Corruption in Brazil:

Transcription

1 Corruption in Brazil: An analysis based on audits of municipalities Gabriela Moretzsohn Pereira Nunes Master of Philosophy in Economics Department of Economics University of Oslo November 2017

2 Corruption in Brazil: An analysis based on audits of municipalities II

3 Gabriela Moretzsohn Pereira Nunes 2017 Corruption in Brazil: an analysis based on audits of municipalities Gabriela Moretzsohn Pereira Nunes Press: Reprosentralen, Universitetet i Oslo III

4 Abstract In this master thesis I will study the determinants of corruption in Brazilian municipalities for the years 2006 to To perform this analysis I first build a database using data from the CGU audit of municipalities program, different IBGE databases and data from the Brazilian Treasury transparency website. Both the building of the database and the definition of my hypotheses are based on the literature on corruption and on my previous knowledge of Brazil. I used the software Stata (version 14) to build the data and run my regressions. The main model is estimated with a pooled OLS with robust standard errors estimates. My analysis is based in 4 main hypotheses which derive from the tradeoff that the choice to corrupt involve: the potential benefit from extracting rents, and the cost from being caught and facing legal consequences. The hypotheses being: 1) corruption is positively correlated with natural resources related royalty transfers; 2) corruption is negatively correlated with other constitutional municipal transfers; 3) corruption is negatively correlated with human development; and 4) the corruption level in a municipality is correlated with it being audited before. My model controls for the main economic and socio-demographic characteristics of the municipalities and for state and year fixed effects. The main results corroborate the hypotheses. They show robust evidence of a significant positive correlation between average corruption and oil and gas royalties per capita, and a significant negative correlation between average corruption and human development index for An increase of R$1, in oil and gas royalty transfers per capita is correlated with an increase of 1.58 corrupt irregularities per inspection. While the results indicate that when a municipality moves from 0.5 to 0.6 in the human development index (HDI), there is an associated reduction of 1.2 corrupt irregularities per inspection. Robustness checks are performed, giving further support to the model specification and estimation strategy choices. I believe this thesis will contribute to the literature by providing a better understanding of the CGU data and by identifying factors that significantly correlate with corruption in Brazil. IV

5 Acknowledgments This Master s thesis was supervised by Professor Yikai Wang from the Department of Economics of the University of Oslo. I would like to thank him for his guidance and insightful feedbacks throughout the research process and writing of this thesis. I would also like to thank the Department of Economics from the University of Oslo. I am grateful for the opportunity of fulfilling my Master Degree in this institution and for everything I have learned. This thesis is a result of my study efforts and an opportunity to apply the knowledge I have acquired through the Master in Economics Research Programme. I would like to thank my family and friends who contributed to the fulfilment of this Master with their support and companionship. First and foremost, I would like to thank my parents, Myriam and Luiz. Thank you for always believing in my potential, for being proud of my achievements and for always supporting me for investing in my education and career. I thank my boyfriend, Jon, who is my greatest friend and have helped me so much through the stress and worries from exams and from writing the thesis. I would like to thank Marit and Tor, who are my dear Norwegian family. I thank my colleagues from the Master s programme, who have shared this experienced with me and have been of great help through the writing of this thesis. Finally, I thank all my friends in Norway and in Brazil for always being here for me. My sincere gratitude to all of you. Gabriela Moretzsohn November 10 th, Oslo, Norway V

6 Table of contents Abstract... IV Acknowledgments... V List of tables and figures... VII Abbreviations... VIII 1 Introduction Status of knowledge Literature on corruption... 4 Theoretical background... 4 The determinants of corruption... 6 Corruption and natural resources abundance... 7 Literature reviews Corruption in Brazil Corruption effect on electoral outcomes and the role of accountability Corruption and natural resources abundance in Brazil Corruption and the efficiency of the public sector in Brazil Methodology Data Measuring corruption Other data sources Overview of the data Hypotheses Model Results Regressing the model Main results Royalty transfers and total transfers per capita Human Development Index Have been audited before Other results Population and share of urban population Unemployment Mismanagement irregularities per inspection State and year dummies Robustness checks Standard errors correction Sample outliers Model specification Discussion Oil and corruption in Brazil The Petrobrás scandal The effectiveness of the audit of municipalities in leading to the punishment of corrupt individuals Conclusion References Appendix VI

7 List of tables and figures Tables: Table 1. Irregularity s classification by the CGU Table 2: Data description Table 3. Summary of statistics Table 4: Regression outcomes Table 5: Robustness check for the standard errors estimation Table 6: Robustness check for sample outliers Table 7. Robustness check for the model specification Figures: Figure 1. Corruption incidence per lottery Figure 2. Corruption per lottery, averaged by the number of inspections Figure 3. Corruption per year Figure 4. Corruption per year, averaged by the number of inspections Figure 5. Corruption incidence per state Figure 6. Corruption per state, averaged by the number of inspections Figure 7. Corruption incidence per department Figure 8. Corruption incidence per department, averaged by the number of inspections Tables in Appendix: Table A1: Region tabulation Table A2: State tabulation Table A3: Department tabulation Table A4. Lottery year tabulation Table A5. Dummy variable Have been audited before tabulation Table A6. Lottery distribution per year and frequency of inspections Table A7. Correlation matrix including all variables in the data Table A8. Regression outcomes including state and dummy variables coefficients VII

8 Abbreviations ANP: Agência Nacional do Petróleo, gás natural e biocombustíveis (In Portuguese. Nacional agency of oil, gas and biofuels) CGU: Corregedoria-Geral da União. (In Portuguese. Comptroller General) CFH: Compensação Financeira de recursos hídricos (In Portuguese. Water resources financial compensation) CFM: Compensação Financeira de recursos minerais (In Portuguese. Mineral resources financial compensation) CIDE-combustíveis: Contribuição de Intervenção no Domínio Econômico - Combustíveis (In Portuguese. Contribution for intervention in the economic domain - fuels) CNJ: Conselho Nacional de Justiça (CNJ) FEX: Auxílio Financeiro para Fomento das Exportações (In Portuguese. Financial aid for export promotion) FEP: Fundo Especial do Petróleo (In Portuguese. Oil special fund) FJP: Fundação João Pinheiro (In Portuguese. João Pinheiro Foundation) FPM: Fundo de Participação de Municípios (In Portuguese. Municipal participation fund) GDP: Gross Domestic Product HDI: Human Development Index IBGE: Instituto Brasileiro de Geografia e Estatística (In Portuguese. Brazilian geography and statistics institute) IPEA: Instituto de Pesquisa Econômica Aplicada (In Portuguese. Institute for applied economic research) IOF-ouro: Imposto sobre Operações Financeiros ouro (In Portuguese. Tax on financial operations gold) IPI- exportação: Imposto sobre Produtos Industrializados exportação (In Portuguese. Tax on industrialized products - exports) ITA: Royalties de Itaipú recursos hídricos (In Portuguese. Itaipú royalties water resources) MPF: Ministério Público Federal (In Portuguese. Federal Prosecution Service) OLS: Ordinary Least Squares PEA: Participação Especial de Petróleo e Gás Natural (In Portuguese. Oil and gas special Participation) PT: Partido dos Trabalhadores (In Portuguese. Labor Party) UNDP: United Nations Development Programme VIII

9 1 Introduction Corruption is a worldwide phenomenon that has received more attention from the media and academic world, especially since the 1990 s. The main actors contributing to this have been the World Bank and Transparency International. However, due to its illegality and the many forms corruption assumes, there is no direct way of measuring it. Many different methods of estimation have been proposed in the literature, many of them perception indexes, as the Corruption Perception Index from the Transparency International. These indexes have been long criticized due to their bias and researchers have been working to find other methods to measure corruption (e.g. Svensson, 2005; Treisman 2007; Olken and Pendergast, 2016). In countries where monitoring efforts and enforcement of the law have been used to fight corruption in the public sector, more data has been collected and so new opportunities have arisen for research. Brazil is perceived as one of the most corrupt countries in the world, according to the Transparency international. It is often quoted in the media for its many public corruption scandals since the end of the military dictatorship in As in other developing countries, the magnitude of corruption in Brazil indicates that it is pervasive and a significant challenge to its economy. The latest uncovered corruption scandal related to the state oil company Petrobrás is being investigated now by the Ministério Público Federal (MPF) (In Portuguese. Federal Prosecution Service) and there is evidence of 10 years of corrupt activity, with a so far accounted diversion of billions of dollars. Although it has been a long period of what seems to be systematic and increasing corruption in Brazil, it is true that more is known about corruption in Brazil now because a greater investigative effort has been in course 1. In 2003, the Brazilian government created the Corregedoria-Geral da União (CGU) (In Portuguese. Comptroller General) responsible for internal control, public auditing and prevention and fighting of corruption. On September 30, 2016 the CGU was incorporated in the new Ministério da Transparêcia e Corregedoria-Geral da União. (In Portuguese. Ministry of Transparency and Comptroller General). Since 2003, 1 THE ECONOMIST. Fighting Corruption, 04/ Available from: < > [Accessed 07/ ] 1

10 besides other investigations, the CGU organizes a yearly lottery of municipalities to audit the use of federal transferred funds. The reports from these audits are available to the public 2. In addition, in 2011 the government approved the Lei de Acesso à Informação (In Portuguese. Access to Information law) 3 and in 2013 the Lei Anti-corrupção (In Portuguese. Anticorruption law) 4. Both enhanced transparency and facilitated the access to data from the public sector in Brazil. Using both the access to information law and the evidence from the audit of municipalities performed by the CGU, this thesis will study the determinants of corruption in Brazilian municipalities for the years 2006 to To perform this analysis I first built a database using data from the CGU audits of municipalities program, different databases from the Instituto Brasileiro de Geografia e Estatística (IBGE) (In Portuguese. Brazilian Institute of Geography and Statistics) and from the Brazilian Treasury transparency website. Both the building of the database and the definition of my hypothesis are based on the literature on corruption and on my knowledge of Brazil. I used the software Stata (version 14) to build the data and run my regressions. The main model is estimated with a pooled Ordinary Least Squares (OLS) regression with robust standard errors estimates. My analysis is based in 4 main hypotheses which derive from the tradeoff that the choice to corrupt involve: the potential benefit from extracting rents, and the cost from being caught and facing its legal consequences. I measure corruption in average per inspection, and expect it to be positively correlated to the amount of transfers received by the municipality due to its abundance of natural resources. My main measure for that being oil and gas royalty transfers per capita. Conversely, I expect the remaining transfers, measured by total constitutional transfers per capita, to negatively correlate with corruption per inspection, since these transfers are endogenously determined. I also expect corruption per inspection to be negatively correlated with standard of living, measured by the Human Development Index (HDI) for Finally I expect the effect of being audited before to be significant to the corruption level per 2 The CGU information was accessed on their website on different pages. Available from: < ciclo> and < [Accessed 07/ ] 3 Access to information Law. Available from: < [Accessed 07/ ] 4 Anti-corruption Law. Available from: < [Accessed in 07/ ] 2

11 inspection, although I do consider that such variable could end up being insignificant due to its different channels cancelling out the effect on corruption. The model includes the main economic and socio-demographic characteristics of the municipalities. It found robust evidence that corruption correlates positively with oil and gas royalty transfers, and negatively with the HDI for 2010 and total constitutional transfers per capita. The results for total constitutional transfers per capita were not as robust in my best fitted specification and being audited before was not significant. Population size, share of urban population and unemployment rate were positive and significantly correlated with corruption per inspection. Mismanagement per inspection also showed to be highly significant and negatively correlated with corruption per inspection. I perform further robustness checks to the correction of the standard errors, the outliers in the sample, and the model specification. The results from these robustness checks give further support to the model and the estimation strategy chosen as baseline. I believe this thesis will contribute to the literature by building a database with information from the CGU audits of municipalities program and by studying the determinants of corruption in Brazil. Although many have used the CGU data for investigating accountability, political outcomes and public spending, it has not yet been used to investigate the determinants of corruption. A better understanding of corruption in Brazil is not relevant only for our domestic policy making. Brazil is an important producer and consumer in the global market, both for its role as an exporter of minerals, fuel and food products as for its big population. Besides, to understand corruption in Latin America it is crucial to understand Brazil, which due to its history, geography and resources, differs a lot from the other Latin-American countries. The thesis is divided in 6 sections plus the Appendix. Section 1 is this introduction. Section 2 presents the status of knowledge on corruption. Section 3 discusses the methodology, which includes a brief explanation of the building of the database and a data summary, the presentation of the main hypotheses and the model derived from these hypotheses. Section 4 presents the regression results and the robustness checks. Section 5 is dedicated to the discussion, where I will address some aspects that I could not investigate further in this thesis, such as the current political crisis in Brazil due to corruption scandals. Section 6 concludes the thesis. The appendix provide tables and figures related to the building of the database. 3

12 2 Status of knowledge This section will briefly present the main papers in the theory of corruption that relate to the scope of this thesis. Although the literature is quite extensive, not many papers have been dedicated to the study of the determinants of the corruption in the world and none have had such approach for the case of Brazil. This section will be divided in two parts. The first subsection discusses theoretical models, cross country evidence and literature reviews on corruption. The second subsection is dedicated to empirical studies of the Brazilian case that used data from the CGU audits of municipalities. 2.1 Literature on corruption Theoretical background The pioneering and most important research on corruption made by economists are those who made an effort to incorporate corruption in the framework of our basic microeconomic models. They try to understand the dynamic between rent-seeking behaviour, market failures and government intervention, in order to find in which conditions corruption happens and, given the economical and institutional scenario, what its level will be. Articles such as Rose- Ackerman (1975), Shleifer and Vishny (1993), Acemoglu and Verdier (1998; 2000) and Ades and Di Tella (1999) have accessed these topics. They are the main reference for the empirical works on corruption. These theories have in common that they can be summarized by the following proposition: corruption is positively correlated with incidence of high rent activities and negatively with the monitoring efforts. Although the approaches and further conclusions vary, this is a constant base in corruption research. Among the models for corruption, is worth mentioning Acemoglu and Verdier (1998). Their main contribution is the discussion that some level of corruption may be optimal. Their model considers that, in order to avoid corruption from bureaucrats, it would the required the hiring of another group of bureaucrats, higher in the hierarchy chain, which would be entitled with monitoring the work of the remaining bureaucracy. For the size of this monitoring bureaucracy to be sufficient to assure complete control of corruption, it would be too costly for the state. On the other hand, if this monitoring bureaucracy is too small, the state would not be fulfilling 4

13 its role, so that society would be better off without a public sector. As in Rose-Ackerman (1975), their conclusion is that you cannot avoid corruption completely. Ades and Di Tella (1999) also derive a model that refers to Rose-Ackerman (1978) and Shleifer and Vishny (1993). Their approach is mainly based on the literature of compensation. They model an efficiency wage model to capture the effect of rents on wages and find an equilibrium level of corruption, as in Acemoglu and Verdier (1998). They conclude that the efficiency wage of a bureaucrat must include the amount in rents he may receive if corrupt but also the probability of detection. The authors argue that higher-than-average wages are usually too costly for the state to implement, even though it would reduce corruption. However, if competition increases, wages are lowered. The main conclusion of the authors is that competitiveness has a positive effect in reducing corruption. Regarding the level of wages, Acemoglu and Verdier (2000) further show that when the probability of being caught is high and the rent prospect from the bribe is low, corruption will be easier to prevent with the same level of wages used to attract workers to the private sector. The exact opposite scenario will give agents the most incentive to practice corruption, so that the government will have to hire more agents and pay the optimal size rent, internalized in their wages, to prevent corruption. Most of the research on corruption that followed this theoretical framework was concerned with developing ways to test these hypotheses empirically, so that they could find evidence to support policy makers in fighting corruption. As summarized by the World Bank (e.g. Gray and Kaufman 1998), corruption is a worldwide problem with serious consequences to the economy and politics of the countries affected, mainly: Bribery is widespread ; Bribery raises transactions costs and uncertainty in an economy; Bribery leads to inefficient economic outcomes. It impedes long-term foreign and domestic investment, misallocates talent to rent-seeking activities, and distorts sectoral priorities and technology choices It pushes firms underground, undercuts the state s ability to provide essential public goods, including the rule of law. A vicious circle of increasing corruption and underground economic activity can result; Bribery is regressive, falling heavily on trade and service activities undertaken by small enterprises; Corruption undermines the state s legitimacy. (Gray and Kaufman, 1998, p.1) Research is interested in understanding the causes of corruption, its costs and how to fight it. 5

14 The determinants of corruption One of the first and most important empirical works is Barro (1991), although not on the determinants of corruption but on the determinants of growth. His biggest contribution is his remark that there is something specific in Latin America that is lacking in his model to explain the regions low long term growth rate. Latin America s high levels of corruption is a plausible explanation. Corruption could be an omitted variable which is causing bias in his model. The study of the effect of corruption on growth and the channels through which corruption impacts growth is the main goal of many researches that followed him, such as Mauro (1995), La Porta et al. (1999), Leite and Weidmann (1999), Ades and Di Tella (1999), Treisman (2000), Gyimah-Brempong and Camacho (2006), Clarke, Pendergast and Van Kooten (2011). Another important reference in this research field is La Porta et al. (1999).The authors investigated the determinants of government quality, given the importance it has for economic growth and found evidence that both bureaucratic delays and tax compliance are positively correlated with corruption and that corruption is negatively correlated with the relative wages in the public sector. Ades and Di Tella (1999) perform an empirical analysis based on the model presented in the theoretical background. The results from their empirical analysis corroborate the proposition that competition is an important determinant of corruption, its increase leading to a reduction in the level of rents and consequently of the corruption levels. This evidence reinforces the argument that making markets more competitive is fundamental when fighting corruption. Treisman (2000) contributed to this discussion by revising the previous works in the area and adding new variables as determinants of corruption. His results are in line with most of the findings in Mauro (1995), La Porta et al. (1999) and Ades and Di Tella (1999). From the economic theory and the evidence from these authors, Treisman (2000) wants to investigate as determinants of corruption the following factors: legal system, religion, democracy, freedom of press, economic development, and wages in the public sector. The author believes the legal system to have a great impact on the quality of the government and, therefore, on the level of corruption, due to the legitimation of the government for ensuring property rights, as in Acemoglu and Verdier (1998). The main conclusion from Treisman (2000) is that there is strong evidence that supports historical variables (the colonizer country, religious tradition and the duration of democracy) as the most relevant determinants of corruption in the cross country analysis. 6

15 Gyimah-Brempong and Camacho (2006) have contributed to the field by studying not only how corruption affects growth but also how it affects income distribution. The results showed that corruption is negatively and significantly correlated with growth. A 10% increase in corruption is found to decrease the growth rate of per capita income by 1.7%. For the regional effect, the authors created interaction terms with the dummies for region and the corruption variable, the reference being the Organization for Economic Co-operation and Development (OECD) countries. The interaction terms for Africa and Latin America were positive and significant. They found evidence that in Africa and Latin America this effect from corruption on growth is higher due to its impact on the physical capital investment. Their results also showed evidence that corruption has a significant effect on income distribution. A one standard deviation increase on corruption increases the Gini coefficient by 0.25 points in Africa and 0.33 in Latin America. Their final conclusion is that the effect of reducing corruption will have the greatest impact for Africa and Latin America, both for increasing the growth rate of per capita income as for reducing inequality. Corruption and natural resources abundance Besides historical factors, many researchers found evidence that resource abundance can explain cross-country differences in corruption levels and long-run growth rates (e.g. Barbier 2003; Sachs and Warner 1995; Leite and Weidmann 1999; Pendergast et al. 2011). Barbier (2003) focus on the importance of natural resources to a country, specifically the importance of their management for a sustainable economic development. The main question he addresses is why the majority of middle and low income countries abundant in natural resource have such low growth rates. With a good management of the resources, these countries were expected to depend on their natural resources extraction only while supporting their transition to an economy focused on industry and services. The cross-country reference shows that this expected takeoff is not happening though. His conclusion, is that this is due to corruption and bureaucratic inefficiency. Sachs and Warner (1995) also study the determinants of economic growth, including resource abundance as one of the determinants in their model. They develop a simple OLS growth model to investigate how it is affected by: natural resources, trade openness, investment, the bureaucratic efficiency, and external terms of trade and income inequality. The authors found evidence that all of these variables but external terms of trade and income inequality were significant. They conclude that resource abundance has a direct effect on growth, beside its 7

16 indirect effects as indicated by the theory. The indirect channel being investment, trade openness and bureaucratic efficiency, which is a measure of corruption. Leite and Weidmann (1999) follow Mauro (1995) in studying empirically the determinants of corruption and its effect on growth. They believe corruption affects growth through other channels than investment. Economic theory predicts the negative effect of the Dutch disease to last only in the short run, although that is not what cross-country evidence has shown. Leite and Weidmann (1999), as Barbier (2003), believe that the missing piece to explain the evidence of lower growth in natural resource abundant countries is corruption. They also believe that this negative correlation between growth and abundance in resources can be due to the endogeneity of corruption. For the purpose of their research, Leite and Weidmann (1999) separate the Dutch disease effects on the economy into direct and indirect effects. They believe corruption and institutions to be the channels through which natural resource abundance affects growth. Besides, the authors believe that different resources will impact corruption differently and they want to account for this difference. Thus, they divide natural resources into the categories: fuel, ores and metals, agricultural products, and food. Their results found a negative impact from fuel and ores, meaning that they increase corruption, which is in line with the theory, those being capital intensive activities. The trade openness was found to have a positive impact in reducing corruption. Rule of law, which measures the government monitoring efforts, was also significant, indicating it has a positive effect on corruption. Where monitoring efforts are greater, corruption is lower. Corruption has a positive and significant coefficient, meaning the greater the corruption the lower the long-term growth. Given this results, the authors conclusion is that corruption affects growth through trade policy, quality of institutions and political stability. Looking at the growth effect of natural resources specifically, Leite and Weidmann (1999) found that for developing countries, downturn in commodity markets has a negative impact on the long-run growth rate. In line with the negative and significant effect of government consumption, they believe the explanation comes from the rents from the natural resources representing a direct increase in revenue to the government, who increases its consumption instead of investing it. The magnitude of the effect of fuels and minerals in their analysis was lower than what was found by other researches, this effect being due to their impact on corruption. When discussing the effectiveness of anti-corruption policies they point out that it 8

17 is more effective for less developed countries to invest on monitoring institutions and technologies, while for developed countries, it is more effective to increase penalties. Clarke, Pendergast and Van Kooten (2011) are in line with Barbier (2003) in identifying that corruption affects economic growth more than the endowment of natural resources itself. They believe that the typical measure used in the literature for natural resources abundance, the share of primary exports in GDP, is actually a measure of the dependence on those natural resources. This is so because natural abundance is known to be positive for the growth of a country, while the increase on the dependence to natural resources, meaning a growth in the share of primary sector in spite of the secondary and tertiary, is negative. The authors recur to structural changes development theories, which states that countries must transition from predominant primary activities to secondary and tertiary, that have higher value-added. In that way a country can develop and achieve higher standards of living. This also explains why countries that are more dependent on natural resources/primary sector will be worse off. Cross-country evidence that most countries developed by industrializing enforces these theories. Besides, they also argue that when countries have low growth rate but high standard of living, they are not considered to have the curse. So the focus of Clarke, Pendergast and Van Kooten (2011) is to understand the relation between natural resources dependence and standard of living. The authors consider standard of living, development and well-being as somehow analogous measures that consist of three aspects: material, physical and psychological. They believe that the HDI from the United Nations is a perfect proxy for it, since it puts equal weights on income levels, health and education. So they use GDP per capita and the HDI to measure well-being. They are also concerned with their model being in per capita terms to account for the country s population size. A country very rich in resources but also in amount of population will have more difficulty in transforming the rents from natural resources exploitation into an increase of the standard of living. The results in Clarke, Pendergast and Van Kooten (2011) showed that latitude is positive and significantly correlated with well-being and control of corruption is positive and significantly correlated with HDI. Both confirming the first part of the resource curse predictions. On the corruption regression, different types of resources showed different effects. Ores, metals and fuel exports had a negative effect on control of corruption while forestry had a positive effect. It seems that ores, metals and fuels are the most important for the resource curse. They also found evidence of a positive relation between control of corruption and regulatory quality, meaning better institutions help fighting corruption even when there is 9

18 a resource curse. Finally, the authors conclude that control of corruption has a greater impact on well-being than investment and trade openness. Therefore, they find evidence that the effect of resources, mainly minerals and fuels is negative for the fight against corruption, and corruption has a negative impact on development/well-being. Literature reviews Even though I believe to have covered the main articles, I do quote three literature review articles that discuss briefly the main advances in the corruption research regarding its measurement and results. The most important studies on corruption until the beginning of the 2000 s have been revised by Svensson (2005). His main discussion revolves around the limitations of perception based and ordinal measures of corruption, although those were the most commonly used measures of corruption in the literature then. The Corruption Perception Index from the Transparency International being one of those. His main remarks are to the data limitation on the field and the lack of robust conclusions regarding the relation between corruption and some economic variables due to the measure and endogeneity of the variables. He also points to the need of further research for detecting which are the most effective measures against corruption and for which countries, since not necessarily the same strategy may work for every country. Treisman (2007) performs a literature review focused on the many cross-country papers from the 1990 s and 2000 s that studied corruption using subjective indexes as a measure. He wants to review this works in terms of their data choice, the methods employed in the analysis and the robustness of the results. Regarding the methodological approach, Treisman (2007) discusses the endogeneity between corruption and its determinants, which makes it hard to establish the causality direction. Besides, he mentions that the identified correlation between them could actually be caused by a third factor, omitted in the model. An additional cause for concern, is that the variables that seem to have a significant effect on corruption are highly correlated among themselves, making it difficult to isolate their individual effects. To evaluate the methodological issue further, the author reproduce other researchers models with perception data for different periods of time and adjust the time period for the rest of the variables used by them. His results confirmed the robustness of the correlation between economic development and perceived corruption, though arguing that only with a good instrument the causality could be 10

19 defined. He criticizes the instrument used by Mauro (1995), the ethnolinguistic fractionalization. The results about the effect of democracy on corruption are also discussed in Treisman (2007). He finds that long lasting democracies with strong freedom of the press are perceived as less corrupt. The endogeneity is again a problem. Not only democracy may be a determinant of corruption, but corruption can have an effect on democracy. The author also considers that corruption could be caused by the same factors that weakens democracy. Furthermore, he discusses the endogenous challenges of measuring the effect of market structure. He identified no significant correlation between state regulatory intervention and perceived corruption. For trade openness there was significant support for the robustness of the year the country opened up to international trade, since trade openness takes time to affect the economy, and exports of fuels. Finally, a more recent literature review is presented by Olken and Pande (2012). Their perspective is more in the line of Svensson (2005) than of Treisman (2007). They want to investigate all the datasets and methodologies available in the literature and how it has advanced, also in what yields anticorruption strategies. Under the approaches they discuss are the recent literature on corruption in Brazil: Ferraz and Finan (2008, 2011) and Brollo et al. (2010). These articles use data from official audits of municipalities in Brazil. Olken and Pande s (2012) critique to the use of this data is that they believe that not only does this measure capture corruption incidence, but it also captures the inability of corrupt agents in hiding from the auditors. One interesting aspect of the review made by Olken and Pande (2012) is about the literature that studies the impact of corruption on public goods. According to the authors, the effect is mainly in increasing the cost of the government provision of them, due to the cost of corruption. Also new efficiency costs will appear due to the distortions caused by corruption. The result will be the reduction on the provision of public goods, since many projects may become noncost-effective. It can also happen that the government will not foresee this corruption cost, allowing an amount of projects to be run and once the cost of corruption arises, underfund some of those. One example is from Ferraz et al. (2010) which provides evidence of the direct efficiency loss caused by corruption in municipalities in Brazil. They show significant evidence of a worse performance by students from municipalities that had education funds diverted in comparison to the control group. 11

20 From this part of their review, the most interesting aspects are: the impact of punishment on corruption, which factors are relevant for the decision of becoming a bureaucrat/politician and its effect on corruption. From Ferraz and Finan (2008; 2011) the authors present the conclusions that accountability and re-election prospects have a significant effect in reducing corruption. Nevertheless, the authors do not mean that infinite possibilities of re-running for office is a solution, as the effect of the punishment if caught (not being re-elected) would be lost. Regarding selection, also from Ferraz and Finan (2010) the authors quote that there is evidence, although weak, that higher salaries attract better candidates to run for office. Higher salaries also seem to improve the performance of the mayor during the term. To conclude, Olken and Pande (2012) mention the importance given for transparency and the benefits of technology for anti-corruption strategies. Transparency is mainly based on the idea of accountability: if citizens can have access to information regarding government projects, spending and performance, they can monitor and demand more of the elected politicians, with their main power being their vote. It will also allow them to signal their interests by public manifestation, for example. It also gives access to information on what are the citizens rights, with the prospect of making the population more politically aware and engaged. Evidence in the literature has shown support to that strategy. In regard to technology, it can allow the development of software, among other tools, that will help fighting corruption. For example, by making tempering with information difficult or stopping agents from acting outside the procedures and hierarchies in firms/public sector. It can also provide better administration of public data, for example taxes. From all the possible advantages, the authors call attention to online procurement by the government. There is evidence since Rose-Ackerman (1975), that procurement and contracts are the most vulnerable to corruption. Finally, the authors mention the creation of high-profiled independent anticorruption agencies with prosecutorial powers to fight corruption nationally. They also mention that there is not enough evidence to judge if these agencies are effective in fighting corruption. 2.2 Corruption in Brazil The development of the corruption literature has also been observed in research focused in the study of the Brazilian case. Brazil is known worldwide for its endemic corruption, given its periodic uncovering of corruption schemes in the media. There is also evidence from the crosscountry literature that corruption is significant and negatively correlated with long term growth rate, and Brazil, together with other Latin-American countries, is considered a remarkable 12

21 example where high corruption levels are associated with low long run growth rates. It makes researchers believe there is something special about Latin-America in general, not being captured by their models (e.g. Barro 1991; Gyimah-Brempong and Camacho 2006; La Porta et al. 1999). Recent anti-corruption efforts in Brazil led to the creation of new monitoring agencies and new laws of access to information and transparency. Additionally, new data has become available regarding the auditing of governmental transfers and programmes to municipalities in Brazil. This combination made Brazil a very attractive and convenient country for the study of corruption. The most important papers on corruption literature in Brazil are: Ferraz and Finan (2008, 2011), Caselli and Michaels (2009), Monteiro and Ferraz (2010), Ferraz et al. (2012), Arvate and Tavares (2012), Brollo et al. (2013) Boas, Hidalgo and Richardson (2014), Britto and Fiorin (2016) and Avis, Ferraz and Finan (2016). All of these papers use the data from the CGU s audits of municipalities, which have been performed in randomly selected municipalities from 2003 to The audits inspect the municipality s use of federal funds transferred by specific areas or projects. The auditors report on the irregularities they find, classifying it according to their severity level. These irregularities have been used in the literature to proxy corruption for the study of different purposes. Corruption effect on electoral outcomes and the role of accountability Ferraz and Finan (2008) became the most quoted paper in the literature of corruption in Brazil for being pioneers in using data from the CGU s audit of municipalities program. The authors use data from the 13 first lotteries of the program to estimate the effect of the release of the audit results on electoral accountability. The authors have read every report from these lotteries, which happened between 2003 and 2005, and classified as corruption those irregularities associated with fraud in procurement, diversion of funds or over-invoicing were used as a proxy for corruption. The authors also gathered the information of the amount of federal transfers received by the government area audited and the amount of funds involved in irregularities. They identify the majority of the municipalities in this sample as being poor and with small density. On average, 38% of the population in the sample is rural, 21% is illiterate and 79% of the households own a radio. To study the effect of political accountability, the authors compare the electoral outcome in two groups of municipalities, both audited by the CGU and where majors were eligible for reelection. In the first group, the CGU audit reports were released before the elections, while in 13

22 the second, the reports were released after the elections. In their model the dependent variable is a dummy for whether the mayor running for re-election won in Their independent variables include the number of corrupt irregularities, a dummy to indicate if the municipality was audited before October 2004 (when elections took place in Brazil), variables to measure municipality and mayor s characteristics, state fixed effects and an interaction term between the dummy for being audited before 2004 and the number of irregularities. Their coefficient of interest is the one for this interaction term, which is capturing the impact of being audited, given the municipality level of corruption, on the electoral outcome. In a second specification, the authors run the same regression adding the presence of local media as the number of local AM radio stations. Their focus in this specifications is the coefficient of the interaction term between the dummy for being audited prior to 2004, the municipality s corruption level and the presence of local media. This will capture the effect of the audit in the electoral outcomes, given the corruption level and the media presence in the municipality. Ferraz and Finan (2008) found that where no evidence of corruption was found by the auditors, 53% of incumbents were re-elected while, when 3 cases of corruption were reported, only 20% of the mayors were re-elected. This evidence implies that voters do punish corrupt politicians. Their results indicate that the probability of re-election is reduced with every increase in cases of corruption. They conclude that not only do voters punish politicians, but they punish them harder in the presence of local media. Ferraz and Finan (2011) present a new approach to the electoral accountability problem studied in Ferraz and Finan (2008). In this previous work, they compared incumbents eligible to reelection and the impact of the disclosure of corruption by the governmental audits on their reelection probabilities. Ferraz and Finan (2011) estimate the impact of reelection incentives by mayors on the corruption level. Thus, for this analysis, they will compare incumbents that are already on their second term, and therefore are not eligible for re-election, to those in their first term. When doing so they take into consideration the accountability effect they studied before. The results from the different specifications (OLS with different controls, matching and Tobit) showed that being a first term mayor has a negative and significant impact on the corruption level in all specifications. The specification controlling for state and lottery intercepts and mayor and municipality characteristics found that having a first term mayor in power leads to a 27% reduction in the share of resources involved in corruption in relation to second term mayors. 14

23 Therefore, both articles show evidence that electoral accountability is a mechanism that discipline politicians, with robust results that indicate that first-term mayors are considerably less corrupt than second-term mayors. The challenge, once accountability is known to be positive in reducing corruption, is how to increase the first. The most recent paper of the same authors, together with Eric Avis, is Avis, Ferraz and Finan (2016). In this paper, they investigate the effect of the audits in reducing corruption. There is a great concern in the literature, as expressed by Svensson (2005), in understanding the effectiveness of the most commonly adopted anti-corruption strategies in the world. With this paper, the authors not only contribute to the study of the determinants of corruption, but also show evidence that the creation of monitoring agencies and auditing programs are effective in reducing the levels of corruption. Avis, Ferraz and Finan (2016) compare the level of corruption in municipalities that are audited for the first time, which is their control group, and the municipalities that have been audited before, their treatment group. The authors expect that being audited before may affect corruption through 4 possible mechanisms. When considering accountability, there may first be an effect of political selection, as voters are expected not to re-elect corrupt politicians. This effect may also work through changing the perceived probability of detection, so that mayors who wish to re-run may reduce their level of corruption. Third, they argue that it could be that there are non-electoral costs involved, such as reputation or legal costs. And finally, the reduction may happen through a political entry effect, changing the mayor type interested in competing for office. Their study suggests that 94% of the disciplinary effect on reducing corruption is due to non-electoral costs. At the same time, the audits may have spill-over effects, reducing the neighboring municipalities level of corruption due to the local media coverage. The data from the CGU is used to create two measures: corruption and mismanagement. Differently from their previous articles, here corruption includes all the irregularities classified by the authors as severe and moderate, while mismanagement includes those administrative and procedural irregularities. First the authors run an OLS regression with corruption as dependent variable. The independent variables include a dummy of whether the municipality was audited before time t when corruption is measured and the number of service order per municipality. The more service orders that are investigated, the more irregularities will be found. The authors include variables to account for municipal characteristics such as income 15

24 per capita, income inequality, population density, share of urban population and share of the population that is literate, besides including state fixed effects, to capture that the randomization is by state, so the probability of a municipality being audited before will be the same for all municipalities in the same state. They include lottery fixed effects as well, since the CGU has changed its methodology from the 20 th lottery on, focusing the audit in selected areas and programmes. Given that the audits are random, the coefficient of the dummy for being audited before will measure the effect of audits on corruption. Secondly, Avis, Ferraz and Finan (2016) want to estimate the effect of audits on the investigation of mayors by the federal police Special Operations, which is not in the scope of this thesis but is in the line of what I believe should be the focus of further research. With this dataset, the authors create a dummy indicating if the municipality was subjected to a crackdown from the Federal Police in the year of interest, and another dummy indicating if the mayor was involved in the irregularities investigated. Data on mayors charged for misconduct in public office is also used for their analysis. A list with all the names of individuals charged for misconduct in public office is available from the Conselho Nacional de Justiça (CNJ) (In Portuguese. National Council of Justice) 5. Their dependent variable is whether there was a Special Operation run in the municipality or if the mayor was investigated and registered in the CNJ. The independent variables are a dummy for if the municipality was audited, municipal year effects, year fixed effects and error term clustered at the level of municipality. The coefficient of the dummy for being audit will measure the effect of the audit on the likelihood of incurring a legal action. For both first and second specifications, the authors run different variants. The results from the OLS regressions indicate for the first set of regressions that being audited in the past has a negative effect on corruption, reducing the number of irregularities found. When controlling for socioeconomic variables and number of service orders, the authors found the number of irregularities to drop by 5.8 percent. Under the most complete specification, being audited in the past had a significant effect in reducing corruption by 7.9 percent compared to the municipalities that were not audited before. They also found municipal characteristics to be determinants of corruption: income per capita is negative and significantly associated with corruption. The authors run the same specifications using mismanagement as dependent 5 A person listed in this registry is not allowed to run/hold office for the next 5 years. 16