Law enforcement, municipal budgets and spillover effects: Evidence from a quasi-experiment in Italy

Accepted Manuscript

Law enforcement, municipal budgets and spillover effects: Evidence from a quasi-experiment in Italy Sergio Galletta PII: DOI: Reference:

S0094-1190(17)30057-8 10.1016/j.jue.2017.06.005 YJUEC 3090

To appear in:

Journal of Urban Economics

Received date: Revised date: Accepted date:

6 April 2016 14 June 2017 25 June 2017

Please cite this article as: Sergio Galletta, Law enforcement, municipal budgets and spillover effects: Evidence from a quasi-experiment in Italy, Journal of Urban Economics (2017), doi: 10.1016/j.jue.2017.06.005

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT

Law enforcement, municipal budgets and spillover effects: Evidence from a quasi-experiment in Italy✩ Sergio Gallettaa,b a

Institute of Economics (IdEP), University of Lugano (USI) Barcelona Economic Institute (IEB), University of Barcelona (UB)

CR IP T

b

Abstract

In this paper, I empirically investigate the presence of spillover effects resulting from the strengthening of law enforcement against corruption in local governments. Specifically, I take

AN US

advantage of an Italian law that empowers the central government to replace democratically elected municipal officials who are potentially connected with the mafia with a commission of non-elected administrators. Fixed-effects model estimates that focus on a sample of municipalities from three Italian regions (Campania, Calabria and Sicilia) for the period 1998 to 2013 show that in municipalities where the city council is dismissed because of the presence of mafia-connected officials, there is a reduction in public investments in neighboring

M

municipalities. Additional empirical evidence suggests that this reduction may be because law enforcement spillovers reduce misconduct in neighboring municipalities.

ED

JEL Classification: D73, E62, H72, K42

AC

CE

PT

Keywords: Horizontal interaction, Italy, Mafia, Corruption

✩

I am grateful for the useful comments and suggestions provided by Massimo Bordignon, Gianmarco Daniele, Ruben Durante, Mario Jametti, Stephan Litschig, Paolo Pinotti, Hannes Mueller, Albert Solé-Ollé, Raphaël Parchet, Pilar Sorribas-Navarro and Juan Carlos Suarez Serrato as well as the editor (Stuart Rosenthal) and two anonymous referees. I have also benefited from comments by participants at the Sinergia Workshop (Lugano), CESifo Workshop (Dresden), SIEP (Ferrara), IIPF (Lake Tahoe) and seminar at IAE/UAB Barcelona. Financial support from the Swiss National Science Foundation (grants Early Postdoc.Mobility - 15860 and Advanced Postdoc.Mobility - 167635) is gratefully acknowledged. Email address: [email protected] (Sergio Galletta)

1

ACCEPTED MANUSCRIPT

1. Introduction To be successful, modern economies must efficiently allocate public spending. The ineffective use of public funds could be explained by different factors such as the presence of mediocre institutions or the incompetency of the political class. Alternatively, the poor governance of public resources is significantly influenced by political corruption to the detriment of the econ-

CR IP T

omy. (Mauro, 1995, 1998; Del Monte and Papagni, 2001; Arin et al., 2011; Hessami, 2014). This negative effect is potentially amplified in countries where sub-national governments have considerable autonomy, as a decentralized setting could exacerbate corruption by increasing officials’ rent-seeking opportunities (Brueckner, 2000; Bardhan and Mookherjee, 2006; Fan et al., 2009). Against this backdrop, several countries have implemented reforms aimed at increasing the monitoring of local governments and punishing dishonest public officials in order

AN US

to reduce their misconduct and increase the effectiveness of public spending. Recent empirical studies provide convincing evidence of the beneficial direct effects of anti-corruption policies.1 However, estimating their overall effectiveness is difficult, because few studies have examines their potential spillover effects.2

This paper therefore assesses the presence of spatial spillover effects that might be triggered by anti-corruption policies. The analysis focuses on Italy, and links exogenous breaks in

M

the governance of municipalities whose representatives’ decisions are affected by mafia-type organized crime groups on the amount of local public investment in both intervened and

ED

neighboring jurisdictions.3

For decades, Italian organized crime has drained public resources by interfering in both

1

PT

the procuring and supplying ends of the allocation of public contracts (Caneppele and MarFor instance, Olken (2007) finds that top-down monitoring reduced missing expenditures in road construc-

tion in Indonesian local jurisdictions. Similarly, Litschig and Zamboni (2011) show that a higher probability

CE

of external audit decreases the occurrence of mismanagement of local procurement in Brazil. In a similar vein, Di Tella and Schargrodsky (2003) report a decrease in the input prices paid by hospitals in the city of Buenos Aires after a monitoring policy was introduced following a crackdown on corruption. 2 For example, Silva (2010) and Avis et al. (2016) account for the potential presence of spillovers by studying

AC

the effect of a random auditing policy on neighbors’ corruption in Brazil, Acemoglu et al. (2015) study the spillover effects of local state capacity in Colombia, and Durante and Gutierrez (2015) investigate how interjurisdictional cooperation among Mexican municipalities can prevent crime. 3 In Italy, political corruption is mainly driven by mafia-type organized crime groups. Therefore it resembles conditions that are more common to emerging democracies (e.g., Colombia and Mexico) than to Western European countries. It is worth noting that the results from the Eurobarometer Survey (2006) (issue 245) show that more than 50% of those interviewed believe that most corruption is caused by organized crime. Importantly, Italy has the highest percentage of people who agree with this statement (71%).

2

ACCEPTED MANUSCRIPT

tocchia, 2014).4 In 1991, in an attempt to control corruption in public administration, the Italian Parliament approved a law (D.L. 31/05/1991, n. 164) that gives the central government the power to dismiss the city council of a municipality if local officials are found to have a potential relationship with the mafia.5 During the dismissal period (usually 18 months, with a maximum of 24 months), a commission of three technocrats governs the municipality in an attempt to enhance law enforcement. The new commission reviews the dismissed council’s

CR IP T

budget and decisions, which often involves reassigning contracts with mafia-infiltrated firms to other businesses or, when possible, directly providing services through various municipal departments.6 These interventions have a significant effect on the budget. I show that investments are reduced, on average by nearly 15%, during the three years following a council dismissal. These results are explained by an increase in the efficiency of public spending, which in turn improves the quality of the services provided (Ministry of the Interior, 2008, 2015).7

AN US

In order to test whether this anti-corruption policy also triggers spillover effects, I use a fixed-effects model that evaluates the consequences of a municipality’s city council dismissal on the investment spending of neighboring municipalities. To reduce concerns about potential selection bias due to the non-random intervention of the central government, my estimates focus on the sample of treated municipalities (i.e., those with at least one neighbor that

M

experienced a city council dismissal during the study period). This allows me to estimate the spillover effects by exploiting both the time-series and cross-sectional variation in the sample. I use data from municipalities in the regions of Campania, Calabria and Sicilia (which have

ED

the strongest historical presence of mafia-type organized crime) during the period 1998 to 2013. More than 90% of the council dismissals that took place within the study period were in municipalities located in these three regions. Initial estimates suggest that spillovers exist: the

PT

compulsory administration of a municipality is associated with an average 6% annual decrease

4

CE

in public investment in its neighbors for three years after a city council dismissal. To provide For instance, the mafia bribes or intimidates local administrators to assign contracts to firms owned by

individuals with direct or indirect ties to the criminal organization (Della Porta and Vannucci, 1999). Estimates

AC

from the Corte dei Conti, the Italian Court of Auditors, suggest that nearly 40% of the cost of large-scale public investments is attributable to corruption. 5 The 1991 law was largely integrated into a 1993 law (D.L. 20/12/1993, n. 529), which details the commission’s powers. In 2000 (D.Lgs. 18/08/2000, n. 267) it was merged with existing laws regulating the activities of local jurisdictions (Testo Unico Eenti Locali). Additional changes occurred in 2009 (L. 15/07/2009, n. 94). 6 Di Cataldo and Mastrorocco (2016) provide empirical evidence that collusion between organized crime and local politicians affects the use of public resources. 7 The municipality of San Luca (in the region of Calabria) is a case in point. At the beginning of 2017, citizens asked the ministry not to hold new elections for the city council, and to leave the commissioner in power because of his “good administration”.

3

ACCEPTED MANUSCRIPT

a causal interpretation of the results, the timing of a council dismissal in the neighboring area must be random, conditional on controls. I test the robustness of the identifying assumption i) by showing that unobservables need to be relatively large – compared to observables – to invalidate my findings (Altonji et al., 2005; Oster, 2015), and ii) by performing a placebo test that suggests the absence of anticipatory effects. The main results come from a reduced-form approach that yields no clear evidence about

CR IP T

which mechanism is driving these spillover effects. However, there are several reasons why an intervention directed at fighting political corruption in a specific area or jurisdiction could affect its neighbors’ public budgets. For instance, the effective control of illegal activities in the administration of a locality can yield “beneficial” spillover effects because it serves as a credible threat that councils in neighboring areas could also be dismissed (Sah, 1991; Rincke and Traxler, 2011). In other words, individuals committing crimes in the surrounding area

AN US

may become more concerned about the risk of being detected themselves, and may reduce their exploitation of public contracts. Alternatively, similar to what happens with other criminal operations, dishonest conduct in public administration may simply relocate to other jurisdictions, and produce an increase in public spending there (Knight, 2013; Dell, 2015). These two kinds of spillover effects are more likely to arise where the infiltration of criminal organizations

M

in local jurisdictions is endemic rather than an isolated event, as is the case in the south of Italy. In addition, anti-corruption policies that have an important effect on the public budget can favor either positive or negative fiscal/economic spillovers that might eventually affect

ED

local investment regardless of the existence of the previous two channels (Besley and Case, 1995; Case et al., 1993; Peltzman, 2016). For instance, a strong reduction in public spending in a given municipality could trigger changes in the economic conditions of a wider area, and

PT

eventually affect neighboring public investment. Alternatively, yardstick competition models relate a government’s policy decisions to the performance of its neighbors (Besley and Case,

CE

1995). In this context, politicians might have incentives to mimic the behavior of a new commission established in the surrounding area to avoid a reduction in electoral support. In order to disentangle the channels driving my estimates, I apply two different approaches

AC

that suggest law enforcement spillovers are the main determinants of the baseline results. First, I assess whether the main effect depends heterogeneously on a set of municipal characteristics that may reveal whether fiscal/economic spillovers are likely to exist (e.g., the extent of economic ties between neighbors, relative size of the municipality experiencing the council dismissal and the electoral cycle). While none of these characteristics seems to matter, the effect becomes weaker for more isolated municipalities. Importantly, I find that mafiainfiltrated municipalities are mainly responsible for the estimated reduction in investments.

4

ACCEPTED MANUSCRIPT

This evidence suggests that compulsory administration causes law enforcement spillovers, as the public administrations that are more likely to be corrupted are also the ones that experience a more significant budget reduction, while the others, if anything, are marginally positively affected. As a second strategy, I apply a falsification test focusing on city councils that are dismissed for other reasons (i.e., not because of links with the mafia). Although these types of dismissals also have an important direct negative effect on public investments,

CR IP T

they do not result in spillovers; this result is independent of whether the municipality is mafia ridden. In other words, the presence of spillover effects crucially depends on the reason why the focal municipality experiences a reduction in public spending, reinforcing the evidence on the marginal role played by fiscal or economic spillovers in the case of study.

This paper contributes to two separate strands of the literature. First, it relates to the literature empirically assessing the effects of anti-corruption policies on public decisions (Di Tella

AN US

and Schargrodsky, 2003; Olken, 2007; Litschig and Zamboni, 2011). My results complement this literature by suggesting that anti-corruption policies can induce budget reductions that spread beyond the geographic limits of the intervened jurisdiction, although they may be minimally offset by rent relocation. Second, the current research supports the literature on spatial spillovers by adding evidence regarding the significant interdependence of policies and deci-

M

sions from different jurisdictions (Case et al., 1993; Besley and Case, 1995; Brueckner, 2003; Bordignon et al., 2003; Baicker, 2005; Solé-Ollé, 2006; Isen, 2014). Unlike most previous studies, the paper uses a quasi-experimental framework that in principle produces causal estimates

ED

and avoids some of the limitations of standard spatial econometrics methods. The remainder of the paper is organized as follows. Section 2 illustrates the institutional framework, and Section 3 presents the data used in the empirical estimation. Section

PT

4 describes the initial results regarding the direct effect of compulsory administration on the municipal budget and explains the potential mechanisms that trigger spillovers. Section 5

CE

describes the empirical strategy. Section 6 reports and discusses the results, and Section 7 concludes.

AC

2. Institutional Background 2.1. Municipal-level institutions and budget Italy has three different sub-national tiers of government. There are 20 regions (regioni),

110 provinces and around 8,000 municipalities (comuni). This paper focuses on the lowest administrative unit of government, municipalities. Italian municipalities are administrated by the mayor (sindaco) together with the executive branch (giunta) and the city council (consiglio comunale), both of which are elected for a term of 5 years. The mayor has a two-term limit 5

ACCEPTED MANUSCRIPT

and chooses the members of the executive branch. In general, local elections are dominated by national parties, though in smaller municipalities independent candidates and local parties have substantial influence. However, national and local parties commonly form coalitions and stand for elections with the same mayoral candidate. Italian municipalities are responsible for several aspects of local service provision, including those related to the environment (e.g., water and waste management, pollution monitoring,

CR IP T

regulation and preservation of urban green spaces), education (e.g., kindergarten, supplementary services for primary school), transportation (e.g., road maintenance, public transportation), welfare (e.g., social housing, aid to needy people) and culture (e.g., libraries, museums). In 2013, almost 30% of the country’s aggregated municipal expenditures were allocated to administrative costs, while policy related to traffic and transportation, and to environmental services accounts, accounted for 20% and 13% of total expenditures, respectively. Moreover,

AN US

around 70% of total expenditures were used for current spending, while 16% went to investments. The remaining 14% came from either loan repayments or expenditures on behalf of third parties.

These expenditures are largely covered by transfers from other tiers of government or by the municipalities’ revenues, which mainly come from property tax and a surcharge on the

M

national income tax.8 In 2013, roughly 40% of the revenue was derived from local taxation, while 15% came from transfers and another 15% from non-tax revenues. The remaining 30%

ED

came from sales of public assets, loans or revenue on behalf of third parties. 2.2. The law against mafia infiltration

Italy suffers from a pervasive presence of organized crime, which has created problems in

PT

the real economy, and through infiltration in the public administration that has inflated public budgets.9 Caneppele and Martocchia (2014) describe in detail how the mafia takes advantage

CE

of public procurement by influencing the design and execution of contracts, as well as the tender stage. At the design stage, local administrations can favor the start of projects that require the involvement of firms own by individuals connected to criminal organizations. During

AC

the tender process, criminal organizations often use threats or violence to deter competitors. Finally, the execution of the contract is either made directly by firms connected to the criminal 8

The property tax was abolished and reintroduced several times during the period of analysis. In the

years in which it was not collected, municipalities usually received transfers from the central government to compensate for the missing revenues. 9 Barone and Mocetti (2014) show that in Italy, the effect of financial aid to areas affected by natural disasters depends on the intensity of corruption in the area before the disaster.

6

ACCEPTED MANUSCRIPT

group or by “clean” firms which, however, are expected to transfer a portion of the revenue to the mafia (pizzo). It is clear that the systematic use of these practices fosters an inefficient use of public resources, which could be expected to have non-negligible effects on several aspects of local economies.10 During the 1980s and early 1990s, several laws were passed in Italy to address the damage that mafia-type organizations were inflicting on the economy.11 Importantly, in 1991 the Italian

CR IP T

Parliament approved a law (D.L. 31/05/1991, n. 164) to break the links between the local public administration and the mafia. This new policy empowers the central government to replace corrupt city councils with a three-person commission composed by experienced official to fight political corruption in local administration. The commission is appointed for an in initial period of usually 18 months, which can be extended to a maximum of 24 months.12 Typically, the process starts with a police investigation that identifies the presence of contacts

AN US

between municipal officials and organized crime. Most such investigations are initiated due to matters unrelated to the mafia.13 This information is then passed to the provincial-level Ministry of Interior representative, the prefetto, who assigns a commission to report, within 3 months, on whether the local government is likely to be liable to prosecution. After the final report is drafted, the prefetto has 45 days to notify the Ministry of Interior, which decides

M

whether there are grounds to dismiss the council. If so, the president of the republic confirms this decision by decree. At the end of the period of compulsory administration, new municipal elections are held.

ED

From 1991 to 2013 there were 243 city council dismissals involving a total of 191 municipalities.14 Figure 1 shows the number of dismissed councils by year. Of them, 147 were dismissed once, 36 twice and 8 three times. Further, the city council dismissal has been eventually

PT

declared unjustified (and hence revoked) 19 times. The fact that almost one-fourth of these

10

CE

municipalities experienced more than one dismissal suggests that this law does not permaFor instance, Brueckner (1982) suggests that inefficient public spending affects the housing market by

decreasing aggregate property values. 11 Laws have tended to pass immediately after very violent attacks against people who symbolized the war

AC

against the mob. For instance, article 416-bis of the penal code that targets the offense of mafia-type association was approved in September 1982 in reaction to the assassination of Gen. Dalla Chiesa, who was the Ministry of the Interior’s representative in the province of Palermo. 12 The 1991 law was largely integrated into a 1993 law (D.L. 20/12/1993, n. 529), which details the commission’s powers. In 2000 (D.Lgs. 18/08/2000, n. 267) it was merged with existing laws regulating the activities of local jurisdictions (Testo Unico Eenti Locali). Additional changes occurred in 2009 (L. 15/07/2009, n. 94). 13 Commissione parlamentare d’inchiesta sul fenomeno delle mafie e sulle altre associazioni criminali, anche straniere, 2005. 14 The complete list of municipalities is reported in Table A.1 in Appendix A.

7

ACCEPTED MANUSCRIPT

nently banish mafia activity from public administration. Indeed, many critics opposed the law because of its very limited long-term effects. For this reason, the paper focuses on identifying temporary effects that correspond to the presence of the compulsory administration. Table 1 indicates where these municipalities are located. As expected, almost all of them are in southern regions, which traditionally have active mafia-type criminal organizations: Mafia in Sicilia, ’Ndrangheta in Calabria, Camorra in Campania and Sacra Corona Unita in

CR IP T

Puglia. However, Liguria, Lazio, Piemonte and Lombardia have also recently experienced city council dismissals. 3. Data

I assembled a database that compiles information from a variety of sources for all munic-

AN US

ipalities belonging to the Italian regions of Campania, Calabria and Sicilia over the period 1998–2013. As shown, compulsory administration due to mafia infiltration is more frequent in the south of Italy. Indeed, the three regions analyzed account for more than 90% of all dismissals during the study period.15 This allows me to achieve a certain degree of homogeneity in the sample composition, which helps reduce bias stemming from using heterogeneous municipalities in terms of unobservables. In addition, for data availability reasons, I have to

M

restrict my dataset to 1998–2013, which excludes the first 7 years after the law’s passage. The missing data, however, should not significantly impact my analysis, as more than half of all

ED

dismissals occurred during this 16-year period. I first gathered municipal spending data from the Ministry of Interior’s website.16 I collected data on total expenditure, current expenditure and investment expenditure.17

PT

Second, I compiled information about the political status of each municipality in the sample from the Interior Ministry’s “Anagrafe degli Amministratori Locali e Regionali” database, which contains information on public officials in sub-central-level jurisdictions.18 I next con-

CE

structed the variable council dismissal, which equals 1 if a city council was dismissed in that

AC

year (i.e., the first year of compulsory administration), and 0 otherwise. In the empirical anal15 16

Golden and Picci (2005) suggest that these are also the Italian regions with the highest levels of corruption. I am thankful to Luca Repetto for sharing his data on municipal expenditures and for allowing me to check

and complement the database. 17 In a few cases, the website was missing information on municipal budgets. Although this slightly reduces the total number of observations in the sample, it should not affect my analysis, as these occurrences are likely to be random. 18 Since this database does not always clearly report whether a municipality is under compulsory administration, I checked the original decrees of the president of the republic and complemented the database information accordingly.

8

ACCEPTED MANUSCRIPT

ysis I also consider an alternative definition of this variable, which accounts for the three-year period starting from the year in which a city council was dismissed, since the length of compulsory administration can vary from 12–24 months.19 This is the more conservative approach to studying the effect of the compulsory administration, as it considers all years in which either the whole or part of the budget process was affected by appointed commissioners. In order to create the main regressor, I relate the presence of compulsory administration

CR IP T

to information on the level of neighborliness among municipalities. The variable neighbor’s council dismissal equals 1 when at least one neighboring municipality has the dummy council dismissal equal to 1, and 0 otherwise. I use a similar approach when considering the three-year period after a city council dismissal. The measure of neighborliness is constructed using data from the “Matrici di contiguità, distanza e pendolarismo” database provided by the Italian statistical office, which reports geographical data for all Italian municipalities. Following the

AN US

standard approach in the research related to spillovers, I consider geographical proximity to be the most relevant characteristic.20 Hence, my definition defines neighbors of municipality i as municipalities that share a border with it. Indeed, I expect physical closeness among municipalities to account for the relevant theoretical channels supporting my empirical analysis, which are highlighted in Section 4.2.21

M

I also compute additional variables to capture the economic closeness, relative spatial isolation, and relative size between dismissed municipalities and their neighbors. First, I create the dummy variable neighbors economically relevant that equals 1 if at least one bordering

ED

municipality experiences a city council dismissal that is economically relevant. I proxy the level of economic linkages between each pair of municipalities as the number of total commuters between them. For each municipality I define “highly connected” neighbors as those belonging

PT

to the top tertile of the distribution of the total number of commuters. Second, to measure the distance between municipalities I use the driving time from their centroids to the city hall cells

CE

(neighbors’ driving distance). Third, I account for the relative size of bordering municipalities that experience a city council dismissal by considering their population (neighbors’ population). Additionally, I create the dummy variable other council dismissal to account for a council

AC

dismissal in a municipality for reasons other than mafia infiltration (i.e., if elected officials resign, the annual budget is not approved or the mayor dies). While this variable equals 1 only for the year in which the dismissal occurred, I also use an alternative definition considering a 19

Most of the time it lasts for at least 18 months, and there is usually a 1-year delay for budget decisions to

take place. 20 See, for example, Case et al. (1993), Bordignon et al. (2003) and Lyytikäinen (2012). 21 Summary statistics for each variable are reported in Table 2.

9

ACCEPTED MANUSCRIPT

two-year period from the beginning of the dismissal because in these cases, the commissioner’s term is only 1 year.22 Non-mafia-related dismissals are clearly more common than compulsory administration for mafia infiltration. The variable commissioner accounts for whether in a given year a municipality is run by commissioners (either because of mafia infiltration or other reasons) or by elected officials. I again use the “Anagrafe degli Amministratori Locali e Regionali” database to compute

CR IP T

more variables that I use as additional controls. Local party takes a value of 1 if a municipality is governed by a mayor who has no party affiliation or who belongs to a list with no clear connection to a national party, and 0 otherwise. Electoral cycle indicates the strength of electoral incentives by counting the number of years since the last election (i.e., electoral incentives are stronger when elections are closer).23

Further, I collected time-varying control variables from the “Atlante statistico dei comuni”

AN US

database provided by the Italian statistical office, including municipal population and dependency ratio. I construct the latter using the percentage of old and young citizens (share pop > 64 and share pop < 15 ).

Finally, I define the time-invariant variable mafia.24 This dummy variable identifies municipalities in which the mafia is more likely to be active, which increases the probability of

M

having a corrupt municipality. Thus, following Buonanno et al. (2015), mafia is coded 1 for each municipality that experienced at least one real estate or firm seizure related to crimes committed by mafia-type organizations from the 1980s until the end of 2012.25 . Figures 3, 4

ED

and 5 show the locations of mafia-ridden municipalities and city council dismissals.

PT

4. The Consequences of a City Council Dismissal 4.1. The direct effect of compulsory administration

CE

Council dismissals are primarily designed to eliminate political corruption and ensure the smooth functioning of the local administration. Thus, their presence promotes an increase

AC

in local law enforcement. All actions taken by the appointed commissioners seek to further 22

23

Using the strategy described above, I also compute the variable neighbors’ other council dismissal. This variable reports values higher than 4, which is supposed to be the highest possible value, because

elections are held every 5 years. This is because some commissioners arrive at the end of the term. 24 Unfortunately, no study has quantified over-time variation in mafia activity at the municipal level. Papers that study the municipality level (see, for example, Barone and Narciso (2015)) focus on cross-section analyses, while those that assess mafia activity over time have aggregated information at the provincial level (see, for example, Acconcia et al. (2014) or Pinotti (2015)). 25 I use data provided by the Agenzia Nazionale per l’amministrazione e la destinazione dei beni sequestrati e confiscati alla criminalita’ organizzata.

10

ACCEPTED MANUSCRIPT

this goal. Typically, the commission cancels any illegitimate contracts or concessions granted by the corrupt council. Several Ministry of Interior reports suggest that the commissioners improve the overall governance and functioning of municipalities in which they intervene by reviewing the budget as well as by introducing, for instance, zoning or public procurement legislation (Ministry of the Interior, 2008, 2015).26 Immediate budget cuts are normal, as the commissioners need time to gather more information on the general status of expendi-

CR IP T

tures. However, after the first year of compulsory administration there is usually an increase in expenditures due to the implementation of new decisions taken by the commission. Interestingly, Acconcia et al. (2014) provide both anecdotal and empirical evidence of the effect of compulsory administration on public decisions, showing that provinces with at least one council dismissal reduced their total investments in infrastructure more than those with no dismissals.27

AN US

To make sure that compulsory administration has direct impacts on different categories of municipal spending, in line with existing research and more anecdotal evidence, below I report some useful descriptive evidence. I run two sets of regressions in which I link municipal annual budgets and compulsory administrations, focusing on the 124 municipalities from the sample that experienced at least one dismissal. The first set investigates what happens to the budget

M

during the first year the commission takes office. In the second set, I test how the commissioners’ presence affects spending in the following three-year period. Table 3 shows the results from the ordinary least squares (OLS) estimates, separately using per capita municipal total

ED

expenditure, current expenditure and expenditure on investment as dependent variables. The results, reported in Column (1) of Table 3, show that current expenditures are not affected in the first year after a city council dismissal, while investment is more than 45% lower than av-

PT

erage during the other periods. This strong reduction in investments affects total expenditure, which also experiences a slight decrease. Column (2) highlights that spending on investment

CE

is reduced, on average, by nearly 15% in each year of the three-year period after the council dismissal. Finally, total and current expenditure are not affected. These results are not meant to determine the direct effect of corruption on public spending,

AC

though the presence of a decrease only in investment, and not in current spending, is in line with previous research showing that corruption can lead to inefficient over-investment in the 26

Murray and Frijters (2016) study a similar process in the Australian context, where a state authority took

zoning control away from local politicians to facilitate development. Their results suggest that zoning policy is strongly affected by the political influence of landowners. 27 In a recent contribution, Cingano and Tonello (2016) provide evidence that compulsory administration is associated with a persistent decline in petty crimes.

11

ACCEPTED MANUSCRIPT

public sector (Mauro, 1998; Tanzi and Davoodi, 2000). Yet the previous findings (Bandiera et al., 2009) may be explained by passive, rather than active, waste. For instance, the senior officials appointed to run a compulsory administration are very likely to be more competent than those previously in power.28 Moreover, from a political economy perspective, one can argue that the dismissal of a city council because of mafia infiltration replaces politicians with bureaucrats. As is well known in the literature, the decisions of politicians (who have

CR IP T

electoral incentives) are different from those of bureaucrats (Besley and Coate, 2003; Alesina and Tabellini, 2007; Garmann, 2015), since the latter are less prone to be influenced by lobbies or enact policy simply to please voters. 4.2. Compulsory administration and spillovers

Compulsory administration, along with an implicit increase in law enforcement, produces

AN US

a negative effect on public investment. Since local spending decisions could be responsive to both of these consequences of a crackdown in a neighboring municipality, I discuss a number of relevant competing channels.

First, Sah (1991) suggests that law enforcement per se can produce spillovers. He argues that an individual’s perceived risk of punishment, which directly affects the probability that he or she will engage in criminal activities, is endogenously determined – and depends on

M

whether similar criminal activity is detected in the vicinity.29 In the present context, this would suggest that if a municipality falls under compulsory administration, this might affect

ED

agents’ risk perceptions, and thus fraudulent activities, in neighboring jurisdictions. Individuals performing illegal activities in the surrounding area may become more apprehensive about being detected themselves. Criminals may also fear that if commissioners are in the

PT

neighborhood, they may identify and report illegal activities occurring nearby to the enforcement authorities.30 Therefore, one should expect that a city council dismissal would produce

CE

a decrease in the occurrence of irregularities not just in the affected municipality, but also in its neighboring municipalities, potentially resulting in a reduction in neighboring municiDaniele and Geys (2015) study the effect of compulsory administration due to mafia infiltration on the

AC

28

education levels of the newly elected representatives after the end of the appointed commission’s term. They find that the new local officials are, on average, more educated than those from the dismissed council. 29 Sah’s (1991)’s model relaxes the assumption of exogeneity of the risk and perception of detection, which is standard in the literature that builds on Becker (1968)’s influential model of crime. 30 In a recent study, Rincke and Traxler (2011) provide initial empirical evidence confirming Sah’s (1991)’s predictions. They show that Austrian households’ compliance with paying TV license fees increases in communities where some households experience enforcement in the form of visits from licensing inspectors. Importantly, they show increasing compliance among both treated and untreated households.

12

ACCEPTED MANUSCRIPT

palities’ budgets. Alternatively, one might expect the criminal organization involved with the dismissed municipal government to temporarily relocate to other municipalities to compensate for the reduced revenues from public resources.31 In this scenario, the mafia would have an incentive to meddle with public activities in neighboring municipalities. Indeed, if this were the mechanism at work, one should expect an increase in spending on investment in these municipalities.

CR IP T

Second, as compulsory administration produces a reduction in public investment, fiscal spillover might also be observed. These competing channels focus on the well-established idea in public finance that a government’s spending decisions should be affected by those of its neighbors (Besley and Case, 1995; Case et al., 1993).32 Classical benefit spillovers could be expected if public goods or services provided by a jurisdiction increase the utility of citizens from neighboring municipalities because they consume them free of charge. For instance,

AN US

when local spending has an element of complementarity, a decrease in spending in a given jurisdiction might force its neighbors to increase spending. The opposite would be true if the public goods are substitutes (Case et al., 1993). Similarly, one could also expect the presence of economic spillovers, for example if a strong negative shock to public investment in a given municipality induced changes in other economic conditions (e.g., local wages or land rents),

M

which eventually affected neighboring public investment decisions (Peltzman, 2016). Finally, yardstick competition models link the quality of a government to the performance of its neighbors (Besley and Case, 1995). Assuming that neighboring municipalities have

ED

similar shocks, voters can infer the quality of a government by comparing its performance with that of its neighbors. Hence, office-oriented politicians will adjust their fiscal decisions to those of their neighbors in order to maximize their chances of re-election or political consensus.

PT

In this context, the dismissal of a municipality produces two consequences: first, it reveals the (poor) quality of former local officials; second, it improves the quality of local officials by

CE

assigning power to the commission. These information shocks increase citizens’ awareness of the quality of their own politicians, who may react by mimicking the behaviors of the commissioners. Therefore, one may expect that public investment will decrease not only in

AC

municipalities subjected to compulsory administration, but also in neighboring jurisdictions. Overall, city council dismissals should produce enough changes that if spillovers exist, they

would impact budget decisions in neighboring municipalities. However, both the mechanism through which the externality should occur – and whether this is positive (i.e., decrease in 31

See Knight (2013) and Dell (2015) for recent empirical examples of how law enforcement affects crime

relocation to neighboring jurisdictions. 32 See also, for example, Brueckner (1998), Baicker (2005), Bordignon et al. (2003) and Solé-Ollé (2006).

13

ACCEPTED MANUSCRIPT

neighbors’ expenditure) or negative (i.e., increase in neighbors’ expenditure) – are empirical matters. 5. Methodology 5.1. The estimation strategy

CR IP T

Most of the literature on spillover effects has focused on standard spatial econometric models. In principle, these models improve the results of OLS estimates by taking care of the omitted variable bias caused by spatial correlation (Anselin, 1988). However, in a recent contribution, Gibbons and Overman (2012) contend that these models are correctly identified only if very strong assumptions hold, and therefore their results could be misleading.33 The policy under study can be considered a source of exogenous variation that is likely to overcome

AN US

issues rising from spatial dependence in the error terms. Hence, a reduced-form approach based on OLS models, which satisfy specific identifying assumptions, should allow estimates with causal interpretation. My strategy is composed of two steps. First, I evaluate whether a council dismissal in a municipality affects public spending in neighboring municipalities. While this analysis emphasizes emphwhether spillovers are likely to exist, it reveals nothing about the mechanism through which city council dismissals should affect neighbors’ budgets. Second,

M

I complement this approach with further checks that aim to understand which channels are likely to play a role.

ED

Formally, in the baseline analysis I estimate the following model: Yipt = βTit + αXit + δi + χpt + it ,

(1)

PT

where i denotes a municipality, p the province and t the year. Yipt , the dependent variable, refers to expenditure in the investment of municipality i in year t. Tit , depending on the

CE

specification, is a dummy that either (1) takes a value of 1 in the year in which at least one neighbor experiences a council dismissal or (2) takes a value of 1 if at least one neighbor

AC

of municipality i experiences a city council dismissal in year t or has experienced it in the 33

Typically, these models use an instrumental variable strategy in which the policy implemented by a juris-

diction (e.g., tax rate or public expenditure) is regressed against a weighted average of those implemented by its neighbors. Eventually, socio-economic characteristics of these neighbors are used to instrument the endogenous variable. These instruments, however, do not seem to satisfy the required exclusion restriction. Indeed, it is likely that either (1) the instruments directly affect the main dependent variable or (2) there are unobserved characteristics that are correlated with both the instruments and the dependent variable. An alternative is the maximum likelihood methodology, which still relies on strong assumptions related to both exogeneity and distributional and functional form.

14

ACCEPTED MANUSCRIPT

previous two years (i.e., in t, t − 1 or t − 2). Xit are time-varying municipal controls, while δi

and χpt are municipality and province-year fixed effects, respectively. Finally, it is the error term. 5.2. Identification issues As already suggested, under certain conditions the estimates of the β coefficient from

CR IP T

the fixed-effects model of Equation (1) can mimic a randomized experiment and represent the causal effect of compulsory administration on the local public spending of neighboring municipalities.

A first concern is that municipalities that neighbor a municipality put under compulsory administration could be systematically different from the other municipalities in the sample. In this case it would be difficult to make a comparison, because the city council dismissal

AN US

would affect “treated” and “untreated” municipalities in very different ways. In order to avoid this potential selection bias, I focus all my regressions on municipalities that experienced the treatment at least once (i.e., have at least one neighboring municipality put under compulsory administration during the study period). Although this reduces the number of municipalities considered from 1,348 to 400, it makes it possible to produce causal estimates under weaker identifying assumptions. Given this sample restriction, the estimates will provide causal ev-

M

idence if the timing of city council dismissals among neighboring municipalities is random conditional on the controls. For a municipality, the occurrence of a city council dismissal has

ED

some elements of exogeneity, as the decision is made by the central government. Nevertheless, one may still expect that municipalities that are put under compulsory administration have engaged in behaviors that triggered the city council dismissal. Yet, this event is clearly exoge-

PT

nous and potentially unexpected for their neighboring municipalities, which I am interested in. However, in order to consistently reduce further concerns about potentially omitted variable

CE

bias, I control for fixed and time-varying municipal characteristics. The model includes municipality fixed effects to control for distinctive municipal features that are fixed over time. This removes concerns about specific municipal characteristics that

AC

could bias the estimates because they are correlated with budget decisions. For instance, they account for the historical presence of organized crime in a given municipality and its effect on public spending. Importantly, the main specification also controls for all time-varying characteristics at the provincial level. The inclusion of province-year fixed effects is a very strong element of my specification, as it controls for attributes that vary over time for a rather small geographical area (e.g., GDP shocks or changes in the provincial government). They control for potential variability, both within and between provinces, in the prefetto’s ability to run his office, and thus properly use the policy of interest. Additionally, as described in 15

ACCEPTED MANUSCRIPT

the data section, I include a set of time-varying control variables that might help increase the precision of my estimates. Finally, I use two different approaches to test the identifying assumptions. First, I look at the relative size of the omitted variable bias by examining how the inclusion of additional controls affects the size of the coefficient of interest. On the one hand, the introduction of controls can produce significant variation in the main coefficient. Since this implies that the

CR IP T

estimated coefficient is likely to be affected by the introduction of even more controls, the resulting bias from omitting controls could confound my estimate. On the other hand, the inclusion of further controls can have a very limited impact on the size of the main coefficient. In this case, the stability of the coefficient would favor a causal interpretation of the results. I follow the approaches of Altonji et al. (2005) and Oster (2015) to produce a formal test. Therefore, I quantify how large the effect of unobservables must be to neutralize the estimated

AN US

effect under the assumption that the selection on observables is proportional to the selection on unobservables. Second, I present evidence from a placebo test that controls for anticipatory effects by including leads that represent dummies for future compulsory administration in neighboring municipalities. This should reveal whether municipal spending was affected by

M

the presence of unlawful conduct in an eventually dissolved neighboring municipality. 6. Results

ED

6.1. Baseline results

Table 4 reports the effect of compulsory administration in a municipality on public investment in neighboring municipalities. These coefficients identify the average treatment effect on

PT

the treated. The first three columns (1 to 3) account for the very first year of compulsory administration, while the following three (4 to 6) focus on a three-year period after the city

CE

council dismissal.

The main regressor is not statistically significant in any of the first three columns, though the coefficient is consistently negative. However, these results are different if I look at the effect

AC

occurring in the three years after the city council dismissal. Column (4), which includes only the main regressor with no additional controls, the coefficient would suggest an 8% reduction in investment when a neighbor is put under compulsory administration. However, this effect is not statistically significantly different from zero.34 In Column (5), I add municipality fixed 34

In all regressions I use standard errors clustered both by municipality and by year. In addition, I report

standard errors obtained by clustering the errors at either the municipality or province level, or by province and by year, in Table A.2 in the Appendix.

16

ACCEPTED MANUSCRIPT

effects and province-year fixed effects. The estimated effect is now 6% and is significantly different from zero at the 5% level. Column (6), which introduces further municipal timevarying controls, again suggests a nearly 6% reduction in investment and has the same level of significance as the previous estimate. The estimates of the effect of covariates show interesting results as well. In particular, there is evidence of a political budget cycle, as spending increases near elections. Importantly,

CR IP T

the presence of a commissioner (either because of mafia infiltration or for other reasons) has the expected negative effect on public investment. It is also interesting that municipalities governed by local parties tend to spend less.

In conclusion, compulsory administration reduces investment in both municipalities that experience a dismissal (Table 3) and their neighbors. With a rough calculation I can approximate the aggregate effect and value the relevance of spillovers. On average, a municipality from

AN US

my sample spends 18,720,000 euro on investment over a three-year period (390 p/c investment X 16,000 inhabitants X 3 years). To evaluate the direct effect, one should consider that investment decreases by an average of 14% per year, which means that a municipality receiving an intervention would invest 16,099,200 euro instead of 18,720,000 (i.e., a 2,620,800 reduction). To estimate the spillover effects it is important to take into account that each municipality has,

M

on average, 6 adjacent neighbors. Because of the council dismissal, each of these neighbors has a 6% reduction in investment per year; thus investments over a three-year period will fall from 112,320,000 (18,720,000 X 3 X 6) to 105,580,800 (i.e., 6,739,200 reduction). Therefore,

ED

in aggregate the policy reduces spending by 9.360.000 – i.e., from 131,040,000 (18,720,000 + 112,320,000) to 121,680,000 (16,099,200 + 105,580,800). Put differently, the overall reduction is equal to half of a three-year period municipality investment. Finally, spillovers account for

PT

around 70% of the overall policy effect. The picture that emerges from my analysis should encourage policymakers not to underestimate the wide-ranging consequences of anti-corruption

CE

policies.

6.2. Checking the identifying assumptions

AC

In this section, I test the identifying assumptions (i.e., whether the city council dismissal

of a neighboring municipality is random conditional on the controls). The first element to emphasize is the relative stability of the coefficients shown over the different specifications in Table 4. If one considers the whole three-year period of compulsory administration, the effect decreases from 8% in the first specification to 6% when I include all controls, while for the first year it moves from 7.4% to 3.4%. More formally, Table 5 reports the results of a test in the

17

ACCEPTED MANUSCRIPT

spirit of Altonji et al. (2005) and Oster (2015).35 The key information is in the last two rows, which display the parameter δ and the identifying bounds. The former indicates how large the selection on unobservables has to be in order to cancel out the results, while the latter reports the set of values the coefficient can take, assuming a δ = 1. Interestingly, the computed δ for the estimates in Columns (5) and (6) are much higher than 1 – 11.20 and 7.88 in Columns (5) and (6), respectively – indicating that the results are robust (Oster, 2015). Therefore, the

CR IP T

effect of unobservables needs to be at least 7.5 times stronger than that of observables in order to completely nullify the effect. Also, the identification sets estimated include only negative values. The results shown in Columns (2) and (3) are also robust, as the δ are still greater than 1. Overall, these results do not seem to be biased by omitted variables that are possibly correlated with both the dependent variable and treatment.

As a further robustness check, I investigate the presence of anticipatory effects by intro-

AN US

ducing leads of neighbors’ council dismissal up to 4 years, to the full controlled regression (i.e., Column (6) of Table 4). These results are reported in Table 6. Expenditure on investments in year t is not associated with the city council dismissal in future years. This is true both if one looks at each individual year separately or at the overall effect of the following 4 years. To conclude, both robustness checks seem to confirm the identifying assumptions, thus reinforcing

6.3. Identifying the channel

M

the causal interpretation of the baseline results.

ED

While the baseline specification aims to estimate the effect of the council dismissal on neighboring municipalities’ spending, it does not address the alternative theoretical reasons why this might occur. Although it is arguably difficult to empirically distinguish among

PT

the different determinants of spillovers, in this section I complement the basic results with additional estimates.

CE

I begin by looking at how the estimated effect depends on certain municipal characteristics, of either the municipality of interest or its neighbors experiencing a city council dismissal. I provide a set of results from several interaction models in which the main regressor is interacted

AC

each time with a different variable. First, I test for the presence of law enforcement spillovers. I can test the plausibility of this

channel by verifying whether the “predetermined” presence of mafia activity in a municipality makes the potential spillover stronger or weaker.36 Thus, I exploit the differences in the 35

For further details and a formal derivation of δ and the identification set, see Oster (2015). All calculations

are performed using the PSACALC Stata module by Oster (2013). 36 Several studies on the origins of organized crime in the south of Italy have confirmed that the current

18

ACCEPTED MANUSCRIPT

likelihood that a municipality will be mafia ridden using the variable mafia. Second, another relevant channel that could explain spillovers from the city council dismissal to neighbors’ spending decisions is yardstick competition. The variable electoral cycle accounts for the different electoral incentives during the years of each term. That should help me verify whether the effect of a city council dismissal on expenditure in a neighboring municipality increases the closer the dismissal occurs to the next election. Third, I test for the presence of economic

CR IP T

spillovers by verifying whether the reduction in public investments is larger when the neighbor that experiences the city council dismissal is economically important to the municipality of interest. Finally, I provide two additional results showing whether either the relative size or distance between municipalities is relevant for determining the sensitivity of public investment to the treatment.

Table 7 presents the results from the estimated interaction models. Columns (1) and

AN US

(6) report that the interaction between neighbors’ council dismissal and mafia is negative and significantly different from zero at the 1% level.37 Specifically, expenditure investment decreases by 12% in mafia-ridden municipalities that share a border with a municipality that experienced a compulsory administration. Interestingly, a similar result also appears when I consider the regressor that identifies only the first year of the council dismissal, though in

M

the baseline estimations reported in Table 4 there was no significant effect; the reduction in investment was 11%.38 These findings support the hypothesis that law enforcement spillovers could be driving the overall effect emphasized in the baseline analysis. The idea is that, ceteris

ED

paribus, municipalities that have a history of mafia activity react more strongly to the dismissal of a city council in a neighboring municipality than those in which the mafia is less likely to be present, because they are the ones that should most “fear” a potential intervention in their own

PT

municipality. Indeed, for municipalities with a relatively low probability of having a corrupt government in the territory (i.e., mafia=0), the occurrence of a city council dismissal in a

CE

neighboring municipality seems to have a somewhat positive, though not significant, effect. Columns (2) and (7) show the effect of a city council dismissal interacted with the number of years since the last election, which proxies for the strength of electoral incentives. Since

AC

the interaction terms of both specifications are not significantly different from zero, it seems

locations of mafia activity have arisen from the specific conditions of decades or centuries ago (Buonanno et al., 2015; Gambetta, 1996). 37 Although mafia is fixed, my specification implicitly controls for variation in mafia activities over time at

the provincial level (province-year fixed effects). 38 These come from summing the two reported coefficients in Columns (1) and (4) of Table 7, respectively. The p-value of the joint significance of the two coefficients in Column (1) is 0.007, and in Column (4) it is 0.009.

19

ACCEPTED MANUSCRIPT

to rule out the presence of spillovers due to yardstick competition. Still, the electoral cycle variable is always significant and positive, confirming the presence of a political budget cycle as public investment increases as elections approach. The outcomes in Columns (3) and (8) show that economic closeness between municipalities, proxied by the number of commuters, does not seem to increase the size of the spillover. Once again, the interaction terms for both specifications are insignificant. This is initial evidence

CR IP T

that potential changes in economic conditions due to a reduction in public investment in a neighboring municipality do not play a relevant role in explaining the spillover effects.

Finally, I show additional results testing whether the spillover effects are heterogeneous depending on both the level of municipal isolation (i.e., driving distance), Columns (4) and (9), and the size (i.e., population), Columns (5) and (10), of the municipalities in which the city council was dismissed. Interestingly, Column (9) suggests that the closer the neighboring

AN US

municipality that experienced a city council dismissal, the stronger the spillover effect. Although this finding is in line with several of the various proposed channels, it is important, as it emphasizes the spatial dimension of the spillovers. By contrast, the population size of the municipality under compulsory administration does not seem to be important in this analysis. 6.4. Timing of the treatment effects

M

Since I observe that neighbors’ reactions to a council dismissal are delayed, on average, I also analyze the timing of the effect. In order to tease out the differential effects over time, I

ED

estimate the baseline model including separate lags with respect to the year of the treatment.39 I report in Figure 2 the coefficients of interest and the associated 95% confidence intervals for estimates considering, alternatively, the sample of all (Panel A), only mafia-ridden (Panel B)

PT

or only non-mafia-ridden municipalities (Panel C). Panel A suggests that the effect found for the three-year period is mainly driven by a drop in investments in the second year after

CE

the city council dismissal. However, this finding appears to be driven by the sum of two opposite effects at the beginning of the treatment period. Panel B shows that the effect is positive for municipalities that are unlikely to be mafia ridden, while Panel C shows that the

AC

effect is negative for mafia-ridden municipalities. These additional results seem to confirm the importance of law enforcement spillover, yet they point to the potential displacement of rent activities to the detriment of less-corrupt municipalities. 39

Formally, I estimate the following equation:

Yipt = βTit + βTit−1 + βTit−2 + βTit−3 + βTit−4 αXit + δi + χpt + it

20

ACCEPTED MANUSCRIPT

6.5. Falsification tests In order to more rigorously rule out the presence of economic or benefit spillover from public goods, I show the results of a falsification test in which I focus on the effect generated by nonmafia-related city council dismissals.40 I conduct this test to examine how local spending responds to the occurrence of such an event in a neighboring municipality and compare it to

CR IP T

the effect estimated in previous sections. To do so, I replicate part of the results using either other council dismissal or neighbors’ other council dismissal as main regressors instead of council dismissal and neighbors’ council dismissal, respectively. If the intervention of commissioners who are in power for reasons other than mafia infiltration produces a similar direct effect on investment (i.e., a reduction in expenditure on investment in municipalities that experience a council dismissal), then I can

AN US

emphasize the presence of fiscal/economic spillovers that are not associated with a crackdown on corruption. In other words, law enforcement spillovers should not be observed in this scenario.

Table 8 reports the results of the falsification test. Columns (1) and (4) present the direct effect of a non-mafia-related city council dismissal. This is estimated considering only municipalities from the full sample that experienced at least one non-mafia-related city council

M

dismissal during the study period. There is a drop in public investment in both the year of the dismissal and in the two years after the dismissal, probably because the commissioner often takes bold measures to improve the municipality’s financial stability. These findings are

ED

comparable to what happens after a council dismissal for mafia infiltration, reported in Table 3. The results shown in Columns (2–3) and (5–6) of Table 8 are estimated for a restricted

PT

sample of neighbors. I include only municipalities from the main analysis with a neighbor that experienced at least one non-mafia-related city council dismissal (n = 368). These estimates suggest that non-mafia-related city council dismissals do not significantly affect the spending

CE

on investment of neighboring municipalities. Interestingly, it does not seem to matter whether a municipality is identified as mafia ridden. These findings suggest that spillover effects are

AC

not related to fiscal/economic conditions. 7. Conclusion Anti-corruption policies, such as monitoring and auditing public officials’ activities, seem

to be effective mechanisms for addressing corruption and misconduct in lower levels of government. Nevertheless, when illegal activities regularly occur in local public administration, 40

Acconcia et al. (2014) and Daniele and Geys (2015) conduct a similar test.

21

ACCEPTED MANUSCRIPT

anti-corruption policies are more likely to produce indirect consequences for neighbors that may affect their efficiency. I use the case of Italy here to provide the first empirical evidence of the existence of such spillover effects. I find that the presence of a commission brought in to replace corrupt city councils significantly reduces municipal spending on investments in both the affected municipality and its neighbors. During the first three years after a neighbor’s city council is dismissed, mu-

CR IP T

nicipal spending on investments decreases by an average of 6% annually. Moreover, I show that only mafia-ridden municipalities (i.e., those with a higher probability of corruption in public administration) are negatively affected by the arrival of commissioners in neighboring municipalities, and that compulsory administration for other reasons do not produce spillover effects. I also observe that non-mafia-ridden municipalities experience a marginal increase in spending when there are commissioners running the city councils of neighboring municipalities.

AN US

These findings are consistent with the presence of law enforcement spillover effects indirectly affecting mismanagement in neighboring municipalities, as modeled by Sah (1991). The evidence reported here represents an initial step towards better understanding the indirect consequences of anti-corruption policies. Future research should aim to determine whether the effect on the budget is due to active or passive waste (Bandiera et al., 2009). In

M

addition, it is important to evaluate whether criminal organizations compensate for the reduction in revenue due to effective anti-corruption policies by increasing other illegal activities (i.e., drug trafficking or racketeering). Finally, further investigations are needed to evaluate

AC

CE

PT

ED

whether my findings can be generalized to anti-corruption policies in other countries.

22

CR IP T

0

Number of dismissed municipalities 20 30 10

40

ACCEPTED MANUSCRIPT

1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013

AC

CE

PT

ED

M

AN US

Figure 1: Council dismissals by year.

23

ACCEPTED MANUSCRIPT

0

1

2

3

CR IP T

-.3

-.2

-.1

0

.1

.2

.3

Panel A. All municipalities

4

AN US

Year since city council dissolution in a neighboring municipality

M

-.3 -.2 -.1

0

.1

.2

.3

Panel B. Mafia-ridden municipalities

1

ED

0

2

3

4

Year since city council dissolution in a neighboring municipality

-.3 -.2 -.1

AC

0

.1

.2

CE

.3

PT

Panel C. No mafia-ridden municipalities

0

1

2

3

4

Year since city council dissolution in a neighboring municipality

Figure 2: Timing of the estimated spillover effects

24

Figure 3: Campania

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

25

Figure 4: Calabria

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

26

Figure 5: Sicilia

AC

CE

PT

ED

M

AN US

CR IP T

ACCEPTED MANUSCRIPT

27

CR IP T

ACCEPTED MANUSCRIPT

Table 1: City council dismissals by region and province

1

Lombardia

Catanzaro

8

Avellino

2

Benevento

Crotone

4

Caserta

33

Reggio Calabria

45

Napoli

51

Vibo Valentia

14

Salerno

6

Piemonte 1

Lazio

Cosenza

Torino

4

3

Roma

Liguria

1

Imperia

1

Puglia

Sicilia

Bari

5

Agrigento

7

Lecce

2

Caltanissetta

6

CE

PT

ED

Milano

Campania

M

Matera

Calabria

AN US

Basilicata

Catania

10

Messina

3

Palermo

26

Ragusa

1

Siracusa

1

Trapani

7

AC

Notes: The table displays the number of municipality dismissals for the period 1991-2013 by region and province.

28

2

ACCEPTED MANUSCRIPT

Table 2: Summary statistics

Mean

Std. Dev.

N

Total expenditure p/c

1323.38

977.66

6303

Current expenditure p/c

700.86

297.41

6303

Investment expenditure p/c

388.55

786.71

6303

Council dismissal

0.015

0.121

6303

0.037

Neighbors council dismissal

0.112

0.19

6303

0.315

6303

Neighbors other council dismissal

0.198

0.399

6303

Local party

0.378

0.485

6303

Population

16325.86

61661.39

6303

Dependency ratio

0.349

0.036

6303

Commissioner

0.066

0.249

6303

AN US

Other council dismissal

CR IP T

Variable

Commissioner (mafia infiltration)

0.029

0.167

6303

Commissioner (no mafia infiltration)

0.037

0.19

6303

Electoral cycle

1.866

1.408

6303

0.572

0.495

6303

3.718

8.26

6303

5246.58

16764.12

6303

0.026

0.159

6303

Mafia Neighbors population

M

Neighbors driving distance

AC

CE

PT

ED

Neighbors economically relevant

29

ACCEPTED MANUSCRIPT

Table 3: Compulsory administration on local spending

(1)

(2)

First year

Three-year period

Coef.

Stand. Error

Coef.

Stand. Error

Total expenditures

-0.071***

0.026

-0.031

0.038

0.009

0.017

-0.018

0.024

-0.466***

0.097

-0.148*

0.084

Current expenditures Investment expenditure

CR IP T

Dependent Variable

Notes: The table reports estimates of regressions where the dependent variable is the Log of the reported category of expenditures per capita. These estimates consider annual information

for all municipalities from the regions Campania, Calabria and Sicilia, with at least one city council dismissal in the period from 1998-2013. The main regressor in the estimates are: in column (1) a dummy equal to 1 the year a municipality is put under compulsory

AN US

administration, and 0 otherwise; in column (2) a dummy equal to 1 for the year, and the following two years a municipality experiences a council dismissal, and 0 otherwise. Standard errors in parenthesis. Standard errors clustered two ways by municipality and by year. *p <

AC

CE

PT

ED

M

0.1, **p < 0.05 and ***p < 0.01.

30

ACCEPTED MANUSCRIPT

Table 4: Municipal investment and council dismissals in neighboring municipalities

First year

Neighbors counc. dism.

(2)

(3)

-0.074

-0.040

-0.034

(0.079)

(0.048)

(0.048)

Dependency ratio

ED

Commissioner (no mafia infiltration)

Province × year fixed effects

N

CE

N mun

PT

Municipality control variables

R2

M

Electoral cycle Commissioner (mafia infiltration)

(4)

(5)

(6)

-0.087

-0.067**

-0.061**

(0.057)

(0.033)

(0.031)

-0.078**

AN US

-0.079**

Population

Municipality FE

Three-year period

(1)

Local party

CR IP T

Bordering municipalities

(0.032)

(0.032)

-0.000**

-0.000**

(0.000)

(0.000)

3.380**

3.372**

(1.444)

(1.452)

0.072***

0.072***

(0.009)

(0.009)

-0.330***

-0.330***

(0.065)

(0.065)

-0.434***

-0.432***

(0.085)

(0.084)

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

No

No

Yes

No

No

Yes

0.001

0.382

0.391

0.001

0.382

0.391

6303

6303

6303

6303

6303

6303

404

404

404

404

404

404

Notes: The dependent variable is the Log of investment expenditure p/c. In columns (1 to 3) neighbors council dismissal is a dummy taking the value of 1 the year in which a neighboring municipality experiences a council dismissal, while in

AC

columns (4 to 6) is a dummy taking the value of 1 if at least one of the neighbors of the municipality i experiences a city council dismissal in year t or have experienced it in the previous two years (t − 1 or t − 2). Municipality control variables are: local party, population dependency ratio, electoral cycle, commissioner (mafia reasons) and Commissioner (no mafia

reasons). Standard errors in parenthesis. Standard errors clustered two ways by municipality and by year. *p < 0.1, **p < 0.05 and ***p < 0.01.

31

AN US

CR IP T

ACCEPTED MANUSCRIPT

Table 5: Selection on unobservables (2)

(3)

(4)

(5)

(6)

Controlled

Controlled

Uncontrolled

Controlled

Controlled

Coefficient

-0.074

-0.040

-0.034

-0.087

-0.067

-0.061

R-squared

0.001

0.382

0.391

0.001

0.382

0.391

3.92

δ

[-0.034,-0.022]

2.84

11.20

7.88

[-0.041,-0.022]

[-0.067,-0.061]

[-0.061,-0.054]

ED

Identified Set

M

(1) Uncontrolled

Notes: The dependent variable is expenditure on investment. The reported coefficients are those of Neighbors counc. dism. for both the first year and three-year period analysis. Each column replicates the specification used in Table 4. δ is calculated assuming

AC

CE

PT

Rmax = 1.3R-squared and β = 0. The identified set is calculated assuming Rmax = 1.3R-squared and δ = 1.

32

ACCEPTED MANUSCRIPT

Table 6: Pre-adoption effects

(1)

(2)

(3)

(4)

(5)

Bordering municipalities 0.006

0.006

(0.056)

(0.062)

F2 Neighbors counc. dism.

0.018 (0.038)

F3 Neighbors counc. dism.

Overall effect Joint significance (p-value)

(0.052)

-0.044

-0.033

(0.037)

(0.048)

0.042

0.046

(0.047)

(0.053)

AN US

F4 Neighbors counc. dism.

0.041

CR IP T

F Neighbors counc. dism.

0.060

0.611

Notes: The dependent variable is the Log of investment expenditure p/c. The regression includes Leads of the variable Neighbors council dismissal, which is a dummy taking the value of 1 the year in which at least one neighboring municipality experiences a council dismissal. The estimations include both municipal and province-year fixed effects and the following municipality control variables: local party, population dependency ratio, electoral

M

cycle and commissioner. Standard errors in parenthesis. Standard errors clustered two ways

AC

CE

PT

ED

by municipality and by year. *p < 0.1, **p < 0.05 and ***p < 0.01.

33

34 0.391 6303 404

R2

N

N mun

404

6303

0.391

Yes

Yes

Yes

(0.010)

0.065***

M 0.000

Yes

404

6303

0.391

Yes

Yes

Yes

0.071***

404

6303

0.391

Yes

Yes

Yes

404

6303

0.391

Yes

Yes

Yes

0.000

404

404

6303

0.392

Yes

Yes

Yes

Yes

404

6303

0.392

Yes

Yes

CR IP T

6303

0.392

Yes

Yes

(0.009)

404

6303

0.392

Yes

Yes

Yes

(0.003)

AN US

0.006**

0.002

(0.058)

-0.141**

(9)

(0.004)

0.015 (0.052)

-0.058

(0.025)

(0.034)

-0.066*

(8)

(0.084)

0.026 (0.018)

0.012

(0.045)

-0.111**

(7)

Three-year period

404

6303

0.392

Yes

Yes

Yes

(0.000)

0.000

(0.037)

-0.068*

(10)

year. *p < 0.1, **p < 0.05 and ***p < 0.01.

control variables are: local party, population, dependency ratio, electoral cycle and commissioner. Standard errors in parenthesis. Standard errors clustered two ways by municipality and by

to the closest neighbors experiencing a city council dismissal. Neighbors population sum the population of the bordering municipalities who experience a city council dismissal. Municipality

least one border municipality that experiences the city council dismissal which is economically relevant, while Neighbors driving distance account for driving time (in minutes) from the focal

related to crimes committed by mafia-type organizations. Electoral cycle counts the number of years from the last election. Neighbors econ. relevant is a dummy equals to 1 if there is at

in year t or have experienced it in the previous two years (t − 1 or t − 2). Mafia is a dummy taking the value of 1 for each municipality that experienced at least one real estate or firm seizure

municipality experiences a council dismissal, while in columns (6 to 10) is a dummy taking the value of 1 if at least one of the neighbors of the municipality i experiences a city council dismissal

Notes: The dependent variable is the Log of investment expenditure p/c. In columns (1 to 5) neighbors council dismissal is a dummy taking the value of 1 the year in which a neighboring

Yes

Province × year fixed effects

Municipality control variables

Yes Yes

Municipality FE

Electoral cycle

Neighbors counc. dism. X Neighbors population

Neighbors counc. dism. X Neighbors driving distance

(0.047)

0.032

(6)

(0.061)

(0.062)

-0.073

(5)

-0.156***

(0.083)

-0.064

(4)

(0.074)

(0.056)

-0.015

(3)

-0.228***

-0.057 (0.068)

0.112

(2)

(0.073)

(1)

First year

Bordering municipalities

Table 7: Testing the mechanism

ED

Neighbors counc. dism. X Neighbors econ. relevant

Neighbors counc. dism. X Electoral cycle

Neighbors counc. dism. X Mafia

PT

CE

Neighbors counc. dism.

AC

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT

Table 8: Falsification test: Municipal investment and council dismissals not for mafia infiltration

First year Direct

Spillover

Direct

Spillover

effect

effect

effect

effect

(2)

-0.354***

(4)

(5)

(6)

-0.356***

(0.070) Neighbors other counc. dism.

(3)

CR IP T

(1) Other counc. dism.

Two-year period

(0.056)

0.029

0.011

-0.001

-0.019

(0.028)

(0.045)

(0.031)

(0.044)

Neighbors other counc. dism. X Mafia

0.032

0.055

(0.068)

(0.061)

No

Yes

Yes

No

Yes

Yes

Province × year fixed effects

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

No

Yes

Yes

R2

0.010

0.400

0.400

0.006

0.400

0.400

N

6854

5741

5741

6854

5741

5741

N mun

439

368

368

439

368

368

AN US

Municipality FE Municipality control variables

M

Notes: The dependent variable is the Log of investment expenditure p/c. In column (1) other council dismissal is a dummy taking the value of 1 the year a municipality is put under compulsory administration for reasons other than mafia, while in column (4) is a dummy taking the value of 1 if at least one of the neighbors of the municipality i experiences a city council

ED

dismissal for reasons other than mafia in year t or have experienced it in the previous year t − 1. In columns (2) and (3)

neighbors council dismissal is a dummy taking the value of 1 the year in which a neighboring municipality experiences a

council dismissal, while in columns (5) and (6) is a dummy taking the value of 1 if at least one of the neighbors of the municipality i experiences a city council dismissal in year t or have experienced it in the previous two years (t − 1 or

PT

t − 2). Mafia is a dummy taking the value of 1 for each municipality that experienced at least one real estate or firm

seizure related to crimes committed by mafia-type organizations. Municipality control variables are: local party, population

dependency ratio, electoral cycle and commissioner. Standard errors in parenthesis. Standard errors clustered two ways

AC

CE

by municipality and by year. *p < 0.1, **p < 0.05 and ***p < 0.01.

35

ACCEPTED MANUSCRIPT

A. Appendix Table A.1: List of municipalities experiencing a city council dismissal (1991-2013) Condofuri

Montecorvino Pugliano

San Luca

Aci Catena

Corigliano Calabro

Nardodipace

San Paolo Belsito

Adrano

Cosoleto

Nettuno

San Procopio

Afragola

Crispano

Nicotera

San Tammaro

Africo

Delianuova

Niscemi

Santa Flavia

Altavilla Milicia

Ercolano

Nocera Inferiore

Santa Maria La Carita’

Amantea

Fabrizia

Nola

Sant’Andrea Apostolo dello Ionio

Ardore

Ficarazzi

Orta di Atella

Sant’Antimo

Arzano

Frattamaggiore

Ottaviano

Sant’Antonio Abate

Augusta

Frignano

Pagani

Sant’Ilario dello Ionio

Bagaladi

Furnari

Pago del Vallo di Lauro

Santo Stefano in Aspromonte

Bagheria

Gallipoli

Pantelleria

Sant’Onofrio

Bardonecchia

Gela

Parghelia

Sarno

Bordighera

Gioia Del Colle

Partanna

Scafati

Borgia

Gioia Tauro

Pignataro Maggiore

Scicli

Boscoreale

Giugliano in Campania

Pimonte

Sedriano

Botricello

Gragnano

Bova Marina

Grazzanise Guardavalle

Burgio

Isca sullo Ionio

Caccamo

Isola delle Femmine

Calanna

Isola di Capo Rizzuto

Calatabiano

Lamezia Terme

Caltavuturo

Lascari

Camini

Leini

Campobello di Licata

Licata

Seminara

Plati’

Siculiana

Poggiomarino

Siderno

Polizzi Generosa

Sinopoli

Pollina

Soriano Calabro

Pomigliano d’Arco

Stefanaconi

Pompei

Strongoli

Portici

Surbo

Pozzuoli

Taurianova

Quarto

Terlizzi

Quindici

Terme Vigliatore

Liveri

Canicatti’

Lusciano

Capaci

Marano di Napoli

Recale

Teverola

Marcedusa

Reggio di Calabria

Torre Annunziata

Marcianise

Riesi

Torre del Greco

Marina di Gioiosa Jonica

Rivarolo Canavese

Torretta

Casalnuovo di Napoli

Mascali

Rizziconi

Trabia

Casaluce

Mascalucia

Roccaforte del Greco

Trani

Casamarciano

Mazara del Vallo

Roccamena

Tufino

Casandrino

Melito di Napoli

Roghudi

Vallelunga Pratameno

Casapesenna

Melito di Porto Salvo

Rosarno

Ventimiglia

Carinola

PT

Casal di Principe

ED

Campobello di Mazara

Careri

Racalmuto

Termini Imerese

Ragalna

Terzigno

Mileto

S. Maria La Fossa

Vicari

Casola di Napoli

Misilmeri

S.Lorenzo Maggiore

Villa di Briano

Casoria

Misterbianco

Salemi

Villa Literno

Castel Volturno

Modugno

Samo

Villabate

Castellammare del Golfo

Molochio

San Calogero

Villaricca

Castello di Cisterna

Monasterace

San Cipriano d’Aversa

Volla

Castrofilippo

Mondragone

San Ferdinando

Cerda

Mongiana

San Gennaro Vesuviano

Cesa

Monopoli

San Giovanni La Punta

Cinisi

Montalbano Jonico

San Giuseppe Vesuviano

Ciro’

Montebello Jonico

San Gregorio d’Ippona

CE

Casignana

AC

AN US

Gricignano di Aversa

Brusciano

Piraino

M

Briatico

CR IP T

Acerra

Notes: This is a list of all municipalities that experienced at least one city council dismissal starting from the law approval in 1991 until the end of 2013.

36

CR IP T

ACCEPTED MANUSCRIPT

Table A.2: Robustness check of main results (Table 4) - different st. errors clustering Bordering municipalities First year

Neighbors counc. dism.

-0.074

(2)

(3)

(4)

AN US

(1)

Three-year period (5)

(6)

-0.040

-0.034

-0.087

-0.067

-0.061

{0.061}

{0.034}

{0.036}

{0.050}

{0.030}**

{0.028}**

(0.057)

(0.030)

(0.031)

(0.053)

(0.035)*

(0.034)*

[0.050]

[0.050]

[0.045]*

[0.038]*

[0.037]

[0.055] No

Province × year fixed effects

No

Yes

Yes

No

Yes

Yes

Yes

Yes

No

Yes

Yes

No

No

Yes

No

No

Yes

R2

M

Municipality FE

0.001

0.382

0.391

0.001

0.382

0.391

N

6303

6303

6303

6303

6303

6303

404

404

404

404

404

404

ED

Municipality control variables

N mun

Notes: The dependent variable is the Log of investment expenditure p/c. In columns (1 to 3) neighbors council

PT

dismissal is a dummy taking the value of 1 the year in which a neighboring municipality experiences a council dismissal, while in columns (4 to 6) is a dummy taking the value of 1 if at least one of the neighbors of the municipality i experiences a city council dismissal in year t or have experienced it in the previous two years (t − 1

or t − 2). Municipality control variables are: local party, population dependency ratio, electoral cycle, commissioner

CE

(mafia reasons) and commissioner (no mafia reasons). Standard errors clustered two ways by province and by

year in braces. Standard errors clustered by province in parenthesis. Standard errors clustered by municipality in

AC

brackets. *p < 0.1, **p < 0.05 and ***p < 0.01.

37

ACCEPTED MANUSCRIPT

Bibliography References Acconcia, A., Corsetti, G., Simonelli, S., 2014. Mafia and public spending: Evidence on the fiscal multiplier from a quasi-experiment. American Economic Review 104 (7), 2185–2209.

CR IP T

Acemoglu, D., Garcia-Jimeno, C., Robinson, J. A., 2015. State capacity and economic development: A network approach. American Economic Review 105 (8), 2364–2409.

Alesina, A., Tabellini, G., 2007. Bureaucrats or Politicians? Part I: A single policy task. American Economic Review 97 (1), 169–179.

Altonji, J. G., Elder, T. E., Taber, C. R., 2005. Selection on observed and unobserved variables:

AN US

Assessing the effectiveness of catholic schools. Journal of Political Economy 113 (1), 151–184. Anselin, L., 1988. Spatial Econometrics: Methods and Models. NATO Asi Series. Series E, Applied Sciences. Springer.

Arin, K. P., Chmelarova, V., Feess, E., Wohlschlegel, A., 2011. Why are corrupt countries less successful in consolidating their budgets? Journal of Public Economics 95 (7-8), 521–530.

M

Avis, E., Ferraz, C., Finan, F., 2016. Do Government Audits Reduce Corruption? Estimating the Long-Term Impacts of Exposing Corrupt Politicians, mimeo.

ED

Baicker, K., 2005. The spillover effects of state spending. Journal of Public Economics 89 (2-3), 529–544.

PT

Bandiera, O., Prat, A., Valletti, T., September 2009. Active and passive waste in government spending: Evidence from a policy experiment. American Economic Review 99 (4), 1278–

CE

1308.

Bardhan, P., Mookherjee, D., 2006. Decentralization, corruption and government accountabil-

AC

ity. Edward Elgar Publishing, Inc. Barone, G., Mocetti, S., 2014. Natural disasters, growth and institutions: A tale of two earthquakes. Journal of Urban Economics 84 (C), 52–66.

Barone, G., Narciso, G., 2015. Organized crime and business subsidies: Where does the money go? Journal of Urban Economics 86, 98 – 110. Becker, G. S., 1968. Crime and punishment: An economic approach. Journal of Political Economy 76, 169. 38

ACCEPTED MANUSCRIPT

Besley, T., Case, A., 1995. Incumbent behavior: Vote-seeking, tax-setting, and yardstick competition. American Economic Review 85 (1), 25–45. Besley, T., Coate, S., 2003. Elected versus appointed regulators: Theory and evidence. Journal of the European Economic Association 1 (5), 1176–1206. Bordignon, M., Cerniglia, F., Revelli, F., 2003. In search of yardstick competition: A spatial 199–217.

CR IP T

analysis of Italian municipality property tax setting. Journal of Urban Economics 54 (2),

Brueckner, J. K., December 1982. A test for allocative efficiency in the local public sector. Journal of Public Economics 19 (3), 311–331.

Brueckner, J. K., November 1998. Testing for Strategic Interaction Among Local Governments:

AN US

The Case of Growth Controls. Journal of Urban Economics 44 (3), 438–467.

Brueckner, J. K., 2000. Fiscal decentralization in developing countries: The effects of local corruption and tax evasion. Annals of Economics and Finance 1 (1), 1–18. Brueckner, J. K., 2003. Strategic interaction among governments: An overview of empirical

M

studies. International Regional Science Review 26 (2), 175–188.

Buonanno, P., Durante, R., Prarolo, G., Vanin, P., 2015. Poor institutions, rich mines: Re-

ED

source curse in the origins of the sicilian mafia. The Economic Journal 125 (586), F175–F202. Caneppele, S., Martocchia, S., 2014. Italian mafias, public procurement and public works in southern italy. In: Caneppele, S., Calderoni, F. (Eds.), Organized Crime, Corruption and

PT

Crime Prevention. Springer International Publishing, pp. 293–299. Case, A. C., Rosen, H. S., Hines, J. J., 1993. Budget spillovers and fiscal policy interdependence

CE

: Evidence from the states. Journal of Public Economics 52 (3), 285–307. Cingano, F., Tonello, M., 2016. Law enforcement, social control and organized crime. Evidence

AC

from local government dismissals in Italy, mimeo.

Daniele, G., Geys, B., 2015. Organised crime, institutions and political quality: Empirical evidence from italian municipalities. The Economic Journal 125 (586), F233–F255.

Del Monte, A., Papagni, E., 2001. Public expenditure, corruption, and economic growth: the case of italy. European Journal of Political Economy 17 (1), 1–16.

39

ACCEPTED MANUSCRIPT

Dell, M., June 2015. Trafficking networks and the Mexican drug war. American Economic Review 105 (6), 1738–79. Della Porta, D., Vannucci, A., 1999. Corrupt Exchanges: Actors, Resources, and Mechanisms of Political Corruption. Social Problems and Social Issues. Transaction Publishers. Di Cataldo, M., Mastrorocco, N., 2016. Organised crime, captured politicians and the alloca-

CR IP T

tion of public resources, mimeo.

Di Tella, R., Schargrodsky, E., 2003. The role of wages and auditing during a crackdown on corruption in the city of buenos aires. Journal of Law and Economics 46 (1), 269–92.

Durante, R., Gutierrez, E., 2015. Fighting crime with a little help from my friends: Political alignment, inter-jurisdictional cooperation and crime in mexico. CEPR (Discussion Paper

AN US

#10769).

Fan, C. S., Lin, C., Treisman, D., 2009. Political decentralization and corruption: Evidence from around the world. Journal of Public Economics 93 (1-2), 14–34. Gambetta, D., 1996. The Sicilian Mafia: the business of private protection. Harvard University

M

Press.

Garmann, S., 2015. Elected or appointed? How the nomination scheme of the city manager

ED

influences the effects of government fragmentation. Journal of Urban Economics 86 (C), 26–42.

Gibbons, S., Overman, H. G., 2012. Mostly pointless spatial econometrics? Journal of Regional

PT

Science 52 (2), 172–191.

Golden, M. A., Picci, L., 2005. Proposal for a new measure of corruption, illustrated with

CE

italian data. Economics and Politics 17, 37–75. Hessami, Z., 2014. Political corruption, public procurement, and budget composition: Theory

AC

and evidence from oecd countries. European Journal of Political Economy 34, 372–389.

Isen, A., 2014. Do local government fiscal spillovers exist? evidence from counties, municipalities, and school districts. Journal of Public Economics 110 (C), 57–73.

Knight, B., 2013. State gun policy and cross-state externalities: Evidence from crime gun tracing. American Economic Journal: Economic Policy 5 (4), 200–229.

40

ACCEPTED MANUSCRIPT

Litschig, S., Zamboni, Y., 2011. Audit risk and rent extraction: Evidence from a randomized evaluation in Brazil. Economics working papers 1270. Lyytikäinen, T., 2012. Tax competition among local governments: Evidence from a property tax reform in finland. Journal of Public Economics 96 (7-8), 584–595. Mauro, P., 1995. Corruption and growth. The Quarterly Journal of Economics 110 (3), 681–

CR IP T

712.

Mauro, P., 1998. Corruption and the composition of government expenditure. Journal of Public Economics 69 (2), 263–279.

Ministry of the Interior, 2008. Relazione sull’attivita’ svolta dalla gestione straordinaria dei comuni i cui consigli comunali sono stati sciolti per condizioni (anno 2008). Relazione Par-

AN US

lamentare.

Ministry of the Interior, 2015. Relazione sull’attivita’ svolta dalla gestione straordinaria dei comuni i cui consigli comunali sono stati sciolti per condizioni (anni dal 2010 al 2014). Relazione Parlamentare.

M

Murray, C. K., Frijters, P., 2016. Clean money, dirty system: Connected landowners capture beneficial land rezoning. Journal of Urban Economics 93, 99 – 114.

ED

Olken, B. A., 2007. Monitoring corruption: Evidence from a field experiment in indonesia. Journal of Political Economy 115, 200–249. Oster, E., 2013. Psacalc: Stata module to calculate treatment effects or bounds under propor-

PT

tional selection of observables and unobservables. Statistical Software Components, Boston College Department of Economics.

CE

Oster, E., 2015. Unobservable selection and coefficient stability: Theory and validation, mimeo. Peltzman, S., 2016. State and local fiscal policy and growth at the border. Journal of Urban

AC

Economics 95, 1 – 15.

Pinotti, P., 2015. The economic costs of organised crime: Evidence from Southern Italy. The Economic Journal 125 (586), F203–F232.

Rincke, J., Traxler, C., 2011. Enforcement spillovers. The Review of Economics and Statistics 93 (4), 1224–1234.

41

ACCEPTED MANUSCRIPT

Sah, R. K., 1991. Social osmosis and patterns of crime. Journal of Political Economy 99 (6), 1272–95. Silva, P., 2010. Learning to fear the inspector-general: Measuring spillovers from anti-corrupt policies, mimeo. Solé-Ollé, A., 2006. Expenditure spillovers and fiscal interactions: Empirical evidence from

CR IP T

local governments in spain. Journal of Urban Economics 59 (1).

Tanzi, V., Davoodi, H. R., 2000. Corruption, growth, and public finances. IMF Working

AC

CE

PT

ED

M

AN US

Papers (00/182).

42