Economics 191. Economics of Corruption. Professor Fred Finan. March 13, PDF Free Download

Economics 191 Economics of Corruption Professor Fred Finan March 13, 2012

Lecture 4 outline Overview of the corruption literature How do we measure corruption? What determines corruption? What determines corruption? Case Studies: Ferraz and Finan (2008;2011)

Corruption Research on corruption can be divided into three main questions: 1. What is corruption? or How do we measure corruption? 2. Does corruption matter for development? 3. What determines corruption? or How can we reduce or prevent corruption?

What is corruption? Corruption comes in many forms Transparency International (TI) and others: the abuse of public office for private gain Some examples are fairly obvious:

What is corruption? Corruption comes in many forms Transparency International (TI) and others: the abuse of public office for private gain Some examples are fairly obvious: Mobutu Sese Seko looted the treasury for $5 billion equal to Zaire s entire external debt in 1997 Mohamed Suharto and Ferdinand Marcos allegedly embezzled between 10 to 35 billion dollars $1 billion of oil revenues, or $77 per capita, vanished from Angolan state coffers in 2001 3 humanitarian aid received by Angola in 2001

What is corruption? Other situations are less obvious: Related lending (La Porta et al 2002)

What is corruption? Other situations are less obvious: Related lending (La Porta et al 2002) Earmarks for education are used to pay health expenditures

What is corruption? Other situations are less obvious: Related lending (La Porta et al 2002) Earmarks for education are used to pay health expenditures What about legal payments that involve lobbying, campaign contributions, and gifts? Are these so different from bribes?

Measuring corruption Most common measures of corruption are the cross-country indices International Country Risk Guide (ICRG) - likelihood that high government officials will demand special payments Corruption Perception Index (Transparency International) - averages ratings reported by a number of perception-based sources Control of Corruption (World Bank) - similar to TI but with a different aggregation technique Correlation between TI and World Bank is 0.97; TI and ICRG is 0.75 Main difference is in coverage both in terms of countries and years

Cross-country measures of corruption

Cross-country measures of corruption Advantages Allows for cross-country and time comparisons

Cross-country measures of corruption Advantages Allows for cross-country and time comparisons Disadvantages Subjective measures Perceptions given by the private sector, primarily foreign investors Other characteristics or institutional features of the country may influence corruption rankings

Measuring corruption - direct approach Svensson (2003) QJE - surveyed Ugandan firms directly 80 percent of firms reported needing to pay bribes Bribes amounted to 8 percent of the costs (conditional mean) 20 percent that did not pay bribes chose to minimize contracts with the public sector Ferraz and Finan (2007) QJE - exploit audit reports Brazilian government audited municipal expenditure of federal funds Corruption was detected in 78 percent of municipalities On average 9 percent of all federal funds were stolen Olken and Barron (2009) JPE - direct observation of bribes Truck drivers in Indonesia spend about 13 percent of total cost on bribes

Direct observations - Olken and Barron 2009

Measuring corruption - indirect approach Reinikka and Svensson (2004) QJE - compare the amount of education funds disbursed from the central government to the amount funds received (based on a survey of 250 schools) Over 1991-1995, schools received only 13 percent of central government spending on program Olken (2007) JPE - compares actual versus reported expenditures on World Bank funded road projects 29 percent of the funds allocated to the road building project were stolen Di Tella and Schargrodsky (2003) JLE - compare prices paid for basic homogenous inputs at public hospitals before and after a corruption crackdown prices paid fell by 15 percent during the first nine months of a crackdown on corruption in 1996-1997

Does corruption matter for development? In terms of economic growth, the only thing worse than a society with a rigid, over-centralized, dishonest bureaucracy is one with a rigid, over-centralized and honest bureaucracy -Samuel Huntington

Corruption need not distort the short-run efficiency Producer surplus goes to the bureaucrat instead of the treasury Pareto-optimal allocation is unaffected

Corruption need not distort the short-run efficiency Producer surplus goes to the bureaucrat instead of the treasury Pareto-optimal allocation is unaffected If bribing is competitive, then the lowest cost firm can provide the largest bribe still have the efficient allocation Theory of the second best approach With excessive taxes and overly restrictive regulation, speed money or grease payments may also improve economic efficiency

Problems with the efficiency argument 1. In many countries, corruption is illegal

Problems with the efficiency argument 1. In many countries, corruption is illegal Bureaucrats will induce substitution into the goods in which bribes can be more easily collected without detection distorting allocations Corruption contracts are not enforceable in a court of law Bureaucrats can create holdup problems

Does corruption matter for development? - Empirical evidence

Does corruption matter for development? - Empirical evidence At the macro-level corruption does appear negatively correlated with:

Does corruption matter for development? - Empirical evidence At the macro-level corruption does appear negatively correlated with: Direct/Foreign investment

Does corruption matter for development? - Empirical evidence At the macro-level corruption does appear negatively correlated with: Direct/Foreign investment Quality of public infrastructure and services Trust in government But these are only correlations there are many concerns with cross-country regressions

Evaulation problem A QUICK DIGRESSION

Counterfactual Suppose we are interested in estimate the effects of anti-corruption program on education outcomes (e.g. Reinkka and Svensson)

Counterfactual Suppose we are interested in estimate the effects of anti-corruption program on education outcomes (e.g. Reinkka and Svensson) To answer this question we need an answer to the counterfactual question What is the counterfactual question?

Evaluation problem Maybe we can compare the same person over time? Schooling 0.3 (Observed) Is this your impact? 0.2 (Observed) 2007 PROGRAM 2008 time

Evaluation problem Maybe we can compare the same person over time? Schooling 0.3 (Observed) Is this your impact? 0.2 (Observed) 2007 PROGRAM 2008 time What s the problem with this design? What is the counterfactual assumption for this design?

Evaluation problem We need to identify what would have happened in the absence of the program Schooling 0.3 (Observed) (What would have happened without the program) 0.2 2007 2008 time

Evaluation problem Pre-post comparison Counterfactual assumption Enrollment status in 2008 would have been the same as 2007 if program had not taken place Other determinants of schooling enrollment change over time (e.g. schooling cost, schooling preferences) We would be attributing to the program, enrollment changes due to other factors = E[Schooling 2008] E[Schooling 2007] = Program+Other stuff

Evaluation problem We need a comparison or a control group

Evaluation problem We need a comparison or a control group But how do we find a comparison group?

Evaluation problem We need a comparison or a control group But how do we find a comparison group? One idea is to use children in places where the anti corruption program did not reach.

Evaluation problem We need a comparison or a control group But how do we find a comparison group? One idea is to use children in places where the anti corruption program did not reach. What s the problem with this idea?

Evaluation problem Depends on the program selected which places to go (selection problem)

Evaluation problem Depends on the program selected which places to go (selection problem) In general those who are exposed to a program differ from those who are not exposed to the program

Evaluation problem Depends on the program selected which places to go (selection problem) In general those who are exposed to a program differ from those who are not exposed to the program These differences then may matter for the outcome and it confounds our estimates of the effects (Selection bias)

Selection bias Enrollment Treatment Group Difference prior to intervention 2007 2008 time

Potential outcomes Now that the intuition is clear (hopefully...if not ask now because it will only get harder) let s fix some ideas with math.

Potential outcomes Now that the intuition is clear (hopefully...if not ask now because it will only get harder) let s fix some ideas with math. Y T i - schooling of student i if he receive the grant program

Potential outcomes Now that the intuition is clear (hopefully...if not ask now because it will only get harder) let s fix some ideas with math. Y T i Yi C - schooling of student i if he receive the grant program - schooling of student i if he does not receive the grant program We are interested in the difference of: Y T i Y C i

Missing data problem Person Treatment Control 1 1. 2 1. 3. 1 4. 1 5. 0 6 0. 7. 0 8. 0 9 1. 10 0. Central question: How do we fill in these missing values?

Counterfactual problem Given the data structure, what we observe is: Y i = (1 W i )Y C i + W i Y T i where W i is an indicator variable for whether child i received the program.

Counterfactual problem Given the data structure, what we observe is: Y i = (1 W i )Y C i + W i Y T i where W i is an indicator variable for whether child i received the program. Ok let s make the problem slightly easier. Say we are interested in the following (notice the expectation operator): E[Y T i Y C i ] That is the average effect of the program. So how do we proceed with this?

Potential outcomes Suppose we computed the difference in average outcomes between those who received the program and those who did not receive the program, i.e. D = E[Y T i = E[Y T i Program] E[Yi C No Program] W = 1] E[Y C i W = 0] Now let s add and subtract a term D = (E[Yi T W = 1] E[Y C + (E[Y C i i W = 1]) W = 1] E[Yi C W = 0])

Potential outcomes D = (E[Yi T W = 1] E[Yi C W = 1]) +(E[Yi C W = 1] E[Yi C W = 0]) }{{} treatment effect first difference is the treatment effect (or more specifically, the treatment effect on the treated (TOT)) - treatment effect for those that participated in the program

Potential outcomes D = (E[Y T i W = 1] E[Y C i W = 1])+(E[Y C } i W = 1] E[Yi C W = 0]) {{} selection bias E[Y C i W = 1] E[Yi C W = 0] is the selection bias

Potential outcomes D = (E[Y T i W = 1] E[Y C i W = 1])+(E[Y C } i W = 1] E[Yi C W = 0]) {{} selection bias E[Y C i W = 1] E[Yi C W = 0] is the selection bias Difference between individuals who would have received the grants and those that did not (in the absence of the treatment)

Potential outcomes With selection bias we cannot estimate the causal impact of the grants on schooling

Potential outcomes With selection bias we cannot estimate the causal impact of the grants on schooling Moreover it may be difficult to sign the bias.

Potential outcomes With selection bias we cannot estimate the causal impact of the grants on schooling Moreover it may be difficult to sign the bias. For instance, program kids may have had a higher propensity to go to school on average even if they had not received the program This would be true if the program targeted families where parents considered education important. The opposite would be true if the program instead targeted the poorest children.

Gold standard One solution is to perform a randomized evaluation

Gold standard One solution is to perform a randomized evaluation If you randomly assign treatment to subjects, then Y i will be independent of W i (mean independent).

Gold standard One solution is to perform a randomized evaluation If you randomly assign treatment to subjects, then Y i will be independent of W i (mean independent). D = TOT + E[YI C W = 1] E[YI C W = 0]

Gold standard One solution is to perform a randomized evaluation If you randomly assign treatment to subjects, then Y i will be independent of W i (mean independent). D = TOT + E[Y C I W = 1] E[Y C I W = 0] How can we rewrite this equation if Y i and W i are independent?

Gold standard Randomization assures us that in the absence of treatment, children that would have received the program and those that would have not received it are the same (in expectation) Also note that in randomization, average treatment effect and treatment effect on the treated are the same: E[Y T i Yi C ] = E[Yi T Yi C W = 1] In sum, randomization solves the missing data problem. We can compute the average effect for the population without observing the individual counterfactuals.

Does corruption matter for development? - Empirical evidence Bertrand et al (2006) conducted an experiments among individuals who were interested in getting a driver s license in India. They randomly assigned people into three groups: bonus (offered a financial reward if they could obtain their license fast) lesson (offered free driving lessons upfront) control

Does corruption matter for development? - Bertrand et al 2006

Long run impact of corruption Corruption in the public sector can reduce the provision and quality of key public services (e.g. education, health, infrastructure) If these human resources are important for development and economic growth, then corruption can impose serious long-run costs Ferraz, Finan, and Morreira (2009) examine this possibility using the audits data in Brazil

Ferraz, Finan, and Morreira 2009 PISA 2006 reading residual -100-50 0 50 100 ISL FIN HKG NZL AUS NLD GER GBR SWE CHE NOR DNK AUT LUX IRL BEL EST FRA JPN CHL ESPPRT SVN URY ISR KOR HUN MAC CZE POL SVK LVA GRC LTU HRV TUR ITA THA ROMEX JOR BRA ARG BGR COL TUN -3-2 -1 0 1 Country corruption measure

Ferraz, Finan, and Morreira 2009 PISA 2006 reading 350 400 450 500 550 600 MAC URY ROM CHL ARG CZE KOR FIN HKG IRL NZL AUS NLD ESTBEL GER GBR FRA NOR HRV LUX LVA SVK LTU GRC ESP TUR ISR THA MEX JOR BRA COL TUN POL SWE JPN CHE AUT DNK SVN HUN ISL ITA PRT BGR 10 15 20 25 Spending on primary education/gdp per capita

Ferraz, Finan, and Morreira 2009 Test scores 100 150 200 250 300 0 500 1000 1500 2000 Expenditure per pupil bandwidth =.8 Test scores 100 150 200 250 0 500 1000 1500 2000 Expenditure per pupil bandwidth =.8 Panel A: Mathematics Panel B: Portuguese

Ferraz, Finan, and Morreira 2009 kdensity mat_4a 0.01.02.03 kdensity port_4a 0.01.02.03 100 150 200 250 300 100 150 200 250 No corruption Corruption No corruption Corruption Panel A: Mathematics Panel B: Portuguese

Ferraz, Finan, and Morreira 2009 Dependent variable: Mathematics Portuguese Dropout rates Failure rates (1) (2) (3) (4) Corruption in education -0.28-0.279 0.034 0.027 [0.120]** [0.100]*** [0.011]*** [0.012]** Corruption in other sectors 0.023 0.014 0.011 0.012 [0.116] [0.096] [0.010] [0.011] Student characteristics Yes Yes Yes Yes Municipal characteristics Yes Yes Yes Yes Number of schools 1251 1251 1251 1251 R-squared 0.48 0.54 0.29 0.17

What determines corruption? Ambiguous laws and regulations (Ades and di Tella 1997; Shleifer and Vishny 1993) Opportunity for abuse of power (Ferraz and Finan 2009) Low income or education (Lipset 1960) Poor enforcement of property rights and laws Ethnic heterogeneity (Alesina, Bagir, Easterly 2002) Historical and cultural factors (Kranton 1996; Fisman and Miguel 2006) Market structure (Shleifer and Vishny 1993)

What determines corruption? Consider the decision of a bureaucrat to engage in corruption w - wage from government v - outside option p - probability of being detected if corrupt b - bribe (exogenous) d - utility cost of being dishonest if undected

What determines corruption? If the bureaucrat engages in corruption, he gets pv + (1 p)(w + b d) If the bureaucract does not engage in corruption, he gets w. He will be corrupt iff w v < 1 p (b d) p

What determines corruption? w v < 1 p (b d) p Simple model provides several predictions w - reduces corruption (wages) p - reduces corruption (monitoring) d - reduces corruption (social norms) Suppose that we replaced d with d i, what does this imply about selection effects?

Information and accountability - Ferraz and Finan 2007 Corruption is major obstacle to economic development Information asymmetries are a major contributing factor to widespread prevalence of systemic corruption Numerous countries have adopted anti-corruption policies predicated on transparency In a functioning democracy, the provision of information can have two effects: 1. Discipline policy makers 2. Empower citizens to select better policymakers

Information and accountability - Ferraz and Finan 2007 Research question: Does disclosing local government corruption practices affect the re-election success of mayors in municipal elections?

Information and accountability - Ferraz and Finan 2007 Research question: Does disclosing local government corruption practices affect the re-election success of mayors in municipal elections? A point about the theory The effects of providing information ultimately depend on people s prior beliefs Voters may punish corrupt politicians but it assumes voters care about corruption politicians committed more corruption than expected If corruption is revealed but less than voters expected then information may actually improve re-election chances

Information and accountability - Ferraz and Finan 2007 What did the previous empirical literature have to about this question?

Information and accountability - Ferraz and Finan 2007 What did the previous empirical literature have to about this question? Not much! For at least two reasons:

Information and accountability - Ferraz and Finan 2007 What did the previous empirical literature have to about this question? Not much! For at least two reasons: 1. Identification problems associated with the non-random nature of information dissemination

Information and accountability - Ferraz and Finan 2007 Our approach exploit an anti-corruption policy in Brazil Randomly selects municipalities for audits Public disseminates the findings to both the municipality and the general media Measures a mayor s corruptness Role of media in disseminating the information

Timing of the release of the audits - Ferraz and Finan 2007 Treatment: Pre-election audits 213 municipalities Control: Post-election audits 165 municipalities Funds audited Jan-01/Sept-03 Funds audited Jan-01/Dec-03 July-03 October-04 Elections June-05

Brazil s anti-corruption policy - Ferraz and Finan 2007 Brazil is one of the most decentralized countries in the world Local governments provide health, primary school education, infrastructure, sanitation Mostly paid for from federal block grants Concerned with extent of local corruption, in May 2003 the Federal government began to audit federal funds transferred to municipalities Each month the Controladoria Geral da Uniao (CGU), joint with the national lottery, draws 60 municipalities randomly across 5000 municipalities

Lottery - Ferraz and Finan 2007

Information and accountability - Ferraz and Finan 2007 10-20 auditors are immediately sent to examine the allocation of the federal funds Local governments are required to provide proof of purchase for any public good Talk to contractors and suppliers, members of the communities, program beneficiaries Goal: To produce evidence that could be used in a court of law After a week of inspections, a detail report describing all the irregularities is submitted to Brasilia A summary of the findings is posted on the internet and disclosed to the mass media

Documentation - Ferraz and Finan 2007

Key aspects of the program - Ferraz and Finan 2007 Municipalities are randomly selected Address identification issues Audit reports are publicly available Measure corruption and the information voters received Media was used to disseminated audit findings program to have a differential effect in municipalities with local media

Anecdotal evidence - Ferraz and Finan 2007 The conclusions from the CGU were used extensively in the political campaigns, by not only the opposition parties but those that received positive reports as well...the reports were decisive in several cities. In the small city of Vicosa, in Alagoas, where a lot of corruption was found, the mayor Flavis Flaubert (PL) was not re-elected. He lost by 200 votes to Pericles Vasconcelos (PSB), who during his campaign use pamphlets and large-screen tv in the citys downtown to divulge the report. Flaubert blames the CGU for his lost. (Diario de Para)

Anecdotal evidence - Ferraz and Finan 2007 Giovanni Brillantino from Itagimirim, in Bahia, who just before the elections claimed that We knew that the opposition party would exploit this information in the election. (Folha de Sao Paulo) In Taperoa, Bahia, where several incidences of fraud were uncovered, the local legislator Victor Meirelles Neto (PTB) claimed that the population was shocked when this information was revealed (Agencia Folha 12/06/2003).

Coding corruption - Ferraz and Finan 2007 Malhada de Pedras, BA (lottery 5): Fraud, diversion of funds, and use of fake receipts associated with the Fundef program: the auditors identified R$100,000 in fake receipts used by the municipal government to account for Fundef related expenditures. Based on interviews conducted by the auditors, all twelve firms that appear as product suppliers on the receipts claimed to have never done business with the local government. The auditors also discovered that more than R$610,000 of Fundef funds, were used irregularly between 2002 and 2003. The funds were used to pay wages of persons not associated with education.

Measuring corruption - Ferraz and Finan 2007 Based on the audit reports, we define corruption as any irregularity associated with: Fraud in procurement Diversion of public resources Over-invoicing Measure: Number of irregularities associated with corruption

Distribution of corrupt violations - Ferraz and Finan 2007 0.35 0.3 0.25 Pre-election Post-election 0.2 0.15 0.1 0.05 0 0 1 2 3 4 5 6 7 Number of corrupt violations

Data sources - Ferraz and Finan 2007 Corruption data Audit reports Election data Results for 2000 and 2004 mayor elections, mayor characteristics, measures of political competition, and electoral performance Municipal data 1999 municipal survey: general characteristics of the municipality including laws and regulations Economic data 2000 population census: measure of per capita income, Gini, demographic characteristics

Summary stats - Ferraz and Finan 2007 Post-election audit Pre-election audit Difference Standard error (1) (2) (3) (4) Panel A: Political characteristics Re-election rates for the 2004 elections 0.413 0.395 0.018 0.045 Re-election rates for the 2000 elections 0.423 0.443-0.020 0.040 2004 re-election rates, among those that ran 0.585 0.559 0.026 0.044 Ran for re-election in 2004 0.707 0.707-0.001 0.060 Number of parties in 2000 2.881 2.933-0.052 0.140 Margin of victory in 2000 0.142 0.131 0.012 0.019 Mayor's vote share in 2000 0.529 0.525 0.004 0.013 Panel B:Mayor characteristics: Age 47.5 48.0-0.5 0.9 Years of education 12.2 12.0 0.3 0.3 Male 0.96 0.94 0.02 0.03

Summary stats - Ferraz and Finan 2007 Post-election audit Pre-election audit Difference Standard error (1) (2) (3) (4) Panel C: Municipal characteristics: Population density (Persons/km) 0.57 0.73-0.16 0.33 Literacy rate (%) 0.81 0.80 0.01 0.03 Urban (%) 0.62 0.62 0.00 0.05 Log per capita income 4.72 4.66 0.06 0.15 Income inequality 0.55 0.54 0.00 0.01 Zoning laws 0.29 0.21 0.08 0.07 Economic Incentives 0.66 0.58 0.07 0.06 Paved roads 58.99 58.30 0.69 7.74 Size of public employment 42.45 42.76-0.32 1.53 Municipal guards 0.20 0.21-0.01 0.07 Small claims court 0.38 0.34 0.04 0.08 Judiciary district 0.59 0.56 0.03 0.07 Number of Newspapers 3.58 2.21 1.37 0.79 Municipalities with a radio stations 0.31 0.24 0.07 0.06 Number of radio stations, conditional on having one 1.37 1.29 0.08 0.11 Number of corrupt violations 1.952 1.584 0.369 0.357 Total resources audited ($R) 5,770,189 5,270,001 500,188 1,361,431

Sample of interest - Ferraz and Finan 2007 CGU audits municipalities with a population of less than 450,000 inhabitants (excludes 8 percent of Brazilian municipalities) Mayors that are eligible for re-election Excludes second-term mayors Focus is on mayors and not political parties Municipalities that were audited

Results - Ferraz and Finan 2007 All incumbent Only mayors that ran for reelection Change in Change in win Pr(re-election) Pr(re-election) Vote share Win margin vote share margin (1) (2) (3) (4) (5) (6) (7) Preelection Audit (1/0) -0.036-0.036-0.059-0.055-0.020-0.032-0.028 [0.053] [0.052] [0.065] [0.072] [0.027] [0.018]+ [0.027] Observations 373 373 263 263 263 263 263 R-squared 0.05 0.17 0.22 0.16 0.22 0.39 0.31 State fixed effects Yes Yes Yes Yes Yes Yes Yes Municipal characteristics No Yes Yes Yes Yes Yes Yes Mayor characteristics No Yes Yes Yes Yes Yes Yes

Results - Ferraz and Finan 2007 Reelection rates.2.3.4.5.6 0 1 2 3 4+ Number of Corrupt Violations

Results - Ferraz and Finan 2007 Reelection rates.2.3.4.5.6 0 1 2 3 4+ Number of Corrupt Violations Postelection Audit Preelection Audit

Results - Ferraz and Finan 2007 Semiparametric Corruption Corruption Linear Quadratic 5 4 (1) (2) (3) (4) (5) (6) Preelection audit 0.029 0.030 0.126 0.084 0.068 0.086 [0.083] [0.082] [0.101] [0.104] [0.087] [0.088] Preelection audit Number of corrupt violations -0.038-0.038-0.200-0.070-0.088 [0.035] [0.035] [0.090]* [0.041]+ [0.043]* Preelection audit Number of corrupt violations² 0.034 [0.017]* Preelection audit Corruption = 0 0.010 0.003 [0.156] [0.036] Preelection audit Corruption = 2-0.253 [0.148]+ Preelection audit Corruption = 3-0.321 [0.192]+ Preelection audit Corruption = 4+ -0.159 [0.168] Observations 373 373 373 373 362 351

Results - Ferraz and Finan 2007 Dependent variables: Pr(re-election) Margin of victory Vote share Change in vote share Corruption Corruption Corruption Corruption Full sample 5 Full sample 5 Full sample 5 Full sample 5 (1) (2) (4) (5) (7) (8) (10) (11) Preelection audit 0.045 0.072 0.037 0.053 0.078 0.104-0.014 0.006 [0.095] [0.099] [0.037] [0.039] [0.102] [0.106] [0.027] [0.027] Preelection audit Corrupt violations -0.06-0.086-0.034-0.049-0.078-0.104-0.01-0.029 [0.039] [0.046]+ [0.015]* [0.019]** [0.041]+ [0.048]* [0.012] [0.013]* Number of corrupt violations -0.016 0.001 0.011 0.019-0.002 0.014-0.001 0.01 [0.030] [0.036] [0.012] [0.014] [0.032] [0.039] [0.010] [0.010]

Results - Ferraz and Finan 2007 Treatment effect by corruption With one violation, the audit policy reduced re-election rates by 4.6 percentage points With 3 violations, the audit policy reduced re-election rates by 17.7 percentage points Interpretation Voters priors are that the average politician is corrupt Politicians are punished only when found to be extremely corruption Politicians that are not corrupt are reward at the polls

Threats to identification - Ferraz and Finan 2007 Municipalities were randomly selected!

Threats to identification - Ferraz and Finan 2007 Municipalities were randomly selected! But, this is does NOT guarantee that the audits themselves were not corrupt Mayors affiliated with the state or national party might have received more favorable audits Mayors engaged in tightly contested elections may have a higher incentive to bribe auditors Unlikely Corruption levels were balanced Interviews Robustness tests

Robustness - Ferraz and Finan 2007 Full sample Corruption 5 Corruption 4 Number of corrupt Dependent variable: violations Pr(re-election) (1) (2) (3) (4) (5) (6) (7) (8) Preelection audit -0.332-0.231 0.067 0.079 0.043 0.096 0.056 0.111 [0.261] [0.298] [0.121][0.132] [0.110] [0.125] [0.115] [0.129] Preelection audit Number of corrupt violations -0.21-0.180-0.08-0.071-0.09-0.088 [0.091]*[0.090]*[0.040]+[0.039]+ [0.043]*[0.041]* Preelection audit Number of corrupt violations² 0.035 0.031 [0.017]*[0.017]+ Preelection audit Member of the governor's coalition -0.155-0.155 0.056 0.055 0.06 0.059 0.1 0.103 [0.256] [0.388] [0.134][0.132] [0.136] [0.134] [0.140] [0.138] Preelection audit Margin of victory in 2000 elections -0.638-0.09-0.198-0.22 [0.868] [0.311] [0.316] [0.315] Preelection audit PT -0.004-0.034 0.269 0.299 0.28 0.3 0.186 0.208 [0.861] [0.864] [0.286][0.278] [0.290] [0.278] [0.280] [0.267] Preelection audit PMB 0.157 0.132 0.19 0.141 0.145 0.073 0.106 0.033 [0.389] [0.398] [0.130][0.128] [0.134] [0.130] [0.136] [0.134] Preelection audit PFL 0.064 0.052-0 -0.01-0.08-0.101-0.02-0.033 [0.445] [0.455] [0.153][0.147] [0.157] [0.149] [0.160] [0.151] Preelection audit PSDB -0.456-0.471-0.28-0.25-0.48-0.533-0.52-0.566 [0.989] [0.978] [0.262][0.295] [0.244]*[0.241]* [0.249]*[0.248]* Preelection audit PSB 0.093 0.073-0.33-0.44-0.32-0.46-0.29-0.422 [0.628] [0.637] [0.262][0.253]+[0.262] [0.253]+ [0.264] [0.255]+ Preelection audit PTB -0.549-0.562 0.324 0.272 0.295 0.232 0.274 0.216 [0.591] [0.594] [0.207][0.221] [0.212] [0.227] [0.216] [0.231] Observations 373 373 373 373 362 362 351 351 R-squared 0.35 0.35 0.19 0.28 0.21 0.27 0.22 0.28 F-test of the additional interaction terms (P-value) 0.97 0.97 0.20 0.39 0.09 0.08 0.15 0.13

Robustness - Ferraz and Finan 2007 Dependent variable: Vote share in 2000 Margin of victory in 2000 Corruption Corruption Corruption Corruption Full Sample 5 4 Full Sample 5 4 (1) (2) (3) (4) (5) (6) (7) (8) Preelection audit -0.001 0.007 0.000 0.001-0.011-0.003-0.012-0.011 [0.014] [0.016] [0.014] [0.015] [0.022] [0.027] [0.023] [0.024] Preelection audit Number of corrupt violations -0.003-0.015-0.003-0.004 0.000-0.014 0.001 0.000 [0.006] [0.015] [0.006] [0.007] [0.010] [0.024] [0.010] [0.012] Preelection audit Number of corrupt violations² 0.002 0.003 [0.003] [0.005] Number of corrupt violations 0.003 0.009 0.005 0.005-0.001 0.005 0.001 0.002 [0.005] [0.011] [0.005] [0.006] [0.007] [0.018] [0.008] [0.009] Number of corrupt violations² -0.001-0.001 [0.002] [0.003] Observations 369 369 358 347 369 369 358 347 R-squared 0.42 0.42 0.42 0.42 0.17 0.18 0.18 0.18 State fixed effects Yes Yes Yes Yes Yes Yes Yes Yes Municipal characteristics Yes Yes Yes Yes Yes Yes Yes Yes Mayor characteristics Yes Yes Yes Yes Yes Yes Yes Yes

Mechanism - Ferraz and Finan 2007 So far...convincing evidence of a reduce-form results The audits and their release reduced the likelihood of re-election among mayors found to be corrupt Mechanisms Information Campaign contributions Incumbents platform

An information story - Ferraz and Finan 2007 Reelection rates 0.2.4.6.8 0 1 2 3 4+ Number of corrupt violations Preelection Audit - No Radio Postelection Audit - No Radio Preelection Audit - Radio Postelection Audit - Radio

An information story - Ferraz and Finan 2007 Dependent variable: Pr(re-election) sample 5 interactions w/ radio (1) (2) (3) (4) (5) Preelection audit -0.059-0.033 0.296 0.208-0.954 [0.091] [0.096] [1.121] [1.247] [0.629] Number of corrupt violations -0.034-0.013-0.13-0.069-0.161 [0.029] [0.035] [0.224] [0.288] [0.194] Number of radio stations -0.131-0.150-0.216-0.253 [0.064]* [0.063]* [0.073]** [0.083]** Preelection audit Number of radio stations 0.229 0.271 0.356 0.449 [0.099]* [0.104]** [0.115]** [0.129]** Preelection audit Number of corrupt violations 0.007-0.018-0.236-0.412 0.458 [0.038] [0.044] [0.402] [0.430] [0.229]* Number of corrupt violations Number of radio stations 0.050 0.058 0.082 0.09 [0.026]+ [0.025]* [0.025]** [0.028]** Preelection audit Corrupt violations Radio stations -0.118-0.157-0.185-0.238 [0.045]** [0.067]* [0.051]** [0.064]** Proportion households with radio -0.834 [0.782] Preelection audit Households w/ radio 1.225 [0.752] Number of corrupt violations Households w/ radio 0.181 [0.243] Preelection audit Corrupt violations Households w/ radio -0.645 [0.292]* Full Corruption Demographic interactions Demographic and institutional Households

Other mechanims - Ferraz and Finan 2007 We do not find any evidence that the audits work through other mechanisms such as: Changes in incumbents platforms Type of candidate that the opposition party ran Campaign contribution

Conclusions - Ferraz and Finan 2007 Our findings lend strong support for the value of information in enhancing political accountability How this information is consequently interpreted depends on voters prior beliefs These results also highlight the influence media have on political outcomes, and particularly in helping to screen out bad politicians and promoting good politicians. (Besley and Burgess 2004; Stromberg 2004; Besley, Pande, and Rao 2005)

Economics 191. Economics of Corruption. Professor Fred Finan. March 13, 2012