CAF - Internal Control as State Capacity

C A F - W O R KI N G PA P E R # 2 0 2 4 / 0 8

F i r s t v e r s i o n : M a r c h , 2 0 2 4

T h i s v e r s i o n : O c t o b e r 1 7 , 2 0 2 4

Internal Control as State Capacity Guillermo Toral1 1Assistant Professor; School of Politics, Economics and Global Affairs; IE University. guillermo.toral@ie.edu While previous research has thoroughly examined the consequences of external control on subnational governments, less is known about the drivers and consequences of internal control. This paper presents a series of findings on Brazilian municipalities’ internal control systems, leveraging administrative data and quasi-experimental designs. First, I use cross-sectional data to document that there remains significant variation in the resources, mandates, and actions of internal control systems. Second, I use panel data and a causal event study design to show that legal reforms of internal control systems lead to an increase in hires of internal controllers. Third, I show these legal reforms have downstream consequences in the control of patronage, leading to a reduction in the size of the municipal workforce, and a higher incidence of civil service contracts. Fourth, I show that randomized federal anti-corruption audits have significant effects on municipal control systems. Together, these findings highlight the relevance of local internal control institutions and call for further research on the causes and consequences of their strength. K E Y W O R D S governance, state capacity, corruption, local governments, Brazil Small sections of text that are less than two paragraphs may be quoted without explicit permission as long as this document is acknowledged. Findings, interpretations and conclusions expressed in this publication are

the sole responsibility of its author(s) and cannot be, in any way, attributed to CAF, its Executive Directors or the countries they represent. CAF does not guarantee the accuracy of the data included in this publication and is not, in any way, responsible for any consequences resulting from its use. ©2024 Corporación Andina de FomentoC A F - D O C U M E NT O D E T R AB A J O # 2 0 2 4 / 0 8

E s t a v e r s i ó n : 1 7 d e o c t u b r e d e 2 0 2 4

Control Interno como Capacidad Estatal Guillermo Toral1 1Assistant Professor; School of Politics, Economics and Global Affairs; IE University. guillermo.toral@ie.edu Aunque investigaciones previas han analizado en detalle las consecuencias del control externo sobre gobiernos subnacionales, sabemos menos sobre las causas y las consecuencias de los sistemas de control interno. Este artículo presenta una serie de resultados sobre los sistemas de control interno en los municipios brasileños, aprovechando datos administrativos y diseños cuasi-experimentales. Primero, utilizo datos transversales para documentar que existe una variación significativa en los recursos, mandatos y acciones de los sistemas de control interno. Segundo, utilizo datos de panel y un diseño de estudio de eventos causal para mostrar que las reformas legales de los sistemas de control interno llevan a un aumento en las contrataciones de controladores. Tercero, demuestro que estas reformas legales tienen consecuencias a largo plazo en el control del patronazgo,

conduciendo a una reducción en el tamaño de la fuerza laboral municipal y a una mayor incidencia de contratos por oposición. Cuarto, muestro que las auditorías federales anti-corrupción aleatorias tienen efectos significativos en los sistemas de control municipal. En conjunto, estos hallazgos destacan la relevancia de las instituciones locales de control interno e invitan a futuras investigaciones sobre sus causas y consecuencias. K E Y W O R D S gobernanza, capacidad estatal, corrupción, gobiernos locales, Brasil Pequeñas secciones del texto, menores a dos párrafos, pueden ser citadas sin autorización explícita siempre que se cite el presente documento. Los resultados, interpretaciones y conclusiones expresados en esta publicación son de exclusiva responsabilidad de su(s) autor(es), y de ninguna manera pueden ser atribuidos a CAF, a los miembros de su Directorio Ejecutivo o a los países que ellos representan. CAF no garantiza la exactitud de los datos incluidos en esta publicación y no se hace responsable en ningún aspecto de las consecuencias que resulten de su utilización. ©2024 Corporación Andina de FomentoTORAL 2 1 | INTRODUCTION Decentralization efforts across the world face the twin challenge of insufficient capacity and rent-seeking among subnational governments. This challenge is particularly acute in the middle income countries of Latin America, which have significant variation in capacity and serious corruption problems at the local level. National governments have at their disposal a number of policy tools to address these challenges and ameliorate compliance and policy implementation by subnational units, including audits, fiscal rules, procurement rules, and performance-based financing. Recent studies have shown positive effects of these external

control institutions, including audits (Avis et al., 2018; Zamboni and Litschig, 2018), fiscal rules (Grembi et al., 2016) and procurement rules (Lewis-Faupel et al., 2016; Coviello and Mariniello, 2014). In contrast to the accumulation of experimental and quasi-experimental evidence in favor of these external control mechanisms, we know very little about the effects of internal control institutions. This gap is worth addressing because internal control is potentially a more direct and thus more effective way to limit rent seeking and improve the quality of public policies, for at least three reasons. First, internal control institutions intervene “at the source”, identifying and addressing issues as they emerge, before malfeasance or ill-designed programs can hurt citizen welfare. Second, internal control institutions have the potential to intervene on a wider range of dimensions than external control ones. Third, internal control institutions can acquire better and more frequent information about public management, and are potentially better positioned to exercise fire-alarm accountability. Despite the potential gains of internal control, the evidence base about their effectiveness is thin, particularly in middle-income contexts. Two recent studies have examined the role of internal auditors in the control of corruption in Italy (Vannutelli, 2024) and Spain (Fernández-Vázquez, 2018), but the treatment of those studies (nationally-deployed auditors) is not representative, and the context of high-income, established democracies may make their findings hard to translate to the Latin America region. One important reason to explain the scarcity of (causal) evidence on internal control institutions at the local level is that they rarely vary at the local level and, when they do, they do so endogenously. The Brazilian context provides a privileged setting in which to study the strength of internal control institutions, for a variety of reasons. First, Brazil is a large federal democracy

with significant variation in the strength and insulation of internal controllers at the local level. Second, there are rich administrative datasets available in Brazil about internal control and other dimensions of local governance. Third, there is variation across time and space in reforms of internal control systems. 2 | INSTITUTIONAL SETTING Brazil has 5,570 municipalities which are distributed across 26 states and a federal district. Municipal governments are responsible for delivering basic services in policy areas including healthcare, education, and social assistance. Together, these three areas of social policy make up over 55% of municipal government spending in the average municipality. As a result of these responsibilities in service delivery, municipal workforces are typically large. On average, municipal governments hired 4.9% of all local residents and 38.2% of thoseTORAL 3 employed in the formal labor market in 2016.1 Municipal jobs are typically highly desirable because of their stability (when in the civil service), because municipal workers enjoy a wage premium relative to the private sector (Colonnelli et al., 2020, 3090), similarly to other Global South contexts (Finan et al., 2017), and because employment opportunities in the typical municipality (which is small and relatively poor) are scarce.2 Mayors (who are elected by majority rule every four years and can only be re-elected once) and the secretaries they appoint have some discretion over the hiring and firing of bureaucrats. Such discretion differs significantly between the civil service and other hiring modes with fewer employment protections. The Constitution mandates all permanent staffing needs to be filled with civil service contracts, 3 which have tenure for life after a short probationary period.4 Approximately a third of municipal employees are hired on temporary contracts,5 which can legally be used to hire political appointees for management or leadership positions, or to fill short-term or urgent staffing needs.6 In practice, temporary

hiring is often used where the civil service should prevail, often as a vehicle for political appointments.7 The abuse of hiring outside the civil service has been shown to help politicians reward political supporters after getting to office (Colonnelli et al., 2020), build legislative coalitions (Mignozzetti et al., 2022), and mobilize supporters ahead of a re-election campaign (Toral, 2023). It has also been shown to be detrimental for citizen welfare; bureaucrats selected under temporary contracts often have worse qualifications and experience (Colonnelli et al., 2020; Toral, 2024a), and their turnover, induced by political turnover, depresses the quality of public services (Akhtari et al., 2022; Toral, 2024b). Given the centrality of public employment in local politics and the strong legal basis for mandating hiring in the civil service, prosecutors often seek to constrain local officials’ use of temporary contracts. For politicians, the consequences of being charged for violating public employment rules are potentially very severe. If found guilty, they are subject to penalties, including the loss of their post, having their political rights suspended, substantive fines, and even imprisonment. 8 In practice, it is not uncommon for politicians to be charged for corruption. Lambais and Sigstad (2023) estimate that about 7.7% of mayoral election winners or runner-ups are involved in a court case accused of corruption charges.9 Bento et al. (2021) document 1,716 judicial cases involving mayors and former mayors between 1992 and 2016 in the southern state of Rio Grande do Sul, which has 497 municipalities. In the state of São Paulo, 40% of municipalities had their mayors or former mayors convicted of corruption charges in just one year; 83% of them had to pay fines, and 68% of them had

1Figures are from the administrative labor data described below. 2According to the 2010 census, the median municipality had fewer than 12,000 inhabitants and a per capita income of less than 500 Brazilian reais (about USD284 at the exchange rate at the time). 3Article 37.II of the Constitution. 4Tenured employees can only be fired in extraordinary circumstances, e.g., after a corruption conviction. 5I use the term temporary contracts to refer to all non-civil service contracts. These contracts can use a variety of labor regimes, all of which lack tenure. 6Article 37.IX of the Constitution. 7In a face-to-face survey of bureaucrats I did in Rio Grande do Norte in 2018, 58% of respondents expressed the highest level of agreement with the statement “political connections influence the hiring of temporary bureaucrats” (Toral, 2024a). 8The legal penalties for breaches of public employment laws are contained in the Constitution, the Administrative Impropriety Law, the Electoral Law, and the Penal Code, among other legal instruments. 9The authors also find that politicians who lose their reelection bid are 65 percent more likely to receive a penalty than those who are reelected.TORAL 4 their political rights suspended (Anuário da Justiça, 2016).10 Local governments and politicians are also subjet to the control of other horizontal accountability institutions. Key among them is the Federal Comptroller’s Office (CGU, Controladoria-Geral da União), which every year audits the accounts of a subset of municipal governments. Between 2003 and 2015, the CGU targeted its audits of municipal governments through public, randomized lotteries. Municipalities selected through the lottery are visited by a team of federal auditors to review local spending of federal transfers. The CGU releases the results of the audits to the media and to other accountability actors like the federal

prosecutor’s office, the audit court, and the police, as well as to the municipal legislative chamber. Details of these randomized audits are described by Avis et al. (2018). These randomized audits have been found to decrease corruption in the use of federal funds (Avis et al., 2018; Zamboni and Litschig, 2018), limit electoral clientelism (Bobonis et al., 2024), increase the prevalence of civil service hiring (Lauletta et al., 2022), improve service delivery (Funk and Owen, 2020), and reduce child mortality (Ramos et al., 2022). 3 | DESCRIBING VARIATION IN MUNICIPAL INTERNAL CONTROL UNITS I use data from the federal 2019 Munic survey to describe cross-sectional variation in municipal internal control units. Munic is a yearly survey of municipal governments done by the federal government, with separate thematic focuses each year. The 2019 survey included a module about internal control, with a series of rich questions about the resources, the mandates, and the actions taken by internal control units / structures in the municipality. All but 2 municipalities submitted data to IBGE. Descriptive data summarized in Figure 1 contains a number of insights. First, high levels of legal formalization coexist with relatively low levels of professionalization of internal control units. For example, 94.4% of the municipalities have formal internal control units and, in 88% of the cases, those units are formalized by law. On the other hand, 37.9% of the municipalities have a political appointee leading the internal control unit or system, and only 25% have a specific civil service track for control staff. When it comes to the formal mandates of internal control systems, there seems to be high levels of isomorphism. For example, an overwhelming majority of municipal internal

control systems cover municipal spending (90.6%), municipal revenues (87%), municipal budget execution (81.3%), and municipal personnel (75%). Yet few of them have functions to make them effective control institutions, including auditing functions (45.2%) and anticorruption functions (33.2%). Finally, when asked about specific control actions they have taken in the last 2 years, we find much less promising responses. While around two thirds of municipal internal control systems report having conducted some form of control (be it ex ante, ex post, or both), accounts of specific accountability actions are much less common. For example, only 18.6% of internal control units have reported irregularities to the state audit court in the last 10The number of cases where politicians are investigated for corruption is of of course larger. I scraped all news reports mentioning the word “mayor” in the website of the largest state prosecutor’s office (that of São Paulo) and found 483 reports in the 10-year period from 2013 to 2022. Of those, 62% refer to cases of mismanagement and corruption.TORAL 5 carreira_concurso_ultimos_2a carreira_concursada head_law_econ_mgt_acct head_no_highered head_appointed head_concurso_geral head_concurso_especifico has_specific_staff subordinate in_mayors_office own_secretariat area_without_law area_by_law area_formalized Resources Mean 0.0 0.2 0.4 0.6 0.8 1.0 carreira_concurso_ultimos_2a carreira_concursada head_law_econ_mgt_acct head_no_highered head_appointed head_concurso_geral head_concurso_especifico has_specific_staff subordinate in_mayors_office own_secretariat

area_without_law area_by_law area_formalized 0.0 0.2 0.4 0.6 0.8 1.0 functions_include_anticorruption functions_include_ouvidoria functions_include_corregedoria functions_include_controladoria functions_include_auditoria covers_access_to_information_law covers_transparency covers_corruption_prevention covers_accounting covers_bidding_oversight covers_targets_compliance covers_budget_execution covers_assets covers_personnel covers_spending covers_revenue Mandates Mean 0.0 0.2 0.4 0.6 0.8 1.0 functions_include_anticorruption functions_include_ouvidoria functions_include_corregedoria functions_include_controladoria functions_include_auditoria covers_access_to_information_law covers_transparency covers_corruption_prevention covers_accounting covers_bidding_oversight covers_targets_compliance covers_budget_execution covers_assets covers_personnel covers_spending covers_revenue 0.0 0.2 0.4 0.6 0.8 1.0 resulted_last_1y_ressarcimento_erario resulted_last_1y_personnel_dismissals resulted_last_1y_personnel_sanctions resulted_last_1y_personnel_fine did_last_2y_tomada_de_contas_especial did_last_2y_processo_administrativo_responsabilizacao did_last_2y_processo_administrativo_disciplinar reported_irregularities_last_2y_mp reported_irregularities_last_2y_tce actions_last_2y_controle_posterior actions_last_2y_controle_concomitante actions_last_2y_controle_previo Actions Mean 0.0 0.2 0.4 0.6 0.8 1.0

resulted_last_1y_ressarcimento_erario resulted_last_1y_personnel_dismissals resulted_last_1y_personnel_sanctions resulted_last_1y_personnel_fine did_last_2y_tomada_de_contas_especial did_last_2y_processo_administrativo_responsabilizacao did_last_2y_processo_administrativo_disciplinar reported_irregularities_last_2y_mp reported_irregularities_last_2y_tce actions_last_2y_controle_posterior actions_last_2y_controle_concomitante actions_last_2y_controle_previo 0.0 0.2 0.4 0.6 0.8 1.0 F I G U R E 1 Descriptive Prevalence of Resources, Mandates, and Actions among Municipal Internal Control Units in 2019. Descriptive data are from IBGE’s 2019 Munic survey, which has data for 5,566 municipalities. 2 years.11 Less than 25% report having started an administrative disciplinary procedure in the past 2 years. And if we look for evidence of actual sanctions against local government officials, we find that internal control systems report having imposed sanctions, dismissals, or fines in the past year only in 5.2%, 5% and 2% of the municipalities, respectively. Based on these descriptive data from the Munic survey of municipal governments, it seems that while formally most municipalities have formally consolidated internal control systems, they vary considerably in their degree of political independence and professionalization, their anti-corruption attributes, and their actual incidence over within-government accountability. In any case, these data are just descriptive. To better assess the causal impacts of internal control systems, I turn to a causal event study design exploiting variation across municipalities and years in legal reforms strengthening internal control systems.

4 | THE CONSEQUENCES OF INTERNAL CONTROL LEGAL REFORM To measure the impact of internal control legal reform on local governance, I leverage variation in the passing of such reforms across years and municipalities. Using a novel, imputation-based causal event study approach (Liu et al., 2024) that bypasses some of the well-known issues with two-way fixed-effects specifications (Baker et al., 2022; Roth et al., 2023),12 I identify the causal effect of internal control legal reform on two sets of outcomes. First, the hiring of control staff (auditors and accountants) and the prevalence of civil service contracts among them. Second, as proxies for good governance, I examine whether legal reform reduces the size of the municipal workforce (relative to the local population) and increases the prevalence of the civil service among municipal employees. 11The figure is even lower for reports to the prosecutor’s office, 7.9%. 12Other recently developed approaches to causal inference in panel settings are inappropriate in this setting because they assume a balanced panel (Imai et al., 2023; De Chaisemartin and d’Haultfoeuille, 2020). The Liu et al. (2024) method is very similar to the imputation event study method developed in parallel by Borusyak et al. (2024). Other methods assume staggered adoption of treatment without reversals, and are therefore applicable in this setting (Callaway and Sant’Anna, 2021; Sun and Abraham, 2021). Chiu et al. (2023) replicate 38 panel data studies in top political science journals and show that, in practice, these new estimators often lead to estimates that are similar to each other and to the traditional two-way fixed-effects specification.TORAL 6 The findings, which leverage administrative data for all municipalities in Brazil, suggest that legal reform does cause an increase in the number and insulation of controllers, as well

as a reduction of the municipal workforce at the expense of temporary employees. 4.1 | Identification The Liu et al. (2024) imputation-based estimator of treatment effects is unbiased and consistent13 under a strict exogeneity assumption.14 This assumption involves the absence of time-varying confounders, anticipation effects, and carryover effects (Imai and Kim, 2019).15 A key advantage of the imputation method of Liu et al. (2024) is that we can test for the validity of these identifying assumptions. 4.2 | Estimation The Liu et al. (2024) estimator follows an imputation procedure. Using only untreated observations (i.e., municipality-year observations without an internal control law), municipality and year fixed effects are fitted and then used to impute the counterfactual potential outcomes for treated units under control. The individual treatment effect for each treated observation (ITE) is estimated by taking the difference between its observed outcome and its imputed counterfactual outcome: ˆτit = Yit − ˆαi − ˆβt.16 ITEs are then aggregated to obtain dynamic treatment effects for each period (DTEs) and overall average treatment effects on the treated (ATT). 4.3 | Inference For inference, the Liu et al. (2024) procedure uses non-parametric block bootstrap clustered at the unit level for inference. In this case, that implies clustering at the municipality level, which is appropriate as municipalities are where treatment is assigned (Abadie et al., 2023). The block bootstrap procedure essentially entails resampling many times, 17 with replacement, an equal number of municipalities from the original sample and re-doing the imputation and estimation of the ATT and DTEs with each bootstrap. Standard errors and confidence intervals are obtained from the standard deviation and percentiles of the

resulting distributions of block-bootstrapped estimates. 4.4 | Data The treatment in this design is defined by the year at which a municipality approved its (latest) law of internal control, as reported in IBGE’s 2019 Munic survey. Munic is a yearly survey of municipal governments done by the federal government, with separate thematic focuses each year. The 2019 survey included a module about internal control. The treatment 13Assuming spherical errors, the imputation method is also the most efficient among all linear and unbiased estimators (Borusyak et al., 2024). 14The strict exogeneity assumption implies the parallel trends assumption. 15The assumption of no anticipation effects would be violated if past potential outcomes directly affected current treatment. The assumption of no carryover effects would be violated if past treatment directly affected current potential outcomes. 16The imputation method can be used with more complex specifications (e.g., with an interaction between unit and period fixed effects) and with control variables. I follow the guidance of Liu et al. (2024) and use the simple two-way fixed effects specification since results pass diagnostic tests. 17I use 1,000 block bootstraps.TORAL 7 history is reported below. I count a municipality as treated the year it passes its internal control law and thereafter. 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Y ear No reform Internal control law passed Internal control law reforms F I G U R E 2 Passing of internal control laws across municipalities To obtain time-varying measures of the strength of internal control units I use detailed administrative data on municipal employment. I use the Ministry of the Economy’s Annual Social Information Report (RAIS, Relação Anual de Informações Sociais). All formal employers

–including municipal governments– are legally obliged to report all their contracts to the Ministry of the Economy every year. RAIS therefore contains data on the universe of municipal employees, including contract type, start and end dates, salary, reason for termination, and professional category, among other variables. I count the number of municipal employees in the categories "auditor" and "accountant", as identified by the Brazilian Classification of Occupations.18 I also use the share of such employees that has a civil service contract.19 To measure impacts on broader governance outcomes, I look at the number of municipal employees per 100 residents, and at the share of municipal employees who have civil service contracts. 4.5 | Results Causal event study estimates reported below suggest that the passing of a municipal law focused on internal control causes large and significant increases in the number of control employees, and in the prevalence of civil service contracts among them (see Table 1 and Figure 3). In turn, results suggest that these legal reforms cause a small decrease in the size of the municipal workforce (relative to the local population) and a marginally insignificant increase in the prevalence of civil service contracts, as shown in Table 2 and Figure 4. 18CBO codes 252205 and 252210 correspond to auditor and contador, respectively. Because this classification was not implemented until 2003, analyses of these outcomes disregard data from before that date. 19I code contracts in the regime jurídico único de servidores públicos as civil service, and all others (which use a variety of legal regimes) as temporary. Unfortunately, RAIS does not allow a reliable identification of politically appointed employees (e.g., cargo comissionado, função de confiança).TORAL 8 TA B L E 1 Average Treatment Effect on the Treated Estimates of Internal Control Legal Reform on the Number of Control Staff for Every 100 Employees Control employees Control employees in the civil service

dATT 0.040 ∗∗∗ 0.029∗∗∗

(0.009) (0.006) [0.023, 0.057] [0.016, 0.042] Mean outcome under control 0.116 0.078 Columns 1 and 2 report, respectively, the ATT estimates for the number of control employees for every 100 employees – with any contract and with civil service contract, respectively. ∗∗∗p<0.001. The municipality-clustered standard errors are in brackets, and the 95% confidence intervals, in square brackets, below them. Both are estimated through the block bootstrap. ATT estimates come from aggregation of dynamic treatment effect estimates reported in Figure 3. This reduction in hiring outside the civil service is meaningful because temporary hires in this context have been shown to help politicians reward political supporters after getting to office (Colonnelli et al., 2020), build legislative coalitions (Mignozzetti et al., 2022), and mobilize supporters ahead of a re-election campaign (Toral, 2023). Accordingly, Lauletta et al. (2022) have shown that anti-corruption audits constrain the size of the municipal workforce, and increase the prevalence of civil service hires. F test p−value: 0.165 3522 −0.10 −0.05 0.00 0.05 0.10 −10 −5 0 5 10 Y ears since internal control law passed Control staff for every 100 employees F test p−value: 0.327 3522 −0.05 0.00 0.05 0.10 −10 −5 0 5 10 Y ears since internal control law passed Control civil servants for every 100 employees

F I G U R E 3 Dynamic Treatment Effect Estimates of Internal Control Legal Reform on the Number of Control Staff for Every 100 Employees. Each subplot presents the estimated dynamic treatment effects (DTE) for switcher municipalities in each period (indexed relative to the year of internal control legal reform) as a dot, and its block-bootstrapped 95% confidence interval as a vertical line. The bar plot at the bottom represents the number of treated units in each period. The F test p-value reported in the upper left corner of each plot corresponds to the test of no pre-trend.TORAL 9 TA B L E 2 Average Treatment Effect on the Treated Estimates of Internal Control Legal Reform on the Number of Employees per 100 Residents and the Share of Employees in the Civil Service Employees per 100 residents Share employees in the civil service dATT -0.072 ∗ 0.011· (0.031) (0.007) [-0.134, -0.010] [-0.002, 0.023] Mean outcome under control 4.571 0.667 Columns 1 and 2 report the ATT estimates for the number of overall employees per 100 residents and the share of overall employees who hold civil service contracts, respectively. · p<0.10; ∗p<0.05. The municipality-clustered standard errors are in brackets, and the 95% confidence intervals, in square brackets, below them. Both are estimated through the block bootstrap. ATT estimates come from aggregation of dynamic treatment effect estimates reported in Figure 4. F test p−value: 0.296 4004 −0.3 −0.2 −0.1 0.0 −10 −5 0 5 10 Y ears since internal control law passed Employees for every 100 residents F test p−value: 0.175

4004 −0.04 −0.02 0.00 0.02 −10 −5 0 5 10 Y ears since internal control la