criminal_procurement

RULES, DISCRETION, AND CORRUPTION IN PROCUREMENT:

EVIDENCE FROM ITALIAN GOVERNMENT CONTRACTING

Francesco Decarolis

Raymond Fisman

Paolo Pinotti

Silvia Vannutelli

WORKING PAPER 28209

NBER WORKING PAPER SERIES

RULES, DISCRETION, AND CORRUPTION IN PROCUREMENT:

EVIDENCE FROM ITALIAN GOVERNMENT CONTRACTING

Francesco Decarolis

Raymond Fisman

Paolo Pinotti

Silvia Vannutelli

Working Paper 28209

http://www.nber.org/papers/w28209

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue

Cambridge, MA 02138

December 2020, Revised February 2023

We thank seminar audiences at Kellogg School of Management - Northwestern University,

University of Montreal and SIOE 2019. We also thank Juan Ortner and Giancarlo Spagnolo for

helpful comments. Decarolis gratefully acknowledges financial support from the European

Research Council (ERC-2015-StG-679217). The views expressed herein are those of the authors

and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been

peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies

official NBER publications.

reserved. Short sections of text, not to exceed two paragraphs, may be quoted without explicit

permission provided that full credit, including © notice, is given to the source.

Rules, Discretion, and Corruption in Procurement: Evidence from Italian Government Contracting

Francesco Decarolis, Raymond Fisman, Paolo Pinotti, and Silvia Vannutelli

NBER Working Paper No. 28209

December 2020, Revised February 2023

JEL No. D72,D73,H57,K42

ABSTRACT

The benefits of bureaucratic discretion depend on the extent to which it is used for public benefit

versus exploited for private gain. We study the relationship between discretion and corruption in

Italian government procurement auctions, using a confidential database of firms and procurement

officials investigated for corruption by Italian enforcement authorities. We show that discretionary

procedure auctions (those awarded based on negotiated rather than open bidding) are associated

with corruption only when accompanied by limits to competition. We further show that, while

these “corruptible” discretionary auctions are chosen more often by officials who are themselves

investigated for corruption, they are used less often in procurement administrations in which at

least one official is investigated for corruption. These findings fit with a framework in which more

discretion leads to greater efficiency as well as more opportunities for theft, and a central monitor

manages this trade-off by limiting discretion for high-corruption procedures and locales. Overall,

our results suggest that competition may allow procurement authorities to extract the benefits of

discretion while limiting the resultant risks of abuse.

Francesco Decarolis

Bocconi University

Via Sarfatti 25

Milan, 20136

Italy

and EIEF

[email protected]

Raymond Fisman

Department of Economics

Boston University

270 Bay State Road, 304A

Boston, MA 02215

and NBER

[email protected]

Paolo Pinotti

Bocconi University

Via Rontgen 1

20136 Milan

ITALY

and fRDB and also BAFFI-CAREFIN Centre

[email protected]

Silvia Vannutelli

Kellogg Global Hub

Northwestern University

2211 Campus Drive

Evanston, IL 60208

and NBER

[email protected]

I Introduction

Governments often face a trade-oﬀ in the oversight and constraints they impose on

lower-level bureaucrats in carrying out their function s. Oﬃcials may use discretion to

better serve the public’s interests, or exploit it for personal gain. The appropriate level of

discretion depends on the beneﬁts of an agent’s informational a d vantage relative to the

costs from his exploiting discretion for personal gain. From a public welfar e perspective,

the agency problem is complicated by yet another layer of delegation – politicians or high-

level oﬃcials who determine the extent of discretion available to lower-level oﬃcials may

be overly risk-averse, to the extent th at the electorate is more attentive to corruption

scandals rat h er than an eﬃcient provision of public goods. Such incenti ves – whether

electoral or p r om o ti o n -r el at ed – may then lead to insuﬃcient delegation and discretion.

In this paper, we study both the determinants and consequences of discretion in the

context of government procurement in Italy. Procurement accounts for a la rg e fract i o n of

government expen diture worldwide; for example, for OECD co u ntries the procur em ent-

to-spending ratio held steady at around 30 percent during 2007-2015 (OECD [2017]).

urthermore, corrupti on is thought to result in substantial “leakage” from procur ement

expenditures, even in more developed (an d less corrupt) countries.

Thus, understanding

how procurem ent ru l es m i ght im p a ct co r ru p t i o n is of interest in its own right, in addition

to serving as an apt setting for studying the trade-oﬀs associated with discretion in

government bureau cr a ci es more generally.

Our work i s enabled by the use of a conﬁdential database obtain

ed from the Agenzia

Informazioni e Sicurezza Interna (AISI), the Italian equivalent of the FBI. The database

lists individ u al s that have been ﬂagged by the AISI as su spected of various crimes, in-

cluding corruption. By linking this list to administrative data on the top employees and

owners of Italian companies, we classify a ﬁrm as investigated for corruption if at least one

employee or owner was ﬂagged by the AIS I for suspected corruption. We then link the

resultant ﬁrm-level database to information on over 200,0 00 procurement auctions for the

construction and maintenance of public infrast r u ctur e held thr ou g h o u t Italy duri n g 2000-

2016. The data inclu d e the near-universe of auctions involving the two most fr equ ently

procured types of contracts: those involving either civi c buildings or roads, highways, and

bridges. These data allow us to observe whether investigated ﬁrms participated in or won

each auction. Finally, we complement these ﬁrm-level data wi t h similar information on

A study sponsored by the European Commission reports that, in projects t h at were found to have

been corr u pt ed , 13 percent of expenditures were lost due to corruption (Ferwerda and Deleanu [ 2013] ) .

Agency problems in procurement extend beyond governments. For example, in a recent working

paper, Bergman et al. [2021] documents the adverse consequences of favoritism in hospitals’ pr

ocurement

of medical services.

investigations for corruption cha r ges involving the publ i c oﬃcials in charge of awarding

(and follow-on monitoring) the contracts in our data (we use the same terminol o gy of

“investigated” an d “clean,” or “non-investiga t ed , ” that we use for businesses also for the

public oﬃcials in charge of the auctions). We know o f no other database of corruption

risk for ind i v i d u al s and organizations that is com p ar a b l e to ours.

The scale and richness of our data are such that we may employ a range of ﬁxed

eﬀects and controls, which helps to rule out a number of alternative interpretati on s,

which inevita b l y ar i se in cor r el at i o n al r esu l t s. For example, in our anal y ses tha t l oo k

at the characteristics of auct i on s won by ﬁrms under investigation for corruption, we

may include over 6,000 procurement authority (PA) ﬁxed eﬀects, so tha t we identify the

relationship based on the selection of diﬀerent auct i on mechanisms by the sa m e entity

(e.g., a municipality), or PA-year ﬁxed eﬀects so that we identify the relationship based

on the selection of diﬀerent auction mechanisms in the same place during the same year.

The latter speciﬁcation allows us to account for any potential unobser ved time-varying

shocks at the procurem ent authority level.

We begin by examining the types of auctions that are m o st often won by investigated

ﬁrms. We show that two aucti on arrangements are signiﬁcantly more likely to lead to

a contract being awarded to an investigated ﬁrm: ﬁr st , so-called scoring rule auctions,

which involve (potentially subjective) non-price selection criteria that may restrict com-

petition, are 1 percentage point (6 percent) more likely to be won by investigated ﬁrms,

relative to ﬁr st-p r i ce (non-d i scr eti o n a ry ) aucti o n s. Au cti o n s that use “negotiated ” proce-

dures in which procurement oﬃcia l s invite b i d d er s ( r a th er than allow for open bidding)

are no more likely to be won by ﬁrms investigated for corruption, relative to open auc-

tions. However, when we look at the subset of negotiated auctions in which oﬃcials fail

to invite th e req u i si t e number of bidders (which we take to be an ind i ca t i on o f a b u se

of discretion), we ﬁnd a 1.9 percentage point (11 per cent) higher probability of an in-

vestigated winner. While mor e at risk of selecting investigated ﬁrms, we also ﬁnd that

scoring rule auctions are associ a t ed with lower cost overruns and higher award prices,

while negotiated procedures are associat ed with lower delays and higher award p r i ces.

In lin e with evidence provided by Bosio et al . [2022] , we interpret these features as an

ndication of improved contract execution.

We then link the choice of discretionary auctions to the chara ct er i sti cs of procure-

ment administrators that dep l oy them. In particula r , we look at whether the choice of

discretion is aﬀected by whether the auction was administered by an individual that the

AISI has ﬂagged as suspect ed of corruption, and also whether the auction occurred in a

municip al i ty in which the AISI has identiﬁed at least one such oﬃcial. The ﬁrst of these

analyses aims to examine whether individual procurement oﬃcials prone to corrupti o n are

more likely to select (corruptible) discretionary auctions; the second examin es whether

locales where suspected corruption is present tend to use “corruptible” discretionary auc-

tions. Our results show eﬀects that go in opposite directions: public o ﬃ ci al s suspected of

corruption are 2.9 percentage points more likely to use one of the two discretionary auc-

tion types we ﬂag for concern (discretionary criteria or di scret i on a r y procedures with too

few invited part i ci p a nts). By contrast, discretiona r y aucti on s are 1.9 percentage points

less common in “corruption-suspected” munici p a l i t i es.

We describe how our results may ﬁt with an intuitive explanation based on classic

models of d eleg a ti o n put forward by Holmstrom et al. [1982] and applied to the bureau-

ratic delegat i o n problem by Epstein and O’hallo ran [1994]. In our context, greater dis-

retion allows for more eﬃcient implementation of government projects by well-informed

and well-i ntentioned procur em ent oﬃci al s, which must be traded oﬀ again st the higher

probability of leakage by corrupt oﬃ ci a l s. If the choice of auction d esi g n i s o n e o f the pri-

mary means of oversight by a (non-cor r u p t) central monitor, then less discretion will be

allowed in locales where the probability of corru p t i on is higher. When possible, however,

corrupt oﬃcials deploy discretion, to th e beneﬁt of cor r u p t ﬁrms.

Overall, the empirica l ﬁndings in t h i s study oﬀer a new, detailed assessment of the

extent of – and the mechanisms involved in – corruption i n infrastructure procurement.

On the fundamental question of whether a central legislat u r e or senior bureaucrat chooses

to impose excessively strict constrai nts on lower-level oﬃcials, while our analyses do

not allow for deci si ve welfare calculations, we argue that the data provide suggestive

evidence of overly strict constraints. This argument is exempliﬁed by the con seq u en ces

of a mid-2000s reform in which the Italian legislature loosened regulations governing

the use of negotiated procedures. Whereas such contracts could onl y be deployed for

relatively small projects (under e300,000) in the early 2000s, by 2011 the l i m i t had been

raised to e1,000,000. This change, motivated by the government’s desire to stimulate the

economy by reducing the procedural times to award public contracts, led t o a massive

increase in the share of auctions held via negotiated p r oced u r es, from 10 percent in 2006

to 60 percent by 2012. Yet the vast majority of these (83 percent) were conducted

using mechanisms that preserved supplier competition (i.e. , with th e legally required

number of b i d d ers) , and hence the lo o sen i n g of rules had at most a very small eﬀect

on the fraction o f contracts awarded to ﬁrm s under investigation for corruption. And in

locations in which oﬃcials might have exploited discretion, their use was relatively limited.

Indeed, calculations based on our estimates imply a 0.05 percent incr ease in investigated

winners overal l between the periods before and after the increase of the t h r esh ol d for using

negotiated procedures. This appears to be a small cost when compared to improvements

in contracting quality from discretion, such a s a 14 percent reduction in delays.

Thus, the primary implication of our analysis is that supplier competition may play a

central role in curtailing the corruption risk that may accomp any greater buyer discretion .

II Liter at ure

Our paper sits at the intersection of several distinct l i t era t u r es, and we organize our

discussion of this related work around what we see as our ﬁve main contributions.

Taken as a whole, our results suggest that greater discretion had only a limited

impact on corruption (but did reduce d el ays, and plausibly also costs). This ﬁrst contri-

bution is relevant to our understanding of the eﬃciency-corruption trade-oﬀ in del ega t i on .

The seminal study of Banﬁeld [1975] observed t h at reducing discretion may limit corrup-

ion, albeit at the expense of constraining honest public oﬃci al s from exercising their

judgment to the beneﬁt of public welfare. This links to the rich and exten si ve literature

on government decentralization and delegation. Hu ber and Shipan [20 0 6] and Bend or

t al. [ 2 0 01 ] provide earlier overviews of th i s body of research; we see our work as cor-

responding to their models of “ex-ante constraints” (as in the reduced use of discretion

that we study here) rather than ex-po st monitorin g . More closely related, Bo si o et al.

[2022] shows, using cross-country data, that constraints on discr

etion are asso ci at ed with

better procurement outcomes, but o n l y in countries with low public sector capacity.

Our second contribution is a new measurement of corru p t i on in

public contracts that

is plausibly more credibl e and more accurate than pri o r measures. There is a vast and

growing body of work on th e political and economic analysi s of corruption ( see Olken and

ande [2012] and Burguet et al . [2016] for recent surveys of corruption that review and

synthesize various models of delegation), which reﬂects the potential importance of cor-

ruption to the functi o n i n g of government, and the correspondingly substantia l resources

devoted to ﬁghting corruption. Thus, we see it as a useful contribution to be able to quan-

tify that 17 percent of public works in Italy are awarded to investigated ﬁrms.

Our third

Related work by Bandiera et al. [2020] investigates delegation in public procurement by experi-

mentall y varying the amount of autonomy granted to procurement oﬃcers. They ﬁnd that shifting

decision-making rights from monitors to oﬃcers reduces procurement pri ces. Wh i l e our analysis also

indicates that discretion improves procurement outcomes, our study focuses on a diﬀerent type of chan-

nel (the choice of award procedures and criteria) and a diﬀerent out come (the risk of selecting criminal

contract ors) . Similarly, two studies concurrent with our own provide evidence that the expanded use

of discretion can improve [Carril, 2019] or worsen citepszucs2018 procurement outcomes. These studies

ocument signiﬁcant bunching of co ntracting activities below the discretionary threshold, suggesting a

role for buyers manipulation, while we do not detect any such distortions. More broadly, given the

monitoring function of higher-level governments, our ﬁndings also relate to the literature on th e costs

and beneﬁts of decentralization (e.g., Bardhan and Mookherjee 2006).

ur work al so relates to studies lin ki n g procurement to ﬁrm political connections, although our

measure of corruption risk is clearly distinct. Mironov and Zhuravskaya [2016] document how ﬁrms

ith public procurement revenue increase the tunneling of funds to poli t i ci ans around elections. They

contribu tion concerns the strengths and weaknesses of diﬀerent procurement method s to

limit corruptio n risks. Our ﬁndin g that discretion has limited impact overall on corruption

is in line with Band i era et al. [2009], who analyze centralized versus decentralized public

rocurement and sh ow that excessive payments for standardized goods are driven more

by ineﬃciency than corruption.

Our result s provide evidence on a well-deﬁned source of

neﬃciency, namely excessively rigid contracting procedures. Several other studies link

procurement methods and oversight to project ou t com es. Notable contributions include

Brierley [2020], who shows that greater oversight may backﬁre if politician

s themselves

are corrupted (a result in t h e spirit of the classic study of hierarchical corruption in Indian

canals by Wade [1982]), Lewis-Faupel et al. [2016], who document the positive impact

f e-procurement on road quality in India and on execution time Indonesi a, possibly by

limiting interactions with corr u p t public oﬃcials, and Djankov et al. [2017], who docu-

ment the correlation across countries in procurement rules and practices and link these

to survey-based measures of road quality. The central role of competition in curtail-

ing corruption that we uncover parallels the recent work of Colonnelli and Prem [20 1 7] ,

hich also points to the role of limited competition in creating rent-seeking beh avior in

Brazilian procurement. At the macroeconomic level, these are key results for the larger

objective of assessing the quality of ﬁscal policy, as underscored by the recent interest in

opening up the black box of “Big G” [Cox et al., 2020].

he fourth contribution relates to th e heterogeneous impact of pr ocu r em ent rules

across diﬀerent public organizations. In particular, we show that discretionary auctions

are rel at i vely rare in high-corruption areas, but are commonly deployed by individual

administrators under investigation for corrupt i on . While these two ﬁndings are, at least

superﬁcially, in tension with one another, as we discuss below they follow from a simple

model that is very much in line wi t h standard theories of delegation.

Overall, our results indicat e that governments ar e aware of the trade-o ﬀ created

by discret i o n , and take it into a ccou nt i n the extent to which it i s allowed in diﬀer ent

areas. This latter ﬁnding was suggested by Coppier et al . [2013], who noted that there is

reater discretion in (low-corruption) U.S. and U.K. procu rem ent. Coviello et al. [2017],

in their investigation of the economic im p a ct s of allowing greater discretion in the public

also document that more corrupt local es tend to award contracts to less productive ﬁrms. Auriol et al.

[2016] show that politically connected companies are more likely to win auctions with limited competition,

which they take to be an indication of corruption. A similar approach is taken by Balt r u nai t e et al. [2018]

in the setting of Italian auctions, in linking political connections to discretionary auctions. Brogaard

t al. [2016] show that contracts won by politically connected ﬁrms i n the U.S . te nd to have p oorer

performance. Our work is distinct from these earlier eﬀorts in a number of ways. Most importantly, we

have an unusual country-wide measure that allows us to identify ﬁrms as potentiall y corrupt.

Along simil ar lines, several recent studies have shown that limiting the discretion of procurement

oﬃcials is most valuable when the skills or abilities of the public b uyers are lower; see Best et al. [2019],

Bucciol et al. [2020] and Decarolis et al. [2020].

procurement of works in Italy, also notice that higher-corruption provinces tend to use

less discretionary auctio n procedures. We are, to our knowledge, the ﬁrst to identify this

relationship systematical l y based on l ocal variation in corrupti on .

Our ﬁfth and ﬁnal contribution is to the debate on anti-corruption poli ci es in public

procurement. While there is much th eo r eti ca l work in this area (see, e.g., Ortner and

hassang [2018], for one recent contribution), th er e ar e scant emp i ri ca l ﬁ ndings. The

few exceptions include Olken [2007], which provides a compar at i ve an al y si s of centralized

udits versus grassroots participation in monitoring; Di Tella and Schargrodsky [2003],

which presents eviden ce on the combined eﬀect of p u b l i c oﬃci a l s’ wages and corr u p t i o n

audits; and Avis et al . [2018], which provides causal estimates of the eﬀects of past

nti-corrup t i o n audits on subsequent corruption levels. Our ﬁndings on the role of ﬁrm

competition to limit the corruption risk of discretionary auction procedures and criteria

are relevant for this policy-r el evant research agenda. We return to policy considera t i on s

in our concl u si o n.

III Background and data collection

III.A Institutional details on Italian procurement

In recent years, Italian regula ti o n s that govern public procurement und erwent a

number of reforms as a result of, among other thi ng s, the passage of European Union

Procurement Directives aimed at creating a common set of rules for public procurement

in the EU. In particular, these reforms aimed to improve the design of source selection sys-

tems, i.e., the process for evaluating bids. We study public contracts under the “or d i n a r y

regime,” which sets the procurement r u l es for most projects, excluding secret military

contracts and some str at eg i c infrastructure projects.

Source selection systems under the o r d i n ar y regim e vary along two main dimension s:

the awarding procedure and the selection criterion. S ta r ti n g with the ﬁrst dimension ,

there are two primary procedures for awarding contracts: open aucti on s and negotiations.

Open auctions are “ordinary” procedures for the assignment of procurement contracts

in which all ﬁrms eligible to execute public contracts can bid. In these procedures, the

contractin g oﬃcer overseeing the project has little discretion in the choice of contract o r.

These auctions presu m e the feasibility of accurately deﬁning, from the outset, th e relevant

scope and technical speciﬁcations of the contract.

Negotiated procedures are, by contrast, marked by sign i ﬁ cant discretionary powers.

The contracting oﬃ cer consults a set of prospective contractors and may negotia t e the

conditions of the contract wit h one or more of them. Given their discretion ar y nature,

negotiated p r ocedures are treated as exceptional, and admissible only under speciﬁc con-

ditions: for the most part, they are permitted only for contracts below a given monetary

threshold. Above this threshold, negotiations are allowed only when there is some ur-

gency in fulﬁlling the contract, or when a previous attempt to run an open auction failed

to elicit any bids.

The second key aspect of contracting is the speciﬁcation of the criterion for determin-

ing the winner. Both open and negotia ted procedures can use either the “lowest pri ce”

criterion or a “scoring rule” criterion (also known as the “most economi ca l l y advantageous

tender” criterion). In the ﬁrst case, the enterprise oﬀering the lowest price is awarded the

contract, provi d ed t h a t t h i s oﬀer i s ju d ged to be reliable, that is, th e oﬀer i s n o t so l ow

as to be unrealistic. The second approach allows for the accounting of a broader range

of considerations beyond price, as speciﬁed in the call for ten d er . Non-price parameters

of a b i d may incl u d e both hard and soft elements. An exa m p l e of a quantitative (hard)

parameter could be the number of engineers that will work on the speciﬁc project, while

an example of a soft element is the aesthetic quality of the proposed solu t i o n . There are

a few limits that regulation s place on the choice of parameters. In particular, criteria

must all pertain to the bid and not the ﬁrm, so that past performance cannot be used

as a parameter. But procurement oﬃcials enjoy wide margins of discretion in setting the

parameters (possibly to the advantage of speciﬁc ﬁr m s) and their associated weights.

The inclusion of multiple parameters ca n be used to restrict c

ompetition, to the extent

that only a narrow set of ﬁ rm s may be able or willing t o participate in the ‘restricted’

auction. Indeed, our data conﬁrm a l ower level of competition in scoring rule auctions –

see Figure A3.

As one might expect, the full set o f regulations governing proc

urement i s far more

complex than we can describe here, and we defer to Decarolis and Giorgiantonio [2015] for

more in-depth discussion. However, we observe that, beyond some modest diﬀerences,

the set of procedures and criteria governing Italian procurement are quite general, allow-

ing only for limited regional variation. Thi s is impor t a nt, in particular, as it is diﬃcult

for individual regions to create r u l es that favor local ﬁrms, which would present a con-

found for our analysis. Indeed, given the const ra i nts laid down by the Eur o pea n Union,

Italian procurement rules also characterize the institutional framework in the EU more

An illustrative example may help convey t hi s point. In 2007, the Italian Supreme Court conﬁrmed

the conviction of a group of public oﬃcers and business owners for rigging multiple scoring rule auctions

in the Santa Maria Capua Vetere municipality. The scheme involved public oﬃcials drafting calls for

tenders following the recommendations of favored ﬁrms: parameter s in the scoring formula emphasized

elements that advantaged pre-identiﬁed ﬁrms, e.g., by specifying the use of a speciﬁc b ran d of machinery.

The extensive use of scoring rules in favoring bribe-paying ﬁrms has been well-documented by Cam-

os et al. [2021], in thei r analysis of the massi ve corruption case involving Brazilian m

ultinational Ode-

brecht.

generally. But they also reﬂect procurement rules in a much broader set of co u ntries, as

documented in a recent survey by the

World Bank [2 01 7] .

One particular feature of procurement rules does warrant further elaboration, given

our focus on del eg at i o n and discretio n by individual procurement oﬃcial s. Whenever not

expressly constrained by national or loca l rules, the choice of both the awardi n g procedure

and the selection criterion is delegated to the contracting oﬃ cer overseeing each contract

(the “Responsabile Unico del Procedimento”, or RUP). This publ i c oﬃcia l is selected from

among management-level bureaucrats in the relevant public administration (PA), unless

none is available for this ro l e (in which case special rules apply). The RUP is nominated

via a formal act by the PA’s top oﬃcial, which in municip al i t i es is the mayor.

The RUP is in charge of man a gi n g the entire contracting process, from the project

deﬁnition phase, throu gh the bidding phase, to the awarding and realization of the con-

tract. Thus, the RUP has consi d er a b l e control over how the contract is structured. But

this discretion has to be exercised within the regulatory constraints imposed by Euro-

pean, nati on a l , and local regula ti o n s, and it is subject to oversight both internally within

the PA, and from third-party auditors (at the local, national and, in certain cases, Eu-

ropean level; see Figure

A1). A RUP who wishes to use a discretionary procedure or

riterion may aim to be appointed to oversee auctions that are amen ab l e to such meth-

ods, and conditional o n the project may select more discretionary approaches. However,

it is diﬃcult to make strong inferences about a RUP’s intent merely from the selection

of discretionary auctions. A socially-motivated procurement oﬃcia l may also choose a

negotiated procedure to exped i te project execution and (with the interests of the munic-

ipality at heart) even manipulate contract amounts to facilitate their use. We thus rely

on detailed data on RUP and ﬁrm s described bel ow to discern whether di scr eti o n is more

plausibly used for self-serving reasons.

III.B Data

III.B.1 Procurement Data

Our procurement data come from a database provided by the Public Contracts Ob-

servatory at the Italian Anticorruption Authority (ANAC), the public entity that oversees

public procur em ent in Italy. Since 2000, ANAC has mon i to r ed all public contracts above

the threshold reserve price of at least e150,000 until 2010, and e40,000 thereafter.

Our dataset contains the universe of ANAC data for the years 2000-2016 for public

infrastructure.

Amongst these, t h o se involving civic bui l d i n g s (OG0 1 ), or tra

nsporta-

Italian procurement more broadl y can be divided into three categories: works, goods, and services.

Our focus is on public works, which represent around 25 % of the value and over 30% of the total

tion infrastructure such as roads, highways, and bridges (OG03), a r e th e m ost relevant

categories which, combined tog et h er , rep resent more than half of all contracts, both in

terms of number of contracts as well as money spent.

For each contract, we have de-

ailed informatio n about the contracting phase, including the start and end date of the

bidding process, the type of contracting authority, the au ct i o n procedur e used to award

the contract, the selection criterion, the number of bidders, and the identity of the win-

ning bidder. The data also include information on auction outcomes, such as the initial

project value, the winni n g discou nt and total eﬀective costs, the expected and eﬀective

contract duration and, for auctions hel d after 2010, we observe all of t h e bids.

We observe 5 types of contractin g author i t i es in the data: centr al admin i st r a ti o n s,

municip al i t i es, other local admini st ra t i on s (regions and provinces), state-owned enter-

prises, and “decentralized administ ra t i on s” (speciﬁcal l y, hospi t a l s and universities). For

each author i ty, we know the i d entity of the RUP managi n g each contract, and for most

contracts we also know the exact geographic location (the except i o n s include central gov-

ernment administrations, decentralized regional administrations (such as hospitals and

universit i es) , and al so highways and railways that span geographic boundaries). Local

institutions – municip a l i t i es in particula r – play the largest role in public works pro-

curement. Local governments account for 72% of total projects awarded (53% municipal

councils, 14% provincial councils, 3% regional governments). While about half of the

contracts in our da ta b a se are awarded by municipal councils, they are relatively small

projects, with an average value of e5 2 7, 0 00 , as compared to an average value of e847,000

for p rovincial an d regional governments, and over e1.5 million for hospi t al s and univer-

sities. There is also a wid e range in the number of contracts per contracting authority.

There are 1,266 municipal cou n ci l s that awarded only a single contract (mean population

of 1,404), wh er eas the municipality of Rome alone awarded 3, 51 9 contract s.

As previously not ed , the contracting authority ca n choose between two main types

of awarding procedures, open and negotiated. If the latter is selected, we addi t i on a l l y

observe the number of ﬁrms invited to participate in the auction, and for all auctions,

we see the number of ﬁrms that present oﬀers (the number of bi d d i n g ﬁrms is, by def-

inition, less tha n or equal to the number of invited oﬀers). Furthermo r e, we observe

the identity of the winning ﬁrm and, for auctions held after 2010, also the i d entities of

number of procurement contracts in Italy. As noted below, the average contract size is around e985,000,

larger than the average size for goo d s contracts but almost half of the average size for ser v ice contracts.

Consistent with the size diﬀerence, negotiated procedures are used signiﬁcantly more frequently for public

works (they represent around 72 % of the total) than in the procurement of services, where they are used

in only 60% of contracts.

The procurement of public infrastru ct ur e is subdivided by law into 13 job types (OG01,...,OG13).

Although the data contain codes that refer to more detailed sub-categories, OG codes are more reliable

since this latter classiﬁcation is the only one requi r ed by law.

all participants. Under normal circumstances, negoti a ted procedur es require a minimum

number of invitations. When we observe fewer than the legally mandated number of

invitati o n s, we ﬂag the auction as involving potential abuse of discretionary procedures

(denoted by the variable DiscretP roc

lowN

). Conversely, we denote as D

iscretP roc

highN

negotiated auctions with the legally mandated number of bidders. Finally, we denote all

negotiated p r oced u r es (both highN and lowN) by the variable DiscretP roc. Note that

a below-minimum number of invited bidders does not automatically indicate abuse – it

may instead result fr o m a contract’s urgency or a lack of qualiﬁed ﬁrms.

Auctions m ay be awarded based on a price-only system or on e that i

ncorporates

a wider set of considerations (i.e., scoring rule auctions).

Since scoring rule auctions

llow for a range of non-price (and potentiall y subjective) pa r am et er s set by the RUP and

thus involve more discretion than ﬁrst-price auctions, we deﬁne an auction as having a

discretionary criterion (denoted by the variable DiscretCrit) if i t is awarded via a scoring

rule.

To capt u r e the two types of discretionary auctions we will emphasize, we deﬁne a

summary measure, Discretion, which denotes a u cti o n s for which DiscretP roc

lowN

= 1

r DiscretCrit = 1. While in principle DiscretP r oc

lowN

and DiscretCrit can both

occur simultaneously, th i s is rarely the case in practi ce since the regulations tend to favor

negotiations for smaller value (or urgent) contracts, while the scoring rule system is used

for complex projects and requires more time to award the contract since a commission,

and not just the RUP, evalu a t es the bids.

Beyond our measu r es of au cti o n procedure and cr i t er i on , we include a number of

other auction attributes as controls. Most importantly, we control for the auction re-

serve price (Reserve), which is the monetary value, reported in the call for tenders,

above which the PA is unwilling to pay for the contract. Price bids are expressed as

discounts over this reserve price. In ou r analysis, the r eser ve price will enter linearly (in

logarithm) as a control in many of our speciﬁcations, as well as via a series of dummy

variables for contracts in various r eserve price rang es, which corresp on d to thresholds

that triggered stricter rules and/or mo n i t or i n g of an auct i on , with cutoﬀs of e100, 000;

150, 000; 300, 000; 500, 000; 1, 000, 00 0 , and 1, 500, 000. At these t h r esh ol d values, both

the publicity requirements of the call for tenders and the set of eligible bidders change.

The auction database provides us with additional information that we will exploi t

To the extent that this is the case, the link between discretion and corruption will be underestimated.

A third alternative is also available, the so-called average bid auction (ABA). The ABA is a variant

of the ﬁrst-price auction in which the winner is the ﬁrm oﬀering the lowest price among a subset of “non-

excluded” oﬀers. The ABA induces higher participation as well as bid coordi n at ion (Decarolis [ 2018] ,

Conley and D ecar ol i s [2016]), but for our analysis, we simply view it as a non-discretionary auction

Hence, we will not treat it separately from the other ﬁrst-price auctions.

in the analysis. In particular, we observe the identity of the ﬁrm wi n n i n g the auction.

Information on each ﬁr m includes its name and the location where it was incorporated,

as well as a unique social security identiﬁer, which provides the link to the criminal inves-

tigations da t a. Finally, we also observe some standard procurement auction o u t com es,

including delivery time, price and - for about half of our sample of auctions

- the total

osts for completion. Data on the expected contractual duration as well as the eﬀective

total co m p l et i on time allow us to construct a measure of contractual delay (Delay) and

cost overrun (Extra Cost). Since Delay can be positive or nega ti ve and has extreme

outliers, we use an inverse hyp erbolic sine transformation. The ﬁnal price of the winning

bid is expressed as a discount over th e reserve price (Discount) and, similarly, Extra

Cost is calculated as the diﬀerence between the ﬁnal price and the awarding price, over

the initial reserve price.

III.B.2 Criminal Investigations Data

A contribution of this study i s to i ntroduce a new measure of criminality in pu b l i c

procurement. As previously noted, in the procurement data we observe bidders’ identities.

For each ﬁrm, we then obtained the full list of i ts owners and top managers through th e

Company Accounts Data System. Thi s is a proprietary database maintained by CERVED

Group that we observe for four separate years: 2006, 2011, 2014, and 2016.

For ea ch

ﬁ

rm, the union of all owners and managers recorded in any of these four period s represents

the set of individuals connected to the ﬁrm in our analysis. For each indivi d u al , their

record of criminal investigations (which we will describe shortly) was cod ed , and th i s

information was aggregated across ﬁr m -l i n ked individuals to obtain a ﬁrm-level measure of

potential criminal status. We use the same criminal investigations database to determine

the suspected cr i m i n al i ty of each RUP in our data.

Records of individu a l s’ criminal investigations were analyzed for us by AISI (Italy’s

internal intelligence and security agency) using a centralized archive, the Sistema D’Indagine

Interforze (SDI), wh i ch is a primary source of information that police oﬃcers and intelli-

gence agencies use to identify potential targets for further investi ga t i on .

This database

ontains reports of all individua l s investigated by any of the Italian police forces: state po-

lice (Polizia di Stato), ﬁnance pol i ce (Guardia di Finanza), military police (Carabinieri),

For a detailed discussion of the reasons behind li mit ed d at a availability, see Decaroli s and Palumbo

[2015].

In Online Appendix B, we describe in greater detail how we obtained each of the data sources we

employ. We note that most of our data are proprietary so that, while we can provide contact information

for interested researchers, we cannot provide t he data itself.

The SD I data have bee n previously used in research by Pinotti [2017]. Our access to the data is

nabled via a framework agreement between AISI and Bocconi University.

and environmental police (Guardia Forestale).

An entry in the SDI d a ta b a se typically occurs after a po l i ce force, b ased on a pre-

liminary investigation, determines that there is suﬃcient evidence to open a formal in-

vestigatio n . This investigation might or might n o t lead to a court case and, if so, to a

conviction . Therefore, court cases are clearly a subset of th e entries in the SDI datab a se

(see Figure A2). The resulting sample of suspect oﬀenders thus includes indi

viduals that

were convicted, acquitted, or never charged. The latter two groups plausibly comprise a

large number of oﬀenders whose guilt could not be proven in court. Indeed, corruption

cases are generally complex, and convictions are relatively rare. This is particularly true

in Italy, where the trial must go through three levels of judgment (Primo grado, Appello,

and Cassazione) within a relatively short statute of limitation – between 6 and 12 years.

For example, in the well-known “clean-hands” case, out of the 2,565 people investigated

for corruption, 1,408 were convicted, 544 were acquitted for lack of conclusive evidence,

and 488 due to t h e statute of limitations Davigo and Mannozzi [2007]. For these vari-

us reasons, oﬃcial data on (convicted) oﬀenders may greatly understate the extent of

corruption.

Although the SDI data do not suﬀer to the same extent from the u n d

er-reporting

problem tha t aﬄicts judicial data, they may instead include false positives. While in gen-

eral one may be concer n ed that investigations overstate the extent of underlying crimes,

there are featur es of corruption and also our speciﬁc context that make this less likely to

occur. Because the cr ed i b i l i ty of these data is central to our empirical exercise, we now

explain these reasons in some detail.

We begin by noting that part i ci p ants in procurement corruption in pa r ti cu l a r are

all likely to beneﬁt to some extent, with th e costs borne by society. Hence, in contrast

with many other types of crimes in which a v i ct i m is directly involved–and motivated

to inform investigators–this is less likely to be the case for corruption. Indeed, as d i s-

cussed earlier, investigations are often initiated based on information that enforcement

authorities receive from competitors of ﬁrms engaged in corruption.

In the speciﬁc context of o u r study, several features further reinforce a t i g ht conn ec-

tion between investigations and the underlying crimes. Understanding why this is the

case requires an elaboration of what Italian law deﬁnes as corruption crimes. There are

two main articles in the Italian penal cod e devoted to corruption. The ﬁrst is Article 318,

which states that any public oﬃ ci al who, in conducting their d u t i es, receives money or

other ben eﬁ ts for himself or for a third party (or accepts a promise of future beneﬁts), is

Decarolis and Giorgiantonio [2019] analyze the universe of court sentences for cor ru pt i on in public

auctions ﬁnding that only 2% of the ﬁrms awarded public contracts were thus implicated. In the same

set of auctions, our measure ﬂags 17% of contract winners as potentially criminal (not e that Decarolis

d Giorgiantonio [2019] use a smaller and diﬀerent set of auctions than the one used in our paper).

to be imprisoned for one to six years. In our context, thi s mi ght include a procurement

oﬃcial demanding payment from ﬁrms to be permi t t ed to bid on a contract – for exam-

ple, some co u rt records have revealed instances of corrupt oﬃcials deli berately creating

impediments to ﬁrms’ ability to conduct the worksite inspections that ar e compulsory

before being eligible to bid in an auction. The second, Article 319, concerns more serious

acts of corruption, speciﬁcally, the omission or delay of oﬃcial duties, or performing acts

that run contrary to oﬃcia l duties, in exchange for beneﬁts (or promise of beneﬁts) for

themselves or a third party (e.g., a procurem ent oﬃcial awarding a contract to an un d er -

qualiﬁed bi d d er ) . Such acts are punishable with a si x to ten year prison sentence. The

law draws a distinction between the less-serious case of “improper corruption” (Art. 318)

in wh i ch the oﬃcial recei ves a beneﬁt to perform duties that are within their purview, as

compared to the more serious case of “proper corruption (Ar t . 319), in whi ch the oﬃcial

acts contrary to his duties.

The two types of corruption have distin ct implication s for re

porting incentives, but

we argue in neither case is there an obvious motive for frivolous whistle-blowing. Under

proper corruption, a public agent’s actions may be revealin g of corruption because they

violate his oﬃcial duties, but none of the p ar t i es involved ha s a direct motive to act as

an infor m ant. By contrast, un d er improper corrupt i on , the act i o n s by the public agent

are less likely t o raise suspicions (since they involve his regular dut i es) ; however, the

parties involved may be more inclined to inform investigators, given that they are forced

to pay for something (like the op po r t u n i ty to bid on a contract) that they are legally

entitled to. Despite this repor t i n g incentive, the best response of involved participants

may nonetheless be to remain si l ent, if there are potential reput at i onal eﬀects that could

impact interactions with other oﬃcials, and especially if they may interact with the same

oﬃcial again, either over time or across contracts.

Several other features of Italian p rocurement result in a very high bar for initiating in-

vestigatio n s. First, unlike some other countries, Italy has no leniency program to enco u r -

age one party to denounce the other. Moreover, according to Article 321 of Italy’s penal

code, punishm ents for all implicated parties are symmetric–e. g. , a bribe-payer faces the

same penalty as the recipient. This inability to secur e lenient treatment reduces reporting

incentives.

Second, detecti n g p r ocurement corruption i n Italy is widely

considered to be

As we explain later, we also include in our measure two other types of crimes which, broadly speaking,

represent forms of corruption, but for which the Italian law uses ad hoc deﬁnitions – (”peculato” and

”concussione”). Such crimes are less prevalent than those falling under the deﬁnitions of articles 318 and

319: for instance, in 2016 th ere were 126 individuals imprisoned for corruption (un der Articles 318 and

319), 26 for ”peculato”, and 11 for ”concussione.”

Reforms to encourage better local governance and whistleblowers to come forward have proven

ineﬀective. The governance changes require that each PA (municipality etc.) nominate an oﬃcial to be

responsible for anti-corruption and that each PA provides an annual anti-corruption plan. In practice, few

harder during our sample p er i od , as a result of the “mani pulite” (clean hands) scandal

of the 1990s. In the period before this scandal , corruption in public contracts was sys-

tematic and served as an unoﬃcial means of ﬁnancing national politi ca l parties’ election

campaigns. This systematic corruption was disrupted by the “mani pulite” revelations,

and as a result bribery in procurement became a more loca l i zed phenomenon, based on a

plethora of small-scale partnerships between individual public oﬃcials an d ﬁ r m s. To the

extent that it is more diﬃcult for police to detect smaller, local i zed cases of corruption ,

this development further r ed u ces the problem of over-reportin g in our data.

Finally, it

s extremely diﬃcult to collect evidence to initiate a corruption investigation because of,

for example, limits on police powers to monitor the communications of suspected parties;

the police may only do so if there is clear eviden ce from the outset of “major guil t . ” As

a result, among cases reported to the police, there wi l l rarely be suﬃcient evidence even

to open a case and thus appear in our data.

We suggest that the preceding a r gu m ents indicate that our inv

estigations-based mea-

sure is unlikely to be overly aﬄicted by false positives. However, we arg u e that any

measure based on court convictions would be plagued by an excess of false negatives.

There are two main reasons for this. First, particu l a rl y in Italy, th e burden of proof for

corruption convictions is very high, requiring that the plaintiﬀ show convincing evidence

of: i) the beneﬁt directly (or in d i rect l y ) given (or promised) by t h e public oﬃcial to a

counterpar t , ii) the delivery by the counterpart of money or other ben eﬁ t s to the public

oﬃcial (or to a person or entity connected to them), and iii) the l i n k between th ese two

actions. There are numerous factors that make it diﬃcult to meet this bar. For example,

there is a signiﬁcant time lag between the ben eﬁ t received by a ﬁrm (say, the awarding

of a public contract) and any payback. Moreover, the latter is often hard to detect and

prove as it could involve indirect forms of beneﬁt, such as t h e hiri n g as consultants or

subcontract or s of persons linked to the p u b l i c oﬃcial (family members, friends or ﬁ g-

ureheads).

The high burden of the proof, coupled with the li m i ta ti o n s on ev

idence

PAs have adopted eﬀective programs. The new whistleblower law does not guarantee the full anonymity

of the whistleblower, a serious limitation in encouraging anyone to come forward to report corruption.

An example of this evolution in the nature of public contract corruption comes from the President of

the Italian Antitrust Authority: ”[Whereas cor r up t i on in] the First Republic was elevated to a “system,”

today micro-corruption prevails, and perhaps for this reason appears more pervasive. I don’t share

the view that “the thieves have won” nor that today is worse th an then. The numbers prove it. The

Enimont Aﬀair, which was called “the mother of all bribes,” involved around 140 billion lire. For cases

today, as anyone can verify on the Istat website, we are talking about 120 million. The bribes paid to

carry out the Mose, one of the biggest scan d als of recent years, amounted to a few million Euros.” See

https://luz.it/en/spns_article/intervista-cantone-corruzione/.

For this r eason some policymakers have in recent years proposed extending to corruption crimes the

same powers of investigation that the police have for maﬁa-related crimes, see https://formiche.net/

2016/10/libro-corradino/.

Corruption expert s often raise the related concern that under Italian law the presence of an inter-

collection described earlier, makes it particularly challenging to prosecute Italian corrup-

tion cases. Given th ese challenges, it is perhaps un su r pr i si n g that a ny investigations that

are opened proceed very slowly, often lasti n g for many years.

This leads to a further

eason why there are so few corruption convictions: Italy’s relatively short statute of

limitations. If an investigation is stil l ongoing as the statute o f limitations approaches,

the plaintiﬀ must decide whether to go to court or simply di sm i ss t he case. In the latter

case, there i s of course no conviction; in the for m er , a rushed case will likely be weak an d

a conviction unli kely, and very likely accounts for the relatively high rate of acquittals

for corruption cases.

In concluding our discussi on of the investigation da t a, we note t

hat the investigated

individuals are unaware that they are u n d er investigation, unless the case is formally

brought to a crimin a l court. For the same reason, unless a formal court case has begu n ,

a PA cannot exclude ﬁrms from auctions even if their owner s/ m an a ger s are investigated

for corruption charges.

To obtain these data for ﬁrms, AISI searched the SDI database for al l managers

and owners we identiﬁed as associated with each ﬁrm, and ﬂagged those who had been

investigated for corruption and other r el a ted crimes. Speciﬁcally, the foll owing categories

of crime were considered: corruption, malfeasance, and embezzlement; abuse of power

and und ue inﬂuence; and violations in public auctions. Based on the individual-level

records extracted from SDI, suspected crimi n a l s in 3,848 ﬁrm s winning a contract over

mediary between the public oﬃcial and the entity beneﬁting from its acts creates a further complication

in proving corruption, as relations among the various parties to thi s type of multi-layered structure need

to be proved. Hence, even direct proof of a payment to a public agent by an intermediary is insuﬃcient

as evidence of corruption unless it can also be proved that such payment can be related to an action

taken by the public agent to favor a business on whose behalf the intermediary made the payment.

This is in contrast, for in st ance, t o what is requi r ed unde r the 1977 U.S. Foreign Corrupt Practi ces Act

which states: “It shall be unlawful (...) to make () oﬀer, payment, promise to p ay, or authorization of

the payment of any money, or oﬀer, gift, promise to give, or authorization of t he giving of anything of

value to any person, while knowing that all or a portion of such money or thing of value will be oﬀered,

given, or promised, directly or indirectly, to any foreign oﬃcial.” Under the latter piece of legislation,

it is thus very clear that it is unlawful to make any payment that has the potential to be used in full,

or in part, as a bribe to a pub l i c oﬃcial. This is a substantially more at t ai nabl e burden of proof than

that required under Italian law. For a discussion of this issue and its application to the court acquittal

of the top management of the Italian oil company ENI in an alleged corru pt i on scandal, see

https:

//www.giurisprudenzapenale.com/wp-

content/uploads/2020/05/Scollo_gp_2020-5.pdf, last ac-

essed February 5, 2023.

See, e.g., https://www.dirittoconsenso.it/2021/06/25/la-durata-delle-indagini-preliminari/,

last accessed February 5, 2023.

The media often blames the statute of limitations for the lack of corrup-

tion convictions; see, e.g., https://www.lastampa.it/cronaca/2016/01/27/news/

italia-

ancora-bocciata-per-corruzione-ma-i-condannati-in-carcere-sono-appena-126-1.

36555629/, last accessed February 5, 2023. As one particularl y prominent example, out of the 36 court

cases in which Silvio Berlusconi has been accused of some crime, three involved corruption charges and,

out of these three, one is still ongoing and two ended due to t h e st at u t e of limitations.

the per i od 2000-201 6 were identiﬁed (9.8% of all ﬁrms winning at least one contract). We

deﬁne InvestigatedWinner as an indicator variable denoting that an au cti o n was won by

a ﬁ r m ever associated through o u t our sample peri od (via empl oyment or ownership) with

at least one individual present in the SDI database. Under our agreement with AISI, we

were unab l e to obtain year-speciﬁc inform at i o n on whether an investigated individual was

associated with a given ﬁrm – our measure thus varies only across ﬁrm s and not over time.

This approach is conservative, as the date at which suspect oﬀenders are reported in the

SDI provides little information – if any – on the date an oﬀense was actually committed.

The SDI data also allow us to ﬂag procuring agencies and publi c administrators as

suspected of corruption. For each auction, we observe the agency procuring the con-

tract and, wi th i n the administration, the RUP in charge of the speciﬁc contract. AISI

searched the SDI database for all RUPs, ﬂagging those suspected of the same types of

crimes used to ﬂag man a g ers and owners (i. e. , corrup t i o n , abuse of power, and so forth).

Overall, 6% of the RUPs in our sample (mana g i n g 9.7% of all contracts) were ﬂagged

as “ i nvestigated.” We use this list to identify auctions administered by an investigated

RUP (InvestigatedRUP ) and also administrations in which at least on e investigated

RUP was employed during our samp l e period (16% of all public administratio n s, denoted

by InvestigatedP A, managing 40% of the contracts).

In concluding our discussion of the criminality data, it is important to discuss two,

related potential problems: reverse causality and sorting. In our setting, reverse causality

could occur if, for instan ce, a ﬁrm woul d become more likely to be labeled as a suspect

when winning negotiated procedures (with few participants) due t o the po l i ce concen-

trating its (limited) monitoring eﬀorts on t h ese typ es o f pr oced u r es. We bel i eve that, if

anythin g, the opposite is in fact li kely to be tr u e in our data , based on exten si ve discus-

sions with the AISI rep resentatives who helped us in accessing the police data. These

oﬃcials gave n o indicati on that police monitoring eﬀorts are concentrated on public ten-

ders characterized by t h e criteria and procedures analyzed in this study. Furthermore,

they emph asi zed that investigations typically result from complaints to the police from a

losing b id d er , which are less likely for negotiated procedures, for two reasons. First, there

are simply fewer ﬁrms in negotiated procedures. Second, since procurement oﬃcers open

themselves up to scrutiny when bidd er s complain, it is also reasona b l e to assume that

oﬃcials will use their discretion in negotiated procedures to avoid i nviting ﬁrm s which,

for any reason, are more likely to report concerns to the police (this is even more t h e case

if the public oﬃcial is himself corrupt and h a s a favored ﬁrm among the pa r t i ci p ants).

Thus, while we cannot rule out reverse causation entirely, we believe that if a diﬀerential

monitoring intensity between negotiated and open procedures is present, in our context

it would most plausibl y imp l y that the estimates we present below represent a conser-

vative assessment of the increased corruption risks associat ed with reduced competition

and discretion.

Finally, on the issue of sorting, it could i nvolve both suppliers an d contracting oﬃ-

cers. For ﬁrm sorting, one might worry that ﬁrms that expect to be awarded contracts

through discreti on a r y systems might exert ad d i t i on a l eﬀort to avoid being detected as

potential l y corrupt. Such eﬀorts might in cl u d e using ﬁgureheads as com p any owners and

managers.

However, as mentioned above, it is not the case that certain typ

es of pro-

curements are more systematically investigated th an others by law enforcement agencies.

Since the controls placed on ﬁrms are lower for smaller contract va l u es, we shoul d expect

a greater presence of investigated ﬁr m s participating in and winning lower-valued pro-

curements. However, since lower-value contracts are also those for whi ch discreti o n ar y

procedure auctions can be used , this could mechanically lead to us to ﬁnd a positive

association between discretion on corruption. Simi l ar l y, an obvious concern about con-

tracting oﬃcers is whether the RUP might manipulate the contract value to make it

eligible for the use of discretionary procedures. Such behavior is illegal, as it is expressly

forbidden by procurement law. A corrupt RUP might nevertheless choose to take this

risk if discretionary procedures were instrumental for r ent-seeking activities. In this case,

the presence of manipulat i o n should, if anyt h i n g, increase the probability of detect i n g an

eﬀect of discretion on corruption, assuming that bureaucrats who sor t below t h e thresh-

old are using this leeway to beneﬁt investigated ﬁrms. Overall, it is very unlikely that

sorting by either suppliers or contracting oﬃcers can explain why our estimates below

show that discretion does not lead to more corruption .

III.B.3 Descriptive Evidence

We begin by presenting an overview of some of the main features of the data.

While in our main analysis we exploit with i n -municipality variation over time or (in

some cases) within-region variation across municipalities, the patterns in th i s subsection

explore t r en d s across time and broad regional diﬀerences in procurement practices at a

relatively high level of aggregation.

One important feature of our institutional setting is that the maxi mum reserve price

for negotiated contracts was increased from e100, 000 to e500, 000 in 2008, and then to

However, one important observation from the AISI is that monitoring eﬀorts are concentrated in

geographical areas where the presence of criminal organizations has been previously detected, and as a

result we will need to take care in interpreting results involving variation at the municipality level in

the presence of investigated ﬁrms. (Though to the extent that these factors are time-invariant, our ﬁxed

eﬀects speciﬁcations account for these geographic factors.)

This behavior is found by Daniele and Dipoppa [2019] in the context of ﬁrm subsidy allocation in

taly. The extent of likely ﬁgureheads is substantially larger for ﬁrms obtaini n g subsidies below the

threshold value that triggers the need for special certiﬁcations on probity of the owners and managers.

e1 million in 2011. As we show in Figure 1, this led to an increase in negotiated contracts;

the fraction of contracts awarded via scoring rule (the complement of ﬁrst-price auctions)

remains roughly constant.

Did this chan g e result in more contracts awarded to investigated ﬁrms? In Figure 2,

e examine whether there is any obvious evidence in favor of this view in the aggregate

data. The ﬁgur e plo t s the fractio n of contracts won by investigated ﬁrms for three groups,

based on the relevant t h r esh o l d s for th e 2008 an d 2011 expansions: contract s less th an or

equal to e500, 000, those between 500,000 and 1 million, and contracts above 1 million. If

discretion led to greater co r ru p t i o n , we would expect a relati ve increase in t h e fraction of

contracts won by investigated ﬁrms in the e150, 000 to 500, 000 range in 2008 and 500, 000

to 1 million range in 20 11 .

However, we observe no d i scer n i b l e change in any reser ve

rice interval after either reform (see Appendix Figure A4). Given that the contract size

is endogenous – we observe sorting aro u n d each of the thresholds in every year in our

sample – it is not possible to p r ovide a sh a rp interpretation of this “non-result.” But

at t h e same time, it does ﬁt with our overall set of ﬁnding s that we document in the

remainder of the paper – discretion in itself does not necessarily p r o m ot e corruption, and

monitors may take steps to ensure that its use is lim i t ed in locales in which discretion is

mostly likely to be abused.

To provide a preview of why greater di scr et i on might not have increased co r r u p -

tion, we consider two further cuts of the data. First, instead of compari n g the fractio n

of investigated winners by the contract reserve price (as in Figure 2), we present in

gure 3 the fraction of investigated winners for thr ee types of more d

iscretionary auc-

tions: those with negotiated procedures and the l eg a l l y mandated number of invited

bidders (DiscretP roc

highN

); t h ose with negot i at ed procedures and “too few” invited bi d-

ders (DiscretP roc

lowN

); and scoring rule au ct i on s (D

iscretCrit). Over the full sampl e

period, we observe that negotiated procedures are only associated with criminal winners

for auctions when there are fewer than the legally man d at ed number of bidders. Scoring

rule auctions (whi ch have potential l y discretionary selection cri t er i a) have the highest

rate of investigated winners. Combining these patterns with the general prevalence of

each type of auction, on e may see why the increased use of negotiated procedures had

no discernible impact on the rate of investigated winners – as can be seen in Figure 4,

he increase came primarily from auctions that preserve competition, i.e., those with the

legally mandated number of invited bidders, a category for which we see a relatively low

Note that these reforms were not associated with any other substantial changes concer ni n g b ur eau-

crats’ discretion as, for instance, the 2011 reform came about not as an organic reform of the procurement

code generally, but as a targeted measure of t h e Berlusconi government t o promote economic growth

by expanding the use of the less bureaucratic-intensive negotiated procedures. See Art. 4, sub. r, Law

Decree 70/2011, modifying Art. 122, sub. 7, Legislative Decree 163/2006.

rate of cor r u p t i on . Naturally, in comparing the corruption of diﬀerent auction types, we

wish to control for a range o f municipality and aucti on attributes in comparing various

types of auction mechanisms, which we will do in our regression analyses.

We next take advantage of the richness of our data to explore some patterns in the

data that will pr ovide the r ead er with a broader sense of where corruption – as captured

by investigated ﬁrms and investigated RUPs – is most prevalent.

In Appendi x Table A1, we show the frequency of investi g at ed RUPs overseeing auc-

ions and the freq u en cy that investigated ﬁrms that win auctio n s, for the two most

common sectors in our database, roads and building construction. For both RUPs and

ﬁrms, investigations are more common in roa d -b ui l d i n g . It is perhaps tel l i n g that Bosio

t al. [2022] use road construction as their hypotheti ca l contract to study the oversight

of procurement processes. Anticipating our later results, we ﬁnd the opposite pattern for

contracts that we classi fy as prone t o corruption (i.e., Discretion = 1): these are more

common in t h e buildings sector.

We next examine whether contracts are m o r e likely to involve investigations based

on whether the oﬃcial overseeing the contract was born in that locality, in Appendi x

Table A1, which might serve as a proxy for access to local networks that

might facilitate

corruption. We include this compari so n in the second part of Appendi x Table A1, where

we show that locally-born RUPs are indeed more likely to be investigated. Paralleling

the prior analysis, we a l so show t h a t discretion is l ower in contracts overseen by local

RUPs.

Finally, we turn to a geographi c comparison of auction procedures and outcom es,

where we again explor e bot h t h e prevalence of investigated ﬁrms and RUPs, and also

anticipat e the limits to discretion that may exist if corruption is more common. In

our geographic comparisons, we can more plausibly take as given that diﬀerent parts of

Italy have historically been associated with higher corruption. Speciﬁca l l y, in Table 1 we

ompare auction characteristics for South, Central, and North Italy over our full sample

period, 2000-2016. Given the South’s long history with, and reputation for, corru p ti o n ,

it is perhaps uns ur p r i si n g that the fraction of auctions overseen by procurement oﬃcials

suspected of corruption is notably higher in the South relative to Central and North

Italy (ﬁrst row). In the second row, we show the mean fraction of auction s won by

ﬁrms suspect ed of corrupti on . Again, there is a North-South gr ad i ent: investigated ﬁrms

are more likely to win in the South relative to the North and Central regions, though

the diﬀerence is much more modest than for RUPs. We next turn t o th e sel ect i o n of

auctions that, in the preceding ﬁgure, were associated with high er levels of corruption,

i.e., Discretion = 1 auctions (recall t h ese are DiscretCrit = 1 and DiscretP roc

lowN

= 1

uctions). Notably, these are far more common in the (r el at i vely less corrupt) North

(third row). In the last two rows, we look at the North-South choice of discreti on for

auctions administered by investigated procurement oﬃcials and clean (non-investigated)

oﬃcials. Interestingly, across all areas investigated administr a to r s select discretion more

often. The relative rarity of “corruptible” auct i on procedures in th e high-corruption

South suggests another potential explan a t i on for the muted link b etween the in cr ea se in

negotiated auctions and investigated winners: problematic auctions are used less often in

locales where th ey are more a p t to be corr u p ted .

Naturally, these patterns are merely presented as motivation – there are many factors

that could acco u nt for the No r t h -S ou t h diﬀerences we observe. We will attempt to account

for these factors when we focu s on within-PA variation in our regressions. But overall,

the patterns in Table 1 and Appen d i x Table A1 oﬀer descriptive evi d en ce that is broadly

onsistent with the regression analysis reported in the next secti o n , and which will be

useful for u n d er st an d i n g how Italian authorities may have limited the extent to whi ch

discretion can be exploited by oﬃcials for p r i vate gai n .

Before proceeding to our regression results, we conclude this section with a presen-

tation of the summary statistics for our dat a in Tabl e 2. Panel (A) provides summary

tatistics at the auction level for the wh o l e sample of just over 200,000 auctions. Of

these, 37% a r e done using negotiated procedu r es, and 83% of auction s use the price-only

criterion. Investigated ﬁrms are awarded 17% of t h e contracts and investigated RUPs

administer 10% of all auctions. The average number of bidders across all auctions is 27,

but for negotiated procedures, the average number of invited b i d d er s is 7. Relative to an

average reserve price of nearly e1 million, the ﬁnal p r i ce entails, on average, a 7% cost

overrun (relative to the initial reserve price), and the average delay is 63% relat i ve to the

originally speciﬁed contractual duration.

Panel (B) repor t s summary statistics at the level of the public administrations award-

ing contracts. We observe 14,024 administra t i on s out of which 16% have at least one

RUP suspected of corruption. 52% of public administrations are in the North, 35% in

the South, and 13% in the Center. In t er m s of administration type, local PAs award

most contracts, with munici p al it i es representing 57% of the PAs in the dataset (thoug h

they administer only 53% of auctions). Of the 7,985 municipalities observed, 67% have

fewer than 5,000 inhabitants, while only 1% of municipalities have more th a n 60,000 in-

habitants. The average administration awards 15 contracts over the sa m p l e period, with

an average total value of nearly e1.5 milli o n .

IV Empirical Analysis

We now turn to examine the relationship between the choice of auction mechani sm

to ﬁrms and oﬃcials suspected of corrup t i on . We ﬁrst examine t h e link from the type

of auction to whether it is won by an investigated ﬁrm, and then turn to look at the

choice of auction types by investigated public oﬃci al s. We will then use the framework

in Section V to interpret these pat t er n s in term s of the tradeo ﬀ invoked by

expanding

discretion.

IV.A Discretionary auctions and investigated winners

We employ throughout variants on the following speciﬁcation:

InvestigatedF irm

xay

= β

Discretion

xay

+ Controls

xay

+ α

+ γ

+ ε

xay

(1)

for auction x conducted by contracting authority a in year y. We include contra ct i n g

authority ﬁxed eﬀects to account for local diﬀerences in the choice of procurement mech-

anisms as well as (localized) diﬀerences in corruption; the year ﬁxed eﬀects absorb shifts

over time i n the prevalence of discretionary contracts as well as cor r u p t i on . Finally, as

controls, we include a lin ea r term for the logarith m of the r eser ve price as well as a set

of ﬁxed eﬀects for various size th r esh o l d s.

We use robust standard errors clustered at

he level of the contracting authority throughout.

Because this expr essi on employs a large number of contracting authority ﬁxed eﬀects,

our empirical approach might raise concer ns i f di scr eti o n on l y varies within a small,

selected poo l of administrations. However, as shown in Table 3, this i s not the case:

any adm i n i st r at i o n s experience variation in the various measures o f discretion analyzed

and, moreover, these administrations do not appear to be selected in any obvious way.

We present these resu l t s in Table 4. In columns (1) and (2) we show results using

iscretP roc

lowN

and DiscretCrit respectively as our measure of discretion, an d in col-

umn (3) we includ e both as covariates. The coeﬃcient on each variable is stable across

all speciﬁcations and signiﬁ cant at least at the 1% level in all cases. The coeﬃcient

In practice, the point estimates we report below are quite insensi t i ve to the inclusion/exclusion of

these covariates. For exampl e, if we include only year ﬁxed eﬀects as controls, the estimate is about 0.003

higher than what we report below, a diﬀerence of about 30 percent as compared to the fully saturated

speciﬁcations. Finally, we note that our results are unaﬀected by the inclusion of a control that captures

whether a ﬁrm is connected to a politician at the local, regional, or n at ion al level. We prefer not to

include this variable in our main speciﬁcations, as we believe it suﬀers from a bad control problem (ﬁrms

intent on engaging in cor rup t i on will coopt politicians), but present result s that include it in Appendix

Table A6.

on DiscretP roc

lowN

of 0.02 implies that auctions employing negotiated procedures with

“too few” invited bidders are associated with a 12% higher probability of being won by

an investigated ﬁrm. The coeﬃcient on DiscretCrit is approximately half as large.

column (4) we add the variable, DiscretP roc

highN

, as a covariate, which denotes aucti on s

hat are done via discretionary procedures, but with the requisite number of bidd er s. The

coeﬃcient on DiscretP roc

highN

is very small (0.0013), and we can reject at the 99% level

that it is even half as l a r ge as the coeﬃcient on DiscretP roc

lowN

. (We can reject at

he 0.1% level that the two coeﬃcients are equal). Finally, in column (5) we use the

summary discretion measu r e, Discretion, pooli n g together both DiscretP roc

lowN

and

DiscretCrit. The coeﬃcient of 0.012 implies that more discretionary auctions are asso-

ciated with a 7% higher probability of being won by a crimina l ﬁrm. Columns (6) – (10)

repeat these analyses, limiting the sample to au ct i on s administered by municipal councils,

as this is the sample we will focus on in analyzing whether the patterns we docum ent are

robust to controls for municipal attributes. The patterns are broadly similar, th o u gh the

coeﬃcients on the two distinct discretion variables are much closer in magni t u d e, and the

coeﬃcient on the pooled discretion measure is larger.

The correlation between the choice of discretionary auctions and the selection of

an investigated ﬁrm as wi n n er is robust to a range of considerations. In addition to

procurement administration ﬁxed eﬀects, we may include reg i on × year o r even province ×

year ﬁxed eﬀects (a total of 1,770 additional ﬁxed eﬀects), and the point estimates remain

quite similar. We may also amend the deﬁnition o f InvestigatedW inner to make it more

– or less – inclusive. In Appendix Tab l e A2, we show the results usin g a deﬁn i t i o n th a t

ocuses more narrowly on corruption (restricting attention only to ﬁrms investigated for

(i) corruption, malfeasance, and embezzlement or (ii) abuse of power and u n d u e inﬂuence,

but excluding those investigated for (iii) violations in public auctions) and in Appendix

Table A3, we expand the deﬁnition to include ﬁrms associated with indiv

iduals su spected

of waste management crimes. The inclusion of the latter group is at the suggestion of

anti-corru p t i on autho r i t i es, who indicated to us that it i s a common area for organized

crime and corruption. In both cases, we observe broadly similar patterns to those reported

in Table 4. While we see a measure of corruption based on i nvestigations r

ather than

realized convictions as preferable, since the former includes cases of likely malfeasance that

nonetheless cannot be prosecuted, we also consider a speciﬁcation in which the outcome

is an indica to r variable denoting that the aucti o n winner was convicted for corruption.

Note that conviction is a much rarer event relative to investigations – the mean conviction

rate is only 0.017 (stan d ar d deviation 0.13) as compa r ed to 0.17 (standard deviati on 0.38)

One possible expl an ati on for this weaker relationship is that ﬁrst-price auctions with few bidders also

aﬀord opportunities for directing a contract to very speciﬁc ﬁrms via the t ail or i ng of the requirements

to make a bid, rather than the criteria used to evaluate t he bids.

for investigations. Given the low conviction rate, the point estimates i n Table A4 a r e

commensurately smaller relative to tho se in our main results, but the broad patterns are

similar, even if the estimates do not generally reach stat i st i cal sig n i ﬁ can ce at co nventional

levels. Fi n a l l y, in Appendix Table A5 we include procurement-authority-by-year ﬁxed

ﬀects. While being more demanding and restricti ve, this speciﬁcation greatly improves

identiﬁca t i on , as it allows us to take into account any uno b ser ved time-varying shocks

at the authority level. Notably, results are remarkably similar to the on es of Table

4.In Appendix Ta b l e A7, we explore whether the higher r at e of investigated winners f

DiscretP roc

lowN

and DiscretCrit auctions is the result of selection into the pa r ti ci p a nts’

pool or selection of the winner ( con d i t i o n al on the pool of bidders). We run a speciﬁcation

analogous to the one in equation (1), but now using data a t the bidder level:

nvestigatedBidder

ixay

= β

Discretion

xay

+ Controls

ixay

+ α

+ γ

+ ε

xay

(2)

As noted in our data description, bidder-level data are only availab l e starting in 2011.

We observe a positive coeﬃcient on DiscretP roc

lowN

across all speciﬁcations, with a value

f 0.011 – 0. 01 2 (signiﬁcant at th e 1 percent level). No other variable is signiﬁcant. These

ﬁndings provide some suggestive evidence that (uncompetitive) negotiated procedures

may be corrupted by directing invitations to investigated ﬁrms, whereas scoring rule

auctions may be corrupted by tailoring the selection criteria to favored ﬁrms, rather tha n

foreclosing entry into bidding.

IV.B Investigated administrators and the choice of discretion

In Table 5, we explore the choice of discretion as an auction mechanism. W

e begin

with result s that most closely parallel those of th e preceding section, with public ad-

ministration ﬁxed eﬀects. In col um n 1 the dependent variable is Discretion, whi l e in

columns 2 and 3 we di sti n g u i sh between the eﬀect on DiscretP roc

lowN

and D

iscretCrit.

In all cases, the coeﬃcient on Inve stigatedRUP is positive (signiﬁcant at least at the 5%

level), indicating a higher use of discretionar y auctions; comparing columns 2 an d 3, the

point estimate is more than twice as high for discretionary criterion auctions, though the

base rate of discretionary criterion auctions is also much higher.

In the remainder of the table, we introduce I

nvestigatedP A as a covariate. Since

this variable varies only at the PA level, we can include only coarser ﬁxed eﬀects. In

Table 5 we employ ﬁxed eﬀects for each of the country’s 20 regions, and i

n Appendix

In Table A8, we explore the direct eﬀect of InvestigatedRU P on investigated winner. The eﬀect

is p osi ti ve and signiﬁcant, albeit small in magnitude. The estimates for the other coeﬃcients remain

qualitatively identical to those in the baseli ne estimates in Table 4.

Table A9 we use a ﬁner partition, with ﬁxed eﬀects for each of 110 provinces. (Recall

that, for a subset of procurement authorities (hospit a l s, hig hways, and so forth), we do

not have a mapping to a speciﬁc geographic location; thus auctions conducted by these

PAs are dropped from speciﬁcati o n s with region or province ﬁxed eﬀects.) In columns

4 and 5 we in cl u de InvestigatedRUP and Investigate d P A respectively as covariates,

with Discretion as the outcome variable. Note that, by deﬁnition, these variables are

positively correlated (ρ = 0.45). It is intriguing, therefore, that thei r coeﬃcients are of

opposite sign (signiﬁcant at the 1% level). Speci ﬁ ca l l y, PAs that have had at least one

administrator suspected of corruption are 7.7% less likely to use discretionary auct i o n s

(a coeﬃcient of 0.017 rel a ti ve to a base rate fo r Discretion of 0.22) whi l e, for a given

municip al council, a corrupt administrator is 8.6% mor e likely to use a di scret i o n ar y auc-

tion (0.019/0.22). In column 6, we include both variables – a s might be expected given

their strong positive correlation, in this speciﬁcation the magnitude of each coeﬃcient

increases, nearly doubling for both InvestigatedRUP and InvestigatedP A. Columns

7 and 8 repeat the speciﬁcations from colum n 6, which include both InvestigatedP A

and Investigated RUP , but using our two distinct discretion variables as the ou t co m es,

DiscretP roc

lowN

and D

iscretCrit. In these speciﬁcations, the relationship s between

both variables and discretion are driven by the selection of DiscretCrit auctions (though

we refer back t o column 1 to emph a si ze that, with ﬁner ﬁxed eﬀects, there is a dis-

cernable p o si t i ve relationship between InvestigatedRUP and the choice of discretionary

procedures).

IV.C The direct beneﬁts and costs of discretionary auctions

e now turn to describe the beneﬁts of discretion. The main oﬃcial motivation

for encou r a gi n g negotiated procedures is speeding up administrative procedures. The

administrative burden is lighter for negotiat ed procedures than with open auctions: PAs

can publish shorter, less detailed calls for tenders, and these calls have shorter minimum

mandatory publicity periods (about half of the 52 days typically required for open tenders,

but even less if certain conditions are met). The selection of t h e winning bid i s also faster,

as typically the RUP selects the winner dir ectl y from among a small set of b i dd er s. At

the opposite end of the spectr u m , scoring rule auctions requir e the creation of ad hoc

commissions to evaluate bids and select winners.

A diﬀerent margin along whi ch discretion can beneﬁt PAs is by helping to reduce the

adverse selectio n eﬀect s of open, competitive bidding. As mentioned ear l i er , incomplete

contracts and non -co ntractible quality are a near-deﬁn i n g feature o f contract procure-

Replicating the speciﬁcations in Table 5 using as dependent variable DiscretP roc, we ﬁnd no rela-

tionship between investigated RUPs or PAs and this outcome; see the Appendix Table A10.

ment. A ﬁrst-price open auction can be the most problematic allocation mechanism

when even just one oppo rt u n i st i c ﬁrm p a r t icipates. Although several institutional fea-

tures in the system are geared toward limiting the problem of “too good to be true” bids,

discretion i n selecting parti ci p a nts and bids can be a powerful tool (it is indeed the pillar

of private contracting).

We provide some indication of these potential ben eﬁ t s of discretion in Table 6. The

able presents the results of speciﬁcations that parallel those presented above, usi n g the

inverse hyperbolic sine of the contract’s delay in implementation (Asinh(Delay)), the

discount oﬀered by the winning ﬁrm, and the extra cost realized at the end o f the con-

tract as outcomes, in place of Investigate dW inner.

While delay is a hig h l y imperfect

ndication of performance – for example, it makes little sense to include DiscretCrit as

an explanatory variable, since execution time may be part of the scorin g rule to evalu-

ate contract s – in the absence of ex-post qual i ty evaluations o f contracts, it nonetheless

provides one objective indication of the winning ﬁrm’s performance.

Table 6, column (1) in cl u d es D

iscret as an explanatory variable, along with ﬁx ed

eﬀects for procurement administration and year, and ﬂexible reserve pri ce controls. As

would be expected if discretion speeds the completion of a contract, the coeﬃcient on

Discret is negative, though small in magnitude and onl y borderline signiﬁcant (p <

0.07). We distinguish between DiscretCrit an d DiscretP ro c

lowN

in column (2), and ﬁnd

hat there is a much stronger negative relationshi p for negotiated procedures – recall

that, as we not ed above, it is hard to interpret the relationship between discretionary

criterion and delay, as completion time may be a component of the scoring ru l e used to

evaluate bids. In co l u m n (3) we add a control for negotiated procedures – recall that this

captures auctions i n which bidders must be invited to participate i n the au cti o n , whereas

DiscretP roc

lowN

denotes negotiated procedure auctions in which “too few” par

ticipants

are invited . Interestingly, once one accounts for wheth er a n auction is a negot i at ed

All three outcomes are available only for a subsample of aucti ons . Therefore, we also test the

robustness of our main results in this restricted sample. Speciﬁcally, Table A12 replicates the results of

able 4 for the subsample of auctions for which we have either Delay, Discount, or Extra Cost information.

As an additional check, i n Table A11 we show that neither InvestigatedW inner nor InvestigatedRUP

predict the presence of outcome data – indeed if anything such data are more likely to be available

in these cases. Finally, si nce we present our earlier results for the full sample of PAs as well as for

munici pal i t i es only, in Table A13 we repeat the analysis but limiting the attention to contracts awarded

y local authorities. Results are very similar to those reported below.

The absence of quality evaluation imposes a li mi t on the interpretation of our results, but no such

data are typically available. With a few notable ex cep t ion s where di rect evidence on quality of the

procured contracts is observed, time delays and cost overruns are generally used as proxies for quality

by government agencies and most of the academic literature when the focus is on complex contracts (as

opposed to simpler contracts for the procurement of standardised goods). For instance, delays are the

main outcome in Lewis and Bajari [2011], while cost overruns are the pr oxy for quality in Mohamed

t al. [2011], Iimi [2013], Bajari et al. [2014], Schoenherr [2019], Jung et al. [2019].

procedure – which itself is associated with much shorter delays – there is little incremental

eﬀect of DiscretP roc

lowN

on delay.

The remaining columns of Table 6 repeat the regression analysis for the two other

outcomes. We observe a clear negative and economically large impact of discretion on

winning discou nts: the coeﬃcient on Discret implies a 4 percentage point lower discount,

relative to an average winn i n g discount of 18 percent. Column 6 shows that mo st of

the dr o p is associated with discretionary criteria and, to a lesser extent, discretionary

procedures with too few bidders. Negotiated procedures with the appropriate number of

bidders more generally are associated wit h lower discounts, as indicated by the negative

coeﬃcient on DiscretP roc

highN

, but the size of the eﬀect is about half of that of the

iscretionary criterion. Thus, it appears that discretion has a direct impact on in crea si n g

the price paid by PAs by a signiﬁcant amount, which could result fr o m discretion limiting

competition, or if discretion is used to select hi gh er quality bids. In the next and ﬁna l

section of the paper, we will relate this increase of public cost to the (potential) beneﬁt

for a corrupt RUP.

Finally, notice that the ﬁnal price, i n cl u si ve of cost overruns, i s essentially unaﬀected

by the choice of discreti o n , as the estimated coeﬃcients are either not si g n i ﬁ ca nt or, in

the case of discretionary criterion, signiﬁcant and negative, but small i n magnitude.

V Conceptual Framework: Corruption and O vers ight

n this sectio n , we lay out a very simple and intuitive model to interpret our empir i ca l

results. Naturally, given the correlational nature of our analysis, we cannot l i n k our

ﬁndings deﬁnitively to a particular interpretati o n ; rather, th e goal of this section is to

illustrate that our dispara t e ﬁndings can be ex p l a i n ed via a very standard principal-agent

framework.

The patterns docu m ented above may be organized through the lens of the theory

of delegati o n , originally laid down by Holmstrom et al. [1982] and applied t o political

conomy settings in particular as outlined in Bendor et al. [2001] and Huber and Shipan

[2006]. Holmstrom et al. [1982] in particular describes the cl a ssi cal optimal delegation

roblem with no transfers: a central monitor (the pr i n ci p al ) trades oﬀ the beneﬁts of an

We are implicitly taking the assignment of a contract by an investigated RUP or assignment to an

investigated ﬁrm as social harmful in itself, and exploring the extent to which other beneﬁts or costs arise

as a result of the types of auctions we associate with corrupti on. We may in addition l ook at the direct

correlation between auction outcomes and whether a RUP or ﬁrm is investigated, as some indication of

whether corruption imposes a direct social cost. We provide these analyses in Appendix Table A14. As

ith discretion, both investigated RUPs and investigated ﬁrms are associated with smaller discounts.

Interest i ngl y, t her e is no oﬀsetting beneﬁt in terms of delay.

agent’s discretion against the costs of self-dealing, without being allowed to link transfers

to the realized ou t com es. This framework plausibly r esembles the situation of the pro-

curement oﬃcers in our data, whose wages and careers are only weakly associated with

the performance of the contracts they supervise.

Our simpliﬁed versi on of this style of model considers the task

of a central monitor-

ing authority (such as a regio n a l government) that aims to limit corruption. Discretion

makes it easier for oﬃcials to abuse their po si t i on s if they choose to do so, but also

empowers civic-minded oﬃcials to execute contracts more eﬃciently. The principal has

limited information on the infrastructure needs of lower-level governments (e.g. munici-

palities), and hence receives a noisy signal as to the beneﬁts of running an auction using

discretionary methods. As a resu l t , infrastr u ctu r e provision may be more eﬃcient if lo-

cal o ﬃ ci al s – who have a stronger local presence and/or expertise – choose the auction

format. Th e misalignment results from potential self-dealing by corrupt local oﬃcia l s.

More speciﬁcall y, we assume that a central authority may choose whether to allow

procurement oﬃcials in administration a to run an auction with gr eater discretion. Let

d be a parameter that captures the potential beneﬁt from discretion in implementing the

project so that, for example, the value of the project is v in the absence of discretion and

v + d if discretion is allowed. While v is perfectly observed, d is known only to the oﬃcial

overseeing the project; others (inclu d i n g enforcement oﬃcials) observe only

d = d + ǫ. It

is possible that d < 0, so that discretion is socially destr u ct i ve, whereas monit o r s may

still receive a positive signal. This a ssu m p t i on allows for the case that a civic-minded

oﬃcial will choose n ot to use a discretionary auction.

A further cost of discretion is that it provides op po r t u n i ti es for self-dealing, which

may be obfuscated precisely because of uncer t ai nty in the value of discretion. We do

not aim, at this l evel of abstracti on , to model the ﬁrm-oﬃcial interaction. In our simple

framework, one can think of corrupt oﬃcials extracting kickbacks from ﬁrms, or p r ospec-

tive bidders corrupting procurement oﬃcials by oﬀering bri bes. For a potentially corrupt

administrator, we think of their theft d eci si on as dictated by the private returns from

stealing s, less a punishment cost which is a function of detection probability e

, which is a

ublic-administration-speciﬁc parameter, so that his payoﬀ function wi l l be: π = s− e

n the internal solution, thi s payoﬀ function leads to a theft choice of s

∗

= 1/2e

We assume that the monitoring authority may const rai n a publ i c admi n i st r a ti o n

from utilizing discretionary auctio n s by setting a threshold for the signal of discretion’s

beneﬁt, accounting for both stealing (which is a function of the pu b l i c administrat i on ’ s

Our model is also very much in the spirit of the framework of Bosio et al. [2022], which documenting

how limits to discret i on are eﬀective in low-public sector capacity countries, but not in high-capacity

ones. Our results ﬁt in this framework, to the extent that the r isk of pu b li c corruption is negatively

correlated with public sector capacity.

enforcement eﬀorts, e

) and the probability that a contract is corrupted (which depends

on the share of corrupt public oﬃcials in administration the administration, p

). A

risk-neutral monitor seeking to maximize the pr oject value will then set a threshold

∗

= p

/2e

his model captures the simple intuition that, in locations with weaker enforcement

or a higher prevalence of corrupt agents (which plausibly are cor r el at ed ) , there will be a

higher threshold set for the use of discretionary auctions. Hence, diﬀerences among ad-

ministrations in (p

, e

) might lead to instan ces in whi ch the monitor restra i n s discre

tion

in situations in which it would be socially optimal to allow for it. But it also follows

that corrupt oﬃcials will use discreti o n ar y auctions more often since, by deﬁnition, non-

corrupt oﬃcials use discretion only when d > 0 whereas corrupt o n es will do so whenever

the monitor allows it (i.e., the threshold is hi g h enough).

V.A Re-evaluating the overall eﬀect of increased discretion

In Section IV.A, we found th a t negotiated contracts wi th many bidders – whi ch

onstitute the vast majority of auctions with discretion – were won by investigated ﬁrms

at the same rate as open price-only auctions. While negotiated contracts with ”too

few” bid s and scoring rule auctions were won more often by investigated ﬁrms, we also

observed in Section IV.B that regional governments may t ake steps to limit the use of

hese mechanisms in locales that are vulnerable to corruption.

These ﬁndings natu r a l l y return us to the question of whether the limits to discretion

imposed by procur em ent regula ti o n s were too strict. Procurement regulati o n s are the

result of a com p l ex web of rules determined by the European Procurement Directives,

Italy’s national procurement law a nd , in most cases, local rules (at the regional, provincial,

and even municipality levels; see Figure A1). At the local level, there a re many exa m p l es

f rules eit h er limiting or expandin g RUP’s discretion : for instance, Calabria, Campania,

and Sicily, the three regions wi t h endemic criminal organization s, passed various regional

guidelines and regulations limiting the use of discretionary procedures or criteria.

The most straightforward setting to explore the aggregate consequences of chang i n g

the limits to discretion is to focus on the nationwide reforms that loosened the rules

on th e use of negotiated procedures during the late 2000s. While our ear l i er discussio n

emphasized the role of a local (regional) m o n i t or that could set the minimum required

expected beneﬁt from discretion to activate it (

∗

, in the model above), we also discussed

he existence of national rules which set strict monetary thresho l d s on contract va l u es to

determine which ones may be awarded via discretionary methods. This type of rule is

typical in procurement regulations, and in d eed a simi l a r setup i s present in the US for

accessing the Si m p li ﬁ ed Acquisition Procedure.

The motive behind this form of regulation can be easily understood if one presumes

that the national regulator does not even observe the signal of the value of discretion for

a speciﬁc project, and we further augment our basic model to assume that the ben eﬁ t s

to the agent from stealing increase wi t h project size.

In this augmented framework,

etting a maximum project value beyond which discretion is forbidden can serve to limit

the risks of stealing.

Note, however, that this ad d i ti o n a l rigidity im posed at the national level comes

at the cost of limiti n g discretion for local ad mi n i st r at i o n s and RUPs that would use it

for public beneﬁt. This rigidity may fur t h er be excessive (relative to the social welfare

optimum ) if political economy con si d er at i o n s lead to a large weight on theft by national

bureaucrats and politicians.

A similar argument may be applied to a bureaucrat with

areer concern s and reduced performance incentives: discretion will be under-utilized if it

increases the probability that an oﬃcial will face a corruption investigation which, in the

Italian context, woul d defer any pr om o t i on until acquittal, without suﬃcient oﬀsetting

rewards.

These changes led to only a modest increase in either of the auc

tion types that

we have ﬂagged as associa t ed with corrupti on . For example, comparing auctions held

prior to 2008 versus those held in 2011 and later, the fraction of auctions for which

DiscretP roc = 1 or DiscretCrit = 1 increases from 20.5% to 2 3 . 6% : while discretionary

In the US, since the Federal Acquisition Streamlining Act of 1994, Simpliﬁed Acquisition Procedures

(SAP) were introduced to p r omot e eﬃciency and economy in contracting by reducing admin ist r at i ve

costs and unnecessary burdens for agencies and contractors. Under the SAP, contracting oﬃcers can

select private contractor s in more informal ways, for instance by getting oral (rather th an written)

quotes and selecti ng the winner without the need for a formal comparative assessment among quotes.

The SAP applies to purchases of sup p li es or services whose anticipated dollar value does not exceed

the S i mpl i ﬁed Acquisiti on Threshold, which has increased over time, reaching $ 150,000 as of 2014, and

making purchases under the SAP an even larger portion of federal procurement.

Under this modiﬁcation, the optimal stealing would become s

∗

, where v is the baseline project

size, as in Section V.

For example, reelection con cern s may lead a politician to l i mit stealing per se – beyond its impact

on project outcomes – because of the negative pu bl i ci ty from revelations of corru pt i on in public works.

The responsiveness of politicians to corrupt i on scandals has been documented, in particular, through

a ser i es of papers exploiting the richness of Brazilian data on corruption au di t s, including Avis et al.

[2018] and Ferraz and Finan [2011]. The former study documents a signiﬁcantly lower rate of corruption

n municipalities in which mayors can run for r eel ecti on , while the latter estimates a structural model

of agency which illustrates that the reduction in corruption after an audit comes primarily from the

perceived non-electoral costs of engaging in corruption.

This is the well-known problem of low-powered incentives f or public employees, which has been

documented across many countries and institutions (see, for instance, the analysis of Indian bank na-

tionalizations by Banerjee et al. [2004]). The problem may be exacerbated by the initial selection of

ndividuals choosing to become bureaucrats (as analyzed, for instance, through a randomized study of

initial public sector wage oﬀers in Mexico by Dal Bo et al. [2013]) as well politicians (see the recent

eview by Dal Bo and Finan [2018]).

procedure auctions increased substantially (from 0% to 12.7%) this increase was largely

oﬀset by a substitution away from discretionar y criterion (scoring rule) auctions. Taken

at face value, our regressio n coeﬃcients imply a 1.5 percentage p o i nt increase in auctions

won by investigated ﬁrms for the incr emental 3.1% of auctions conducted via discre-

tionary procedure or criterion. This calculation leads to a 0.05% increase in investigated

winners overall (0.031 × 0.015). Given ou r proposed framework, these results are u n su r -

prising. Indeed, recall that the increase in negotiated procedure auctions with the legally

mandated number of bidders is about 5 0 % between 2008 and 2011. Thus, if these led to

even small eﬃciency gains relative to open ﬁrst-price auctions, it would more than oﬀset

the loss from the very sm al l increment in corrupted auction s. We ﬁnd this to be quite

plausible given our ﬁndings on the improvements in contracting quality from discretion,

such as a 14 percent reduction in delays.

VI Conclusions

We pr esent evidence suggesting that discretion, to the extent that it limits com-

petition, is associated wit h higher suspected corruption in procurement. We show that

these auctions ar e chosen more often by oﬃ ci a l s suspected of corruption, and less often

in public administrations in which at least o n e procurement oﬃcial has been investigated

for corruption.

We see several main takeaways from our ﬁndings. First, given the central role played

by competition in the patterns we document, our results argue against certain classes of

models which emphasize bribery as a means of competing with other b i d d er s, and those

that model corruption as the outcome of a competitive (and eﬃcient) bidding process in

which the best ﬁrm is willing to bribe the most to secure a contract. Second, presuming

there is enough competition (i.e., suﬃcient bi dders), rigid constrai nts on auction oﬃcials’

discretion (e.g., via minimum contract size thresholds) may be costly tools t h at , at least

based on our measure, have a modest impact on corruption. Indeed, our rough assessment

based on the cost s and ben eﬁ ts of discret i on suggest t h a t it is likely under-utilized in our

setting. In our view, this result is unexpected, particularly for a country like Italy,

which has been traditionally characterized by high levels of corruption, given its level of

development.

We also see a number of avenues for future research. For example, we wish to better

understand the costs invoked by rules to limit corruption as a step to further clarifying the

trade-oﬀs that result from anti-corruption policies. Furthermore, in this ﬁrst assessment

of the link between discretion and corruption, we have taken a broad view of the data, and

done so in a correlational framework. We hope that the patterns we document may oﬀer

inspiration for future work wit h a clearer causal desi gn or equilibrium analysis, to further

probe our basic ﬁndings and proposed framework. In a similar spirit, future research may

also provide deeper insights into the speciﬁc mechanisms that underlie the correlations

we document.

Finally, our ﬁnding s have a number of policy implications. In particular, the diﬀer-

ence in outcomes of negotiated auctions with “many” versus “few” bidders is potentially

important for assessing the overall costs and beneﬁts of discretion. Indeed, our ﬁndings

suggest that discr eti o n itself is n ot necessarily problemat i c, b u t rather discretion com-

bined with foreclosure of competition: sco ri n g r u l e auct i on s li m i t co m pet i t i o n by tailoring

contract terms to a speciﬁc ﬁrm’s ca p a b i l i ti es, while negotiated contracts with few in-

vited bi d d er s by construction limit the competitive bidding process. Hence, the use of

more discretionary auctions shou l d go hand in hand with more stringent requirem ents

for fostering ﬁ r m participation.

More generally, in both developed and developin g cou ntries, the legal and regulatory

frameworks governing public procur em ent have a profou n d impact on the interactions

between governments and pri vate sector ﬁrms, and ultimately on the eﬀectiveness of

government service delivery. In 2013, the World Bank began publishing an annual stud y

– Benchmarking Publ i c Procurement – which analyzes the publi c procurement regulations

of about 180 economies; these reports real considerable heterogeneity across countries.

Our results help to explain why such a variety of systems exist, as we argue that trad e-o ﬀs

in the choice of p r ocu r em ent rules (in particular the ext ent to whi ch discretion is allowed)

depend criti cal l y on the local conditions (in particul a r the extent of corr u p t i o n and also

the monitoring eﬀectiveness).

By th e same reasonin g, the same rules may have highly heterogeneo u s eﬀects, de-

pending on the context where they are used. In this respect, one noteworthy element

of our analysis for policy design is the ﬁnding of higher corruption risks associ at ed with

scoring rule auctions. In the European Union, after 10 years of negotiations between

member stat es, a new Procurement Directive was published in 2014. At its core, it fea-

tures a switch from the previous highl y rigid system of p ri ce-o n l y open auct i o n s to a

more discretionary system, in which scori n g rule auctions are eﬀect i vely the default. The

eﬀects of this change have still to be st u d i ed , as its full implementation is quite recent.

Member sta t es are permitted an adjustment perio d to adopt the Directive in their leg-

islation and Italy, for instance, implemented the new rules only in Ap r i l 2016. However,

our resu l t s indicate that the goal of creating a common legi sl a t i ve framework in the EU

to foster economic integration and cross-bord er procurement may come at a cost of re-

quiring regu l at i o n s that are not necessarily well-suited to al l institution a l environments

– the new rules may result in r eg u l at i o n s that for some areas lead to substantially higher

corruption risk, while for other areas, the one-size-ﬁts-all regulations may not allow for

suﬃcient discretion. Our estimates are a ﬁrst step in qua ntifying the elements of this

important trade-oﬀ.

References

Emmanuell e Auriol, St´ephane Straub, a n d Thomas Flochel. Public procurement and

rent-seeking: the case of paraguay. World Development, 77:395 – 40 7, 2016.

Eric Avis, Claudio Ferraz, and Frederico Finan. Do government audit s reduce corruption?

estimating the impacts of exposi n g corrupt politici a n s. Journal of Political Economy,

forthcoming, 2018.

Patrick Bajari, Stephanie Houghton, and Steven Tadelis. Bidding for incomplete con-

tracts: An empirical analysis of adaptation costs. American economic review, 104(4):

1288–1319, 2014.

Audinga Baltrunaite, Cristi n a Giorgiantonio, Sauro Mocet t i , Tomm a so Orlando, et al.

Discretion and supp l i er selection in public procurement. Technical report, Bank of

Italy, Economic Research a n d Interna t i on a l Relations Area, 2018.

Oriana Bandiera, Andrea Prat, and Tommaso Valletti. Active and Passive Waste in

Government Spendin g : Evidence from a Policy Experiment. The American Economic

Review, 99:1278–1308, 2009.

Oriana Bandiera, Michael Best, Adnan Khan, and Andrea Prat. The allocation of au-

thority in organizations: A ﬁeld experiment with bureaucrats. mimeo, 2020.

Abhijit Banerjee, Esther Duﬂo, and Shawn Cole. Ban k i n g reform in india. mimeo, 2004.

Edward C. Banﬁeld. Corruption as a feature of governmental organization. Journal of

Law and Economics, 18(3):3, 1975.

Pranab Bardhan and Di l i p Mookherjee. Decentralization, corruption and government

accountabi l i ty. International handbook on the economics of corruption, 6:161–188, 2006.

Jonathan Bendor, Amihai Glazer, and Thomas Hammond. Theories of delegation. Annual

review of political science, 4(1) : 2 35 –2 6 9, 2001.

Alon Bergman, Matthew Grennan, and Ashley Swanson. Lobbying physicians: Payments

from ind u str y and hospital procurement of medical devices. Working Paper 29583,

National Bureau of Econ o m i c Research, December 2021. URL http://www.nber.

rg/papers/w29583.

Michael Carlos Best, Jonas Hjort, and David Szakonyi. Individuals and organizations

as sources of sta t e eﬀectiveness. NBER Working Paper 23350, National Bureau of

Economic Research, Cambridge, MA, 2019. URL http://bit.ly/2dlCTyo.

Erica Bosio, Simeon Djankov, Edward Glaeser, and Andrei Shleifer. Public procurement

in l aw and practice. American Economic Review, 112(4):1091– 1 11 7, April 2022. doi:

10.1257/aer.20200738. URL https://www.aeaweb.org/articles?id=10.1257/aer.

20200738.

arah Brierley. Unprincipled principals: Co-opted bureaucrats and corruption in ghana.

American Journal of Political Science, 64(2):209–222, 2020.

Jonathan Brogaard, Matthew Denes, and Ran Duchin. Politi cal inﬂuence and government

investment: Evidence from contract-level d at a . mimeo, 2016.

Alessandro Bucciol, Riccardo Camboni, and Paola Valbonesi. Purchasing medical de-

vices: the role of buyers’ competence and di scr eti o n . Journal of Health Economics,

forthcoming, 2020.

Roberto Burguet, Juan-Jose Ganuza, and Jose G. Montalvo. The microeconomics of

corruption . A rev i ew of thirty years of research . Barcelona GSE Working Paper

Series, 908(May), 2016.

Nicol´as Campos, Eduardo Engel, Ronald D Fischer, and Alexander Galetovic. The ways

of corruption in infrastructure: Lessons from the odebrecht case. Journal of Economic

Perspectives, 35(2):171–90, 2021.

Rodrigo Carril. Rules Versus Di scr eti on in Public Procurement. Working Paper, 2019.

Matias D. Cattaneo, Michael Jansson, and Xinwei Ma. Simple local polynomial den-

sity estimat or s. Journal of the American Statistical Association, 0(0):1–7, 2019. doi:

10.1080/01621459.2019.1635480. URL https://doi.org/10.1080/01621459.2019.

635480.

Emanuele Colonnelli and Mounu Prem. Corruption and ﬁrms: Evidence from randomized

audits in brazil. Technical report, 2017.

Timothy G. Conley and Francesco D ecar o l i s. Detecting bidders groups in collusive auc-

tions. American Economic Journal: Microeconomics, 8(2):1–38, May 2016.

Raﬀaella Coppier, Mauro Costantini, and Gustavo Piga. The ro l e of monitoring of cor-

ruption in a simple en d og en o u s growth model. Journal of Labor Economics, 51(4):

1972–1985, 2013.

Decio Coviello, Andrea Guglielmo, and Giancarlo Spagnolo. The eﬀect of discretion on

procurement performance. Management Science, 64(2):715–738, 2017.

Lydia Cox, Gernot J. Mller, Ernesto Pasten, Raph ael Schoenle, and Michael Weber. Big

g. Chicago Booth Research Paper No. 20-04, Fama-Miller Working Paper, University

of Chicago, Becker Friedman Institute for Economics Working Paper No. 2020-36,

2020.

Ernesto Dal Bo and Frederico Fina n . Progress and perspectives in the study of political

selection. Annual Review of Economics, forthcoming, 2018.

Ernesto Dal Bo, Fred er i co Finan, and Martin A. Rossi . Strengthening State Capabiliti es:

The Role of Financi a l Incentives in the Call to Public Service. The Quarterly Journal

of Economics, 128(3):1169–1218, 2013.

Gianmarco Daniele and Gemma Dipo p p a. Business below the line: Screening, maﬁas

and public funds. Technical Report No. 2018-98, BAFFI CAREFIN Centre Research

Paper, June 2019.

P. Davigo and G. Mannozzi. La corruzione in Italia: percezione sociale e controllo penale.

Libri del tempo Laterza. Laterza, 2007. ISBN 9788842083863. URL https://books.

oogle.it/books?id=sIEEAQAAIAAJ.

Francesco Decarolis. Comparing procurement auctio n s. International Economic Review,

59(2):391–419, 2018.

Francesco Decarolis and Cristina Giorgiantonio. Local public procurement regul at i o n s:

The case of italy. International Review of Law and Economics, 43:209–226, 2015.

Francesco Decarolis and Cristina Giorgiantonio. Cor r u p t i on red ﬂags in public procure-

ment: Evidence from italian calls for tenders. 2019.

Francesco Decarolis and Giuliana Palu mbo. Renegotiation of public contracts: An empir-

ical analysis. Economics Letters, 132:77 – 81, 2015. ISSN 0165-1765. doi: https://doi.

org/10.1016/j.econlet.2015.04.025. URL http://www.sciencedirect.com/science/

rticle/pii/S0165176515001767.

Francesco Decarolis, Leonardo M Giuﬀri d a, Elisabetta Iossa, Vincenzo Mollisi, and

Giancarlo Spagnolo. Bureaucratic Competence and Procurement Outcomes. The

Journal of Law, Economics, and Organization, 05 2 02 0. ISSN 8756-6222. doi:

10.1093/jleo/ewaa004. URL https://doi.org/10.1093/jleo/ewaa004. ewaa004.

afael Di Tella and Ernesto Schargrodsky. The role of wages and auditing during a

crackdown on corruption in the city of buenos aires. The Journal of Law and Economics,

46(1):269–292, 2003.

Simeon Djankov, Tania M. Ghossein, A si f Mohammed Islam, and Federica Saliola. Pub-

lic procurement regulation and road quality. Technical report, National Bureau of

Economic Research, 2017.

David Epstein and Sharyn O’halloran. Administrative procedures, infor m a ti o n , and

agency discretion. American Journal of Political Science, pages 697–722, 1994.

Claudio Ferraz and Frederico Fin a n . Elector al accountability and corruption: Evi d en ce

from the audit s of local governments. American Economic Review, 101(4):1274–1311,

June 2011.

Joras Ferwerda and Ioana Deleanu. Identifying and redu ci n g corruption in public pro-

curement in the eu, 2013.

Bengt Holmstrom et al. On the theory of delegation. Northwestern University, 1982.

John D Huber and Charles R Shipan. Politics, delegation, and bureaucracy. In The

Oxford handbook of political science. 2006.

Atsushi Iimi. Testing low-balling strategy in rural road procurement. Review of Industrial

Organization, 43:243–261, 2013.

Hojin Jung, Georgia Kosmopoulou, Carlos Lamarche, and Richard Sicotte. Strategi c

bidding a n d contract r en ego t i at i o n . International Economic Review, 60(2):801–820,

2019.

Gregory Lewis and Patrick Bajari. Procurement co ntracting with ti m e incentives: Theory

and evidence. The Quarterly Journal of Economics, 126(3):1173–1 21 1 , 2011.

Sean Lewis-Faup el, Yusuf Neggers, Benjamin A. Olken, and Rohini Pande. Can electronic

procurement improve infrastructure provisio n ? evidence from public works in india and

indonesia. American Economic Journal: Economic Policy, 8(3):258–83, August 2016.

doi: 10. 1 25 7/ pol.20140258.

Maxim Mironov an d Ekaterina Zhuravskaya. Corruption in procurement and the polit i ca l

cycle i n tunneling: Evidence from ﬁnancial transactions data. American Economic

Journal: Economic Policy, 8(2):287–321, May 2016.

Khaled A Mohamed, Shaﬁk S Khoury, and Sherif M Hafez. Contractors deci si on for bid

proﬁt reduction within opportunist i c b i d d i n g behavior of clai m s r ecovery. International

journal of project management, 29(1):93–107, 2 01 1 .

OECD. Government at a glance 2017. 201 7 .

Benjamin A. Olken. Monito r i n g Corruption: Evi d en ce from a Field Experiment in In-

donesia. Journal of Political Economy, 115:200–249, 2007.

Benjamin A. O l ken and Rohini Pande. Corrupti o n in developing countries. Annual

Review of Economics, 4(1):479–509, 2012.

Juan Or t n er and Sylvain Chassang. Making corruption harder: Asymmetric information,

collusion, and crime. Journal of Political Economy, 126(5):2108–2133, 2018.

Paolo Pinotti. Clicking on heaven’s door: The eﬀect of immigrant legalization on crime.

American Economic Review, 107(1):138–6 8 , 2017.

David Schoenherr. Political connections and allocative distortions. The Journal of Fi-

nance, 74(2):543–586, 2019.

Robert Wade. The system of administrative and p o l i ti ca l corruption: Canal irrigation in

south india. The Journal of Development Studies, 1 8( 3 ) : 28 7– 3 28 , 1982.

World Bank. Benchmarking Public Procurement. 2017.

Figure 1: Procedures and criteri a over time

Note: The graph shows, by year, the share of contracts awarded through, respectively, ﬁrst-price auctions as well as the

subset of ﬁrst-price auctions via negotiated procedure, and scoring rule auctions.

Figure 2: Share of contracts won by investigated ﬁrms, by reser

ve price

.1 .15 .2 .25 .3

Share Investigated Firms Winning

2000 2006 2008 2011 2015

year

150-500k 500k-1mil

above 1 mil

Note: The graph depicts the share of contracts awarded to investigated ﬁrms, separately by the reserve price: e150,000-

500,000; e500,000-1,000,000; and over e1,000,000.

Figure 3: Share of contracts won by investigated ﬁrms, by type of procedure

.1 .15 .2 .25

Share investigated ﬁrms Winning

2000 2004 2008 2012 2016

year

DiscretProc, lowN DiscretCrit

DiscretProc, highN

Type of Procedure

Note: The graph shows the share of contracts awarded to investigated ﬁrms, by type of procedure. In particular, the red

(diamond) line indicate the share of contracts awarded using DiscretCrit as an awarding criterion won by investigated ﬁrms,

the blue line (circles) indicates the share of contracts awarded using Discretproc

lowN

as procedure won by investigates

ﬁrms, and ﬁnally the green (square) line indicates the share of contracts awarded using Discretproc

highN

as procedure

won by investigates ﬁrms.

Figure 4: Discretionary procedures over time

Note: The graph shows the share of contracts awarded through, respectively, Discretionary Criterion, overall Discretionary

Procedures and Discretionary Procedures with few bidders, over time.

Table 1: Summary statistics by geographical area

(1) (2) (3)

South Center North

Investigated RUP 0.164 0.122 0.0697

(0.370) (0.328) (0.255)

Investigated Winner 0.175 0.161 0.168

(0.380) (0.367) (0.374)

Discr. Auction 0.149 0.125 0.298

(0.356) (0.331) (0.457)

Discr. Auction, Investigated RUP 0.178 0.138 0.323

(0.382) (0.345) (0.468)

Discr. Auction, Clean RUP 0.143 0.124 0.303

(0.350) (0.329) (0.460)

Note: The sample refers to the universe of contracts awarded by municipalities or other local authorities: 27 % of contracts

awarded in the South, 23 % in the Center and 50% in the North. InvestigatedRU P is an indicator equal to 1 if the public

oﬃcial in charge of the auction has been investigated. InvestigatedW inner is an indicator equal to 1 if the ﬁrm winning

the auction has been investigated. Discr.Auction denotes auctions for which either a discretionary procedure with fewer

than the legally mandated number of bidders (DiscretP roc

lowN

) or a discretionary criterion (DiscretCrit) has been used

to award the auction.

Table 2: Summary statistics fo r the full data

A. Auction Level

(1)

Mean Median S.D. N

Discretion 0.22 0.00 0.42 211,507

DiscretCrit 0.17 0.00 0.38 211,507

Discre tP roc

lowN

0.06 0.00 0.24 211,507

iscretP roc

highN

0.31 0.00 0.46 211,507

DiscretProc 0.37 0.00 0.48 211,507

Price Only Auction 0.83 1.00 0.38 211,507

investigated Firm 0.17 0.00 0.38 200,092

Investigated RUP 0.10 0.00 0.30 211,507

No. Bidders 26.93 10.00 41.64 210,405

No. Invi ted Bidders 7.48 4.00 16.78 103,205

Reserve Price (mil) 0.92 0.30 14.14 195,718

Winning D i sco u nt 18.22 16.88 11.58 192,362

Extra Cost (wrt Base) 7.01 3.37 13.85 83,088

Contra ctu a l Duration 239.91 180.00 224.98 144,942

Delay (days) 135.08 73.00 220.48 108,663

B. Admini str a t i on Level

(1)

Mean Median S.D. N

Investigated PA 0.16 0.00 0.37 14,024

Area==North 0.51 1.00 0.50 9,328

Area==Center 0.13 0.00 0.34 9,328

Area==South 0.35 0.00 0.48 9,328

Total N. Auctions, by PA 15.06 4.00 68.25 14,024

Total Value (in bil), by PA 148.00 17.89 2,061.68 14,024

type==Central Admin 0.02 0.00 0.14 14,024

PA type==Other Local PA 0.05 0.00 0.22 14,024

PA type==Cities 0.57 1.00 0.50 14,024

PA type==Transportatio n s 0.03 0.00 0.16 14,024

type==Hospit al s & University 0.1 7 0.00 0.38 14,024

PA type==Other 0.17 0.00 0.37 14,024

Popul at i o n == Pop. u p to 5k 0.67 1.00 0.47 7,004

Popul at i o n == 5-10 k 0.16 0.00 0.37 7,004

Popul at i o n == 10-2 0 k 0.09 0.00 0.29 7,004

Popul at i o n == 20-6 0 k 0.06 0.00 0.23 7,004

Popul at i o n == 60-2 5 0k 0.01 0.00 0.11 7,004

Popul at i o n == above 250k 0.00 0.00 0.04 7,004

Note: DiscretP roc denotes all negotiated procedures. DiscretP roc

highN

denotes negotiated procedures with at least

the legally mandated number of bidders. DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally

mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions for which either

DiscretP roc

lowN

=1 or DiscretCrit=1. Winning Discount is measured as a percentage of discount relative to the initial

reserve price. ExtraCost is measured as a percentage of the initial reserve price. ContractualDuration and Delay are

both measured in days.

Table 3: Summary statistics fo r identi ﬁ cat i o n

All PAs Cities

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

South Center North

Total PAs 14,384 2,374 937 4,098

Total PA, > 1 Au cti o n 10,439 2,140 863 3,573

At least 1 Discret 6,845 1,372 530 2,653

At least 1 DiscretCrit 5,993 1,290 473 2,226

At least 1 DiscretP roc

lowN

3,214 341 224 1,593

A w. Vari an ce in Discret 6,387 1,323 526 2,495

PA w. Variance DiscretCrit 5,667 1,243 470 2,125

PA w. Variance in DiscretP roc

lowN

3,156 341 223 1,581

Note: DiscretP roc denotes negotiated procedures. DiscretP roc

lowN

denotes negotiated procedures with fewer than the

legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions for which

either DiscretP roc

lowN

=1 or DiscretCrit=1.

Table 4: Auction-level regressions, investigated winner

all procurement authorities all city councils

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

DiscretCrit 0.012 2

∗∗∗

0.0132

∗∗∗

0.0133

∗∗∗

0.0191

∗∗∗

0.0199

∗∗∗

0.0197

∗∗∗

[0.00325] [0.00328] [0.00328] [0.00400] [0.00401] [0.00403]

DiscretProc

lowN

0.0215

∗∗∗

0.0229

∗∗∗

0.0222

∗∗∗

0.0127

∗∗

0.0152

∗∗∗

0.0163

∗∗∗

[0.00495] [0.00500] [0.00512] [0.00592] [0.00589] [0.0058

DiscretProc

highN

0.00183 0.00326 -0.00321 -0.00336

[0.00316] [0.00312] [0.00425] [0.00423]

Discretion 0.0147

∗∗∗

0.0199

∗∗∗

[0.00304] [

0.00367]

Dep. Var. Mean 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170

Observations 199089 199089 199089 199089 199089 107994 107994 107994 107994 107994

R-sq 0.118 0.118 0.118 0.118 0.118 0.130 0.129 0.130 0.130 0.130

Note: In all speciﬁcations, the dependent variable is an indicator equal to 1 if an investigated ﬁrm is awarded the contract.

DiscretP roc

highN

denotes negotiated procedures with at least the legally mandated number of bidders. DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule

auctions. Discretion denotes auctions for which either DiscretP roc

lowN

=1 or DiscretCrit=1. All regressions include PA

and Year ﬁxed eﬀects, a linear control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds

(up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics:

4 dummies for category type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract

was awarded under urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust standard

errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table 5: Auction-level regressions, choice of procedure

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

Discretion DiscretProc

lowN

DiscretCrit Discretion Discretion Discretion DiscretProc

lowN

DiscretCrit

Investigated RUP 0.0298

∗∗∗

0.00996

∗∗

0.0210

∗∗∗

0.0189

∗∗∗

0.0339

∗∗∗

0.000439 0.0330

∗∗∗

[0.00805] [0.00402] [0.00766] [0.00650] [0.00854] [0.00419] [0.00780]

Investigated PA -0.0170

∗∗∗

-0.0257

∗∗∗

0.00372 -0.0291

∗∗∗

[0.00639] [0.00754] [0.00461] [0.00598]

Dep. Var. Mean 0.222 0.222 0.222 0.222 0.222 0.222 0.0589 0.169

Observations 206421 206421 206421 166768 166768 166768 166768 166768

R-sq 0.325 0.257 0.321 0.210 0.210 0.211 0.131 0.196

Geog. FE PA PA PA Region Region Reg i on Region Region

Note: The dependent variable is indicated on top of each column. DiscretP roc denotes all negotiated procedures.

DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit

denotes scoring rule auctions. Discretion denotes auctions for which either DiscretP roc

lowN

=1 or DiscretCrit=1.

InvestigatedRUP is an indicator equal to 1 if the public oﬃcial in charge of the auction has been investigated.

InvestigatedP A is an indicator equal to 1 if any of the public oﬃcials in the PA have been investigated. All regres-

sions include Year ﬁxed eﬀects, a l inear control for reserve price (in log) Price and 5 dummies for diﬀerent contract size

thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract char-

acteristics: 4 dummies for category type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether

the contract was awarded under urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust

standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table 6: Auction-level regressions, outcomes

Delay (Asinh) Winning Disco u nt Extra Cost

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Discretion -0.142

∗∗∗

-4.031

∗∗∗

-0.312

[0.0469] [0.267] [0.28 2]

DiscretProc

lowN

-0.259

∗∗∗

-0.129

∗

-3.965

∗∗∗

-3.023

∗∗∗

0.396 0.492

[

0.0766] [0.0756] [0.422] [0.356] [0.509] [0.520]

DiscretCrit -0.0778 -0.0837 -3.971

∗∗∗

-4.117

∗∗∗

-0.640

∗∗

-0.656

∗∗

[0.0538] [0.0535] [0.241] [0.251] [0.268] [0.270]

iscretProc

highN

-0.340

∗∗∗

-2.426

∗∗∗

-0.276

[0.0635] [0.356] [0.215]

Dep. Var. Mean 3.296 3.296 3.296 18.11 18.11 18.11 7.035 7.035 7.035

Observations 107067 107067 107067 191053 1 9 10 53 191053 81439 81439 81439

R-sq 0.250 0.250 0.251 0.443 0.444 0.448 0.219 0.219 0.219

Note: The dependent variable is indicated at the top of each column. Delay is the inverse hyperbolic sine transformation

of the number of days between the expected contractual duration and the eﬀective total completion time. W inning

Discount is the ﬁnal price of the winning bid expressed as a discount over the reserve price (Discount) and ExtraCost

represents excess completion costs, calculated as the diﬀerence between the ﬁnal price and awarding price, over the initial

reserve price. DiscretP roc

highN

denotes negotiated procedures with at least the the legally mandated number of bidders.

DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit

denotes scoring rule auctions. Discretion denotes auctions for which either DiscretP roc

lowN

=1 or DiscretCrit=1. All

regressions include PA and Year ﬁxed eﬀects, a linear control for reserve price (in log) Price and 5 dummies for diﬀerent

contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls

for contract characteristics: 4 dummies for category type (Civil Building, Roadworks, Specialized Works or Others), 1

dummy for whether the contract was awarded under urgency and 1 dummy for whether the object of the contract entailed

maintenance. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Appendix: For Online Publication Only

A Additional Tables and Figures

Figure A1: Regulatory Constraints and the Procurement Audit System

Internal Audit

System

- First Level Audit

Management Audit

(Adg)

Control Audit (Adc)

- Second Level Audit

(Ada)

European Procurement Directives

Directives 2004/17/EC and 2004/18/EC

National Procurement Law

Legislative Decree 12/4/2006, n. 163

Local Procurement Regulation

Regional, Provincial, Municipal

Contractin g Authority Guidelines

Contractin g Oﬃcer Discretion on

Selection Procedure and Award

Criterion

External Audit

System

- Anticorruption Au-

thority

- Court of Auditors

- Third Level Audit

(IGRUE) and EU-level

audit, on ly for certain

EU funded projects

Note: The ﬁgure illustrates the set of regulatory constraints and audit oversight, sub j ect to which a contracting oﬃcer

exercises discretion over the supplier selection procedure and contract awarding criterion. At any point in time, the

exact set of regulations and audit processes applicable depend on the contract reserve price, job characteristics, source of

project funding, and the identity of the contracting authority. The system has changed over time, but for most of the

contracts in our sample, the relevant regulations are the European Procurement Directives 2004/17 and 2004/18 and Italian

procurement law (L.D. 163/2006). For the typical contract, the audit process has two levels and is also subject to scrutiny

by external auditors. When the project is at least in part funded by the EU, there is a third audit level conducted by the

regional oﬃces of the Ministry of the Treasury (IGRUE) and, possibly, further levels of European audits as well.

Figure A2: The Investigation Process

One of the country’s four police forces is notiﬁed of

potential crimes by private citizens or publi c oﬃcials

If the prelim i n ar y evidence is deemed suﬃcient,

the potential crime is registered in a centralized

database (SDI) and a police investigation begins

(under supervision of a public prosecutor (P.M.))

If there are suitable elements to pro-

ceed, the P.M. requests the su pervising

judge for Prelim i n a r y Inquiries (G.I.P.)

to refer the case to the court for a pre-

liminary hearing before a judge (G.U.P.)

The defendant is notiﬁed of pre-

liminary hearings and has the

right to be defended by a lawyer

The G.U.P. consid er s the argu-

ments brought by the prosecutor

and defendants lawyer and de-

cides whether to dismiss or

begin a forma l criminal trial

The case is brought before

the First Inst an ce Court

Note: The ﬁgure shows the various steps in the investigation process in Italy. Our data comes from the second step,

highlighted in red.

Figure A3: Distribution of number of bidders, by type of awarding criterion

0 .02 .04 .06 .08 .1

Density

0 4 8 12 16 20 24 28 32 36 40

No. Bidders

Lowest Price DiscretCrit

Note: The ﬁgure represents two histograms of the number of bidders, for auctions using lowest price or scoring rule

(DiscretCrit) as awarding criteria. For ease of visualization, the plot is limited to auctions with up to 40 bidders, which

represent 80% of auctions in our sample.

Figure A4: Regression discontinuity plots

0 1.0e-06 2.0e-06 3.0e-06 4.0e-06

Density

300000 400000 500000 600000 700000

Reserve Price

0 1.0e-07 2.0e-07 3.0e-07 4.0e-07

Density

800000 900000 1000000 1100000 1200000

Reserve Price

0 .1 .2 .3 .4

300000 400000 500000 600000 700000

Sample average within bin 2th order global polynomial

Regression function ﬁt

0 .1 .2 .3 .4 .5

600000 800000 1000000 1200000 1400000

Sample average within bin 2th order global polynomial

Regression function ﬁt

Note: These graphs depict the results of our analysis using a Regression Discontinuity Design. The top panels display the

density of contracts with reserve price around the e500,000 and e1,000,000 cutoﬀs, respectively. The green bands depict

conﬁdence intervals f or the of the estimated density function. The bottom panels display the average fraction of contracts

awarded to investigated ﬁrms across equally-sized bins of the reserve price, and ﬁtted polynomials functions on each side

of the cutoﬀ. All estimates are performed using optimal bandwidth selection procedure by Cattaneo et al. [2019].

Table A1: Summary Statistics by Secto r and RUP type

(1) (2)

Roads Buildings

investigated Firm 0.192 0.127

(0.394) (0.333)

Investigated RUP 0.113 0.103

(0.316) (0.303)

Discretion 0.238 0.282

(0.426) (0.450)

Discr. Auction, Investigated RUP 0.221 0.287

(0.415) (0.452)

Discr. Auction, clear RUP 0.243 0.285

(0.429) (0.452)

(1) (2)

Local RUP Not Local RUP

Investigated RUP 0.127 0.112

(0.333) (0.315)

investigated Firm 0.170 0.159

(0.376) (0.366)

local Firm 0.226 0.122

(0.419) (0.327)

Discretion 0.185 0.245

(0.389) (0.430)

Discr. Auction, Investigated RUP 0.160 0.252

(0.366) (0.434)

Discr. Auction, clear RUP 0.189 0.245

(0.392) (0.430)

Note: InvestigatedRU P is an indicator equal to 1 if the public oﬃcial in charge of the auction has been investigated.

InvestigatedW inner is an indicator equal to 1 if the ﬁrm winning the auction has been investigated. Discr.Auction

denotes auctions for which either a discretionary procedure with fewer than the legally mandated number of bidders

(DiscretP roc

lowN

) or a discretionary criterion (DiscretCrit) has been used to award the auction.

Table A2: Auction-level regressions, investigated winner - Restrictive deﬁnition

all cities

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

DiscretCrit 0. 0 0 98 3

∗∗∗

0.0107

∗∗∗

0.0109

∗∗∗

0.0143

∗∗∗

0.0149

∗∗∗

0.0150

∗∗∗

[0.00275] [0.00281] [0.00281] [0.00324] [0.0032 6] [0.00326]

DiscretProc

lowN

0.0181

∗∗∗

0.0193

∗∗∗

0.0163

∗∗∗

0.00979

∗∗∗

0.0117

∗∗∗

0.0110

∗∗∗

[0.00408] [0.00418] [0.00426] [0. 00 3 45 ] [0.00352 ] [0.00364]

DiscretProc

highN

0.00773

∗∗∗

0.00864

∗∗∗

0.00209 0.00180

[

0.00230] [0.00228] [0.00287] [0.00277]

Discretion 0.0119

∗∗∗

0.0148

∗∗∗

[0.00253] [0.00281]

Dep. Var. Mean 0.170 0.170 0.170 0.170 0.170 0.170 0 . 17 0 0.170 0.170 0.170

Observations 199089 199089 199089 199089 199089 107994 107994 107994 107994 107994

R-sq 0.103 0.103 0.103 0.104 0.103 0.112 0. 11 2 0.112 0.112 0.112

Note: In all speciﬁcations, the dependent variable is an indicator equal to 1 if an investigated ﬁrm is awarded the contract.

In this table, we restrict the deﬁnition of investigated ﬁrms to those investigated for (i) corruption, malfeasance and

embezzlement or (ii) abuse of power and undue inﬂuence, (i.e., we do not include in our deﬁnition those investigated for

violations in public auctions. DiscretProc denotes negotiated procedures. DiscretPro c

lowN

denotes negotiated procedures

with fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes

auctions for which either DiscretProc

lowN

=1 or DiscretCrit=1. All regressions include PA and Year ﬁxed eﬀects, a linear

control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k,

300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category type

(Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under urgency

and 1 dummy for whether the object of the contract entailed maintenance. Robust standard errors clustered at the PA

level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A3: Auction-level regressions, investigated winner - Br

oad deﬁnition

all cities

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

DiscretCrit 0.0170

∗∗∗

0.0181

∗∗∗

0.0182

∗∗∗

0.0203

∗∗∗

0.0210

∗∗∗

0.0212

∗∗∗

[0.00369] [0.00371] [0.00372] [0.00470] [0.00470] [0.0047

DiscretProc

lowN

0.0212

∗∗∗

0.0231

∗∗∗

0.0206

∗∗∗

0.0125

∗

0.0152

∗∗

0.0143

∗∗

[0.00557] [0.00559] [0.00588] [0.00714] [0.00711] [0.00723]

DiscretProc

highN

0.00650

∗

0.00719

∗∗

0.00278 0.00224

[0.00378] [0.00362] [0.00504] [0.00496]

Discretion 0.0180

∗∗∗

0.0201

∗∗∗

[0.00337] [

0.00424]

Dep. Var. Mean 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170

Observations 199089 199089 199089 199089 199089 107994 107994 107994 107994 107994

R-sq 0.138 0.138 0.138 0.138 0.138 0.148 0.148 0.148 0 . 1 48 0.148

Note: In all speciﬁcations, the dependent variable is an indicator equal to 1 if an investigated ﬁrm is awarded the contract.

In this table, we extend the deﬁnition of investigated ﬁrms to include ﬁrms investigated for waste management crimes.

DiscretProc denotes negotiated procedures. DiscretProc

lowN

denotes negotiated procedures with fewer than the legally

mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions for which either

DiscretProc

lowN

=1 or DiscretCrit=1. All regressions include PA and Year ﬁxed eﬀects, a linear control for reserve price

(in log) Price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-

1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category type (Civil Building, Roadworks,

Specialized Works or Others), 1 dummy for whether the contract was awarded under urgency and 1 dummy for whether

the object of the contract entailed maintenance. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A4: Auction-level regressions, Convicted Winner

all cities

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

DiscretCrit 0.00190 0.00205 0.00210 0.00237 0.00252 0.00265

[0.00205] [0.00209] [0.00212] [0.00336] [0.00346] [0.00357]

DiscretProc

lowN

0.00305

∗

0.00327

∗

0.00251

∗

0.00281 0.00313 0.00220

[0.00160] [0.00173] [ 0 . 0 01 44 ] [0.00234] [0.00266] [0.00201]

DiscretProc

highN

0.00199 0.00208 0.00269 0. 0 026 8

[

0.00158] [0.00147] [0.00282] [0.00261]

Discretion 0.00221 0.00255

[0.00188] [0.00321]

Dep. Var. Mean 0.0169 0.0169 0.0169 0.0169 0.0169 0.0169 0.0169 0.170 0.0169 0.0169

Observations 199089 199089 199089 199089 199089 107994 10 79 94 107994 107994 10 7 99 4

R-sq 0.129 0.129 0.129 0.129 0.129 0.157 0.157 0.157 0.157 0.157

Note: In this table, in all speciﬁcations, the dependent variable is an indicator equal to 1 if a ﬁrm ever convicted for

corruption is awarded the contract. DiscretPro c denotes negotiated procedures. DiscretProc

lowN

denotes negotiated

procedures with fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion

denotes auctions for which either DiscretProc

lowN

=1 or DiscretCrit=1. All regressions include PA and Year ﬁxed eﬀects,

a linear control for reserve price (in log) Price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k,

150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category

type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under

urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust standard errors clustered at

the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A5: Auction-level regressions, PA X Year ﬁxed eﬀects

all cities

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

DiscretCrit 0.00752

∗

0.00791

∗

0.00795

∗

0.0186

∗∗∗

0.0188

∗∗∗

0.0187

∗∗∗

[0.00455] [0.00456] [0.00456] [0.00647] [0.00648] [0 . 00 64

DiscretProc

lowN

0.0236

∗∗∗

0.0239

∗∗∗

0.0224

∗∗∗

0.0176

∗∗

0.0180

∗∗

0.0196

∗∗∗

[0.00572] [0.00575] [0.00602] [0.00760] [0.0 07 6 0] [0.00758]

DiscretProc

highN

0.00375 0.00559 -0.00476 -0.00451

[0.00415] [0.00407] [0.00636] [0.00633]

Discretion 0.0116

∗∗∗

0.0206

∗∗∗

[0.00410] [

0.00538]

Dep. Var. Mean 0.1 70 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170

Observations 170210 170 2 10 170210 170210 170210 86195 86195 86195 86195 86195

R-sq 0 . 24 1 0.241 0.241 0.241 0.241 0.289 0.289 0.289 0.289 0.289

Note: In all speciﬁcations, the dependent variable is an indicator equal to 1 if an investigated winner is awarded the

contract. DiscretProc denotes negotiated procedures. DiscretProc

lowN

denotes negotiated procedures with fewer than

the legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions for which

either DiscretProc

lowN

=1 or DiscretCrit=1. A ll regressions include PA*Year ﬁxed eﬀects, a linear control for reserve price

(in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-

1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category type (Civil Building, Roadworks,

Specialized Works or Others), 1 dummy for whether the contract was awarded under Urgency and 1 dummy for whether

the object of the contract entailed maintenance. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A6: Auction-level regressions, investigated winner, controll i n g for connections

all cities

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

DiscretCrit 0.00853

∗∗∗

0.00937

∗∗∗

0.00934

∗∗∗

0.0162

∗∗∗

0.0168

∗∗∗

0.0166

∗∗∗

[0.00318] [0.00321] [0.00321] [0.00391] [0.00392] [0.00393

]

Connected 0.140

∗∗∗

0.140

∗∗∗

0.140

∗∗∗

0.140

∗∗∗

0.140

∗∗∗

0.129

∗∗∗

0.129

∗∗∗

0.129

∗∗∗

0.129

∗∗∗

0.129

∗∗∗

[0.00445] [0.00445] [0. 0 04 44 ] [0.00444] [0.00444] [0.00552] [0.00553] [0.00552] [0.00551] [0.00551]

DiscretProc

lowN

0.0176

∗∗∗

0.0186

∗∗∗

0.0189

∗∗∗

0.0101

∗

0.0123

∗∗

0.0140

∗∗

[0.00481] [0.00485] [0.00494] [0.00600] [0.00598] [0.0058

DiscretProc

highN

-0.00102 0.000425 -0.00501 -0.00511

[0.00312] [0.00310] [0.00418] [0.00419]

Discretion 0.0110

∗∗∗

0.0167

∗∗∗

[0.00295] [

0.00362]

Constant -0.377

∗∗∗

-0.383

∗∗∗

-0.380

∗∗∗

-0.379

∗∗∗

-0.379

∗∗∗

-0.203

∗∗∗

-0.209

∗∗∗

-0.204

∗∗∗

-0.200

∗∗∗

-0.201

∗∗∗

[0.0583] [0.0579] [0.0581] [0.0589] [0.0589] [0.0748] [0.0753] [0.0750] [0.0755] [0. 07 5 4]

Dep. Var. Mean 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170

Observations 199089 199089 199089 199089 199089 107994 107994 107994 107994 107994

R-sq 0.143 0.143 0.143 0.143 0.143 0.151 0.151 0.151 0 . 1 51 0.151

Note: In all speciﬁcations, the dependent variable is an indicator equal to 1 if an investigated ﬁrm is awarded the contract.

DiscretP roc

highN

denotes negotiated procedures with at least the legally mandated number of bidders. DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule

auctions. Discretion denotes auctions for which either DiscretP roc

lowN

=1 or DiscretCrit=1. All regressions include PA

and Year ﬁxed eﬀects, a linear control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds

(up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics:

4 dummies for category type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract

was awarded under urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust standard

errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A7: Bidder-level regressions, participants’ pool

participant auction

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

DiscretCrit 0.00248 0.00240 0.002 32 0.0199

∗∗

0.0198

∗∗

0.0197

∗∗

[0.00292] [0.00292] [0.00292] [0.00924 ] [0.00923] [0.0093 0 ]

DiscretPro c

lowN

0.0125

∗∗

0.0125

∗∗

0.0143

∗∗∗

0.0221

∗∗

0.0220

∗∗

0.0221

∗∗

[0.00533] [0.00533] [0.00538] [0.00896 ] [0.00896] [0 . 00 91 2]

DiscretPro c

highN

-0.00364 -0.00282 -0.000338 0.000571

[0.00392] [0.00392] [0.00801] [0.00774]

Discretion 0.00114 0.0223

∗∗∗

[0.00228] [0.00738]

Dep. Var. Mean 0.163 0.163 0.163 0.163 0.163 0.161 0.161 0.161 0.161 0.161

Observations 462821 462821 462821 462821 462821 24197 24197 24197 24197 24197

R-sq 0.0562 0.0563 0.0563 0.0563 0.0562 0.223 0.223 0.223 0.223 0.223

Note: In columns 1-5, the dependent variable is an indicator equal to 1 if an investigated ﬁrm participates in the auction.

The unit of observation is the auction participant, so we have multiple observation per auction. Columns 6-10 replicate

columns 6-10 of Table 4, but restricts the sample to auctions for which we have information on the participants. Across

all columns, we restrict attention to contracts awarded by municipal councils. DiscretProc denotes negotiated procedures.

DiscretProc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit

denotes scoring rule auctions. Discretion denotes auctions for which either DiscretProc

lowN

=1 or DiscretCrit=1. All re-

gressions include controls for participant ﬁrms’ characteristics, and in particular ﬁrm net worth, ﬁrm size, proﬁts, operating

margin, negative operating margin dummy, change in operating margin. Regressions also include PA and Year ﬁxed eﬀects,

a linear control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k,

150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category

type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under

urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust standard errors clustered at

the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A8: Auction-level regressions, investigated winner on investigated RUP

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

Investigated RUP 0.0095

∗

0.0095

∗

0.0092

∗

0.0092

∗

0.0093

∗

0.0098

∗

0.0096

∗

0.0095

∗

0.0097

∗

0.0093

∗

[0.0052] [0.0052] [0.0052] [0.0052] [0.0052] [0.0052] [0.0052] [0.0052] [0.0052] [0.0052]

DiscretCrit 0.0130

∗∗∗

0.0140

∗∗∗

0.0141

∗∗∗

0.0130

∗∗∗

[0.0033] [0.0033] [0.0 03 3 ] [0.0033]

iscretProc

lowN

0.0215

∗∗∗

0.0230

∗∗∗

0.0224

∗∗∗

0.0215

∗∗∗

[0.0050] [0.0050] [0.0051] [0.0050]

DiscretProc

highN

0.0015 0.0029 0.0043 0.0029

[0.0032] [0.0032] [0.0032] [0.0032]

Discretion 0.0154

∗∗∗

0.0154

∗∗∗

[0.0031] [

0.0031]

PA FE No No No No No Yes Yes Yes Yes Yes

Dep. Var. Mean 0.170 0.1 7 0 0.170 0.170 0.170 0.170 0.170 0.170 0.170 0.170

Observations 1 95 158 195158 195158 195158 195158 195158 195158 195158 195158 195158

R-sq 0.118 0.118 0.118 0.118 0.118 0.117 0.118 0.11 8 0.117 0.118

Note: This table is the counterpart of Table 5 but including Investigated RUP among the regressors.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A9: Auction-level regressions, choice of procedure, pro

vince FE

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

Discretion DiscretProc

lowN

DiscretCrit Discretion Discretion Discretion DiscretProc

lowN

DiscretCrit

Investigated RUP 0.0298

∗∗∗

0.00996

∗∗

0.0210

∗∗∗

0.0207

∗∗∗

0.0391

∗∗∗

0.00167 0.0381

∗∗∗

[0.00805] [0.00402] [0.00766] [0.00731] [0.0100] [0.00443] [0.00888]

Investigated PA -0.0170

∗∗∗

-0.0297

∗∗∗

0.00124 -0.0318

∗∗∗

[0.00608] [0.00786] [0.00420] [0.00589]

Dep. Var. Mean 0.222 0.222 0.222 0.222 0.222 0.222 0.0589 0.169

Observations 206421 206421 206421 110618 110618 110618 110618 110618

R-sq 0.325 0.257 0.321 0.228 0.228 0.229 0.143 0.212

Geog. FE PA PA PA Province Province Province Province Provin ce

Note: This Table is the counterpart of table 5 but using a ﬁner partition for the geographic ﬁxed eﬀects, one for each of

Italy’s 110 provinces. DiscretProc denotes negotiated procedures. DiscretP roc

lowN

denotes negotiated procedures with

fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions

for which either DiscretProc

lowN

=1 or DiscretCrit=1. Investigated RUP is an indicator equal to 1 if the public oﬃcial

in charge of the auction has been investigated. Investigated PA is an indicator equal to 1 if any of the public oﬃcials

in the PA have been investigated. All regressions include Year ﬁxed eﬀects, a linear control for reserve price (in log)

price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil,

over 1.5mil) as well as controls for contract characteristics: 4 dummies for category type (Civil Building, Roadworks,

Specialized Works or Others), 1 dummy for whether the contract was awarded under urgency and 1 dummy for whether

the object of the contract entailed maintenance. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A10: Auction-level regressions, choice of DiscretProc procedures

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

Investigated RUP 0.0008 5 2 0.00873 0.0008 46

[0.0101] [0.00738] [0.00972]

Investigated PA 0.0123 0.0120

[0.0101] [0.0121]

Dep. Var. Mean 0.222 0.222 0.222 0.222

Observations 109511 110269 11 0 26 9 110269

R-sq 0.574 0.500 0.500 0.50 0

Geog. FE PA Region Region Region

Note: The dependent variable across columns is DiscretProc, which denotes all negotiated procedures. Investigated RUP

is an indicator equal to 1 if the public oﬃcial in charge of the auction has been investigated for corruption. Investigated

PA is an indicator equal to 1 if at least one RUP in the PA has been investigated. All regressions include Year ﬁxed eﬀects,

a linear control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k,

150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category

type (Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under

urgency and 1 dummy for whether the object of the contract entailed maintenance. Robust standard errors clustered at

the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A11: Auction-level regressions, predicting the presenc

e of outcomes’ data

Delay (Asinh) Winning Di sco u nt Extra Cost

(1) (2) (3) (4) (5) (6) (7) (8) (9)

investigated Firm 0.00286 0.000646 0.00531

[0.00350] [0.00200] [0.00350]

Investigated RUP 0.00920 -0.000575 -0.00588

[0.0121] [0. 0 03 9 4] [0.0115]

Investigated PA -0.0356

∗∗∗

-0.00934

∗∗∗

-0.0159

∗

[0.0117] [0.00359] [0.00935]

onstant -0.281

∗∗∗

-0.284

∗∗∗

-0.106 -0.336

∗∗∗

-0.336

∗∗∗

-0.372

∗∗∗

-0.0595 -0.0608 -0.0569

[

0.0704] [0.0704] [0.0649] [0.0521] [0.0520] [0.0562] [0.0538] [0.0541] [0.0548]

Dep. Var. Mean 0.487 0.487 0.487 0.0918 0.0918 0.0918 0.608 0.608 0.608

Observations 155574 155574 155574 155574 155574 155574 155574 155574 15557 4

R-sq 0.295 0.295 0.153 0.147 0.147 0.0577 0.290 0.290 0.163

Note: The outcomes in this table are dummies for the presence of information on the outcomes used in Table 6. Regressions

in columns (1), (2), (4), (5), (7), (8) include PA and Year ﬁxed eﬀects, a linear control for reserve price (in log) price and 5

dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil)

as well as controls for contract characteristics: 4 dummies for category type (Civil Building, Roadworks, Specialized

Works or Others), 1 dummy for whether the contract was awarded under urgency and 1 dummy for whether the object

of the contract entailed maintenance. Regressions in columns (3), (6), (9) include Region instead of PA ﬁxed eﬀects

as the main regressor only varies at the PA level. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A12: Auction-level regressions, subsample of auctions with outcomes’ data

all cities

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

DiscretCrit 0.0168

∗∗∗

0.0179

∗∗∗

0.0178

∗∗∗

0.0254

∗∗∗

0.0266

∗∗∗

0.0266

∗∗∗

[0.00585] [0.00585] [0.00586] [0.00733] [ 0 . 00 730 ] [0.00737]

DiscretProc

lowN

0.0278

∗∗∗

0.0291

∗∗∗

0.0293

∗∗∗

0.0279

∗∗∗

0.0302

∗∗∗

0.0302

∗∗∗

[0.00774] [0.00777] [0.00803] [0.00887] [0.00890] [0.00898]

DiscretProc

highN

-0.000638 0.00111 -0.000131 0.000603

[

0.00520] [0.00505] [0.00693] [0. 00 67 5]

Discretion 0.0211

∗∗∗

0.0299

∗∗∗

[0.00491] [0.00599]

Dep. Var. Mean 0.161 0.161 0.161 0.161 0.161 0.161 0.161 0.161 0.161 0.161

Observations 66458 66458 66458 66458 66458 37311 37311 37311 37311 37311

R-sq 0.145 0.145 0.145 0.145 0.145 0.165 0.165 0. 1 6 5 0.165 0.165

Note: This table is analogous to Table 4, but restricting the sample to the subset of auctions for which we have information

on the outcomes used in Table 6. In all speciﬁcations, the dependent variable is an indicato

r equal to 1 if an investigated

ﬁrm is awarded the contract. DiscretProc denotes negotiated procedures. DiscretProc

lowN

denotes negotiated procedures

with fewer than the legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes

auctions for which either DiscretProc

lowN

=1 or DiscretCrit=1. All regressions include PA and Year ﬁxed eﬀects, a linear

control for reserve price (in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k,

300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category type

(Civil Building, Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under urgency

and 1 dummy for whether the object of the contract entailed maintenance. Robust standard errors clustered at the PA

level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A13: Auction-level regressions, outcomes (municipalit

ies only)

Delay (Asinh) Winning Disco u nt Extra Cost

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Discretion -0.156

∗∗∗

-4.313

∗∗∗

-0.530

∗

[0.0508] [0.388] [0.272]

DiscretProc

lowN

-0.462

∗∗∗

-0.334

∗∗∗

-3.153

∗∗∗

-2.418

∗∗∗

0.276 0.242

[0.0825] [0.0862] [0.571] [0.401] [0.438] [0.428]

DiscretCrit -0.0417 -0.0586 -4.667

∗∗∗

-4.829

∗∗∗

-0.776

∗∗

-0.768

∗∗

[0.0601] [0.0595] [0.316] [0.342] [0.301] [0.310]

iscretProc

highN

-0.358

∗∗∗

-2.105

∗∗∗

0.108

[0.0626] [0.601] [0.309]

Dep. Var. Mean 3.296 3.296 3.296 18.11 18.11 18.11 7.035 7.035 7.035

Observations 58071 58071 58 07 1 104628 104628 104628 46276 46276 46276

R-sq 0.260 0.260 0.261 0.437 0.439 0.442 0.249 0.249 0.249

Note: The dependent variable is indicated on top of each column. Delay is the inverse hyperbolic sine transformation of

the number of days between the expected contractual duration and the eﬀective total completion time. Winning Discount

is the ﬁnal price of the winning bid expressed as a discount over the reserve price (Discount) and Extra Cost represents

excess completion costs, calculated as the diﬀerence between the ﬁnal price and awarding price, over the initial reserve

price. DiscretProc denotes negotiated procedures. DiscretProc

lowN

denotes negotiated procedures with fewer than the

legally mandated number of bidders. DiscretCrit denotes scoring rule auctions. Discretion denotes auctions for which

either DiscretProc

lowN

=1 or DiscretCrit=1. All regressions include PA and Year ﬁxed eﬀects, a linear control for reserve

price (in log) price and 5 dummies for diﬀerent contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-

1mil, 1-1.5mil, over 1.5mil) as well as controls for contract characteristics: 4 dummies for category typ e (Civil Building,

Roadworks, Specialized Works or Others), 1 dummy for whether the contract was awarded under urgency and 1 dummy

for whether the object of the contract entailed maintenance. Robust standard errors clustered at the PA level are in

parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

Table A14: Auction-level regressions, direct eﬀect of Investigated RUP and Investigated

winner on ou t com es

Delay (Asinh) Winning Discount Extra Cost

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Investigated Firm -0.00965 -0.00419 -0.460

∗∗∗

-0.471

∗∗∗

-0.233

∗

-0.264

∗

[0.0337] [0.0343] [0.0776] [0.0783] [0.138] [0.140]

Investigated RUP -0.0743 -0.0764 -0.734

∗∗∗

-0.731

∗∗∗

0.198 0.159

[0.0771] [0.0776] [0.223] [0.224] [0.318] [0.319]

Constant -5.360

∗∗∗

-5.659

∗∗∗

-5.436

∗∗∗

6.713

∗∗∗

10.21

∗∗∗

7.875

∗∗∗

-7.139

∗∗

-6.798

∗∗

-7.406

∗∗

[0.678] [0.682] [0.690] [2.417] [2.725] [2.343] [3.392] [3.293] [3.415]

Dep. Var. Mean 3.296 3.296 3.296 18.11 18.11 18.11 7.035 7.035 7.035

Observations 101346 105102 99400 180469 187674 177123 77015 79984 75570

R-sq 0.2 49 0.249 0.248 0.435 0.429 0.434 0.222 0.219 0.223

Note: The dependent variable is indicated at the top of each column. Delay is the inverse hyperbolic sine transformation

of the number of days between the expected contractual duration and the eﬀective total completion time. W inning

Discount is the ﬁnal price of the winning bid expressed as a discount over the reserve price (Discount) and ExtraCost

represents excess completion costs, calculated as the diﬀerence between the ﬁnal price and awarding price, over the initial

reserve price. DiscretP roc

highN

denotes negotiated procedures with at least the the legally mandated number of bidders.

DiscretP roc

lowN

denotes negotiated procedures with fewer than the legally mandated number of bidders. DiscretCrit

denotes scoring rule auctions. Discretion denotes auctions for which either DiscretP roc

lowN

=1 or DiscretCrit=1. All

regressions include PA and Year ﬁxed eﬀects, a linear control for reserve price (in log) Price and 5 dummies for diﬀerent

contract size thresholds (up to 100k, 100-150k, 150-300k, 300-500k, 500k-1mil, 1-1.5mil, over 1.5mil) as well as controls

for contract characteristics: 4 dummies for category type (Civil Building, Roadworks, Specialized Works or Others), 1

dummy for whether the contract was awarded under urgency and 1 dummy for whether the object of the contract entailed

maintenance. Robust standard errors clustered at the PA level are in parentheses.

∗

p < 0.1,

∗∗

p < 0.05,

∗∗∗

p < 0.01.

B Additional Details o n the Investigations Data

1. The sour ce of the procu r em ent data is the Public Contracts Observatory at t h e

Italian Anticorruption Authority (ANAC). We accessed these data through a di-

rect request to the Observatory, but t h e data have recently (September 2020) been

made available as open data through the portal accessible here: https://dati.

nticorruzione.it/opendata/dataset. The porta l lists all of the datasets avail-

able. To replicate our data, a researcher needs to select contracts for public works

involving either civic buildings (cod e: OG01), or tran sportation infrastructure such

as roads, highways, and bridges (code: OG03) and then select the foll owing vari-

ables: the start and end date of t h e biddin g process, the identity and type of

contractin g author i ty, the auction pro ced u r e used to award the contract, the selec-

tion criterion, the number of bidders (both invited and participating), the identity

of the winning bidder, initial project value, t h e winning discount, the total eﬀec-

tive costs, and the expected and eﬀective contractual duration. The identity of

the RUP i n charge of each contract i s considered sensitive information and must

be obtain ed through an ad hoc request to the Observatory motivated by research

purposes. Similarl y, the full list of ﬁrms part i ci p at i n g in the auction i s sensitive

information and must be requested to the Observatory as the list of ﬁr m s paying

for the bid d i n g fee (”contributo partecipazione”).

2. The source of in fo rm a t i on on ﬁrm owners and managers is the Company Ac-

counts Data System, a proprietary database maintained by a private compa ny,

the CERVED Group. Among the procurement data described above, the d at a set

“aggiudicatari” contains for each contract the name and social security number

(“codice ﬁscale”) of the winner. The latter variable uniquely identiﬁ es ﬁrm s in

the Company Accounts Data System and can thus be used to retrieve information

on the identi ty of their owners and managers.

We used the data observed for

our separate years: 2006, 2011, 2014 and 2016. For each ﬁrm, the union of a l l

owners and managers recorded in any of these four period s represents the set of

individuals connected to the ﬁrm in our analysis. We performed the same proce-

dure also for the ﬁrms participating in the auct i on . Access to the Company Ac-

counts Data System is available for a fee from the CERVED Group by contacting:

More precisely, we consider all of the individuals who either own shares of the ﬁrm or o ccu py at least

one of the positions monitored by CERVED: the board of directors, auditors, general managers, an d the

heads of legal technical oﬃces (the main rol es are: AMMINISTRATORE; AMMINISTRATORE DEL-

EGATO; AMMINISTRATORE UNICO; CONSIGLIERE; CONSIGLIERE DELEGATO; CURATORE

FALLIMENTARE; DELEGATO AL RITIRO CAPITALE VERSATO; DIRETTORE GENERALE; DI-

RETTORE TECNICO; INSTITORE; LEGALE RAPPRESENTANTE; PRESIDENTE; PRE SID E NTE

DEL COLLEGIO SINDACALE; PROCURATORE; RESPONSABILE TECNICO; SINDACO; SOCIO )

https://www.cerved-online.com/contatti.

3. Information on the identities o f ﬁr m owners and mana g ers ( a n d a l so RUPs’ iden-

tities) were used to retrieve their records of criminal investigations from which, by

aggregating up at the ﬁrm level, ﬂags were created for ﬁrms with at least one ﬁrm-

linked person under investigation. We could not directly link individuals to their

criminal records, and thus th e process of generating ﬁrm-level ﬂags for ﬁrms linked

to investigated individuals was per for m ed by AISI, Italys internal intelligence and

security agency. Our access to the data is enabled via a n agreement between AISI

and Bocconi University. AISI used a centralized archive, the Sistema DIndagine In-

terforze (SDI) Centro Elabora zio n e Dati (CED), which is a primary source of infor-

mation that police oﬃcers and intelligence agenci es u se t o i d entify potential ta rg et s

for further investigation (https://www.poliziadistato.it/articolo/37262), at

ll four of t h e Italian police forces: state police (Polizia di Stato), ﬁnance po-

lice (Guardia di Finanza), military police (Carab i n ieri), and environmental police

(Guardia Forestale).

The staﬀ of the Police Forces is r eq u i r ed by law (Art. 16 L. 121/81) to send to the

SDI, without delay, any information acq u i r ed through “administrative activi ti es”

or “activit i es of p r evention or repression of crimes .” Hence the SDI must cover

information on every investigation undertaken by the pol i ce forces. For an individ-

ual, the ﬁrst entry in the SDI database for a part i cu l a r allegati o n occurs when a

police force, based on a preliminary investigation, determines that there is suﬃcient

evidence to open a formal i nvestigation. See Figure A.2.

Hence, based on the list of individuals that we communicated to the AISI, i t cre-

ated a sample of suspect oﬀenders, includi n g those individuals that were convicted,

acquitted, or never charged (but nonetheless investigated) for the following crimes:

corruption, malfea sa n ce and embezzlement; abuse of power and undue inﬂ u en ce;

and violations in public auct i on s.

The SDI data also allow us to ﬂag RUPs who are under investigation for corruption

and related charges. By ﬂagging these RUPs we can also det er m i n e which procuring

agencies are suspect (i.e., those employing at least one suspect RUP).