfortunately, the �rst two of these variables are not available in the extended dataset used here, and including the latter reduces my sample by more than half. However, because these variables vary little over time, they are likely to be controlled for by using �rm �xed effects. I test for this by performing the regressions with the length of distribution lines included in the translog function and �nd the variable to be in- o signi�cant, with no signi�cant changes in my results. Moreover, using the dataset from Dal B´ 39 and Rossi (2007), I �nd that their results are insensitive to the removal of the service area and transformer capacity from the labor demand function. 11 Spiller and Tommasi (2005), for instance, argue that regulatory independence is more likely to occur when there is a greater division of powers and less preference homogeneity across government. 12 Firm and year �xed effects are not included in this graphical analysis. 13 The coef�cients suggest that if both output measures were to double, then the increase in labor required would be 59%. This is very close to the value obtained using the data from o and Rossi (2007), which suggest that a doubling of outputs requires a 62% increase in Dal B´ employees. 14 The model also suggests that there may be an interaction among corruption, ownership, and regulatory autonomy. However, the introduction of this extra term is found to be insignif- icant. This result is likely due to the lack of �rms in the sample that are privately owned and not regulated by an IRA. 15 The 30/70 split was chosen because it maximizes the difference between the coef�cients on the two terms. The difference between the two coef�cients, however, remains strongly signi�cant for a range of other splits. I have also tried entering the ERGI directly, which returns similar results. I have explored breaking down the ERGI into different governance components, but no particular component is more successful in explaining �rm performance than the ERGI measure. Similarly, no governance component consistently improves upon the IRA dummy when interacting with corruption. See Wren-Lewis (2010, pp. 207-213) for more details. 40 16 One factor in this result may be the greater commitment ability of well-governed regula- tors, as detailed by Levy and Spiller (1994) 17 errez For the telecoms sector, I use an index of regulatory governance constructed by Guti´ (2003b). For water, I use a simple dummy that indicates whether an IRA exists, which I take from Estache and Goicoechea (2005) 18 This is constructed from the World Bank’s PPI Project Database. The four sectors are water, gas, railways, and sea ports. 19 o Evidence for this assumption is provided, for example, by Chowdhury (2004). Dal B´ and Rossi (2007) use imports over GDP as an instrument for corruption. This is not a strong instrument for the full sample that I use, but when I restrict my sample to the period that they analyze (before 2002), I �nd that the instrument provides signi�cant results similar to those reported in Table IV. 20 The full results from the �rst-stage regressions can be found in the online appendix 21 For more details of this test, see Baum, Schaffer, and Stillman (2003). 22 See Dethier, Hirn, and Straub (2008) for a survey of the literature using �rm survey data on corruption. 23 For locations for which I have at least one observation, I use linear interpolation to �ll in missing values between observations and take the value of the observation closest in time otherwise. 24 Unlike the prior analysis, I do not regress the dependent variables on a function of �rm outputs. Doing so does not affect the results signi�cantly. 41 25 I also enter each quality variable into the baseline regression and �nd each to be insigni�- cant, with no signi�cant changes to the other coef�cients of interest. 26 Wren-Lewis (2010, pp.204-207) considers in more detail how the effect of IRA creation and privatization changes over time. In terms of privatization, productivity does appear to increase prior to the change in ownership, but the interaction with corruption is insigni�cant. 42 Appendix: Selection of additional country control variables Control variable Description Source GDP per capita Constant 2000 US$ World Bank (2009) Workers’ compensation Employee compensation / GDP World Bank (2009) Fuel Exports % of merchandise exports World Bank (2009) Urbanization Urban population / total World Bank (2009) Trade Imports & exports / GDP World Bank (2009) Shadow Economy Share of total GDP Schneider (2007) Length of of�ce Yrs ruling party in power Beck et al. (2001) Executive orientation Left-wing/central/right-wing Beck et al. (2001) Separation of powers Does the party of the executive control Beck et al. (2001) legislature? Elections Dummy for election year Beck et al. (2001) World Bank presence Number of WB projects Boockmann and Dreher (2003) IMF presence IMF agreement dummy Dreher (2006b) Legislative effectiveness Index Norris (2009) General strikes Number of strikes Norris (2009) Workers rights Index Teorell, Holmberg, and Rothstein (2009); Cingranelli and Richards (2009) Government de�cit % of GDP Teorell, Holmberg, and Rothstein (2009); Easterly (2001) Accountability Index Kaufmann, Kraay, and Mastruzzi (2009); ICRG Political stability Index Kaufmann, Kraay, and Mastruzzi (2009); ICRG Regulatory quality Index Kaufmann, Kraay, and Mastruzzi (2009); ICRG Rule of law Index Kaufmann, Kraay, and Mastruzzi (2009); ICRG Judicial independence Index Gwartney and Lawson (2009) Property rights Index Gwartney and Lawson (2009) Credit market regulation Index Gwartney and Lawson (2009) Labor market regulation Index Gwartney and Lawson (2009) Business regulation Index Gwartney and Lawson (2009) Financial development Various measures Beck, Demirguc-Kunt, and Levine (2000) Employment elasticity ∆ Employment /∆ GDP ILO (2009) Unemployment % of population ILO (2009) Aid Total aid/GDP Roodman (2005) Education Various measures Barro and Lee (2001) inflation ECLAC (2009) Legal origin Porta, de Silanes, and Shleifer (2008) Economic freedom Various indices Holmes, Feulner, and O’Grady (2008) Political rights Index Freedom House Civil liberties Index Freedom House Freedom of the press Index Freedom House Globalization Various indices Dreher (2006a) Democracy Various indices Marshall and Jaggers (2007) Government spending Government share of real GDP Heston, Summers, and Aten (2009) 43 References Andres, Luis, Sebastian Lopez Azumendi, and J. Luis Guasch. 2008. “Regulatory Governance and Sector Performance: Methodology and Evaluation for Electricity Distribution in Latin America.� World Bank Policy Research Working Paper 4494, World Bank, Washington DC. Andres, Luis, J. Luis Guasch, Makhtar Diop, and Sebastin Lopez Azumendi. 2007. “Assessing the Gov- ernance of Electricity Regulatory Agencies in the Latin American and Caribbean Region: A Bench- marking Analysis.� World Bank Policy Research Working Paper 4380, World Bank, Washington DC. Bai, Chong-En, David D. Li, Zhigang Tao, and Yijiang Wang. 2000. “A Multitask Theory of State Enterprise Reform.� Journal of Comparative Economics 28 (4):716–738. Bardhan, Pranab. 1997. “Corruption and Development: A Review of Issues.� Journal of Economic Literature 35 (3):1320–1346. URL http://links.jstor.org/sici?sici=0022-0515% 28199709%2935%3A3%3C1320%3ACADARO%3E2.0.CO%3B2-6. Barro, Robert J. and JongWha Lee. 2001. “International data on educational attainment: updates and implications.� Oxford Economic Papers 53 (3):541–563. Baum, Christopher F, Mark E. Schaffer, and Steven Stillman. 2003. “Instrumental variables and GMM: Estimation and testing.� Stata Journal 3 (1):1–31. URL http://ideas.repec.org/a/tsj/ stataj/v3y2003i1p1-31.html. Beck, Thorsten, George Clarke, Alberto Groff, Philip Keefer, and Patrick Walsh. 2001. “New Tools in Comparative Political Economy: The Database of Political Institutions.� The World Bank Economic Review 15 (1):165–176. Beck, Thorsten, Asli Demirguc-Kunt, and Ross Levine. 2000. “A New Database on the Structure and Development of the Financial Sector.� The World Bank Economic Review 14 (3):597–605. 44 Bergara, Mario E., Witold J. Henisz, and Pablo T. Spiller. 1998. “Political Institutions and Electric Utility Investment: A Cross-National Analysis.� California Management Review 40:3–35. Boehm, F. 2009. “Anti-Corruption in Regulation - A Safeguard for Infrastructure Reforms.� Competition and Regulation in Network Industries 10 (1):45–76. Boockmann, Bernhard and Axel Dreher. 2003. “The contribution of the IMF and the World Bank to economic freedom.� European Journal of Political Economy 19 (3):633–649. Boubakri, Narjess, Jean-Claude Cosset, and Omrane Guedhami. 2008. “Privatisation in Developing Countries: Performance and Ownership Effects.� Development Policy Review 26 (3):275–308. Chowdhury, Shyamal K. 2004. “The effect of democracy and press freedom on corruption: an empirical test.� Economics Letters 85 (1):93–101. Cingranelli, David L. and David L. Richards. 2009. “Human Rights Dataset.� http://www.humanrightsdata.org. Clarke, George R. G. and Lixin Colin Xu. 2004. “Privatization, Competition, and Corruption: How Characteristics of Bribe Takers and Payers Affect Bribes to Utilities.� Journal of Public Economics 88 (9-10):2067–2097. Cubbin, John and Jon Stern. 2006. “The Impact of Regulatory Governance and Privatization on Elec- tricity Industry Generation Capacity in Developing Economies.� The World Bank Economic Review 20 (1):115–141. o, Ernesto. 2006. “Regulatory Capture: A Review.� Oxford Review of Economic Policy 22 (2):203– Dal B´ 225. o, Ernesto and Mart´ Dal B´ ın A. Rossi. 2007. “Corruption and Inef�ciency: Theory and Evidence from Electric Utilities.� Journal of Public Economics 91 (5-6):939–962. 45 Dethier, Jean-Jacques, Maximilian Hirn, and Stphane Straub. 2008. “Explaining Enterprise Performance in Developing Countries with Business Climate Survey Data.� World bank policy research working paper. Dreher, Axel. 2006a. “Does globalization affect growth? Evidence from a new index of globaliza- tion.� Applied Economics 38 (10):1091. URL http://www.informaworld.com/10.1080/ 00036840500392078. ———. 2006b. “IMF and economic growth: The effects of programs, loans, and compliance with conditionality.� World Development 34 (5):769–788. Easterly, William. 2001. “Growth Implosions and Debt Explosions: Do Growth Slowdowns Cause Public Debt Crises?� Contributions to Macroeconomics 1 (1). ECLAC. 2009. “CEPALSTAT.� http://www.eclac.org/estadisticas/default.asp?idioma=IN. Estache, Antonio and Ana Goicoechea. 2005. “A ’Research’ Database on Infrastructure Economic Performance.� Policy Research Working Paper 3643, World Bank. Estache, Antonio, Ana Goicoechea, and Lourdes Trujillo. 2009. “Utilities reforms and corruption in developing countries.� Utilities Policy 17 (2):191–202. Estache, Antonio and Lourdes Trujillo. 2009. “Corruption and infrastructure services: An overview.� Utilities Policy 17 (2):153–155. Estache, Antonio and Liam Wren-Lewis. 2009. “Toward a Theory of Regulation for Developing Coun- tries: Following Jean-Jacques Laffont’s Lead.� Journal of Economic Literature 47:729–770. ———. 2011. “Anti-corruption policy in theories of sector regulation.� In The International Handbook on the Economics of Corruption II, edited by Susan Rose-Ackerman and Tina Soreide. Ferraz, Claudio and Frederico Finan. 2011. “Electoral Accountability and Corruption: Evidence from the Audits of Local Governments.� American Economic Review 101 (4):1274–1311. 46 Guasch, J. Luis and Stphane Straub. 2009. “Corruption and concession renegotiations.: Evidence from the water and transport sectors in Latin America.� Utilities Policy 17 (2):185–190. errez, Luis H. 2003a. “The Effect of Endogenous Regulation on Telecommunications Expansion Guti´ and Ef�ciency in Latin America.� Journal of Regulatory Economics 23:257–286(30). ———. 2003b. “Regulatory Governance in the Latin American Telecommunications Sector.� Utilities Policy 11 (4):225–240. Gwartney, James and Robert Lawson. 2009. “2009 Economic Freedom Dataset, published in Economic Freedom of the World: 2009 Annual Report.� Tech. rep., Economic Freedom Network. Henisz, Witold J., Bennet A. Zelner, and Mauro F. Guillen. 2005. “The Worldwide Dif- fusion of Market-Oriented Infrastructure Reform, 19771999.� American Sociological Re- view 70:871–897(27). URL http://www.ingentaconnect.com/content/asoca/asr/ 2005/00000070/00000006/art00001. Heston, Alan, Robert Summers, and Bettina Aten. 2009. “Penn World Table Version 6.3.� http://pwt.econ.upenn.edu/. Holmes, Kim R., Edwin J. Feulner, and Mary Anastasia O’Grady. 2008. “Economic Freedom Indices.� http://www.heritage.org/index/. ILO. 2009. “Key Indicators of the Labour Market (KILM), Sixth Edition.� www.ilo.org/KILM. Joskow, Paul L. 2005. “Regulation and Deregulation after 25 Years: Lessons Learned for Research in Industrial Organization.� Review of Industrial Organization 26 (2):169–193. Kaufmann, Daniel, Aart Kraay, and Massimo Mastruzzi. 2009. “Governance Matters VIII: Aggregate and Individual Governance Indicators, 1996-2008.� Policy Research Working Paper 4978, World Bank. URL http://ssrn.com/paper=1424591. 47 Kumbhakar, Subal C. and Lennart Hjalmarsson. 1998. “Relative performance of public and private ownership under yardstick competition: electricity retail distribution.� European Economic Review 42 (1):97–122. Laffont, Jean-Jacques and David Martimort. 1999. “Separation of Regulators against Col- lusive Behavior.� The Rand Journal of Economics 30 (2):232–262. URL http: //links.jstor.org/sici?sici=0741-6261%28199922%2930%3A2%3C232% 3ASORACB%3E2.0.CO%3B2-Y. Laffont, Jean-Jacques and Jean Tirole. 1991. “The Politics of Government Decision-Making: A Theory of Regulatory Capture.� The Quarterly Journal of Economics 106 (4):1089–1127. Lambsdorff, Johann Graf. 2005. “Consequences and causes of corruption - What do we know from a cross-section of countries?� In International Handbook on the Economics of Corruption, edited by Susan Rose-Ackerman. Cheltenham: Edward Elgar, 3–51. Levy, Brian and Pablo T. Spiller. 1994. “The Institutional Foundations of Regulatory Commitment: A Comparative Analysis of Telecommunications Regulation.� Journal of Law, Economics and Organi- zation 10 (2):201–246. Marshall, Monty G. and Keith Jaggers. 2007. “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2006.� http://www.systemicpeace.org/polity/polity4.htm. ephane Straub. 2009. “Infrastructure privatization and changes in corruption Martimort, David and St´ patterns: The roots of public discontent.� Journal of Development Economics 90 (1):69–84. Megginson, William L. and Natalie L. Sutter. 2006. “Privatisation in Developing Countries.� Corporate Governance: An International Review 14 (4):234–265. Montoya, Miguel . and Francesc Trillas. 2007. “The Measurement of the Independence of Telecommu- nications Regulatory Agencies in Latin America and the Caribbean.� Utilities Policy 15 (3):182–190. 48 Neuberg, Leland Gerson. 1977. “Two Issues in the Municipal Ownership of Electric Power Distribution Systems.� The Bell Journal of Economics 8 (1):303–323. Norris, Pippa. 2009. “Democracy Time series Dataset.� http://www.hks.harvard.edu/fs/pnorris/Data/Data.htm. Olken, Benjamin A. 2009. “Corruption perceptions vs. corruption reality.� Journal of Public Economics 93 (7-8):950–964. Parker, David and Colin Kirkpatrick. 2005. “Privatisation in Developing Countries: A Review of the Evidence and the Policy Lessons.� Journal of Development Studies 41:513–541. Porta, Rafael La, Florencio Lopez de Silanes, and Andrei Shleifer. 2008. “The Economic Consequences of Legal Origins.� Journal of Economic Literature 46:285–332. URL http://www.ingentaconnect.com/content/aea/jel/2008/00000046/ 00000002/art00001;http://dx.doi.org/10.1257/jel.46.2.285. Prado, Mariana Mota. 2012. “Implementing independent regulatory agencies in Brazil: The contrasting experiences in the electricity and telecommunications sectors.� Regulation & Governance 6 (3):300– 326. URL http://dx.doi.org/10.1111/j.1748-5991.2012.01142.x. Roodman, David. 2005. “An Index of Donor Performance.� Working Paper 67, Center for Global Development. Schneider, Friedrich. 2007. “The Shadow Economies in Middle and South America and their Influence on the Of�cial Economy: What do we know?� http://www.econ.jku.at/members/Schneider/�les/publications/ShadEconomySouthAmerica.pdf. Shleifer, Andrei and Robert W. Vishny. 1994. “Politicians and Firms.� The Quarterly Journal of Eco- nomics 109 (4):995–1025. 49 Spiller, Pablo and Mariano Tommasi. 2005. “The Institutions of Regulation: An Application to Public Utilities.� In Handbook of New Institutional Economics, edited by Claude Menard and Mary Shirley. Springer US, 515–543. Spiller, Pablo T. 1990. “Politicians, Interest Groups, and Regulators: A Multiple-Principals Agency Theory of Regulation, or �Let Them Be Bribed�.� Journal of Law and Economics 33 (1):65–101. Teorell, Jan, Sren Holmberg, and Bo Rothstein. 2009. “The Quality of Government Dataset.� http://www.qog.pol.gu.se. World Bank. 2009. “World Development Indicators.� http://data.worldbank.org/. Wren-Lewis, Liam. 2010. Regulation of Utilities in Developing Countries. Ph.D. thesis, University of Oxford. Zhang, Yin-Fang, David Parker, and Colin Kirkpatrick. 2008. “Electricity Sector Reform in Developing Countries: An Econometric Assessment of the Effects of Privatization, Competition and Regulation.� Journal of Regulatory Economics 33 (2):159–178. 50 Online Appendix for: ‘Do infrastructure reforms reduce the effect of corruption? Theory and evidence from Latin America and the Caribbean’ Abstract This online appendix provides supplementary material to complement the main essay. In particular, Section 1 provides tables for the regressions undertaken as described in section 5.1. Section 2 then presents the �rst-stage regressions for the IV estimation carried out in section 5.2. 1. Additional control variables This section provides regression tables for the baseline regression carried out with additional control variables, as described in section 5.2 of the paper. In Tables (1) to (5), these additional control variables are interacted with corruption. In tables (6) to (14), the variables are interacted with private ownership and independent regulator indicator variables. In each case, the regression is the same as the basline - i.e. the log of employment is the dependent variable, with �rm �xed effects and a translog function included on the right hand side. Table 1: Additional control variables, interacted with corruption Corruption -0.65* 0.24*** 0.21*** 0.0034 0.24*** 0.20*** 0.22*** (0.35) (0.027) (0.032) (0.095) (0.040) (0.050) (0.029) Corruption × Private -0.095*** -0.086*** -0.076*** -0.12*** -0.086*** -0.081*** -0.100*** (0.025) (0.022) (0.024) (0.030) (0.024) (0.025) (0.024) Corruption × IRA -0.14*** -0.14*** -0.15*** -0.15*** -0.16*** -0.17*** -0.15*** (0.030) (0.027) (0.028) (0.027) (0.033) (0.036) (0.026) Private dummy -0.26*** -0.28*** -0.24*** -0.26*** -0.27*** -0.33*** -0.24*** (0.034) (0.036) (0.036) (0.036) (0.035) (0.066) (0.033) Bad IRA dummy 0.11*** 0.091*** 0.11*** 0.12*** 0.10*** 0.14** 0.14*** (0.036) (0.031) (0.037) (0.036) (0.038) (0.067) (0.037) Good IRA dummy -0.10** -0.13*** -0.12*** -0.11*** -0.100** -0.12** -0.094** (0.041) (0.041) (0.038) (0.039) (0.040) (0.049) (0.037) ln(GDP per capita) 0.40* (0.24) GDP per cap × cor 0.097** (0.039) Compensation/GDP 0.0062 (0.0087) Compenstation × cor -0.010** (0.0040) Fuel exports / exports 0.0014 (0.0017) Fuel exports × cor 0.00013 (0.00075) Urbanisation -0.0035 (0.013) Urbanisation × cor 0.0031** (0.0013) Exports / GDP -0.0054*** (0.0020) Exports × cor -0.00046 (0.00088) Shadow economy 0.0068 (0.0082) Shadow economy × cor 0.00066 (0.00090) Party’s years in office -0.0044*** (0.0013) Years in office × cor 0.0020 (0.0016) Observations 1325 1324 1315 1325 1325 687 1292 Number of �rms 153 153 153 153 153 152 153 R2 0.3865 0.3877 0.3743 0.3837 0.3845 0.4456 0.3986 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level,* Signi�cant at the 10% level 2 Table 2: Additional control variables, interacted with corruption Corruption 0.16*** 0.22*** 0.20*** 0.16*** 0.18*** 0.18* 0.14*** (0.035) (0.029) (0.030) (0.031) (0.043) (0.097) (0.040) Corruption × Private -0.086*** -0.10*** -0.087*** -0.12*** -0.084*** -0.099*** -0.11*** (0.024) (0.027) (0.024) (0.024) (0.024) (0.029) (0.025) Corruption × IRA -0.11*** -0.15*** -0.15*** -0.14*** -0.12*** -0.13*** -0.080** (0.030) (0.027) (0.028) (0.029) (0.041) (0.029) (0.035) Private dummy -0.25*** -0.25*** -0.26*** -0.26*** -0.25*** -0.22*** -0.22*** (0.036) (0.032) (0.035) (0.034) (0.033) (0.034) (0.034) Bad IRA dummy 0.13*** 0.14*** 0.11*** 0.12*** 0.13*** 0.10** 0.11** (0.041) (0.038) (0.036) (0.041) (0.040) (0.049) (0.049) Good IRA dummy -0.11*** -0.11*** -0.11*** -0.13*** -0.11*** -0.12*** -0.11*** (0.039) (0.037) (0.039) (0.037) (0.037) (0.039) (0.038) Political colour -0.035*** (0.012) Political colour × cor 0.015** (0.0070) Seperation of powers -0.095*** (0.023) Seperation of powers × cor 0.068** (0.032) Legislative election held -0.0030 (0.014) Legislative election × cor 0.057*** (0.017) World Bank projects -0.0031 (0.0032) WB projects × cor 0.014*** (0.0037) IMF programme dummy -0.018 (0.016) IMF × cor 0.031 (0.023) Legislative effectiveness 0 (0) Leg effectivenss × cor 0.0096 (0.031) General strikes -0.0049 (0.0098) Strikes × cor 0.046** (0.018) Observations 1338 1272 1359 1110 1238 975 975 Number of �rms 153 153 153 153 153 151 151 R2 0.3783 0.4091 0.3827 0.4295 0.4048 0.4227 0.4274 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 3 Table 3: Additional control variables, interacted with corruption Corruption 0.23*** 0.36*** 0.24*** 0.25*** 0.24*** 0.24*** 0.24*** (0.038) (0.043) (0.052) (0.049) (0.045) (0.049) (0.058) Corruption × Private -0.11*** -0.15** -0.075** -0.071** -0.071** -0.065** -0.055** (0.026) (0.072) (0.031) (0.028) (0.030) (0.027) (0.022) Corruption × IRA -0.14*** -0.20*** -0.18*** -0.19*** -0.18*** -0.18*** -0.17*** (0.030) (0.034) (0.049) (0.045) (0.042) (0.046) (0.046) Private dummy -0.24*** -0.022 -0.29*** -0.28*** -0.29*** -0.29*** -0.34*** (0.034) (0.068) (0.055) (0.053) (0.053) (0.053) (0.095) Bad IRA dummy 0.12*** 0.28*** 0.11** 0.10** 0.100** 0.11** 0.015 (0.042) (0.041) (0.044) (0.040) (0.042) (0.046) (0.051) Good IRA dummy -0.12*** -0.30*** -0.091* -0.079 -0.085* -0.091* -0.16** (0.037) (0.070) (0.053) (0.052) (0.049) (0.054) (0.065) Workers rights -0.013 (0.024) Workers rights × cor -0.019 (0.022) Government budget de�cit / GDP 0.028*** (0.0083) Budget de�cit × cor 0.011* (0.0057) Voice and accountability -0.0053 (0.064) Voice × cor 0.021 (0.022) Political stability -0.046 (0.039) Political stability × cor -0.0034 (0.014) Regulatory quality -0.057 (0.050) Reg. qual. × cor 0.0055 (0.022) Rule of law -0.011 (0.075) Rule of law × cor -0.0069 (0.018) Judicial independence 0.00018 (0.010) Judicial independence × cor -0.0039 (0.0058) Observations 1110 302 959 959 959 959 889 Number of �rms 153 94 153 153 153 153 153 R2 0.4204 0.3331 0.3349 0.3357 0.3360 0.3343 0.3577 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 4 Table 4: Additional control variables, interacted with corruption Corruption 0.17*** 0.16 0.20*** 0.22*** 0.23*** 0.20*** 0.16*** (0.056) (0.100) (0.065) (0.078) (0.039) (0.030) (0.045) Corruption × Private -0.066*** -0.058** -0.055** -0.053** -0.066** -0.080*** -0.093*** (0.024) (0.024) (0.023) (0.021) (0.028) (0.024) (0.023) Corruption × IRA -0.17*** -0.17*** -0.17*** -0.17*** -0.13*** -0.14*** -0.13*** (0.046) (0.041) (0.043) (0.045) (0.028) (0.027) (0.032) Private dummy -0.35*** -0.33*** -0.36*** -0.36*** -0.25*** -0.26*** -0.28*** (0.085) (0.065) (0.088) (0.087) (0.036) (0.036) (0.036) Bad IRA dummy 0.016 0.086 0.030 0.015 0.11*** 0.11*** 0.11** (0.049) (0.056) (0.054) (0.055) (0.032) (0.035) (0.043) Good IRA dummy -0.16** -0.14** -0.14** -0.16** -0.094** -0.097** -0.10** (0.066) (0.062) (0.063) (0.066) (0.039) (0.038) (0.039) Property rights -0.023* (0.013) Property rights × cor 0.011* (0.0058) Credit market regulation -0.029 (0.020) Credit regulation × cor 0.0073 (0.0096) Labour regulation -0.0090 (0.014) Labour regulation × cor 0.0035 (0.0072) Business regulation -0.019 (0.014) Business regulation × cor 0.00021 (0.010) Bank deposits / GDP -0.46** (0.21) Bank deposits × cor -0.11 (0.11) Employment elasticity -0.044*** (0.014) Employment elasticity × cor -0.010 (0.024) Unemployment -0.0082 (0.0051) Unemployment × cor 0.0050 (0.0035) Observations 899 924 899 899 1325 1359 1184 Number of �rms 153 153 153 153 153 153 151 R2 0.3640 0.3562 0.3602 0.3609 0.3837 0.3826 0.3966 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 5 Table 5: Additional control variables, interacted with corruption Corruption 0.22*** 0.18*** 0.25*** 0.25*** 0.15 0.22*** 0.13 (0.028) (0.038) (0.041) (0.043) (0.11) (0.036) (0.12) Corruption × Private -0.079*** -0.078*** -0.088*** -0.090*** -0.10*** -0.082*** -0.088*** (0.023) (0.023) (0.024) (0.025) (0.026) (0.023) (0.025) Corruption × IRA -0.14*** -0.13*** -0.14*** -0.14*** -0.14*** -0.14*** -0.14*** (0.028) (0.027) (0.029) (0.027) (0.031) (0.027) (0.029) Private dummy -0.26*** -0.26*** -0.26*** -0.26*** -0.24*** -0.26*** -0.27*** (0.034) (0.036) (0.035) (0.036) (0.033) (0.035) (0.036) Bad IRA dummy 0.10*** 0.11*** 0.13*** 0.12*** 0.14*** 0.13*** 0.11*** (0.035) (0.037) (0.040) (0.035) (0.046) (0.037) (0.032) Good IRA dummy -0.12*** -0.11*** -0.11*** -0.10*** -0.12*** -0.10** -0.12*** (0.037) (0.038) (0.040) (0.038) (0.037) (0.040) (0.039) Aid / GDP -2.39 (1.68) Aid × cor -1.41** (0.59) Inflation -0.00023 (0.0013) Inflation × cor 0.0018 (0.0018) Political rights -0.014 (0.0095) Political rights × cor -0.013 (0.013) Civil liberties 0.023 (0.018) Civil liberties × cor -0.011 (0.013) Democracy 0.0073 (0.011) Democracy × cor 0.0062 (0.013) Press freedom -0.0027*** (0.0010) Press freedom × cor -0.00056 (0.00079) Globalisation 0.0078** (0.0039) Globalisation × cor 0.0013 (0.0021) Observations 1325 1359 1359 1359 1095 1359 1359 Number of �rms 153 153 153 153 153 153 153 R2 0.3883 0.3787 0.3791 0.3789 0.4263 0.3812 0.3814 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 6 Table 6: Additional control variables, interacted with regulation and ownership Corruption 0.19*** 0.21*** 0.19*** 0.16*** (0.035) (0.028) (0.029) (0.047) Corruption × private -0.078*** -0.074*** -0.063*** -0.079*** (0.023) (0.022) (0.023) (0.023) Corruption × IRA -0.13*** -0.14*** -0.13*** -0.085* (0.033) (0.028) (0.028) (0.046) Private dummy -0.13 -0.25*** -0.19*** -0.19 (1.34) (0.040) (0.041) (0.32) Bad IRA dummy 2.12* 0.017 -0.035 0.73*** (1.18) (0.041) (0.027) (0.23) Good IRA dummy 1.91 -0.20*** -0.14*** 0.57** (1.19) (0.043) (0.038) (0.25) ln(GDP per capita) 0.58** (0.25) ln(GDP per capita) × pri -0.016 (0.15) ln(GDP per capita) × IRA -0.23* (0.13) Compensation/GDP -0.060*** (0.016) Compensation/GDP × pri -0.013 (0.013) Compensation/GDP × IRA 0.054*** (0.016) Fuel exports / exports -0.0032 (0.0036) Fuel exports × pri -0.0051** (0.0024) Fuel exports × IRA 0.0066*** (0.0017) Urbanisation 0.0026 (0.012) Urbanisation × pri -0.00089 (0.0044) Urbanisation × IRA -0.0086*** (0.0031) Observations 1325 1324 1315 1325 Number of �rms 153 153 153 153 R2 0.3858 0.4015 0.3844 0.3889 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 7 Table 7: Additional control variables, interacted with regulation and ownership Corruption 0.20*** 0.26*** 0.26*** 0.22*** (0.042) (0.051) (0.022) (0.030) Corruption × Private -0.068*** -0.083*** -0.098*** -0.078*** (0.022) (0.025) (0.023) (0.023) Corruption × IRA -0.14*** -0.20*** -0.18*** -0.14*** (0.043) (0.045) (0.022) (0.029) Private dummy -0.14*** -0.064 -0.27*** -0.13* (0.045) (0.31) (0.035) (0.076) Bad IRA dummy 0.011 -0.17 0.065 0.27** (0.050) (0.26) (0.051) (0.11) Good IRA dummy -0.16*** -0.44* -0.19*** 0.0030 (0.035) (0.26) (0.042) (0.10) Exports / GDP -0.0083*** (0.0030) Exports / GDP × pri -0.0062*** (0.0015) Exports / GDP × IRA 0.0039 (0.0025) Shadow economy 0.0022 (0.0077) Shadow economy × pri -0.0071 (0.0067) Shadow economy × IRA 0.0082 (0.0057) Party’s years in office -0.027*** (0.0096) Party’s years in office × pri 0.0083** (0.0035) Party’s years in office × IRA 0.022** (0.0096) Political colour 0.027 (0.033) Political colour × pri -0.051* (0.027) Political colour × reg -0.048 (0.035) Observations 1325 687 1292 1338 Number of �rms 153 152 153 153 R2 0.3966 0.4472 0.4071 0.3855 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 8 Table 8: Additional control variables, interacted with regulation and ownership Corruption 0.21*** 0.21*** 0.20*** 0.19*** (0.031) (0.028) (0.038) (0.039) Corruption × Private -0.084*** -0.085*** -0.10*** -0.085*** (0.023) (0.023) (0.027) (0.024) Corruption × IRA -0.13*** -0.14*** -0.14*** -0.13*** (0.030) (0.027) (0.037) (0.039) Private dummy -0.26*** -0.27*** -0.24*** -0.26*** (0.031) (0.037) (0.044) (0.038) Bad IRA dummy 0.16*** 0.12*** 0.15*** 0.12*** (0.042) (0.043) (0.053) (0.040) Good IRA dummy -0.055 -0.11** -0.090 -0.12*** (0.037) (0.041) (0.061) (0.040) Seperation of powers -0.012 (0.057) Seperation of powers × pri 0.13** (0.054) Seperation of powers × IRA -0.18*** (0.058) Legislative election held -0.00097 (0.043) Legislative election held × pri 0.025 (0.029) Legislative election held × IRA -0.023 (0.043) World Bank projects 0.0012 (0.0045) WB projects × pri 0.0011 (0.0056) WB projects × IRA -0.0047 (0.0046) IMF agreement dummy -0.084* (0.046) IMF × pri 0.058 (0.035) IMF × IRA 0.044 (0.047) Observations 1272 1359 1110 1238 Number of �rms 153 153 153 153 R2 0.4144 0.3783 0.4205 0.4074 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 9 Table 9: Additional control variables, interacted with regulation and ownership Corruption 0.19*** 0.20*** 0.22*** 0.26*** (0.035) (0.033) (0.031) (0.049) Corruption × Private -0.10*** -0.10*** -0.095*** -0.072 (0.028) (0.027) (0.027) (0.067) Corruption × IRA -0.11*** -0.13*** -0.15*** -0.17*** (0.032) (0.031) (0.031) (0.036) Private dummy -0.23 -0.24*** -0.36*** 0.038 (0.33) (0.033) (0.038) (0.073) Bad IRA dummy -0.063 0.12** -0.029 0.36*** (0.12) (0.050) (0.081) (0.12) Good IRA dummy -0.29** -0.10*** -0.21*** -0.39*** (0.11) (0.036) (0.068) (0.078) Legislative effectiveness 0 (0) Leg. effec. × pri 0.0070 (0.16) Leg. effec. × IRA 0.085* (0.049) General strikes -0.0044 (0.016) Strikes × pri 0.046 (0.032) Strikes × IRA -0.039 (0.026) Workers rights -0.12** (0.048) Workers rights × pri 0.13*** (0.034) Workers rights × IRA 0.088* (0.052) Government budget de�cit / GDP 0.082* (0.046) Budget de�cit × pri 0.038** (0.016) Budget de�cit × IRA -0.068 (0.045) Observations 975 975 1110 302 Number of �rms 151 151 153 94 R2 0.4233 0.4268 0.4424 0.3540 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 10 Table 10: Additional control variables, interacted with regulation and ownership Corruption 0.26*** 0.26*** 0.24*** 0.23*** (0.058) (0.050) (0.035) (0.048) Corruption × Private -0.065** -0.074** -0.056** -0.063** (0.026) (0.029) (0.026) (0.028) Corruption × IRA -0.19*** -0.19*** -0.17*** -0.17*** (0.057) (0.045) (0.033) (0.046) Private dummy -0.29*** -0.25*** -0.32*** -0.22*** (0.059) (0.063) (0.054) (0.072) Bad IRA dummy 0.14*** 0.096** 0.042 -0.068 (0.049) (0.042) (0.054) (0.071) Good IRA dummy -0.039 -0.085* -0.14** -0.19*** (0.080) (0.047) (0.055) (0.058) Voice and Accountability 0.25 (0.22) Voice × pri -0.065 (0.12) Voice × IRA -0.23 (0.19) Political stability -0.090 (0.063) Political stability × pri 0.066 (0.079) Political stability × reg 0.0072 (0.057) Regulatory quality -0.28* (0.16) Reg quality × pri 0.19*** (0.070) Reg quality × IRA 0.12 (0.15) Rule of law 0.28 (0.17) Rule of law × pri 0.17 (0.12) Rule of law × IRA -0.39** (0.16) Observations 959 959 959 959 Number of �rms 153 153 153 153 R2 0.3363 0.3374 0.3460 0.3446 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 11 Table 11: Additional control variables, interacted with regulation and ownership Corruption 0.23*** 0.21*** 0.21*** 0.23*** (0.044) (0.052) (0.058) (0.046) Corruption * Private -0.057*** -0.055** -0.060** -0.064*** (0.021) (0.022) (0.023) (0.023) Corruption * IRA -0.18*** -0.16*** -0.16*** -0.17*** (0.042) (0.049) (0.054) (0.045) Private dummy -0.32*** -0.38*** -0.25 -0.080 (0.093) (0.11) (0.24) (0.26) Bad IRA dummy 0.14 -0.062 -0.34* 0.18 (0.19) (0.28) (0.20) (0.42) Good IRA dummy -0.052 -0.23 -0.56** 0.028 (0.14) (0.25) (0.23) (0.44) Judicial independence 0.026 (0.024) Judicial independence X pri -0.0036 (0.029) Judicial independence X IRA -0.024 (0.033) Property rights -0.046 (0.047) Property rights X pri 0.0057 (0.020) Property rights X IRA 0.018 (0.057) Credit market regulation -0.068*** (0.024) Credit regulation X pri -0.013 (0.031) Credit regulation X IRA 0.066* (0.036) Labour regulation 0.070 (0.088) Labour regulation X pri -0.053 (0.043) Labour regulation X IRA -0.035 (0.088) Observations 889 899 924 899 Number of �rms 153 153 153 153 R2 0.3585 0.3623 0.3677 0.3631 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 12 Table 12: Additional control variables, interacted with regulation and ownership Corruption 0.20*** 0.23*** 0.19*** 0.22*** (0.056) (0.031) (0.032) (0.034) Corruption * Private -0.041** -0.085*** -0.078*** -0.090*** (0.020) (0.023) (0.026) (0.024) Corruption * IRA -0.16*** -0.16*** -0.13*** -0.15*** (0.053) (0.030) (0.030) (0.033) Private dummy -0.63*** -0.32*** -0.26*** -0.23*** (0.14) (0.094) (0.048) (0.083) Bad IRA dummy -0.091 -0.082 0.080 -0.034 (0.21) (0.086) (0.060) (0.080) Good IRA dummy -0.25 -0.26*** -0.11** -0.23*** (0.20) (0.075) (0.042) (0.075) Business regulation -0.065** (0.026) Business regulation X pri 0.052* (0.029) Business regulation X IRA 0.015 (0.034) Bank deposits / GDP -1.23*** (0.40) Bank deposits X pri 0.19 (0.22) Bank deposits X IRA 0.55** (0.25) Employment elasticitiy -0.080 (0.052) Employment elasticitiy X pri 0.011 (0.041) Employment elasticitiy X IRA 0.034 (0.053) Unemployment -0.018** (0.0070) Unemployment X pri -0.0046 (0.0092) Unemployment X IRA 0.015** (0.0072) Observations 899 1325 1359 1184 Number of �rms 153 153 153 151 R2 0.3664 0.3874 0.3829 0.3969 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 13 Table 13: Additional control variables, interacted with regulation and ownership Corruption 0.22*** 0.25*** 0.20*** 0.18*** (0.027) (0.032) (0.029) (0.038) Corruption * Private -0.083*** -0.094*** -0.081*** -0.082*** (0.023) (0.021) (0.024) (0.024) Corruption * IRA -0.15*** -0.17*** -0.13*** -0.10*** (0.027) (0.032) (0.028) (0.039) Private dummy -0.26*** -0.24*** -0.32*** -0.32*** (0.038) (0.040) (0.061) (0.11) Bad IRA dummy 0.090*** 0.20*** 0.20*** 0.49*** (0.034) (0.049) (0.068) (0.11) Good IRA dummy -0.14*** -0.057 -0.031 0.29*** (0.033) (0.043) (0.059) (0.11) Aid / GDP -6.20*** (1.49) Aid X pri -1.57 (1.27) Aid X IRA 4.15*** (1.33) Inflation 0.0027** (0.0011) Inflation X pri -0.0021 (0.0021) Inflation X reg -0.0057*** (0.0012) Political rights 0.0020 (0.021) Political rights X pri 0.019 (0.018) Political rights X IRA -0.027 (0.019) Civil liberties 0.10*** (0.035) Civil liberties X pri 0.023 (0.031) Civil liberties X IRA -0.11*** (0.028) Observations 1325 1359 1359 1359 Number of �rms 153 153 153 153 R2 0.3897 0.3914 0.3798 0.3848 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 14 Table 14: Additional control variables, interacted with regulation and ownership Corruption 0.18*** 0.20*** 0.20*** (0.037) (0.029) (0.031) Corruption * Private -0.097*** -0.081*** -0.074*** (0.027) (0.023) (0.021) Corruption * IRA -0.12*** -0.13*** -0.13*** (0.037) (0.027) (0.029) Private dummy -0.13 -0.26*** 0.44 (0.13) (0.091) (0.29) Bad IRA dummy -0.19* 0.18* -0.047 (0.094) (0.096) (0.26) Good IRA dummy -0.42*** -0.056 -0.26 (0.097) (0.089) (0.28) Democracy -0.024 (0.014) Democracy X pri -0.014 (0.016) Democracy X IRA 0.041*** (0.012) Press freedom -0.0013 (0.0022) Press freedom X pri -0.000052 (0.0024) Press freedom X IRA -0.0014 (0.0024) Globalisation 0.011* (0.0058) Globalisation X pri -0.012** (0.0053) Globalisation X IRA 0.0026 (0.0048) Observations 1095 1359 1359 Number of �rms 153 153 153 R2 0.4296 0.3812 0.3860 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 15 2. First-stage of instrumenting This section presents the �rst-stage regression results of the instrumental variable estimation carried out in section 5.2 of the main paper. Table (15) presents the results for when corruption and interaction terms involving corruption are estimated, table (16) for private ownership and (17) for the regulator dummies. Table 15: IV First Stage LHS: Corruption Bad IRA Good IRA Private Corruption × private Corruption × IRA Press Freedom 0.73** 0.21*** 0.72*** 0.22 -0.11 0.33 (0.31) (0.080) (0.10) (0.13) (0.25) (0.31) Telecoms regulation index -1.24 0.73*** 0.95*** 0.41 -0.26 -2.38** (1.00) (0.26) (0.34) (0.43) (0.81) (1.01) Water IRA dummy -0.14 0.68*** 1.84*** 0.0039 -0.99 -3.07*** (0.94) (0.25) (0.32) (0.41) (0.77) (0.95) Rail PPI dummy 2.18*** 0.14 0.93*** 1.36*** 0.18 3.02*** (0.69) (0.18) (0.23) (0.30) (0.56) (0.70) Ports PPI dummy 1.19* -0.68*** 0.76*** -0.81*** 0.95 1.09 (0.72) (0.19) (0.24) (0.31) (0.59) (0.73) Gas PPI dummy ex 0.59 -0.40* 1.25*** -0.95** 0.74 0.093 (0.89) (0.24) (0.30) (0.39) (0.73) (0.90) Water PPI dummy 1.55** 0.085 -0.71*** 0.20 0.40 1.95*** (0.74) (0.19) (0.25) (0.32) (0.60) (0.74) Press × Rail -0.71*** -0.086* -0.24*** -0.34*** -0.060 -0.89*** (0.18) (0.047) (0.061) (0.078) (0.15) (0.18) Press × Ports -0.21 0.23*** -0.17** 0.22*** -0.26 -0.21 (0.19) (0.051) (0.066) (0.084) (0.16) (0.20) Press × Gas -0.043 0.093 -0.32*** 0.24** -0.18 0.091 (0.24) (0.063) (0.081) (0.10) (0.19) (0.24) Press × Water PPI -0.44** 0.025 0.14** -0.061 -0.13 -0.54*** (0.20) (0.053) (0.068) (0.087) (0.16) (0.20) Press × Telecoms 0.37 -0.23*** -0.24** -0.038 0.12 0.73** (0.29) (0.077) (0.099) (0.13) (0.24) (0.30) Press × Water IRA 0.0086 -0.22*** -0.53*** -0.053 0.32 0.75*** (0.27) (0.071) (0.091) (0.12) (0.22) (0.27) Observations 1,359 1,359 1,359 1,359 1,359 1,359 R-squared 0.1246 0.2544 0.1936 0.1427 0.0218 0.1299 F-stat 12.14 27.13 19.51 9.31 1.61 12.64 *** Signi�cant at the 1% level , ** Signi�cant at the 5% level, * Signi�cant at the 10% level 16