Public Procurement and Rent-Seeking: The Case of Paraguay
and THOMAS FLOCHEL
Toulouse School of Economics (ARQADE, IDEI, and IAST), France
World Bank, USA
Summary. —A model of entrepreneurial choices in an economy with a corrupt public procurement sector is built, providing predictions
along two dimensions. First, corrupt public institutions operate by oﬀering contracts without competition and more corrupt entities
channel larger share of their budget in this way. Second, these ﬁrms enjoy extra returns, so that procurement related activities attract
the best entrepreneurs. A large-scale microeconomic database, including all public procurement operations over a 4-year period in
Paraguay, amounting annually to approximately 6% of the country’s GDP, is then used to corroborate these predictions.
Ó2015 Elsevier Ltd. All rights reserved.
Key words — procurement, corruption, rent-seeking, development
Public procurement of goods and services is one of the main
areas at risk of corruption in developing countries where reg-
ulations and legal enforcement are weak. On top of the static
cost of corruption and fund embezzlement, systematic depar-
tures from competition in the attribution of public markets
are likely to have a devastating impact on economic agents’
incentives and as a result on these countries’ productive
structure. This paper presents the ﬁrst large-scale micro-level
evidence on the channels of rent-seeking and its impact on eco-
nomic development, using a unique database of nearly 50,000
public procurement operations in Paraguay, covering the per-
iod 2004–07. In a nutshell, we show that in Paraguay corrupt
behavior in the allocation of public contracts is a key channel
for rent-seeking. This large-scale network of favoritism, some-
times coined ‘‘la patria contratista”,
has deeply damaging
economic consequences: public institutions buy goods and
services at inﬂated prices, and the set of incentives facing
potential entrepreneurs is biased towards unproductive
To guide the analysis, we model the choice of formal entre-
preneurs with idiosyncratic cost levels, between serving private
consumers competitively, or joining a rent-seeking sector,
where they sell to public institutions. In this rent sector, con-
tracts are attributed by corrupt oﬃcials, who distort allocation
rules in exchange for bribes. Firms willing to do business with
the Government must therefore be proﬁtable enough to cover
their production costs as well as the bribes. We derive from the
model two main sets of predictions that are sustained by the
data, revealing the following story.
First, we establish that in Paraguay the main channel for
corruption in procurement is the systematic use of an ‘‘excep-
tional”purchase mechanism, which bypasses legally required
minimum standards of transparency and competition and is
used much more frequently than what should be expected
from international best practice. Using the whole panel dimen-
sion of the data, we show that this type of corruption is used
more by institutions–ﬁrms pairs that trade repeatedly and in
large volumes. Moreover, we exploit a natural experiment,
linked to an exogenous change in public monitoring following
the widely publicized release midway through our period of
study of an NGO report ﬂagging up exception as a key
channel for corruption, and show that its use decreased
signiﬁcantly for these pairs boasting frequent interactions.
Finally, we also provide evidence that this channel dominates
other more ‘‘traditional”ways to rig procurement contracts,
such as the breaking down of lots in amounts that escape
the obligation for open tenders.
Second, this implies that ﬁrms making more business with
the State, those in the so-called rent sectors, enjoy above nor-
mal rates of return and are the most eﬃcient ones. We provide
evidence of these two aspects, by showing that ﬁrms selling
more to the public sector, as well as those selling more through
the exception channel, have higher proﬁt margins, despite the
fact that they trade mostly in standard goods and should face
competition for the market.
As a result, public intervention in markets distort ﬁrms’
incentives by inducing additional entry in activities with an
important procurement component. To the extent that this
self-selection process pushes some of the best potential entre-
preneurs towards rent sectors, it generates a misallocation of
talents across the economy. Indeed, we document this strong
selection bias by exploiting an original econometric strategy
using ﬁrms’ names.
The paper concludes that in the case of Paraguay, the release
of the 2006 TI report and the subsequent change in exposition
to public scrutiny had an important positive impact on the
overall eﬃciency of the public procurement process. As it
made obvious the involvement of the civil society in monitor-
ing the use of public funds, and spurred an increase in the
interest of the media, it generated a change in the behavior
of the public sector with respect to the use of exceptional
procedures, especially so where the use of exception was
motivated by corrupt deals.
The rest of the paper is structured as follows. Section 2
reviews the main strands of related literature and spells out
*We are grateful to Marı
´a del Pilar Callizo and Oscar Gavila
Transparencia Paraguay for making the procurement data available and
for their support during the stay of the third author in Paraguay. We
thank the editor, three referees, and many seminar participants at
Berkeley, CERDI, Edinburgh, Guanajuato, Namur, Oxford, Paris,
Rome, Stockholm, Toulouse, the World Bank and Yale for useful
comments. Financial support from grant RENTSEP ANR-09-BLAN-
0325 is gratefully acknowledged. Final revision accepted: September 5,
World Development Vol. 77, pp. 395–407, 2016
0305-750X/Ó2015 Elsevier Ltd. All rights reserved.
the contributions of the paper. Section 3describes the Para-
guayan institutional environment and reviews procurement
practices over our period of study. Section 4develops the
model and derives empirical predictions. Section 5presents
the data. Sections 6 and 7 present the results related to the
two main sets of theoretical predictions, and Section 8con-
2. LITERATURE REVIEW
The idea that rent-seeking behavior has important social
and economic costs is a relatively long-standing one in the eco-
nomic and political science literature. Early contributions such
as Tullock (1967), Buchanan (1980), Krueger (1974) and
Bhagwati (1982), were concerned, mostly in a theoretical
framework, with the diﬀerent types of costs associated with
the transfer of rents and the waste generated by agents engag-
ing time and resources in competing for rents, for example
through political lobbying or corruption.
More recently, some papers have provided explanations for
ways in which rent-seeking entails dynamic costs. Baumol
(1990) and Murphy, Shleifer, and Vishny (1991) focus for
example on the resulting dysfunctional allocation of talents.
In this approach, potential investments in physical or human
capital are directed to rent-abundant sectors (such as those
stemming from political favors, corruption or exploitation of
natural resources), while investments in innovative activities,
which have greater growth potential, become relatively less
attractive and are discouraged. As supporting empirical evi-
dence, Murphy et al. (1991) present cross-country growth
regressions augmented with country-level proportions of engi-
neering and law students, where the former are said to corre-
spond to investments in productive activities while the latter
are considered rent-seekers. Baumol’s evidence, on the other
hand, is based on historical accounts from Rome, Ancient
China, and the Middle Ages.
To date, there is still limited micro-evidence on the actual
channels and consequences of rent-seeking in developing econ-
omies. Following Fisman’s (2001) seminal contribution, some
papers have stressed the performance premium of connected
ﬁrms (Hellman, Jones, & Kaufmann, 2003; Fries, Lysenko,
& Polanec, 2003; Slinko, Zhuravskaya, & Yakovlev, 2005).
Other contributions have documented the importance of polit-
ical connections in securing access to key economic inputs,
such as credit (Khwaja & Mian, 2005; Li, Meng, Wang, &
Zhou, 2008), tax advantages or foreign exchange (Hsieh,
Miguel, Ortega, & Rodriguez, 2011), or favorable regulations
(Agrawal & Knoeber, 2001). At a more general level, the large
literature on corruption that developed since the 1990s is also
relevant here, and especially the strand of more recent papers
using microeconomic evidence to directly measure corruption
and its eﬀects on outcomes.
A few contributions have dealt speciﬁcally with public pro-
curement. Hyytinen, Lundberg, and Toivanen (2007), who
study the eﬀects of politics on municipal cleaning contracts
in Sweden, show that the lowest bidder does not win 58% of
the time and that the choice of the winner is subject to political
considerations; Goldman, Rocholl, and So (2013) show that
US companies connected, through the composition of their
boards, to the winning party in both legislative and presiden-
tial elections (in 1994 and 2000) are signiﬁcantly more likely to
have experienced an increase in procurement contracts. Refer-
ences dealing explicitly with corruption include Di Tella and
Schargrodsky (2003), who document the impact of a crack-
down on corruption in Argentinean hospitals, and Bandiera,
Prat, and Valletti (2009), who disentangle the eﬀect of passive
(ineﬃciency) versus active waste (corruption) in Italy, ﬁnding
that the former accounts for about four times the eﬀect of
the latter, and Mironov and Zhuravskaya (2012), who analyze
the link between corrupt procurement and campaign ﬁnancing
in Russia. Relatedly, Davis (2004) and Deiniger and Mpuga
(2004) analyze the impact of corruption on public service
delivery in South Africa and Uganda respectively.
With respect to this literature, our paper provides two main
contributions. First, we have data not only on the expenses
realized by public institutions, but also on the ﬁrms that are
on the selling side. This enables us to capture the eﬀect of
large-scale corrupt practices on the proﬁtability of ﬁrms and
hence on the industrial structure of the economy.
evidence of the distortive eﬀects of rent-seeking in terms of
economic eﬃciency, by showing that it implies an ineﬃcient
specialization of the more able entrepreneurs in imports and
procurement activities. Second, we document one of the most
prevalent channels of corruption in procurement activity,
namely the use of purchase mechanisms circumventing stan-
dard competitive rules, and uncover the economic characteris-
tics of the institutions and sectors more prone to it.
3. RENT-SEEKING AND CORRUPTION IN
Paraguay is considered to be one of the most corrupt coun-
tries in the world.
Our period of study is part of a non-
interrupted 61-year spell, including the 1954–89 Stroessner
dictatorship, in which the Colorado party governed the coun-
try. At the heart of the system was the distribution of public
employment and contracts to its supporters, and the exclusion
of its opposition.
An important channel for corruption,
which we focus on here, was the allocation of public contracts
to ﬁrms that in most cases were created with the sole purpose
of supplying the state, often by selling a wide variety of
imported goods. As a result of the ample anecdotal evidence
of corruption in public procurement, and under pressure from
international organizations, a law regulating public procure-
ment practices (law 2051/03) was enacted in 2003 by the gov-
ernment of the newly elected president Nicanor Duarte
Frutos, with the announced intention of promoting transpar-
ency and eﬃciency in public purchases. The most signiﬁcant of
its provisions were the creation of a public procurement
watchdog (the National Directorate of Public Procurement,
or DNCP), the design of a menu of purchase mechanisms to
regulate procurement procedures, and the compulsion to make
all information (calls, providers, award, etc.) public. This last
proviso was accompanied by the creation of the DNCP web
site where this information is available, but in practice access
is often intermittent and the interface is impractical.
There are strong indications however that improvements in
the regulatory framework did not translate quickly into clea-
ner procurement practices, partly because many oﬃcials did
not comply with the new law and the wrongdoings continued.
The main mechanism through which ﬁrms were favored is the
use of the exceptional purchase mechanism, by which speciﬁc
regulations, such as the obligation to organize public tenders
above certain amounts, were disregarded (see details in Sec-
tion 5below). In 2006, Transparencia Paraguay (TP), the
local chapter of Transparency International, published an
extensive report sponsored by the Inter-American Develop-
ment Bank (IDB) focusing on the excessive use of exceptional
procedures, which was clearly identiﬁed as one of the main
irregularities in the procurement process.
396 WORLD DEVELOPMENT
State-owned enterprise ANDE has for instance been pointed
out for buying large numbers of electric transformers in this
way over the years, despite the fact that these are routinely
required by the ﬁrm for network repairs. Firm oﬃcials recog-
nize that this practice usually generates excess pricing of
between 17% and 27%.
Because the report was given ample coverage in the local
media and through public presentations, and its recommenda-
tions were subsequently relayed by the IDB–World Bank Par-
aguayan public procurement evaluation panel, the oﬃcials in
charge of procurement in public institutions and ﬁrms became
more cautious. Indeed, in 2004 and 2005 purchases made
through the ‘‘exceptional”procedure amounted to nearly
24% of the total procurement spending. In the period
2006–07 the share of purchases made through the ‘‘excep-
tional”procedure decreased to 13%. We exploit this ‘‘natural
experiment”, by showing that the speciﬁc pairs of institutions
and ﬁrms that we identify as irregularly using exception in the
ﬁrst subperiod, subsequently experienced the largest reduction
in its use.
The next section builds a model of corrupt procurement,
from which we derive predictions that we exploit to conduct
an empirical analysis of corruption in public purchases in
4. THE MODEL
The model focuses on the formal sector of the economy as
formal ﬁrms are the only ones allowed to compete for public
markets. The production functions involve constant returns
to scale technologies.
The cost function of a producer oper-
ating in the formal sector is CðqÞ¼cq, where by assumption
Assumption A1. cis independently and uniformly distributed
Entrepreneurs in the formal sector have the choice between
procuring commodities for the public sector, where, as corrup-
tion prevails, they make rents, or doing business in the private
sector where they serve consumers competitively. As a bench-
mark, we ﬁrst brieﬂy discuss the corruption-free equilibrium.
In the absence of rent-seeking opportunities, entrepreneurs
serve market demand competitively and make no rents. Com-
petitive pressure helps to select the best available technology
so that in equilibrium the price equates the lowest marginal
cost, p¼0, and quantities traded in the private sector are
DðpÞ. Welfare is maximized.
Rent-Seeking. We now turn to the case, denoted by the
superscript r, where corruption prevails in public purchase.
While in practice corruption and bribes are not observed,
the model based on assuming the existence of corruption will
generate a number of testable predictions discussed below.
Optimal procurement rules specify that for large purchase
above given thresholds it is mandatory to organize a compet-
itive tender (see Auriol, 2006), and to advertise the calls to
encourage submissions by ﬁrms. In practice it is not always
possible to organize a competitive tender, for instance because
the commodity is patented, or there is an emergency. To deal
with these speciﬁcs cases, public procurement rules include an
‘‘exceptional purchase”procedure. When they rely on this type
of procedure, public purchasers are able to bypass the compe-
tition phase. This is the easiest way to favor a ﬁrm in exchange
of a bribe. Auriol (2006) shows that for large purchases a cor-
rupted procurement oﬃcial favors limited tendering proce-
dures (i.e., exceptional purchase), thereby maximizing the
price of the purchase and his bribe.
We thus expect corrupt
Paraguayan public institutions to rely on the exceptional pur-
chase mechanism to collect bribes over large purchases.
Firms selection in the rent sector. The public purchaser
chooses how many ﬁrms should be included in the rent sector
and creates a lot for each of them out of his/her total budget.
A ﬁrm invited to procure one of these lots through exceptional
purchase is in a monopoly position. It can ask for the highest
possible unit price c. In the rent sector a contract of size q>0
hence costs cq. The government oﬃcials choose b2½0;1, the
share of cq they take as a bribe in exchange for giving this
market to a ﬁrm without competition.
We deduce that, in
the rent sector, the proﬁt of a ﬁrm with cost c2½0;cand
lot size q>0 is:
To access the rent sector the ﬁrm must pay bcq. Since bcq is
independent of the cost of the ﬁrm, the bribe is equivalent to
a ﬁxed cost which screens out the less eﬃcient ﬁrms. Let
crðbÞ2½0;cdenotes the ﬁrm that is just indiﬀerent between
the rent and the private sector: PðcrðbÞÞ ¼ 0. Since the bribe
is a fraction of the market value (i.e., it is linear in cq), the
value crðbÞis independent of q. It is straightforward to check
We deduce that if c<crðbÞthen PðcÞ¼qc
the public sector rents are made and diﬀerent types of ﬁrms
coexist because corruption creates artiﬁcial barriers to entry.
By contrast entrepreneurs with costs higher than crðbÞwould
make a loss, and so prefer to serve private demand. It is intu-
itive that the share of ﬁrms in the rent sector, crðbÞ
decreases with b. The more greedy government representatives
are, the more proﬁtable ﬁrms need to be to do business with
them: they need to be able to cover their cost plus the bribes
and still make non-negative proﬁt.
Optimal bribe rate. By choosing b2½0;1the public pur-
chaser chooses how many ﬁrms enter the rent sector. Then
she attributes to each ﬁrm with a marginal cost lower than
crðbÞa lot. In our database the lots qij >0 vary depending
on the identity of the purchaser (institution j) and of the seller
(ﬁrm i). As shown by Eqn. (2) our results are independent of
the exact size of the lots and on their distribution among the
ﬁrms in the rent sector, so we leave them unspeciﬁed. The only
constraint is that the lots are positive in value and that their
sum is equal to the available budget. To compute the optimal
bribe rate, public oﬃcials internalize the risk of corruption
being detected and punished.
We assume that the greedier
the public purchaser (i.e., the higher bthe bribe rate) the
higher the risk of detection.
That is, the probability of detec-
tion is GðbÞ2½0;1, where GðbÞis a distribution function and
gðbÞ¼G0ðbÞ>0 is the associated density function deﬁned
over ½0;1satisfying the monotone hazard rate property
Assumption A2. 1GðbÞ
gðbÞis decreasing in b.
Consistently with empirical evidence we focus on weak pun-
ishment: in case of detection the bribe is lost to the oﬃcials.
For a purchase of total size Q, the net expected rent of the
public purchaser then writes B¼bcQ 1GðbÞðÞ. To avoid
being excluded from the lucrative bribes business, the public
purchaser needs to split the budget among enough ﬁrms
(i.e., to choose a low enough b). This captures in a simple
way the fact that Paraguay at the time of the study was a
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 397
patronage economy. Indeed, the Colorado party was able to
maintain itself in power for 61 years by sharing among its fol-
lowers the windfall from power through corrupted deals.
We deduce the next result.
Proposition 1. The bribe rate, br2ð0;1Þ, is solution to:
Entrepreneurs choose the rent sector if and only if
Under Assumption A2 it is easy to check that brsolution of
(3) exists and is unique.
Proposition 1 indicates that the most productive entrepre-
neurs choose the rent sector where there is no competition
and commodities are overpriced, so that they make rent
rðbÞc½>0 for all c<ð1brÞc). The model
captures the essence of the redistribution mechanism among
the Paraguayan elite: only the ﬁrms with the lowest cost
are included in the corrupted deals. Firms with high cost
(i.e., those with little physical and social capital, run by
poorly educated and connected managers, etc.) are left to
serve private demand. We deduce that entrepreneurs who
choose to do business with the government are the most
eﬃcient ones and they make rents. Compared to a
corruption-free economy, prices are higher so that the quan-
tities consumed in equilibrium are smaller, leading to lower
In practice procurement activities are decentralized at the
institution level (ministries, state enterprises, etc.), so for the
empirical analysis bshould be thought of as institution-
speciﬁc. Corruption detection varies from one institution
to another. They diﬀer in their level of exposure to public
scrutiny, depending for example on how many people are
harmed by corruption or on how politically sensitive their
activities are. They also diﬀer in their capacity to realize
and hide corrupt acts. A simple way to formalize this in
the context of the model is in terms of hazard rate domi-
nance, which implies stochastic dominance.
easily the next result.
Proposition 2. Let Gð:Þand Kð:Þbe two distributions of
probability of corruption detection such that gðbÞ
1KðbÞ8b2½0;1. Let br
Kbe deﬁned in Eq. (3)with
distributions Gð:Þand Kð:Þrespectively. Then br
Institutions characterized by a lower probability of
detection (i.e., lower hazard rate) will be less cautious to hide
corruption and will ask for more bribes. It implies that the
average lot size, q¼Q
1b, which increases with b, will be larger
for those institutions.
It is intuitive that when the public
oﬃcials are more greedy there are less ﬁrms that are able to
survive in the rent sector and that their lots size is then larger.
At the institution level, we hence expect a correlation between
the frequency of exceptional purchases, which is our marker
for corruption, and the average market shares attributed to
Summary of empirical predictions. Our theoretical results
lead to two main sets of predictions that we take to the data
in the following order, using several complementary empirical
First, according to Proposition 2, institutions character-
ized by a lower probability of detection rely more heavily
on exceptional purchase and have larger lots size attributed
to their providers. In Section 6, we test this prediction in
two steps. In the ﬁrst one, we discuss in details why in
Paraguay the use of exceptional purchase mechanisms
can be considered a marker for corruption. We then
establish that more corrupt institutions do indeed attribute
larger lots to their providers through the exceptional
Second, following Proposition 1, entrepreneurs who
enter the procurement sector are the most eﬃcient ones
and they are making positive rents. We test this prediction
in Section 7.
5. THE DATA
The main data set tracks all the pro-
curement transactions made over the period 2004–07 between
73 public entities (representing over 90% of total Paraguayan
public spending and employment) and 5,517 diﬀerent private
These 47,615 public purchases include all types
of goods and services, from stationary to machinery, oil pur-
chases, food, services, etc. There are good reasons to believe
that no public procurement operations escape registration
as, under the new system, contracts need to be registered
and executed before the corresponding funds are released.
Total public spending over the whole period amounts to Gs.
12,400 bn. (approx. US$ 2,235 m),
between 5.5% and 6.9% of Paraguay’s yearly GDP.
The distribution of contract values has a fat left-hand tail
with 84% of purchases costing less than 2000 minimum daily
wage (mdw), while 5.5% of contracts costing over 10,000
mdw make up 86% of the total spending. The sample mean
is approximately US$ 47,000, equivalent to 36 times the
national per capita GDP at the time.
Each observation in the procurement data set contains the
name and type of the public entity, the name and legal regis-
tration number (RUC) of the supplying ﬁrm and its owner,
and information on the purchase including the nature of the
good or service categorized in 25 diﬀerent groups, the total
cost in local currency, and the purchase mechanism used.
Purchase mechanisms are a key provision of the 2003 public
procurement law, regulating the procedures to be followed in
allocating contracts depending on their total value. There are
ﬁve legal purchase mechanisms with gradually increasing con-
straints on the minimum number of oﬀers, the mode and
length of publication of the call for oﬀers, and the attribution
Finally, these guidelines can be disregarded in cases of emer-
gency, such as natural disasters or health epidemics, for the
purchase of patented and copyrighted goods, or for purchases
requiring defense secrecy. In those extraordinary circum-
stances, public oﬃcials can skip all formal purchase require-
ments through the so-called exceptional purchase
mechanism. In our sample, exceptional purchases are quite
common for certain categories of goods or services, such as
rentals, advertisement, consultancy and transport.
Tax ranking data. We use annual rankings of top taxpayers
published on the Ministry of Finance’s web site. Firms’ ranks
are determined by their total payments on all taxes.
public ﬁrms are excluded, we have information for 748 ﬁrms
in 2004, 459 ﬁrms in 2005, 482 ﬁrms in 2006, and 478 ﬁrms
398 WORLD DEVELOPMENT
Import–export data. We also include annual rankings from
the Customs’ SOFIA oﬃcial data base. These include the full
universe of importers from 2004 to 2007, including the total
free on board (FOB) value imported, and of exporters for
the same period, including the cost, insurance, freight (CIF)
Institution-level corruption indices. We introduce institution-
level corruption indices for a subset of 13 institutions in our
sample. In total, this covers 15,640 of our initial observations,
equivalent to 32.8% of the total. These indices were developed
by the NGO Transparencia Paraguay during 2004–08.
6. CORRUPTION AND THE STRUCTURE OF
It is useful to discuss the claim that the use of exceptional
purchases is the main channel for corruption in the context
we study. Three aspects are relevant here. First of all, as men-
tioned in Section 3, there is ample anecdotal evidence support-
ing this claim, in particular the report by Transparencia
Paraguay (2005), which has been widely publicized. Second,
the use of such procedure in the Paraguayan context vastly
exceeds comparable ﬁgures from around the world. For
instance, in a sample of Brazilian health procurement con-
tracts between 2004 and 2009, exceptions amount to approxi-
Similarly, from 2006 to 2010, only 7% of EU
procedures were ‘‘negotiated without publication”, corre-
sponding to 5% of the value share of total procurement con-
tracts above the EU threshold for mandatory call for
By means of comparison, the equivalent ﬁgure in
Paraguay in the two-year period before the publication of
Transparencia Paraguay (2006), looking at contracts above
the 2,000-mdw threshold for mandatory call for tender, is
20.4% of all awards. In terms of share of total contract value,
it climbs to 23.8%.
Finally, it is worth noting that estimations not shown here
to save space provide evidence that none of our results go
through when using as a marker of corruption an other com-
monly mentioned channel for corruption in procurement,
namely the practice of breaking down contracts in smaller lots
so that they can be attributed without competition. This is true
using as dependent variable an indicator of whether the con-
tract lies in a 10% or 20% value band just below the threshold
that implies mandatory open tender. This shows that in the
context of generalized corruption prevailing in Paraguay, pub-
lic purchasers were careless and used the simplest procedure
available to favor ﬁrms.
Going to the estimations, note that our unit of observation
is the individual purchase. Corruption being in most cases not
observed, our identiﬁcation strategy, following the model,
relies on documenting the link between the frequency of
non-competitive procedures (i.e., exceptions), the intensity of
ﬁrm–institution relationships (i.e., the lot size in the model),
as well as institution-level detection proneness (i.e., the detec-
tion functions). Each of the 47,615 purchases available corre-
sponds to a pair composed of a ﬁrm iand an institution j. Our
sample of 73 institutions and 5,517 ﬁrms generates a total of
13,693 diﬀerent ‘‘active”pairs, with an average number of con-
tracts equal to 3.5 (std. dev. 10.5), a minimum of 1 (for 7,215
pairs) and a maximum of 460.
We estimate the following basic model:
excijkt ¼hjþhkþhtþXijtb1þeijkt ;ð5Þ
where exc is a binary variable equal to 1 if the contract over
good kprocured from ﬁrm iby institution jin year tis made
through the exception, and h0sare institution (j), good (k), and
year (t) ﬁxed eﬀects. We use a linear probability model to esti-
mate the model above.
The ﬁxed eﬀects allows us to capture any systematic deter-
minants of exceptional purchase that would correspond to
characteristics of the goods (patented or monopolistic goods,
exclusive dealing) and the institutions (speciﬁcally dedicated
to attend emergencies, involved in defense deals, etc., possibly
with changes over time), as well as speciﬁc time ﬂuctuations or
trend in the use of exceptions. Once these ﬁxed eﬀects are
introduced, we expect no additional features to be signiﬁcant
if procurement rules are applied correctly.
Our explanatory variables of interest Xijt are ﬁrm–
institution-level ones: the total value of each pair’s transac-
tions and the proportion of an institution’s transactions done
with each particular provider, year by year. These are two
measures of the lot size described in the model. Our model sig-
nals as a symptom of corruption the fact that institutions,
which do more exception for that purpose do that through
the attribution of larger lots to their providers (i.e., a more
corrupt public purchaser asks for larger bribes which reduces
the number of ﬁrms able to pay them, and increases their aver-
age lot size see Proposition 2).
Table 1 contains the results from (5). In column 1, which
report the estimation with institution, goods and year ﬁxed
eﬀects, the larger the share of an institution’s transactions
done with each particular provider, the higher the probability
that the contract is made through exception, an eﬀect signiﬁ-
cant at the 5% level. This is robust to adding both
institution-year and goods-year ﬁxed eﬀects in column 2.
The marginal eﬀect implies that an institution increasing the
share of its total procurement volume allocated to a particular
ﬁrm from 0.9% to 3.8% of its portfolio (that is by one standard
deviation above the sample mean), would increase the share of
its contracts with that particular ﬁrm made through the excep-
tion by 36%.
In Column 3, where we add a full set of ﬁrm
ﬁxed-eﬀect, the signiﬁcance is lost, which is not very surprising
considering the number of additional parameters (over 5,000).
Proposition 2 suggests that institutions more exposed to
public scrutiny use less exceptional purchase. We use the
institution-level corruption indices described in the data sec-
tion to proxy for this exposure. We measure corruption with
a synthetic index equal to the arithmetic mean of three original
indices, namely the evaluations based on the Comptroller
General’s report, the number of administrative indictments
in any given institution, and the number of newspaper articles
mentioning corruption in each institution.
We introduce the following speciﬁcation, where we expect b2
to be positive if the corruption story is relevant
excijkt ¼hjþhkþhtþhjt þXijtb1þXijt Inst Corr jt
The results are in columns 4–6 of Table 1. In column 4, con-
tract value becomes negative, while its interaction with corrup-
tion is positive and signiﬁcant at the 1% level. This conﬁrms
that the link between frequent interactions and exceptional
purchases is mediated by corruption. This eﬀect remains signif-
icant with institution- and goods-speciﬁc time trends in column
5 and, more strikingly, with ﬁrm ﬁxed eﬀects, in column 6.
This ﬁrst set of result support the corruption hypothesis and
gives us more conﬁdence that corruption is the relevant expla-
nation. Of course, despite being more satisfactory than usual
subjective indicators, we cannot completely rule out endogene-
ity of the corruption measure used here, and ﬁnding appropri-
ate instruments appears diﬃcult.
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 399
Alternative interpretations for the results that frequent pair
interactions lead to more contracts through the exception
include ﬁrst a ‘‘reputation”eﬀect. In circumstances where pub-
lic institutions need to use exceptional mechanisms, for exam-
ple because of some social emergency, they may turn to ﬁrms
they have had frequent interactions with, because they know
these are more reliable. Another explanation involves simple
ineﬃciency or passive waste, as Bandiera et al. (2009) docu-
ment in the case of public procurement in Italy. Here, the
argument is that procurement oﬃcials simply award contracts
to ﬁrms they already know, because they do not internalize the
new rules or because they are lazy and it is the solution that
requires less eﬀort.
To sort out the diﬀerent explanations, our strategy is to
exploit the shock created by the publication of the Transpar-
ency Paraguay report at the beginning of 2006, which
attracted attention to what was considered the excessive use
of exception in public contracting. This prompted debates
and comments in the media and the political arena, as well
as incentives for investigative journalism in the country to
inquire speciﬁcally on procurement done through the excep-
tional procedure. Our hypothesis is that if exception is indeed
a mean to make corrupt deals, the institution–ﬁrm pairs that
were using exception frequently should display greater reduc-
tions in its use as a result of this shock. In other words, we
hypothesize that where the use of exception was motivated
by corrupt deals, the higher risk of being discovered and sanc-
tioned brought about by increased public scrutiny should have
generated a stronger decrease in its use than where it was moti-
vated by justiﬁable reasons.
The resulting diﬀerence-in-diﬀerences strategy thus should
display a positive coeﬃcient based on this diﬀerential eﬀect,
even in the case where the report implied an overall reduction
in exceptional purchase. From a theory point of view, this cor-
responds to the monitoring shock shifting the detection func-
tion for a subset of ﬁrm–institution pairs, as described in
Proposition 2 above. Alternatively, if the explanation for high
levels of exception is simply ineﬃciency, or if the decrease
results from a learning process, we should observe a reduction
in its use across the board, and the diﬀ-in-diﬀ estimator should
not be signiﬁcant.
Table 2, in which we run the basic estimations of Table 1,
columns 1–3, over the separate subperiods 2004–05 and
2006–07, provides a ﬁrst illustration. Over the period 2004–
05, the variables proxying for frequent/intense ﬁrm–institution
relationship, i.e., the total value of pair’s transactions and the
share of institutions’ transactions done with each particular
provider, are signiﬁcant ﬁve times out of six, including in spec-
iﬁcations with ﬁrm ﬁxed eﬀects, in column 3. On the contrary,
they are almost never signiﬁcant in the period 2006–07, and
even become negative with ﬁrm ﬁxed eﬀects in column 6.
In the ﬁrst sub-period, a one standard deviation in value of
pair’s transactions reduces the average use of exception by
between 3% and 4%, while the same eﬀect is more than an
order of magnitude smaller in the second sub-period. The
magnitude of the eﬀect is similarly smaller when looking at
In Table 3, we present the results from a formal diﬀerences-
in-diﬀerences based on these insights. We estimate the follow-
ing regression, where b3is the diﬀ-in-diﬀ estimator:
excijk ¼b0þb1D0607 þb2DTþb3ðD0607 DTÞþeijk >0;ð7Þ
For the sake of robustness, we consider three alternative ways
to construct our treatment group of institution–ﬁrm pairs:
based on high-volume, high share of institutions’ budget,
and high use of exception in the 2004–05 period (see Online
Appendix for details).
The results in Table 3 are striking. In column 1, the treatment
group includes institution–ﬁrm pairs with total transactions
value above the median in the ﬁrst period; the diﬀ-in-diﬀ
estimator is negative and signiﬁcant at the 1% value, and indi-
cates that the treatment group reduces its average contracting
by exception by close to 4%. The result is more telling when
translated in terms of amounts: for the treatment group, the
value of transactions made by exception goes from representing
23.2% of the total in the ﬁrst period to only 12.2% in the second
period. This 9 percentage points (pp) reduction should be com-
pared with that of the control group which is only 4.3 pp (from
16% to 11.7% of the total).
A similarly strong result emerges in column 2 when using
instead the share of institutions’ budget as our treatment crite-
ria. There is a decrease between the two periods, signiﬁcant at
the 1% level. The size of the eﬀect is similar, as the treatment
group experiences a reduction in the value share of exception
of 10.2 pp, while the corresponding control group reduction is
of 3.4 pp.
Table 1. Exceptional purchase determinants and institution-level corruption
(1) (2) (3) (4) (5) (6)
Total value of pair ij 0.002 0.002 0.003 0.123 0.001 0.030
contracts (0.002) (0.002) (0.002) (0.056)
Share of institution j0.335 0.360 0.032 1.349 0.690 0.099
contracts with ﬁrm i(0.137)
Total value of pair ij 0.049 0.024 0.026
contracts x corruption (0.012)
Share of institution j0.272 0.145 0.059
contracts with ﬁrm icorruption (0.155) (0.170) (0.067)
Instit. F.E. Yes Yes Yes Yes Yes Yes
Goods F.E. Yes Yes Yes Yes Yes Yes
Years F.E. Yes Yes Yes Yes Yes Yes
Instit. years F.E. Yes Yes Yes Yes
Goods years F.E. Yes Yes Yes Yes
Firms F.E. Yes Yes
Observations 47,615 47,615 47,615 15,489 15,489 15,489
R-Squared 0.09 0.12 0.53 0.13 0.17 0.60
Note: In columns 1–6, the dependent variable is an dummy equal to 1 if a contract was made by exception, and 0 otherwise. Corruption is an institution-
level corruption index. Robust standard errors in parentheses, clustered at the institution level. Stars indicate signiﬁcance at the 10% (*), 5% (**), and 1%
400 WORLD DEVELOPMENT
In column 3, we experiment with yet another deﬁnition of
the treatment group, based on the use of exception in the base
period. The result again strongly support our hypothesis: the
diﬀ-in-diﬀ estimator is negative and signiﬁcant at the 1% level,
with a probability reduction of 14%. In terms of amounts, the
treatment group value share of exception diminishes from
50.9% to 7.3%, while for the control group it experiences an
increase from 1.4% to 19.9%.
Columns 4 presents the case where the treatment is deﬁned
on the basis of the share of exception in a pair’s transaction
being above the average. Again, the diﬀ-in-diﬀ coeﬃcient is
negative and signiﬁcant at the 1% level, indicating a probabil-
ity decrease of 21%. The changes in terms of amounts are very
close to those of column 3. Finally, column 5 provides an addi-
tional robustness check, by restricting the base sample to pairs
that have strictly positive exception use in the period 2004–05,
with again very similar results: a probability decrease of 15%,
signiﬁcant at the 1% level.
Finally, in the last line of Table 3, we also report the results
from similar speciﬁcations with a continuous version of the
treatment variables. The complete set of results are on line
with those of the discrete version.
Additional technical details and robustness tests, including
the way treatment groups are constructed, and a comparison
of pre-treatment trends between treatment and control groups,
supporting the idea that the report’s publication can be viewed
as a natural experiment, are also provided in Online Appen-
In the next section, we turn to show how the corrupt prac-
tices documented above distort the proﬁtability of ﬁrms.
Table 2. Exceptional purchase determinants – sub periods
(1) (2) (3) (4) (5) (6)
2004–05 2004–05 2004–05 2006–07 2006–07 2006–07
Total value of pair ij 0.037 0.035 0.032 0.001 0.001 0.002
(0.001) (0.001) (0.002)
Share of institution j0.339 0.297 0.085 0.209 0.259 0.097
contracts with ﬁrm i(0.178)
(0.146) (0.129) (0.139)
Instit. F.E. Yes Yes Yes Yes Yes Yes
Goods F.E. Yes Yes Yes Yes Yes Yes
Years F.E. Yes Yes Yes Yes Yes Yes
Instit. years F.E. Yes Yes Yes Yes
Goods years F.E. Yes Yes Yes Yes
Firms F.E. Yes Yes
Observations 22,179 22,179 22,179 25,436 25,436 25,436
R-Squared 0.09 0.11 0.51 0.11 0.14 0.62
Note: In columns 1–6, the dependent variable is an dummy equal to 1 if a contract was made by exception, and 0 otherwise. See Table 2 notes for variables
deﬁnitions. Robust standard errors in parentheses, clustered at the institution level. Stars indicate signiﬁcance at the 10% (*), 5% (**), and 1% (***)
Table 3. Exceptional purchase – diﬀerences-in-diﬀerences
Pair-level treatment criteria (1) (2) (3) (4) (5)
Share of transactions
in institution budget
Share of transactions
done by exception
Share of transactions
done by exception
(only > 0 obs.)
0.012 0.003 0.000 0.023 0.086
(0.011) (0.015) (0.012) (0.015) (0.010)
0.022 0.034 0.216 0.384 0.332
0.037 0.030 0.138 0.211 0.151
Observations 37,453 37,453 37,453 37,453 13,709
R-Squared 0.46 0.46 0.47 0.52 0.52
0.003 0.007 0.080 0.092
(0.018) (0.010) (0.049) (0.046) (0.057)
Continuous 0.187 0.235
treatment (0.365) (0.132) (0.099) (0.047) (0.050)
Note: In columns 1–6, the dependent variable is an dummy equal to 1 if a contract was made by exception, and 0 otherwise. D
is a dummy variable
taking value 1 for the period 2006–07. D
is the treatment group dummy; for each column, the line pair-level treatment criteria indicates the way D
deﬁned at the institution–ﬁrm pair level. All estimations include ﬁrms, institutions, goods, institutions * years, and goods * years ﬁxed eﬀects, as well as a
control for the amount contracted. The last two lines show the diﬀ-in-diﬀ coeﬃcients and standard errors from similar estimations on the 2004–05 sample
only, where D
is a dummy variable taking value 1 for the year 2005.
Robust standard errors in parentheses, clustered at the institution level. Stars indicate signiﬁcance at the 10% (*), 5% (**), and 1% (***) respectively.
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 401
7. THE PROFITABILITY OF FIRMS
The model’s second set of predictions is that, as a result of the
corrupt practices unveiled above, entrepreneurs doing business
with public institutions are more proﬁtable than their counter-
parts serving private demand. Moreover, the most able entre-
preneurs are expected to self-select into the more proﬁtable
procurement activities, as only they are eﬃcient enough to
aﬀord both the production cost and the bribes to public oﬃcials.
Proﬁtability in the rent sector. We ﬁrst perform a reduced
form analysis of the eﬀect on ﬁrms’ proﬁts of a number of vari-
ables, derived from the results in the previous section. As a
proxy for the share of ‘‘favored”contracts in the ﬁrm’s port-
folio, we use the share of a ﬁrm’s contracts made through
the exception, and the weighted average level of corruption
of the institutions it deals with (where the weights are the share
of the sales to these institutions in the ﬁrm’s total sales). In
addition, we also use ﬁrms’ amount and number of contracts.
The amount of taxes paid provides a reasonable approxima-
tion for proﬁts because the tax rate on gains is ﬂat and uni-
form in each period (30% in 2004, 20% in 2005, 10% in 2006
and 2007). While the inclusion of other taxes (among which
custom duties are by far the largest component) introduces
some noise in the mapping between proﬁts and taxes paid,
we control for total imports in all estimations to minimize this
issue. The model we want to estimate is:
Git ¼aþb1Zit þb2Mit þXitb3þhtþeit ;ð8Þ
where Git denotes the net gains of ﬁrm iin year t,Zit is the var-
iable of interest (alternatively, the share of sales through the
exception, average corruption of buyers, total sales to the
state, number of contracts), Mit is the total amount imported,
Xit is a vector of control variables, and htare time ﬁxed eﬀects.
However, the income tax and other taxes are amalgamated
in the tax data, so we only observe:
Tit ¼xtGit þdiMit þmit;ð9Þ
where xt¼0:3 for 2004, xt¼0:2 for 2005, and xt¼0:1 for
2006 and 2007. In order to obtain the ﬁrms’ net gains we
therefore divide the total amount paid in taxes by the corre-
sponding tax rates.
We thus test the following speciﬁcation:
Tit=xt¼aþb1Zit þðb2þdi=xtÞMit þXitb3þhtþeit þmit
under the assumption that Zit is uncorrelated with mit .
The distribution of proﬁts resulting from the available data
is truncated at a strictly positive point. Moreover, the set of
ﬁrms for which we have non-zero tax data is not constant over
time. This forces us to restrict the panel to the subset of strictly
positive tax observations.
As a result, we obtain an unbal-
anced panel of 2,167 observations across 4 years for 1,017 pri-
One worry is that unobserved ﬁrm characteristics might be
correlated both with the amount of taxes paid and with some
of the Zit variables on the right hand side.
For example, more
eﬃcient entrepreneurs might be more successful in general,
hence pay more taxes, and also win more procurement con-
tracts or be more frequently favored through exception because
of their good reputation. Another concern is related to ﬁrm
size. Indeed, bigger ﬁrms may have larger overall proﬁts and
also be in a better position to win procurement contracts or
to respond to emergency calls from public institutions. To
address such issues, we add ﬁrm-level ﬁxed eﬀects hito (10),
exploiting the panel dimension of the data to wash out any time
invariant ﬁrm-level unobserved characteristics.
The results in Table 4 support our hypotheses. Column 1
shows that ﬁrm’s proﬁts are signiﬁcantly increasing in the
share of its contracts made by exceptional purchase. The aver-
age marginal eﬀect implies that a 1% increase in the share of
contracts made by exception corresponds to Gs. 28 millions
(US$ 5,600) additional proﬁts.
In column 2, the correlation between the average level of
corruption of public buyers and ﬁrms’ proﬁtability is positive
but only nearly signiﬁcant at conventional levels, which is not
surprising given that the sample size is reduced to 261 since
corruption indices are not available for all institutions.
Finally, in columns 3 and 4, we look directly at the correla-
tion between ﬁrms’ proﬁts and their procurement activity. The
coeﬃcients of both the amounts sold and the number of
contracts are positive and signiﬁcant. In terms of marginal
Table 4. Procurement and proﬁtability of ﬁrms
(1) (2) (3) (4)
Random eﬀects Fixed eﬀects Fixed eﬀects Fixed eﬀects
Imports 0.001 0.003 0.000 0.000
Exports 0.000 0.024 0.001 0.001
Exceptional purchase 2.834
Corruption index 1.205
Amount sold 0.293
Number of contracts 0.154
Time F.E. Yes Yes Yes Yes
Firms F.E. No Yes Yes Yes
Observations 476 261 2167 2167
R-Squared 0.46 0.66 0.25 0.25
Hausman chi2 0.40 0.00 0.00 0.00
Standard errors in parentheses. Stars indicate signiﬁcance at the 10% (*), 5% (**) and 1% (***) respectively.
Note: In all columns the dependent variable is ﬁrm-level gains, deﬁned as in Eqn. (10). All data adjusted for yearly price variations. In each case, we test
the appropriateness of the random versus the ﬁxed eﬀect model, using the standard Hausman test. We report only the speciﬁcation supported by the test.
402 WORLD DEVELOPMENT
eﬀects, every additional Gs. sold to the state translates into a
Gs. 0.29 increase in proﬁts, i.e., a rate of return on procure-
ment operations of nearly 29%, while a ﬁrm obtaining an addi-
tional contract increases its proﬁts by Gs. 154 millions
(approx. US$ 30,800).
These results, together with those of the previous Sec-
tion showing that ﬁrms with bigger procurement portfolios
are more likely to enter in side deals, imply that average
proﬁtability should be higher in procurement than in private
In turn, this is likely to distort ﬁrms’ incentives
and induce additional entry of potential entrepreneurs into
these sectors. Next, we provide evidence of this self-selection
Misallocation of talents. An important point of the model is
that ﬁrms’ unobserved attributes (entrepreneurial or network-
ing skills, eﬃciency, etc.) should explain part of their increased
proﬁtability due to a self-selection process. Some of the best
entrepreneurs are attracted to sectors where they can beneﬁt
from the corrupt allocation of procurement contracts,
resulting in a misallocation of talents in the economy.
The following test explicitly addresses the process of
self-selection into the procurement sector, using a procedure
proposed by Wooldridge (2002, p. 631).
estimating ﬁrst a probit model to explain the fact that ﬁrms
intervene in the procurement sector or not:
where Yiis a dummy variable equal to 1 if the ﬁrm sells to
public institutions at any point during the sample period, Xi
is a vector of ﬁrm-level observables, and Siis a set of instru-
ments. From (11), we derive b
/, the predicted density and b
the corresponding predicted cumulative density. We then
estimate, for each year, the following tobit model:
Git ¼max 0;aþb1Zit þXitb2þb3Yib
Remember that Git denotes the net gains of ﬁrm iin year t,Zit
is either total ﬁrm’s sales to the state or its total number of
contracts, and Xit is a vector of control variables. We are inter-
ested in the statistical signiﬁcance of the two last regressors, as
an indication of self-selection, as well as in how their inclusion
will aﬀect the coeﬃcient b1.
Table 5. Self-selection into procurement and ﬁrms’ proﬁtability
(1) (2) (3) (4) (5) (6) (7) (8)
Tobit Tobit Tobit Tobit Tobit Tobit Tobit Tobit
Gains 2004 Gains 2004 Gains 2005 Gains 2005 Gains 2006 Gains 2006 Gains 2007 Gains 2007
Volume of contracts 0.009 0.008 0.041 0.041 0.129 0.105 0.075 0.068
(0.006) (0.019) (0.016)
Import dummy 3.701 4.101 10.653 11.527 27.689 29.632 30.800 33.454
Import volume 0.000 0.000 0.002 0.001 0.001 0.001 0.001 0.001
Export dummy 1.387 1.440 4.173 4.658 25.662 29.741 17.759 17.168
Export volume 0.000 0.000 0.001 0.000 0.001 0.001 0.001 0.001
(0.000) (0.000) (0.000)
(0.000) (0.001) (0.001) (0.000)
Mills1 1.035 1.552 10.834 10.103
Mills2 3.816 6.741 24.041 25.150
0.185 0.232 0.275 0.289 0.133 0.162 0.156 0.178
Observations 12,759 12,759 12,759 12,759 12,759 12,759 12,759 12,759
Number of contracts 0.010 0.007 0.032 0.026 0.270 0.187 0.256 0.150
(0.008) (0.009) (0.016)
Import dummy 3.696 4.101 10.625 11.511 27.721 29.748 30.746 33.512
Import volume 0.000 0.000 0.002 0.001 0.001 0.001 0.001 0.001
Export dummy 1.381 1.440 4.148 4.653 25.895 29.942 17.438 17.196
Export volume 0.000 0.000 0.001 0.000 0.001 0.001 0.001 0.001
(0.000) (0.000) (0.000)
(0.000) (0.001) (0.001) (0.000)
Mills1 1.020 1.501 10.328 9.694
Mills2 3.817 6.741 24.171 25.232
0.186 0.232 0.275 0.289 0.135 0.162 0.157 0.178
Observations 12,759 12,759 12,759 12,759 12,759 12,759 12,759 12,759
Note: Tobit speciﬁcations with left truncation at the lowest observed proﬁt level in each year. For each speciﬁcation, explanatory variables correspond to
the relevant year. Robust standard errors in parentheses (bootstrapped with 500 replications when Mills ratios are included). Stars indicate signiﬁcance at
the 10% (*), 5% (**), and 1% (***) respectively.
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 403
The crucial point is the availability of suitable instruments,
that would predict access to the procurement sector, while
being excludable from the second Eqn. (12). To generate these,
we exploit the fact that apart from raising the cost of procure-
ment and changing the identity of sellers, corruption also dis-
torts the sectorial abundance of ﬁrms. We capture this bias by
exploiting ﬁrms’ names, which are speciﬁc to the procurement
categories where a large number of ﬁrms are active.
The probit model shows that our names variables are very
strong predictors of ﬁrms being active in procurement, while
there is no reason to think that names inﬂuence ﬁrms’ proﬁt-
ability directly, supporting the excludability requirements.
Table 5 shows the results from estimating (12) on a sample
of 12,759 ﬁrms. For each year, we ﬁrst display the results from
a standard tobit estimation and then provide the results
including self-selection correction terms, with bootstrapped
standard errors. Panel 1 uses the total volume of procurement
contracts as our variable of interest Zit, while panel 2 uses the
total number of contracts.
The correction terms are strongly (jointly) signiﬁcant (at the
1% level) in all estimations. Moreover, their inclusion system-
atically induces a reduction in the estimated coeﬃcients of the
variables of interest. The marginal eﬀect of ﬁrms’ contract vol-
ume on their proﬁtability is reduced by between 9 and 19%
(except in 2005, when it remains constant), and loses signiﬁ-
cance in the last three years. Similarly, the marginal eﬀect of
the number of contracts is reduced by between 19% and
42%, and becomes insigniﬁcant in the 2005 sample. There is
thus a strong composition eﬀect, meaning that the proﬁtability
advantage of better entrepreneurs stems from an unobserved
We thus conclude that part of the link between procurement
and ﬁrms’ proﬁtability relates to unobserved self-selection of
the best entrepreneurs into activities that oﬀer privileged
access to the procurement sector. This provides the ﬁnal ele-
ment of our story, in which would-be entrepreneurs are likely
to be disproportionately attracted to sectors in which strong
demand from corrupt public buyers generate opportunities
We have illustrated the fact that rent-seeking is costly to
development, by showing how entrepreneurs’ economic incen-
tives are distorted toward less productive activities as the
result of favoritism in the allocation of public contracts in
Paraguay. After building an entrepreneurial choice model,
we have used a large-scale microeconomic database including
all public procurement operations over a 4-year period to test
the predictions of the model. Firms have a greater probability
of obtaining a contract directly through an exceptional proce-
dure from an institution with which they have a strong con-
tractual relation, both in terms of the total value and
frequency of transactions, particularly when dealing with
more corrupt State entities. This is supported by the evidence
from a natural experiment, which exogenously increased pub-
lic scrutiny over procurement practices, and especially excep-
tion, halfway through our period of study. Firms trading
more with the public sector are more proﬁtable, even when
controlling for their unobserved characteristics. This overall
picture embodies the consequences of a systematic misalloca-
tion of talents a
`la Murphy et al. (1991).
While the results must be qualiﬁed because of the intrinsic
limitations of the data, in particular those related to corrup-
tion and to the proﬁtability of ﬁrms, we think that the paper
points to two main conclusions. First, rent-seeking is costly
because it destroys the development potential of the best entre-
preneurs. Indeed, the Paraguayan entrepreneurial class is in its
large majority imports-oriented, with over 90% of the top 500
taxpayers being importers. Over the period 1996–2005, the
commercial balance displayed an average deﬁcit of 8.5% of
GDP. Large rents linked to the resale of imported goods to
the State and the historical absence of an import-
substitution strategy have contributed to make Paraguay one
of the least industrialized economy in South America as, apart
from the soybean and meat sectors, its entrepreneurs have sys-
tematically specialized in commercial intermediation, often
with the public sector, rather than in production.
The costs of this productive atrophy and biased specializa-
tion are reﬂected in the poor record of economic growth. After
a period of signiﬁcant growth in the 1970s and early 1980s,
linked to the massive construction projects including the
hydroelectric dams, the average rate of growth of per capita
income was only 0.8% over the 1980s and strictly negative
after that (0.1% and 0.6% over the 1990s and 2000s). Over
the last two decades, the Paraguayan Central Bank indicates
that 92% of growth ﬂuctuations were due directly to ﬂuctua-
tion in agricultural production and exports. As a result, per
capita income was lower in real terms in 2005 than it was at
the beginning of the 1980s.
Second, the release on an NGO report on the abuse of
exceptional procedures, appears to have had a signiﬁcant eﬀect
as it translated in an improvement in the following period. In
that sense, civil society involvement in monitoring the public
sector use of funds appears as a crucial check on corruption.
1. The ‘‘contracting homeland”, see for example Alfredo Boccia Paz,
Diario Ultima Hora, Asuncion, March 4, 2009.
2. Authoritative surveys on corruption include Bardhan (1997), Rose-
Ackerman (1999), Svensson (2005), Pande (2008), and Olken and Pande
(2012) among others. There is a large macro literature, staring with Mauro
(1995), while micro-econometric papers include Reinikka and Svensson
(2004), Olken (2007), Bertrand, Djankov, Hanna, and Mullainathan
(2007), Ferraz and Finan (2008), and Sequeira (2014) to mention only a
3. Related papers are Rama (1993), who tracks the number of foreign-
trade rent-seeking regulations over the XXth century in Uruguay and
relates these to political and economic variables, and Fisman and Sarria-
Allende (2010), who present cross-country, industry-level evidence of the
eﬀect of regulatory distortions on the industrial structure.
4. It has lingered in the bottom 4% of surveyed countries included in
Transparency International’s Corruption Perception Index since its
inclusion in 2002. It had for instance a score of 2.1 in 2005, placing it
144th out of 158, and the same score in 2009 (154th out of 180). See Online
Appendix for general information on the country.
5. See Straub (2014) for evidence on the change in the relevance of
political connections for public procurement around the 2008 elections, in
which the Colorado party lost power.
404 WORLD DEVELOPMENT
6. For example during the period 2004–05 public ﬁrms awarded close to
90% of their advertisement contracts through exceptions. As for speciﬁc
institution, the Oﬃce of the First Lady spent respectively 40% and 93% of its
budget in these two years using the exceptional mechanism. Some cases have
made headlines, such as the use of this procedure to pay close to US$ 100,000
to a consulting ﬁrm formerly owned by the President, for the organization of
the XIIIth conference gathering Americas’ First Ladies in 2005 in Asuncio
(Diario Ultima Hora, Asuncio
´n, June 7, 2007). See Mironov and
Zhuravskaya (2012) for similar stories in the Russian context.
7. Diario ABC Color, Asuncio
´n, January 3, 2010. These ﬁgures are
consistent with the estimation by Auriol (2006) and with the results in
8. This assumption is consistent with existing evidence on manufacturing
and service ﬁrms in developing countries, whether they belong to the
formal or the informal sector (see Tybout, 2000). It is also consistent with
the nature of activities included in our procurement database.
9. Empirically Chong, Klien, and Saussier (2013) ﬁnd a positive
relationship between the use of negotiated procedures without prior
notiﬁcation and the weakness of governance across the European Union:
in countries more prone to corruption, public purchasers use more often
exceptional purchase. This is consistent with earlier ﬁndings by Della
Porta and Rose-Ackerman (2002) who show that in the 1990s in Italy
public authorities were abusing emergency procurement procedures to
10. For instance Tran (2011), exploiting an Asian trading ﬁrm’s records
of the bribes it paid over the year to secure public contracts, shows that the
average kickback was 14.7% of the product cost when auctions were not
required. Ufere, Perelli, Boland, and Carlsson (2012) provides insights
about the supply of bribes by ﬁrms.
11. This is a common assumption in the corruption literature, going back
to the Becker and Stigler, 1974 crime-deterrence model. See for example
Besley and McLaren (1993) and Mookherjee and Png (1995).Di Tella and
Schargrodsky (2003) is an empirical application. Wade (1985) is an early
reference on the behavior of corrupt government oﬃcials, and
Elbahnasawy (2014) looks at the impact of e-government in deterring
12. Equivalently the probability of detection increases with the share of
ﬁrms that are excluded from the rent sector, which by virtue of Eqn. (2)is
equal to ccrðbÞ
c¼b. When the number of ﬁrms that are excluded from the
rent sector bis large it increases the probability of outcry and detection
13. The monotone hazard rate property is equivalent to the log-concavity
of the reliability function 1-G(b). It turns out that most reliability
functions of standard random variable are log concave. This is true for
distributions such as uniform, normal, logistic, extreme-value, Chi-Square,
Chi, exponential, Laplace, Weibull, power function, gamma, beta, Pareto,
log-normal, Student’s t, Cauchy and F distributions (Bagnoli &
Bergstrom, 2005; Borzadaran & Borzadaran, 2011).
14. In Paraguay at the time of our study, there had been very few cases of
prosecution or indictment for corruption leading to jail time or ﬁnes.
15. Indeed under Assumption A2, 1GðbÞ
gðbÞdecreases in b, while bincreases
in ½0;1. Moreover 1
gð0Þ>0 and 1Gð1Þ
gð1Þ¼0<1 so that these two functions
cross once and only once in ð0;1Þ.
16. In the public segment, corruption implies that unit price is c. In the
private segment price is pr¼cð1brÞas the less eﬃcient ﬁrms are left to
serve private consumers. Since pr>p, quantities exchanged in the private
sector fall so that the formal productive sector shrinks.
17. Let Kð:Þand Gð:Þbe two probability functions so that gðbÞ
8b2½0;1, then it implies that GðbÞ6KðbÞ8b2½0;1(e.g., see Nanda &
18. For instance if all procurement contracts are of similar size, we have
cdc, which by virtue of Eqn. (2) is equivalent to Q¼qð1bÞ.
19. See Online Appendix for more details on the diﬀerent types of data.
20. The data we use were initially painstakingly compiled by Transpar-
encia Paraguay (TP), the national chapter of the international NGO
Transparency International, using the information published on the
DNCP web site.
21. The Guarani–US$ exchange rate over the period ﬂuctuated in the
range 5,021–6,178 Gs. for 1$.
22. As a way to index it to the general evolution of prices, contract values
and thresholds are expressed in minimum daily wage (mdw) units, so for
example a 1,000,000 Gs. contract expressed in multiples of a legal mdw of
50,000 Gs. would amount to 20 mdw. See Online Appendix for more
23. Systematic data on total sales, proﬁts, etc., for the whole universe of
ﬁrms could not be accessed due to conﬁdentiality restrictions.
24. FOB is the standard way to report import values without costs of
transport and other taxes, while CIF is the standard way to report export
values, including cost, insurance and freight to the national border.
25. Barbosa and Straub (2014).
26. See European Commission (2011).
27. The inclusion of ﬁxed eﬀects prevents us from using a probit
estimation, while a conditional logit would imply eliminating any pair for
which there is no within variation, therefore reducing the ﬁnal sample by
28. Here, identiﬁcation arises from both cross-sectional and time
variation of the amount of exception used across institution–ﬁrm pairs,
controlling for a time trend, and institution and goods time-invariant
levels (column 1) as well as institution- and good-speciﬁc time trends
29. Note that the identiﬁcation of the nature of the relationship between
public buyers and suppliers is beyond the possibilities of our data. It
maybe for example that public oﬃcials or their family members have
direct stakes in the supplying ﬁrms, as the anecdotal evidence suggests, or
that they operate at arm-length and share bribes.
30. Note that institution-year ﬁxed eﬀects absorb the direct eﬀect of
corruption, which is measured year by year. The pairwise correlations
between the Corrjt and the Xijt variables are 0.003 and 0.04,
respectively, so we are not worried by potential multicollinearity.
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 405
31. Note that running the speciﬁcations of columns 1–3 on the smaller
sample of columns 4–6 conﬁrms a positive (approximately an additional
10% for a one standard value increase) and signiﬁcant eﬀect for the total
pairs’ contract value.
32. The news index might be particularly subject to caution, as press
coverage of speciﬁc institutions, based for example on journalists inquiries
or on denunciations, is likely to be inﬂuenced by the nature of the
institutions and their past behavior in procurement or other activities.
Using only the mean of the evaluations based on the Comptroller
General’s report and on the number of administrative indictments in any
given institution yields similar results though.
33. The exception if the coeﬃcient for the share of institutions’ budget
made with the ﬁrm, in column 5, which is signiﬁcant at the 10% level.
34. Note also that the other coeﬃcients support our previous insights:
the control group-speciﬁc trend b1is not signiﬁcantly diﬀerent from 0,
supporting the idea that the meaningful reduction occurs in the identiﬁed
treatment group, while b2is positive and signiﬁcant, supporting our
previous results on the fact that pairs with frequent/intense interactions
make more use of the exception in the period 2004–05.
35. Using all the observations to measure the variations in net gains, we
would have some positive measurement errors (when a ﬁrm’s tax
observation is out of the sample and therefore set at zero for one year
and is positive the following one), some negative ones (in the reverse case),
and more generally errors going either way for ﬁrms that do not make it to
the ranking of top taxpayers.
36. Note however that such endogeneity concerns are much less obvious
for variables such as the average level of corruption.
37. We do not have additional ﬁrm-level data to control for such general
characteristics. Fixed eﬀects will take care of the size issue as long as it is
reasonably constant over the period of study.
38. Results not shown here to save space indicate that the results in
columns 1 and 2 are robust to systematically controlling for the amounts
of ﬁrms’ sales to the State.
39. A technical concern has to do with tax evasion. Indeed, it is
likely that Paraguayan ﬁrms do not report all of their sales for tax
purpose, possibly biasing our estimations. One could think that sales
to the State, because they are publicly registered, imply lower rates of
evasion than other sales, in which case we may be facing an upward
bias in our estimations. However, strong anecdotal evidence rather
suggest that well-connected ﬁrms use their inﬂuence to evade a bigger
share of their tax obligations. This leads us to think that our estimates
should be considered as a lower bound on the true returns of these
ﬁrms. In other words, the fact that we still ﬁnd a positive eﬀect of
public contracts on proﬁts leads us to consider that the true eﬀect is
probably even larger.
40. The procedure, which aims to correct for the failure of the
ignorability-of-treatment assumption, is a kind of extended Heckit,
where sample selection is viewed as an omitted variable bias,
addressed by the inclusion of the Mills ratio as additional
regressors. Fafchamps and La Ferrara (2012) apply this technique
to control for individuals’ self-selection into self-help groups based
on unobservable characteristics.
41. See Online Appendix for the details on how the instruments are
deﬁned, descriptive statistics and ﬁrst stage estimations.
42. The dummies for the ‘‘construction”and ‘‘consultancy”categories
correlate negatively with proﬁts. Moreover, the results are robust to
excluding the import–export category.
43. This has also fueled a ﬂourishing and illegal reexportation business to
the neighbors Brazil and Argentina. See Straub (1998) for more details on
Agrawal, A., & Knoeber, C. (2001). Do some outside directors play a
political role?. Journal of Law and Economics, 44, 179–198.
Auriol, E. (2006). Corruption in procurement and public purchase.
International Journal of Industrial Organization, 24(5), 867–885.
Bagnoli, M., & Bergstrom, T. (2005). Log-concave probability and it’s
applications. Economic Theory,26(2), Springer.
Bandiera, O., Prat, A., & Valletti, T. (2009). Active and Passive waste in
government spending: Evidence from a policy experiment. American
Economic Review, 99, 1278–1308.
Barbosa, K., & Straub, S. (2014). Revolving doors in Brazilian health
Bardhan, P. (1997). Corruption and development: A review of issues.
Journal of Economic Literature, 35(3), 1320–1346.
Baumol, W. J. (1990). Entrepreneurship: Productive, unproductive, and
destructive. Journal of Political Economy, 98(5), 893–921.
Becker, G., & Stigler, G. J. (1974). Law enforcement, malfeasance and
compensation of enforcers. Journal of Legal Studies, 3, 1–18.
Bertrand, M., Djankov, S., Hanna, R., & Mullainathan, S. (2007).
Obtaining a driving license in india: an experimental approach to
studying corruption. The Quarterly Journal of Economics, 122(4),
Besley, T., & McLaren, J. (1993). Taxes and Bribery: The role of wage
incentives. Economic Journal, 103, 119–141.
Bhagwati, J. (1982). Directly unproductive, proﬁt seeking (DUP) activ-
ities. Journal of Political Economy., 90, 998–1002.
Borzadaran, G. R. M., & Borzadaran, H. A. M. (2011). Log-concavity
property for some well-known distributions. Surveys in Mathematics
and its Applications, 6, 203–219.
Buchanan, J. M. (1980). Rent-seeking and proﬁt-seeking. In J. M.
Buchanan, R. D. Tollison, & G. Tullock (Eds.), Towards a theory of
the rent-seeking society (pp. 3–15). College Station: Texas A&M
Chong, E., Klien, M., & Saussier, S. (2013). The quality of governance and
the use of negociated procurement procedures: Some (un)surprising
evidencefrom the European Union. MIMEO.
Davis, J. (2004). Corruption in public service delivery: Experience from
South Asia’s water and sanitation sector. World Development, 32(1),
Deiniger, K., & Mpuga, P. (2004). Does greater accountability improve
the quality of public service delivery? Evidence from Uganda. World
Development, 33(1), 171–191.
Della Porta, D., & Rose-Ackerman, S. (Eds.) (2002). Corrupt exchanges:
Empirical themes in the politics and political economy of corruption.
Di Tella, R., & Schargrodsky, E. (2003). The role of wages and auditing
during a crackdown on corruption in the city of Buenos Aires. Journal
of Law and Economics, 46(1), 269–292.
Elbahnasawy, N. G. (2014). E-government, internet adoption, and
corruption: An empirical investigation. World Development, 57,
European Commission. (2011). Public procurement in Europe. Cost and
eﬀectiveness. Study prepared for the European Commission by PwC,
London Economics and Ecorys.
Fafchamps, M., & La Ferrara, E. (2012). Self-help groups and mutual
assistance: Evidence from Kenyan slums. Economic Development and
Cultural Change, 60(4), 707-734.
406 WORLD DEVELOPMENT
Ferraz, C., & Finan, F. (2008). Exposing corrupt politicians: The eﬀect of
Brazil’s publicly released audits on electoral outcomes. Quarterly
Journal of Economics, 123(2), 703–745.
Fisman, R. (2001). Estimating the value of political connections. American
Economic Review, 91(4), 1095–1102.
Fisman, R., & Sarria-Allende, V. (2010). Regulation of entry and the
distortion of industrial organization. Journal of Applied Economics, 13
Fries, S., Lysenko, T., & Polanec, S. (2003). The 2002 business environment
and enterprise performance survey: Results from a survey of 6,100 ﬁrms.
Working Paper No. 84, European Bank for Reconstructions and
Goldman, E., Rocholl, J., & So, J. (2013). Politically connected boards of
directors and the allocation of procurement contracts. Review of
Finance, 17(5), 1617–1648.
Hellman, J. S., Jones, G., & Kaufmann, D. (2003). ‘Seize the state, seize
the day’: State capture, corruption, and inﬂuence in transition. Journal
of Comparative Economics, 31, 751–773.
Hsieh, C.-T., Miguel, E., Ortega, D., & Rodriguez, F. (2011). The price of
political opposition: Evidence from Venezuela’s Maisanta. American
Economic Journal: Applied Economics, 3(2), 196–214.
Hyytinen, A., Lundberg, S. & Toivanen, O. (2007). Politics and procure-
ment: Evidence from cleaning contracts. HECER Discussion Paper No.
Khwaja, A. I., & Mian, A. (2005). Do lenders favor politically connected
ﬁrms? Rent provision in an emerging ﬁnancial market. Quarterly
Journal of Economics, 120(4), 1371–1411.
Krueger, A. (1974). The political economy of the rent seeking society.
American Economic Review, 64, 291–303.
Li, H., Meng, L., Wang, Q., & Zhou, L.-A. (2008). Political connections,
ﬁnancing and ﬁrm performance: Evidence from Chinese private ﬁrms.
Journal of Development Economics, 87(2), 283–299.
Mauro, P. (1995). Corruption and growth. The Quarterly Journal of
Economics, 110(3), 681–712.
Mironov, M., & Zhuravskaya, E. (2012). Corruption in procurement and
shadow campaign ﬁnancing: Evidence from Russia. MIMEO.
Mookherjee, D., & Png, I. P. L. (1995). Corruptible law enforcers: How
should they be compensated?. Economic Journal, 105, 145–159.
Murphy, K. M., Shleifer, A., & Vishny, R. W. (1991). The allocation of
talent: Implication for growth. Quarterly Journal of Economics, 106(2),
Nanda, A. K., & Shaked, M. (2001). The hazard rate and the reversed
hazard rate orders, with applications to order statistics. Annals of the
Institute of Statistical Mathematics, 53(4), 853–864.
Olken, B. (2007). Monitoring corruption: Evidence from a ﬁeld experi-
ment in Indonesia. Journal of Political Economy, 115(2), 200–249.
Olken, B., & Pande, R. (2012). Corruption in developing countries. Annual
Review of Economics, 4, 479–505.
Pande, R. (2008). Understanding political corruption in low income
countries. In T. Schultz, & J. Strauss (Eds.). Handbook of development
economics (Vol. 4). North-Holland: Elsevier.
Rama, M. (1993). rent-seeking trade policy. A time-series approach. Policy
research working paper 1142. Washington, DC: The World Bank.
Reinikka, R., & Svensson, J. (2004). Local capture: Evidence from a
central government transfer program in Uganda. Quarterly Journal of
Economics, 119(2), 679–706.
Rose-Ackerman, S. (1999). Corruption and government: Causes, conse-
quences, and reform. Cambridge: Cambridge University Press.
Slinko, I., Zhuravskaya, E., & Yakovlev, E. (2005). Laws for sale:
Evidence from Russia. American Law and Economics Review, 7(1),
Sequeira, S. (2014). Tariﬀs and Corruption: Evidence from a Tariﬀ
Liberalization program. MIMEO.
Svensson, J. (2005). Eight questions about corruption. Journal of
Economic Perspectives, 19(5), 19–42.
Straub, S. (1998). Evolucio
´mica del Paraguay 1989–1997:
Burbuja de Consumo y Crisis Financiera. Revista de la CEPAL, 65,
Straub, S. (2014). Political ﬁrms, public procurement, and the democrati-
zation process. TSE Working Paper, no. 14-461.
Tran, A. (2011). Which regulations reduce corruption? Evidence from the
Internal records of a bribe-paying ﬁrm. Working paper.
Transparencia Paraguay. (2006). Indice de desempen
˜o e integridad en
´blicas 2004–2005. Asuncion.
Tullock, G. (1967). The welfare costs of tariﬀs, monopolies, and theft.
Western Economic Journal, 5, 224–232.
Tybout, J. R. (2000). Manufacturing ﬁrms in developing countries: How
well do they do and why?. Journal of Economic Literature, 38(1),
Ufere, N., Perelli, S., Boland, R., & Carlsson, B. (2012). Merchants of
corruption: How entrepreneurs manufacture and supply bribes. World
Development, 40(12), 2440–2453.
Wade, R. (1985). The market for public oﬃce: Why the Indian state is not
better at development. World Development, 13(4), 467–497.
Wooldridge, J. (2002). Econometric analysis of cross section and panel data.
APPENDIX A. SUPPLEMENTARY DATA
Supplementary data associated with this article can be
found, in the online version, at http://dx.doi.org/10.1016/j.
Available online at www.sciencedirect.com
PUBLIC PROCUREMENT AND RENT-SEEKING: THE CASE OF PARAGUAY 407