Organised Crime, Captured Politicians and the Allocation of Public Resources

Marco Di Cataldo and Nicola Mastrorocco

TEP Working Paper No. 0420

January 2020

Trinity Economics Papers Department of Economics

Organised Crime, Captured Politicians and theAllocation of Public Resources

Marco Di Cataldo∗ Nicola Mastrorocco†

January 27, 2020

Abstract

What is the impact of organised crime on the allocation of public resources and ontax collection? This paper studies the consequences of collusion between members ofcriminal organisations and politicians in Italian local governments. In order to capturethe presence of organised crime, we exploit the staggered enforcement of a national lawallowing the dissolution of a municipal government upon evidence of collusion betweenelected officials and the mafia. We measure the consequences of this collusion by usingnewly collected data on public spending, local taxes and elected politicians at the locallevel. Difference-in-differences estimates reveal that infiltrated local governments spendmore on average for construction and waste management, less for municipal police, andcollect fewer taxes for waste and garbage. In addition, we uncover key elements of localelections associated with mafia-government collusion.

Keywords: Organised crime, Elections, Collusions, Public Spending, Italy.

Jel Classification: K42, H72, D72.

We are particularly grateful to Riccardo Crescenzi, Simon Hix, Valentino Larcinese, Andres Rodriguez-Pose, and Stephane Wolton for theirhelp. We also thank Alberto Alesina, Charles Angelucci, Oriana Bandiera, Paolo Buonanno, Pamela Campa, Micael Castanheira, Cyrus Samii,Decio Coviello, Melissa Dell, Torun Dewan, Paola Giuliano, Jeffrey Frieden, Dominik Hangartner, Asim Khwaja, Horacio Larreguy, John Marshall,Luigi Minale, Henry Overman, Gerard Padro y Miguel, Nicola Persico, Imran Rasul, Livia Schubiger, Shanker Satyanath, Jim Snyder, Olmo Silva,Daniel Sturm, Guido Tabellini, Edoardo Teso, Petra Todd, Janne Tukiainen and all the participants of the seminars at the LSE, Harvard, NYU,Yale, Princeton, Chicago, Bocconi University, Ca’ Foscari University, of the NARSC Conference in Portland, the ERSA Congress in Vienna, theOrganised crime Workshop at Bergamo University, the Applied Economics Workshop at Petralia Sottana, the Northwestern Political EconomyConference, the UEA meeting in Copenhagen, and the AISRe Congress in Bolzano for very helpful insights and suggestions. We are particularlygrateful to Paolo Pinotti for sharing data on mafia homicides in Italian provinces. We also thank the Italian Ministry of Interior and specificallyto Dr. Giancarlo Verde and Dr. Pasquale Recano for precious help in data collection. All mistakes in this paper are our own.

∗London School of Economics (m.di-cataldo@lse.ac.uk)†Trinity College Dublin, Department of Economics (n.mastrorocco@tcd.ie).

1

1 Introduction

Organised crime is detrimental to the efficiency of any democratic or economic system (Gam-betta and Reuter 1995; Pinotti 2015; Acemoglu, Robinson and Santos 2013). Its presencereflects institutional failure and has the potential to influence key aspects of legal economicactivity, ultimately undermining the long run development of any society (Shleifer and Vishny1993; Mauro 1995). Its strength, as well as its influence on the legal economy, relies on thediffused external complicity, i.e. an increasingly close relationship between organised crimegroups and public officials such as national or local politicians and public administrators(Dickie 2005). Thanks to the development of such networks, organised crime has becomehighly pervasive and fully integrated into the everyday socio-economic and political life ofmany countries in the world (Trigilia 2001; Allum and Siebert 2003).

Yet, understanding the extent to which these dynamics condition the choices and ac-tivities of policy-makers is far from easy. What impact does collusion between membersof criminal organisations and politicians have on the allocation of public resources and onthe collection of fiscal revenues? In this paper, we tackle this question by investigating aparticular aspect of organised crime activity: its infiltration within local governments. Suchinfiltration occurs when criminal groups manage to capture local politicians who in turnmanipulate policy decisions in their favour. We study the case of Italy by using a yearlymunicipal-level dataset for the three Italian regions where organised crime is most widespreadand rooted: Calabria, Campania and Sicily.1

In order to measure the presence of organised crime, we exploit the staggered enforce-ment of National Law 164/1991, which allows for the dissolution of a municipal governmentupon evidence of collusion between elected officials and criminal organisations.2 We exploitthe enforcement of this policy to identify cases of criminal infiltration, and compare mu-nicipal governments with and without infiltration before and after such infiltration occurs.Difference-in-differences estimates reveal that the capture of local governments by organisedcrime does not affect the total level of public spending but does have consequences both for

1A focus on southern regions rather than on Italy as a whole has the advantage of restricting the sample toa relatively homogenous area in terms of unobservable elements such as culture or social capital, traditionallyconsidered as highly diversified across this country (Putnam 1993) Municipalities are chosen as the unit ofanalysis because infiltrations often occur at the local level, where the central States control over electoraland legislative processes is weaker (Cantone and Di Feo 2014).

2The enforcement of this law within a given municipality at a specific point in time represents a suddenshock to both the local political establishment and the organised crime group, given that its occurrence andtiming is solely determined at the national level and kept secret until its implementation.

2

the allocation of public resources and the collection of fiscal revenues. In particular, infil-trated local governments modify capital account expenditures in sectors that are strategic tothe interests of organised crime. According to our estimates, infiltration leads to a 13% in-crease in the share of total investment in construction and waste management. This effect iseconomically sizeable since it translates into approximately an additional 45 euros per capitaallocated to this spending component, annually. In addition, infiltration leads to a decreasein the annual share of investment in police force. Moreover, infiltrated municipalities exhibita 17% decrease in revenue inflows from the waste and garbage tax translating into a loss of79 euros per capita in collected revenues on a yearly basis.

We provide evidence of absence of pre-trends between treatment and control municipal-ities and we show that our results are robust to exogeneity checks, changes in specification,placebo tests and to the introduction of a rich set of controls.

Our estimates could pick up some non-mafia related effects (e.g. low quality of politicians,unstable governments) or be determined by political characteristics of the municipal electionscorrelated with infiltrations. We perform a series of further tests, ensuring that our resultsare driven by mafia collusion and not by any of these potentially unobserved components.More specifically, we identify a set of political characteristics of municipal elections thatcould be correlated with the infiltration. Although descriptive, this exercise is noteworthyin that it shows how elections represent the main opportunity for criminal organisations toinfiltrate local government.3 We uncovers several interesting empirical correlations, namely arelationship between infiltrations and elections where (1) there is just one candidate runningfor office, (2) the mayor is running for her second and last term, and (3) the right-wing partywins the election. Using our difference-in-differences setting, we show that none of thesefactors have an impact on public spending or on revenue collection.

We are not the first to empirically study the presence and effect of organised crime.A recent but growing literature has measured the presence and intensity of mafia activityby employing proxies such as the number of mafia-related crimes, murders, and violentattacks (Alesina, Piccolo and Pinotti 2018; Daniele and Marani 2011; Olivieri and Sberna2014; Barone and Narciso 2015), historical or geological indicators (Acemoglu, De Feo andDe Luca 2018; Bandiera 2003; Dimico, Isopi and Olsson 2017; De Feo and De Luca 2017;Buonanno, Vanin, Prarolo and Durante 2015; Buonanno, Vanin and Prarolo 2016), forced

3 Following (Dal Bo, Dal Bo and Di Tella 2006), election can be seen as a recruitment process whereby anew bargaining table between criminal and politicians is established. This might particularly be the case inSouthern Italy where political turnover is very high: 71% of local administrators leave local politics within5 years and 93% within 10 years

3

displacement policies (Scognamiglio 2018) or artificial constructs for counterfactual analysis(Pinotti 2015).

An important strand of these studies has focused on the impact of mafia-governmentlinkages on political and electoral outcomes. For example, Alesina, Piccolo and Pinotti(2018) and Daniele and Dipoppa (2017) investigate how criminal organisations strategicallyuse violence to influence elections and get captured politicians elected. Pinotti and Stanig(2016) exploit as-if random variation in the presence of organised crime in northern Italy,so as to study its impact on the quality of local governance. Other studies have examinedhow criminal organisations choose their political counterparts (Dal Bo, Dal Bo and Di Tella2006; Buonanno et al. 2015), uncovering different strategies. De Feo and De Luca (2017)argue that the mafia sells votes to the party that has more core supporters and it is thereforeexpected to win. Buonanno, Vanin and Prarolo (2016) find a systematic correlation betweenthe strength of Cosa Nostra and the proportion of votes for the main Italian conservativeparty. Most of these studies measure the presence of criminal organisations with proxiesof violence. This is certainly a crucial, distinctive and important behaviour of criminalactivities but, we argue, it only represents the tip of the iceberg of the real strength of theseorganisations. Particularly in developed countries, organised crime has evolved over time,progressively reducing the use of violence and becoming increasingly integrated within theboundaries of democratic society, to the point that mafia activities may no longer even berecognisable as criminal enterprises. While in conflict with the State, criminal organisationsdo not wish to displace the latter but rather to co-exist with it through the creation ofa network based on mutual interests. As a magistrate member of the AntiMafia DistrictDirectorate (DDA) commented,“Today’s mafia no longer kills, no longer makes noise, andthis makes it less identifiable as a criminal group. Our fight against them has therefore neverbeen so difficult”.4 In particular, criminal organisations use violence only as a last resortwhen previous strategies have failed. This implies that we still ignore the consequences ofsuccessful criminal behaviours that did not employ any use of violence. By focusing oncollusion between organised crime and politicians, we aim to shed light on this more silentbut equally dangerous phenomenon and in doing so, assess its impact on economic andpolitical outcomes.

Although there exists a large body of evidence on the distortion effect of corruptionand the quality of governance for government spending (e.g. Tanzi and Davoodi 1997;

4Interview with Giuseppe Borrelli, member of the Naples AntiMafia District Directorate (DirezioneDistrettuale Antimafia), on the Italian television program ’Report’, May 5th, 2016.

4

Mauro 1998; Gupta, de Mello and Sharan 2001; Shleifer and Vishny 1993; Rajkumar andSwaroop 2008; Bandiera, Pratt and Valletti 2009; Gennaioli and Onorato 2010; Covielloand Mariniello 2014; Crescenzi, Di Cataldo and Rodrıguez-Pose 2016), empirical researchinvestigating the rent-seeking behaviour of organised crime is relatively scarce. A notableexception is Barone and Narciso (2015), which argues that the presence of organised crimeaffects the distribution of national public funds to firms. However, the degree to whichthe allocation of public resources is influenced by organised crime remains a puzzle. Ourpaper contributes to this literature by providing the first empirical analysis of the impactof collusion between organised crime and local politicians on public spending, showing thatrather than aiming to affect the overall level of public spending, or engage in patronage byproviding jobs in the public administration, the main objective of illegal organisations is tore-direct resources towards specific investment sectors.

The national law 164/1991 examined here has previously been employed in the empiricalliterature (Acconcia, Corsetti and Simonelli 2014; Daniele and Geys 2015, 2016; Galletta2017).5 Our approach differs, however, from previous studies in that we aim to capture theimpact of organised crime infiltrations within local governments rather than evaluate theeffect of the 1991 law. More specifically, our focus is on the period before the enforcementof the law, i.e. before the dissolution of mafia-infiltrated municipalities took place.

The rest of the paper is organised as follows: section 2 focuses on the institutional settingused as a basis for the difference-in-differences analysis and provides background on organisedcrime infiltrations and political capture; section 3 discusses our identification strategy, thedata and the estimating equation; section 4 presents the main results; section 5 and section6 present respectively important exogeneity tests and a set of robustness tests; section 7concludes.

5Acconcia, Corsetti and Simonelli (2014) exploit temporary contraction in public investment occurringin post-dissolution periods to obtain estimates of the fiscal multiplier for Italian provinces. Daniele and Geys(2015, 2016) provide an assessment of the impact of the 1991 law on different post-dissolution outcomes,such as elected politicians′ levels of education and turnout at local elections. Galletta (2017) empiricallyinvestigates the presence of spillover effects resulting from the strengthening of law 164/1991.

5

2 Institutional setting

“Organised crime has globalised and turned into one of the worlds foremost economic andarmed powers”.6 Its illegal activities and illicit flows affect the entire world, and they gen-erate revenues for $870 billion. In Italy, according to recent estimates, the total combinedannual earnings are about 10.7 billion euros (Figure A1), which generate a turnover of ap-proximately 1.6% of the Italian GDP (Transcrime, 2013). The main sources of revenue areillegal activities such as drug trafficking, extortion and corruption (Figure A2). However, asSchelling would put it, “burglars may operate in the underworld, but they seek to govern thereal world”(Schelling 1971). His words are a perfect description for the evolution of organisedcrime in recent years. Since the 1970, organised crime groups understood the importanceof re-investing a substantial part of their profits within the “legal economy” (Figure A3).They therefore became increasingly sophisticated and their business model shifted from onebased on extortion to one based on entrepreneurship (Le Moglie and Sorrenti 2017). As aconsequence, the nature of the relationship between the mafia and the State also changed:rather than representing an enemy to fight, the government has instead become an oppor-tunity to exploit. Italy has experienced this change prima facie. Since the 80s, the countryexperienced a rise in mafia infiltration and collusion within politics, particularly at the locallevel. This led the central government to introduce a tougher set of anti-mafia measures inthe 1990s. In an effort to tackle cases of collusion between local politicians and members oforganised crime a new law was introduced in 1991, imposing the dissolution of a city councilupon evidence of “mafia infiltration” within the local government; that is, electoral competi-tion contaminated by the mafia and/or policy decisions taken by the government but clearlyrigged by a criminal organisation (D.L. 31/05/1991n.164).7 This criminal strategy can beperpetrated in different ways. It can, for example, occur directly, as in the case of Pompei (inthe province of Naples) where “the speaker of the municipal council has been identified as themain link between the local administration and the local mafia boss, who has also been arrested

6Statement from Antonio Maria Costa, Executive Director of the United Nations Office on Drugs andCrime (UNODC) at the launch of a new UNODC report on The Globalization of Crime: A Transnational or-ganised crime Threat Assessment: https://www.unodc.org/unodc/en/press/releases/2010/June/organized-crime-has-globalized-and-turned-into-a-security-threat.html

7Some of the most common reasons for dissolving a local government under law 194/1991 include: admin-istrators or bureaucrats having an affinity with/kinship relation to members of the criminal organisations orindividuals with recurrent criminal records; permits awarded illegitimately due to bid rigging; severe infringe-ment of building regulations; absence of rigorous inspections in the execution of public works; significantflaws in tax collection; cases of clientelism; illegal elections.

6

in the same investigation”.8 Alternatively, it can be indirect, such as through contaminationof the electoral competition. This was the case in Plati′ (in the province of Reggio Calabria),where “the party winning the electoral competition benefitted from electoral favours from thelocal mafia group, who was able to divert a large number of votes and aimed to maintainpolitical control of the territory”.9 Finally, infiltration can occur simply through the use ofthreats and intimidations. To this regard, Africo (in the province of Reggio Calabria) wasdissolved because “the policy decisions of the municipal council were not made freely andwithout bias because local politicians were repeatedly intimidated and threatened by criminalorganisations”.10 These examples are crucial to clarify just how infiltration is defined: it isnot simply the physical presence of criminal members within the local government, but alsoany direct or indirect link between criminal organisations and politicians.

The Law 164/1991 states that the national government can decree the dissolution ofa municipal government “when evidence emerges regarding direct or indirect links betweenmembers of the local government and criminal organisations [...] jeopardising the free willof the electoral body and the sound functioning of the municipal administration”.11 The dis-solution of a local government requires a number of steps. First, a proposal for dissolutionmust be put forth by the provincial prefect, who has been informed by either magistratesor the police of the risk of infiltration of a municipal government. The prefect then estab-lishes a commission composed of the vice-prefect and officials from different law enforcementbodies (the Polizia di Stato), the military Carabinieri and the Guardia di Finanza). Thecommission investigates the local government′s activity over a period of three to six months,producing a report which the prefect sends to the Ministry of Interior. Any proposal fordissolution signed by the Minister must also be approved by the National government Cabi-net (Council of Ministers - Consiglio dei Ministri) and the President of the Republic beforebeing implemented. Municipalities where the local government is dissolved are thereforethose where the mafia infiltration has been attested to by the Italian judicial system andconfirmed by multiple political institutions. Importantly, infiltrated municipalities are un-aware that they are under investigation, as the process of dissolution is kept fully secret untilits implementation. Once the investigation is concluded, both the members of the criminal

8Official Gazette (Gazzetta Ufficiale) - Decree of the President of the Republic no. 133 of June 2001:http://www.gazzettaufficiale.biz/atti/2001/20010223/01A10530.htm

9Official Gazette (Gazzetta Ufficiale) - Decree of the President of the Republic no. 119 of Marzo 2012:http://www.gazzettaufficiale.biz/atti/2012/20120093/12A04237.htm

10Official Gazette (Gazzetta Ufficiale) - Decree of the President of the Republic:http://www.gazzettaufficiale.biz/atti/2014/20140194/14A06583.htm

11 http://www.gazzettaufficiale.biz/atti/2001/20010223/01A10530.htm

7

organisation and the all the local politicians involved are arrested.

Upon removal of the infiltrated local administration, the central government appointsthree non-elected, external commissioners, who govern the municipality for a period of 12 to24 months. Since non-elected, the commissioners only take care of the ordinary management.They cannot neither approve a new budget nor make any investment (Acconcia, Corsettiand Simonelli 2014; Galletta 2017). At the end of the transition period, regular elections areheld.

As shown in Figure 1, the large majority (and in some years all) of the dissolutionsoccurred in the three regions which form the focus of our study. Figure 2 illustrates thenumber of dissolved municipal governments due to mafia infiltration from the introductionof the law up until 2015. In total, there have been 258 detected cases of mafia infiltrationinto local governments over this period. From 1998 to 2013, our period of analysis, 182municipalities have been dissolved in Campania, Calabria, and Sicily.

Within these three regions, the geographical distribution of dissolution varies significantly.As shown in Figure 1, detected cases of mafia infiltration tend to be clustered in severalspecific areas within these regions. In Campania, the large majority of dissolutions occurredin the north-west, particularly in the provinces of Caserta and Naples - the area wherethe Camorra is traditionally stronger. Similarly, in the region of Calabria most detectedinfiltrations were located in the south, in the provinces of Reggio Calabria and Vibo Valentia,where the ′Ndrangheta is known to be centred. Finally, while dissolutions in Sicily are morewidespread, the majority are concentrated in the province of Palermo, the heart of CosaNostra.

3 Empirical strategy

3.1 Identification strategy

Our identification strategy is based on a difference-in-differences (DiD) design that exploitsthe time and geographical variation of dissolutions over time. We rely on law 164/1991to identify cases of mafia infiltration within local governments of the municipalities in oursample regions. That is, we use the dissolution of a municipal government to identify ourtreatment period. A timeline helps to clarify our approach: Figure 3 shows the case ofthe municipality of Casoria, in the province of Naples (Campania), which has held local

8

elections in 2002. The elected government was later dissolved at the end of 2005. After that,commissioners sent from Rome took over until the following elections, at the beginning of2008. Our treatment period ranges from the election in 2002 to the dissolution in 2005. Wewant to identify the period of time during which organised crime was plausibly colluded withthe local government.

The years from 2005 to 2008 are always dropped from the sample because during thisperiod no investment can be made since non-elected commissioners can only handle theordinary management until new elections are held (Acconcia, Corsetti and Simonelli 2014).

The definition of the control group is crucial in this type of design design (Borusyakand Jaravel 2017). We employ two different approaches. In the first, our control groupis composed by never-dissolved municipalities and by municipalities having experienced adissolution, before the infiltration started and after it was ended. Going back to the exampleof Figure 3, all years before 2002 and after 2008 make up the control period. However, localgovernments formed after a dissolution may be different precisely because of the enforcementof the policy. Hence, including the years after 2008 in the previous example might biasour results. To take this into account we propose a second approach where the controlperiod is composed by municipalities that have never been dissolved and, for the dissolvedgovernments, only to the years before the election, i.e. excluding from the control group allthe years after the dissolution. With reference to Figure 3, the years that will be part of thecontrol groups are only before 2002 (when our treatment begins).

Importantly, some municipalities have been dissolved multiple times. We drop from thesample all the municipalities dissolved before 1998. This is to eliminate any post-treatmenteffect that might bias our results.12

Finally, unlike classic DiD strategies, our treatment switches on in different points intimes. We can exploit this time variation to restrict the sample to only those municipalitiesthat have experienced a dissolution due to criminal infiltration. Performing this samplerestriction is important because, as seen in Figure 1, the geography of dissolutions reflectssignificant concentrations in specific provinces of the sample regions. The figure indicatesthat the intensity of mafia activities in some territories is lower with respect to the core areaswhere the criminal organisations are primarily based.

Threats to identification. Organised crime infiltration within local governments is not12There are also 10 municipalities that experienced two dissolutions within our period of analysis, 1998-

2013. We show in the empirical analysis that our results are robust to their inclusion or exclusion in theestimating sample.

9

random. As a consequence, there are some potential concerns relative to our identificationstrategy. First, the application of law 164/1991 may be imperfect. Some municipalitiescould have been infiltrated but not dissolved because judicial authorities did not detect thecollusion. Similarly, some dissolutions may have been done erroneously if there was no realinfiltration.13 These issues should not represent a concern for our estimation strategy asinfiltrated municipal governments that are not dissolved would belong entirely to the controlgroup, causing attenuation bias in the empirical results.14

Second, our measure of infiltration might be correlated with our outcome variable, i.epublic spending. This would generate a problem of simultaneous causality: criminal groupsdecide to infiltrate in municipalities that exhibit a particular spending pattern. To tacklethis concern, Section 5 shows that infiltrated and non-infiltrated government were on aparallel trend with respect to the outcome variables of interest. An additional concern forour analysis is that judicial investigators might start their investigations precisely in thosemunicipalities that present anomalies on their balance sheets. In this case, selection intotreatment (i.e. being infiltrated and later dissolved by the national government) would becorrelated with the outcome variables (i.e. public spending and revenue collection) creatingan endogeneity problem. We tackle this issue in Section 5 by showing that our results donot change when we exclude from the sample those municipalities for which the main motivedriving the investigation and dissolution was related to either public spending or revenuecollection.

Another potential issue with our empirical setting is that it assumes infiltration as binaryobjects, that switches on during the election and off when the government is dissolved. Whileour hypothesis may be true for many infiltrated municipalities where electoral manipulationbrought to power local governments subject to the conditioning of the mafia from the verymoment they took office, it may not hold for all of them. Criminal infiltrations can becontinuous in time and, as a consequence, be already present in some form before the elec-tions.15 To empirically test for this we investigate whether we find any effect on our outcome

13Periods of erroneously detected infiltration would instead belong to the treated years, again biasing theestimated impact of infiltrations towards zero. This means that the point estimate of regression coefficientsis likely to be larger (in absolute value) than the one observed.

14This is a relevant point for the interpretation of our results. The likelihood of having some infiltratedmunicipalities that have not been caught might suggest that we may be measuring the effect of the ′low-skilled′ mafia organisations, i.e. those that have been detected. We cannot discard this interpretation, whichis indeed a limitation of our measure. However, conditional on heterogeneous treatment effects, we arguethat the estimated coefficients indicating the impact of infiltrations represent a lower bound of the true effect.

15 Statistically, this should generate a contamination in the control group that would downward bias ourestimates.

10

variables in the years preceding the elections.16 We deal with this issue by introducing a fullset of lags in section 5.

A final potential concern for our estimates could arise if the dissolution of municipalgovernments has been manipulated politically. This distorted use of law 164/1991 is, how-ever, unlikely to happen for several reasons. First, the dissolution process is initiated andcarried forward by the Italian Anti-Mafia Investigation Directorate (Direzione InvestigativaAntimafia), one of the most efficient investigative bodies of the Italian State.17 This is anorganisation composed of highly-trained and specialised individuals from the three main po-lices forces (Polizia di Stato, Carabinieri and the Guardia di Finanza), whose experience isoften valued and requested by other countries and institutions needing consults on the fightagainst organised crime.18

In addition, the multiplicity of actors involved in the dissolution decision, from nationalMPs to the Minister and the Cabinet to the President of the Republic, makes any formof manipulation of the law improbable. However, in order to provide as much evidence aspossible on this point, we perform a test to attenuate the concern of systematic politicalmanipulations. If dissolutions were manipulated, we would expect to observe that the polit-ical colour of provincial and national governments is significantly associated to the politicalcolour of dissolved municipal governments. As shown in Tables A1 and A2, which refer tothe restricted sample of dissolved municipalities in the 1998- 2013 period, there is no statis-tically significant correlation between the colour of national or provincial governments andthat of municipal governments. Indeed, given the political cost generated by a dissolutionfor the national government - i.e., high national media coverage and political competitors

16It is equally important to check whether the infiltration started after the elections. The official docu-ments motivating dissolutions (published by the Ministry of Interior) explicitly indicate the election of thedissolved governments as the key moment when the infiltrations take place. Additionally, by law, munici-palities have to decide the budget allocation for major investment projects within the first year of each newlegislature. Hence, criminal organisations have the possibility to condition such political decision only ifalready infiltrated from the first year of the new government.

17The Anti-Mafia Investigation Department (DIA) was founded in 1991, under the authority of the Minis-ter of Interior and the coordination of the National Anti-Mafia Directorate (Direzione Nazionale Antimafia).The DIA′s operations include pre-emptive investigations and judicial investigations. They study the charac-teristics, objectives, and methods of organised crime as well as examine the latter′s domestic and internationaloperations.

18For example, the Italian Prosecutor Antonio Ingroia, who extensively investigated the Sicilian mafia,was appointed Director of the United Nation-backed judicial watchdog in Guatemala, the InternationalCommission against Impunity in Guatemala (CICIG). Another example is judge Giovanni Falcone, whoin 1989 established the Anti-Terrorism Unit in Quantico, Virginia. It is unlikely that these professionals,together with other judges and prosecutors, would allow their investigations to be strategically used bypoliticians.

11

exploiting the latter by asking for the government′s resignation - it is unlikely that the na-tional government would strategically choose to dissolve municipal governments governed byopposing parties. It is important to note, however, that even if this was true, our estimatesshould be downward biased, since strategically manipulated dissolutions would be coded astreated and cause an attenuation bias of coefficients.

3.2 Data

Local public spending. Our primary data source is the Italian Ministry of Interior′sFinancial Statement Certificates (Certificati Consuntivi) database, which contains yearlystatistics on the public finances of Italian municipalities for a number of different spendingcategories. The full dataset is disaggregated into capital account and current account expen-ditures. These are further disaggregated into six specific spending categories. These differentcategories reflect the services and functions to which the resources have been allocated andspent and include: general administrative functions, social sectors, construction and wastemanagement, transportation, public education and municipal police. Capital expenditureand current expenditure are further sub-divided into three spending sections: spending deci-sions, year-over-year spending and residuals. Spending decisions correspond to the amountof financial resources a municipality plans to spend over the course of the following year,determined at the end of the current year. Year-over-year spending refers to that which themunicipal government has actually spent, calculated at the end of the year. Residuals con-sist of the resources that have not been spent. Throughout our analysis, we adopt spendingdecisions as a spending proxy, given that data on residuals and year-over year spending ismuch more fragmented, less reliable and less homogeneous19. This dataset is available forthe 1998-2013 time period.

Table 1 and Appendix A2 illustrate average per capita spending for the municipalities inour sample over the 1998-2013 period. The resources spent by the municipalities amount toa yearly average of 543 euros per inhabitant for the capital account (i.e. investments) anda yearly per capita average of 731 euros for the current account (i.e. salaries and services).Summing these two figures we obtain the average total spending per municipality, 1,274

19In addition, quite often year-over-year spending includes expenditures planned by previous governments,while our intention is to capture the conditioning role of mafia infiltrations on policy decisions taken specif-ically by the infiltrated governments. On the contrary, this is not the case with ’spending decisions’ which,by law, cannot contain allocated budget from previous years. Additional details on the local public financesystem of Italian municipalities in Appendix A2.

12

euros per inhabitant. As shown in Table 1, the spending function to which the most annualresources are allocated is construction and management, which makes up 34% of the annualcapital account budget. As for the current account, spending is highest for administration,followed by construction and waste management. The municipalities are also responsible fortendering and awarding public procurement contracts to the contractor company in chargeof carrying out the work.

Infiltrated municipalities. In order to measure the infiltration of organised crimewithin local governments, we identified all municipalities that experienced government dis-solution due to mafia infiltration from 1991 to 2013, exploiting information on the date of thedissolution available from the Ministry of the Interior. The treatment variable was created asa dummy taking value 1 from the year of the last regular election before the dissolution untilthe moment in which the municipal government was dissolved, and zero otherwise. Data onthe date of local elections before dissolutions were obtained from the Historical Archive ofLocal Elections, publicly available from the Italian Ministry of the Interior.

Control variables. A number of municipal level time-varying characteristics were ob-tained from the ISTAT Censuses including municipal population, unemployment rate, per-centage of industry employment and percentage of tertiary education degree holders. De-scriptive statistics for these variables are in Appendix A4.

Criminal violence. We collected the data from Avviso Pubblico, an Italian NGO thatsystematically record news of attacks to local Italian politicians. On a daily basis, volunteersfrom AvvisoPubblico consult and register news of attacks on Italian politicians and publicofficers at all levels of government that appear in national or local newspapers or that arecommunicated to the NGO directly through first-hand sources. Our dataset covers Calabria,Campania and Sicily for the years 2010 to 2015.

Electoral results. To study the relationship between criminal organisations and localpolitical parties we have collected data on municipal election results from 1998 to 2013. Thedata source is the Archivio Storico delle Elezioni of the Ministry of Interior.

3.3 Estimating equation

We exploit a difference-in-differences setting to test whether mafia infiltrations have an im-pact on the public spending allocations of local governments in Campania, Calabria andSicily. To this end, we compare municipal governments with and without infiltration beforeand after such infiltration is ended by the national government through the application of

13

law 164/1991.

We estimate various versions of the following model:

Ym,c,t+1 = α + β Infiltrationm,t + δ Xm,t + ϕm + τt + εm,c,t (1)

Where Ym,c,t+1 refers to public spending in municipality m at time t+ 1.20

Ym,c,t+1 is Sc,m,t+1∑cSm,t+1

, i.e. the spending allocated to component c as a share of the totalspending committed to the next financial year. Total spending is calculated per capita.21

The key variable in the model is Infiltrationm,t, a dummy taking value one if the localgovernment is infiltrated by the criminal organisation and zero otherwise. The coefficient ofinterest is β which captures the impact of the infiltration at time t on the public spendingallocation at time t+ 1.

Vector Xm,t denotes a set of control variables. This includes socio-economic and demo-graphic characteristics of municipalities in the sample regions.

In the empirical analysis we perform two types of estimations: one with the full sampleof municipalities of Calabria, Campania and Sicily, and one with the restricted sample ofmunicipalities having experienced a dissolution (dissolved sample). The latter specificationis our preferred one. Adopting this restricted sample allows us to focus on more similarmunicipalities and minimise any unobserved heterogeneity.

The data are drawn from the 1991, 2001 and 2011 ISTAT Censuses interpolated overtime.

The model is completed by municipality dummy variables, controlling for time-invariantunobservables correlated with the timing of the infiltration (ϕm), and time fixed effects, con-trolling for year-specific shocks (τt). Finally, εm,t is an idiosyncratic error term. Throughoutthe empirical analysis we cluster standard errors at the municipal level.

20The time lead derives from the fact that our dependent variable is based on spending decisions definedat the end of the financial year. This reduces issues of reverse causation as our main variable of interest ismeasured at time t.

21∑

cSm,t+1

popm,tis the total per capita spending allocated by a municipal government

14

4 Estimation results

4.1 Does the infiltration of organised crime affect the overall levelof public spending?

We begin by presenting the estimates of the effect of mafia infiltration on total municipalspending (Table 2). Panel A presents our main specification, comparing municipal govern-ments with and without infiltration before and after such infiltration occurs. As explainedabove, the control group is composed by municipalities that have never been dissolved and bydissolved municipalities before and after the infiltration is terminated by the national govern-ment. Panel B presents an alternative specification, where we exclude from the control groupnot only the commissioning years (as in Panel A), but also all the all post-commissioningyears of dissolved municipalities. In columns 1 and 2 we focus our attention on total spendingper capita. The model is initially estimated for the full sample of 1,350 municipalities fromCalabria, Campania and Sicily (column 1). In column 2 we restrict the sample to a group ofmore homogeneous municipalities - those 182 having experienced a government dissolutionfor mafia infiltration. In the following columns, we sub-divide total overall spending intototal capital account spending per capita (columns 3-4) and total current account spendingper capita (columns 5-6). All estimations include municipality fixed effects, year fixed effectsand control variables. Outcome variables are expressed in logarithm.

Throughout all the different specifications, the coefficients of the infiltration dummyare not statistically significant. Hence, the results provide evidence that, other things equal,infiltration periods tend not to be associated with significant variations in the total amount oflocal government expenditures, either for public investments (capital account) or for servicesand maintenance (current account).22

One interpretation of our results is that when mafia groups infiltrate local governments,they are not interested in affecting the overall aggregate spending.23 Indeed, if municipalgovernments were running constant budget deficits during infiltration periods, they wouldrisk being taken over by the central government for reasons of financial instability, thus leav-

22As discussed in Section 3, while these are the observed results, we acknowledge that a number ofmeasurement issues in the infiltration dummy may bias our estimates downward.

23Our findings differ from those of Olivieri and Sberna (2014), who report a positive relationship betweenpre-electoral mafia violence and total public investment in local municipalities in Southern Italy. The differ-ence may be due to the fact that we do not focus on violent attacks on the part of organised crime, but oncriminal infiltration within politics.

15

ing the criminal organisation without reliable political connections within the local council.24

Rather, a way to coercively condition the public financing of infiltrated governments may beto modify their investment policies precisely in those sectors that are strategic to protectingthe interests of organised crime. One might thus ask, when infiltrating local governments,does the mafia engage in patronage behaviours? Or does it bias the allocation of resourcestoward specific spending components? In an effort to answer these questions, we break downtotal spending into different items of expenditure.

4.2 Does the infiltration of organised crime affect specific spend-ing components?

In this section, we test whether criminal infiltrations significantly affect the allocation ofpublic resources. In Tables 3 and 4, we compare each of the spending items of infiltratedgovernments with those of non-infiltrated governments before and after this infiltration isended by the national government. Table 3 reports estimates for capital expenditure com-ponents, while Table 4 focuses on current expenditure chapters. Each spending item ismeasured as a share of the total annual spending. As explained in subsection 3.1, and simi-larly to Table 2, we present estimates for our main specification (Panel A) and also estimateswhen excluding post dissolution years from the control group (Panel B).

For each spending category, the model is estimated both for the full sample of municipal-ities and for the restricted sample of municipalities who have had their government dissolvedat least once. Most of the current spending components (Table 4) display insignificant co-efficients. Particularly interesting is the administration spending component. If organisedcrime had invested in patronage behaviour, thus inflating the hiring of public administrationpersonnel, the coefficient would have been positive and significant (Gambetta 1993). We donot observe this effect. The only significant effect is on municipal police.

When we turn our attention to capital spending (Table 3), i.e. investments, we find thaton average infiltrated municipalities spend more on construction and waste management(columns (5) - (6)) and less on municipal police (columns (11) - (12)). Panel B reportsalternative estimates excluding all post-commissioning years from the control group. All theprevious findings are confirmed. In addition, when focusing on dissolved municipalities, the

24Article 244 of the Unified Text Governing Local Authorities (Testo Unico Enti Locali - TUEL) foreseesthe possibility of declaring a municipalities as financially unstable (dissesto finanziario) when they are unableto provide basic functions, services and public goods.

16

coefficient for transport and public lighting is negative and (weakly) significant. Importantly,in Appendix A3 we replicate these estimates to study how the coefficients change when wegradually introduce a number of controls, fixed effects and linear time trends.

With any testing of a null hypothesis, a false-positive result could occur. This is partic-ularly true when performing a number of hypothesis tests with multiple comparisons. Wecontrol for this in two ways. First, in Table A9, we decrease the number of tests that werun by aggregating together capital and current expenditure for each spending componentand expressing it as share of total spending. All the results are confirmed. Secondly, we fitseemingly unrelated regression model to test the joint significance of our parameter of inter-est. Focusing on the restricted sample of dissolved municipalities, we report the estimates inTable A10. Reassuringly, both the coefficients for construction and waste management andpolice remain significant and stable in magnitude.

Interestingly, a first look at these results indicates that organised crime groups infiltratelocal governments to bias the allocation of public resources towards sectors which are ofstrategic interest for them. We provide a more comprehensive interpretation of these resultsbelow.

Construction and waste management. According to the estimates in Table 3, infil-trated governments increase investment spending on construction and waste management.With an estimated coefficient of 0.0503 (Column 6, Table 3), the effect is economically signif-icant. Infiltrated municipalities increase spending on constructions and waste managementby 13% compared to the average spending on construction and waste management in non-treated municipalities (equal to 0.34). This is a relatively large figure if we consider thatfunctions related to construction and waste management already account for the largest partof the capital account budget (Table 1).

Therefore, any additional resources these governments allocate to this sector during aperiod of infiltration can be substantial. In per capita terms, given an average yearly totalspending of 346 euro per capita, infiltrated municipalities redirect an additional 45 eurosto construction and waste management (Table 1). This particular spending item includesall expenses for waste collection and the construction of new buildings, bridges, streets andhighways. Investing in this sector represents a strategic decision for criminal organisationsfor many important reasons.

First, criminal groups need to find an outlet for profits obtained from illegal activities and24At first look, our equations seem unrelated, but they might be related through the correlation in the

errors.

17

the construction sector represents an easy and highly profitable option for money laundering.This is not just because of the financial returns that this sector might generate, but alsobecause technological and financial barriers to entry are relatively low, making this an idealarea for long-term investment.

Second, the area of construction and waste management is associated with a set of activ-ities that are deeply embedded within the local territory. Seizing control of these activities iscrucial for the mafia, so as to establish and expand a wide network of relationships, allowingthe latter not only to survive, but to prosper. The construction of new buildings and thecollection of waste involve many agents: political leaders in charge of awarding public worktenders, contractor enterprises responsible for delivering the project, and a labour pool tocarry out the work. Members of organised crime groups may be involved at all levels of thischain, and very similarly to the most traditional interest groups, they exploit the politicalconnections they have in order to rig public work bids to the advantage of the enterprisesthey control, or intend to favour. This is indeed what emerges from the message of a Camorralead figure to the Mayor of San Cipriano d′Aversa (August 2012): “I need the Mayor andthe City Council to approve more public constructions projects and I need them to let meknow in advance so that I can alert my firms.... Tell them that, in order to avoid misunder-standing [...], if there is any other public work they will let me know immediately. Right?Bye”. Mafias invest in local firms and, in order to generate profits, they need investmentopportunities and public funding. To provide a first descriptive evidence of this, we havecollected unique information on the municipal level location of firms connected to criminalorganisations and on their disaggregation by sector. Table 6 exhibits the correlation betweendissolved municipalities and the presence of a firm seized because of connection with criminalorganisations. The coefficient is positive, statistically significant and remains stable when wecontrol for mafia-related violence and for a full battery of socio-demographic controls at themunicipal level. Figure 4 breaks down mafia firms by their business sector. The majority ofthem operate in the sector where infiltrated municipalities re-direct resources: constructionand waste management. This exercise might represent evidence of a precise modus operandi:control over local politicians facilitates the capture of public procurement contracts, in turnenabling criminal organisations to provide business opportunities to their controlled firms aswell as reinvest liquidity generated from illicit activities and, more broadly, strengthen theircontrol over the local territory by offering employment opportunities and cheap services.

Municipal police. The second significant variation in the local public finances of infil-trated governments is spending on municipal police. A significant decrease is seen both for

18

the capital account and for the current account spending in this sector. Our estimates in Ta-ble 3 report an annual reduction in the share of total capital account spending for municipalpolice. While this might seem like a low figure, it should be compared to the average shareof investment in local police forces made by municipal governments in our sample. As shownin Table 1, the proportion of capital account resources that local governments allocate tothis sector is about 0.3% of the total for the full sample of municipalities, and 0.7% for themunicipalities who had their government dissolved at least once. Therefore, an estimatedcoefficient of 0.003 (column 12), represents a considerable change, equal approximately toa annual reduction of 64% compared to the baseline mean. In practice, given that policeexpenditures are typically very low, they are thus nearly absent in infiltration years.

According to the estimates in Table 4, infiltrations also lead to a significant reduction inspending on municipal police as part of the current expenditure. This corresponds, however,to a less radical change in budget decisions with respect to that reported for capital account,given the average share of current account expenditures allocated to municipal police (Table1). Nevertheless, this result loses significance as we exclude post-dissolution years in Table4, Panel B.

If we add up the current and the capital account effects, a clear pattern emerges indicatingthat infiltrated governments tend to refrain from making expenditures on local police forces.A reduction of resources directed towards law enforcement bodies such as the municipalpolice may directly benefit the criminal organisations, facilitating their illegal activities.Indeed, the local police are responsible for maintaining public order and security, a taskshared with the national police (Polizia di Stato) and low-quality equipment may imply alesser ability to fight crimes such as drug trafficking, usury and murders. Perhaps mostimportantly, local police are also responsible for so-called ′administrative police′ functions,including surveillance over construction works and abidance with building regulations. Giventhat a lack of compliance with building regulations is one of the most frequent motivationsfor government dissolutions, allocating fewer resources to municipal police forces may alsobe one of the ways in which corrupt local politicians attempt to prevent dissolutions.

Public transport and lighting. When excluding the post dissolution years from thecontrol group, we find a weakly significant reduction in capital account spending in publictransport and lighting (Column 8, Panel B). Investment in this chapter of the balance sheetincludes expenditures for local public transportation (e.g. school buses), public lighting, aswell as any improvements in the management of road traffic. The coefficient is significantat 10% level in our preferred sample, only focusing on municipalities having been dissolved

19

(Table 3, column (8)). Reducing investment in this municipal budget item may be beneficialfor captured politicians as it may help achieving two distinct but related goals. First, sincethese public goods have little (if any) direct connection with the main activities of criminalorganisations, they represent an ideal set of available resources for them to re-direct towardsother spending sectors without running a budget deficit at the yearly level. Second, areduction in capital investment in public transport and lighting may not be perceived bythe municipal residents in the short term. Hence, this would avoid the risk of electoralbacklashes. However, the result is barely significant and it should be regarded with caution.

4.3 Does the infiltration of organised crime affect local revenuecollection?

We now turn to whether infiltration also has an impact on the ability of the local governmentsto collect fiscal revenues. Given the quasi-federal structure of the Italian State, municipalitiesare expected to maintain a certain level of independence and autonomy in collecting theirown financial resources. Hence, local taxes represent an important source of income formunicipalities.25

In order to assess the performance of municipal governments, we follow Drago, Galbiatiand Sobbrio (2014), constructing a measure of efficiency in revenues collection calculated asthe ratio between collected revenues and the total amount of forecasted revenues that themunicipality should collect within the budget year. We focus on the two main local taxes, i.e.property tax and waste tax, and on total taxes and total collected revenues.26 Property tax andwaste tax are the main source of income in the municipal budget representing respectivelythe 35% and 22% of the total budget.

Exploiting our difference-in-differences setting, we present our analysis in Table 5. Theestimation includes municipal fixed effects, time fixed effect and a wide range of controlvariables. The table is subdivided into Panel A, reporting our main specification, and PanelB, excluding post-commissioning years. For each outcome variable, we estimate our modelwith the full and the restricted sample. The coefficient on waste tax (columns (3) and (4)) isnegative and significant. The effect is economically sizeable. According to our estimates, in-

25Local revenues correspond on average to 52% of the entire budget for Italian municipalities (IFEL,2014).

26Our data come from Certificates on Financial Statements (Certificati Consuntivi - quadro 2). Thisanalysis is based on a panel dataset beginning in 1999 and ending in 2012. Total taxes represent the totalfiscal inflows for a municipality. Total revenues also include transfers from the National Government.

20

filtrated municipalities collect 17% less taxes on waste and garbage compared to the averageof non-treated municipalities (baseline average is 0.14), translating into a loss of approxi-mately 79 euros per capita every year. The result is stable to the inclusion of our set ofcontrols and to the restriction of the sample. Panel B confirms these findings. The interpre-tation of this result is twofold. First, the direct or indirect presence of criminal organisationswithin the municipal government has an impact on the performance of the local government.Indeed, tax evasion generates significant losses and distortions in government revenues; theability to efficiently enforce tax collection is one of the fundamental components of statecapacity (Casaburi and Troiano 2016). Waste tax represents 22% of the municipal budget(total revenues are on average 2.8 million euros per year). Second, lower fiscal revenues mightcorrespond to a precise strategy on the part of criminal organisations who aim to weakenthe presence and reputation of the State in order to open up the possibility of substitutingit through a system of provision of private favours (Trocchia 2009). Together with the ev-idence on spending on construction and waste management uncovered in section 4.2, thisresult seems to confirm the well-known presence of criminal organisations within the wastemanagement sector.27 The reduction in waste taxes may determine a general reduction infiscal revenues of municipalities. Indeed, Panel B, column (6) reports a negative significanteffect of infiltrations on total fiscal revenues collection.

5 Exogeneity checks

In this section, we present a set of tests of the robustness of our design and our estimates.

Infiltration period starts with the elections. As discussed in section 3, the startingassumption of our identification strategy is that the period of infiltration begins at themoment of the election of later-dissolved governments and ends with the dissolution. Wetest the validity of this assumption in Table 7, where we perform a placebo experiment. Ifsignificant variation in both public investments and revenue collection starts in the periodpreceding infiltration, the decision to infiltrate a government might be taken as a result of this

27The connection between the waste hauling industry and organised crime dates back decades. In the U.S.,Cosa Nostra has been part of New York′s commercial sanitation system since at least the 1950s (personaltrash is hauled by the city Department of Sanitation). “Carters”, or trash haulers, have always been able tocarve out and sell routes to one another, making the system vulnerable to strong-arm tactics. The Camorrais said to have controlled garbage in the city of Naples since the early 1980s. The poorly run system attractedworldwide attention when, back in 2008, uncollected garbage piled up on the city′s streets for more than twoweeks because the Mafia had closed the dumps.

21

variation. This would occur if the criminal organisation were selecting municipalities whereto extract rents on the basis of pre-determined variations in public expenditures or localtaxes, made by governments with no links with organised crime. If this is the case, publicspending decisions are the cause, not the consequence, of organised crime infiltrations. Ourplacebo test verifies the spending behaviour and the revenue collection of those governmentslater dissolved for mafia infiltration. In Table 7, for each of our outcome variables, column1 reports the results of our full model as expressed in Tables 3 and 5 in the main sectionof the paper. In column 2, we drop the infiltration dummy and introduce a variableequal to 1 in the year before the election of a later dissolved government and 0 otherwise.We expect to find no significant correlation between pre-infiltration governments and anyform of public spending or revenue collection distortion. The infiltration coefficients incolumn 2 are indeed not significant for our 2 spending categories and waste tax. However,given that the average period of infiltration is longer than one year we might worry thatthe estimates are more precisely estimated for the period after infiltration compared to theperiod before. This would undermine the results of column 2. To overcome this, in column3 we define a pre-infiltration period comparable in length to the actual infiltration period.Prior Legislature is a dummy variable equal to 1 for the legislature preceding the infiltrationperiod. Reassuringly, none of the coefficients is significant in neither our estimation with thefull sample nor with the dissolved sample.

Although we cannot reject with full certainty the possibility that infiltrations begin be-fore elections, the results of our placebo test seem to follow the theoretical framework ofDal Bo, Dal Bo and Di Tella (2006), according to which elections constitute a “recruitmentprocess” whereby a new bargaining table between crime and politics is established. Thismight particularly be true in Southern Italy where the political turnover is very high: 71%of local administrators leave local politics within 5 years and 93% within 10 years (Danieleand Geys 2015). In this context, elections are crucial because they can constitute a turningpoint whereby the “criminal interest groups” select the political counter-parties that bestsuit their interests Wolton (2016). Hence, the striking difference in all the coefficients fromcolumn 1 to columns 2-3 in Table 7 might be explained as a newly renovated agreementbetween mafia members and politicians which in turn leads to a distortion in the allocationof public resources and revenue collection.

Parallel trend - full dynamic specification. The difference-in-differences model lendsitself to a test of causality in the spirit of Granger (Angrist and Pischke 2008), which allowsto verify the parallel trend assumption and to analyse the year-by-year evolution of local

22

spending and revenues collection during infiltration years. In this context, Granger causalitytesting means performing an event study to check whether there is any statistically significantdifference between infiltrated and non-infiltrated municipalities before the infiltration takesplace. In order to do this, a set of dummy variables is created for each and every year of thetreatment period, i.e. the period from the governments′ election to their dissolution. Similardummy variables are also constructed for pre-treatment years.

Formally, we estimate the following equation:

Ym,t+1 = ϕm + τt +ρ∑

τ=0δ−τInfiltrationm,t−τ +

q∑τ=1

δ+τDm,t+τ +Xm,tβ + εm,t (2)

Where q represents the post-infiltration effect and ρ represents the anticipatory effect.28

We have re-estimated the model for the main dependent variables (capital account spendingfor construction and waste management, municipal police and waste tax) by including thisset of leads and lags, again controlling for municipal fixed effect, time fixed effects and socio-demographic characteristics. Following Borusyak, Hull and Jaravel (2018), we perform thetest exploiting the full sample. We exclude post-dissolution years (including the commis-sioning period) from the control group. The omitted category is the year before the electionof a later-dissolved government (1y prior).

Figure 5 displays the result of the analysis for both public spending (construction andwaste management and police) and for revenues collection (waste tax). We assess the evolu-tion of municipal spending up to 5 years before the election of an infiltrated government and2 years after the election. Each point in the Figure refers to the estimated coefficient fora given year. We limited the post period at 2 years after the elections because the averageinfiltration period is 2.7 years. However, in Figure A4, we replicate the analysis expandingthe post period to 4 years after the elections.29

Importantly, for all our results, the estimates reveal no statistical difference in the pre-infiltration trends between control and treatment group. In all the three plots, the 5 pre-treatment years show that, before the infiltrated governments, there is very limited and notsignificant variation in either the share of public investments (in construction and waste man-agement and police) or in the collection of fiscal resources. Hence, there is no evidence that

28All municipalities with more than one infiltrated government in the 1998-2013 period have been excludedfrom the sample for this test.

29However, this test is less rigorous and reliable because t + 3 and t + 4 are identified by very fewmunicipalities.

23

the significant change in the proportion of investments and revenues precedes the election ofan infiltrated government.

Figure 5 shows a significant change in spending for construction in the first year aftermunicipal elections (i.e. second year of legislature). This is likely due to the fact thatthe second budget year is the one in which a municipal government can promote three-year construction works and hope to see the result of this investment while still in office.Three-year plans concern all public works worth more than 100,000 Euros and these medium-term investment initiatives are potentially very appealing to the mafia due to their highermonetary value as compared to single-year plans.

Test for selection into treatment correlated with outcome variable. Our resultsindicate that infiltrated local governments spend on average more on construction and wastemanagement and less on municipal police and transport. One concern, however, is thatjudicial investigators might choose to investigate precisely those municipalities that presentanomalies in their balance sheets. If this is the case, selection into treatment would becorrelated with the dependent variable, and there would therefore be bias.

In order to tackle this issue, we reproduced our analysis excluding from the sample allthose municipalities for which the main reason for dissolution was related to distortion in theallocation of resources.30 Table A.11 reports the results, showing that reductions in policeinvestment and the increases in construction and waste management remain significant andstable. Columns 1 and 3 provide the point estimates for capital spending in police, publictransport and construction. All the coefficients remain significant and similar in magnitude.Hence, these estimates suggest that our results are not driven by a bias in the selection intotreatment.

6 Robustness checks

In the previous section we provided some fundamental tests for our identification strategy. Inthis section we explores some alternative mechanisms that might affect our empirical results.

Placebo test: organised crime - unrelated dissolutions. One concern related to30To perform this test, we exploit official statements on the dissolutions. These documents contain precise

descriptions not only of the final reason for the dissolution, but also why the investigation started in the veryfirst place. We exclude from our sample all the municipalities for which a) the investigation started and/orb) the reason for the dissolution was due to spending related distortions. In doing so we excluded 14% ofthe sample.

24

the changes in the public spending of infiltrated governments is that, rather than beingcaused by the mafia, they might be driven by some inherent characteristics of dissolvedlocal governments. These may include the degree of political instability, or the qualityof politicians governing these local councils. In order to test for this, we exploit the factthat in Italy local governments can be dissolved for reasons unrelated to mafia infiltrations,including: failure to approve the financial budget, resignation of the mayor, resignation ofmore than 50% of the council members, vote of no confidence. These dissolutions are infact relatively common in our sample and time-span - in the period from 1998 to 2013 therewere 463 cases of municipal government dissolutions unrelated to the mafia within the threeRegions of analysis. We use these dissolutions as proxies for unstable governments and for lowquality of elected politicians, replicating the estimates of model (1) using Mafia-unrelateddissolutionm,t as the main explanatory variable, a dummy taking value 1 for all years inwhich governments dissolved for mafia-unrelated reasons were ruling the municipalities.31 Ifthe results in section 4 were driven by local government characteristics unrelated to the mafia- rather than by infiltrations - we would expect to obtain similar effects as those presentedabove.

The results of this placebo test are presented in the Appendix, from Table A12 to TableA15. We exclude all infiltrated governments and compare dissolved governments for mafia-unrelated reasons with other governments, before and after the dissolution takes place. We doso using the entire sample of municipalities from Calabria, Campania and Sicily from 1998to 2013, controlling for time and municipality fixed effects, and all other controls. TableA13 includes our main results as outcome variables. There are no statistically significantcoefficients, suggesting that the observed differences between infiltrated and non-infiltratedgovernments is truly produced by the presence of the mafia.

Test for spillover effects. Previous research has demonstrated that municipalitiesneighbouring those dissolved for mafia infiltration tend to respond to their neighbours′ dis-solution by reducing the overall level of public investment (Galletta 2017). If some munic-ipalities artificially lower their spending as a response of the dissolution of a neighbouringmunicipality, there is a risk of bias in the control group which may affect our results. Inorder to discard the possibility that our results are driven by such spillover mechanism, wereplicate our estimates by excluding from the sample all municipalities having at least one

31This type of dissolution is indubitably a bad outcome for a newly elected local government. When thegovernment is dissolved for mafia-unrelated reasons the elected politicians cannot run again in the followingelection. Thus, politicians have every incentive to avoid this scenario.

25

neighbour dissolved for mafia infiltration at any moment during the 1998-2013 period. Thisentails dropping 268 municipalities from our full sample. The results are displayed in TableA16. The coefficient of total capital expenditures is larger (in absolute value) than thatof Table 2. This suggests that indeed, on average, municipalities which are neighbours ofdissolved local governments tend to exhibit a decrease in public spending. Yet, as in Table2, the coefficient of the treatment dummy is not statistically different from zero, indicatingthat total capital expenditures during infiltrations is not significantly different from non-infiltrated periods. All other key results remain unaffected by the exclusion of neighbouringmunicipalities.

Attacks to politicians. Next, we test for the relationship between mafia violence andcriminal infiltration. To do so, we adopt municipal level data on violent attacks againstlocal politicians and administrators perpetrated by organised crime groups.32 We regressthis measure on our standard Infiltration variable. The results are reported in Table A17,where the variable Attacks to Politicians is an indicator equal to 1 if there has been violenceagainst local administrators in a given municipality.33 The results indicate that there is nostatistical correlation between the capturing of local politicians (i.e. Infiltration) and mafias′

attacks against them. In column 2 we also test whether there is any violence against politi-cians in the year before the infiltration to the municipal council, finding no evidence of it.While one may expect that criminal infiltrations give rise to an increase in violent activities- as criminal groups may impose their will to administrators through violent means - engag-ing in violent behaviour when infiltrated may be counterproductive for mafia organisations,because it may undermine their potential to bias the allocation of public resources. Thismay be the reason for the insignificant coefficient observed in Table A18 where, exploitingour difference-in-differences setting, we run our main specification (equation 1) measuringorganised crime with the Attacks to Politicians variable. Attacks to Politicians is uncorre-lated with the spending components that are so strongly affected by the criminal infiltrationin local councils.

Organised crime and politics. A legitimate question is whether our results are trulydriven by criminal infiltration or, simply, by some unobserved political characteristics. We

32The dataset is constructed using information collected by Avviso Pubblico, an Italian NGO that sys-tematically record news of attacks to local Italian politicians. The data covers the years 2010-2015. Asstudied by Daniele and Dipoppa (2017) the violence targeted at politicians is not sporadic. From 2010 to2014 there were, on average, 277 attacks against Italian politicians.

33We are not exploiting the intensity of the attacks in this analysis. In some municipalities attacks aremore than in others and they are different in typology (from threatening letters to murders).

26

provide an answer to this concern by investigating the conditional correlations between a setof political factors and criminal organisations. We focus on the 182 municipalities that haveexperienced one dissolution for mafia infiltration between 1998 and 2013 and estimate thefollowing linear probability model.34

Political Factorsm,t = α + β Infiltrationm,t + δ NatGovt + ϑXm,t + ϕm + τt + εi,t (3)

Where, as in equation 1, Infiltrationm,t is an indicator equal to 1 if the local governmentis infiltrated by the mafia. NatGovt is a dummy variable controlling for the political colourof the national government at time t - left or right-wing governments. Xm,t is a vector ofcontrols, including socio-demographic variables and a variable on mafia homicides. ϕm andτt respectively represent municipality and time fixed effects. εm,t is an idiosyncratic errorterm. Standard errors are clustered at the municipality level.

Political Factorsm,t is sub-divided into a set of variables referring to key political fea-tures of the local government. First, electoral competition (Schleiter and Voznaya 2014),with a particular focus on the cases where there is only one candidate potentially eligiblefor mayor because no other electoral lists were presented (Single candidate).35 Second, weexploit the fact that up until 2014 all mayors had a term limit of two mandates to test a stan-dard accountability mechanism (Besley and Case 2003; List and Sturm 2006; Alt, Bueno deMesquita and Rose 2011; Ferraz and Finan 2011), that is, whether binding term limits affectthe behaviour of politicians (Last mandate).36 Finally, infiltration may be systematicallycorrelated with the political colour of the government.37 We investigate this by testingwhether there is a particular political party recurrently chosen by the mafias (Right party,Centre party, Left party and Civic list).38 Table A19 reports linear probability estimates

34We exploit the same dataset presented in the Section 3, augmented with data on election characteristicsfor all the municipalities of Calabria, Campania and Sicily from 1998 to 2013. Our primary data sourceis the Historical Archive of Local Elections of the Italian Ministry of Interior. Descriptive statistics for allpolitical variables are displayed in Table A3 in the Appendix.

35In such cases, the only condition necessary to valid the election is a voter turnout above 50%.36While mayors can run for a third term after a term break, third-term candidacies are rare.37Recent evidence has shown that the mafia sells votes to the party that has more core supporters and

it is therefore expected to win (De Feo and De Luca 2017). In Sicily, the strongest political relationshipsdeveloped by the mafia have been with the Christian Democrats (Democrazia Cristiana, DC) (De Feo andDe Luca 2017) and then after the DC′s demise in 1994, with the conservative party Forza Italia (Buonanno,Vanin and Prarolo 2016). Daniele and Geys (2016) find that in the elections immediately following thedissolution, turnout is lower.

38Theoretically, there are different political characteristics that might be associated with cases of collusion.

27

from this regression. Results show that criminal infiltrations are correlated with politicalcharacteristics. In particular, both the coefficients for Single candidate (indicator equal to 1if there is just one candidate running) and Last mandate (indicator equal to 1 if the mayoris running for her last mandate) are positive and significant. This seems to suggest thatorganised crime infiltrations are associated with absence of electoral competition and withpoliticians finishing their mandate. Furthermore, column 3 shows how criminal capturing ismore likely when right-wing parties are in power.

All the political elements discussed so far may not only be correlated with infiltrations,but also with the investment decisions of local governments. This would imply that our mainresults are actually driven by some omitted variables rather than criminal infiltrations. Totest for this, we replicate our main model (equation 1) controlling for the political elementscorrelated with infiltration and for the color of the National government. We estimate:

Ym,t = α+β Infiltrationm,t+γ Political Factorsm,t+δ NatGovt+ϑXm,t+ϕm+τt+εi,t (4)

The results, displayed in Table A20, report no significant correlation between key po-litical factors and the public finance components varying during infiltration periods. Thissuggests that, as hypothesised, the variations in public spending are not determined by anyof the political elements linked with infiltrations. Table A20 shows that our main results arerobust to the introduction of political controls. To investigate in a more causal fashion thelink between organised crime, politics and public spending we focus on the uncovered cor-relation between right party and mafia infiltration and implement a regression discontinuitydesign (RDD) based on close elections. We proceed in two steps. First, we whether theinfiltration is more likely to take place when the right-wing party wins by a narrow margin.Appendix A6 reports the RD estimating equation and its standard due diligence. Table A22and Figure A5 report the estimates and reveal that criminal infiltrations are more likelywhen right-wing parties marginally win the election. Second, we then restrict the sample toonly infiltrated municipalities and replicate RDD estimates by using our spending results asoutcome variables. This is an additional important robustness test to show that our previousresults were driven by a conservative government (and not mafia collusion). Table A23 andFigure A6 report the results. Reassuringly, the insignificant coefficients of right-wing parties

We do not argue that this list is exhaustive, but certainly electoral competition, accountability and partycolour are important dimensions to controls for.

28

reveal that there is no statistically significant variation in spending patterns.

7 Conclusion

Collusion and corruption distort the correct functioning of democratic systems. Such institu-tional failures have the potential to influence key aspects of economic activity, underminingthe long run development of any society (Shleifer and Vishny 1993; Mauro 1995; Glaeser andSaks 2006). A particularly dangerous form of corruption is that perpetrated by organisedcrime. Differently from the more common white-collar crimes, criminal groups seek profitthrough illegal business and frequently employ physical intimidation. Illegal and secretiveagreements between elected officials and colluding parties may alter the legislative process,compromising the definition of policies aimed at the welfare of citizens. Yet the mechanismsthrough which this negative impact takes place are still unclear. In this paper, we exploredone possible channel: collusion between organised crime and politicians. We analyse boththe conditions that make such collusions more likely and their possible consequences.

Using disaggregated municipal data from three regions of Southern Italy, we find that thecollusion between organised crime and politicians affects the allocation of public resourcesand the ability of local governments to collect resources. Our analysis suggests that whilethe overall amount of financial resources that local governments spend remains unaltered,expenditures for specific components of public finance vary significantly as a result of infiltra-tions. In particular, difference-in-differences estimates reveal that infiltrated municipalitiesinvest higher shares of resources in construction and waste management and reduce annualinvestment shares in municipal police forces. Moreover, infiltrated municipalities collect onaverage fewer waste taxes. These results are robust to changes in specifications and to aseries of robustness checks.

Furthermore, we have identified a set of political characteristics of municipal electionsthat are correlated with infiltrations. We find that both the absence of competition atlocal elections, as well as having a mayor running for her second and last mandate, arelinked with infiltrations. This seems to suggest that there may be some recurrent electoralpatterns associated with mafia-government collusions. Importantly, we find no evidenceof a correlation between these political conditions and spending decisions. This providesadditional evidence in favour of the hypothesis that variations in public spending decisionsare determined by infiltrations.

29

In conclusion, this paper provides an assessment of the strategy of organised crime whenit endeavours to take control of local politics and consequences for local state capacity.Criminal groups neither seem to impose generalised inflations of public expenditures, nor dothey seem to be interested in conditioning the current account budget. Rather, local financesare modified only in the key and strategic sectors where the mafia has interests to protect.In addition, we show that there may be some political parties that are systematically morelikely to collude with organised crime.

In sum, our analysis has unveiled the important distortionary effect that mafia infil-trations may have on politics and policy choices. Our study may help to gain a deeperunderstanding of such phenomena and possibly aid in the prevention of mafia infiltrations.

30

References

Acconcia, A., G. Corsetti and S. Simonelli. 2014. “Mafia and Public Spending: Evidence onthe Fiscal Multiplier from a Quasi-experiment.” American Economic Review 104(7):2185–2209.

Acemoglu, D., G. De Feo and G. De Luca. 2018. “Weak States: Causes and Consequencesof the Sicilian Mafia.” NBER Working Paper No. 24115 .

Acemoglu, D., J.A. Robinson and R.J. Santos. 2013. “The monopoly of violence: evidencefrom Columbia.” Journal of the European Economic Association 11(1):544.

Alesina, A., S. Piccolo and P. Pinotti. 2018. “Organized crime, violence, and politics.” TheReview of Economic Studies 86(2):457–499.

Allum, F. and R. Siebert, eds. 2003. Organized Crime and the Challenge to Democracy.Routledge.

Alt, J., E. Bueno de Mesquita and S. Rose. 2011. “Distangling Accountability and Compe-tence in Elections: Evidence from U.S. Term Limits.” Journal of Politics 73(1).

Angrist, J.D. and J. Pischke. 2008. Mostly Harmless Econometrics: An Empiricists Com-panion. Princton, NJ: Princton University Press.

Bandiera, O. 2003. “Land reform, the market for protection, and the origins of the Sicilianmafia: theory and evidence.” Journal of Law, Economics, and Organization 19(1):21 44.

Bandiera, O., A. Pratt and T. Valletti. 2009. “Active and Passive Waste in GovernmentSpending: Evidence from a Policy Experiment.” American Economic Review 99(4).

Barone, G. and G. Narciso. 2015. “Organized Crime and business subsidies: Where does themoney go?” Journal of Urban Economics 86:98–110.

Besley, T. and A. Case. 2003. “Political Institutions and Policy Choices: Evidence from theUnited States.” Journal of Economic Literature 41(1):7–73.

Borusyak, K., P. Hull and X. Jaravel. 2018. “Quasi-Experimental Shift-Share ResearchDesigns.” NBER Working Paper No. 24997 .

Borusyak, K. and X. Jaravel. 2017. “Revisiting Event Study Designs.”.

31

Buonanno, P., P. Vanin and G. Prarolo. 2016. “Organized Crime and Electoral Outcomes.Evidence from Sicily at the turn of the XXI Century.” European Journal of Political Econ-omy 41:61–74.

Buonanno, P., P. Vanin, G. Prarolo and R. Durante. 2015. “Poor institutions, rich mines:resource curse and the origins of the Sicilian mafia.” Economic Journal 125(586):F175 202.

Cantone, R. and G. Di Feo. 2014. I Gattopardi. Mondadori.

Casaburi, L. and U. Troiano. 2016. “Ghost-House Busters: The Electoral Response to aLarge Anti Tax Evasion Program.” Quarterly Journal of Economics 90:1133–53.

Coviello, D. and M. Mariniello. 2014. “Publicity requirements in public procurement:evidencefrom a regression discontinuity design.” Journal of Public Economics 109.

Crescenzi, R., M. Di Cataldo and A. Rodrıguez-Pose. 2016. “Government Quality and theEconomic Returns of Transport Infrastructure Investment in European Regions.” Journalof Regional Science 57(814-839).

Dal Bo, E., P. Dal Bo and R. Di Tella. 2006. “Plata o plomo? Bribe and punishment in atheory of political influence.” American Political Science Review 100(1):41 53.

Daniele, G. and B. Geys. 2015. “Organised Crime, Institutions and Political Quality: Em-pirical Evidence from Italian Municipalities.” Economic Journal 125(586):F233–F255.

Daniele, G. and B. Geys. 2016. “Exposing politicians? Ties to criminal organizations: theeffects of local government dissolutions on electoral outcomes in southern italian munic-ipalities.” Working Paper (Documents de Treball de l’IEB 2015/41) Institut d’Economiade Barcelona .

Daniele, G. and G. Dipoppa. 2017. “Mafia, Elections and Political Violence.” Journal ofPublic Economics 154:10–33.

Daniele, V. and U. Marani. 2011. “Organized crime, the quality of local institutions andFDI in Italy: a panel data analysis.” European Journal of Political Economy 27(1):13 42.

De Feo, G. and G. De Luca. 2017. “Mafia in the ballot box.” American Economic Journal:Economic Policy 9(3).

Dickie, J. 2005. Cosa Nostra: a history of the Sicilian Mafia. London: Macmillian.

32

Dimico, A., A. Isopi and O. Olsson. 2017. “Origins of the Sicilian Mafia: The Market forLemons.” The Journal of Economic History 77(4):1083–1115.

Drago, F., R. Galbiati and F. Sobbrio. 2014. “The Political Cost of Being Soft on Crime:Evidence from a Natural Experiment.” Journal of the European Economic Association .

Ferraz, C. and F. Finan. 2011. “Electoral Accountability and Corruption: Evidence fromthe Audits of Local Governments.” American Economic Review 101(4).

Galletta, S. 2017. “Law enforcement, municipal budgets and spill-over effects: Evidence froma quasi-experiment in Italy.” Journal of Urban Economics 101:90–105.

Gambetta, D. 1993. The Sicilian Mafia: The Business of Private Protection. HarvardUniversity Press.

Gambetta, D. and P. Reuter. 1995. Conspiracy among the Many: The Mafia in Legiti-mate Industries. In The Economics of Organised Crime, ed. Gianluca Fiorentini and SamPeltzman. Cambridge: Cambridge University Press pp. 116–160.

Gennaioli, C. and Onorato. 2010. “Public spending and organized crime: The case of the1997 Marche and Umbria Earthquake.” Unpublished.

Glaeser, EL. and RE. Saks. 2006. “Corruption in America.” Journal of Public Economics90(6-7):1053–1072.

Gupta, S., L. de Mello and R. Sharan. 2001. “Corruption and Military Spending.” EuropeanJournal of Political Economy 17:749–777.

Le Moglie, M. and G. Sorrenti. 2017. “Mafia Inc.: When Godfathers Become Entrepreneurs.”Mimeo .

List, J. and D. Sturm. 2006. “How Elections Matter: Theory and Evidence from Environ-mental Policy.” Quarterly Journal of Economics 121(4).

Mauro, P. 1995. “Corruption and Growth.” Quarterly Journal of Economics 110(3):681–712.

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

Olivieri, E. and S. Sberna. 2014. “Set the night on fire! Mafia Violence and Elections inItaly.” SSRN Working Paper 37 .

33

Pinotti, P. 2015. “The economic consequences of organized crime: evidence from southernItaly?” Economic Journal 125(586):F203–32.

Pinotti, P. and P. Stanig. 2016. “Sowing the Mafia: A Natural Experiment.” Working Paper.

Putnam, R.D. 1993. Making Democracy Work: Civic Traditions in Modern Italy. Princton,NJ: Princeton University Press.

Rajkumar, A. and V. Swaroop. 2008. “Public Spending and Outcomes: Does GovernanceMatter?” Journal of Development Economics 86:96–111.

Schelling, T. 1971. “What Is the Business of Organized Crime?” The American Scholar40(4):643–652.

Schleiter, P. and A.M. Voznaya. 2014. “Party system competitiveness and corruption.” PartyPolitics .

Scognamiglio, A. 2018. “When Mafia comes to town.” European Journal of Political Economy55:573–590.

Shleifer, A. and R. Vishny. 1993. “Corruption.” Quarterly Journal of Economics 108(3).

Tanzi, V. and H. Davoodi. 1997. “Corruption, Public Investment, and Growth.” IMF workingpaper 97/139 .

Trigilia, C. 2001. “Social Capital and Local Development.” European Journal of Social Theory4(4):427–442.

Trocchia, N. 2009. Federalismo criminale. Viaggio nei cmuni sciolti per mafia. Nutrimenti.

Wolton, S. 2016. “Lobbying, Inside and Out: How Special Interest Groups Influence PolicyChoices.” Working Paper .

34

Figures and Tables

Figures

Figure 1: Geographical location of the dissolutions

!

Italy

Campania

Calabria

Sicily

Source: Italian Ministry of Interior. Maps are authors′ own elaboration. The figure shows the geographical location of thedissolved municipalities. The first panel provides the locations at the national level. Most of the dissolvements took place inthe 3 regions that we study. The second panel shows the geographical breakdown at the municipal level.

35

Figure 2: Number of dissolved municipal governments for mafia infiltration

010

2030

1991 1994 1997 2000 2003 2006 2009 2012 2015year

number of dissolutions of which: in Calabria, Campania and Sicily

Source: Italian Ministry of Interior. The figure shows the yearly number of municipal dissolvements. The series stops at 2015,but since then 40 municipalities have been dissolved. To date, Calabria, Campania and Sicily (the 3 regions object of ourstudy), have experienced the highest number of dissolutions. Precisely, 104 in Campania, 103 in Calabria and 72 in Sicily.

Figure 3: Identification of the treatment period

Electoral history of the municipality of Casoria (Campania)

elections dissolution elections

1991 2002 2005 2008 2013

infiltration commissioners

control

control

TREATMENT

Note: In this example (Casoria), the treatment period begins with the election of Casoria’s local government in 2002 to itsdissolution in 2005. During this period, organised crime was colluding with members of the local government.

36

Figure 4: Organised crime controlled firms investments by sector

0% 5% 10% 15% 20% 25% 30% 35% 40%

Other

SocialServices

Agricolture/fishing

RealEstate

HotelsandRestaurants

RetailBusinesses

Construc;on/WasteManagement

Source: Authors′ own elaboration with Transcrime (2013) data. The figure shows the breakdown by firms seized because ofinfiltration with criminal organisations

37

Figure 5: Dynamic panel estimation (Event study)

Construction & waste management-.1

-.05

0.0

5.1

.15

2y prior3y prior4y prior5y prior 1y prior Infiltration 1y after 2y after

Police

-.005

0.0

05.0

1.0

15

2y prior3y prior4y prior5y prior 1y prior Infiltration 1y after 2y after

Waste tax

-.1-.0

50

.05

2y prior3y prior4y prior5y prior 1y prior Infiltration 1y after 2y after

Note: Event study. Dots refer to point estimates, spikes to 95% confidence intervals. We include dummy variables relative to 5 years prior to and 2 years after the electionof the infiltrated government. We use the restricted sample of municipalities that have been dissolved. Importantly, all years after the dissolution of the infiltratedgovernments are excluded. The omitted category is the first year before the infiltration. The estimation includes time and municipality dummies, socio-demographiccontrols and municipal population. Standard errors clustered at the municipal level.

38

Tables

Table 1: Descriptive statistics - public spending

Full sample Dissolved sample

Obs Mean Std dev Obs Mean Std dev

Total per capita expenditures

Total 21,156 1273.8 1129.9 2,678 1020.3 930.96Capital expenditures 21,156 542.82 1002.7 2,678 354.98 821.98Current expenditures 21,156 730.97 394.43 2,678 665.3 267.2

Capital expenditures (share of total)

Administration 21,037 0.152 0.217 2,813 0.164 0.211Social sector 20,901 0.063 0.134 2,789 0.056 0.124Construction & waste management 21,137 0.342 0.292 2,817 0.321 0.277Public transport and lighting 21,090 0.232 0.242 2,818 0.228 0.233Education 20,844 0.084 0.153 2,799 0.106 0.166Municipal police 20,477 0.003 0.019 2,751 0.007 0.026

Current expenditures (share of total)

Administration 21,240 0.429 0.095 2,842 0.400 0.093Social sector 21,243 0.073 0.058 2,842 0.086 0.061Construction and waste management 21,239 0.228 0.085 2,842 0.269 0.090Public transport & lighting 19,909 0.082 0.040 2,664 0.068 0.036Education 18,557 0.083 0.041 2,480 0.075 0.038Municipal police 21,239 0.059 0.027 2,842 0.058 0.023

Municipal revenues (collected/forecasted)

Total revenues 17,596 0.494 0.362 2,381 0.478 0.425Total taxes 18,692 0.573 0.192 2,524 0.563 0.157Property tax 18,703 0.463 0.210 2,524 0.437 0.202Waste tax 17,314 0.137 0.250 2,330 0.097 0.198

Note: ’Full sample’ refers to all 1350 municipalities of Campania, Calabria and Sicily. ’Dissolved sample’ refers to the 182municipalities of these regions that experienced at least one government dissolution due to mafia infiltration. The sum of themeans of all capital account or current account spending components does not sum up to 1 due to the fact that there are someother minor spending items.

39

Table 2: Infiltration and total per capita spending

Totalexpenditures

Totalcapital expenditures

Totalcurrent expenditures

(1) (2) (3) (4) (5) (6)

Panel A - Main specification

Infiltration -0.012 -0.0029 -0.047 -0.036 0.014 0.0145(0.019) (0.019) (0.069) (0.070) (0.0134) (0.012)

Observations 20,888 2,582 20,889 2,582 20,890 2,582R-squared 0.534 0.532 0.357 0.351 0.754 0.560

Panel B - Excluding post dissolution years

Infiltration 0.0014 -0.009 0.0192 -0.0105 0.017 0.018(0.0263) (0.0293) (0.0764) (0.0891) (0.0205) (0.0232)

Observations 20,380 2,074 20,381 2,074 20,382 2,074R-squared 0.532 0.530 0.357 0.370 0.756 0.541

Controls Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesFull sample Yes No Yes No Yes NoDissolved sample No Yes No Yes No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis comparesthe spending categories of non-infiltrated government with infiltrated governments, before and after the infiltration is terminatedby the national government. Commissioning period always excluded. Infiltration is a dummy equal to 1 if the government isinfiltrated by organised crime. Outcome variables: Log total per capita spending (columns (1)-(2)); log total per capita capitalexpenditures (columns (3)-(4)); log total per capita current expenditures (columns (5)-(6)). Controls: industry employment,tertiary education degree holders, municipal population and unemployment rate. Full sample: 1350 municipalities of Campania,Calabria and Sicily; dissolved sample: 182 municipalities that experienced one government dissolution for mafia infiltration. InPanel B, the analysis excludes the post-commissioning years: the control group is composed by municipalities that have neverbeen dissolved and, for the dissolved municipalities, only by the years before the infiltration.

40

Table 3: Infiltration and capital spending categories

Administration Social sectorConstruction &

waste managementPublic transport

& lighting Education Police

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A - Main specification

Infiltration -0.0116 -0.016 -0.0045 -0.0061 0.0444** 0.0503*** -0.0196 -0.0204 0.0057 0.0069 -0.00268** -0.00289**(0.0143) (0.0140) (0.0075) (0.0077) (0.0175) (0.0174) (0.0132) (0.0134) (0.011) (0.011) (0.0012) (0.0012)

Observations 20,676 2,554 20,545 2,535 20,777 2,559 20,729 2,559 20,484 2,541 20,120 2,496R-squared 0.260 0.218 0.135 0.138 0.205 0.225 0.174 0.152 0.116 0.138 0.170 0.232

Panel B - Excluding post-commissioning years

Infiltration -0.0092 -0.013 -0.0043 -0.00024 0.0461** 0.0429* -0.0231 -0.0333* 0.0086 0.0135 -0.00023 -0.00337**(0.017) (0.0192) (0.009) (0.010) (0.0211) (0.0238) (0.0170) (0.019) (0.0135) (0.0142) (0.0012) (0.00162)

Observations 20,169 2,047 20,038 2,028 20,270 2,052 20,222 2,052 19,977 2,034 19,613 1,989R-squared 0.261 0.219 0.138 0.169 0.208 0.251 0.176 0.169 0.118 0.166 0.161 0.243Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesFull sample Yes No Yes No Yes No Yes No Yes No Yes NoDissolved sample No Yes No Yes No Yes No Yes No Yes No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis compares capital spending categories of non-infiltrated government with infiltrated governments, before and after the infiltration is terminated by the national government. Commissioning period always excluded.Infiltration is a dummy equal to 1 if the government is infiltrated by organised crime. The dependent variables are yearly capital and current expenditure spendingfor Administration (columns (1)-(2)); social sector (columns (3)-(4)); construction and waste management (columns (5)-(6)); public transport and lighting (columns(7)-(8)); education (columns (9)-(10)) and police (columns (11)-(12)). Spending components calculated as share of total spending. Controls: industry employment,tertiary education degree holders, municipal population and unemployment rate. For each outcome variables we perform two estimations: one with full sample of 1350municipalities and one with the dissolved sample, i.e. 182 municipalities that have been dissolved.

41

Table 4: Infiltration and current spending categories

Administration Social sectorConstruction &

waste managementPublic transport

& lighting Education Police

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A - Main specification

Infiltration -0.0062 -0.00534 -0.0014 -0.00116 0.0058 0.00462 -0.00087 -0.00073 0.00021 0.000692 -0.0026** -0.0029**(0.0049) (0.00483) (0.0051) (0.0051) (0.0049) (0.00496) (0.00193) (0.00193) (0.00168) (0.00167) (0.00131) (0.00126)

Observations 20,875 2,579 20,878 2,579 20,874 2,579 19,579 2,427 18,229 2,242 20,874 2,579R-squared 0.739 0.697 0.651 0.612 0.733 0.687 0.753 0.752 0.816 0.787 0.623 0.660

Panel B - Excluding post-commissioning years

Infiltration -0.0060 -0.0029 -0.00047 -0.00237 0.0084 0.00841 -0.00395* -0.0031 -0.0022 -0.0019 0.00122 -0.00099(0.00578) (0.00641) (0.0049) (0.00485) (0.0055) (0.00583) (0.00214) (0.00227) (0.00218) (0.00237) (0.00150) (0.00173)

Observations 20,367 2,071 20,370 2,071 20,366 2,071 19,142 1,990 17,791 1,804 20,366 2,071R-squared 0.741 0.716 0.654 0.633 0.736 0.714 0.753 0.768 0.817 0.804 0.624 0.674Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesFull sample Yes No Yes No Yes No Yes No Yes No Yes NoDissolved sample No Yes No Yes No Yes No Yes No Yes No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis compares current spending categories of non-infiltrated government with infiltrated governments, before and after the infiltration is terminated by the national government. Commissioning period always excluded.Infiltration is a dummy equal to 1 if the government is infiltrated by organised crime. The dependent variables are yearly capital and current expenditure spendingfor Administration (columns (1)-(2)); social sector (columns (3)-(4)); construction and waste management (columns (5)-(6)); public transport and lighting (columns(7)-(8)); education (columns (9)-(10)) and police (columns (11)-(12)). Spending components calculated as share of total spending. Controls: industry employment,tertiary education degree holders, municipal population and unemployment rate. For each outcome variables we perform two estimations: one with full sample of 1350municipalities and one with the dissolved sample, i.e. 182 municipalities that have been dissolved.

42

Table 5: Infiltration and local revenue collection

Property tax Waste tax Total taxes Total revenues

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A - Main specification

Infiltration 0.0301 0.032 -0.0216** -0.019** -0.00048 -4.73e-05 -0.0128 -0.0157(0.0411) (0.0424) (0.0090) (0.0089) (0.0111) (0.011) (0.0111) (0.0114)

Observations 17,383 2,170 17,103 2,122 18,475 2,299 18,464 2,299R-squared 0.395 0.351 0.521 0.472 0.670 0.647 0.316 0.388

Panel B - Excluding post-dissolution years

Infiltration -0.0118 -0.0286 -0.0423*** -0.0143 -0.0192 -0.0258* 0.000299 -0.00157(0.0219) (0.0220) (0.0113) (0.0149) (0.0134) (0.0136) (0.0124) (0.0138)

Observations 17,045 1,832 16,739 1,758 18,088 1,912 18,077 1,912R-squared 0.416 0.475 0.526 0.528 0.674 0.680 0.318 0.421Controls Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes YesFull sample Yes No Yes No Yes No Yes NoDissolved sample No Yes No Yes No Yes No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. In Panel A, the analysiscompare the revenues collection of non infiltrated government with infiltrated governments, before and after the infiltrationis terminated by the national government. Commissioning period always excluded. Infiltration is a dummy equal to 1 ifthe government is infiltrated by organised crime. The outcome variables are: “Property tax” (columns (1)-(2)), “Waste tax”(columns (3)-(4)), “Total taxes” (columns (5)-(6)), and “Total revenues” (columns (7)-(8)). Controls: industry employment,tertiary education degree holders, municipal population and unemployment rate. Full sample: 1350 municipalities of Campania,Calabria and Sicily; dissolved sample: municipalities that experienced a government dissolution for mafia infiltration. In PanelB, the analysis excludes the post 182 dissolution years for the dissolved municipalities, i.e. the control group in the estimationare composed by municipalities that have never been dissolved and, for the dissolved municipalities, only by the years beforethe infiltration.

Table 6: Mafia-infiltrated firms and dissolved municipalities

Mafia firms

(1) (2)

Municipal dissolution 0.438*** 0.320***(0.0290) (0.0327)

Controls No YesObservations 1,356 1,350R-squared 0.089 0.170

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The table shows thecorrelation between municipalities with cases of firm(s) either seized or under investigation for mafia infiltration and the dissolvedmunicipalities. It employs a cross sectional data where mafia firm is a dummy variable equal to 1 if in a given municipalitythere is a firm linked to criminal organisations. Dissolution is a dummy variable taking value 1 if the municipality has beendissolved. In column 2 we include socio-demographic controls at municipal level. Firms′ data are obtained from the ItalianAnti-mafia Directorate.

43

Table 7: Timing of the infiltration

PoliceConstruction &

waste managementPublic transport

& lighting Waste tax

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A - Full sample

Infiltration -0.00268** 0.0444** -0.0196 -0.0216**(0.00126) (0.0175) (0.0132) (0.009)

Infiltration (t-1) 0.0007 0.005 -0.004 0.007(0.002) (0.03) (0.02) (0.02)

Prior legislature 0.0004 0.01 -0.01 0.012(0.002) (0.02) (0.013) (0.012)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Observations 20,120 20,120 20,120 20,777 20,777 20,777 20,729 20,729 20,729 17,103 17,103 17,103R-squared 0.170 0.169 0.169 0.205 0.205 0.205 0.174 0.174 0.521 0.521 0.521 0.521

Panel B - Dissolved sample

Infiltration -0.00289** 0.0503*** -0.0204 -0.019**(0.00122) (0.0174) (0.0134) (0.0088)

Infiltration (t-1) 0.0013 0.005 -0.003 0.011(0.001) (0.027) (0.0221) (0.0175)

Prior legislature 0.0015 0.0057 -0.0056 0.0122(0.0015) (0.0182) (0.0142) (0.012)

Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesObservations 2,496 2,496 2,496 2,559 2,559 2,559 2,559 2,559 2,559 2,122 2,122 2,122R-squared 0.232 0.234 0.234 0.225 0.224 0.224 0.152 0.151 0.472 0.472 0.472 0.472

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. Dependent variables: Capital expenditures for Police (columns(1)-(3)); capital expenditures for construction and waste management (columns (4)-(6)); capital expenditures for transport and lighting (columns (7)-(9)); waste taxcollection (columns (10)-(12)); Infiltration refers to the infiltration dummy; Infiltration (t-1) is a dummy variable taking value 1 for the year immediately beforethe election of a later-dissolved government; Prior legislature is a dummy variable taking value of 1 for the entire legislature before the election of a later dissolvedgovernment. All years coded as infiltration years, from the election to the dissolution, are excluded from the sample in columns (3),(6),(9),(12). The estimation exploitsthe full sample of 1350 municipalities. Commissioning years always excluded.

44

Appendix

Figure A1: Yearly earning (bn EUR) by criminal organisation

0 0.5 1 1.5 2 2.5 3 3.5 4

Other

CosaNostra(Sicilia)

Ndrangheta(Calabria)

Camorra(Campania)

Source: Authors’ own elaboration with data from Transcrime, Gli Investimenti delle Mafie 2013. The figure provides adescriptive estimate of the illegal yearly earnings of criminal organisations. They are measured in billions of Euro and calculatedover a 10-years period, 1994-2004.

45

Figure A2: Organised crime by illegal sector

0.0%

5.0%

10.0%

15.0%

20.0%

25.0%

30.0%

35.0%

40.0%

45.0%

50.0%

Estor/

on

Drugs

Usury

Pros/tu/

on

Forgery

Tobacco

Gambling

Garbages

Weapons

Source: Authors’ own elaboration with data from Transcrime, Gli Investimenti delle Mafie 2013. The figure provides abreakdown of the main illegal activities of criminal organisations.

46

Figure A3: Percentage of illegal profits re-invested into the legal economy

55%Italy

28%USA

35%Japan

0

10

20

30

40

50

60

1950 1960 1970 1980 1990 2000 2010

%ofillegalearningsre-invested

inlegalecono

my

Source: Transcrime and Geo. L.O.C. of Financial Guards. The figure provides a descriptive estimate of the illegal profitsre-invested into the legal economy. After the second world war the trend is constantly increasing. The figure shows a similarpattern (with different levels) for three developed economies, Italy, US and Japan.

47

Figure A4: Dynamic panel estimation (Event study)

Construction & waste management-.1

-.05

0.0

5.1

.15

1y prior2y prior3y prior4y prior5y prior Infiltration 1y after 2y after 3y after 4y after

Police

-.01

-.005

0.0

05.0

1

1y prior2y prior3y prior4y prior5y prior Infiltration 1y after 2y after 3y after 4y after

Waste tax

-.1-.0

50

.05

.1

1y prior2y prior3y prior4y prior5y prior Infiltration 1y after 2y after 3y after 4y after

Note: Event study. Dots refer to point estimates, spikes to 95% confidence intervals. We include dummy variables relative to 5 years prior to and 4 years after theelection of the infiltrated government. We use the restricted sample of municipalities that have been dissolved at least once. Importantly, all years after the dissolutionof the infiltrated governments are excluded. The omitted category is the first year before the infiltration. The estimation includes time and municipality dummies,socio-demographic controls and municipal population. Standard errors clustered at the municipal level.

48

A1 Political Manipulation

Table A1: Correlation between dissolved municipal governments and National government

National Government

Number of dissolutionsMunicipal

Government Right Left67 Right -0.108 0.06143 Left 0.14 -0.0476 Centre -0.07 -0.011

Note: no statistically significant coefficient. Right-wing national governments: Berlusconi 2001-2005 and Berlusconi 2008-2011; Left-wing national governments: Prodi 1998, D’Alema 1999, Amato 2000, Prodi 2006-2007, Letta 2013; Centre nationalgovernments: Monti 2012.

Table A2: Correlation between dissolved municipal governments and provincial governments

Province and Provincial GovernmentsMunicipal Government Caserta Napoli Reggio Calabria Vibo Valentia Palermo

(Right) (Left) Right) (Left) (Right) (Left) (Right) (Left) (Right) (Left)Right -0.143 / 0.28 / 0.233 / N/A / -0.154 /Left / -0.15 / 0.19 / 0.14 / 0.24 / N/a

Note: no statistically significant coefficient. None of these provinces had governments from the ′Centre′ over the 1998-2013period. Vibo Valentia only had left-wing governments while Palermo only had right-wing governments. a / Right-wing municipalgovernments during infiltration period in given province. b / Left-wing municipal governments during infiltration period ingiven province.

Table A3: Descriptive statistics: political variables

All municipalities Infiltration years

Obs Mean Std. Dev. Obs Mean Std. Dev

Single candidate 2,869 -0.023 0.149 437 0.059 0.059Last mandate 2,869 0.023 0.402 437 0.327 0.470Left party 2,869 0.320 0.467 437 0.316 0.465Centre party 2,869 0.077 0.266 437 0.098 0.298Right party 2,869 0.461 0.499 437 0.563 0.497Civic list 2,869 0.510 0.500 437 0.584 0.494

Note: All municipalities: municipalities of Campania, Calabria and Sicily that experienced a government dissolution for mafiainfiltration. Infiltration years: years classified as infiltration for these municipalities

49

Table A4: Descriptive statistics: control variables

Full sample Dissolved sampleObs Mean Std. Dev Obs Mean Std. Dev

Percentage agricultural employment 21,594 4.592 3.382 2,912 4.303 4.066Percentage of citizens holding tertiary education degrees 21,594 6.06 2.62 2,912 5.687 2.272Percentage of industry employment 21,594 6.489 2.128 2,912 5.894 1.693Unemployment rate 21,594 7.609 2.518 2,912 8.89 2.646Mafia-related homicides at the province level 21,600 0.0058 0.0082 2,912 0.0095 0.0092

Note: Full sample refers to all municipalities of Campania, Calabria and Sicily. Dissolved sample refers to municipalities ofthese regions that experienced a government dissolution for mafia infiltration. Source: Istat and Ministry of Interior

A2 Municipal institutional setting and public spending

• Italian municipalities institutional setting. As of 2016, there were 8,010 municipalitiesin Italy, 1,350 of which are found in the regions of analysis, varying considerably by areaand population. The institutional setting of the municipalities is centred on the figureof the mayor, who heads the local government and leads along with the legislative body,the local council, and the executive body, the local giunta. The mayor and membersof the council are elected together by resident citizens. The giunta is chaired by themayor, who appoints its members. Elections of local councils are staggered over timeand not held at the same time for all municipalities.

• Public spending components. General functions of administration include all expensesrelated to the management of offices coordinating the internal activities of the munic-ipality; (2) social sectors include all expenses for the provision of social services andthe creation of infrastructure to that aim (kindergartens, retirement homes, rehab cen-tres); (3) construction and waste management refers to all expenses for urban planning- adoption of construction plans and building regulations, maintenance and construc-tion of all new buildings (all part of capital account spending), waste collection anddisposal (current account spending); (4) public transport and lighting includes expensesto guarantee local public transportation, public lighting, management of road traffic;(5) public education includes all expenses for all education infrastructure, school main-tenance and school transportation; (6) functions of local police include the acquisitionand maintenance of goods and equipment, cars and office structures.

50

Figure A5: Capital and current expenditures over time

200

400

600

800

1000

euro

s pe

r cap

ita

1998 2001 2004 2007 2010 2013year

capital account current account

Source: Italian Ministry of Interior, Division of Local Finance

A3 Robustness checks: main results

Gradually increase control variables. From Table A5 to Table A8 we provide a series of ro-bustness checks for our main results, i.e. total municipal spending (A5), capital expenditure(A6), current expenditure (A7) and taxes (A8). In all estimations, the sample is also re-stricted to the municipalities that have been dissolved. This is important because we controlfor unobserved heterogeneous effects that might be present across municipalities. In the firstcolumn, a parsimonious specification is presented, including municipal fixed effects and noother controls. The second column adds year fixed effects. The third column includes mu-nicipal socio-economic factors as controls. The fourth and last column include municipalityspecific time trends. In practice, the results in column (3) of table A5 - A8 replicate ourmain estimates in Tables 2, 3 and 5. In the fourth column of Tables A5 - A8, we includea full set of municipality specific linear time trends , accounting for any previously omittedfactor potentially affecting the temporal development of municipal governments and corre-lated with infiltrations. Our estimates remain stable with the exception of Public transportand lighting, whose coefficient remains negative and maintains a similar magnitude but losesits significance. The estimate remains economically sizeable.

51

Table A5: Infiltration and total spending - progressive inclusion of controls

(1) (2) (3)

Panel A - Dep var: total spending pc

Infiltration 0.0154 -0.0058 -0.006(0.0215) (0.019) (0.02)

Observations 2,582 2,582 2,582R-squared 0.482 0.518 0.512

Panel B - Dep var: total capital expenditures pc

Infiltration -0.0226 -0.0404 -0.073(0.0786) (0.072) (0.069)

Observations 2,582 2,582 2,582R-squared 0.289 0.313 0.347

Panel C - Dep var: total current expenditures pc

Infiltration 0.042* 0.0121 0.014(0.012) (0.0113) (0.0121)

Observations 2,582 2,582 2,582R-squared 0.448 0.550 0.556Controls No No YesMunicipality dummies Yes Yes YesYear dummies No Yes Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We replicate Table2 by introducing gradually municipality dummies, time dummies, municipal controls. The analysis compares the spendingcategories of non-infiltrated government with infiltrated goverments, before and after the infiltration is terminated by thenational government. Commissioning period always excluded. Infiltration is a dummy equal to 1 if the government is infiltratedby organised crime. The dependent variables are total spending, total capital and total current expenditure. Column 1 controlsfor municipal FE. Column 2 introduces year FE. Column 3 introduces socio-demographic controls at the municipal level. Weperform the analysis on the restricted sample of 182 municipalities that have experienced one dissolution.

52

Table A6: Infiltration and capital expenditures - progressive inclusion of controls

(1) (2) (3) (4)

Panel A - Dep. var.: Administration

Infiltration -0.0113 -0.0143 -0.0162 -0.00346(0.014) (0.014) (0.0141) (0.0166)

Observations 2,554 2,554 2,554 2,554R-squared 0.205 0.214 0.217 0.320

Panel B - Dep. var.: Social sector

Infiltration -0.00603 -0.0059 -0.0059 -0.00921(0.0076) (0.0076) (0.0077) (0.0093)

Observations 2,535 2,535 2,535 2,535R-squared 0.131 0.136 0.138 0.229

Panel C - Dep. var.: Construction & waste management

Infiltration 0.0407** 0.0469*** 0.0501*** 0.0471**(0.0175) (0.0177) (0.0174) (0.0197)

Observations 2,559 2,559 2,559 2,559R-squared 0.209 0.220 0.226 0.333

Panel D - Dep. var.: Public transport & lighting

Infiltration -0.0199 -0.0206 -0.0200 -0.0175(0.0134) (0.0133) (0.0134) (0.0147)

Observations 2,559 2,559 2,559 2,559R-squared 0.142 0.150 0.151 0.257

Panel E - Dep. var.: Education

Infiltration 0.00559 0.00807 0.0059 0.0096(0.0112) (0.0111) (0.0109) (0.0132)

Observations 2,541 2,541 2,541 2,541R-squared 0.117 0.133 0.139 0.218

Panel F - Dep. var.: Police

Infiltration -0.00256* -0.00277** -0.00294** -0.00442*(0.0013) (0.00125) (0.00122) (0.0024)

Observations 2,496 2,496 2,496 2,496R-squared 0.212 0.230 0.234 0.418Controls No No Yes YesMunicipality dummies Yes Yes Yes YesYear dummies No Yes Yes YesTime trends No No No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We replicate Panel A,Table 3 by introducing gradually Municipality dummies, time dummies, municipal controls and municipality-specific time trends.The analysis compares the spending categories of non-infiltrated government with infiltrated goverments, before and after theinfiltration is terminated by the national government. Commissioning period always excluded. Infiltration is a dummy equal to1 if the government is infiltrated by organised crime. The dependent variables are yearly capital expenditure spending. Column1 controls for municipal FE. Column 2 introduces year FE. Column 3 introduces socio-demographic controls at the municipallevel. Controls include industry employment, tertiary education degree holders, municipal population and unemployment rate.Column 4 introduces municipality-specific time trends. We perform the analysis on the restricted sample of 182 municipalitiesthat have experienced one dissolution.

53

Table A7: Infiltration and current expenditures - progressive inclusion of controls

(1) (2) (3) (4)

Panel A - Dep. var.: administration

Infiltration -0.00231 -0.00482 -0.007 0.0035(0.005) (0.005) (0.0045) (0.00504)

Observations 2,579 2,579 2,579 2,579R-squared 0.672 0.691 0.694 0.782

Panel B - Dep. var.: Social sector

Infiltration 0.000363 -0.00184 0.00031 0.0046(0.00514) (0.0051) (0.00483) (0.0045)

Observations 2,579 2,579 2,579 2,579R-squared 0.585 0.602 0.606 0.744

Panel C - Dep. var.: Construction & waste management

Infiltration 0.00803 0.005 0.00546 -0.000584(0.0055) (0.00495) (0.00497) (0.00515)

Observations 2,579 2,579 2,579 2,579R-squared 0.631 0.685 0.687 0.778

Panel D - Dep. var.:Public transport & lighting

Infiltration -0.00272 -0.00066 -0.00103 -0.0012(0.00203) (0.0012) (0.00195) (0.0016)

Observations 2,427 2,427 2,427 2,427R-squared 0.723 0.751 0.752 0.847

Panel E - Dep. var.: Education

Infiltration -0.00632** 0.00055 0.000591 -0.000846(0.00254) (0.00168) (0.00174) (0.0018)

Observations 2,242 2,242 2,242 2,242R-squared 0.395 0.786 0.787 0.874

Panel F - Dep. var.: Police

Infiltration -0.00272** -0.00301** -0.00228* -0.000837(0.00129) (0.00128) (0.00125) (0.00116)

Observations 2,579 2,579 2,579 2,579R-squared 0.649 0.658 0.663 0.774Municipality dummies Yes Yes Yes YesYear dummies No Yes Yes YesControls No No Yes YesTime Trends No No No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We replicate PanelB, Table 3 by introducing gradually Municipality dummies, time dummies, municipal controls and municipality specific timetrends. The analysis compares the spending categories of non-infiltrated government with infiltrated goverments, before andafter the infiltration is terminated by the national government. Commissioning period always excluded. Infiltration is a dummyequal to 1 if the government is infiltrated by organised crime. The dependent variables are yearly current expenditure spending.Column 1 controls for municipal FE. Column 2 introduces year FE. Column 3 introduces socio-demographic controls at themunicipal level. Column 4 introduces municipality-specific time trends. We perform the analysis on the restricted sample of182 municipalities that have experienced one dissolution.

54

Table A8: Infiltration and municipal revenues - progressive inclusion of controls

(1) (2) (3) (4)

Panel A - Dep. var.: Property tax

Infiltration 0.0164 0.0288 0.0339 0.00934(0.0490) (0.0413) (0.0419) (0.0266)

Observations 2,170 2,170 2,170 2,170R-squared 0.155 0.349 0.351 0.489

Panel B - Dep. var.: Waste tax

Infiltration -0.0202** -0.0234*** -0.0231*** -0.00244*(0.00893) (0.00859) (0.00907) (0.00133)

Observations 2,125 2,125 2,125 2,170R-squared 0.451 0.468 0.468 0.489

Panel C - Dep. var.: Total revenues

Infiltration 0.000720 -0.0149 -0.0131 -0.0082(0.0125) (0.0113) (0.0112) (0.0121)

Observations 2,299 2,299 2,299 2,299R-squared 0.263 0.383 0.388 0.488

Panel D - Dep. var.: Total taxes

Infiltration 0.0184 0.000247 -0.00656 0.00077(0.0163) (0.0114) (0.0108) (0.013)

Observations 2,299 2,299 2,299 2,299R-squared 0.153 0.644 0.650 0.731Controls No No Yes YesMunicipality dummies Yes Yes Yes YesYear dummies No Yes Yes YesTime trends No No No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We replicate Table5 by introducing gradually Municipality dummies, time dummies, municipal controls and municipality specific time trends.The analysis compares the spending categories of non-infiltrated government with infiltrated goverments, before and after theinfiltration is terminated by the national government. Commissioning period always excluded. Infiltration (Inf) is a dummyequal to 1 if the government is infiltrated by organised crime. The dependent variables are respectively property tax, wastetax, total revenues and total taxes. Column 1 controls for municipal FE. Column 2 introduces year FE. Column 3 introducessocio-demographic controls at the municipal level. Column 4 introduces municipality-specific time trends. We perform theanalysis on the restricted sample of 182 municipalities that have experienced one dissolution.

55

Table A9: Effect of infiltration on aggregate (capital and current) spending components

Administration Social sectorConstruction &

waste managementPublic transport

& lighting Education Police

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)

Panel A: Main specification

Infiltration -0.0171 -0.0204 -0.00636 -0.00762 0.0496*** 0.0544*** -0.0190 -0.0203 0.00651 0.00837 -0.00508*** -0.00565***(0.0144) (0.0143) (0.00892) (0.00907) (0.0177) (0.0177) (0.0144) (0.0145) (0.0123) (0.0122) (0.00190) (0.00178)

Observations 20,667 2,553 20,538 2,533 20,766 2,557 19,429 2,406 17,833 2,203 20,116 2,495R-squared 0.338 0.278 0.224 0.234 0.251 0.255 0.211 0.179 0.185 0.199 0.498 0.463

Panel B - excluding post dissolution years

Infiltration -0.0143 -0.0143 -0.00751 -0.00560 0.0531** 0.0494** -0.0267 -0.0334* 0.00485 0.0110 0.000851 -0.00464*(0.0173) (0.0197) (0.00999) (0.0107) (0.0212) (0.0248) (0.0181) (0.0200) (0.0154) (0.0164) (0.00217) (0.00262)

Observations 20,160 2,046 20,031 2,026 20,259 2,050 18,992 1,969 17,396 1,766 19,609 1,988R-squared 0.340 0.284 0.227 0.273 0.254 0.280 0.212 0.191 0.190 0.244 0.504 0.501Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesFull sample Yes No Yes No Yes No Yes No Yes No Yes NoDissolved sample No Yes No Yes No Yes No Yes No Yes No Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis replicates Table 3 and 4, but expresses the spendingcategories as aggregates, i.e. sum of current and capital expenditures for each line items divided by total spending. Commissioning period always excluded. Infiltrationis a dummy equal to 1 if the government is infiltrated by organised crime. The dependent variables are: yearly aggregate spending on Administration (columns(1)-(2)); social sector (columns (3)-(4)); construction and waste management (columns (5)-(6)); public transport and lighting (columns (7)-(8)); education (columns(9)-(10)) and police (columns (11)-(12)). For each outcome variables we perform two estimations: one with the full sample of municipalities and one with the restrictedsample, i.e. municipalities that have been dissolved. Panel A reports our main specification. In Panel B, we exclude the post dissolution years. The estimates includesocio-demographic controls: industry employment, tertiary education degree holders, unemployment; municipality fixed effects, year fixed effects. Full sample: 1350municipalities of Campania, Calabria and Sicily. Dissolved sample: 182 municipalities having experienced a government dissolution for mafia infiltration.

56

Table A10: Effect of infiltration on spending components: seemingly unrelated regressionequations

Administration Social sectorConstruction &

waste managementPublic transport

& lighting Education Police(1) (2) (3) (4) (5) (6)

Infiltration -0.0169 -0.0034 0.0498*** -0.0147 -0.00756 -0.00271**(0.0112) (0.0069) (0.0147) (0.0128) (0.0092) (0.0013)

Observations 2,731 2,731 2,731 2,731 2,731 2,731R-squared 0.223 0.131 0.226 0.158 0.133 0.233Controls Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesDissolved sample Yes Yes Yes Yes Yes Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis comparescurrent spending categories of non-infiltrated government with infiltrated governments, before and after the infiltration isterminated by the national government. Commissioning period always excluded.Infiltration is a dummy equal to 1 if thegovernment is infiltrated by organised crime. The analysis fits seemingly unrelated regression models to correct for potentialerrors’ correlation among dependent variables. he dependent variables are yearly capital and current expenditure spending forAdministration (columns (1)-(2)); social sector (columns (3)-(4)); construction and waste management (columns (5)-(6)); publictransport and lighting (columns(7)-(8)); education (columns (9)-(10)) and police (columns (11)-(12)). Spending componentscalculated as share of total spending. Controls: industry employment,tertiary education degree holders, municipal populationand unemployment rate. We perform the analysis on the dissolved sample, i.e. 182 municipalities that have been dissolved.

A4 Capital expenditure components and infiltration by municipalpopulation

One aspect we investigate in this section is whether the intensity of the effect depends on thesize of the municipalities whose governments are infiltrated. We hypothesise that the largestabsolute variation in spending allocations are found in smaller municipalities. Small townsare where the power of the mafia can be more pervasive, due to the high control of territoryit exercises and to the greater distance from the central State felt by the citizens. We testthis by sub-dividing the entire sample into municipalities with less than 2,000 inhabitants,between 2,000 and 5,000 inhabitants, and above 5,000 inhabitants, replicating the mainestimates.

57

Effect of infiltration on capital account spending components by municipal population

Population

Below 2000 Between 2000/5000 Above 5000(1) (2) (3)

Panel A - Dep. var.: Construction & waste managementInfiltration 0.0963** 0.0797** 0.0196

(0.0430) (0.0329) (0.0218)Observations 6,564 6,299 7,258

Panel B - Dep. var.: PoliceInfiltration 0.00294 -0.00174 -0.00338**

(0.00257) (0.00176) (0.00166)Observations 6,817 6,514 7,447

Panel C - Dep. var.: Public transport & lightingInfiltration -0.024* -0.036 -0.0127

(0.0015) (0.024) (0.0175)Observations 6792 6502 7433Controls Yes Yes YesMunicipality dummies Yes Yes YesYear dummies Yes Yes Yes

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The Table replicatesTable 3 and provides a breakdown of the sample by population. Infiltration refers to infiltration dummy; The estimationincludes socio-demographic controls at the municipal level: industry employment, tertiary education degree holders, unemploy-ment. Commissioning years excluded in all specifications. Inflations in capital account spending for construction and wastemanagement are higher, the smaller the population of a municipality. These resources might be taken from Public transportand lighting, whose coefficient is indeed negative and significant in small municipalities. The reduction of the police investmentis larger in cities with greater than 5,000 inhabitants. This result can be explained by the fact that the investment budget forpolice forces managed by large cities is significantly larger than those of small towns. The mafia has more interest in limitingexpenses for law enforcement where the latter can affect the productivity of police investigations.

58

A5 Additional Robustness checks

Table A11: Selection into treatment

PolicePublic transport &

lightingConstruction &

waste management Waste tax(1) (2) (3) (4)

Infiltration -0.0031** -0.022 0.0576*** -0.0713*(0.00136) (0.0141) (0.0185) (0.0101)

Controls Yes Yes Yes YesMunicipality dummies Yes Yes Yes YesYear dummies Yes Yes Yes YesObservations 1,407 1,866 1,452 1,256

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The analysis compare thespending categories of non infiltrated government with infiltrated governments, before and after the infiltration is terminatedby the national government. Commissioning period always excluded. Infiltration is a dummy equal to 1 if the government isinfiltrated by organised crime. Municipalities for which the the main reason for a) dissolution or b) the police investigationin the first place was related to distortions in the balance sheets (anomalies in spending patterns) are excluded from sample.Hence, the analysis is based on all municipalities for which the investigation for dissolution started for reasons unrelated topublic spending. Dependent variables by column: capital expenditure in municipal police (column 1); capital expenditure inpublic transport and lighting (column 2); capital expenditure in construction and waste management (column 3); waste tax(column 4). Estimation based on the dissolved sample (182 municipalities that have been dissolved once).

Table A12: Placebo test: total spending

Totalexpenditures

Totalcapital expenditures

Totalcurrent expenditures

(1) (2) (3)

Mafia-unrelated dissolutions 36.01 21.25 14.76(41.67) (38.19) (12.98)

Controls Yes Yes YesMunicipality dummies Yes Yes YesYear dummies Yes Yes YesObservations 18,305 18,305 18,305

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. Dissolution NoMafiarefers to mafia-unrelated dissolved governments. The estimation includes socio demographic controls (“other controls”) at themunicipal level including municipal population, industry employment, tertiary education degree holders, unemployment. Fullsample of 1350 municipalities from Campania, Calabria and Sicily: infiltration and commissioning years excluded.

59

Table A13: Placebo test: main results

Construction &waste management Police Waste tax

Public transport& lighting

(1) (2) (3) (4)

Mafia-unrelated dissolutions 0.018 -0.001 -0.0033 -0.014(0.011) (0.0009) (0.0086) (0.009)

Controls Yes Yes Yes YesMunicipality dummies Yes Yes Yes YesYear dummies Yes Yes Yes YesR-squared 0.177 0.178 0.811 0.781Observations 18,305 18,305 18,305 16,165

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. Mafia-unrelated dis-solution refers to mafia-unrelated dissolved governments. The estimation includes socio demographic controls (“controls”) atthe municipal level including municipal population, industry employment, tertiary education degree holders, unemployment.The estimation includes municipalities fixed effects and year fixed effects. Full sample of 1350 municipalities from Campania,Calabria and Sicily: infiltration and commissioning years excluded.

Table A14: Placebo test: capital account

Administration Social SectorConstruction &

waste managementPublic transport

& lighting Education Police(1) (2) (3) (4) (5) (6)

Mafia-unrelated dissolutions 0.0027 -0.0022 0.00085 -0.0012 0.0006 -0.00063(0.0023) (0.0017) (0.00213) (0.00091) (0.00112) (0.00076)

Controls Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesObservations 18,296 18,299 18,295 17,152 15,987 18,295

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. Dissolution NoMafiarefers to mafia-unrelated dissolved governments. The estimation includes socio demographic controls (“other controls”) at themunicipal level including municipal population, industry employment, tertiary education degree holders, unemployment. Theestimation is conditional to municipalities fixed effects and year fixed effects. Full sample of 1350 municipalities from Campania,Calabria and Sicily: infiltration and commissioning years excluded.

60

Table A15: Placebo test: current account

Administration Social sectorConstruction

& waste managementPublic transport

& lighting Education Police

Mafia-unrelated dissolutions 0.0027 -0.0022 0.00085 -0.0012 0.0006 -0.00063(0.0023) (0.0017) (0.00213) (0.00091) (0.00112) (0.00076)

Controls Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesObservations 18,296 18,299 18,295 17,152 15,987 18,295

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. Dissolution NoMafia refersto mafia-unrelated dissolved governments. The estimation includes socio demographic controls (“controls”) at the municipallevel including municipal population, industry employment, tertiary education degree holders, unemployment. The estimationis conditional to municipalities fixed effects and year fixed effects. Full sample of 1350 municipalities from Campania, Calabriaand Sicily: infiltration and commissioning years excluded.

Table A16: Test for spillover effect

Totalexpenditures

Totalcapital expenditures

Construction& waste management

Public transport& lighting Police Waste tax

(1) (2) (3) (4) (5) (6)

Infiltration -0.0124 -0.0665 0.585*** -0.0268* -0.00208* -0.020**(0.0223) (0.0784) (0.0197) (0.0147) (0.00123) (0.0102)

Controls Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesObservations 16,740 16,741 16,648 16,622 16,112 13,658R-squared 0.512 0.358 0.203 0.176 0.147 0.0530

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. All municipalitiessharing a border with those dissolved are excluded from the sample. The estimation includes socio-demographic controls at themunicipal level: municipal poopulation, industry employment, tertiary education degree holders, unemployment. The estimationis conditional to municipality and year fixed effects. Estimation performed with the restricted sampale of municipalities fromCampania, Calabria and Sicily that have experienced a dissolution; commissioning years always excluded.

61

Table A17: Violent attacks to politicians and infiltration

Infiltration One year before infiltration(1) (2)

Attacks to politicians 0.011 0.0075(0.014) (0.0146)

Controls Yes YesMuncipality dummies Yes YesYear dummies Yes YesObservations 5,313 3,993

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We test for the statisticalcorrelation between violent attacks towards politicians (“Attacks to Politicians”) in infiltrated municipalities (“Infiltration”).Attacks to Politicians is an indicator equal to 1 if there has been violence against local administrators in a given municipality.Infiltration refers to the infiltration dummy. Controls: municipal population, industry employment, tertiary education degreeholders, unemployment rate. We are using only the years 2010/2011/2012/2013 in the estimation because our main datasetstops in 2013. We are not exploiting the intensity of the attacks in this analysis. In some municipalities attacks are more thanin others and they are different in typology (from threatening letters to murders).

Table A18: Violent attacks to politicians and public spending

PolicePublic transport

& lightingConstruction

& waste management Waste tax(1) (2) (3) (4)

Attacks to politicians -0.003 0.092 -0.00011 -0.336(0.0028) (0.0817) (0.0255) (0.559)

Controls Yes Yes Yes YesMuncipality dummies Yes Yes Yes YesYear dummies Yes Yes Yes YesObservations 5,027 5,021 5,221 5,021

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. We exploit our difference-in-differences setting and we run our main specification (equation 1) on our main spending results, but measuring organisedcrime with a measure of violence of politicians (“Attacks to politicians”). Attacks to Politicians is an indicator equal to 1 ifthere has been violence against local administrators in a given municipality. The outcome variables are: capital expenditurefor municipal police (column (1)), capital expenditures for public transport and lighting (column (2)), capital expenditure forconstruction and waste management (column (3)), waste tax (column (4)). Controls include: municipal population, industryemployment, tertiary education degree holders, unemployment rate. We are using only the years 2010/2011/2012/2013 in theestimation because our main dataset stops in 2013. We are not exploiting the intensity of the attacks in this analysis. In somemunicipalities attacks are more than in others and they are different in typology (from threatening letters to murders).

62

A6 Mafia and Politics

Table A19: Infiltration and political factors

Single candidate Last mandate Right party Left party Centre party Civic list

(1) (2) (3) (4) (5) (6)

Infiltration 0.047** 0.189** 0.091** -0.049 0.033 -0.037(0.019) (0.05) (0.051) (0.049) (0.0356) (0.0398)

Controls Yes Yes Yes Yes Yes YesNatGov (Left) Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes YesObservations 2,471 2,515 2,471 2,471 2,471 2,466R-squared 0.306 0.251 0.483 0.479 0.414 0.633

Note: Clustered standard errors at municipal level in parentheses; *** p<0.01, ** p<0.05, * p<0.1. Infiltration refers to theinfiltration dummy. Outcomes variables: Single candidate takes value 1 if only one candidate is running for elections; Lastmandate takes value 1 if the incumbent is running for the second and last mayoral mandate; Right/Left/Centre party takesvalue 1 for the political colour of the winning party (Right Party, Centre Party, Left Party), governments whose administrationcannot be classified are excluded; Civic list takes value 1 if the winning party is a civic list. Sample of all municipalities fromCampania, Calabria and Sicily that experienced at least one government dissolution for mafia infiltration. The estimationsinclude socio-demographic controls, municipality Fe, year Fe.

Table A20: Infiltration and public spending, controlling for political factors

Construction &waste management Police

Public transport& lighting Waste tax

(1) (2) (3) (4)

Infiltration 0.0526*** -0.003** -0.0236* -0.0213**(0.0160) (0.00142) (0.0131) (0.011)

Single candidate -0.0883* -0.000171 0.0322 -0.00133(0.0486) (0.00243) (0.0429) (0.0334)

Last mandate -0.0064 0.00097 -0.00174 -0.00079(0.0157) (0.00136) (0.0141) (0.0211)

Right party 0.00528 0.000356 0.0182 0.0355(0.0140) (0.00198) (0.0126) (0.0537)

Controls Yes Yes Yes YesMunicipality dummies Yes Yes Yes YesYear dummies Yes Yes Yes YesObservations 2,808 2,736 2,780 2,301R-squared 0.228 0.235 0.151 0.092

Note: Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. Sample of municipalities having experiencedat least one government dissolution for mafia infiltration. Outcome variables: capital expenditure for construction and wastemanagement (column (1)), capital expenditure for municipal police (column (2)), capital expenditure for public transport andlighting (column (3)), waste tax (column (4)). Estimates replicate Table 3 but controlling for political factors.

63

Table A21: Political factors and public spending components

PoliceConstruction &

waste management Waste taxPublic transport

& lighting

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12)Single candidate -0.075 -0.0007 -0.0045 0.026

(0.048) (0.0025) (0.035) (0.042)Last mandate 0.00068 0.0006 -0.00014 -0.0017

(0.016) (0.00136) (0.017) (0.014)Right party 0.0107 0.00011 0.0330 0.0151

(0.014) (0.0019) (0.053) (0.0123)Controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesMunicipality dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesYear dummies Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes YesObservations 2,808 2,850 2,808 2,736 2,772 2,736 2,301 2,330 2,301 2,780 2,819 2,780R-squared 0.224 0.221 0.223 0.234 0.233 0.234 0.092 0.092 0.092 0.149 0.148 0.150

Note: Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. Sample of municipalities having experienced at least one government dissolution formafia infiltration. Outcome variables: capital expenditure for construction and waste management (column (1)), capital expenditure for municipal police (column (2)),capital expenditure for public transport and lighting (column (3)), waste tax (column (4)). single candidate: one candidate running for elections, last mandate: mayorrunning for her last mandate, Right party: elections won by right-wing party.

64

Regression Discontinuity Design

Let Xm,t be the vote share of the right-leaning candidate minus the vote share of the non-right candidate, Rm,t be the treatment dummy variable referring to electoral victories ofright-wing parties, and Pr(Inf)m,t the probability of infiltration. We then have Rm,t = 1if Xm,t > 0 and Rm,t = 0 if Xm,t < 0. We focus on the set of electoral races where Xm,t islower than a bandwidth h, such that the outcome of those races can be considered as good asrandom. Our treatment effect is the average difference between Pr(Inf)m,t of a municipalitywhere the right narrowly wins and Pr(Inf)m,tof a municipality where the right is narrowlydefeated.

Pr(Inf)m,t = αRm,t + f(Xm,t) + εm,t (5)

for all electoral races, such that −h < Xm,t < h

with Rm,t = 1 if Xm,t > 0,and Rm,t = 0 if Xm,t < 0

We estimate

α = E[Pr(Inf)m,t|Rm,t = 1]− E[Pr(Inf)m,t|Rm,t = 0]. (6)

α is estimated both parametrically and non-parametrically. The optimal bandwidth used is0.075, meaning that the sample is made up of governments whose election was characterisedby a difference in votes - between the right-wing party and other parties - below 7.5%.39Inorder to obtain reliable RDD estimates, we need to ensure that there is an absence of non-random sorting around the cutoff. To this end, we perform a McCrary density test, makingsure that there is no significant jump in the density of observations at the cutoff point. FigureA19 exhibits a very small discontinuity at the threshold, which is statistically insignificant.

39In the choice of optimal bandwidth (h), we face a trade-off between efficiency and bias. With very smallbandwidths, we are more likely to approximate the quasi-experimental assignment of the treatment variableand to attain balance in the other observable covariates. Very small bandwidths often, however, lead to smallsample problems and imprecise estimates. To address this issue, we use the optimal bandwidth proposed byCalonico et al. (2014), which addresses the bias in the confidence interval and the point estimator.

65

Figure A6: RDD - right-wing party victory and probability of infiltration

-.05

0.0

5.1

.15

Pr In

filtra

tion

-.1 0 .1Vote_share

Note: Polynomial fit of order 2. vote share>0 refers to elections won by right-wing parties; vote share<0 refers to electionsbarely lost by right-wing parties

66

Figure A7: RDD - right-wing party victory, spending components and waste tax

Note: Polynomial fit of order 2. vote share>0 refers to elections won by right-wing parties; vote share<0 refers to electionsbarely lost by right-wing parties

Table A22: RDD - right-wing winner and infiltration

Non Parametric Parametric

(1) (2) (3) (4) (5)

Right-wing winner 0.075* 0.0846* 0.0722** 0.0722** 0.101*(0.04) (0.0524) (0.0366) (0.0365) (0.0604)

Bandwidth 0.0751 0.0751 0.0751 0.0751 0.0751Observations 911 911 911 911 911

Note: Clustered standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. Forcing variable coefficients not displayed.Column 1: rddrobust Linear; column 2: rddrobust Polynomial; column 3: linear regression with kernel weights; column 4:linear regression varying linear slopes; column 5: polynomial regression of order 2 with interaction with the forcing variable.All the estimations use Calonico, Cattaneo and Titiunik (2014) optimal bandwith.

67

Table A23: RDD - right-wing winner and public spending components

Construction & waste management Police Waste tax Public transport & lighting

(1) (2) (3) (4)

Right-wing winner -0.02 0.048 0.0641 -0.0738(0.0263) (0.0551) (0.0564) ( 0.0524)

Bandwith 0.0751 0.0751 0.0751 0.0751Observations 620 620 620 620

Note: Robust standard errors in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. The outcome variables are respectively capitalexpenditure for construction and waste management, capital expenditure for municipal police and waste tax. Forcing variablecoefficients not displayed. All columns are estimated with rddrobust linear.

Table A24: RDD - Balance of covariates

UnemploymentIndustry

EmploymentHumanCapital Population

TotalSpending

WhiteBallots Turnout

Non ValidBallots

Right-wing victory -0.5940 0.48 -0.0919 -0.269 -0.0195 0.129 -2.397 0.8(0.795) (0.551) (0.670) (0.364) (0.00233) (0.306) (2.428) (0.520)

Observations 620 620 620 620 614 619 621 619

Figure A8: McCrary Test

01

23

45

-.5 0 .5

Note: Clustered standard errors at municipality level in parenthesis; *** p<0.01, ** p<0.05, * p<0.1. McCrary test witht-student at the discontinuity -0.9782 with robust estimation. There is no presence of non-random sorting at the cutoff.

68