Corruption in Procurement and the Political Cycle in

Tunneling: Evidence from Financial Transactions Data

Maxim Mironov and Ekaterina Zhuravskaya
This draft: June 2015

Abstract
We provide evidence of corruption in allocation of public procurement and assess its
efficiency. Firms with procurement revenue increase tunneling around regional elections,
whereas neither tunneling of firms without procurement revenue, nor legitimate business
of firms with procurement exhibits a political cycle. Data are consistent with the corruption channel - cash is tunneled to politicians in exchange for procurement contracts - and
inconsistent with alternative channels. Using the strength of correlation between procurement revenue and tunneling around elections as a proxy for local corruption, we reject the
efficient grease hypothesis: in more corrupt localities, procurement contracts go to unproductive firms.

We thank anonymous referees, Ray Ball, Francesco Caselli, Mark Duggan, Matthew Gentzkow,
Sergei Guriev, Irena Grosfeld, Peter Lorentzen, Christian Leuz, Garen Markarian, Elena Paltseva,
Paolo Porchia, Paola Sapienza, Andrei Shleifer, Jeremy Stein, Marco Trombetta, and the participants of the NES/SITE retreat in Ventspils, the ISNIE Annual Conference, NBER Political
Economics meetings, 2014 ASSA/AEA in Philadelphia, and seminar participants at UTDT, IE
Business School, EBRD, Paris School of Economics, New Economic School, Sciences Po, Toulouse
School of Economics, ESSEC and the University of Cergy-Pontoise for helpful comments.

Mironov is from IE Business School, Madrid; Zhuravskaya is from the Paris School of Economics (EHESS). Please send correspondence to Ekaterina Zhuravskaya, Paris School of Economics, 48 boulevard Jourdan, 75014 Paris, France; Email: Zhuravsk@pse.ens.fr.

Introduction

Does corruption serve as an efficient grease by allowing more productive firms to get around
inefficient red tape (Leff, 1964; Huntington, 1968)? Or, on the contrary, does corruption pave the
way for less productive firms with political connections to bypass regulations and win government
procurement contracts, thus, increasing inefficiency? In theory, both answers are possible (see,
for instance, the survey by Aidt, 2003). Empirical evidence on this question is limited, despite a
sharp increase over the past two decades in research on corruption. The lack of evidence is not
surprising: bribes are unobservable and corruption is difficult to estimate. Due to the secretive
nature of corruption, most empirical work had been based on subjective perception measures,
which is problematic for assessing welfare effects of corruption (Reinikka and Svensson, 2006).
Recognizing the problems with subjective evidence, researchers have recently turned to studying
corruption using policy experiments (e.g., Reinikka and Svensson, 2004; Olken, 2006), natural
experiments (e.g., Caselli and Michaels, 2009), and field experiments (e.g., Bertrand et al., 2007b;
Olken, 2007; Ferraz and Finan, 2008). However, experiments that allow evaluation of the effects
of corruption are rare and often cover a specific area of corrupt economic activities.
The goal of this paper is twofold: first, to reliably estimate the variation in corruption in
allocation of public procurement contracts across localities based on objective data without narrowing the scope, namely, for the near-population of Russias large firms, and second, to test
whether corruption facilitates or hurts efficiency in allocation of government procurement across
firms, i.e., to test the efficient grease hypothesis.
We measure the variation in corruption across localities in Russia by relating illicit financial
transactions around the time of regional elections to the allocation of government procurement
contracts at the local level. Using financial transactions data of the weekly frequency for the
near-population of large firms in Russia, we measure tunneling, i.e., illegal transfers of cash out of
firms (Johnson et al., 2000). With this measure at hand, we document that firms that win government procurement contracts intensify tunneling around regional elections substantially more
than firms without government procurement revenue by estimating the political cycle in tun1

neling using difference-in-differences methodology. Next, we test for the mechanism behind this
relationship. We show that the political cycle in tunneling for winners of procurement contracts
cannot be explained by a political cycle in their legitimate business activity or by a political cycle
in procurement revenue, as neither is sufficiently large. The data are also inconsistent with an
increase in the political risk, associated with elections, of being the driving force of the political
cycle in tunneling, as the political cycle in tunneling decreases with a margin of victory and political turnover. In contrast, the data are consistent with corruption explaining the political cycle in
tunneling, namely, that firms use tunneled cash to bribe politicians in order to win procurement
contracts. The correlation between tunneling around elections and obtaining government procurement contracts is significantly higher in regions deemed to be corrupt, according to Transparency
Internationals Corruption Perception Index (CPI), a perception-based measure of corruption.
As most procurement contracts are allocated at the local level in Russia, and the strength
of correlation between tunneling around elections and the allocation of procurement across firms
varies across localities, we use the strength of this correlation as a proxy for the extent of corruption
in public procurement at the local level. Our proxy for local corruption is strongly correlated with
the regional CPI but has important advantages over it. Unlike the CPI, it is based on objective
data and is available for almost the entire Russian economy at the subregional level. Using
the variation in this measure of corruption across different localities within a region, we show
that in more corrupt environments public procurement contracts are allocated to less productive
firms, controlling for region, industry, and even locality fixed effects. More productive firms lose
competition for public procurement contracts when their allocation depends on illicit payments
to politicians. Thus, we reject the efficient grease hypothesis and conclude that corruption has
negative welfare implications.
The data that made our research possible come from a list (available on the Internet) of banking
transactions of legal entities in Russia over a six-year period. These data were previously used by
Mironov (2013), who developed methodology to identify tunneling (Johnson et al., 2000; Desai et
al., 2007), i.e., the amount of transfers to fly-by-night firms set up to take cash out of companies,

at each point in time for legitimate firms. Legitimate firms are defined as those that pay taxes
and report data to the authorities, whereas fly-by-night firms are those that have revenue but do
not pay taxes, even though they should be doing so according to Russian legislation. Banking
transactions data allow us to observe taxes paid by firms and public procurement revenue of firms,
as both show up among a firms banking transactions. Mironov (2013) measures tunneling and
shows that tunneling across firms is associated with lower firm performance.
Our paper uses the same data and methodology to identify tunneling, but it addresses a
different research question. We contribute to the literature in the following respects. First, our
paper is the first one to focus on the over-time variation in tunneling within firms and relate it
to the proximity to regional elections. We document the political cycle in tunneling for firms
with government procurement revenue and no cycle in tunneling for firms without government
procurement revenue. The political cycle in tunneling for firms with government procurement
contracts allows us to detect corruption, i.e., illegal payments to politiciansbribes and kickbacksin exchange for government procurement contracts, as opposed to tunneling for other
purposes such as tax evasion and income diversion. Second, by considering the allocation rule
for government procurement contracts among firms, we show that in places where government
procurement depends on bribes and kickbacks, the contracts get awarded to less productive firms.
This result differs from that of Mironov (2013), who shows that tunneling is associated with
lower firm performance, because the firms that get procurement contracts are not necessarily
the ones that tunnel more, but rather, the ones that tunnel more around elections conditional
on tunneling outside elections, and the overall level of tunneling is similar between corrupt and
noncorrupt localities.
Our main contribution is to the literature on the welfare implications of corruption (e.g.,
Shleifer and Vishny, 1993, 1994). Our work is particularly related to the recent strand of this
literature, which provides empirical estimates of social costs of political connections and shows
that political connections, in part, determine the allocation of government procurement contracts,

(e.g., Cingano and Pinotti, 2013; Amore and Bennedsen, 2013; Goldman et al., 2013).1 We also
contribute to the empirical literature that attempts to provide systematic evidence of corruption
using objective rather than perception-based measures (e.g., Di Tella and Schargrodsky, 2003;
Reinikka and Svensson, 2004; Bertrand et al., 2007b; Olken, 2007; Fisman and Miguel, 2007;
Butler et al., 2009; Caselli and Michaels, 2009; Cheung et al., 2011; Ferraz and Finan, 2011).
Our work is also related to the literature on opportunistic political cycles (see, for instance, a
survey by Drazen, 2001). This work focuses primarily on the correspondence between election cycles and benefits directed to voters in the form of transfers and social expenditure, e.g., Akhmedov
and Zhuravskaya (2004) document short-lived and very strong political budget cycles for Russian
regional elections. In this paper, we document a political cycle in illegal cash payments that firms
make to politicians in order to obtain procurement contracts. Related to our findings, Burgess et
al. (2012) show a political cycle in another illegal activity that brings cash to politicians, namely,
forest extraction in the tropics. Our finding that firms provide benefits to politicians in the face
of elections is also related to Bertrand et al. (2007a), who document that political connections are
associated with the political cycle in employment granted for the political benefit of incumbent
politicians.
The paper proceeds as follows. In section 2, we describe the data and the main variables
used in the analysis. In Section 3, we present evidence of a political cycle in illegal cash tunneled
from companies that receive public procurement revenue and estimates the size of illicit payments to politicians associated with obtaining procurement contracts. Section 4.1 develops the
methodology of measuring corruption using the association between illicit payments and receiving
procurement revenue across firms and presents a test of this approach. In section 4.2, we use our
measures of corruption show that corruption is associated with an inefficiency in the allocation
of public procurement. In section 5, we conclude.
1

See also work surveyed in Bardhan (1997); Rose-Ackerman (1999); Svensson (2005); Olken and Pande (2012).

Data

2.1

Legitimacy and Reliability of the Banking Transactions Data

Our main data source is the dataset of banking transactions among legal entities in Russia between
1999 and 2004; it was allegedly leaked to the public from the Central Bank of Russia in 2005, and
it is now widely available on the Internet. These data were previously used by Mironov (2013) and
are available either for free or for a symbolic payment from several websites: www.vivedata.com,
www.rusbd.com, www.wmbase.com, www.mos-inform.com, www.specsoft.info, etc. The Russian
press widely discussed the appearance of these data in the public domain.2 The websites that
demand a symbolic payment primarily charge for having formatted the dataset to make the data
more easily accessible rather than for the dataset itself, which is elsewhere available for free.
Though the data appeared in the public domain presumably without official permission from the
Central Bank of Russia, both the Russian government and the Russian Central Bank are aware
that journalists and researchers are using this data, and both have publicly discussed policyrelevant conclusions of analyses based on the data (see, e.g., the transcript of the Conference on
Tax Evasion at the Ministry of Economy, which took place in Moscow in October 2006).3 The
fact that government and Central Bank officials take the results of research based on these data
seriously indicates the reliability of these data.
As for legal issues associated with the use of these data, no lawsuits have been initiated against
any party for using these data in spite of fairly wide circulation of the data and publications
by both journalists and researchers.4 Lawyers within the Ministry of Interior of the Russian
Federation, when commenting in the press on the legitimacy of these data, explained that the
2

See, for instance, the main Russian business daily, Vedomosti, on March 30, 2005.
The authors of this paper received a request from the First Deputy Chairman of the Central Bank of Russia,
Andrei Kozlov, and a deputy chairman, Viktor Melnikov, to write a policy memo explaining the methodology of
identifying fly-by-night firms using the banking transactions data as developed by Mironov (2013). In this request,
the top Central Bank officials refer to the data set from the Internet as a legitimate source of information and
acknowledge that the research department of the Central Bank uses the same data. A copy of the letter is available
from the authors upon request.
4
For an example of a journalistic investigation using these data, see Vedomosti on May 20, 2005; for popular
descriptions of research based on these data, see Vedomosti on July 24, 2006 and October 25, 2011. For research
on similar datasets, see, for instance, Guriev and Rachinsky (2006) and Braguinsky et al. (2011).
3

Central Bank never admitted that it leaked any data, and therefore, from a legal standpoint, all
data sets in the public domain are legal and no dataset is considered illegitimate.5
A detailed description of the dataset on banking transactions and several reality checks on
them are provided in the online appendix to Mironov (2013).6 In the online appendix to this
paper, we reproduce the most important reality checks, which lead to the following conclusions.
First, the banking transactions data for the group of firms that actually pay taxes (i.e., legitimate
firms, as defined precisely in the next section and used as a unit of observation in the analysis
that follows) match rather well with the registry of Russian firms published by Russias official
statistical agency, Rosstat, in 2003. The official registry often has no information on firms that
do not pay taxes (i.e., fly-by-night firms, as defined in the next section), according to the banking
transactions data. Second, for the firms that are present both in the banking transactions data
and in the official registry, firm characteristicsavailable in both data setsare very similar,
which is another important sign that the data are reliable.7

2.2

Sample and Variables

The banking transactions data for 2003 and 2004 were used by Mironov (2013) and come from
www.vivedata.com. The data for 1999-2002 come from www.rusbd.com. The data set contains
513, 169, 660 transactions involving 1, 721, 914 legal entities with information on the date of each
transaction, the payer, the recipient, the amount of each transaction, and the self-reported purpose of it. Our aim is to test for a relationship between transfers to fly-by-night firms from
regular nongovernment firms around elections, and the public procurements contracts that these
5

See, for instance, Financial-economic news published by Interfax on April 1, 2005. See also an article, published in a journal specializing on covering the banking sector, Bankovskoye Obozreniye (Banking Review), No. 11,
November 2005, in which an economist from the Central Bank explains the phenomenon of the leakage of these data
by the excessive regulation of secrecy and the lack of financial transparency regulations, he argues for the need to
make the data officially public, see http://bankir.ru/publikacii/s/provodki-cb-rf-vorovat-nelzya-pokypat-1378429/
(accessed on November 24, 2014).
6
The
full
text
of
the
online
appendix
to
Mironov
(2013)
is
available
at
http://onlinelibrary.wiley.com/doi/10.1111/jofi.12026/suppinfo (accessed on November 24, 2014).
7
The use of the banking transactions data has some methodological parallels with using other unconventional
data sources to provide empirical evidence on the questions that cannot be studied using conventional data (e.g.,
Levitt and Venkatesh, 2000; Guriev and Rachinsky, 2006; Braguinsky and Mityakov, Forthcoming; Braguinsky et
al., 2011; Mironov, Forthcoming). Much of this work focuses on Russia because of data availability.

regular firms receive. Thus, the tunneling revenue and public procurement revenue are the two
main variables in our analysis. Both of these variables are constructed from the list of banking
transactions. (We describe how we construct these variables below.)
Following Mironov (2013), we first clean the banking transactions database. Nearly threefourths of all legal entities in the dataset (72.2%) are either tiny firms (kiosks and small shops) with
monthly revenue of less than 100,000 RUR (about US$3,000) or have fewer than 10 transactions
between 1999 and 2004, which probably means they represent typos in the legal entity ID numbers.
These legal entity IDs are numerous, but their transactions are extremely small: together they
constitute a scant 0.8% of the value of all transactions. Thus, dropping these IDs leaves us with
27.8% of all legal entity IDs and with 99.2% of the value of all transactions.8 As our focus is on
domestic, nonfinancial, nongovernment firms, we take the universe of all remaining legal entity
IDs and eliminate all government and municipal IDs, all firms and agencies with 100% state or
municipal ownership, all financial institutions (such as banks), all foreign companies, and all selfemployed entrepreneurs without a legal enterprise status. Financial institutions constitute 1.1%
of the IDs, but their transactions are large and numerous68.1% of the value of all transactions.
Government and municipal agencies constitute 1% of all IDs, and their transactions constitute
6.9% of the value of all transactions. Foreign firms constitute 0.5% of all legal entities and
their transactions amount to 0.4% of the value of all transactions. We also drop self-employed
entrepreneurs, which constitute 1.4% of IDs and 4.7% of the value of all transactions. Thus, we are
left with 85.8% of all legal entities, a near-population of domestic, nonfinancial, nongovernment
business entities; they are the focus of our analysis. The Panel A of Table A.1 in the online
appendix presents the number of legal entities and the value of transactions corresponding to the
subsamples of transactions in the banking transactions database after consecutively eliminating
each of these categories of IDs.
Mironov (2013) developed the methodology of identifying fly-by-night firms, i.e., firms that
have profitable banking transactions but pay no taxes, in the banking transactions data set.
8

All of Russias legal entities have a unique identifier ID, called an INN, which stands for individual taxpayer
number.

Intuitively, fly-by-night firms pay no taxes despite having transactions that require the payment
of taxes according to Russian law.9 To be precise, firms are defined as fly-by-night when they
satisfy the following three criteria: (i) the ratio of taxes paid to the difference in cash inflows
and outflows is negligible (i.e., below 0.1%); (ii) social security taxes are below the amount that
corresponds to the social security tax for a firm with one employee on a minimum wage (i.e.,
$7.20); and (iii) cash inflows are higher than cash outflows. Mironov (2013) argues that judging
by these firms transactions, one can be reasonably sure that these firms exist in order to tunnel
cash for legitimate firms.
Fly-by-night firms help legitimate firms transfer large amounts of cash illegally for various
purposes, such as tax evasion and diverting cash from shareholders to managers and from minority
shareholders to majority shareholders (see Johnson et al., 2000) or in order to be used as bribes,
which is the focus of this paper. Mironov (2013) provides evidence that fly-by-night firms are
usually registered on stolen passports and do not provide any real services or produce any real
goods. Instead, the legitimate firms sign a legal (from their standpoint) contract with fly-by-night
firms for consulting services and pay them for these services via a banking transaction. Then,
fly-by-night firms withdraw the cash from the bank and give it back to the management of the
legitimate firm (for a fee).10 All tax liabilities associated with such banking transactions rest with
the fly-by-night firms, whereas the legitimate firms get the tunneled cash and are free to use it.
On the basis of firms banking transactions, Mironov (2013) also defines firms that are fully
legitimate (which does not preclude them from making transfers to fly-by-night firms). A firm is
classified as fully legitimate if its ratio of taxes paid to the difference in cash inflows and outflows
is above 1%. The rest of the firms in the banking transactions data (after accounting for the flyby-night firms and fully legitimate firms) are defined as semi-legal firms, as they do have some
regular operations, and do pay some taxes, but also evade taxes without the help of fly-by-night
9

All Russian legal entities, irrespective of their legal form, i.e., including non-for-profits, are supposed to pay
at least some taxes, such as, for instance, payroll tax. Having revenues and no taxes is a clear indication of an
illegal activity.
10
We cannot track what happens to the money after it was transferred to fly-by-night firms precisely because
it is cashed out, as we only observe banking transitions between legal entities.

firms.11 According to the Mironovs criteria, we identified 168,009 regular fully legitimate firms,
99,925 fly-by-night firms and 143,560 semi-legal firms among the private domestic non-financial
business entities in the banking transactions data (as reported in Panel B of Table A.1). Note
that the vast majority of the 168,009 fully legitimate firms and practically all 143,560 semi-legal
firms are small businesses. We are interested in the behavior of fully legitimate large firmswhich
both have commercial transactions and pay taxesbecause only fully legitimate firms can apply
for and win public procurement contracts. Though legitimate firms may engage in illegitimate
tunneling, they appear fully legitimate on paper because they use services of fly-by-night firms.
Following Mironov (2013), we deem all the transfers from fully legitimate firms to fly-by-night
firms as tunneling.
We use the banking transactions data set also to identify those legitimate firms that get procurement contracts from the government. We define revenue from public procurement contracts
as the amount of banking transactions from government entities to regular firms that have the
reported purpose of payment for goods and services.12 Thus, using banking transactions data,
we construct two variables (measured weekly): tunneling, i.e., transfers to fly-by-night firms, and
procurement revenue, i.e., transfers from government agencies with the stated purpose of paying
for provision of goods and services, for each legitimate firm.13
We also collect data on the basic characteristics of legitimate firms, such as location, revenue,
net income, debt, assets, and industry, which we use as control variables and labor productivity
proxied by revenue per worker. All these variables are available only as cross-section in 2003
for the near-population of industrial firms in Russia. The source of these data is the registry of
industrial firms, compiled by Russias official statistical agency (Rosstat).14 We merge legitimate
11

The vast majority of these firms have the ratio of taxes paid to the difference in cash inflows and outflows
between 0.1% and 1%.
12
In the baseline analysis, we exclude payments for utilities such as electricity and water from the list of revenues
from public procurement contracts because the utilities contracts are not allocated on a competitive basis, instead,
they automatically go to local utility monopolists. The inclusion of utilities would bias the results in our favor.
13
Formally, the rules of procurement actions are governed by a federal law and are the same in all regions and
localities. Each procurement auction has to be publicized in the media; and any firm can apply with a bid. The
authorities then consider bids and announce the winner.
14
We do not have access to official statistical data for other years. In addition, note that not all firms in the
registry report employment and, therefore, we can measure labor productivity only for a subset of firms.

firm IDs from the banking transaction database to the Rosstat registry data. Thus, to be in our
sample, a firm should be present both in Rosstats 2003 registry and in the list of legitimate firms
from the banking transactions database.
We use the following additional criteria for our sample of legitimate firms. We focus only on
large legitimate firms, i.e., firms with official reported revenue greater than $1M in 2003. We apply
this criterion in order to obtain a data set of manageable size and because one can reasonably
expect only relatively large firms to engage in bribing in exchange for obtaining government
procurement contracts. We verified that small firms are very unlikely either to get government
procurement contracts or to use services of fly-by-night firms; thus, they would add nothing
to our analysis apart from noise. In addition, the official registry data can be considered as a
near-population representative sample only for large firms, and not for small firms, which often
do not file information with statistical agencies. Furthermore, as the official registry data are
available only in 2003, the best indicator we have of whether the firm exists in a particular year
is that it makes banking transfers (all legitimate firms are supposed to pay taxes). We confine
our analysis to firms that make at least some banking transactions (to any recipient) within one
year of the regional elections. These criteria yield 45, 275 legitimate large firms. Table A.2 in
the online appendix presents information on the total number of firms and the cumulative size of
their revenue for all firms present in the Rosstat registry, all large firms (i.e., over $1M in revenue)
present in the Rosstat registry, and our sample, i.e., the intersection between the list of large firms
in the Rosstat registry and legitimate firms with recorded banking transactions within one year of
the regional elections. The cumulative revenue of firms in our sample equals 77.9% of the revenue
of all firms in the registry.15
In the appendix, we present summary statistics for the entire sample (Table A.3) and separately for each region (Table A.4). All nominal variables are expressed in thousands of constant
2003 U.S. dollars. A detailed description of variables can be found in the data section of the
online appendix.
15

Our sample of 45, 275 firms is the largest data set on Russian firms ever considered.By comparison, in Brown
et al. (2006) 14, 620 firms in Russia are considered as a near-population sample of large industrial firms.

10

Since we are interested in estimating the electoral cycle in tunneling, we focus on the 87
(out of 89) Russian regions that held gubernatorial elections between 1999 and 2004. The two
excluded regions are Dagestan, which has a parliamentary form of government, and Chechnya,
which experienced a severe armed conflict in 1999-2000. In the 87 regions, during the period
under study, 129 elections took place at 48 different points in time.

The Political Cycle in Tunneling

In this section, we test for the political cycle in tunneling for firms with and without government procurement revenue, we find evidence of such a political cycle for winners of procurement
contracts, and we test for several alternative mechanisms driving the results.

3.1

Case Study: Inteko

Inteko was a Moscow-based private company owned by Yelena Baturina, the wife of former
Moscow mayor Yury Luzhkov. It provides an illustration of what the political cycle in tunneling by recipients of government procurement contracts can look like. According to Forbes, in
2010 Baturina was the richest woman in Russia and the third-richest woman in the world. She
made her fortune through procurement contracts and concessions allocated to Inteko by Moscow
city government officials while Luzhkov was the mayor of Moscow (from 1992 to 2010).16 Figure
1 presents transfers from Inteko to fly-by-night firms, calculated from the banking transactions
data, beginning six months before the Moscow mayoral election in December 2003 and ending
six months after the presidential election of March 2004. The data show tunneling activity concentrated around the elections. In particular, the two biggest transfers to fly-by-night firms from
Inteko, for US$2.8 million and $US4 million, took place, respectively, one week before the Moscow
mayoral election, when Baturinas husband stood for reelection, and one month before the presidential election, in which Vladimir Putin was running for his second term in office. Altogether,
16

The New York Times and Forbes magazine published a series of articles on Yelena Baturina and the source of her fortune.
See,
for instance,
the following
links:
http://topics.nytimes.com/top/reference/timestopics/people/b/yelena baturina/index.html ;
http://www.forbes.com/profile/elena-baturina/ and http://sputniknews.com/russia/20100615/159431047.html
(websites accessed on November 24, 2014).

11

in five consecutive months around these two elections from November 2003 to March 2004, Inteko tunneled $US10.3 million. By contrast, Inteko tunneled just US$1.1 million during the five
months preceding November 2003 and the five months starting April 2004.
No information is available on how the cash tunneled out of Inteko was used; one could suspect
that at least some of it was spent on financing the two election campaigns. Indeed, according
to the testimony of Russian tycoon Roman Abramovich in the London High Court of Justice in
2011, the presidential election campaign of Mr. Putin in 2000 was illegally financed with cash
by businessmen close to Putin. In particular, Abramovich together with another tycoon, Boris
Berezovsky, spent US$ 50 million supporting Putins 2000 presidential campaign.17

3.2

Baseline estimation

Our first task is to estimate the electoral cycle in tunneling for firms with and without public
procurement contracts. We regress the tunneling that each firm in our sample makes each week
during the election year (between 1999 and 2004) normalized by the total amount of firms revenue
on a set of dummies indicating the time-distance to the election date, controlling for firm and
week fixed effects. We allow the electoral cycle to vary between two groups of firmsthose that
do and those that do not receive revenue from public procurement contractsas we are interested
in the difference in the magnitude of the electoral cycle between the two groups. Because larger
firms may be more able to tunnel, we also allow for differential electoral cycles depending on the
size of firms revenue. The unit of analysis here is a firm in a particular week. Altogether there
are 6, 286, 011 firm-week observations in the sample, i.e., firm-weeks in each region in the two
years around each election.
To be precise, we estimate the following two equations:
6
X
T unnelf t
1
=
m
Df m D P roc Revf e +
Revf
m=8

+
17

6
X

2
m
Df m
m=8

6
X

1
m
Df m
m=8

log(Revf ) +

2
X

l2 log(Inc transf,tl ) + t + f + f t , (1)

l=0

See the Wall Street Journal article Russian Tycoons Face Off in Court, November 7, 2011).

12

and
6
X
P rocRevf e
T unnelf t
1
=
m
Df m
+
Revf
Revf
m=8

6
X

2
m
Df m
m=8

6
X

1
m
Df m
m=8

log(Revf ) +

2
X

l2 log(Inc transf,tl ) + t + f + f t , (2)

l=0

f indexes firms. t indexes time in weeks. There are 313 weeks over the entire period under study.
Subscript e indexes elections in a particular region. (It is redundant for all regions where there
was only one election and meaningful for regions where there were two elections between 1999
and 2004.) The dependent variable is a measure of tunneling per week. In particular, T unnelf t
stands for the transfer by firm f to fly-by-night firms at time t. We normalize it by the revenue
of firm f in 2003, denoted by Revf .
The index m refers to the time-distance in months to the election date in the region where the
firm f is located, so that m = 1 refers to the month right before the election and m = 1 refers
to the month right after the election. As elections are always held on Sundays, whereas banking
transactions occur on weekdays, we define months, as four-week intervals, and m is never equal
to zero. We estimate the dynamics of tunneling around elections starting eight months before
elections and ending half a year after the elections, because all the interesting dynamics falls into
this interval. Thus, m varies from 8 to 1 and from 1 to 6. Df m is a dummy variable indicating
the time that is m months away from the election date in the region where firm f is located.
The coefficients on Df m estimate the political cycle in tunneling. In equation 1, the political
cycle in tunneling varies with two groups of firms: D P roc Revf e is a dummy variable, indicating
that

P rocRevf e
Revf

is greater than a certain threshold.

P rocRevf e
Revf

is the size of procurement revenues

+/ one year from the election date e in annualized terms (i.e., divided by 2) normalized by firm
revenues. In equation 2, we let the political cycle in tunneling vary with

P rocRevf e
.
Revf

As a baseline,

we consider the 1% threshold. To check the robustness of the results, we repeat all the analysis
for D P roc Revf e indicating firms with 5% of revenue coming from public procurement contracts.
As the political cycle in tunneling may depend on the size of the firm, we control for the

13

distance to elections Df m interacted with a measure of firm size, the logarithm of the firms
revenue in 2003, log(Revf ). In addition, we control directly for cash inflows into the firm bank
account log(Inc transf t ) along with two lags of this variable. This control is needed to make sure
that the timing of incoming bank transfers into the firms bank account are not driving the results
on the dynamics of outgoing transfers to fly-by-night firms. t and f are the full sets of 313 time
fixed effects and 45, 275 firm fixed effects. As elections that we consider are at the regional level,
the error terms f t are assumed to be clustered at the level of each Russian region. The results are
practically identical when we exclude the controls for the incoming transfers, log(Inc transf,t ),
log(Inc transf,t1 ), and log(Inc transf,t2 ), and for the differential political cycle depending on
the size of the firm, i.e., Df m log(Revf ).
2
1
2
1
from equation
and m
from equation 1 and m
and m
Our main parameters of interest are m
1
estimate the difference in the political cycle in tunneling between
2. The series of coefficients m
2
estimate the political
firms with procurement revenue above and below the threshold and m
1
cycle in tunneling for firms with procurement revenue below the threshold. The coefficients m

estimate the additional political cycle in tunneling per additional percentage point of revenue
2
coming from the procurement revenue, whereas m
estimate the political cycle in tunneling for

firms with zero procurement revenue.

Table 1 presents the results of the estimation of equation 1 in Panel A and equation 2 in
Panel B. Figures 2 and 3 illustrate the resulting patterns in estimated coefficients graphically.
1
2
The upper chart of Figure 2 portrays the sum of the point estimates of m
and m
from equation

1 each month m around the election date. These estimates show the dynamics of the tunneling
activity by firms with procurement revenue above the 1% threshold. Similarly, the middle chart
portrays the dynamics of tunneling for firms with procurement revenue below the 1% threshold,
2
i.e., m
coefficients. The lower chart shows the difference between the two, i.e., the estimates
1
1
of m
. The upper chart of Figure 3 presents the estimates of m
and the lower chart presents
2
the estimates of m
. All charts also portray the 95% confidence interval around each estimated

coefficient.

14

We find that starting six months before the election, firms with procurement revenue transfer
more and more cash to fly-by-night firms as elections approach, compared to their usual tunneling
activity. The tunneling increases monotonically until it peaks during the month right after the
election. Starting from the second month after the election on, tunneling returns to its usual
level and fluctuates around it. The political cycle is present for firms with and without pub2
lic procurement revenue. (In particular, m
coefficients are positive and statistically significant

pointing to a political cycle in tunneling of firms without any procurement revenue.) However, the
political cycle in tunneling is much more pronounced in firms with procurement contracts. The
lower charts of figures 2 and 3 demonstrate that the magnitude of the described political cycle
is significantly higher among firms with procurement revenue above the 1% threshold compared
to firms below this threshold and that the cycle is significantly magnified with an increase in
procurement revenue as a share of revenue. Table A.5 and Figure A.1 in the online appendix
establish the robustness of the results of the estimation of equation 1 to choosing an alternative
threshold for procurement revenue of 5% of revenue.
The last two rows in each panel of Table 1 present the results of the F -tests for the joint
significance of all and coefficients as well as the F -tests for the significance of the difference in
the average level of and coefficients within the brief, two-month window around the election,
when the tunneling is most pronounced, compared to the average level of and coefficients,
respectively, outside this two-month election window, i.e., average s and s from the intervals of
[8; 2] and [+2; +6] months away from election. All these F -tests yield statistical significance
well above conventional levels.
The magnitude of the additional tunneling around elections is substantial. During only one
month after the elections, firms that get more than 1% of their revenue from procurement contracts
transfer to fly-by-night firms an additional sum equal to about 0.33% of their annual revenue
(on top of the usual tunneling activity). For comparison, the average tunneling per month for
an average (median) firm is 0.4% (0.1%) of revenue (with a standard deviation of 0.7%). The
total additional tunneling during the interval starting six months before election and ending one

15

month after the election for the firms with procurement revenue above the 1% threshold is equal
to 1% of annual revenue, which is equal to 36% of what an average firm tunnels and 125%
of what a median firm tunnels in any seven months far away from the election. An increase
in political cycle for additional revenue from procurement contracts is substantial: For each
additional percentage point of annual revenue coming from procurement, the firms increase their
tunneling by an additional 3.9% of their annual revenue during the window of [6; +1] months
around elections.
Of the total amount of abnormal tunneling activity of firms with public procurement revenue
in the window starting six months before elections and ending one month after elections, one-half
of the sum is tunneled during the two months around elections, i.e., in the election window of [1;
+1] months around elections, and another half is tunneled during the interval starting six months
and ending one month before elections. Two-thirds of all abnormal tunneling around elections
takes place before the election date; one-third takes place during the month after the election.
To make sure that our standard errors are not too small and that our results are not driven
by differential trends in regions or firms, we reestimate the political cycle in tunneling using
equations 1 and 2 for 200 randomly chosen combinations of placebo election dates in our regions.
We draw placebo election dates randomly from the time interval that our data cover, at least
16 weeks away from the true election dates. Figure A.2 in the online appendix presents the
histograms of the F -statistic from the test of equality of the means of coefficients of interest
1
1
= out[1;1]
in the
1 and 1 inside and outside the election window (i.e., the tests for in[1;1]
1
1
upper chart and in[1;1]
= out[1;1]
in the lower chart of the graph). In each of the charts, the

vertical line indicates the value of the F -statistic for the same test performed on the true data,
which is substantially larger than vast majority of those generated by the placebo treatment. This
experiment shows that the pattern in the data that we uncover is very unlikely to be generated by a
random realization or differential trends between firms with and without government procurement
revenue.
Overall, we find a strong evidence of a political cycle in transfers to fly-by-night firms which

16

is substantially and significantly larger for firms with public procurement revenue.

3.3

Potential mechanisms

How one can explain the political cycle in tunneling? There are four potential explanations. First,
if all business activities exhibit a political cycle, tunneling should also increase around elections, as
a result. Second, if incoming procurement revenues exhibit a political cycle, this may also trigger
tunneling to follow a political cycle. Third, firms may tunnel more cash out around elections
because of increased political risk associated with elections. Fourth, firms may use tunneled cash
to bribe politicians in order to get procurement revenue. In this subsection, we consider each of
these potential explanations and conclude that only the last one is consistent with the data.
3.3.1

The political cycle in legitimate business activity?

Ordinary business activity of legitimate firms may exhibit a political cycle, because of a sharp increase in disbursement of government subsidies in the pre-election month (documented by Akhmedov and Zhuravskaya (2004)). The influx of cash subsidies into the regional economy right before
an election may, in principle, lead to a temporary increase in demand and, thus, to a boost in
economic activity of regional firms, which, in turn, may trigger a political cycle in tunneling.
Importantly, in the baseline regressions, reported in Table 1, we control for each firms incoming
revenue through banking transfers. This control should, to a large extent, account for such an
increase in business activity. We find not only that the cycle in tunneling is large and significant,
controlling for the incoming revenues, but also that including this control does not at all affect
the results on the political cycle in tunneling.
We also can test for a political cycle in outgoing banking transfers to other legitimate firms
to see whether regular business is more intense around elections for firms with government procurement contracts. We focus our attention on banking transfers from our sample of legitimate
firms, which they make to other legitimate firms in industries unrelated to publishing, media, and
services. Publishing houses and firms in the media sector produce political advertising during
election campaigns, and firms in the service sector provide services to voters as part of pork17

barrel politics. Therefore, regular firms may give favors to politicians by ordering (and paying
for) political advertisements for incumbents and ordering (and paying for) the provision of services to voters.18 As we are interested in the regular business of firms, rather than in their favors
to politicians (which would be consistent with the fourth explanation for the political cycle in
tunneling and which we consider below), we focus here on transfers that regular firms make to
other regular firms outside the domain of business related to election campaigns. Thus, we reestimate equation 1 using outgoing transfers to other nonmedia and nonservices firms (i.e., a proxy
for legitimate business activity) normalized by the firms revenue in 2003 (Revf ) as a dependent
variable.
Figure 4 presents the results, it portrays the dynamics of transfers to other legitimate firms
2
1
+m
(i.e., their business partners) for firms with procurement revenue above 1% of revenue (i.e., m

from equation 1) along with the 95% confidence interval. For comparison, we also present the
dynamics of tunneling. It is evident from the figure that regular business activity, in contrast
1
2
to tunneling, exhibits no significant political cycle. The coefficients m
andm
are imprecisely
1
2
estimated. Up until the election, the pattern of (m
+ m
) is completely flat. There is a small in

magnitude and statistically insignificant increase in the business activity of firms with procurement
revenue one month after the election, which slowly returns to normal within six months after the
election. None of the over-time changes in the regular business activity around elections are
statistically significant. The table below the figure confirms that the test for the equality of
coefficients inside and outside the election window of [1; 1] month around elections yields no
significant differences for transfers to legitimate firms (in contrast to transfers to fly-by-night
firms).
Thus, we conclude that the political cycle in tunneling cannot be driven by a general increase
in economic activity during the election time. This result is consistent with previous findings on
electoral cycles in Russia, which show no cycle in real economic indicators and a strong cycle in
governments social transfers and cash subsidies (Akhmedov and Zhuravskaya, 2004).
18
For example, food baskets in Russia are often delivered to pensioners homes before regional elections. Some
of them may be paid for by local businesses.

18

3.3.2

Political cycle in procurement revenues?

As mentioned above, in the baseline regressions we control for the total incoming revenues in
the three preceding weeks. However, one could argue that propensity to tunnel out of procurement revenue may be different from the propensity to tunnel out of nonprocurement revenue.
Therefore, if the incoming procurement revenue exhibits a political cycle due to an increase in
budget spending (i.e., in the political budget cycle), this is what may drive the political cycle in
tunneling.
First, we verify that, indeed, the propensity to tunnel is higher when firms get incoming
procurement revenue compared to other incoming revenue by regressing tunneling on all incoming
revenue and on incoming procurement revenue received by the firm in the last three weeks.
Precisely, we estimate the following equation:
2

T unnelf t X Inc transf,tl X P rocRevf,tl

=
l
+
l
+ t + f + f t .
Revf
Revf
Revf
l=0
l=0

(3)

We keep the same notation. As above, we adjust standard errors for clusters at the level of regions,
and we run this regression on the full sample of 6, 286, 011 firmsweeks observations. Coefficients
estimate the difference in propensity to tunnel out of procurement revenues (P rocRev) compared
to any incoming revenues (Inc trans), which also include procurement revenues. We find
coefficients to be large and statistically significant. The estimates suggest that propensity to
tunnel out of procurement revenues is 40 times larger than out of any incoming revenues. (The
regression output is presented in Table A.6 in the online appendix.) This is not surprising, as
government officials in Russia often require kickbacks for government procurement. However, as
we cannot distinguish between tunneling for paying kickbacks and tunneling for evading taxes, it is
possible that the political cycle in tunneling is driven by a political cycle in incoming procurement
revenue, if the latter exists.
To test for a political cycle in procurement revenue, we estimate the following equation:
6
6
X
X
P rocRevf t
1
2
=
m Df m +
m
Df m log(Revf ) + t + f + f t .
Revf
m=8
m=8

19

(4)

1
Again, all notation is as above. Coefficients m
estimate the political cycle in procurement revenue

m months away from election. As always, we adjust standard errors for clusters at the level of
regions and run this regression on the full sample of 6, 286, 011 firmsweeks observations. We find
no political cycle in incoming procurement revenue. Figure 5 presents the results. The incoming
procurement revenues of firms in our sample do not present a political cycle, on average. More
importantly, however, one needs to test whether those firms that tunnel more around elections
get more procurement revenues around elections, in other words, whether incoming procurement
revenue drives a political cycle of firms with a large procurement-revenue share. In order to test
this, we reestimate equations 1 and 2, controlling for the incoming procurement revenue in each
of the three preceding weeks, i.e., adding the following additional controls:
P rocRevf,t2
.
Revf

P rocRevf t P rocRevf,t1
,
,
Revf
Revf

Qualitatively, the results on the political cycle in tunneling are unaffected by the

addition of these controls. Quantitatively, the estimated political cycle in tunneling actually
becomes somewhat larger in magnitude. The results are presented in Figure A.3 in the online
appendix.
Overall, we conclude that the political cycle in tunneling is not driven by a political cycle in
incoming procurement revenues.
3.3.3

Political risk?

The third potential explanation is that firms tunnel cash out near election time because of an
increase in the political risk associated with a possible change in leadership. The change in
political risk could vary across firms, which would explain the differential cycle between firms
with and without procurement revenue. If political risk is behind the political cycle in tunneling,
one should expect tunneling around elections to be particularly high when elections result in
political turnover, which presents a high risk of changes in the formal regulatory environment or
any informal implicit contracts between the governor and the regional business. In contrast, if
an incumbent wins with a large margin of victory in the first round of elections, one should not
expect much change in the rules of the game between the business and the regional government,
and therefore, these elections should not be associated with any additional political risk. In such
20

elections, then, we should observe no political cycle in tunneling, if political risk is the mechanism
driving the political cycle.
We test these predictions by exploring how the margin of victory and political turnover affect
the presence and the magnitude of the political cycle in tunneling during elections. In 27 of 129
elections, the incumbent ran and lost. In 74 elections, the winner got more than 50% of the firstround vote; in 32 elections, the incumbent got more than 70% of the first-round vote. We test
for the political cycle in tunneling on the subsamples of elections in which the winner got above
50%, below 50% of the vote in the first round, in which the incumbent lost, and in which the
incumbent won, and finally in which the incumbent got above 70% and in which the incumbent
got below 70% of the vote in the first round of elections.
For conciseness of presentation, we estimate a simplified version of equations 1 and 2, in which
we replace 14 Df m dummies indicating distance to elections with just a single dummy DfElection
e
indicating the election window of [1; +1] months around the election in region where firm f is
located. We choose this window, as the deviation of the tunneling activity from the usual level is
the highest during this time (as is evident from the results of the estimation of the full political
cycle.) In particular, we estimate the following equations:
T unnelf t
= 1 DfElection
D P roc Revf e + 1 DfElection
+ 1 DfElection
log(Revf )+
e
e
e
Revf
2
X
+
l2 log(Inc transf,tl ) + t + f + f t , (5)
l=0

and
T unnelf t
P rocRevf e
= 1 DfElection

+ 2 DfElection
+ 1 DfElection
log(Revf )+
e
e
e
Revf
Revf
2
X
+
l2 log(Inc transf,tl ) + t + f + f t , (6)
l=0

in which superscript Election denotes the time period starting four weeks before and ending four
weeks after the regional election, and therefore, DfElection
is a dummy indicating the [1; +1]
e
month election window in election e for firm f , and all the rest of the notation is as above. Table

21

2 presents these results. We report the results of the estimation of equation 5 in Panel A and
of equation 6 in Panel B. As a baseline, as usual, we set the dummy D P roc Revf e to indicate
firms with procurement revenue above 1% of total revenue. (In the online appendix Table A.7,
we present the robustness of the results to choosing the 5% threshold).
We find that the political cycle in tunneling decreases sharply with an increase in political
competition, contrary to the prediction of the political-risk mechanism. There is no political
cycle in tunneling, among firms with procurement revenue above or below the 1% threshold, for
elections in which the winner got less than 50% of the vote in the first round. In contrast, for
elections that ended in one round, as the winner got above 50% in the first round, the political
cycle in tunneling is significant for both groups of firms and is substantially larger for firms
that have procurement revenue above the 1% threshold. The magnitude of the cycle for firms
with procurement revenue above the threshold, i.e., the sum of the coefficients 1 and 2 , is 7.2
times larger for elections where the winner won in the first round than in the rest of the sample.
Similarly, 1 + 2 is 6.4 times larger for elections in which incumbent won compared to elections
in which incumbent lost. It is 2.7 times larger for elections where the incumbent got more than
70% of the vote in the first round compared to elections in which incumbent got below 70%. The
marginal increase in the political cycle in tunneling per unit of procurement revenue is also much
larger for elections where the incumbent won (6 times), got over 70% of votes (1.7 times), and
where there was no second round (7.4 times) compared to the other elections, as reported in Panel
B of the Table 2.
Therefore, we conclude that the change in the political risk in the face of elections cannot be
the driving force of the intensified tunneling around elections.19
19

The differences in the political cycle in tunneling in regions with high and low margin of victory are an
interesting topic for future research. One possible explanation for the absence of political cycle in tunneling among
firms with procurement contracts when elections are competitive is that, in these regions, regional governments
are more accountable due to higher political competition and more democratic institutions and, therefore, there
is less corruption overall and in distribution of public procurement contracts, in particular. It is important to
note that this is only suggestive, as all the results concerning the effect of political competition measured by the
de-facto winning margin on the size of the cycle are subject to a reverse causality problem. The reason for this
is that campaign financing, including informal financing through the shadow economy, should have a direct effect
on the election results, such that incumbents, who run better campaigns, should have a better chance of winning.
At the same time, sure winners do not need to campaign. Thus, one should be careful in interpreting the results

22

3.3.4

Corruption in the allocation of procurement contracts?

Another possible interpretation of the political cycle in tunneling is that the cash is transferred to
politicians. If this is the case, the fact that the tunneling increases during elections much more for
firms that rely on contracts with the government for their business suggests that these transfers
might be used as informal payments, i.e., bribes, for obtaining procurement contracts.
There are two potential non-mutually-exclusive reasons for why bribes paid in exchange for
allocation of government procurement contracts may exhibit a political cycle. First, politicians
need cash the most when they are on the campaign trail. Note that the law severely restricts the
size of legal financing of election campaigns in Russia to the point that the funds raised legally
account for a tiny fraction of the total financing.20 In addition, all regional incumbents, even
those who are sure of winning, run election campaigns and distribute cash, goods, and services
to voters in various forms.21 This is important because the political cycle in tunneling is most
pronounced when incumbents are politically strong, as shown in the previous subsection. Some
campaign spending is realized right after the elections, as many campaign-related services (such
as printing and distribution of advertising leaflets, T-shirts, and posters) are provided up until the
very end of the campaign. This may partly explain abnormal tunneling right after the election.
As we have shown above, one-third of abnormal election-related tunneling takes place one month
after the election. We have no data to assess the relative share of these campaign expenses in
comparison to campaign spending made on the spot, before the election (such as, for instance,
direct vote-buying).
The second potential explanation for a political cycle in bribes to politicians, which, in parpresented in this subsection beyond the rejection of political risk as the alternative explanation for the political
cycle in tunneling. Note that we also explored whether the timing of the political cyclenamely, how much cash
is tunneled before and how much cash is tunneled after the electiondepends on the margin of victory or political
turnover; we found no relationship.
20
For example, according to expert estimates the pre-election budget of the United Russia party for the parliamentary election in 2003 was US$250 million, whereas the maximum limit permitted by law was US$8 million. For
details, see The biggest deal on the political market is the current parliamentary elections, in Novaya Gazeta,
September 18, 2003.
21
One possible interpretation of why incumbents do this, despite being sure of wining, is an implicit contract
between them and voters, who expect to get gifts in the month before elections.

23

ticular, concerns tunneling right after the election and is unrelated to election campaigns, was
suggested to the authors in an informal interview with public officials in the Moscow-city administration.22 Many public procurement contracts are allocated in the beginning of the electoral
term for budgeting reasons and, therefore, the beginning of the electoral term is marked with the
signing of new public procurement contracts. Importantly, the expiration dates of the contracts
are scattered throughout the electoral term. (As we can see from Figure 5, there are (insignificantly) fewer contracts being paid out in the first few months of the new electoral term, which is
consistent with the idea that many procurement contracts are signed during this time.) Again,
if illegal tunneling of cash intensifies when public procurement contracts are being signed, this
would suggest that the cash payments are used as bribes for helping to get procurement contracts.
The corruption explanation for the political cycle in tunneling of firms with procurement
contracts has a testable implication: one should observe a particularly strong correlation between
abnormal tunneling around elections and the probability to get procurement contracts in regions
that are notoriously corrupt. In general, the correlation between tunneling around elections
and the probability of obtaining procurement contracts should increase as regional corruption
increases.
In order to test this proposition, we need a measure of corruption that provides variation
within Russia and does not come from our data. The only available measure is the regional-level
Corruption Perception Index by Transparency International. It is available for 40 (of 89) regions
in Russia.23 This is a perception-based index compiled using enterprise-manager surveys; it was
constructed only once in 2002. For simplicity of interpretation of the magnitudes, we take the
z-score of the index, so that the resulting measure has zero mean and unit variance with higher
values indicating higher perceived regional corruption. We use this index to test whether tunneling
is more closely associated with winning public procurement contracts in the regions named most
corrupt by Transparency International.
22

Moscow city (along with the metropolitan area of St. Petersburg) is a Subject of the Federation, a status
equal to the rest of Russias regions.
23
The index is available at: http://www.anti-corr.ru/rating regions/index.htm (accessed on November 27, 2014).

24

Henceforth, we consider a firm following a particular regional election episode as the unit of
analysis. Thus, our cross-sectional sample consists of all legitimate firms in the year following the
regional election. We estimate the relationship between the probability of obtaining a procurement
contract or the size of the procurement revenue in a year following a particular regional election
by a particular regional firm and tunneling activity of that firm during the preceding elections
(controlling for tunneling outside the narrow election window). Our main focus is whether this
relationship is stronger in regions that a priori are considered more corrupt. More precisely, we
estimate the following two equations:
P rob[

P rocRevf e
> 1%] = 1 L(T unnelfElection
) + 2 L(T unnelfElection
) CP If +
e
e
Revf
+ 3 L(T unnelfNeo Election ) + 4 log(Revf ) + 50 Xf + e + f e (7)

and
P rocRevf e
= 1 L(T unnelfElection
) + 2 L(T unnelfElection
) CP If +
e
e
Revf
+ 3 L(T unnelfNeo Election ) + 4 log(Revf ) + 50 Xf + e + f e . (8)
These two equations differ only in their dependent variable. Equation 7 is a linear probability
model with the dummy, indicating firms that received at least 1% of revenue from public procurement contracts in the year following a particular election e, i.e., firms with

P rocRevf e
Revf

> 1%, as

a dependent variable. (As shown in Table A.3, this dummy equals one in 5.02% of observations.
In the online appendix, we also present results for the 5%, instead of the 1%, threshold.) The
second equation considers the size of procurement revenue normalized by the firms revenue as a
dependent variable. Our aim is to assess how different the correlation is between the procurement
revenue received by the firm and its tunneling activity during the election for regions with high
and low corruption. Thus, our main explanatory variables of interest are the amount of cash
tunneled out of firm f during election e and its interaction with Transparency Internationals
regional Corruption Perception Index (CP I) for the region where firm f is located.

25

T unnelfElection
denotes the average weekly transfer by firm f to fly-by-night firms within the
e
window of [1; +1] months from the election date e in the region where firm f is located.
is nominal tunneling in USD (in constant dollars); and its distribution has a logT unnelfElection
e
normal shape. Following MacKinnon and Magee (1990), we use the inverse hyperbolic sine function L(.), such that L(X) = log(X + (X 2 + 1)1/2 ). (We use this transformation for all variables
that can take the value of zero but should be logged.)24 As we are interested in the political cycle
rather than the size of tunneling, we control for tunneling activity per week outside the election
window in the year prior to elections, we denote this variable by L(T unnelfNeo Election ). To control
for the size of the firms revenue, we include log(Revf ) in the list of controls. Xf is a vector of
additional control variables, namely, net income as a share of revenue, the ratio of debt to assets,
industry (sector), and region dummies. All controls are measured in 2003. e is the year fixed
effect, controlling for multiple elections in a particular year. The error term f e is clustered at
the level of regions.
Our main focus is the coefficient 2 on the interaction between tunneling and the CPI. This
coefficient shows whether the correlation between tunneling around elections and winning public
procurement contracts increases with the level of regional perceived corruption. Note that the
CPI does not vary over time, and therefore region fixed effects control for the direct effect of
regional variation in perceived corruption.
The entire sample comprises 45, 275 firms (just as in the previous section) and 63, 021 firmby-election-year observations. However, the CPI is available only for a subset of regions. The
inclusion of this variable decreases the sample size to 38, 044 firms and 54, 115 observations.
Table 3 presents the results. The first two columns report the results for the probability of
securing procurement revenue above 1% of revenue as a dependent variable (as in equation 7), and
the last two columns report the results for the level of procurement revenue as a share of revenue
as a dependent variable (as in equation 8). In columns 1 and 3, we report the average association
24
The results of regression analysis with L(X) are easier interpreted than that with log(1 + X), as the estimated
coefficients are approximately equal to percents, as if log(X) was used, but there were no zero values in X. All
our results are robust to using log(1 + .) instead of L(.) transformations.

26

between tunneling and procurement in the entire sample (which is the cross-sectional analog of
the relationship presented in section 3.2). On average, a 10 percent increase in tunneling during
elections is associated with a nine-percentage-point increase in procurement revenues as a share of
revenues and a 0.1-percentage-point increase in the probability of obtaining procurement contracts
that bring revenue larger than 1% of a firms total revenue. The association between tunneling
outside the election window and procurement revenue is much weaker. The magnitude of the
coefficient on this variable is 1/9th to 1/3rd of that for tunneling around elections. Moreover, it
is statistically significant only for the linear probability model and not for the size of procurement
revenue.
In columns 2 and 4, we add the interaction term between tunneling around elections and
the regional corruption, which is our main focus. The coefficients on the interaction between
corruption level and tunneling are positive and highly statistically significant. We find that, a
one-standard-deviation increase in perceived corruption increases the effect of tunneling around
elections on procurement revenue by about 30%. The results are robust to using the 5% as
opposed to the 1% threshold for procurement revenue as reported in Table A.8 in the online
appendix.
Firms that do and that do not engage in tunneling may have very different characteristics in
terms of size, profitability, industry, and possibly location. Controlling for these characteristics in
an OLS cross-sectional setting, therefore, may not be enough to identify the association between
tunneling around elections and obtaining government procurement as the two groups of firms
may not overlap in terms of their observable characteristics. Thus, in addition to OLS, we use
Propensity Score Matching (PSM) on all observables (i.e., log(Revf ) and Xf ) in order to have
similar comparison groups. As treatment, we consider a dummy for having tunneling around
elections greater than 10% of revenue and estimate the effect of treatment on the probability to
get procurement contracts and on the size of procurement revenue using PSM for three different
samples: all firms, firms in regions with CPI scores above the median, and firms in regions with
CPI scores below the median. Our results are robust to using the Propensity Score Matching

27

estimation method as reported in Tables 4 and A.9. Tunneling is associated with higher probability of getting procurement contracts and higher procurement revenue, on average, and this
association is entirely driven by the large and significant relationship in the subsample of regions
that Transparency International deems more corrupt.
Overall, consistent with our hypothesis, higher corruption is associated with stronger links
between tunneling around elections and obtaining public procurement. This evidence supports
the hypothesis that corruption is the mechanism behind the political cycle in tunneling among
firms with procurement revenue.25

Test of the efficient grease hypothesis

In the previous section, we established that in corrupt regions, firms that tunnel more around
elections are also the ones that obtain government procurement contracts. This suggests that
we can use the strength of the correlation between tunneling around elections and the public
procurement revenue as a proxy for the extent of corruption.

4.1

A proxy for local corruption

Most of procurement contracts are allocated at local (subregional) level.26 We can calculate the
correlation between allocation of procurement contracts and tunneling around elections at any
level of aggregation. However, our aim is to construct a proxy for the variation in corruption across
localities within a region because between-region variation is confounded by a large number of
25

We also analyzed how other regional characteristics affect the magnitude of the political cycle in tunneling
and found no robust correlations between the magnitude of the cycle with any of the observable characteristics of
the regions, with the exception of a positive association between the political cycle and perception-based measures
of regional corruption, which we report in this section.
26
Precise information on the distribution of procurement contracts among levels of government
is available only starting 2005, when the centralized internet portal of government procurement
(http://zakupki.gov.ru/epz/main/public/home.html accessed on November 27, 2014) was launched. 55% of the
total number of public procurement auctions and 43% of the total value of government procurement contracts are
financed by subregional budgets and extra-budgetary funds, i.e., various governmental/municipal organizations
at the local level. The corresponding figures for the federal and regional levels are as follows: 36% of the total
number of contracts and 45% of the value are federal. 8% of the number of contracts and 12% of the value are
regional. Many of the federal procurement contracts, however, are related to defense and defense industry is not
present in Rosstat registry. Thus, the vast majority of the procurement payments that we observe in our sample
are allocated at the subregional level.

28

unobserved political and economic differences between regions.

We define municipalities by the boundaries of subregional tax districts. We use this definition
of localities because it is observable using the codes of firm identifiers. The number of tax districts
varies from two in the smallest region to 35 in the city of Moscow. Considering the extent of
corruption in procurement contracts at the level of locality is meaningful, as this level is equivalent
to towns and boroughs of large metropolitan areas. A lot of public procurement contracts are
allocated at this level. We index localities with subscript l and estimate the specification analogous
to equation 7. However, instead of using the interaction between TI Corruption Perception Index,
we let the coefficients on tunneling in the election window and outside the election window to
vary across localities, i.e., we include a set of interactions between dummies for each locality l
(denoted by Dl ) and tunneling around elections and tunneling far away from elections. We also
control for the unobserved variation across localities (as well as across regions) with full set of
locality dummies:
P rob[

X
P rocRevf e
1l Dl L(T unnelfElection
)+
> 1%] =
e
Revf
l
X
2l Dl L(T unnelfNeo Election ) + 3 log(Revf ) + 40 Xf + 5 Dlf + e + f e . (9)
+
l

In order to estimate 1l precisely, we restrict the sample to localities with at least 40 firms, out
of which at least one has government procurement contracts. Using the results of this estimation,
we define two alternative proxies for local corruption in allocation of public procurement: (1) a
continuous variable equal to the t-statistic of the 1l coefficient in the locality l, denoted by CorlC ;
and (2) a dummy variable, CorlD , which indicates that the estimate of 1l for the locality l is
positive and statistically significant at the 10% level. Both measures are defined for 64 regions
and 246 localities.27 CorlC varies across all localities in each region. CorlD equals zero in every
locality in 44 regions, and there is a within-region variation in CorlD in 20 regions. Overall, the
27

For localities, where there are no private firms with procurement revenue (for instance, because government
procurement contracts are allocated to state-owned or municipal firms), corruption in distribution of procurement
revenue is undefined.

29

coefficients 1l are positive and significant in 26% of localities.28 Summary statistics for these
variables are presented in Table A.10.
The two measures of local corruption are complementary, as the first one has much larger
variation and the second one is more precise (as variation in the continuous measure, when it
is below statistical significance is not meaningful). The correlation coefficient between the two
measures is 0.76. Both measures correlate strongly with the regional CPI, with the correlation
coefficient of 0.45 for the dummy and 0.48 for the continuous measure. Figure 6 presents the
unconditional scatterplot of our continuous proxy for local corruption (i.e., t-statistic for the
coefficient 1l ) and the regional CPI measure along with a nonparametric regression line, relating
the two variables, for the regions where the CPI measure is available. As is evident from the
figure, the two measures are strongly related. The advantages of our proxies for local corruption
over the CPI measure are as follows: first, they are based on objective data; second, they are
available at the subregional rather than the regional level; third, they are available for a much
larger number of regions.
It is important to note that our proxies for local-level corruption are based on the assumption
that firms that get procurement contracts receive revenues from their business activities (unrelated
to government procurement) officially through bank transfers. This is because the data allow us
to measure cash tunneled out of a firms bank holdings, and corruption is associated with black
cash payments. However, if a firm gets its revenues in cash, there is no need for it to tunnel cash
out, as it can under-report the cash revenue and use the extra for bribes. However, only relatively
large, established firms can apply for and win government procurement contracts; therefore, it is
unlikely that these firms receive their revenues in cash. In addition, bribes can be paid at any time,
we consider political cycle in tunneling in order to rule out alternative reasons for tunneling, such
as tax evasion, but tunneling away from election may also be done in order to bribe politicians.
This is because we can not measure the absolute amount of bribes and focus on the differences
in corruption across localities, rather than their absolute level. In the next section, we use our
28

In a few cases, estimates of 1l are negative and significant. However, this is well within statistical errorthis
occurs in 2.8% of localities for significance at the 5% level and in 6.1% of localities for significance at the 10% level.

30

proxies for the variation in local corruption to test the efficient grease hypothesis.

4.2

Does corruption contribute to inefficiency?

With the help of our proxies for local corruption, we can shed light on the efficient grease
hypothesis, which postulates that in more corrupt environments, the more efficient firms can
use bribes to get around red-tape regulations in order to improve their access to government
procurement. We take the best-available to us proxy for the efficiency of firms, namely, a firms
labor productivity measured as the officially reported revenue per worker in 2003, which comes
from the registry.29 This is a cross-sectional measure available only for a subset of the largest
firms, so that the resulting data set consists of 19, 113 firms in localities, where the corruption
measures are defined.
Our aim is to compare the allocation rule for government procurement among local firms
between localities with high and low corruption. Thus, first, we regress a firm-level dummy for
having won government procurement contracts above 1% of a firms revenue in the year following
elections or, alternatively, the size of government procurement revenue, received during the year
following elections and normalized by firms revenue, on the firms log revenue per worker, log(Yf ),
a measure of local corruption, Corl , and their interaction. To be precise, the following equations
are estimated on the sample of 19, 113 firms:
P rob[

P rocRevf e
> 1%] = 1 log(Yf ) Corl + 2 Corl + 3 log(Yf ) + 4 log(Revf ) + 50 Xf + e + f e
Revf
(10)

and
P rocRevf e
= 1 log(Yf ) Corl + 2 Corl + 3 log(Yf ) + 4 log(Revf ) + 50 Xf + e + f e .
Revf

(11)

Corl is one of the two corruption measures of locality l, in which the firm f is located, derived in the
previous subsection. As above, we control for the size of the firm, net income as a share of revenue,
the ratio of debt to assets, as well as sector and region dummies, and we cluster standard errors at
29

Note that the same information on firms efficiency is available to the government officials, who decide how
to allocate government procurement contracts.

31

the regional level. Our main coefficient of interest is 1 . It estimates whether an increase in local
corruption leads to less productive or more productive firms getting government procurement
contracts (holding everything else constant and provided that unobserved heterogeneity is not
correlated with regressors).
As the interaction term between local corruption and labor productivity is identified even if
we control for all unobserved heterogeneity among localities, we also estimate a similar regression adding locality fixed effects, i.e., dummies for each locality Dl , to the list of controls and
suppressing the direct effect of Corl as it is collinear to locality effects:
P rob[

P rocRevf e
> 1%] = 1 log(Yf )Corl +3 log(Yf )+4 log(Revf )+50 Xf +7 Dl +e +f e (12)
Revf

and
P rocRevf e
= 1 log(Yf ) Corl + 3 log(Yf ) + 4 log(Revf ) + 50 Xf + 7 Dl + e + f e .
Revf

(13)

Note that controlling for regional and industry fixed effects is important because a substantial
variation in the number of procurement contracts and efficiency of firms is driven by unobserved
regional and industry-level factors.
Second, as both the observable covariates and the unobserved heterogeneity may interact
with the level of local corruption, we split the sample into two subsamples with CorlD = 0
and CorlD = 1, i.e., with statistically significant and statistically insignificant local corruption,
namely, 1 from equation 9, and compare how the allocation of procurement revenue depends
on productivity between these subsamples. The specifications estimated on the subsamples are
identical to 10 and 11 with the exception of not having log(Yf ) Corl and Corl among the
covariates.
The results of these estimations are presented in Tables 5 in the main text for the full sample
and A.11 in the online appendix for the subsamples. Panels A of both tables report the results
for the linear probability models and Panels B report the results for the continuous dependent
variable. In addition, Table A.12 in the online appendix reports the robustness of the results of

32

Panel A of Table 5 to increasing the threshold for the procurement revenue to 5% of revenue from
1% of revenue in equations 10 and 12.
All these results draw a consistent picture. Let us first discuss the results of the estimation
on the full sample. In Table 5, for comparison, in every odd column, we suppress the interaction
between the local corruption measure and labor productivity. Columns 1, 2, 5, and 6 use the
continuous measure of corruption, i.e., t-statistic for 1 from equation 9; columns 3, 4, 7, and 8
use the dummy, CorlD indicating significance of 1 .
The first finding is that government procurement contracts are, on average, directed to firms
with lower revenue per worker (as the coefficient on labor productivity is negative and statistically
significant irrespective of specification). This result has two alternative interpretations. One
possibility is that the distribution of government procurement contracts is simply inefficient. The
alternative story, however, is that government procurement is distributed in order to support
higher employment for patronage reasons. The latter interpretation does not necessarily imply
lower efficiency as one can imagine two different technologies: one more labor intensive and the
other less labor intensive. (We do not observe capital stock or value added.) A paternalistic
government with an objective of high employment could chose to allocate procurement contracts
to firms with more labor-intensive technology without losing efficiency. Second, we find that, on
average, the amount of procurement revenue allocated to firms in more and less corrupt localities
does not systematically differ, as can be seen from the coefficients on Corl in columns 1 and 3.
Therefore, any differences in the distribution of government procurement among firms between
more and less corrupt localities cannot be driven by the differences in volume of the government
procurement.
The third finding is our main focus: higher level of local corruption is associated with less
productive firms obtaining public procurement contracts in the locality. The coefficients on the
interaction term between the proxy for labor productivity and both of our corruption measures
are negative and significant in all specifications. Importantly, in contrast to the interpretation
of the direct effect of Yf , the interpretation of the effect of the interaction between Yf and Corl

33

has an unambiguous interpretation of an increase in the inefficiency in distribution of government

procurement contracts, provided that one accepts the main identification assumption that unobserved cross-sectional heterogeneity across firms or localities is not driving these results. There
is no alternative interpretation here because in more corrupt localities, by construction of our
corruption proxies, governments are less paternalistic on average as they distribute procurement
contracts in exchange for tunneling of illegal cash (and, therefore, avoiding paying taxes on it)
rather than in exchange for supporting higher levels of employment.30
As equations 10 to 13 are estimated on a cross-section of firms in very different localities,
the identification assumption is demanding. To make sure that the unobserved heterogeneity
and observed covariates do not interact with local corruption in a way that biases our estimates,
we present the results on subsamples of corrupt and non-corrupt localities in Table A.11 in the
online appendix. These results confirm the full-sample estimation: in regressions, the relationship
between government procurement revenues and productivity is more negative in a subsample of
corrupt localities compared to the subsample of non-corrupt localities. Moreover, we also observe
a similar pattern for the officially reported profitability of firms measured by net income per
revenue. Namely, the profitability of firms that get government procurement contracts relative to
that of firms, which do not get procurement contracts, is also lower in corrupt localities (as can
be seen from the comparison of the magnitudes of the coefficients on net income per revenue in
columns (3) vs (4) and (7) vs (8) of the table).
The magnitude of the estimated effects is as follows. We find that in noncorrupt localities
(i.e., those, where the distribution of procurement contracts is not significantly correlated with
tunneling around elections), government procurement contracts are allocated to firms that have
1.3% lower labor productivity compared to firms that do not win public procurement contracts in
30

An important caveat in interpreting these results is that our measures of corruption, Corl , are themselves
statistics calculated from the data. Ideally, standard errors should be bootstraped to take into account the fact
that some regressors have confidence intervals. In practice, a multi-level structure of our data with restrictions
on both firms and localities to get into the sample makes a clean bootstrap procedure impossible to implement
due to a change in the sample between different iterations. Thus, we have to rely on standard errors adjusted to
clusters at the regional level. With these standard errors, the estimates are very precisely estimated irrespective of
which of the two measures of corruption we use, despite the fact that these measures have very different statistical
properties.

34

the same locality and same industry (and are in all other observable ways similar). In contrast,
in corrupt localities (i.e., those, where the distribution of procurement contracts is significantly
related to tunneling around elections), firms with procurement revenue are 2% less productive
compared to nonrecipients of public procurement contracts (as shown in column 8 of Panel A of
Table 5). The size of government procurement revenue increases twice as fast with a decrease in
firms productivity in corrupt localities (as shown in column 8 of Panel B of Table 5).
Our results suggest that corruption is associated with substantial efficiency losses in the allocation of government procurement contracts, as government procurement goes to less efficient firms
in more corrupt localities. Thus, the data are inconsistent with the efficient grease hypothesis
corruption does not lead to an increase in efficiency of distribution of public procurement contracts.
A plausible explanation for this result is that the managerial skills needed to bribe politicians and
needed to run firms efficiently are substitutes rather than complements.

Conclusions

We use objective financial data for a near-population of large Russian firms to establish that firms
that receive government procurement contracts increase illegal tunneling around elections (i.e.,
exhibit a political cycle), whereas tunneling of firms without government procurement contracts
does not exhibit a political cycle. After considering several potential mechanisms driving this
political cycle, we conclude that corruption in allocation of procurement is the main channel:
tunneled cash is likely to be directed to politicians in exchange for allocation of procurement
contracts. In particular, the political cycle in tunneling cannot be explained by a political cycle
in legitimate business activity, by a political cycle in procurement revenue, or by firms increased
political risk associated with political turnover. We find that the political cycle in tunneling
among firms with government procurement contracts is particularly pronounced in notoriously
corrupt regions. Using the correlation between tunneling around elections and the allocation of
procurement as a proxy for local corruption, we test the efficient grease hypothesis and find that
the data are inconsistent with it, as less productive firms are more likely to win public procurement

35

contracts in more corrupt localities and, therefore, the allocation of public procurement is less
efficient under corruption. Our findings suggest that the managerial skills needed to run a firm
efficiently do not help win procurement contracts in corrupt environments; vice versa, political
capital that helps win procurement contracts is not useful in running firms efficiently.

References
Aidt, Toke S., Review: Economic Analysis of Corruption: A Survey, The Economic Journal, 2003,
113 (491), pp. 632652.
Akhmedov, Akhmed and Ekaterina Zhuravskaya, Opportunistic Political Cycles: Test in a Young
Democracy Setting, Quarterly Journal of Economics, 2004, 119 (4), 13011338.
Amore, Mario Daniele and Morten Bennedsen, The value of local political connections in a
low-corruption environment, Journal of Financial Economics, 2013, 110 (2), 387402.
Bardhan, Pranab, Corruption and Development: A Review of Issues, Journal of Economic Literature, September 1997, 35, 13201346.
Bertrand, Marianne, Francis Kramarz, Antoinette Schoar, and David Thesmar, Politically
Connected CEOs and Corporate Outcomes: Evidence from France, 2007. Unpublished manuscript.
, Simeon Djankov, Rema Hanna, and Sendhil Mullainathan, Obtaining a Drivers License
in India: An Experimental Approach to Studying Corruption, The Quarterly Journal of Economics,
November 2007, 122 (4), 16391676.
Braguinsky, Serguey and Sergey Mityakov, Foreign Corporations and the Culture of Transparency: Evidence from Russian Administrative Data, Journal of Financial Economics, Forthcoming.
http://papers.ssrn.com/sol3/papers.cfm?abstract id=1980581.
, , and Andrey Liscovich, Direct Estimation of Hidden Earnings: Evidence From Administrative
Data, 2011. Mimeo, John F. Kennedy School of Government.
Brown, J. David, John S. Earle, and Almos Telegdy, The Productivity Effects of Privatization:
Longitudinal Estimates from Hungary, Romania, Russia, and Ukraine, Journal of Political Economy,
February 2006, 114 (1), 6199.
Burgess, Robin, Matthew Hansen, Benjamin A. Olken, Peter Potapov, and Stefanie Sieber,
The Political Economy of Deforestation in the Tropics, Quarterly Journal of Economics, 2012, 127
(4), 17071754.
Butler, Alexander W., Larry Fauver, and Sandra Mortal, Corruption, Political Connections,
and Municipal Finance, Review of Financial Studies, July 2009, 22 (7), 26732705.
Caselli, Francesco and Guy Michaels, Do Oil Windfalls Improve Living Standards? Evidence from
Brazil, NBER Working Papers 15550, National Bureau of Economic Research, Inc December 2009.
Cheung, Yan-Leung, Raghavendra Rau, and Aris Stouraitis, Which Firms Benefit from Bribes,
and by How Much? Evidence From Corruption Cases Worldwide, 2011. Mimeo, Hong Kong Baptist
University.
Cingano, Federico and Paolo Pinotti, Politicians at Work: The Private Returns and Social Costs
of Political Connections, Journal of the European Economic Association, 2013, 11 (2), 433465.
Desai, Mihir A., Alexander Dyck, and Luigi Zingales, Theft and taxes, Journal of Financial
Economics, June 2007, 84 (3), 591623.

36

Drazen, Allan, The Political Business Cycle After 25 Years, in NBER Macroeconomics Annual
2000, Volume 15 NBER Chapters, National Bureau of Economic Research, Inc, 2001, pp. 75138.
Ferraz, Claudio and Frederico Finan, Exposing Corrupt Politicians: The Effects of Brazils Publicly
Released Audits on Electoral Outcomes, The Quarterly Journal of Economics, 2008, 123 (2), 703745.
and , Electoral Accountability and Corruption: Evidence from the Audits of Local Governments,
American Economic Review, 2011, 101 (4), 12741311.
Fisman, Raymond and Edward Miguel, Corruption, Norms, and Legal Enforcement: Evidence
from Diplomatic Parking Tickets, Journal of Political Economy, December 2007, 115 (6), 10201048.
Goldman, Eitan, Jorg Rocholl, and Jongil So, Politically Connected Boards of Directors and The
Allocation of Procurement Contracts, Review of Finance, 2013, 17 (5), 16171648.
Guriev, Sergei and Andrei Rachinsky, The Evolution of Personal Wealth in the Former Soviet
Union and Central and Eastern Europe, Working Papers RP2006/120, World Institute for Development Economic Research (UNU-WIDER) 2006.
Huntington, Samuel P., Political Order in Changing Societies, New Haven: Yale University Press,
1968.
Johnson, Simon, Rafael La Porta, Florencio Lopez de Silanes, and Andrei Shleifer, Tunneling, American Economic Review, May 2000, 90 (2), 2227.
Leff, Nathaniel H., Economic Development through Bureaucratic Corruption, The American Behavioral Scientist, 1964, 8 (2), 814.
Levitt, Steven D. and Sudhir Alladi Venkatesh, An Economic Analysis Of A Drug-Selling GangS
Finances, The Quarterly Journal of Economics, August 2000, 115 (3), 755789.
MacKinnon, James G and Lonnie Magee, Transforming the Dependent Variable in Regression
Models, International Economic Review, May 1990, 31 (2), 31539.
Mironov, Maxim, Taxes, Theft, and Firm Performance, Journal of Finance, 08 2013, 68 (4), 1441
1472.
, Should One Hire a Corrupt CEO in a Corrupt Country?, Journal of Financial Economics, Forthcoming. http://papers.ssrn.com/sol3/papers.cfm?abstract id=2023988.
Olken, Benjamin A., Corruption and the Costs of Redistribution: Micro Evidence from Indonesia,
Journal of Public Economics, May 2006, 90 (4-5), 853870.
, Monitoring Corruption: Evidence from a Field Experiment in Indonesia, Journal of Political
Economy, 2007, 115, 200249.
and Rohini Pande, Corruption in Developing Countries, Annual Review of Economics, 07 2012,
4 (1), 479509.
Reinikka, Ritva and Jakob Svensson, Local Capture: Evidence from a Central Government Transfer Program in Uganda, Quarterly Journal of Economics, 2004, 119 (2), 679705.
and , Using Micro-Surveys to Measure and Explain Corruption, World Development, February
2006, 34 (2), 359370.
Rose-Ackerman, Susan, Corruption: Causes, Consequences, and Reform, Cambridge: Cambridge
University Press, 1999.
Shleifer, Andrei and Robert W. Vishny, Corruption, Quarterly Journal of Economics, August
1993, 108 (3), 599617.
and , Politicians and Firms, Quarterly Journal of Economics, 1994, 109 (4), 9951025.
Svensson, Jakob, Eight Questions about Corruption, Journal of Economic Perspectives, 2005, 19
(3), 1942.
Tella, Rafael Di and Ernesto Schargrodsky, The Role of Wages and Auditing during a Crackdown
on Corruption in the City of Buenos Aires, Journal of Law and Economics, 2003, 46 (1), 26992.

37

Figure 1: Tunneling by Inteko, the company of the wife of Moscow mayor

Elelection of Moscow mayor

Dec 7, 2003

1000

2000

3000

Presidential election
Mar 14, 2004

01oct2004

01sep2004

01aug2004

01jul2004

01jun2004

01may2004

01apr2004

01mar2004

01feb2004

01jan2004

01dec2003

01nov2003

01oct2003

01sep2003

01aug2003

01jul2003

01jun2003

01may2003

Weekly transfer, 000 USD

4000

Transfers to flybynight firms out of Inteko, the company of Moscow mayors wife

dat

Note: The figure portrays the amount of money transferred from Inteko to fly-by-night firms.

38

Figure 2: The political cycle in tunneling for firms with procurement revenue above and below
1% of revenue
1
2
The political cycle in tunneling of firms with procurement above 1% of revenue, m
+ m
:
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

!0.40%
!0.60%

2
:
The political cycle in tunneling of firms with procurement below 1% of revenue, m
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

4%

5%

6%

!0.40%
!0.60%

1
:
The difference between the two, m
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

!0.40%
!0.60%

Note: Units of the y-axes: weekly transfers to fly-by-night firms in percentage of annual revenue multiplied by
1
2
1000. The figure portrays coefficient estimates of m
, m
and their sum from the estimation of equation 1 along
with their 95% confidence intervals. The x-axes portray the number of months m from the election date. Full
regression output is presented in Panel A of Table 1.

39

Figure 3: The political cycle in tunneling per procurement revenue

1
:
The political cycle in tunneling per additional unit of procurement revenue, m
4.50&
3.50&
2.50&
1.50&
0.50&
!8&

!7&

!6&

!5&

!4&

!3&

!2&

!1&
!0.50&

1&

2&

3&

4&

5&

6&

!1.50&

2
The political cycle in tunneling of firms with zero procurement revenue, m
:
4.50&
3.50&
2.50&
1.50&
0.50&
!8&

!7&

!6&

!5&

!4&

!3&

!2&

!1&
!0.50&

1&

2&

3&

4&

5&

6&

!1.50&

Note: Units of the y-axes: weekly transfers to fly-by-night firms in percentage of annual revenue multiplied by
1
2
1000. The figure portrays coefficient estimates of m
, m
from the estimation of equation 2 along with their 95%
confidence intervals. The x-axes portray the number of months m from the election date. Full regression output
is presented in Panel B of Table 1.

40

Figure 4: The political cycle in tunneling vs. regular business activity for firms with procurement
revenue above 1% of revenue

1.00%
0.80%

Transfers%to%legi9mate%rms%

0.60%

Tunnelling%

0.40%
0.20%

!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%

1%

2%

3%

4%

5%

6%

!0.20%
!0.40%

Firms
Proc.rev > 1% of rev.:

F -test for cycle

( + )in[1;1] = ( 1 + 2 )out[1;1]

Business activity
F = 2.06
p = 0.15

Tunnelling
F = 24.07
p = 0.000

Proc.rev < 1% of rev.:

2
2
= out[1;1]
in[1;1]

F = 0.48
p = 0.49

F = 18.76
p = 0.000

1
2
Note: The figure portrays the dynamics of total transfers, i.e., m
+ m
from the estimation of equation 1, taking
the transfers to legitimate nonmedia firms (i.e., regular business activity) as dependent variables. We also
present the 95% confidence interval on these coefficients. The number of observations is 6,286,011. The number
of firms is 45,275. The number of regions (clusters) is 87. The y-axis portrays weekly transfers in percentage of
annual revenue multiplied by 1000. The x-axis portrays the number of months m from the election date. (For
comparison, we also present dynamics of transfers to fly-by-night firms (tunneling). The confidence interval for
these estimates is presented in Figure 2.) The table below the figure presents F -statistics for the political-cycle
test in respective transfers among firms with public procurement revenue above and below 1% of their total
revenue.

41

Figure 5: The absence of a political cycle in incoming procurement revenue

1.00%
0.80%
0.60%
0.40%
0.20%

!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

!0.40%

Note: Units of the y-axis: weekly incoming procurement revenue in percentage of annual revenue multiplied by
1
1000. The figure portrays coefficient estimates of m
from equation 4 along with their 95% confidence intervals.
The number of observations is 6,286,011. The number of firms is 45,275. The number of regions (clusters) is 87.
The x-axis portrays the number of months m from the election date.

Figure 6: Local corruption proxy and Transparency Internationals Corruption Perception Index

-1.5

TI Corruption Perception Index (CPI)

-1
-.5
0
.5

Lowess smoother

-4

-2

0
2
Local corruption proxy (t-statistic)

bandwidth = .8

42

Table 1: The political cycle in tunneling

Dependent variable: Transfers to fly-by-night firms as a share of revenue
Panel A:
Estimation results of equation 1
Month from Coef. Std. Err. t-stat Coef. Std. Err.
t-stat
1
2
election, m
m
m
-8
-0.016
0.031
-0.52 -0.214
0.050
-4.29
-7
0.001
0.033
0.03 -0.030
0.074
-0.41
-6
0.186
0.019
9.97 0.044
0.095
0.46
-5
0.087
0.019
4.48 0.062
0.092
0.68
-4
0.061
0.029
2.07 0.097
0.054
1.8
-3
0.182
0.039
4.66 0.279
0.104
2.68
-2
0.090
0.041
2.17 0.205
0.087
2.37
-1
0.137
0.021
6.57 0.221
0.060
3.69
1
0.471
0.102
4.6
0.358
0.097
3.69
2
-0.060
0.018
-3.32 0.009
0.064
0.13
3
-0.022
0.022
-0.99 0.095
0.036
2.61
4
-0.005
0.023
-0.2
0.129
0.027
4.73
5
-0.012
0.034
-0.35 0.116
0.036
3.22
6
-0.109
0.028
-3.95 -0.252
0.213
-1.18
1
2
F -stat
jointly
154 jointly
10
2
2
1
1
19
6= out[1;1]
26
in[1;1]
6= out[1;1]
F -stat
in[1;1]
Panel B:
Month from
election, m
-8
-7
-6
-5
-4
-3
-2
-1
1
2
3
4
5
6
F -stat
F -stat

Estimation results of equation 2

Coef. Std. Err. t-stat Coef. Std. Err.
1
2
m
m
0.101
0.216
0.47 -0.216
0.052
0.240
0.142
1.69 -0.032
0.074
1.167
0.414
2.82 0.046
0.100
0.727
0.109
6.68 0.062
0.092
0.107
0.232
0.46 0.099
0.055
1.173
0.188
6.25 0.281
0.105
1.061
0.172
6.18 0.202
0.083
1.378
0.262
5.26 0.218
0.057
2.984
0.737
4.05 0.365
0.106
-0.364
0.113
-3.22 0.008
0.065
-0.316
0.106
-2.99 0.095
0.036
0.214
0.078
2.74 0.126
0.027
-0.104
0.224
-0.46 0.116
0.036
0.039
0.246
0.73 -0.261
0.219
1
2
jointly
20
jointly
1
1
2
2
in[1;1]
6= out[1;1]
11
in[1;1]
6= out[1;1]

t-stat
-4.18
-0.44
0.46
0.68
1.8
2.67
2.43
3.8
3.46
0.12
2.61
4.72
3.25
-1.19
9
17

Note: The table presents the results of the estimation of equation 1 in Panel A and equation 2 in Panel B.
Coefficients and standard errors are multiplied by 1000. Standard errors are corrected for clusters at the regional
level. The number of observations is 6,286,011. The number of firms is 45,275. The number of regions (clusters)
is 87.

43

44

Note: The table presents the results of the estimation of equation 5 in Panel A and equation 6 in Panel B. Coefficients and standard errors are
multiplied by 1000. Standard errors corrected for clusters at the regional level are in parentheses. Election window is a dummy indicating a window
of two months: one month before and one month after the elections, [-1;1]. Proc. rev. share > 1% is a dummy indicating that procurement
revenue is above 1% of a firms revenue. Proc. rev. share is the share of a firms revenue that comes from procurement contracts. * significant at
10%, ** significant at 5%, *** significant at 1% level.

Dependent variable:
Transfers to fly-by-night firms as a share of revenue
Subsample Winner, the 1st round:
Incumbent:
Incumbent, the 1st round:
of elections:
<50%
>50%
lost
won
<70%
>70%
Panel A:
Estimation of equation 5
(Proc. rev. share > 1%) Election window
-0.004
0.301
0.089
0.301
0.148
0.275
(0.032)
(0.035)*** (0.039)** (0.038)*** (0.066)**
(0.031)***
Election window
0.087
0.300
0.004
0.290
0.106
0.411
(0.079)
(0.061)***
(0.087)
(0.064)*** (0.044)**
(0.113)***
Number of obs 1,833,618 4,452,393
744,847
4,864,401 2,917,418
2,691,830
Number of firms
16,980
33,817
7,460
38,665
25,833
26,521
Number of regions
45
57
27
68
24
63
Mean dependent variable
0.320
0.712
0.325
0.678
0.931
0.354
Panel B:
Estimation of equation 6
Procurement rev. share Election window
0.204
1.507
0.255
1.530
0.686
1.197
(0.101)** (0.257)*** (0.072)*** (0.256)*** (0.351)*
(0.21)***
Election window
0.086
0.309
-0.002
0.299
0.104
0.426
(0.08)
(0.065)***
(0.088)
(0.068)*** (0.045)**
(0.117)***
Number of obs 1,833,618 4,452,393
744,847
4,864,401 2,917,418
2,691,830
Number of firms
16,980
33,817
7,460
38,665
25,833
26,521
Number of regions
45
57
27
68
24
63
Mean dependent variable
0.320
0.712
0.325
0.678
0.931
0.354

Table 2: The political cycle in tunneling, the margin of victory, and political turnover

45

Dummy: Proc. rev. > 1% revenue Procurement revenue/Revenue

(1)
(2)
(3)
(4)
0.011
0.010
0.932
0.924
(0.003)***
(0.002)***
(0.265)***
(0.175)***
0.003
0.251
(0.001)***
(0.074)***
0.003
0.003
0.119
0.070
(0.001)***
(0.001)***
(0.107)
(0.096)
-0.005
-0.005
-0.392
-0.448
(0.004)
(0.004)
(0.296)
(0.313)
-0.034
-0.031
-3.952
-4.247
(0.008)***
(0.008)***
(0.795)***
(0.969)***
-0.027
-0.028
-1.954
-1.845
(0.007)***
(0.009)***
(0.484)***
(0.582)***
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.071
0.069
0.057
0.056
63,021
54,115
63,021
54,115
45,275
38,044
45,275
38,044
87
40
87
40

Note: The table presents the results of the estimation of equation 7 in column 2 and equation 8 in column 4. Standard errors corrected for clusters
at the regional level. TI CPI corruption is a z-score of the Transparency International regional corruption perception index. * significant at 10%,
** significant at 5%, *** significant at 1% level.

Industry dummy
Region dummy
Election year dummy
R-sq
Number of obs
Number of firms
Number of regions

Debt/Assets

Net Income/Revenue

log(Revenue)

L(Tunneling/week), outside election window

TI CPI Corruption L(Tunneling / week), election window

L(Tunneling/week), election window

Dependent variable:

Table 3: Procurement contracts and political cycle in tunneling in more and less corrupt regions

46

Dummy: Procurement revenue > 1% revenue

All
TI CPIMedian TI CPI>Median
(1)
(2)
(3)
0.032
0.008
0.032
(0.005)***
(0.005)
(0.007)***
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.005
0.000
0.003
62,874
26,161
27,838
45,128
19,746
18,206

Procurement revenue/Revenue
All
TI CPIMedian TI CPI>Median
(4)
(5)
(6)
2.828
0.688
3.650
(0.489)***
(0.549)
(0.655)***
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.0043
0.0000
0.0046
62,874
26,161
27,838
45,128
19,746
18,206

Note: Propensity score estimates. The firms are matched on size of revenue, net income as a share of revenue, the ratio of debt to assets, sector, and
region. Standard errors corrected for clusters at the regional level are in parentheses.TI CPI is the Transparency Internationals regional
corruption perception index. * significant at 10%, ** significant at 5%, *** significant at 1% level.

log(Revenue)
Net Income/Revenue
Debt/Assets
Industry dummy
Region dummy
Election year dummy
R-sq
Number of obs
Number of firms

(Tunneling)/(Revenue) > 0.1

Dependent variable:
Sample:

Table 4: Propensity score estimates of the relationship between tunneling around elections and procurement
in more and less corrupt regions

47
t-stat for 1
(1)
(2)
-1.011
-0.791
(0.144)*** (0.182)***
0.011
0.386
(0.097)
(0.206)*
-0.122
(0.046)***
Yes
Yes
Yes
Yes
No
No
Yes
Yes
0.05
0.05
29,302
29,302
19,113
19,113
64
64

-0.002
(0.001)***
Yes
No
Yes
Yes
0.07
29,302
19,113
64

Yes
No
Yes
Yes
0.06
29,302
19,113
64

-0.663
(0.171)***
Yes
No
Yes
Yes
0.06
29,302
19,113
64

dummy: p-value< 10%

(7)
(8)
-1.019
-0.689
(0.16)***
(0.272)**

Yes
No
Yes
Yes
0.07
29,302
19,113
64

-0.007
(0.002)***
Yes
No
Yes
Yes
0.07
29,302
19,113
64

> 1 % of revenue
for 1
dummy: p-value< 10%
(6)
(7)
(8)
-0.013
-0.015
-0.013
(0.003)*** (0.003)*** (0.003)***

Procurement revenue/revenue
dummy: p-value< 10%
t-stat for 1
(3)
(4)
(5)
(6)
-1.010
-0.712
-1.019
-0.774
(0.144)*** (0.259)*** (0.16)*** (0.189)***
0.234
2.040
(0.217)
(0.656)***
-0.598
-0.137
(0.171)***
(0.045)***
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
0.05
0.05
0.06
0.06
29,302
29,302
29,302
29,302
19,113
19,113
19,113
19,113
64
64
64
64

Dummy: government procurement revenue

t-stat for 1
dummy: p-value< 10%
t-stat
(1)
(2)
(3)
(4)
(5)
-0.015
-0.013
-0.015
-0.013
-0.015
(0.002)*** (0.003)*** (0.002)*** (0.003)*** (0.003)***
0.002
0.007
0.003
0.019
(0.001)
(0.002)***
(0.005)
(0.012)
-0.002
-0.005
(0.001)***
(0.003)**
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
0.06
0.06
0.06
0.06
0.07
29,302
29,302
29,302
29,302
29,302
19,113
19,113
19,113
19,113
19,113
64
64
64
64
64

Note: Panel A presents the results of the estimation of equations 10 and 12; Panel B of equations 11 and 13. Standard errors corrected for clusters
at the regional level are in parentheses. Election year fixed effects and industry fixed effects are included as controls in all specifications. Local
corruption stands for one of the two measures of corruption constructed from the association between tunneling around elections and the
distribution of public procurement: the size of the t-statistic or the dummy for p-value<10%. Coefficients and standard errors are multiplied by 1000
in Panel B. * significant at 10%, ** significant at 5%, *** significant at 1% level.

Industry dummy, election dummy

Region dummy
Locality dummy
log(Revenue); Net Income/Revenue; Debt/Assets
R-sq
Number of obs
Number of firms
Number of regions

Log labor productivity Local corruption

Local corruption

Log labor productivity

Panel B: Dependent variable is continuos.

Dependent var:
Local corruption measure:

Industry dummy, election dummy

Region dummy
Locality dummy
log(Revenue); Net Income/Revenue; Debt/Assets
R-sq
Number of obs
Number of firms
Number of regions

Log labor productivity Local corruption

Local corruption

Log labor productivity

Panel A: Dependent variable is a dummy.

Dependent var:
Local corruption measure:

Table 5: Efficiency in allocation of government procurement and corruption

48

479,169
460,156
443,196
435,288
411,494

99.23%
31.18%
24.28%
23.88%
19.20%

100.00%

% of total value
of all transactions

business legal entities:

9.76%
9.94%
5.80%
2.11%
8.34%
7.15%

27.83%
26.72%
25.74%
25.28%
23.90%

Number of legal
% of total
entities
number of legal
entities
1,721,890
100.00%

Out of the 411,494 remaining domestic, non-financial, non-individual

Legitimate firms
168,009
Fly-by-night firms
99,925
Semi-legal (intermediate) firms
143,560

Panel B

Entire banking transactions data base

After dropping IDs with fewer than 10 transactions
or less than 100,000 RUR per month in revenue
After dropping financial legal entities
After dropping government and municipal legal entities
After dropping foreign legal entities
After dropping individual entrepreneurs

Panel A

Table A.1: Sample construction out of the banking transactions data

Online Appendix

49

Average revenue per firm

in the sample, $1000

Percentage of revenue
of firms in the sample
to all firms in Rosstat, $1000
All firms in Rosstat with revenue in 2003 > $100,000
193,399
4,281
100.0%
All firms in Rosstat with revenue in 2003 > $1 million
63,478
12,522
96.0%
Our sample of legitimate firms
45,275
14,246
77.9%
Note: Our sample of legitimate firms consists of all firms that satisfy the following criteria: had revenue in 2003 > $1 million,
are present among legitimate firms in the banking transactions data, have some transactions in the interval +/- one year from
regional elections.

Number of firms

Table A.2: Intersection of the Rosstat and banking transactions data sets

50

Sample: firms elections

Mean
(1)
Revenue 2003, $000s $14,246
Assets 2003, $000s $13,113
Net Income 2003, $000s
$ 784
Net Income / Revenue 2003, %
1.95
Debt / Assets 2003, %
4.37
Annualized transfers to fly-by-night firms 1999-2004, $000s
$ 272
Annualized transfers to fly-by-night firms 1999-2004 / Revenue 2003, %
4.75
Annualized revenue from procurement 1999-2004, $000s
$ 58
Annualized revenue from procurement 1999-2004 / Revenue 2003, %
0.37
Log(1+procurement revenue)
0.572
Proc. Rev >0, %
17.11
Proc. Rev >1% of revenue, %
5.02
Proc. Rev >5% of revenue, %
2.24
L(Tunneling / week, election window), 1999-2004
0.758
L(Tunneling / week, outside election window), 1999-2004
0.828
TI CPI Corruption
0.000
TI CPI Political Corruption
0.032
Labor productivity (revenue per employee)
32.551
Log labor productivity
2.873
Sample: firms weeks
Mean
(1)
Tunneling per week / Revenue
0.001
Procurement revenue / Revenue, %
0.530
Proc. Rev >1% of revenue, %
5.747
Proc. Rev >5% of revenue, %
2.387

Table A.3: Summary statistics

St. dev.
(3)
$ 105,660
$ 173,933
$ 21,760
9.38
14.83
$ 5,909
8.38
$ 1,548
1.99
1.602
37.66
21.84
14.78
1.313
1.232
1.000
0.030
58.625
0.966
St. dev.
(3)
0.003
3.386
23.274
15.265

Median
(2)
$ 2,852
$ 1,317
$ 26
0.81
0.00
$ 36
1.36
$0.00
0.000
0.00
0.00
0.00
0.000
0.125
0.457
0.031
15.008
2.709
Median
(2)
0.000
0.000
0.000
0.000

N of obs
(4)
6286011
6286011
6286011
6286011

N of obs
(4)
63021
63008
62804
63021
63021
63021
63021
63021
63021
63021
63021
63021
63021
63021
63021
54115
44787
44532
44532
N of firms
(5)
45275
45275
45275
45275

N of firms
(5)
45275
45264
45102
45275
45275
45275
45275
45275
45275
45275
45275
45275
45275
45275
45275
38044
30842
31101
31101

Table A.4: Summary statistics by region

Region
Adygeya republic
Bashkortostan republic
Buryat republic
Altai republic
Ingush republic
Kabardino-Balkar republic
Kalmyk republic
Karachaevo-Cherkess republic
Karelia republic
Komi republic
Mari-El republic
Mordovia republic
Sakha (Yakutia) republic
North Osetiya republic
Tatarstan republic
Tuva republic
Udmurtia Republic
Khakasia republic
Chuvash republic
Altai krai
Krasnodar krai
Krasnoyarsk krai
Primorskii krai
Stavropol krai
Khabarovsk krai
Amur oblast
Arkhangelsk oblast
Astrakhan oblast
Belgorod oblast
Bryansk oblast
Vladimir oblast
Volgograd oblast
Vologda oblast
Voronezh oblast
Ivanovo oblast
Irkutsk oblast
Kaliningrad oblast
Kaluga oblast
Kamchatka oblast
Kemerovo oblast
Kirov oblast
Kostroma oblast
Kurgan oblast
Kursk oblast
Leningrad oblast
Lipetsk oblast
Magadan oblast
Moscow oblast
Murmansk oblast
Nizhny Novgorod oblast
Novgorod oblast
Novosibirsk oblast
Omsk oblast
Orenburg oblast
Oryol oblast
Penza oblast
Perm oblast
Pskov oblast
Rostov oblast
Ryazan oblast
Samara oblast
Saratov oblast
Sakhalin oblast
Sverdlovsk oblast
Smolensk oblast
Tambov oblast
Tver oblast
Tomsk oblast
Tula oblast
Tyumen oblast
Ulyanovsk oblast
Chelyabinsk oblast
Chita oblast
Yaroslavl oblast
Moscow city
St. Petersburg city
Evrei autonomous oblast
o/w Aginsk Buryat autonomous
o/w Komi-Permyak autonomous
o/w Koryak autonomous okrug
o/w Nenets autonomous okrug
Taimyr autonomous okrug
o/w Ust-Ordyn Buryat autonom
Khanty-Mansi autonomous okru
Chukotka autonomous okrug
Evenki autonomous okrug
Yamalo-Nenets autonomous okr

Election
Date 1
(1)
2002-01-13
2003-12-07
2002-06-23
2001-12-16
2002-04-07
2002-01-13
2002-10-20
1999-04-25
2002-04-28
2001-12-16
2000-12-03
2003-02-16
2001-12-23
2002-01-20
2001-03-25
2002-03-17
2000-10-15
2000-12-24
2001-12-16
2000-03-26
2000-12-03
2002-09-08
1999-12-19
2000-12-03
2000-12-10
2001-03-25
2000-12-03
2000-12-03
1999-05-30
2000-12-10
2000-12-10
2000-12-24
1999-12-19
2000-12-24
2000-12-03
2001-07-29
2000-11-05
2000-11-12
2000-12-03
2001-04-22
2000-03-26
2000-12-10
2000-11-26
2000-10-22
1999-09-19
2002-04-12
2000-11-05
1999-12-19
2000-03-26
2001-07-15
1999-09-05
1999-12-19
1999-09-05
1999-12-19
2001-10-28
2002-04-12
2000-12-03
2000-11-12
2001-09-23
2000-12-03
2000-07-02
2000-03-26
2000-10-22
1999-08-29
2002-05-19
1999-12-19
1999-12-19
1999-09-19
2001-04-08
2001-01-14
2000-12-24
2000-12-24
2000-10-29
1999-12-19
1999-12-19
2000-05-14
2000-03-26
2000-10-29
2000-12-03
2000-12-03
2001-01-14
2001-01-28
2000-11-19
2000-03-26
2000-12-24
2001-04-08
2000-03-26

Election
Date 2
(2)

2003-08-17

2004-12-19

2004-03-14
2004-12-26
2004-03-14
2004-03-14
2001-05-27
2004-12-19
2004-03-14
2004-12-05
2003-05-25
2004-12-05
2004-12-05
2003-12-07
2004-03-14

2004-03-14
2004-12-05
2003-12-07
2004-11-28
2003-09-21
2003-02-02
2003-12-07
2004-03-14
2003-09-07
2003-12-07
2003-09-07
2003-12-07

2004-11-14
2004-03-14

2003-12-07
2003-09-07
2003-12-07
2003-12-07
2003-09-21

2004-12-05
2004-03-14
2003-12-07
2003-12-07
2003-09-21

2004-03-14
2003-01-26
2004-11-14

N of obs

N of firms

(3)
38
932
125
201
11
44
48
76
209
205
236
173
202
60
578
15
622
133
215
603
2,440
564
697
379
827
117
431
253
577
442
295
823
591
900
147
552
129
584
207
395
506
99
146
147
760
288
167
5,271
381
855
266
1,215
605
446
199
243
343
237
957
494
508
253
404
1,993
234
238
482
371
337
259
445
496
169
807
21,321
4,362
10
11
6
16
13
31
6
389
22
5
132

(4)
38
932
125
201
11
44
48
53
209
205
148
173
202
60
578
15
429
88
215
483
1,589
564
435
379
532
117
288
147
439
262
295
513
403
685
147
552
129
324
125
395
334
99
102
147
546
288
108
3,176
277
855
164
881
428
311
199
243
343
155
957
284
508
253
252
1,387
234
172
331
319
337
259
291
496
109
560
13,783
2,890
10
11
6
10
13
17
5
389
22
5
132

51

Revenue,
$000s
(5)
4,099
15,678
8,717
23,178
47,930
4,984
20,906
7,224
6,544
17,385
4,430
8,005
18,535
6,876
20,799
3,473
10,478
9,736
6,603
6,771
8,523
8,813
9,700
7,685
9,009
9,200
8,350
9,142
12,281
6,978
8,596
12,001
18,131
6,306
6,317
9,928
13,540
7,670
7,706
18,723
6,636
7,315
9,790
10,561
13,045
18,278
6,402
11,006
14,717
14,124
9,120
11,071
21,003
16,233
14,016
5,002
18,899
6,844
8,156
9,575
25,800
14,651
5,866
14,089
8,932
6,081
7,244
13,391
9,795
24,286
18,284
20,278
7,750
8,957
17,592
11,810
4,127
27,739
3,683
20,500
15,570
294,062
1,684
70,841
12,309
6,423
61,224

Paym. to sp.
/ Revenue, %
(6)
0.83
2.10
2.02
6.04
2.97
2.07
9.59
2.87
0.93
2.88
1.93
1.48
2.81
5.23
2.71
1.19
1.49
1.14
1.86
1.62
3.00
2.57
1.52
2.37
2.09
3.23
2.19
2.08
2.65
3.25
2.30
2.52
1.87
2.20
1.67
2.31
1.79
4.07
1.47
1.86
1.61
2.23
1.37
1.97
1.13
2.18
2.21
6.76
1.70
2.65
1.53
2.36
2.26
1.76
2.26
2.32
1.80
2.66
2.91
2.51
1.65
2.62
1.78
1.78
4.00
1.43
3.46
2.40
3.06
2.46
2.57
1.94
2.23
2.06
8.38
2.42
1.29
4.59
0.73
1.86
4.35
5.43
0.56
3.14
3.55
5.34
2.48

Gov orders
/ Revenue, %
(7)
0.00
0.09
0.09
0.31
0.00
0.00
0.45
0.05
0.04
0.00
0.11
0.13
0.16
0.00
0.23
0.00
0.20
0.00
0.15
0.25
0.13
0.10
0.13
0.19
0.14
0.01
0.11
0.09
0.12
0.32
0.52
0.05
0.06
0.16
0.98
0.08
0.15
0.52
0.17
0.05
0.14
0.47
0.35
0.26
0.03
0.09
0.19
0.64
0.15
0.28
0.13
0.16
0.10
0.04
0.24
0.20
0.14
0.37
0.19
0.28
0.14
0.25
0.02
0.09
0.25
0.51
0.29
0.00
0.36
0.07
0.11
0.22
0.10
0.23
0.62
0.30
0.18
0.23
0.00
0.09
0.00
0.15
0.00
0.06
1.04
0.01
0.10

Gov01,
%
(8)
0.00
1.29
0.80
3.48
0.00
0.00
4.17
1.32
0.96
0.00
1.27
2.31
1.49
0.00
2.25
0.00
2.25
0.00
1.40
2.32
1.52
1.06
1.15
2.37
1.57
0.00
0.70
1.58
1.39
2.94
5.42
0.73
0.85
1.33
8.16
0.91
3.88
5.31
1.93
0.51
2.17
4.04
2.05
3.40
0.79
1.39
2.40
10.09
1.57
3.16
0.75
1.89
0.83
0.45
5.53
3.70
2.33
3.38
2.30
2.23
1.38
2.77
0.50
1.25
2.99
5.04
3.11
0.00
4.75
1.93
1.35
2.02
1.78
3.35
8.86
3.26
10.00
18.18
0.00
0.00
0.00
3.23
0.00
0.77
9.09
0.00
1.52

Gov05,
%
(9)
0.00
0.54
0.80
1.99
0.00
0.00
2.08
0.00
0.48
0.00
0.42
0.58
0.99
0.00
1.21
0.00
1.29
0.00
1.40
1.66
0.90
0.53
0.72
1.06
0.85
0.00
0.70
0.79
0.69
2.04
3.73
0.24
0.34
0.89
6.12
0.54
0.78
3.08
1.45
0.25
0.99
4.04
2.05
2.04
0.00
0.35
1.20
3.76
0.79
1.64
0.75
0.99
0.50
0.45
1.51
1.65
0.58
2.11
1.78
2.02
0.79
1.58
0.00
0.50
1.28
3.36
1.87
0.00
2.08
0.39
0.90
1.41
1.18
1.24
3.75
1.90
0.00
0.00
0.00
0.00
0.00
0.00
0.00
0.26
9.09
0.00
0.76

Table A.5: Robustness of Panel A of Table 1 to using 5% threshold for procurement

revenue share
Dependent variable: Transfers to fly-by-night firms as a share of revenue
Estimation results of equation 1
Month from Coef. Std. Err. t-stat Coef. Std. Err.
t-stat
2
1
m
election, m
m
-8
0.005
0.045
0.11 -0.215
0.052
-4.18
-7
0.015
0.061
0.24 -0.031
0.076
-0.41
-6
0.236
0.085
2.77 0.048
0.101
0.48
-5
0.115
0.061
1.87 0.064
0.094
0.68
-4
0.060
0.075
0.8
0.098
0.055
1.78
-3
0.242
0.052
4.65 0.282
0.107
2.64
-2
0.167
0.031
5.42 0.206
0.086
2.39
-1
0.293
0.035
8.44 0.220
0.060
3.68
1
0.690
0.167
4.14 0.367
0.107
3.43
2
-0.040
0.024
-1.63 0.006
0.066
0.09
3
0.000
0.022
0.01 0.093
0.037
2.49
4
0.100
0.022
4.53 0.125
0.027
4.69
5
-0.026
0.054
-0.47 0.116
0.036
3.26
6
-0.005
0.053
-0.09 -0.260
0.218
-1.19
F -stat
1 jointly
5
2 jointly
11
1
1
2
2
F -stat
in[1;1] 6= out[1;1]
40
in[1;1] 6= out[1;1]
48
Note: Coefficients and standard errors are multiplied by 1000. Standard errors are corrected for clusters at the
regional level. The number of observations is 6,286,011. The number of firms is 45,275. The number of regions
(clusters) is 87.

52

Table A.6: Propensity to tunnel out of procurement revenue and out of other revenue
Dependent variable: Transfers to fly-by-night firms as a share of revenue
Estimation results of equation 3
Incoming revenue, t
0.068
(0.013)***
Incoming revenue, t 1
0.035
(0.009)***
Incoming revenue, t 2
0.004
(0.004)
Incoming procurement revenue, t
3.214
(0.663)***
Incoming procurement revenue, t 1
0.906
(0.445)***
Incoming procurement revenue, t 2
0.906
(0.217)***
Firm and week fixed effects
Yes
Number of obs
6,286,011
Number of firms
45,275
Number of regions
87
Note: Coefficients and standard errors are multiplied by 1000. Both dependent and the main independent
variables are normalized by the size o firms revenue in 2003. t is a particular week. * significant at 10%, **
significant at 5%, *** significant at 1% level.

53

54

0.274
(0.145)*
0.105
(0.044)**
2917418
25833
24
0.931

0.450
(0.053)***
0.420
(0.117)***
2691830
26521
63
0.354

share of revenue
Incumbent, the 1st round:
<70%
>70%

Note: Coefficients and standard errors are multiplied by 1000. Standard errors corrected for clusters at the regional level are in parentheses.
Election window is a dummy indicating a window of two months: one month before and one month after the elections, [-1;1]. Proc. rev. share
> 1% is a dummy indicating that procurement revenue is above 1% of a firms revenue. Procurement Rev share is the share of a firms revenue
that comes from procurement contracts. * significant at 10%, ** significant at 5%, *** significant at 1% level.

Dependent variable:
Transfers to fly-by-night firms as a
Subsample Winner, the 1st round:
Incumbent:
of elections: <50%
>50%
lost
won
Specification 1
(Proc. rev. share > 5%) Election window -0.034
0.497
0.053
0.493
(0.056)
(0.077)*** (0.092) (0.083)***
Election window
0.087
0.308
0.000
0.298
(0.079)
(0.065)*** (0.087) (0.068)***
Number of obs 1833618
4452393
744847
4864401
Number of firms 16980
33817
7460
38665
Number of regions
45
57
27
68
Mean dependent variable
0.320
0.712
0.325
0.678

Table A.7: Robustness of Panel A of Table 2 to using 5% threshold for procurement revenue share

Table A.8: Robustness of Table 3 to using 5% threshold for procurement revenue

share
Dependent variable:
L(Tunneling/week), election window
TI CPI Corruption L(Tunneling/week), election window
L(Tunneling/week), outside election window
log(Revenue)
Net Income/Revenue
Debt/Assets
Industry dummy
Region dummy
Election year dummy
R-sq
Number of obs
Number of firms
Number of regions

Dummy: Proc. rev. > 5% revenue

(1)
(2)
0.006
0.006
(0.002)***
(0.001)***
0.002
(0.001)***
0.000
0.000
(0.001)
(0.001)
-0.003
-0.003
(0.002)
(0.002)
-0.025
-0.027
(0.005)***
(0.007)***
-0.012
-0.011
(0.003)***
(0.004)***
Yes
Yes
Yes
Yes
Yes
Yes
0.042
0.041
63,021
54,115
45,275
38,044
87
40

Note: Standard errors corrected for clusters at the regional level. TI CPI corruption is the Transparency
Internationals regional corruption perception index. * significant at 10%, ** significant at 5%, *** significant at
1% level.

55

Table A.9: Robustness of Table 4 to using 5% threshold for procurement revenue

share
Dependent variable:

(Tunneling)/(Revenue) > 0.1

Log(Revenue)
Net Income/Revenue
Debt/Assets
Industry dummy
Region dummy
Election year dummy
R-sq
Number of obs
Number of firms

Dummy: Procurement revenue > 5% revenue

(1)
(2)
(3)
All
TI CPIMedian TI CPI>Median
0.017
0.003
0.022
(0.004)***
(0.004)
(0.005)***
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
0.003
0.000
0.003
62,874
26,161
27,838
45,128
19,746
18,206

Note: Standard errors corrected for clusters at the regional level are in parentheses.TI CPI is the Transparency
Internationals regional corruption perception index. * significant at 10%, ** significant at 5%, *** significant at
1% level.

Table A.10: Summary statistics for local corruption measures

Obs

t-stat on coefficient 1l
Dummy (1l is signif at 5% level)
t-stat on coefficient 1l
Dummy (1l is signif at 5% level)

Mean Std. Dev. Min Max

Locality level
246
0.60
1.36
-4.22 4.40
246
0.21
0.41
0
1
Firm level (observations)
44118 1.16
1.36
-4.22 4.40
44118 0.38
0.48
0
1

56

57

-0.017
(0.002)***
-0.068
(0.018)***
0.008
(0.001)***
-0.049
(0.007)***
0.06
11010
6802
20
-0.020
(0.019)
0.08
18292
12311
64

-0.008
(0.002)***

insignificant

-0.041
(0.009)***
0.06
11010
6802
20

-0.011
(0.001)***

significant

insignificant

significant

insignificant

significant

Dep. var: Procurement revenue/revenue

-0.015
(0.004)***
-0.039
(0.013)***
0.009
(0.004)**
-0.029
(0.018)
0.07
18292
12311
64

significant

insignificant

-0.014
(0.004)***
-0.041
(0.013)***
0.008
(0.004)*
-0.028
(0.018)
0.08
18292
12311
64

insignificant

(7)

-0.0009
-0.0013
-0.0006
-0.0009
-0.0008
(0.0003)*** (0.0001)*** (0.0002)*** (0.0001)*** (0.0003)***
-0.005
-0.007
-0.0053
(0.002)**
(0.001)***
(0.002)**
0.00038
0.00057
0.0003
(0.0003)
(0.0001)***
(0.0004)
0.0501
-0.002
-0.002
0.0071
0.0503
-0.0016
(0.011)***
(0.001)
(0.001)**
(0.007)
(0.011)***
(0.001)
0.0456
0.0556
0.0464
0.0642
0.0495
0.0648
11010
18292
11010
18292
11010
18292
6802
12311
6802
12311
6802
12311
20
64
20
64
20
64
Controls used in each of the two regressions presented in column column

-0.0008
(0.0001)***

significant

-0.040
(0.008)***
0.06
11010
6802
20

-0.010
(0.001)***

insignificant

(3)
(4)
(5)
(6)
Dep. var: Dummy: government procurement > 1 % of revenue

significant

(2)

-0.0014
(0.0001)***
-0.0066
(0.001)***
0.0006
(0.0001)***
-0.0018
(0.001)**
0.0504
11010
6802
20

significant

-0.019
(0.002)***
-0.065
(0.018)***
0.009
(0.001)***
-0.05
(0.007)***
0.06
11010
6802
20

significant

(8)

Industry dummy
Y
Y
Y
Y
Y
Y
Y
Y
Region dummy
Y
Y
Y
Y
N
N
N
N
Locality dummy
N
N
N
N
Y
Y
Y
Y
Election year dummy
Y
Y
Y
Y
Y
Y
Y
Y
Note: The sample is split into two subsamples with significant and insignificant local corruption, i.e., p-value above and below 10 for the coefficient
1 in equation 9, measuring the association between tunneling around elections and the distribution of public procurement. Standard errors
corrected for clusters at the regional level are in parentheses. Election year fixed effects and industry fixed effects are included as controls in all
specifications. * significant at 10%, ** significant at 5%, *** significant at 1% level.

R-sq
Number of obs
Number of firms
Number of regions

Debt/Assets

Log(Revenue)
0.0066
(0.007)
0.0550
18292
12311
64

-0.0007
(0.0001)***

Log labor productivity

Net Income/Revenue

insignificant

Sample: Local corruption

Panel B

R-sq
Number of obs
Number of firms
Number of regions

Debt/Assets

Log(Revenue)
-0.021
(0.019)
0.07
18292
12311
64

-0.008
(0.002)***

Log labor productivity

Net Income/Revenue

insignificant

(1)

Sample: Local corruption

Panel A

Table A.11: Procurement and productivity in subsamples with high and low local corruption

58

Dummy: government procurement revenue

t-stat for 1
dummy: p-value< 10%
t-stat
(1)
(2)
(3)
(4)
(5)
-0.006
-0.005
-0.006
-0.005
-0.006
(0.001)*** (0.001)*** (0.001)*** (0.001)*** (0.001)***
0.000
0.003
-0.001
0.006
(0.001)
(0.002)**
(0.002)
(0.003)*
-0.001
-0.002
(0.0004)**
(0.001)***
0.002
0.002
0.002
0.002
0.002
(0.001)*
(0.001)*
(0.001)*
(0.001)*
(0.001)*
-0.043
-0.044
-0.043
-0.044
-0.042
(0.009)*** (0.009)*** (0.009)*** (0.009)*** (0.009)***
-0.010
-0.010
-0.010
-0.010
-0.010
(0.006)
(0.006)
(0.006)
(0.006)*
(0.006)*
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
Yes
No
No
No
No
No
Yes
Yes
Yes
Yes
Yes
Yes
0.03
0.03
0.03
0.03
0.04
29,302
29,302
29,302
29,302
29,302
19,113
19,113
19,113
19,113
19,113
64
64
64
64
64
-0.001
(0.0005)**
0.002
(0.001)*
-0.043
(0.009)***
-0.010
(0.006)*
Yes
No
Yes
Yes
0.04
29,302
19,113
64

0.002
(0.001)*
-0.042
(0.009)***
-0.010
(0.006)*
Yes
No
Yes
Yes
0.04
29,302
19,113
64

-0.003
(0.001)***
0.002
(0.001)*
-0.043
(0.009)***
-0.010
(0.006)*
Yes
No
Yes
Yes
0.04
29,302
19,113
64

> 5 % of revenue
for 1
dummy: p-value< 10%
(6)
(7)
(8)
-0.004
-0.006
-0.005
(0.001)*** (0.001)*** (0.001)***

Note: Standard errors corrected for clusters at the regional level are in parentheses. Election year fixed effects and industry fixed effects are included
as controls in all specifications. Local corruption stands for one of the two measures of corruption constructed from the association between
tunneling around elections and the distribution of public procurement: the size of the t-statistic or the dummy for p-value<5%. * significant at 10%,
** significant at 5%, *** significant at 1% level.

dummy
dummy
dummy
dummy
R-sq
Number of obs
Number of firms
Number of regions

Industry
Region
Locality
Election year

Debt/Assets

Net Income/Revenue

Log(Revenue)

Log labor productivity Local corruption

Local corruption

Log labor productivity

Dependent var:
Local corruption measure:

Panel A: Dummy.

Table A.12: Robustness of Panel A of Table 5 for using 5% threshold for procurement revenue share

Figure A.1: Political cycle in tunneling for firms with procurement revenue above and below
5% of revenue
The political cycle in tunneling of firms with proc. rev. below and above 5% of revenue,
2
1
:
+ m
m
1.60%
1.40%
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

!0.40%

2
:
The political cycle in tunneling of firms with procurement below 5% of revenue, m
1.60%
1.40%
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

4%

5%

6%

!0.40%

1
:
The difference between the two, m
1.60%
1.40%
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

!0.40%

Note: Units of the y-axes: weekly transfers to fly-by-night firms in percentage of annual revenue multiplied by
1
2
1000. The figure portrays coefficient estimates of m
, m
and their sum from the estimation of equation 1 along
with their confidence intervals. The x-axes portray the number of months m from the election date.

59

Figure A.2: Placebo election dates

.2

Placebo results, equation 1 (200 iterations)

.05

Density
.1

.15

Value of F in the real data

10
20
30
40
F-statistic: (inside election window)=(outside election window)

50

.3

Placebo results, equation 2 (200 iterations)

.1

Density

.2

Value of F in the real data

10
20
F-statistic: (within election window) = (outside election window)

60

30

Figure A.3: Robustness of political cycle in tunneling to controlling for firms incoming procurement revenue
2
Dynamics of m
:
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

2%

3%

4%

5%

6%

2%

3%

4%

5%

6%

2&

3&

4&

5&

6&

!0.40%
!0.60%

1
:
Dynamics of m
1.20%
1.00%
0.80%
0.60%
0.40%
0.20%
!8%

!7%

!6%

!5%

!4%

!3%

!2%

0.00%
!1%
!0.20%

1%

!0.40%
!0.60%

1
Dynamics of m
:

4.50&
3.50&
2.50&
1.50&
0.50&
!8&

!7&

!6&

!5&

!4&

!3&

!2&

!1&
!0.50&

1&

!1.50&

Note: Units of the y-axes: weekly transfers to fly-by-night firms in percentage of annual revenue multiplied by
1
2
1
1000. The figure portrays coefficient estimates of m
, m
and m
along with their 95% confidence intervals from
the estimation of equations 1 and 2 with additional control for the incoming procurement revenue in the three
P rocRevf t P rocRevf,t1 P rocRevf,t2
,
,
. X-axes portray the number of months m from the election
preceding weeks:
Revf
Revf
Revf
date.

61

Data appendix

B.1

Data sources used in addition to banking transactions data

In addition to the list of banking transactions, we use two other data sources. The first source is
the Rosstat (Russias official statistical agency) database of Russian companies provided by Spark
(http://www.ispark.ru/en-US/default.aspx). This database contains a firms ID number (INN),
name, region, date of registration, industry, directors, owners, and other identifying information
about the firm. In addition, it contains basic accounting data, such as revenue, profit, net income,
assets, and debt. According to Russian law, all firms (even small ones) must report their balance
sheets and income statements to Rosstat quarterly. Though this law sets no explicit penalty
for noncompliance, the majority of Russian firms report their data to Rosstat to maintain good
relations with the tax authorities. The Rosstat database contains accounting data for about 1.5
million Russian firms.
The second dataset includes the personal income of Moscow residents. It contains more than 7
million records for 2002, and more than 9 million records for 2003 and 2004. Each entry contains
unique identification data (name, address, identification number) for both employer and employee.
A person can have multiple records if he or she receives income from several sources.

B.2
B.2.1

Description of variables used in the analysis

Sample: firms elections

Revenue, 2003 - Companys book revenue in 2003 taken from Rosstat

Assets, 2003 - Companys book assets in 2003 taken from Rosstat
Net Income, 2003 - Companys net income in 2003 taken from Rosstat
Debt, 2003 - A sum of companys short term debt in 2003 and long term debt in 2003, which
are taken from Rosstat
Annualized transfers to fly-by-night firms 1999-2004 - Total transfers to fly-by-night firms
by a firm in 1999-2004 divided by six. The transfers to fly-by-night firms are calculated
using the banking transaction database. See Mironov (2013) for a detailed description of
the identification procedure of fly-by-night firms. These firms are referred to as spacemen
in Mironov (2013).
Annualized revenue from procurement, 1 year after election - Total transfers from governmentaffiliated entities to a firm, which have the reported purpose of payment for goods and services during one year after the election. These transfers are calculated using the banking
transaction database. In the baseline analysis, we exclude payments for utilities, e.g., electricity and water, from the list of revenues from public procurement contracts because these
contracts are not usually allocated on a competitive basis and are automatically allocated
to local monopolists. (None of the results changes if we keep utilities in definition of public
procurement.) If the election date happened to be in 2004, then we divide the total transfers
by the number of days left until the end of 2004 and multiply by 365 (to get annualized
revenue from procurement).
62

 Tunneling per week, election window - We calculate total transfers to fly-by-night firms by
a firm from 4 weeks before until 4 weeks after the election and divide by 8. If the election
period overlapped with our sample period and is shorter than 8 weeks, then we divide
the transfers to fly-by-night firms by the actual number of weeks presented in our sample
period (1999-2004). For example, if the date of the election is 2004-12-19, then we divide
the transfers to fly-by-night firms by 5.7.
Tunneling per week, outside election window - We calculate total transfers to fly-by-night
firms by a firm starting one year before the election date and ending 4 weeks before the
election. Then, we divide this number by the actual number of weeks presented in our
sample period.
Perceived corruption - see section 4.2 of the paper for description
Tax-agency-level corruption - see section 5 of the paper for description
Number of employees - number of firms employees in 2003 taken from Rosstat
B.2.2

Sample: Firms Weeks

Tunneling per week - Transfers to fly-by-night firms during a specific week.

Tunneling per year - Annualized transfer to fly-by-night firms from one year before until one
year after the election. For example, if the election date is 2004-03-14 then we take total
transfers to fly-by-night firms from 2003-03-14 to 2004-12-31. After that we divide this
number by the number of days in the period from 2003-03-14 to 2004-12-31 and multiply
by 365. If the election date is 2003-12-07 then we take total transfers to fly-by-night firms
from 2002-12-07 to 2004-12-07 and divide them by 2.
Revenue - Companys book revenue in 2003 taken from Rosstat
Procurement revenue - Annualized transfers from government-affiliated entities from one
year before until one year after the election. We include only transactions that have the
reported purpose of payment for goods and services. We exclude payments for utilities, e.g.,
electricity and water, from the list of revenues from public procurement contracts.

B.3

Reality checks on the banking transactions data by Mironov (2013)

In the section, we re-print the abbreviated vesion of the reality checks performed on the banking
transactions data by Mironov (2013), which were published as an online appendix to the article and
can be found at: http://onlinelibrary.wiley.com/doi/10.1111/jofi.12026/suppinfo.31 See pages 913 of the original document:
Because the Russian Central Bank never confirmed the authenticity of the data, I examine
whether the banking database contains transactions of real companies and how complete it is.
31

This site was accessed on November 24, 2014.

63

For this purpose, I use the Rosstat32 database of Russian companies provided by Spark.33 This
database contains a firms INN [comment of authors: INN is the firms ID], name, region, date
of registration, industry, directors, owners, and other identifying information about the firm. In
addition, it contains basic accounting data, such as revenue, profit, net income, assets, and debt.
According to the Russian law, all firms (even small ones) must report their balance sheets and
income statements to Rosstat on a quarterly basis. Although this law does not set any explicit
penalty for firms that do not report, the majority of Russian firms prefer to report their data to
Rosstat to maintain good relations with the tax authorities. Rosstat contains accounting data for
about 1.5 million Russian firms. If the banking data used here are accurate, there should be a
large degree of similarity between the Rosstat data and the banking data.
A. Test 1. Matching Rosstat to Banking Data
Using the Rosstat database, I select up to 3,000 companies with highest revenues from the 20
largest Russian regions. Columns (1) and (2) of Table AI.I [comment of authors: the table is
re-printed below] contain average revenue for 2003 for every region. An average company from
Moscow has about 15 times higher revenue ($98.4M) than that of an average one from outside of
Moscow ($6.3M). This is not surprising because the majority of large Russian corporations that
operate in different regions have their headquarters in Moscow (e.g., Gazprom, United Energy
System, Lukoil , Rosneft, Russian Railways, Aeroflot ). According to expert estimates supported
by Rosstat data, about 80 to 90% of all Russian business activities are concentrated in Moscow.
I match this sample of companies to the banking database by INN. Columns (3) to (5) of Table
AI.I show the results: 71% of the firms from Rostat are also present in the banking database. The
match rate for Moscow city and Moscow region (94 to 97%) is much higher than the match rate
for other regions (49 to 84%, average 69%). [. . . ] Also, an average firm from column (4) has 30%
greater revenue than an average one from column (2). The regional firms caused this difference
in size. [. . . ] See column (6) of Table AI.I for the results of the match weighted by revenue. Even
for firms from outside Moscow, the match rate is 79% to 97% with an average of 90%. Columns
(7) to (10) show firms that have at least 100 banking transactions present in the banking data set.
As expected, these companies are much larger; the average revenue of a firm from column (8) is
$23.2M whereas the average revenue of a firm from column (4) is $13.9M. Even though only 39%
of the firms from the original sample have at least 100 transactions in the banking database, they
account for about 85% of the total revenues. [. . . ]
B. Test 2. Matching Banking Data to Rosstat
I select a random sample of 1,000 firms from the banking dataset. To avoid inclusion of
spacemen [comment of authors: Mironov (2013) refers to fly-by-night firms as spacemen], I
select only those firms that have a ratio of tax paid to total turnover of more than 1%. [. . . ]
spacemen are less likely to provide their financial statements to authorities and therefore are less
likely to be present in the Rosstat database. A comparison of these 1,000 firms with the Rosstat
database by INN matched 676 of the 1,000 firms. Why are 32% of the firms not found in the
Rosstat database? One possible explanation is the presence of typos in the INNs in the banking
database. Although I performed several attempts to clean this dataset, it is possible that not
32
33

Rosstat is an official Russian statistical agency.

spark.interfax.ru

64

everything was cleaned. Another possible explanation is that due to the lack of punishment for
nonreporting, some firms chose not to send their reports to Rosstat.
C. Test 3. Analysis of the Matched Firms
I choose a random sample of 500 firms that are present in both Rosstat and the banking
dataset. Using similar logic as in Test 2, I select only the firms that have a ratio of tax paid to
total turnover of more than 1%. Then I choose all banking transactions for these 500 firms the
banking dataset. In total, I get a sample of about 300,000 transactions. I manually classify each
firms transactions that correspond to revenue receipts and profit tax payments according to the
description of each transaction. Next, I aggregate these transactions at the firm level for 2003
and 2004. [comment of authors: Mironov (2013) constructs a set of financial variables from
banking transactions data and compares them to the corresponding variables in the registry. The
variables constructed from the banking transactions data are denoted by B at the end of their
names.] For revenue (Revenue B) I use cash basis accounting principles (i.e., if a transaction took
place in 2003, it is accounted for in 2003) because it is often impossible to determine the time of
performing the service from the transaction description. For profit tax payments (Profit tax B),
I use the transaction description to identify the period because all tax payments must include the
corresponding tax period in the transaction description. I define Profit B as Profit tax B/Tax rate.
Because a portion of profit tax for 2004 should be paid at the beginning of 2005 and the banking
dataset does not have 2005 transactions, Profit B for 2004 is underestimated. I obtain Reported
revenue (Revenue) and profit before taxes (Profit) for each firm from the Rosstat database. I define
Margin B as Profit B/Revenue B and Margin as Profit/Revenue. See Table IA.II [comment of
authors: the table is re-printed below] for summary statistics and correlations.
The table indicates that sample characteristics for the Rosstat data are similar to those for the
banking data. The correlation between Log(Revenue) and Log(Revenue B) is 0.76 for 2003 and
0.56 for 2004, the correlation between Log(Profit) and Log(Profit B) is 0.87 for 2003 and 2004, and
the correlation between Margin and Margin B is 0.66 for 2003 and 0.49 for 2004. Furthermore,
26% to 28% of the firms have banking revenues within [-10%, +11%] of the reported revenues, and
64% to 71% have their banking and reported revenue within the [-39%, +65%] range. These facts
suggest that the banking data set contains real transactions of real firms. However, it is likely
that some transactions might be missing or mistyped; otherwise correlations between analogous
characteristics should be higher. Figure IA.2 [comment of authors: the figure is re-printed below],
which plots Log(Revenue) against Log(Revenue B) and Log(Profit) against Log(Profit B), provides
additional evidence of similarity between Rosstat and banking transaction data.

65

Note: Table reprinted from Mironov (2013) online appendix.

66

Note: Table reprinted from Mironov (2013) online appendix.

67

Note: Figure reprinted from Mironov (2013) online appendix.

68