57"! . 1. ..

91K

-

..-'l‘.'
...'__-."“_.

..~...

52'.

4. ‘... “a“. .
_. u

1‘

;!'5. ‘\

 

h
'5': 72‘s.?

4.. I0

 

(5-.....

 

 

«I

w

a.

 

. ,...
A 3.1-;-

-

” 339...‘
th'. -

 

“mm:

2 LIBRARY
; nag; Michigan State lte
University

 

This is to certify that the
dissertation entitled

ESSAYS IN FIRM-LEVEL COSTS OF CORRUPTION AND THE
TRANSMISSION OF MONETARY POLICY SHOCKS TO THE REAL
ECONOMY

presented by

Lebohang Lijane

has been accepted towards fulfillment
of the requirements for the

PhD degree in Economics

 

 

Major Professor’s Signature

Date

 

 

MSU is an affirmative-action, equal-opportunity employer

. ~,....—.-.—n---I.--~-o.--.—..u-a-u-uclnn-

PLACE IN RETURN BOX to remove this checkout from your record.
TO AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DAIEDUE

DAIEDUE

DATEDUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

6/07 p:/CIRC/DateDue.indd-p.1

 

ESSAYS IN FIRM-LEVEL COSTS OF CORRUPTION AND THE TRANSMISSION
OF MONETARY POLICY SHOCKS TO THE REAL ECONOMY
By

Lebohang Lijane

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY
Department of Economics

2007

ABSTRACT
ESSAYS IN FIRM-LEVEL COSTS OF CORRUPTION AND THE TRANSMISSION
OF MONETARY POLICY SHOCKS TO THE REAL ECONOMY
By

Lebohang Lijane

By distorting the rule Of law and weakening the institutional foundations on which
economic growth depends, corruption has become one of the greatest Obstacles to economic
and social development. While country level causes and consequences of corruption are
relatively well understood, very little is known about corruption related determinants of firm
performance. The first essay, “Bribery and the Nature of Corruption” studies the impact
of bribery on firms. Specifically, the paper explores how the organization of the corruption
network influences its costs on firms. Using a relatively new dataset, the study empirically
tests the impact of the structure of the corruption network on the incidence and amount of
bribes firms face. Findings of the empirical analysis provide evidence that corruption faced
by firms varies with the structure of its network. Particularly, the incidence and
pervasiveness of corruption is higher where the corruption regime is well organized and
uncertainty regarding delivery of the service that is the object of the bribe increases the cost
to firms.

Building on the theme of the first essay, my second essay “Bribery and Firm
Borrowing Conditions” examines the channels through which bribery constrains firms’
operations and growth. Since corruption induces uncertainty into the operational
environment, it can potentially be a barrier to firms’ access to external finance by lowering

the projects’ expected returns. Using firm-level data from the Business

Environment and Entrepreneur Performance Surveys covering twenty-six transition
economics, the study investigates whether a link exists between bribery and firms’ borrowing
conditions. Empirical results suggest that bribery leads to stringent borrowing conditions. In
particular, in environments characterized by pervasive corruption, the probability of firms
obtaining bank credit is lower; interest rates on which loans are made are higher; and loan
repayment periods are generally shorter.

The last essay, “The Role of U.S. Banks in the Transmission of Monetary
Policy”, examines whether in the U.S. there is a part of monetary policy actions on real
economic activity that can be attributed to banks through the bank-lending channel. The
effects of monetary contractions can not be fully explained by the well-established interest
rate channel, so is it possible that additional effects come from credit markets imperfections?
A bank-lending channel is operative if the effect on firm borrowing is a consequence of a
decline in loan supply rather than loan demand. Evidence on the existence of this channel
has so far proved elusive. Using data from the CALL reports for the period 1991 to 2000, I
estimate the impact of bank size and financial soundness, measured by CAMEL—type ratios,
on bank lending. The results indicate that banks of different sizes and financial positions are
affected differently by monetary policy shocks. Furthermore, unlike existing literature that
shows no effect on large banks, I find that depending on their financial attributes, these
banks are also affected by policy changes. An additional contribution of this study is the

treatment for sample attrition that is a prominent feature of the U.S. banking industry.

ACKNOWLEDGEMENTS

The completion of this dissertation has been made possible by the many people who
supported me in numerous ways throughout the long and rocky process. I am indebted to
each and every one of them.

Firm and foremost, I want to thank my advisor, Professor Ana Maria Herrera, who
willingly took over the supervision Of this dissertation at a very difficult stage. I deeply
appreciate the amount of time she devoted to providing advice, academic direction and
encouragement. It was only after she became my advisor that I started believing that I would
complete this work.

I am also deeply grateful to the other members committee, Dr. Wooldridge for his
invaluable support on all empirical aspects of this work and Dr. jeitschko, especially for his
steadfast encouragement.

Dr. John Strauss, even though no longer a member of this committee, has continued
to support me, especially in times of difficulty. I am particularly grateful to him for restoring
my belief in myself.

My co-workers and friends at the World Bank—jorge, Elene, Xiao, Pamela, Patricia,
Ousman, Ann, Ausi Lesego, Sonia, Delfin and Yutaka—have been an invaluable source of
strength, supporting me emotionally and devoting time to reading and critiquing my work.
As a group they helped me understand that quitting was not an option. Mu Ren deserves
special thanks for all she has helped me with, especially in the last two years. She has listened
to me complain, read through my dissertation and helped me to recognize weaknesses in my

work.

iv

Friends at All Saints, especially Carol Swinehart, Dr. Gail Shafer-Crane and the late
Dr. Jacquie Nickerson, have been my emotional and spiritual pillars. I have needed both
more times than I can count over the years of my academic life.

My family—especially Kamohelo, Lipalesa, Likhabiso, my mother, my sisters,
Libuseng and Matseliso, ‘Mangoane Thakane, ‘Mamokete, Molepo together with the Mbeki
quartet, especially Nobantu and Ausi ‘Mathapelo—through their unwavering belief in me,
made it possible for me to want to find the strength to finish the program. I will forever
remember the sacrifices they have made to help me achieve this goal.

I would be remiss if I did not express my profound appreciation to my manager, Mr.
Benno Ndulu, who made attainment of this degree a personal priority. He has been a true
mentor—encouraging, listening, pushing—whatever the situation required. I would not have
been able to finish this dissertation, definitely not at this time, if I did not have his full
support.

Lastly, appreciation of the two guys whose unconditional love for me helped put
things in the right perspective, Cheng Cheng (Zachary) and Pasqual, and you are right, Zach,

it is so amazing.

TABLE OF CONTENTS

CHAPTER 1
BRIBERY AND THE NATURE OF CORRUPTION .......................................... 1
1 Introduction ................................................................................... 1
2 Micro-Characteristics of Bribery ............................................................... 3
3 Theoretical Underpinnings .................................................................... 6
3.1 Organization of the Corruption Network ......................................... 6
3.2 The Role of the Operational Environment on the Incidence and Size of
Bribes .................................................................................. 8
4 Data Description and Methodology .......................................................... 9
4.1 Data .................................................................................... 9
4.2 Methodological Issues ............................................................... 14
5 Empirical Strategy ............................................................................ 15
6 Estimation Results ........................................................................... 18
6.1 Organized and Disorganized Corruption .......................................... 18
6.2 Quality of Public Services and their Effect on Corruption ..................... 22
6.3 Corruption and Firm Characteristics ............................................. 26
7 Conclusion .................................................................................... 29
REFERENCES ....................................................................................... 32
CHAPTER 2
BRIBERY AND FIRM BORROWING CONDITIONS: IS THERE A
RELATIONSHIP .................................................................................... S8
1 Introduction ................................................................................. 58
2 Data, Summary Statistics and Theoretical Links ............................................ 61
2.1 Bank Lending Conditions ......................................................... 62
2.1.1 Interest Rates ............................................................... 63
2.1.2 Quantity of Loans .......................................................... 64
2.1.3 Collateral Requirements ................................................... 65
2.1.4 Maturity .................................................................... 66
2.2 Corruption Measures .............................................................. 66
2.3 Legal,}udicial and Infrastructure ................................................. 67
2.4 Firm-specific Characteristics and Country-level Variables. . . . . . . . . . . . . . ..........67
3 Estimation Strategy .......................................................................... 68
4 Empirical Results ............................................................................. 71
5 Conclusion ..................................................................................... 75
REFERENCES ....................................................................................... 76
CHAPTER 3
ARE BANKS IMPORTANT IN TRANSMITTING MONETARY POLICY SHOCKS
TO THE REAL ECONOMY? ........................................................................................................ 90
1 Introduction ................................................................................. 90
2 The Bank Lending Channel: Theoretical Framework and Evidence. . . . . . . . . . . .96
2.1 Theoretical Underpinnings .......................................................... 96

vi

2.2 Empirical Studies .................................................................. 98
2.3 Changing Structure of the Banking Industry and the Bank—Lending
Channel ............................................................................ 102
2.4 Some Stylized Facts ............................................................... 105
3 Data ......................................................................................... 109
3.1 Measures of Financial Soundness ............................................... 110
3.2 Other Descriptive Statistics ...................................................... 112
3.3 Potential Problems ................................................................ 1 13
3.3.1 Sample Attrition: Tests and Correction ................................. 113
3.3.2 Endogeneity .............................................................. 117
3.3.3 Correlation between Fixed and Explanatory Variables. . . . .. . . ........119
4 Estimation and Results ................................................................... 120
4.1 Estimation ........................................................................ 120
4.2 Estimation Methodology ......................................................... 121
4.3 Empirical Results ................................................................. 123
5 Conclusion .................................................................................. 131
REFERENCES .................................................................................... 132
APPENDIX ......................................................................................... 138

vii

Table 1-1

Table 1-2

Table 1-3

Table 1-4

Table 1-5

Table 1—6

Table 1-7

Table 1-8

Table 1-9

Table 1-10

Table 1—11

Table 1-12

Table 1-13

Table 1-14

Table 1-15

Table 1-16

Table 1-17

Table 1-18

Table 2-1 a

Table 2-1 b

Table 2-2

Table 2-4

Table 2-5

LIST OF TABLES

Selected Summary Line ............................................................ 34
Selected Correlations ................................................................ 35
First-Stage Rivers and Vuong Regressions ......................................... 37

Second-Stage Rivers and Vuong Regressions (Test for Exogeneity). . . .....38

Corruption as a Constraint to Firm Operations .................................. 40
Corruption as the Single Most Important Obstacle ............................. 41
How Common are Bribes for Getting Things Done ........................... 42
Size of the Bribe (Percentage of Contract Value) ............................... 44
Size of the Bribe (Percentage Of Firm Revenue) .................................. 46
Frequency of Additional Payments: Electric Power Company ................ 48
Frequency of Additional Payments: Telephone Personnel ..................... 49
Frequency of Additional Payments: Licensing Authorities ..................... 50
Frequency of Additional Payments: Tax Inspection Agents ................... 51
Frequency of Additional Payments: Customs Officials ........................ 52
Frequency of Additional Payments: Government Procurement ............... 53
Frequency of Additional Payments: Judicial and Courts ....................... 54
Frequency of Additional Payments: Politicians ................................. 55
Test for Non—Linearities in Infrastructure ......................................... 56
Summary Statistics: All Countries ................................................ 79
Summary Statistics by Country: Selected Variables ............................. 82
Selected Correlations ............................................................... 84
Impact of Corruption on Bank Loans: Inspections ............................. 85
Impact of Corruption on bank Loans: Frequency of Bribes ................... 86

viii

Table 2—6

Table 2-7

Table 2-8

Table 2-9

Table 3-1

Table 3-2

Table 3-3

Table 3—4

Table 3-5

Table 3-6

Table 3-7

Table 3-8

Table 3-9

Firms’ Access to Finance — TObit and Logit Regressions ...................... 87
Impact Of Corruption on Interest Rates .......................................... 88
Impact of Corruption on Collateral Requirements .............................. 89
Impact of Corruption on Loan Term ............................................ 90
Summary Statistics ................................................................. 135
Summary Statistics for Ratios ................................................... 137
Descriptive Statistics by Attrition ................................................ 139
Test with Selection Lead ......................................................... 140
Attrition Probability .............................................................. 141
Correction for Attrition ......................................................... 148
Loan Response to Policy Change Levels ...................................... 149
Loan Response to Policy Shocks — Regressions with Interactions ......... 150
Ability to Issue External Debt ................................................. 153

ix

CHAPTER 1. BRIBERY AND THE NATURE OF CORRUPTION

1 Introduction

The growing consensus that corruption is harmful to economies worldwide has
meant that the recent past has seen corruption drawing increasing attention from
policymakers, politicians and academic scholars. A vast literature has emerged on causes and
consequences of corruption, with virtually all empirical work based on cross country data.
These works have made important contributions to understanding causes and consequences
Of corruption. For instance, the studies have found that corruption slows economy-wide
growth and total investment (Mauro 1995), reduces foreign direct investment (Wei 1997),
drives firms out of the official economy (Kaufmann 1997), and reduces both public sector
budgets and the productivity of a county’s infrastructure (T anzi and Davoodi 1997).

Country level studies however, since they only provide aggregate determinants, tell
us very little about the relationships between corruption and economic agents. They
especially can not explain why individuals or firms facing identical institutional and policy
environments, can and do pay different amounts in bribes for the same services. But until
very recently, micro—level aspects of corruption had largely been overlooked, yet it is through
firms and individuals that economies function. Little attention has been paid to corruption-
related determinants of firm performance, yet the extent to which firms’ operations are
constrained is central to economic outcomes. Neglecting first level operators and then
attempting to draw policy implications from macroeconomic studies is, at best, a risky

endeavor. Moreover, the severity and institution of corruption is vastly different across

counties], therefore we would expect to Observe heterogeneities in the impacts of corruption
on outcome measures. Hence we believe micro-level empirical research is critical for
understanding heterogeneous mechanisms and distributional outcomes Of corruption and
for better informing policy.

This paper is one of the few in the literature that exploit micro data to study impacts
Of bribery.2 It constitutes an empirical investigation of the impact Of corruption on firms.
Specifically, we explore how the organization Of the corruption networks influences its costs
on firms. The objective of the study is to improve the understanding of the workings of
corruption and their consequences at the micro-level. In parn'cular, using a relatively new
dataset, we empirically test the impact of the structure of the corruption network on the
incidence and amount of bribery that firms face.

While our line of inquiry is not new having first been proposed by Schleifer and
Vishny (1993) in a theoretical framework, as far as we know, our study is the first that
empirically investigates the link between the nature of corruption and the cost of bribes at

the firm level.3 This study provides empirical evidence supporting the claim that the cost of

 

' Some studies suggest that corruption is the same everywhere: that its effects on firms are similar
regardless of any national nature (Banfield 1975). Other authors, such as Schleifer and Vishny (1993),
suggest that the nature of corruption varies appreciably across countries. In some countries, corruption is
hierarchical, organized and predictable; one bribe guarantees access to the desired property or service
whereas the opposite is true in other countries.

2 This fairly new and small strand of empirical literature on broad micro level effects of corruption is
beginning to attract scholarly attention. Using the Business Environment Survey conducted by the World
Bank, Gaviria (2002) examines the effects of corruption on growth of sales, investment and employment at
the firm level. Mocan (2005) using micro data from the International Crime Victim Survey on 49 countries
investigates the causes of corruption but specifically incorporates individual specific characteristics, such as
gender, income, education and marital status to determine the probability that the individual will be
exposed to corruption. Hunt and Laszlo (2006) using survey data on households in Peru find that because
rich clients are more likely to use public officials, they are targeted by officials for bribes and are more
likely to pay bribes. However, they find that the service delivery does not necessarily improve with the
payment of bribes. Svensson (2002) based on a survey of Ghanaian firms finds that those firms that deal
with officials whose actions directly affect their operations are more likely to make unofficial payments.

3 There are two empirical papers that also use the organization of the corruption network. Wu (2005) tests
the importance of the network of corruption but unlike our approach, this paper focuses on the relationship
between bribery and firm accounting practices for corporate Asia. Campos, Lien and Pradhan (year??) use

corruption faced by the firm varies with the network through which corruption is organized.
In particular, we find that the incidence and pervasiveness Of corruption is higher where the
corruption network is well organized. Additionally, the frequency of bribes decreases if firms
have effective recourse through government channels or a managerial superior to obtain
proper treatment without agreeing to make unofficial payments.

Finally, this paper offers novel results suggesting that the quality of public services,
especially the judiciary systems and physical infrastructure, matters significantly for
perceptions of corruption. Thus, our results suggest that government policies directed
towards improving these services could meaningfully reduce corruption and thereby fosrer
economic growth.

The remainder of the paper is organized as follows. Section 2 highlights the
difference between country—based and firm-based assessments of corruption. Section 3
presents the theoretical motivation for the empirical work while Section 4 describes the data
and the survey instrument. The primary hypotheses and the estimation procedures are
presented and discussed in Section 5. Section 6 presents the estimation results and Section 7

concludes.

2 Micro-Characteristics of Bribery

Groups like Tramparengl International typically measure the ‘level’ of corruption within
a given country through some combination of surveys of business leaders, politicians and
academicians. The cardinal rankings that result surely capture a broad picture of the relative
level of corruption across countries but may tell us very little about the likely implications of

a particular level of corruption or even what it means for one country to be ‘more corrupt’

 

the same dataset to explore the importance of predictability of corruption in determining its impact on
investment.

than another. Moreover, there are good reasons to believe that the effects of corruption can
be quite different among countries with the same broadly measured level of corruption.
Bhardan (1997) notes in his excellent survey piece that whereas country-level indices
measure Indonesia and India as equally corrupt most observers would agree that firms in the
two countries have vastly different experiences with corruption. Understanding the
particular nature of corruption and the consequent variegated effects is vital to the design of
policies and other efforts aimed at reducing corruption, but require more detail than is
available through one-dimensional measures.

Recently completed firm-level surveys sponsored by the World Bank have for the
first time allowed researchers a valuable examination of the micro-characteristics of corrupt
transactions across a broad group of countries and regions. Such information allows for the
investigation of heretofore unexamined hypotheses regarding the environment within which
corruption takes place. Because corrupt transactions are by their very nature secret, firms
have no recourse to a legally enforceable contract if services are not rendered according to
the implicit contract of the bribe. Thus, the moral hazard risk surrounding any corrupt
transaction is naturally higher than that of an above-board payment for similar services. As a
result, corrupt transactions and their aftermath may represent a major source of uncertainty
for firms (Kaufmann, et. a1. 2000). Moreover, corrupt transactions represent a major
challenge to researchers as they are hidden, largely untraceable and highly significant features
of the commercial environment in many countries.

Corruption certainly imposes significant direct and indirect costs to firms. Direct
costs typically take the form of bribes or kickbacks and are most often monetary payments.

Monetary payments to corrupt officials can be expensive but indirect costs often pose even

greater Obstacles to firm performance.4 The more debilitating indirect costs of corruption
are: the opportunity costs of the resources it consumes, the cost of delaying transactions, the
cost of the resources expended in avoiding common venues for corrupt transactions (e.g.
entry into the underground economy), that it deters or eliminates investment through
reduced profitability, and that it raises the uncertainty of returns on investments. Schleifer
and Vishny (1993) argue that the direct costs of corruption (calculated as a percentage of
revenues) are in many ways like taxes in that they are simply another business expense. As
such, corruption’s direct cosrs may be high and still allow firms to operate normally. If the
analogy to taxes holds, it follows that the uniquely onerous burden of corruption stems from
the illegality and uncertainty of engaging in corrupt transactions. Otherwise identical corrupt
transactions may be undertaken under markedly different circumstances that are critical
determinants of their ultimate effect on a firm.

Kaufmann et a1 (2000) support the suggestion that the indirect costs of corruption
are the more insidious and make the important point that the most important public good
that the state can provide is predictability in the institutional and policy environment.
Similarly, we assert that the predictability of a corrupt environment largely determines its
cost to business. We are not suggesting that perfectly predictable corrupt regimes are not
costly or that bribes in such regimes are in fact just like taxes. Rather we emphasize that the
perception by firms of the predictability and efficacy of the corrupt regimes is a critical
determinant of its full cost to businesses. We investigate how the nature of corrupt regimes
affects the corrupt transaction itself. This paper furthers the understanding of how micro-
characteristics of firms, including their perceptions of the corrupt transactions they engage

in, affect the size and frequency of bribes.

 

" Moreover, most bribes are relatively small especially in comparison with the rents that are often conferred
on successful bribers. This finding is known as the 'Tullock Paradox' in the public choice literature.

3 Theoretical Underpinnings
In this section, we provide a brief theoretical motivation for the determinants of the
incidence and level of corruption. We identify three groups that are major factors: the

corruption regime, the quality of infrastructure and the quality Of institutions.

3.1 Organization of the Corruption Network

As stated in the introduction the framework we base our empirical work on was first
proposed by Shleifer and Vishny (1993), which characterizes two poles of organizational
structures for corruption networks. At one extreme in this framework is the fully
disorganized corruption network. Government agents who control access to related and
unrelated resources act independently and capriciously in an effort to maximize their own
bribe revenue and disregard the effects of their efforts on other officials. The lack of
coordination among corrupt agents works to diminish overall economic activity and lower
total bribe revenues in the same way that high taxes on final goods lower the derived
demand for those goods. More directly, independent providers of complementary goods
(e.g. import and business licenses) set the marginal revenue of a corrupt transaction equal to
the marginal cost (e.g. the official tariff rate or cost of a business license). In this type of
regime, firms are uncertain whom to pay, what to pay and if payment of bribes will result in
property rights over goods they purchased.

The other extreme of the scale is characterized by a well structured regime in which
payments expectations are not only predictable but ensure that the services for which firms
make unofficial payments will actually be delivered. In the Schleifer and Vishny analogy,
under a well-organized regime, bribe collectors act as joint providers and set marginal

revenue below marginal costs to account for the cross elasticities of demand for the goods

and thus provide more of the demanded goods in equilibrium. As a result, per unit bribes
are lower but total bribe revenue is higher in organized regimes.

One implication of the Shleifer and Vishny paradigm is that organized corruption
networks have less adverse effects on firms because government officials internalize some of
the negative effects of their corruption on overall economic activity. In disorganized
corruption networks however, firms are confronted with substantial uncertainty regarding
government-provided rights and services and may often need to bribe multiple agents for
the same service. Consequently, firms will reduce their demand for the objects of bribery
(i.e. contracts, business licenses, public services, etc.) and lower the total amount of bribes
ultimately paid and collected. In hierarchical and organized corruption network, agents’
actions are coordinated so as to maximize total bribe revenue. Through some combination
of interpersonal and institutional controls, the organized corruption network prevents
opportunistic actions by individual agents and thereby prevents the cannibalization of
downstream bribes by upstream agents. Firms may frequently pay bribes but the nature of
the corrupt transaction is more predictable. A greater degree of certainty surrounds the
delivery of rights or services and firms are rarely confronted with unexpected demands for
bribes. As a result of the predictable bribe requests and subsequent delivery of services,

firms conduct more business and pay more in total bribes.5

 

’ While Batra, Kaufmann and Stone (2003) argue that unpredictable corruption is more harmful and
predictable corruption is more like business as usual, Lambsdorff (2002) makes the observation that
predictable corruption causes further corruption as it removes the incentive to seek alternative and legal
arrangements.

3.2 The Role of the Operational Environment on the Incidence and Size

of Bribes

ijrim/Irfmrtmrture: The quality of infrastructure such as power supply,
transportation facilities, telecommunications and water can play an important role in the
pervasiveness and costliness of corruption to firms. While the majority of studies focus on
corruption in big infrastructure projects (see Kenny 2006; Leary 2006), we want to propose a
different hypothesis that centers on the relationship between existing infrastructure and
bribery. In particular, we posit that poor quality infrastructure can actually present
opportunities for officials to extort additional payments. For example in most African
countries, unreliable supply of electricity is constantly identified as a major impediment to
productive activity, (World Bank Investment Climate Surveys). Because this service is so
critical to firms’ ability to function, we can conceivably see how officials could choose not to
provide the service until they were induced with an illicit payment. Thus the impact of
infrastructure quality on the cost of corruption is an interesting empirical question that we
are now in a position to explore.

Imiz'tutional Capaabr. While institutions cover a wide spectrum of activities, here we
are mainly interested in the aspect of effectiveness of the legal and regulatory frameworks.
Sullivan and Shkolnikov (2005) argue that weak legal systems, poor enforcement, complex
regulations and excessive discretionary powers provide a fertile base for corruption. The
confidence firms place in the ability of the legal system to enforce contracts and property
rights, for instance, will have an impact on the pervasiveness of bribery. The quality of the
judicial system can affect the likelihood that officials will punish corruption when it is
exposed and will affect the probability of firms believing they have recourse to seek

assistance from uncorrupt offices when faced with demand for unofficial payments. Bo et al

(2002) point out that honest and effective judiciary systems and courts increase the cost of
corrupt deals whereas if the justice system is Slow in responding to corruption and in
punishing perpetrators, the incentives for corrupt behavior will outweigh the costs. On the
part of regulations, unclear and/ or excessive regulations make it easy for officials to engage
in rent-seeking through extortion. In fact, corruption thrives in environments of complex
regulations, especially if they change frequently, and officials have too much discretion
(Sullivan et al. 2002; Klitgaard 1998). For instance \World Bank (2006) shows that in
Cambodia, overlapping regulations and too many administrative measures had created room
for excess discretion and rent-seeking. This then suggests that streamlining regulatory
procedures and simplifying laws so that there is no room for multiple interpretations or

discretion can help reduce corruption.

4. Data Description and Methodology

4.1 The Data

Trampareng' International, like other well known advocacy groups and NGOs,
produces measures of the ‘level’ of corruption within a given country, but does not collect
data on the organizational characteristics of the corruption regime. These regime-level
organizational characteristics are reflected in the Business Enw’mnment San/6y: (hereafter BES)
conducted by the \World Bank in 1999 and 2000, which provides information on the
frequency and size of the bribes paid by a varied assortment of firms across a broad sample
of countries. The BES followed a smaller survey conducted by the \World Bank in 1997 and

were designed to capture the perception of managers regarding the main obstacles to

production and growth faced by their firms. Approximately 100 managers6 in each of 74
countries located in five broadly defined world regions (i.e., Latin America and the
Caribbean, Eastern Europe, OECD, Asia, and Africa) were interviewed. The survey
spanned a broad set of topics such as infrastructure, regulations, the functioning of the
judiciary system, commercial policies and regulations and the prevalence of crime and
conupfion.

The survey comprises regional questionnaires that were similar in almost all respects
but for questions regarding bureaucratic red tape and “unofficial payments” to public
officials which were not included in the survey for Africa. In addition, the survey allows for
different responses to identical questions (i.e. qualitative versus quantitative) making it
necessary to exclude Eastern Europe from the sample in some of the regressions.

The BES contains a set of questions concerning the degree to which corruption
constitutes an obstacle to the growth of the firm, and a second group of questions regarding
the frequency and size of bribes. The first set of questions comprises: a) whether corruption
is a major, moderate, minor or no obstacle for the operation and growth of the business; and
b) which factor --financing, infrastructure, taxes an regulations, policy instability or
uncertainty, inflation, exchange rate fluctuations, functioning of the judiciary, corruption,
street crime, organized crime or mafia, and anti-competitive practices by government or
private enterprises-- is the single most important obstacle for the operation and growth of
the business. To get an approximate measure of the importance of corruption we create the
dichotomous variable car-mp, which assigns the value of 1 to those firms that answer major or

moderate to question (a), and 0 otherwise. As a measure of the importance of corruption

 

6 Only in a few cases like Belize, where the country size is small, the number of queried managers is
significantly smaller (e.g., 50 firms). Also the observations for estimation is considerably smaller than the
surveyed sample would suggest, mainly because of missing data i.e. non-response.

10

relative to other factors we compute the variable eone/ by assigning the value of 1 to those
firms that answer that corruption is the single most important obstacle and 0 otherwise.

Questions regarding the frequency and size of the bribes can be classified in two
groups. The first group of questions includes: a) how frequent is it for firms in a line of
business to pay some irregular ‘additional payments’ to get things done (freq); b) how many
times in the previous year did a government official request that the company pay an extra
payment to the electric power company (e/eetr), the telephone company (re/cyan), business
license authorities (bur/2r), tax agency inspectors (tartar), government procurement agents
(gozpro), customs, trade, or licensing officials (auto/n), judges or court officials (judges),
politicians (pa/if) or other government officials. The first question, (a), in the list above
generated a discontinuous variable that ranges from 1 (never) to 6 (always) and spans all the
countries in the sample. We transform this multinomial variable in a dichotomous variable
that takes the value of 1 if the answer is always or mostly, and 0 otherwise. The second
question was also included in all the questionnaires but was worded differently for the
Eastern European survey. Given this appreciable difference we elect to estimate the
regressions separately for the Eastern European firms and for the sub-sample with firms
from all other regions (Asia, Latin America and the Caribbean, and a subset of OECD
countries). Questions concerning the frequency of payments are simple count variables
ranging from 0 to the maximum reported number of bribes paid.

Questions regarding the size of “unofficial payments” comprise: (a) the average
percentage of revenues paid per annum in “unofficial payments” (lm'brev); and (b) the
percentage of the contract offered to the government to secure the former (bn'beon). W'e
stratify both questions in order to merge all the regions since in some cases the answer is

open ended and continuous while in others the answer takes on discontinuous values that

11

represent interval responses. The percentage of revenues ranges from 1 which corresponds
to 0%, to 7 for more than 25%, and the percentage of the contract ranges from 1 for 0% to
6 for more than 20%.

Unlike many other corruption data sets, the World Bank data has the advantage of
allowing for the identification of not only the frequency and size of the bribes, but also that
of the environment and expectations surrounding their payments. In particular, the survey
inquires whether: a) the manager knows in advance the Size of the ‘additional payment’
required to obtain the desired object or treatment (advance); b) once the payment is made, the
service is delivered as agreed (rem/e1) ; c) after the firm has made an ‘additional payment’ to a
particular government official the firm will need to pay another official for the same service
(addpgl); d) when a government agent acts against the rules a manager can go to another
official or the superior to get the correct treatment without recourse to unofficial payments
(nolm‘b). In addition, the surveys provide data on managers’ perceptions of the quality of
public services such as customs (qmrtom), judicial services (q/ndge), policy and armed forces
(amend, and infrastructure (qz'nfiar). The answers to these questions take on values ranging
from 1 (very bad quality) to 6 (very good quality). This allows us to explore the relationship
between the quality of the services provided by the government and the level of corruption.

The Banner: Environment Surveys are representative at the country level but do not
correspond to the composition of the country’s enterprises by economic sector or location.
In particular, the selection of the queried firms was designed to include at least 15% small

firms (with 5—50 workers), 60% service enterprises, 15% firms located outside the main

12

cities, and some enterprises with state or foreign participation.7 This fact should be kept in
mind when comparing summary statistics among different countries.

Table 1-1 and Table 1—2 present summary statistics and correlations for the variables
described in the paragraphs above. Several characteristics are noteworthy among the
relationships presented. More than half of all firms consider corruption to be a major or
moderate obstacle to the operation of the firm but only 1% list corruption as the mosr
important obstacle. Not surprisingly, the percentage of firm revenues paid in bribes per
annum (briberev) is highly correlated with the percentage of the contract offered as unofficial
payment to secure a contract with the government (bn'becon), as are the number and amount
paid in bribes. A more interesting set of positive correlations is that between the regularity
of paying bribes (freq) and both advance knowledge of the size of bribe necessary to induce
the desired treatment (advance) and delivery as agreed of the service for which the bribe was
paid (:erdel). Finally, effective recourse to government officials (nolm'b) is negatively
correlated with payments to government agents (bolt'tyfldges, et. al.), the regularity of paying
bribes (freq) and whether corruption is considered a major or moderate obstacle to the
operation of the firm. These correlations suggest a reasonable relationship between the
frequency of bribe payments and the efficacy of those payments and between various
measures of bribery across government agencies. The next section develops and tests
hypotheses designed to explain the nature of corruption across countries and its relationship

to firms’ experiences with bribery.

 

7 In order to deal with this problem we include controls for the aforementioned characteristics in the
regression analysis.

13

4.2 Methodological Issues

The potential unreliability of using perception based variables for estimation is not
new by any means. The main concern is the extent to which inference based on what is
basically subjective data is informative. Although this issue remains largely unresolved, a few
studies have shown that results from surveys on corruption/ governance perceptions can be
informative.8

Another potential concern of relying on the direct experiences of managers is an
individual manager’s perceptual bias. That is, managers’ perceptions may be an insufficient
basis upon which to base a sound assessment of actual corruption. This concern matters
when managers’ responses are systematically biased in one direction or when managers
prefer to hide or obfuscate their actual experiences for fear of reprisal from authorities.
Given the substantial efforts in the instrument design and in the training of the surveyors to
assure managers of the anonymity of their responses, systematically biased responses seem
unlikely and if anything would tend to understate corruption. Moreover, research on Banner:
Environment Sun/9'5 by Hellman et al, (2000) found no evidence that such survey data suffered
from systematic country bias. Finally, utilizing perceptions Of a corrupt regime to make
inferences about actual corruption is appropriate in the sense that the perception of
corruption matters for planning, investment and the ongoing operation of businesses. Thus,
while there are good reasons to believe that the surveys reflect actual conditions, they are
also useful for their depiction Of the business environment.

Finally in most of our estimations, we use information on the frequency of bribes

and the percentages of revenues paid out in additional payments rather than the perception

 

8 Kaufmann’s (2004) investigation on the quality of governance finds that stakeholders’ perceptions matter
in the same way as objective data from official statistics with a striking consensus between beliefs of
commercial risk rating agencies and those of individuals and firms in most countries. See also Kuncoro
(2006) for an interesting experiment on relationships between subjective beliefs and hard data.

14

of corruption. Even though these variables are indirect measures of corruption they
represent close proxies to the ideal objective data. Indeed, it is conceivable that managers

have good knowledge regarding the frequency and amount of the bribes paid.

5 Empirical Strategy

The primary objective of the paper is to study the role of the structure of the
corruption network in determining the costs of bribes borne by firms. Four variables
measure the level of corruption that may directly impact a firms’ cost structure: (a) the
perception that corruption is a major or moderate problem to production growth (cormp); (b)
the perception that corruption is the main single obstacle to the operation of the firm (correl);
(c) the frequency of payment (freq) and pervasiveness of bribes (brbpaj); and (d) the size of

bribes (brbrz'ge).

p(y..
1!

For cormp, corre/ and brbpa}, we estimate a probit model given by:

 

X..,W..,Z..)=G(c.+X..a+W..fl+Z..5+v..) (1)
1] 'J U J 1J 1} 1] ll

where i indexes the firm; J indexes a cluster defined as firms that belong to the

same country, sector, and size; G(°) is the standard normal cumulative distribution

function; Y--is corrup, correl, or freq for manager 1 in cluster 1 ; X - -is a matrix containing the

11 11

variables that reflect the organization of the corruption system; - is the perception of the

1']

quality of various public services; Z - -

1]

contains firm specific characteristics such as age,

government and foreign participation; C - are cluster specific fixed effects; and V" is the

J U

CHOI’ term.

15

Both corrup and cone/ reflect the perception of the managers regarding corruption,
either by itself or relative to other obstacles to growth. Similarly, the measure of
infrastructure represents the perception of the manager with respect to physical facilities and
services such as roads, mail, telephone, electricity, water and sewage (qt/fray). Given that
managers who have to pay bribes to get these services delivered would most likely not rate
them highly, our infrastructure measure could be endogenous. Thus, we use the proportion
of firms that report infrastructure to be the single most important obstacle, relative to the
firms that report other factor to be the main obstacle to instrument for the perceived quality
of infrastructure.

Our estimation strategy combines two procedures, Rivers and Vuong (1989) and IV-
probit. Rivers and Vuong henceforth (RV) is essential a two—stage estimation procedure
whereby in the first stage we perform an OLS regression of the endogenous variable on the
instrument and all the other explanatory variables and obtain the predicted residual. In the
second stage, we do a probit regression of the dependent variable on the predicted residual
from the first stage regression together with the endogenous and exogenous variables. An
appealing attribute of the RV procedure is that the probit t—statistics on the predicted
residual provides a valid test for the exogeneity of infrastructure. We report the test results in
Table 1-3 under the heading Endogenez'g' test. If we find evidence of endogeneity we then
proceed to estimate the IV-probit models via maximum likelihood estimation (MLE). Also
to quantify the impact of the measures of the corruption regime, as well as other factors of
interest on the perception of corruption, we compute and report the average partial effects
(APES) of the variables on the different dependant variables.

With respect to the frequency of bribery by public officials in different sectors, the

dependent variable is a count variable, which represents the answer to the question “how

16

many times in the past year did one of the government officials required bribes?” Therefore,

and firm specific characteristics (Zi -)

ij)’ J

we maintain the same corruption organization (

as covariates, but we specify the model as a Poisson regression of the form

p(y.. =
1J

wherexfl=c.+X..a+Z..5+u.. (3)
J U ’1 1]

 

h Xij’Zij] = exp[— exp(xB)]exp[(xB)]h /h h = 0,1,... (2)

An intriguing message of the Schleifer and V ishny paradigm is that corruption may
be more of a deterrent to commerce and investment where government agents collect less
total bribe revenue. The implication is that true costliness of corruption lies as much or
more so in the uncertainty it creates for firms than in the financial burden it imposes on
them. Bribes paid to organized corruption networks are akin to taxes, costly but predictable,
whereas bribes paid in disorganized networks are smaller but nevertheless debilitating in the
uncertainty that surrounds them.

If the Schleifer and Vishny characterization is accurate, one would expect larger total
bribes payments by a firm to be associated with higher levels of perceived certainty and
efficacy surrounding the corruption network and vice versa. We investigate this
hypothesized relationship by using as the dependent variable a survey question that asks
managers to estimate the percentage of their firrn’s annual revenues typically paid in bribes
or the percentage of a government contract’s value paid as a bribe.9

Recall that we have transformed the ‘bribe size’ variables by grouping the data into

intervals, so as to make the answers comparable across regions. Therefore, while the

 

9 Kaufmann (2000) explains why in this survey instrument firms were asked to estimate bribes as a percent
of annual revenues rather than profit; because there is less uniformity across countries regarding what
counts as profit, revenues are generally easier to estimate and firms have an incentive to misreport profit
estimates.

17

empirical model uses the same covariates as in equation (1), we estimate an interval

regression model whose likelihood function is given by:

bsizegb—c—Xa—Wfl—Zé' bsize’b—c—Xa—Wfl—Zé'

 

 

i'
L = Zlog J — ¢ 1 (4)
u 0 0
Here bsizeg) and bsizegb indicate the lower and upper bounds of bribery paid by
the firm, respectively.
6 Estimation Results
6.1 Organized and Disorganized Corruption

Tables 1-5 and 1- 6 report the estimation results of the probit models, where the
dependent variables are corrup and cone], respectively. For ease of comparison, we report the
results of the OLS and ZSLS regressions together with the probit models and the APES in
these tables. As mentioned previously, given the large differences between the coding of
responses for Eastern Europe and the other regions, we have stratified the sample into two
groups, one comprises Eastern Europe and the other all other regions. \X’ e estimate the
regressions separately for the two sub-samples. With some differences in the importance of
individual factors, overall the results hold across the sub-samples. Because of the differences
in conducting the surveys, the results will only be considered within a sub-sample and cannot
be informative for comparative analysis of the two sub-samples.

Notice first that the t—statistic on the residuals in the RV second-stage regression
indicates that we cannot reject the null that infrastructure is endogenous (Table 1-3). Also

the first stage regression shows that the instrument is highly significant with a t-statistics of

18

2.99. Therefore, whereas we report the OLS and probit estimates as benchmarks, we focus
our discussion on the IV estimations. These results suggest that the organization of the
corruption network plays an important role in the perception managers have of corruption
as an obstacle to the operation of the firm. In particular, the probability of corruption being
reported as an obstacle (corrnp) increases when the size of the unofficial payment is known in
advance and when firms are expected to make multiple payments for the same service. The
coefficients for these variables are positive and significant at the 5% level. Also the first stage
regression shows that the instrument is highly significant with a t-statistics of 2.99.
Therefore, whereas we report the OLS and probit estimates as benchmarks, we focus our
discussion on the IV estimations. In line with our expectation, having recourse to another
official that can provide the service without a bribe decreases this perception. The parameter
estimate is significant at the 5% level with an APE of 0.016. That is, having recourse to an
uncorrupt officer results in a 2 percentage point decline in the probability that a firm
considers corruption a constraint to performance. Given that the average (median) response
ranges from 0.05 (0.02) for OECD countries to 0.48 (0.04) for Eastern Europe, this effect is
not negligible.

Similarly for Transition Europe, the structure of the corruption network matters for
the perception of corruption being a constraint to firm performance. Advance knowledge of
the required payment increases this perception with an APE of -0.052. Note for this
particular case the negative effect implies a positive relationship because the responses for
con-1W start from the lowest rating (corruption is a major constraint) to the highest rating
(not an obstacle) whereas the explanatory variables are ordered in the opposite direction.
Service delivery and recourse to an uncorrupt government agent also have a positive effect

with APEs of 0.030 and 0.067, respectively, and are Significant at the 5% level.

19

The results are much weaker for the belief that corruption is the single most
important constraint to firm operations (correl), with only the possibility of getting correct
treatment having a significant impact.

Consider now the effect of the corruption network on the pervasiveness of bribes
(how common is it for firms to be required to make unofficial payment). The results suggest
a strong effect of the corruption network, although the coefficients on service delivery and
recourse to an uncorrupt official are only significant at the 10% level. Three main
conclusions can be drawn from the estimation results presented in Table 1-7. First, the
pervasiveness of unofficial payments is higher when the corruption network is organized.
The probability that a manager will answer that bribes are common or always paid in her
firrn’s line of business is significantly higher when the size of the payment is known in
advance and when the service that is the object of the bribe is usually delivered as agreed
after the payment is made. With respect to Eastern Europe, there is also evidence of the
importance of the network organization with both advance knowledge of the size of the
bribe and the likelihood of having to make multiple payments Significant at 5%. The APEs
are 0.053 and 0.045 for advance and mn/ngr, respectively; thus suggesting increases of roughly
5 percentage points in the probability that a manager will respond that bribes are common in
her line of business.

Across a wide range of countries and regions, firms seem more willing to pay bribes
when the outcome of doing so is more predictable. This result supports numerous
qualitative studies that have long suggested that corruption is a regular part of business in
many countries and is indeed not unlike taxation. Second, the frequency of the bribes is
lower in environments where the correct treatment can be obtained from a superior or

another government official without recourse to bribes. This result suggests that investment

20

into creating effective avenues of recourse and monitoring systems might curtail the
incidence of bribery and, indirectly, that bribery is more common where it is accepted and
systemic rather than the result of individual opportunism. Third, as one would expect, the
propensity of numerous officials demanding unofficial payments for the same service
increases the frequency of bribes.

One important offering of the results shown in Table 1-7 is that the frequency of
bribes can be large for very different reasons. Regular bribes can result from highly
predictable environments or because multiple bribes are required in uncertain environments.
Thus, knowing only that bribery is equally frequent in two countries does not allow one to
meaningfully regard the two countries as equivalently or even similarly corrupt. However,
our results do suggest meaningful differentiation among apparently similar corrupt
environments and reveal the conditions that determine the frequency of bribery.

Other interesting tests concern how the frequency of bribes differs according to the
service sector. These results are presented in Tables 1-10 to 1-17 and, though not as strong
as the results in Table 1-7, the results of these regressions do provide statistical evidence that
an organized corruption network is related to higher incidence of corrupt transactions. This
relationship is significant for customs and trade officials, business licensing, judges and
court, and tax agency inspectors. This finding is in line with those of other studies. In
particular, Svensson (2002) using survey data on Ugandan firms finds that those firms that
deal with government officials typically paid bribes while those firms that operate in sectors
with no contact with the public sectors paid no bribes. Further, these results are generally
consistent across both sub-samples. Both the ZSLS regression coefficients and the APEs
suggest that economically, advance knowledge of the size of the bribe and the expectation of

multiple payments for the same service tend to be more important across sectors than

21

service delivery and service without bribery (nobrz'lze). The two have APEs of 0.05 and 0.03
respectively compared to APEs of around 0.01 for the other two measures of corruption
organization.

Tables 1-8 and 1—9 show the effect of the corruption network and the perceived
quality of public services on the total size of bribes. In fact, for both Eastern European
firms and the other regions, the results Show that the percentage of a firm’s revenues spent
on bribes is indeed higher the more organized the network of corruption into which bribes
are paid. That is, firms are willing to commit a larger portion of their revenues to unofficial
payments when the size of the payment is known in advance. On the other hand, the
opportunity to obtain the service from another government official without recourse to
bribes reduces the size of the latter.

Overall, our estimation results provide evidence in support of the hypothesis that
organized corruption networks are associated with a higher incidence of corruption,
measured in terms of frequency of unofficial payments and the size of bribes (i.e. percentage
of the revenues and percentage of the contract). In other words, firms are willing to pay
more in total bribes and more frequently pay bribes the higher the certainty regarding the

efficiency of and return from participating in the corruption network.

B. Quality of Public Services and their Eflect on Corruption

Apart from the relationship between the size of bribes and the nature of corruption
networks we consider how some firm and environmental features influence the size and
regularity of bribes. This general line of inquiry has been substantially advanced in recent
years. Using country-level indices of corruption Ades and DiTella (1999) find that
corruption is higher in less open economies and where the supply of bribers is small relative

to available rents. Similarly, Treisman (2000) finds that openness to trade lowers measured

22

corruption as do a host of legal and historical factors: the degree of federal control, tenure as
a democracy, the proportion of Protestants. Another, perhaps more intriguing aspect of
corruption is its effect on the environment where firms operate. Mauro (1999) finds that
corruption skews public investment away from education and health-related investments
towards larger infrastructure projects where kickoffs are larger and easier to obtain. Tanzi
and Davoodi (1997) find that corruption reduces maintenance and operational expenditures
on infrastructure, raises expenditures on new equipment, reduces the productivity of public
investments, and crowds out private investment.

It is not so surprising to find that corruption and poorly maintained public
infrastructure go hand in hand as it is well known that developing countries often exhibit
high levels of corruption (Mauro 1995), strained public sector budgets, and low quality
infrastructures. But weak infrastructures also provide opportunities for corrupt officials to
extract bribes from firms competing for the timely provision of infrastructure services. Like
customs officers, agents who control access to infrastructure services have regular
opportunities to impose costs on firms that do not pay bribes. Indeed, Gaviria (2000) finds
that bribes related to public infrastructure services (cg. telephone and power companies) are
among the most commonly reported category of bribes in Latin American countries. Thus,
it seems reasonable to suggest that corruption both arises from the opportunities created by
weak infrastructures and retards efforts to improve them, thereby creating a vicious cycle of
corruption and poor quality infrastructure. We consider one half of this circular relationship
by analyzing how firm’s perceptions of the quality of various public services influence the
amount of bribe they pay.

In order to study this relationship we include the perception of the managers

regarding the quality of various public services as additional explanatory variables in the

23

aforementioned empirical model. The results presented in Tables 1—5 to 1-17 allow us to
draw some interesting conclusions. First, the results Show that the quality of public services
matter in a non-negligible way for the perception of corruption as a constraint to firm
operations, the proportion of revenue that firms pay to get things done and the perception
of corruption as the single most important obstacle. These results suggest that the incidence
of corruption could be reduced with investment directed at increasing access to public
services, improving the quality of institutions (such as better monitoring, e.g., security) and
improving the efficacy of the judiciary system. For example, the results suggest that corrupt
officials would request and engage in fewer corrupt transactions if they faced a higher risk of
being caught and punished for doing so.

The results presented in Tables 1-5 and 1-6 Show that, for regions other than Eastern
Europe, all measures of the quality of public services have a significant impact on the
probability that corruption will be an obstacle as well as on the likelihood that corruption
will be considered as the single most important constraint. The impacts are also
economically significant as shown by the APES that range from 0.031 for the quality of the
judiciary systems to 0.364 for the quality of infrastructure. For Eastern Europe only the
quality of customs services and judiciary systems matter. Contrarily the results for Eastern
Europe are very Strong for prevalence of bribery. With the exception of the quality of
customs, all other measures of public services have an impact that is significant at 5% with
APES ranging from 0.035 for security to 0.085 for the quality of infrastructure. Only the
quality of custom services is significant in the case of all other regions.

The one aspect of this category that is at first puzzling is the impact of physical
infrastructure. Our hypothesis is that a higher perception of the quality of facilities would be

associated with lower perceptions of corruption. Our results, however, suggest otherwise.

24

For all other regions, the coefficients on infrastructure are positive, highly significant and
economically sizeable with APEs in excess of 0.250 for com/p and correl. For Eastern Europe,
the impact of the quality of infrastructure on the pervasiveness of corruption is also positive
but relatively lower at an APE of 0.085 which is still not at all a small impact. The question
of interest is whether this positive impact of infrastructure on firms’ cost structures leads us
to conclude that improvements in the quality of physical infrastructure necessarily increase
corruption. This is clearly counter-intuitive. \W e offer an alternative explanation for the
somewhat “perverse” results. The estimations could be picking up some non-linearities we
had not accounted for. Consider a situation where the quality of infrastructure is so poor
that even the payment of a bribe would not improve access to the service. In a situation like
this, it is possible that firms would not be willing to make unofficial payments as the agents
would not be able to deliver. However, as the quality of infrastructure improves, ability to
deliver also improves, thus making it more worthwhile for service users to grease palms, up
to some threshold beyond which impediments to access become so few as to negate the
need to bribe. The results for the tests of non-linearities presented in Table 1-18 Show
evidence of the existence of such non-linearities. For regions other than Eastern Europe,
the coefficients on both the linear and the quadratic terms for conic/9 and corre/ are significant
at the 5%. In both cases the level term has a positive effect and the quadratic term a negative
impact. Noteworthy is the fact that the inclusion of the quadratic term wipes out the impact
of the other operational environment indicators. This in essence captures both the strong
correlation between the measures of the operational environment and the dominance of
infrastructure within this class of indicators. However, jointly the measures are significant at
the 5% level. For Eastern European firms, the non-linearity exists for [arbpajI but not for

comp with both the level and quadratic terms significant at the 5% level. The possibility of

25

this relationship suggests that country strategies for growth should necessarily include
investment in infrastructure as a way of ultimately doing away with the need to resort to
making unofficial payments.

The results in Tables 1-8 and 1-9 indicate that the quality of public services matter
for the size of the bribes. In particular higher perceptions of the quality of customs and
security reduce the size of the bribes with coefficients that are significant at the 5% level and

also economically significant at 0.277 and 0.327, respectively.

C. Corruption and Firm Characteristics

We briefly examine three other relationships between firms and bribes: the influence
of national origin, firm size and sectoral affiliation. For many years multinational firms have
sought ways to limit their exposure to corrupt practices through the use of internal codes of
conduct and participation in anti-corruption campaigns.10 Whether solely through internal
governance or in conjunction with multilateral agencies like the OECD, multinational firms
have struggled, often unsuccessfully, to avoid participation in corrupt transactions and their
ill effects (Gordon and Miyake 2001). Still, there are good reasons to believe that foreign
firms have different experiences with corruption than domestic firms do. Large US
multinationals report that their ‘zero-tolerance’ policies on corruption are costly but work
well after a period of just a few years (Gordon and Miyake 2001). Nevertheless, some firms
have decidedly opposite experiences with their foreign operations. Presumably because
foreign firms are often cash—rich, naive regarding local customs, or have less recourse to
domestic courts, they suffer from a ‘liability of foreignness’ and are confronted with more

corruption than their domestic counterparts (Brewer 1997). Hellman,Jones and Kaufmann

 

1" Surely multinational firms have also sought to initiate and profit from corrupt transactions. However,
there is ample evidence to suggest that most multinationals believe the costs of corruption to be far greater
than any potential benefits.

26

(2002) find that the prevalence of foreign firms to engage in corruption depends on the type
of corruption and the pervasiveness of corruption in host countries. Specifically where
kickbacks for procurement are uncommon, foreign firms are more likely to engage in this
type of corruption than domestic firms, whereas in countries with significant state capture,11
the propensity of foreign firms to pay the relevant bribes almost doubles. However, in terms
of the level of bribes, both foreign and domestic firms pay Similar percentages of their
revenues, on average. In the absence of a deterministic model we cannot and do not claim
strong support for one relationship or the other but rather empirically test whether the
amount that foreign firms’ pay is meaningfully different from that paid by domestic firms.
The two remaining hypotheses are straightforward and simple. Compared to small
firms, large firms more frequently possess influence with local authorities and are more
individually important to the local economy. Consequently, large firms are thought to be
less vulnerable to the exploitative demands of corrupt officials.12 Moreover, to the extent
that bribes are rather uniform in size, or that there is an equilibrium bribe rate across all
firms, they will constitute a smaller proportion of the revenues of large companies. This
reasoning suggests that bribes should rise less than proportionately with firm size. Another
possibility is along the finding of Svensson (2003) that the size of the bribe is correlated with
firms’ ability to pay. Assuming that larger firms have a higher ability to pay then it will be the
case that these firms pay more in bribes than smaller firms. Thus, depending on the effect

that dominates, the impact could be positive or negative.

 

1‘ State capture refers to bribes that firms pay to government officials with the specific objective of
influencing laws, regulations and policies of the government to their own advantage.

’2 Svensson (2003) finds that existence of outside options increased firms’ refusal power but higher ability
to pay captured by current and fiiture profitability increased the likelihood of being targeted by officials and
the likelihood that they will pay the bribe. Note this depends on an assumption that officials can correctly
assess the profitability.

27

Lastly, there is reason to believe that being either a manufacturing or service firm
might have a bearing on how vulnerable the firm is to bribery demands, but a priori it is not
clear which sector should be more exposed. Valid arguments can be made for either sector
being more susceptible to corruption. For instance, some service industries can be more
exposed to corruption through their use of physical infrastructure and also their being
subject to inspections. A case in point is the hotel industry and other hospitality services.
Availability of telephone, water and electricity services is absolutely critical. for firms in this
industry to function. Moreover, by their association with the public, it is easy to subject them
to numerous inspections, legitimate and otherwise. Anecdotal and statistical evidence suggest
that officials prefer to prey on hotels and restaurants and on service firms than
manufacturing firms, Kuncoro (2006). On the other hand, Safavian and Graham (2000)
postulate that manufacturing firms’ production is more visible which increases their
vulnerability to regulatory authorities as they have more difficulty hiding their output and
sales making it easy for corrupt officials to assess their ability to meet bribe demands. Also,
manufacturers’ input and outputs more often require the approval of customs officials, their
energy and infrastructure demands are higher, and their operations are more easily and
expensively interrupted (e.g. Specific assets like manufacturing facilities are irreplaceable in
the short run).

The results reported in the bottom rows of the tables suggest that firm size matters
for the size of bribes (measured as a percentage of revenue), but not for firms’ perceptions
of corruption as an obstacle. Specifically, being a medium firm increases the size of bribes,
possibly reflecting the fact that these firms have a higher ability to pay than small firms
rather than their being susceptible to being targeted for bribes. Manufacturing firms pay a

lower percentage of their revenues in bribes.

28

Finally, foreign ownership matters only for the size of the bribe and the perception
of corruption as a constraint, which is higher for foreign firms but only at a 10% significance
level for regions other than Eastern Europe. The positive impact on perception of
corruption can however be the result of different levels of tolerance. Arguments have been
made that foreign firms - particularly when they originate from the first world - tend to be
less tolerant of corrupt practices than domestic firms and as such could reasonably be

expected to have higher perceptions of the incidence and pervasiveness of corruption.

7 Conclusions

We began this paper by asking a question that could not have been addressed just a
few years ago: how does the nature of corruption influence its costliness to firms? Country-
level indices suggest that corruption imposes a degree of burden on economic activity but,
by their very nature they cannot describe the determinants of firms’ experience with
corruption in a given regime. In the absence of the finer details of corrupt transactions we
are unable to characterize differences among countries measured as equally corrupt or even
to be precise about what it means to say that one country is more corrupt than another
country. The firm-level data employed in this Study offers a valuable opportunity to expand
our understanding of the variety of corruption regimes and the characteristics that mediate
firm’s experiences within them.

Our most general finding is straightforward and robust. Across a broad sample of
firms in a diverse group of economies we find strong evidence that the nature of corruption
does influence its costliness to firms: predictable and effectual corruption regimes increase
the frequency of bribery and the total monetary cost of corruption to firms. Specifically, the
probability that a manager will report that bribes are frequently paid in his firm’s line of

business is significantly higher when the size of the payment is known in advance and when

29

the service that is the object of the bribe is usually delivered as agreed after the payment is
made.

Similarly, we find that the monetary costliness of bribes rises with the predictability
of the corrupt regime. This is true whether the total monetary cost of bribery is measured as
a percentage of firm revenues or as a percentage of the value of a government contract.
These results support the general message of Schleifer and Vishny (1993) that organized
corruption regimes are able to extract more bribe revenues than disorganized regimes.
Moreover, the results suggest that, in and of themselves, high measured levels of bribery may
overstate the burden placed upon firms since high levels of bribery go hand in hand with
highly predictable and effectual corruption regimes. The lack of a coordinated network of
government agents will also increase bribe frequency. We find that bribery is more frequent
where multiple officials muSt be bribed for the receipt of the same service. This result does
not imply that total bribes are higher, only that, like predictability, the inefficacy of a single
bribe raises bribe frequency. Not surprisingly, we also find that both the frequency and total
amount of bribe payments fall when there is effective recourse to the superiors of corrupt
government agents. Thus, the monetary costliness of corruption is assailable through
effective monitoring.

In addition to examining the influences of the organization structure of corruption,
our analysis suggests that inadequate or ineffective government expenditures increase the
cost of corruption to firms. Low infrasrructure quality raises both the frequency of bribery
and the total amount firms spend on bribes. This result implies that investment into
infrastructure could decrease the costliness of bribery and, in conjunction with the Tanzi and

Davoodi (1997) relationship, that there may be a circular relationship between bribery and

30

infrastructure quality: corruption leads to lower investment into infrastructure which
increases the opportunities for bribery and cost of corruption to firms.

Finally, we find some evidence that, all else equal, foreign firms pay slightly less in
total bribes than domestic firms do but seem to more frequently pay bribes to various public
service agencies. In addition, manufacturing firms appear to pay less and larger firms more
in total bribes (as a percentage of revenue) than non-manufacturers and smaller firms do.
With regard to paying bribes to government officials, manufacturers appear to pay bribes
more frequently than non-manufacturers do and smaller firms appear to pay bribes less
frequently than larger firms do.

With the continued examination of corruption in its many natures we are sure to
learn much more about the diversity of corrupt regimes throughout the world and how
better to combat them. Our efforts here have provided a step forward in describing firms’
experiences with bribery within such variegated environments. More research on ever finer
and more detailed data is needed to understand how best to design governmental and firm
level policies so as to ease the burden of corruption on firms, peoples and economic

development.

31

REFERENCES

Ades, A. and R. DiTella, (1999), “Rents, Competition and Corruption”, American Economic
Review 89, pp. 982-93.

Bardhan, P. (1997), “Corruption and Development: A Review of Issues”, journal of Economic
Literature 35, pp. 1320-46.

Batra, G., D. Kaufman and A. Stone (2003), “The Firm Speaks: \X’hat the World Business
Environment Survey Tells Us about Constraints on Private Sector Development”,
World Bank, Washington DC.

Besley T. and J. MacLaren (1993), “Taxes and Bribery: The Role of \X’age Incentives”,
Economic journal 103, 119-41.

Friedman, E., Johnson S., Kaufmann D., and Zoido-Lobaton P. (2000), “Dodging the
Grabbing Hand: The Determinants of Unofficial Activity in 69 Countries”, journal of
Public Economic: 76, 459-493.

Frye T., and E. Zhuravskaya (2000), “Rackets, Regulation and the Law”, journal of Law
Economic: and Organization: 16, 478-502.

Gaviria, A. (2000), “Assessing the Effects of Corruption and Crime on Firm Performance”,
Manuscript, Fedesarrollo, Bogota, Colombia.

Hellman, J.S., Jones, G., Kaufmann D., and M. Schankerman (2000), “Measuring
Governance, corruption and State Capture: How Firms and Bureaucrats Shape the
Business Environment in Transition Economies”, Public Poligr Re:earcb Paper 2312,
The World Bank.

Hunt,J and Sonia Laszo (2005) “Bribery: Who Pays, \tho Refuses, What are the Pay-offs?”,
The William Davidson Institute, Working Paper No. 752, University of Michigan.

Johnson, 8., Kaufmann D., McMillan J., and C. Woodruff, (2000), “Why Do Firms Hide
Bribes and Unofficial Activity after Communism?”, journal of Public Economic: 76, 495-
520.

Kaufmann, D and A. Kraay (2003), “Governance Matters III: New Indicators for 1996-2002
and Addressing Methodological Challenges”, World Bank, Washington, DC.

Kaufmann D. and S. Wei (1999), “Does “Grease Money” Speed Up the \Wheels of
Commerce?” NBER Working Paper 7093

Klitgaard, R. (1998), “Strategies against Corruption”, Agencia Espafiola de CooperaciOn
Internacional

32

Kuncor, A. (2004), “Bribery in Indonesia: Some Evidence from Micro-Level Data”, Bulletin
of Indone:za Economic S tua’ie:, Vol. 40, No. 3, pp 329—54.

LaPorta R., Lopez-de—Silanes F., Schleifer A., and R. Vishny, (2000), “The Quality of
Government”, journal of Law Economic: and Organization: 75, pp. 222279

Mauro, P. (1995), “Corruption and Growth”, Quarterl I journal of Economics 110, 681-712.

Mocan, N.H. (2007), “What Determines Corruption? International Evidence from Micro
Data”, NBER Working Paper No. W10460.

Olken, B., (2006), “Corruption Perceptions vs. Corruption Reality”, Harvard University and
NBER

Reinika, R. and J. Svensson, (2002), “Measuring and Understanding Corruption at the Micro
Level”, Unpublished Manuscript

Rose-Ackerman, S. (1978), Comrptz'on: A Study in Political Economy, New York: Academic
Press.

Schleifer, A. and R.W. Vishny (1993), “Corruption”, Quarterbl journal of Economic: 108, 599-
617.

Svensoon,J. (2003), “Who Must Pay Bribes and How Much? Evidence from a Cross-Section
of Firms”, Quarterly journal of Economic: 778, pp. 207-230

Tanzi, V. (1998), “Corruption around the World: Causes, Consequences, Scope and Cures”,
IMF Working Paper 98/ 63.

Tanzi V., and H. Davoodi (1997), “Corruption, Public Investment and Growth”, Ill/IF
Working Paper 97 / 739.

Treisman, D. (2000), “The Causes of Corruption: a Cross-National Study”, journal (f Public
Economic: 76, 399-457.

Voung, Q. (1989). “Likelihood Ratio Tests for Model Selection and Non-nested
Hypothesis”, Econometrica 5 7, 307-333

Wei 5. (2000), “Bribeol in tbe Economies Grea:e or S and? ”, Brookings Institute,
Washington DC.

Wu, X. (2005), “Firm Accounting Practices, Accounting Reform and Corruption in Asia”,
Polly and S oa'egy, 24(3), pp. 53-78.

Wooldridge, JM. (2001), “Econometric Anabm': of Cro:: Section and Panel Data”, The MIT Press.
Cambridge, Massachusetts.

33

Table 1-1
SELECTED SUMMARY STATISTICS

 

Asia O.E.C.D. Transition Latin America
(Excl. Mexico) Europe & Caribbean

Magnitude known in advance (advance)

Always or Mostly 28.8 14.5 10.8 18.2

Average response 3.4 2.8 3.6 2.8
Service delivered as agreed (serdel)

Always or Mostly 25.1 39.7 9.4 47.6

Average response 3.2 4 2.6 4.2
Additional payments required

Always or Mostly 16.6 10 41.1 17.6

Average response 2.9 2.1 4.1 2.7
Can go to another official to get service without paying bribe

Always or Mostly 36.6 35.9 40.7 23.1

Average response 3.7 3.3 3.9 2.9
Corruption as an obstacle to the operation of the firm

Always or Mostly 20.7 16.5 45.2 56.4

Average response 2.3 1.6 2.4 2.8
Corruption as the main single obstacle to the operation of the firm

Average response 1.6 1.9 n.a. 3.9
Common to pay additional payments (freq)

Always or Mostly 20.5 4.8 48 16.6

Average response 3 1.9 4.3 2.6
Percentage of revenues paid in bribes

10% or more 12.5 1.7 15.1 7.1

Average response 2.6 1.3 3.2 1.9
Percentage of the contract paid

More than 15% 4 1.2 27.8 2.5

Average response 2.1 1.2 3.4 1.4
Frequency of bribes - Mean

Power Company 0.2 0.2 n.a. 0.5

Telephone Company 0.2 0.1 5.4 0.5

Business Licensing 0.8 0.2 5.1 0.5

Tax Agency Inspectors 0.6 0.1 5.3 0.7

Government Procurement 0.2 0.1 5.6 0.4

Customs and Trade 1.9 0.1 5.7 1.3

Judges or Court Officials 0.1 0 5.6 0.3

Politicians 0.2 0.2 5.9 0.3

 

34

 

 

888 88.8 88.8 888 $88 88.8 88.8 888. 8:8 888 828 82:5
8:8 8.28 888 88.8 88.8 88.8 82.8 38.8. 888 88.8 $8.8 33:5
898 88.8 :88- 88.8 88.8 28.8 88.8 :88- 88.8 8.8.8 2.8.8 58
88.8 88.8 888 8.8.8 388 88.8- 88.8 888 38.8 888 88.8 883
8.2.8 82.8 888 88 $88 88.8 28.8 28.8- :8 88 228 8925
88.8 83.8 85.8 88.8 82.8 388 $88 88.8- 88.8 88.8 288 89%
_ 88.8 ~88 4.2.8 888 888- .88 88.8- 828 388 R88 893
_ 88.8 38.8 88.8 88.8 888- 888 3.88 :88 88.8 27.3
_ :2 ~88 8.8.8- 888 88.8 88.8 88.8 88.8 £23
_ $8.8 $88 888 2.8.8- 8:8 888 :88 82s
_ 8.8.8 828 88.8- 888 88 58.8 95
_ 3.28 88.8 88.8 88.8 88.8 E8
_ 5.8- 838 $8.8 $8.8 mass
_ $8.8 8.8.8 88.8 €28
_ 8.2.8 88.8 is;
_ 88.8 .83...
w UUCN>UN

$523 2745 .328 boo—o Uni Theo @9950 £50: hump—om 323. 853?“

 

mcocflobou vogue—om

Nun each.

35

5858 8: 888 8 EoEEo>ow 9 3mm 5.8.50 5 598588: 858:5
8:5 mm 35 8.5.8 .5 owSCoubm 588:5

53258: 5 18578 8:5 35¢

£3qu 8585 5 mow—8; .5 5558 8:5 ”mow—oi

£555 wcugu: 5.8 5 28083 .5 @8858 8:5 ”F583

Edema 5558355 58:585w .5 58:58 8:5 ”89.0w

89855: is .3 @8578 8:5 8553

55mm wcmmcou: 88:85 5 @8578 8:5 6:55

55:55 0.5.528 .3 58:58 8:5 ”.528

5.588 5655:. .5 @8858 8:5 ”boo—o

88:85 5 8.: 2.: E m8:5 >8: 9 .5558 58.:

8.2350 8:5 05 9:58 22350 5859:: 808 295 8: m: 5085.55 ”—8.55
535% 9 25.350 885.55 5 57:8 8 Canasta 5.855

8:5 .5 505me 50539 8:38 How 9 35.5 555.3 9 ow :5 8:598
8:5 .5 “55.25 5:... @8578 589:5 1.58:5.“ “>855.“

8:55 589:8: 88mm 58w... m... @8250: 8:68 ”—5.88

08838 c: 5.58.. 5855 5 58585 6855.“

 

. 484.8 88.8 88.8 .888 8:8 82:5
. $28 88.8 838 88.8 85.5

. E88 38.8 88... £2.

. 888 $88 883

. 8:8 89.25

F 895w

 

.58:5 55:5 85% mumpi 8955 895w

 

gown—8800 vogue—om

 

3.5.2.8 3 «ES.

36

Table 1-3
First-Stage Rivers and Vuong Regressions

 

 

Coefficient SE
Instrumental Variable
Q44b_p -0.619 0.207**
Exogenous Explanatory Variables
Advance Knowledge -0.014 0.022
Service Delivery 0.008 0.021
Multiple Payments —0.038 0.016**
Recourse to Uncorrupt Officer 0.006 0.015
Quality of Customs Services 0.148 0.027**
Quality of Judiciary Systems 0.045 0.03
Security 0.246 0.028**
Manufacturing -0.064 0.043
Small 0.133 0067*
Medium 0.008 0.065
Government Ownership 0.082 0.1 12
Foreign Ownership -0.032 0.064
Age 0.002 0.001“
Number of Observations 717
R-squared 0.35

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

37

Table 1-4
Second-StaggRivers and Vuong Regressions (:I' est for Exogeneity)

 

 

 

brbpay corrup

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.283** 0.055 0.189** 0.040
Service Delivery 0.122** 0.058 -0.018 0.043
Multiple Payments 0.170** 0.051 0.167** 0.054
Recourse to Uncorrupt Officer -0.131** 0.049 —0.099** 0.033
Public Services
Quality of Customs Services -0.261 0.185 -0.274** 0.130
Quality ofjudiciary Systems 0.004 0.069 -0.189** 0.054
Quality of Infrastructure 0.924 1.085 2.218** 0.749
Security -0.197 0.267 —0.623** 0.194
Firm Specific Characteristics
Manufacturing 0.108 0.144 0.187 0.116
Small -0.165 0.237 -0.27 0.197
Medium -0.034 0.161 0.032 0.155
Government Ownership -0.434 0.402 -0.655** 0.302
Foreign Ownership -0.092 0.213 0.349** 0.144
Age -0.006 0.004 -0.003 0.003
Residuals -1.064 1.087 -2.254** 0.764
Observations 706 700

 

38

Table 1-4 continued

 

 

 

correl

Coefficient SE
Corruption Regime
Advance Knowledge 0.1 15* 0.064
Service Delivery -0.l6** 0.067
Multiple Payments 0.051 0.088
Recourse to Uncorrupt Officer 0.269** 0.061
Public Services
Quality of Customs Services —0.569** 0.241
Quality of Judiciary Systems -0.207** 0.102
Quality of Infrasrructure 3.176** 1.367
Security -0.692* 0.368
Firm Specific Characteristics
Manufacturing 0.139 0.207
Small -0.785** 0.286
Medium -0.509** 0.184
Government Ownership -0.788** 0.464
Foreign Ownership -0.199 0.244
Age 0.004 0.005
Residuals -3.116** 1.365
Observations 573

 

39

58m 8&53w "Wm
cam 2.. 885% r.
32 a 385mm ..

85:88on 8: :a E 5528: 8« 85:5: .6585

 

 

 

 

 

oodwm N.m3 80H. 0:3— 85::85
S. 82- 3.5.- 82:95 m3
mmd 5858-:
S E 8855
_ E Zn Zn H E. 25.83850 5 35852
voood- mood Hood- mood mood Hood Sod- Sod o %<
oood C yd .L «Nd v _.o 03d oood .L _ _d mvod ...Kod n#588520 $80”:
9 :d- do do imovd- :Nd 3m d- 5 Fd Emd- owod om :o- @285an EoEEoEoO
wood 5 _ d v Hod me do god- wood Sod ovod o 85:82
ovod- m :d om :d- 8 _.o mmod wood wood- nmod hood :«Em
omod onod _o —d :d vmod wmod mmod mmod oood waCBuaochE
moumtoaofiaau 8.55am gm
No ~ d- Sod ivmmd- ovod wood- wood 1.3 y d- F 5d mmod- busuom
vom d vvmd zoom: mod ovod- m cm d *mond mmod mood- 83u8885~ 5 5:80
53- 53 :83- mod ..Nood- owed 886. :3 Rod- “seam Spiro Ego
mvod- oood 30m Hd- vmod Sod omod wood- n Fod a 5d moumtuw 55850 .5 .5:50
3835 85.5
0 5d- Nmod *mmod- mmod 3.05.? o 5d wmod- :od iomod- 8850
58.585 9 8.505%
omod mmod **oo_d vod *mnod mmod ivmod NSd iomod monogram 05292
wood- mmod N 5d- god mood a 5d wood- m 5d wood- 59:39 85%
Nmod ovod to: _d vod saw—3d Edd Immod NSd ...iwvod 0383582 ouca>w<
8:33: €333.50

mE< Mm 5205.50 mm 506550 mm ucflucooou mm 506585

£35 fies—-2 use; 38 So
523% 8.500 :<

 

acct—«808:0 8:"— 9 558250 a we Getaway—50 "oi-«53’ “Geranium
950 GEE so :5382 coumstou 2: 5 50585590 8: 5 Seam
Wan—nah.

40

5tm Envy—8m "Wm
o\om on ocwucmcwfi ”I"
$2 3 deduciwfi *

5068535 of =a E moo—535cm 05 85535 59550

 

 

 

 

oooom mde amp. 05 89395
no. Goo- one? .523: mod
vmd vmd 333d
Fm ww muUum3_U
moo moo moo moo 823.530 do H3:32
mood mood .ood- mood mood _oo.o *mood wood mood 0?
otd- : _.o i omd Ed oEd Koo :Ed- mood *Ed- 35:30 conga
hood oo _.o :movd- :md Sod- mod :od mood wood 952830 6065800
hood- 3 3 :od mid Sod- ~mood wood oood mood and»:
wood mod omod- oood mood mood m:d Ed ozd oaem
mood- Eod oood do wood omod mood- mood Sod margins“:
motmtoouauwao ogoomm Eamon
mood- Sod good- add wood- oood mood- Sod omod- bosom
mood vvmd room; ood ovod- om dd mod vod mod Baueomeoco o0 330
Sod omod toood- sod *mood- mmod iovod omoo *mmod 283% @5650 530
Sod- oood tomod- Eod Sod m ood :mood- mod good. 823% £026 do 380
OODEOW 0:33am
Sod mood *mmod- mood :oood- flod imood mood :Eod “BEG
53553 9 3.580%
odd- mood toood vod *mood Sod omod- oood mmod- 358:5 012:2
oood mood mood- Sod mood mood tomod mmod mod 58:5 85%
mod. wood :2 3 mod 33d mood tomod- wood omod- owogaoav. 85??
“on—Bum Gown—350

ma? mm 620585 mm 306585 mm 35585 mm 585585

Amdso 505-2 flea 8mm 30

 

vacuum 583m

 

acowauomo gm 9 «535650 a on 559650 nofiamuamr «Go—Eamon

$50 85E :0 50502 559500 05 5 cocoficuwuo 2% 5 ouuoom
.250 95—35%

41

55m .355me ”mm
com a 285mm 3
$2 a 885mm *

50.33qu of z“ E «535203 98 85:55 55:00

 

 

 

 

3mm; m5 255
mm. omoo- mommm- €53: woo
ood omd voodoo-d
«w magma—U
E E to E wcooscomoooosoasz
oood- mood mood- wood wood. oood oood- oood Sod- om<
omod- oood omod- momd Ed- Edd Sod- ood vmod- 95830 558m
mood- momd oomd- momd ommd- mood mood- oood omod- o2w.ocaoocue=§o0
mood- mid omod- o_.o mmod- ovod mood- mood mood- 8282
omod- otd om_.o- wood wood. mood mood- mod mood- ooeo
omod mooo oood omod .o-mod mod omod mmod ood mongoose“:
womuwmh0«OdHGa.—U
0500mm 5.3m
mood- ood mood- mood :wod mood mod- ood mod 58%
momd oood wood Sod omod- ommd ommd omod _mod- eaueomeoczo Ego
oood mood mood oood omod mod mood- :od ood 33% 556% oo 3:30
omod- vmod :oomd- mood *Ed- sod flood- mood omod- 825389353 330
mUUEom Oman—3m
omod- mod :88. ovod to:d- mood tomod- mood :omod- 559588: 90389;
mmod mmod 33d omod {moo mood Isod :od tomod 3855 Dana:
mmod .ood to; ood immod mood immod oood tamod 535 85am
mood ood ammo mood ivomd mood ioood mod :mood owned-SQ 852%
oemwom €353.50

92 mo “SUE-you mo Ease-Vow mm 65580 mm 35580

add 58o? ooeo flom W5
53qu 550 =<

 

mo€oQ ago—H. M3200 50 moo—mum 05 55550 35: 5355> o€ov€omon

#535 Z :35550 05 5 covammcwwuo of _u€a :50 83m cougum mEmcog—um
buy 33$.

42

“ohm Eoncsm "mm
o\om ow Ego-«Emma **
$3 3 38¢in *

3239mm“: 2% :1. E vows—up: 8d 3.2556 53on

 

 

 

Sdf m5 2“?
mo. omo _- modem- Rood-$5 m3
hmd fiasco-m
a :w 3286
ooo so so so 822530 do Benz
ooood oood mood oood oood- _oo.o o Sod oood- Uw<
Edd oood Rod Emd Ed- oood omod mood o_o.o- 95.330 awash“
omod- Rd SE. mood ommd- oood omod- oood oood osmoocsoofieegoo
:od woo mod ood mmod- oood oood oood m_o.o 8282
_mo.o oomd mood oood oood- oood m_od oood omod ooeo
ood mzd omod omod mood mood o_o.o omod :Kod mongoose“:
mowmtofiauwsu 05025 gn—
Sod- oood toomd- oood good mood :oSd- m_o.o Sod £88m
Sod oood :Emd- 53 mod. mid :omod omod oood aaueomeoczobooso
mood- oood tom? oood omod omod :Ed- mood Edd- 282m announce Ego
mmod- oood *oood- mood i:.o- Eod Edd- oood :omod- 805038055? ESQ
mUUEOW Dan—3m
Sod- mood oood- oood :oZd- Edd Sod- oood- .Boooaasuc: ooasosm
oood oood toad oood 15d oood {ood :od n.a.-oood afiéfi 032:2
oood mood omod ood immwo oood mood Sod Eod 925 83%
mood ood immmd omod :oomd Eod toood oood ioood owogaoov. 83>?
2.5me Gauge-50

mm? mm “SUE-you mm 256508 mm Eocoooou mm 56580

Ede fem-Z 58m Som So

 

umoosm Snood-m

 

mason amp—EH. ”£800 new «oatm— oua GOEEOU Bo: "vistamr Eo—Eoflofl
x5382 scams-coo ufi .oo covamedqu 05 van omoU SHE :8qu mEmcoufiom
.58 E ozon-

43

Table 1-8

Impact of Corruption Network Organization on Cost to Firms

Dependent Variable: Size of the Bribe (Percentage of Contract Value)

 

 

 

 

All Other Regions
OLS Interval
Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.048 0.031 0.188 0.127
Service Delivery 0.04 0.025 0.137 0.107
Multiple Payments 0.062 0.041 0.252 0.173
Recourse to Uncorrupt Officer -0.033 0.035 -0.16 0.149
Public Services
Quality of Customs Services -0.118** 0.048 -0.483** 0.204
Quality of Judiciary Systems -0.082 0.054 -0.357 0.234
Quality of Infrastructure 0.049 0.074 0.152 0.316
Security -0.072 0.05 -0.307 0.214
Firm Specific Characteristics
Manufacturing -0.161 0.105 -0.721 0.439
Small 0.077 0.137 0.37 0.572
Medium 0.148 0.1 19 0.704 0.499
Government Ownership -0.292 0.208 -1.602** 0.798
Foreign Ownership -0.189 0.157 -0.714 0.658
Age -0003 0.002 -0.011 0.008
Number of Observations 575 574
Clusters 80 80
R-squared 0.18
Log Likelihood -1435.49
Wald Chi2 1303.86

 

Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

44

Table 1-8 cont.

Impact of Corruption Network Organization on Cost to Firms

Dependent Variable: Size of the Bribe (Percentage of Contract Value)

 

Eastern Europe

 

 

 

 

OLS Interval

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge -0.055 0.036 -0.272"l 0.164
Service Delivery 0.005 0.036 0.061 0.164
Multiple Payments -0.065* 0.039 -0.215 0.177
Recourse to Uncorrupt Officer 0089’” 0.034 0.453** 0.157
Public Services
Quality of Customs Services -0.117** 0.041 -0.509** 0.189
Quality of Judiciary Systems 0086* 0.046 0.438** 0.209
Quality of Infrastructure 0.056 0.066 0.357 0.302
Security 0.079 0.052 0.374 0.237
Firm Specific Characteristics
Manufacturing -0. 108 0.099 -0.651 0.453
Small 0.566M 0.194 2.455** 0.892
Medium 0537’” 0.168 2.132Ml 0.77
Government Ownership 0.04 0.139 0.565 0.64
Foreign Ownership -0.195 0.1 85 -0.674 0.847
Age 0005* 0.003 0.021 * 0.012
Number of Observations 533 533
Clusters 80 80
R-squared 0.25
Log Likelihood —882.771
Wald Chi2 1303.86

 

Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

45

Table 1-9
Impact of Corruption Network Organization on Cost to Firms

Dependent Variable: Size of the Bribe (Percentage of Firm Revenue)

 

 

 

All Other Regions
OLS Interval
Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.086" 0.037 0.263** 0.094
Service Delivery 0.189 0.132 0.132** 0.047
Multiple Payments 0.071 0.044 0.151 0.11
Recourse to Uncorrupt Officer -0.038 0.041 -0.103 0.113
Public Services
Quality of Customs Services —0.139** 0.046 -0.277** 0.130
Quality of judiciary Systems 0006 0.049 -0.080 0.142
Quality of Infrastructure —0.038 0.071 -0.074 0.210
Security -0.124** 0.060 -0.327** 0.160
Firm Specific Characteristics
Manufacturing -0.144 0.099 -0.661** 0.253
Small 0.125 0.161 0.167 0.404
Medium 0.397** 0.124 0.980** 0.348
Government Ownership -0.302 0.284 -0.613 0.520
Foreign Ownership -0.128 0.156 -0.650* 0.395
Age 0 0.003 0 0.007
Number of Observations 625 625
Clusters 82 82
R-squared 0.3
Log Likelihood -1367.46
Wald Chi2 2177.51

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

46

Table 1-9 cont.
Impact of Corruption Network Organization on Cost to Firms
Dependent Variable: Size of the Bribe (Percentage of Firm Revenue)

 

Eastern Europe

 

 

 

OLS Interval

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.086** 0.037 -0.481** 0.110
Service Delivery 0.189 0.132 -0.678** 0.127
Multiple Payments 0.071 0.044 0.345** 0.125
Recourse to Uncorrupt Officer -0.038 0.041 0.489** 0.112
Public Services
Quality of Customs Services -0.139** 0.046 -0.144 0140
Quality of Judiciary Sysrems -0.006 0.049 0.065 0.155
Quality of Infrastructure -0.038 0.071 -0.261 0.214
Security -0.124** 0.06 0.988** 0.158
Firm Specific Characteristics
Manufacturing -0.144 0.099 -0.163 0.322
Small 0.125 0.161 1.494** 0.655
Medium 0.397** 0.124 0.416 0.578
Government Ownership -0.302 0.284 0.619 0.473
Foreign Ownership -0.128 0.156 0.853 0.66]
Age 0 0.003 0.001 0.010
Number of Observations 625 625
Clusters 82 82
R-squared 0.3
Log Likelihood -1367.46
Wald Chi2 2177.51

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

47

Table 1-10

Organization of the Corruption Network

Frequency of Additional Payments: Electric Power

 

 

 

Company

All Other Regions

Coefficient SE
Corruption Regime
Advance Knowledge 0037 0.087
Service Delivery -0.127 0.081
Multiple Payments 0.158** 0.080
Recourse to Uncorrupt Officer 0039 0.071
Public Services
Quality of Customs Services —0.171 0.106
Quality of Judiciary Systems 0.095 0.083
Quality of Infrastructure -0.040 0.136
Security -0.128 0.145
Firm Specific Characteristics
Manufacturing -0.304 0.262
Small 0.040 0.454
Medium 0.547 0.371
Government Ownership -1.870* 1.125
Foreign Ownership 0.453 0.446
Age -0.010 0.008
Number of Observations 686
Clusters 82
Log Likelihood -597.59
Wald ChiZ 294.15

 

Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

48

Table 1-11
Organization of the Corruption Network
Frequency of Additional Payments: Telephone Personnel

 

 

 

 

Other Regions Eastern Europe

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0039 0.074 0.026 0.017
Service Delivery 0.135 0.168 0.023** 0.011
Multiple Payments 0.002 0.171 0.003 0.010
Recourse to Uncorrupt Officer 0.292 0.181 0.002 0.014
Public Services
Quality of Customs Services 0.05 0.148 0.041** 0.015
Quality of Judiciary Systems -0.131 0.124 -0.010 0.017
Quality of Infrastructure -0.13 0.244 -0.067** 0.014
Security 0.124 0.276 -0.040** 0.018
Firm Specific Characteristics
Manufacturing -0.442 0.356 -0.022 0.023
Small 0.92 0.616 -0.002 0.047
Medium 0.587 0.473 0.007 0.039
Government Ownership 0.52 0.822 0.093** 0.034
Foreign Ownership 0.21 0.406 0.054 0.036
Age -0.012 0.01 0.001 0.001
Number of Observations 688 877
Clusters 82 61
Log Likelihood -725.752 -1640.3
Wald Chi2 9118.42 2141.69

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

49

Table 1-12
Organization of the Corruption Network
Frequency of Additional Payments: Licensing Authorities

 

 

 

No Transition Europe Eastern Europe

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.200** 0.058 0.007 0.020
Service Delivery 0.005 0.085 0.029 0.019
Multiple Payments 0.117** 0.059 0.006 0.009
Recourse to Uncorrupt Officer -0.077 0.073 -0.089** 0.025
Public Services
Quality of Customs Services -0.212* 0.118 -0.020** 0.010
Quality of Judiciary Systems 0.210** 0.090 —0.024 0.021
Quality of Infrastructure —0.285* 0.169 -0.048** 0.018
Security -0.075 0.098 0.065** 0.028
Firm Specific Characteristics
Manufacturing 0.116 0.264 -0.065* 0.038
Small -0.567 0.414 0.086 0.089
Medium 0.095 0.316 0.200** 0.097
Government Ownership -0.404 0.575 0.136** 0.047
Foreign Ownership 0.450 0.312 0.023 0.049
Age -0.017** 0.007 -0.001 0.001
Number of Observations 669 877
CluSters 81 61
Log Likelihood -1068.76 -1632.4
Wald Chi2 4563.36 912.82

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

50

Table 1435
Organization of the Corruption Network
Frequency of Additional Payments: Tax Insp. Agents

 

 

 

Other Regions Eastern Europe
Coefficient SE Coefficient SE

Corruption Regime
Advance Knowledge 0054 0.089 -0.003 0.016
Service Delivery 0.201 ** 0.074 0034* 0.020
Multiple Payments 0.288** 0.098 0.000 0.009
Recourse to Uncorrupt Officer -0.248** 0.067 —0.080** 0.022
Public Services
Quality of Customs Services 0.057 0.075 -0.001 0.020
Quality ofJudiciary Systems 0.146 0.125 -0.033 0.020
Quality of Infrastructure -0.275** 0.139 -0.052** 0.018
Security -0.084 0.141 0.027 0.019
Firm Specific Characteristics
Manufacturing 0297* 0.176 -0.015 0.038
Small -0.353 0.253 -0.046 0.055
Medium -0.261 0.185 0.060 0.040
Government Ownership -0.661 1.029 -0.012 0.039
Foreign Ownership 0.241 0.171 -0.010 0.056
Age -0.013** 0.006 -0.001 0.001
Number of Observations 674 877
Clusters 82 61
Log Likelihood -1238.65 -1622.25
Wald Chi2 7833.52 945.46

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

51

Frequency of Additional Payments Required: Customs OEicials

Table 1-14
Organization of the Corruption Network

 

No Transition Europe

Eastern Europe

 

 

 

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.186 0.114 0.007 0.014
Service Delivery 0.228** 0.097 0.002 0.013
Multiple Payments 0.105 0.076 0.030** 0.012
Recourse to Uncorrupt Officer -0.173* 0.102 -0.048** 0.022
Public Services
Quality of Customs Services -0.214** 0.106 -0.013 0.014
Quality of Judiciary Systems 0.012 0.103 -0.044 0.028
Quality of Infrastructure -0.117 0.123 -0.025 0.026
Security 0.099 0.131 0.011 0.018
Firm Specific Characteristics
Manufacturing 0.15 0.254 —0.024 0.041
Small -0.785** 0.308 0.116* 0.066
Medium -0.711** 0.322 0.137** 0.054
Government Ownership -0.807 0.737 -0.023 0.045
Foreign Ownership 0.587** 0.192 -0.124* 0.070
Age -0.022** 0.009 0.000 0.001
Number of Observations 651 877
Clusters 82 61
Log Likelihood -2471.142 -1696.07
Wald Chi2 9805.3 705.01

 

Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

52

Table 1-15
Organization of the Corruption Network
Frequency of Additional Payments: Government Procurement

 

No Transition Europe Eastern Europe

 

 

 

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0249* 0.129 0.027** 0.012
Service Delivery 0.137 0.129 0.029** 0.009
Multiple Payments 0.093 0.129 0.016 0.011
Recourse to Uncorrupt Officer 0.071 0.190 -0.066** 0.022
Public Services
Quality of Customs Services -0.170 0.142 -0.020 0.013
Quality of Judiciary Systems —0.071 0.157 -0.001 0.029
Quality of Infrastructure 0.028 0.386 -0.011 0.026
Security -0.459** 0.149 -0.014 0.025
Firm Specific Characteristics
Manufacturing -0.297 0.314 -0.085 0.065
Small -1.246** 0.564 0.077 0.105
Medium 0315 0.376 0.068 0.082
Government Ownership -O.657 0.830 0.099 0.062
Foreign Ownership -0.164 0.447 0.004 0.064
Age -0.014 0.013 -0.006 0.004
Number of Observations 669 877
Clusters 81 61
Log Likelihood -900.63 -1769.53
Wald Chi2 3927.96 4377.8

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

53

Table 1-16
Organization of the Corruption Network
Frequency of Additional Payments Required: Judicial and Courts

 

 

 

 

No Transition Europe Eastern Europe

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.030 0.122 0.005 0.015
Service Delivery -0.093 0.066 0.010 0.013
Multiple Payments 0.185** 0.091 0.027** 0.008
Recourse to Uncorrupt Officer 0.050 0.115 -0.041** 0.009
Public Services
Quality of Customs Services 0.100 0.091 -0.060** 0.029
Quality of Judiciary Systems -0.281** 0.138 -0.026** 0.008
Quality of Infrastructure 0.416* 0.222 0.020 0.037
Security -0.338** 0.105 0.008 0.027
Firm Specific Characteristics
Manufacturing -0.045 0.292 -0.045 0.055
Small 0221 0.443 0.141* 0.079
Medium -0.01 0.395 0.140* 0.079
Government Ownership -0.896 0.747 0.051 0.039
Foreign Ownership 0.237 0.254 0.042 0.048
Age -0.012 0.014 0.000 0.001
Number of Observations 669 877
Clusters 81 61
Log Likelihood -549.23 -1676.79
Wald Chi2 298.81 427.89

 

Country dummies are included in all the regressions
* significant at 10%

** significant at 5%

SE: Standard Error

54

Table1-17
Organization of the Corruption Network

Frequency of Additional Payments Required: Politicians

 

 

 

 

All Other Regions Eastern Europe

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge 0.139 0.103 0.009 0.006
Service Delivery 0.077 0.135 -0.001 0.010
Multiple Payments -0.012 0.123 0.007 0.008
Recourse to Uncorrupt Officer -0.154 0.131 -0.013** 0.005
Public Services
Quality of Customs Services -0.228 0.153 0.002 0.006
Quality of Judiciary Systems —0.055 0.1 11 -0.007 0.009
Quality of Infrastructure 0.153 0.252 -0.006 0.016
Security 0.100 0.129 0.011 0.012
Firm Specific Characteristics
Manufacturing -0.265 0.389 0.003 0.018
Small 0.074 0.415 0.155** 0.049
Medium -0. 128 0.372 0.130** 0.052
Government Ownership —0.681 0.721 0.028 0.032
Foreign Ownership 0602* 0.344 0.012 0.043
Age —0.01 0.01 -0.001* 0.001
Number of Observations 673 877
Clusters 82 61
Log Likelihood -963.971 -1666.94
Wald Chi2 642.84 4820.13

 

Note. Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

55

 

 

 

Q. on No 3385
o: ooo N: 822.5320 52832
oooo vooo- vooo _oo.o mooo o woo.
momo _oo.o- Eo Nooo oflo .132 @3830 comes;
oooo woo- ooNo ..ooo- moo oovo- @3830 US$5.60
mm _.o oooo- oflo isomo- :so omoo 8232
oto oooo- 5o toSo- oflo oooo- :85
2 _.o zoo oooo ooo- oo _.o mooo- woo—Bosses:
momomteoaflfiu $.30on 5.5m
too too- omoo torso- Eoo too—o- 3&5
vooo an oomo toovm Smo toe; assesses .8 530
Eoo Eoo Nooo woo. woo ...oooo- 38%
too woo woo too- Goo oooo- 52% baggie Ego
Boo its. _.o- :oo ...vooo- Eoo oooo 882.com seesaw co E30
wOOEDW Om—haflnm
Vmoo :2 _.o- oooo imfo Nvoo oooo- 8&0 £358: 9 888;
mooo ...LQo woo oooo- Roo :Soo 353$ “12::
omoo iofio goo oooo- ovoo _oo.o 5.5.5 “atom
omoo iommo oooo mooo Nooo ivio 8338M 8:23
25mg— cougar—.50

mm 38808 mm £20808 mm Baotou

>833 958

 

mcommom EEO =<

 

a?“ Dish.

ouBoEuoodcm E moutmucfléoz we “ooh.

56

Table 1-18 continued

Test for Non-Linearities in Infrastructure

 

Eastern Europe

 

 

 

corrup brbpay

Coefficient SE Coefficient SE
Corruption Regime
Advance Knowledge -0.159 0.167 0.026 0.054
Service Delivery 0.09 0.128 0.013 0.052
Multiple Payments -0.061 0.12 0.023 0.041
Recourse to Uncorrupt Officer 0.197 0.48 -0.042 0.048
Public Services
Quality of Customs Services -0.172 0.307 —0.012 0.063
Quality of Judiciary Systems 0.132 0.666 -0.090** 0.039
Security -0.114 0.476 -0.104** 0.048
Quality of Infrastructure 1.707 35.699 8.130** 0.904
Infrasq -0.211 5.09 -1.151* 0.148
Firm Specific Characteristics
Manufacturing 0.011 0.56 0.174** 0.082
Small 0.274 0.795 -0.044 0.176
Medium 0.023 1.247 0257* 0.145
Government Ownership 0.022 0.394 -0.084 0.122
Foreign Ownership -0.508 1.022 0.021 0.151
Age 0.006 0.011 0.002 0.003
Number of Observations 832 832
Clusters 61 61

 

Country dummies are included in all the regressions

* significant at 10%
** significant at 5%
SE: Standard Error

57

CHAPTER 2. BRIBERY AND FIRM BORROWIN G COSTS: IS THERE A

RELATIONSHIP?

1 Introduction

What are the true costs that administrative bribery imposes on firms? How are these
transmitted throughout the economy? This paper investigates possible channels through
which, the effects of bribery are transmitted throughout the economy and magnified beyond
the value of the bribe. In particular, we explore whether bribery imposes indirect costs on
firms through its impact on firms’ ability to raise external finance.

In the last few decades, corruption has become a highly prominent topic of debate
among policy makers, academic scholars and other practitioners in developmental issues and
the universal consensus is that it is a public bad that should be eradicated. In almost all
countries, especially the developing and transition economies, there are ongoing policy
initiatives aimed at addressing the problem. The outcome of this intense interest in the topic
has been an extensive research on causes and consequences of corruption. This literature
provides valuable insights into its impact on economic activity. For instance, the studies
have found that corruption reduces growth and investment throughout the economy
(Mauro1995), reduces foreign direct investment (Wei 1997), drives firms out of the official
economy (Kaufmann 1997), reduces both public sector budgets and the productivity of a
county’s infrastructure (T anzi and Davoodi 1997) and lowers stock values (Lee and Ng
2002)

With this rich literature, do we know enough about the impact of corruption?
Specifically, does the current research provide any knowledge about the outcomes of

corruption at the micro-level? For instance, do we know enough about the costs that

58

corruption imposes on firms, how these effects are transmitted or the distributional effects
on different agents? The answer here is emphatically negative. The bulk of existing empirical
work is based on cross country data, which while it provides valuable insights into
understanding economy-wide determinants of corruption, can not inform us of micro level
outcomes. Thus, the research agenda is far from complete. Micro-level empirical analysis is
critical to complement the existing country level studies for any policy reform agenda to be
effective. Yet it was only very recently that research effort is being devoted to micro-level
empirical studies. This research has yielded some useful and very informative findings in

areas such as firm-level costs of bribery,1314

the importance of the organization of the
corruption regime15 and identifying the agents most vulnerable to corruption”. However,
the very newness of this strand of literature suggests that there is still a lot to uncover on the
relationships between corruption and economic agents.

The purpose of this paper is to add to this new but growing research, specifically
examining if other channels exist through which bribery could be imposing costs on firms
and thereby constraining their performance. Specifically, we investigate the possibility that
bribery could negatively impact firms’ ability to raise external finance.

Access to credit has long been recognized as key to firm performance and growth.
Furthermore, firms with access to external financing generally grow faster than firms that
rely on internal funds. At the same time access to finance is a major obstacle to firms’

operations, especially small and medium firms, who employ the larger proportion of

populations in developing and transition economies. For instance, Beck, Demirguc—Kunt

 

1’ Svensson (2003) found bribe payments constituted about 8% of firm costs in Nairobi while in Kuncoro
(2006) reports payments as high as 10.8% of operating costs in Indonesia.

1“ Indirect costs, i.e. all other costs related to corruption such as the increased amount of time that firms
spend dealing with government officials (Henderson and Kuncoro, 2004 find that Indonesian firms spent
over 10% of management time with bureaucrats).

1’ See Herrera, Lijane and Rodriguez, 2007 and Wu 2005.

’6 See Mocan 2005 and Svensson 2002.

59

and Levine (2005) find that financial and legal constraints and, corruption affect firm growth
but also find that small firms are the most affected by these obstacles. Beck, Demirguc-
Kunt, Laeven and Levine (2007) study the impact of financial development on firm growth
and find that under developed financial systems are more detrimental to the growth of small
firms.17 Bank concentration is another factor that increases financing obstacles and decreases
the probability of firms receiving bank finance (Beck, Demirguc-Kunt and Maksimovic
2003).18

To the best of our knowledge, this study is the first to explore the link between
administrative corruption and explicit borrowing outcomes. However, there are a number of
articles that explore different aspects of the relationship between corruption and credit.
Specifically, Depken, LaFountain and Butters (2006) test the impact of corruption on a
country’s creditworthiness. They find that creditworthiness measured by sovereign credit
ratings is a decreasing function of corruption. Beck, Demirguc-Kunt and Levine (2005) look
specifically at the effects of bank corruption on firms’ ability to raise external finance, and
further assess the effectiveness of different supervision approaches in lowering this obstacle.
Finally, Ciocchini, Durbin and Ng (2002) study the relationship between emerging market
bond spreads and corruption. They find that governments and firms in countries where
corruption is widespread have higher default risks and consequently higher spreads.

By identifying one way firms’ operations and their growth opportunities are affected
by corruption, the study contributes to an improved understanding of the workings of
bribery. We find evidence that pervasive bribery is associated with more stringent borrowing

conditions for firms. In particular, in environments characterized by high levels of bribery,

 

‘7 Love (2003) confirms this result through a study that shows financial development improves availability
of external and consequently increases firms’ investment.

‘8 See also Cetorelli and Strahan (2004) who find that increased bank competition increased access to
finance.

60

bank lending activity is low, interest rates on which firms borrow tend to be higher and loan
maturities are generally shorter.

Our results also have policy implications that emanate from the complementarities of
governance and financial market development. In particular, it suggests that efforts to ease
credit constraints for firms without genuine and concerted efforts to reduce corruption
could potentially have low probability of success.

The rest of the paper proceeds in this manner. Section 2 is devoted to data,
descriptive analysis and theoretical links. We present the empirical strategy in Section 3,

results in Section 4 and conclude with Section 5.

2 DATA, SUMMARY STATISTICS AND THEORETICAL LINKS

The primary dataset we use is from the Business Environment and Entrepreneur
Performance Surveys, hence forth BEEPS that the World Bank conducts in transition
Europe and Turkey. The three wave surveys (1999, 2002, and 2005) cover over 10, 000 firms
in 26 countries. In this study we use only the last two surveys, primarily because some of the
variables of interest are not included in the 1999 survey.

BEEPS are meant to identify factors that constrain firm performance and growth in
the target countries. Thus, the surveys contain a large number of questions on the obstacles
to growth and their severity. Questions cover a wide array of issues that pertain to firms’
ability to function including different aspects of corruption, financing constraints and public
services such as physical infrastructure and institutional frameworks. In addition, the surveys
also include questions on firm-specific characteristics such as ownership, origin, size, primary
activity and age.

These types of surveys are relatively new but have already provided a platform for

carrying out the much needed firm level empirical research. While other surveys e.g. \WEBS

61

2000 have provided a basis for informative analysis, they have tended towards perceptions
which still attract controversy on reliability of inference. BEEPS have a very appealing
feature in that the questions are more and more moving from perception-kind responses to
more objective data. For example, in WEBS 2000, respondents were asked how they rated
the quality of different infrastructure services with possible responses ranging from 1 for
very poorly, to 6. In BEEPS, the respondents are asked for the number of times they
experienced service interruption and the duration of the interruptions. With respect to
administrative corruption, the perception-type questions are Still used. For instance, a
question is how common it is for firms in a particular business to have to pay bribes to get
things done with the responses having six possible outcomes. However, some new variables,
that nicely proxy for the level of corruption, are now available. For example, in addition to
asking firms to state the number of times they made unofficial payments for different public
services, firms are asked for the number of times officials from these sectors inspected the

firm over the same time frame.

2.1 Bank Lending Conditions

The primary objective of this analysis is to test whether a link exists between
corruption and firm borrowing conditions. The kind of effect we envisage emanates mainly
from the impact of corruption on firms’ cash flows and consequently on their default risk.

In addition to the costs of paying bribes, studies have shown that corruption can
reduce legal protection of creditors as well as regulatory oversight (e.g., Garmaise and Liu,
2005). Moreover, delivery of essential services, such as utilities, becomes uncertain in the
presence of corruption as government officials may withhold services even for those who
pay the bribes, particularly if the corruption regime is unorganized. For instance, Tanzi

(1988) presents evidence suggesting that corruption tends to lower efficacy of service deliver.

62

Deficient service provision interferes with firms’ functionality and consequently reduces
their revenues. On the other hand, firms that pay bribes might get services that they would
otherwise not have access to, which could increase their profits. In general however,
payment of bribes increases firms’ operation costs, consequently lowering the profitability of
projects and raising the default risk, which in turn leads to tighter borrowing conditions. The
expected return to banks depends on the probability of loans’ repayment and as such
projects’ expected returns, which directly affect ability to repay are a key criteria used to
determine lending conditions.19 Depending on the assigned default risk, the bank will
determine the conditions on which firms borrow and these are generally more stringent the
higher the default risk is. There are several means by which banks can impose conditions on
borrowing. In this paper we have four different variables that capture the stringency of

borrowing conditions faced by firms. W’e discuss each of these variables below.

2.1.1 Interest Rates

In making loans, banks are concerned about the interest they receive and the
riskiness of the loans as these two are key components of their expected returns. The main
determinants of this riskiness are the uncertainty of production, projects’ profitability as well
as the severity of information asymmetries. The riskiness of projects, in turn, determines the
monitoring effort banks expend. Monitoring, on the other hand, results in an external
finance premium, which Gilchrist and Zakrajsek (1995) show to be highest where the risk of
opportunistic behavior is difficult to mitigate. Environments with pervasive corruption
therefore, lead to increased risk of making investments and consequently to a higher risk

premium that is reflected in high interest rates. Thus, to the extend that pervasive bribery

 

1’ For a full discussion on project’s expected return and lending, see F reixas and Rechet, 1997.

63

increases default risk by lowering cash flows, we should expect to firms to face higher
interest rates.

Two questions in the surveys provide information regarding the prevailing interest
rate faced by firms and the possible effect interest rates have on firms’ access to credit. First,
firms that hold bank loans are asked for the actual interest rate they are charged. Second,
firms who have not sought bank lending are asked to indicate whether interest rates were a
factor in their not applying for loans. Summary statistics reported in Tables 2-1a and 2-1b
show an average interest rate faced by borrowing firms of 17%, with the country-averages
varying widely from as low as 7.6% for Slovenia to as high as 34.6% for Belarus. With
respect to firms that did not seek bank funding, 22% of firms indicated that high interest

rates were a determining factor.

2.1.2 Quantity of Loans

In their seminal paper, Stiglitz and W’eiss (1981) show that banks can only increase
interest rates up to a certain point beyond which safe borrowers will be driven out of the
market leaving only a pool of high risk projects with a low probability of success. There is an
interest rate level at which expected returns on loans become negative. In this case, banks
may choose to ration credit, instead of increasing interest rates. This situation is very
common in developing economies where banks voluntarily hold liquidity in excess of that
required statutorily, despite the fact that (by western standards) statutory requirements are
already high, imposing an implicit tax on banks. Thus, by contributing to increases in interest
rates, bribery then can cause credit rationing.

Firms in the survey were asked for the percentage of their working capital and/ or
investment that was financed with funds from commercial banks. For firms that do not have

a loan, the surveys have asked additional questions on whether the firm had applied and was

64

turned down or did not seek funding from banks. For instance, firms were asked whether
they do not have loans because they were rejected or had not applied. A significant number
of firms that did not apply indicated that while they might have needed bank loans, they did
not apply because they felt that the conditions were too stringent, e.g. collateral requirements
(16% of firms) and interest rates (22% of firms) were too high, while others were intimidated
by the complexity of the application procedures (16% of respondents). A small percentage
did not apply because they thought their applications would be rejected.

The descriptive statistics reported in Table 2-1a show that about 20% of firms had
loans from commercial banks. However, across countries the means vary widely, from as

low as 5% for Tajikistan to a high of 35% for Latvia (Table 2-1b).

2.1.3 Collateral Requirements

Covenants meant to reduce banks exposure to risk in case the borrower defaults are
an option that can offset the above outcomes. Collateral can help mitigate the problems of
information asymmetries. In particular, collateral acts as a signal that induces borrowers to
reveal their default risk (Bester, 1985; Besanko and Thakor, 1987) and it provides firms with
an incentive to exert effort in the project. However, for collateral to be meaningful, banks
must be assured that the legal system is able and/ or willing to enforce contracts. By
weakening the rule of law and where it permeates the legal systems, corruption can
undermine the use of collateral. Strong protection of creditor rights is also essential. De Haas
and Peeters, 2006 show however, that the process of seizing and disposing of collateral once
the borrower has defaulted tends to be very slow in transition and developing countries. As a
result, collateral requirements can be prohibitive as banks seek to compensate for the time

lapse.

65

The average collateral requirement for all countries is 149.65% of the loan amount,
and for individual countries, the requirements range from 119% for Turkey to 217% for

Georgia (see Tables 2-1a and 2-1b).

2.1.4 Maturity

Finally, with a high level of uncertainty where future policy direcrion is not clear,
banks may confine their decisions to the immediate future by maintaining short planning
horizons. The overall mean loan term for all countries is about 27 months (see Table 2-1a)
but country-wise comparisons in Table 2-1b show a wide disparity in average loan terms
with the lowest repayment period in Turkey at about 15 months while Croatia has the

highest at 46 months.

2.2 Corruption Measures

The survey has a number of questions that ask firms to indicate the extent to which
corruption is an obstacle to their operation. The severity of corruption is captured by firms’
responses to four questions: a) is it common for firms in the line of business to have to
make irregular payments to get things done? (brbpay); b) Do firms in the line of business
know in advance the magnitude of the additional payment? (advance); c) How many times do
firms have to make the unofficial payments to obtain service in different sectors? (brbfreq)
and d) How problematic corruption is for firm operation and growth (corrup). Answers for
the first three questions vary between 1 (never) and 6 (always) while for corrup four different
outcomes are possible from 1 (no obstacle) to 4 (for major obstacle). In addition firms were

asked for the number of times officials from different service sectors inspected their

66

operations (mmjmgbr). Whereas inspections could represent a necessity for the business

development, beyond some threshold, they are likely to indicate red tape faced by firms.20

2.3 Legal, Judicial and Infrastructure

While our chief interest is in testing whether corruption influences lending
outcomes, it is also true that these also depend on the quality of institutional and physical
infrastructure. For example, banks may be unwilling to extend loans if the legal framework is
such that contract enforcement is problematic while on the part of the physical
infrastructure, particularly telecommunications have been shown to mitigate problems of
transacting with banks (Kamel, 2005; Honohan et a1, 2006). The survey includes a number of
variables that proxy for the quality of these services. Firms were asked to evaluate the quality
of the courts in terms of their being a) fair, b) honest, c) affordable, c1) efficient and e)
effective. In addition, firms were asked to indicate their confidence in the legal system. We
have used these six individual measures to construct an index of the quality of the legal
system.

With regard to physical services, we use firms’ use of different information and

communications technology services.

2.4 Firm-specific Characteristics and Country-level Variables
Firm—specific traits also have a bearing in the conditions on which firms borrow. For
example, literature shows that small firms generally attract punitive conditions because of

among others, their inability to prove their credit worthiness.21 We have therefore, included

 

’0 Various studies argue that excessive red tape is a deliberate outcome of corrupt officials’ efforts to create
environments that facilitate opportunistic behavior (for example, see Sullivan and Shkolnikov, 2004;
Sherif, 2005). Hence, inspections can be a valid proxy for existence of corruption.

2'For example, see Beck, Demirguc-Kunt and Levine (2004), Beck, Demirguc-Kunt, Laeven and Levine
(2007) and Gilchrist and Zakraj sek (1995).

67

dummies for firm size and its principal activity. Summary statistics show that 69 percent of
firms are small, while 39 percent are manufacturing firms. \We also included dummies that
cover ownership. Specifically we control for whether a firm is government owned in which
case the variable govern takes the value of one and zero otherwise and, for foreign ownership
the variable foreign takes on the value of one if the firm is foreign owned. Of the surveyed
firms, 12% are government owned while foreign entities control 34% of the sample.

Finally, since borrowing conditions, which are our main outcomes, also depend on
the general state of the economy, we include some country-level variables, specifically, GDP
per capita growth and inflation.

In Table 2-2 we present a set of selected pair-wise correlations and a few of these are
of particular interest. Not surprisingly, the perception that corruption is an obstacle to the
operation of the firm is negatively correlated with the probability that firms obtain finance
from commercial banks. As should be the case, the quality of the legal framework is
positively correlated with firms’ probability of getting finance while service interruption is
negatively correlated but the rate at which firms use information and communication
technology is positively correlated. Bribe frequency is negatively correlated to bank lending
but more interesting is the positive and relatively large correlation between frequency of
bribes and the number of inspections firms are subjected to. Finally, all measures of
corruption are positively correlated with the interest rate banks charge individual borrowers
and negatively related to the maturity of the loan but also with collateral requirements.

Overall, the correlations provide a reasonable basis for testing the stated hypotheses.

3 Estimation Strategy
Our primary objective is to test whether administrative corruption has a role in

determinin firms’ borrowin costs. We have four variables that ca ture borrowin costs,
8 g 8

68

namely the quantity of loans, the interest rates that individual firms pay, the collateral

required by banks and the maturity of loans. We start with a general model of the form:
Yit = Aity + Wita + 2:15 + Ci + vit (1)

where A” is the matrix of variables that measure the level of administrative corruption,

Wit contains variables that capture the efficiency and effectiveness of the operational

environment, Z1. t is a matrix of firm specific characteristics that could potentially have a

bearing on firms’ borrowing conditions, Ci are country specific fixed effects, and vit is a

random error term.

One aspect of our choice of variables needs to be explained further. In the variables
measuring the intensity of administrative corruption, we have chosen to use the number of
inspections a firm is subjected to by the various bodies as the main explanatory variable. We
believe that this variable is likely to reflect the reality better than variables that entail firms
admitting to participating in what are essentially illicit activities. Furthermore, the
relationship between firms’ borrowing cost and the measure of corruption is likely to be
nonlinear. The reasoning for this non-linearity is as follows: inspections are legitimate and
essential elements of well functioning systems that ensure adherence rules, regulations and
laws and, ensures a level playing field for firm operations. Thus at low levels, they portray a
desirable environment in which all parties are comfortable operating. However, the higher
the number of inspections, the more likely they are to become harassment through which
officials can extort payments and some firms will succumb to the pressure.22 Thus beyond

some threshold, inspections become a symptom of administrative corruption, generating

 

2’ This is in line with the findings that bureaucrats designed complex regulations with the objective of
creating an environment that provided opportunities for rent seeking (Sullivan 2002; Sullivan and
Shkolnikov, 2005).

69

uncertainty and consequently to credit rationing that show up in low quantities of loans or a
lower probability of obtaining financing from banks.

Regarding the estimation strategy for quantity of loans, recall that firms were asked
for the proportion of the operations financed by bank loans. The variable has a pile up at
zero (72% of responding firms did not receive loans) and is roughly continuous to the right.
Thus we have used the model using tobit to estimate the probability that a firm will get a
loan with a model is given by:

Ait +Wita+Zit6+ci
Fruit > 0' Ait’Wz‘t’Zitci’) =1—(D(—

 

) (2)

O'
whereCD is the cumulative standard normal distribution.

To provide additional insight into the probability that a firm will have access to
credit we also estimated the probability that a firm is able to obtain financing from banks.
We have therefore, transformed the variable into a binary that takes the value of one in any
period a firm obtains a loan and zero otherwise. The probability that a firm will have a bank
loan is thus given by the following logit model:

Ply” =1): Aity+Wita+Zit5+ci+vit (3)
To account for country fixed effects, we estimate the model with country dummies.

Interest, loan term and collateral are all continuous thus are estimated using standard

panel data methods. A-priori, it is likely that the corruption measure xit is highly correlated

with unobserved country-level characteristics (Ci ), thus rendering the fixed effects model

70

more appropriate. Yet, to test whether the fixed effects or random effects model are
appropriate we use the Hausman’s test.23 Thus the model is given by:
y it = Air” + Wita + Zita + Ci + vit (4)

We have also estimated the model including an interaction term between the country
dummies and a time period indicator, which allows us to test the importance of time varying
country level effects. Unreported results (available upon request) are essentially identical to
those obtained for the standard fixed effects model without the interaction term. The only

apparent difference is the lack of statistical significance of foreign ownership does not

appear to have a significant effect.

4 Empirical Results

(i) What Factors Determine Firms Access to Credit?

Tables 2-4 and 2-5 present the estimation results for the probability that a firm will
obtain a loan from the banks. In the first regression, we use number of the inspections faced
by each firm as an indicator of the presence of corrupt tendencies. In the second, we use the
most direct measure but potentially subject to selective misreporting, frequency of bribes in
different sectors. In both we also include other measures of corruption. Although we have
run different regressions with and without country dummies we focus our discussion on the
former as there is clear evidence of country fixed effects (see Table 2-4).

The estimation results suggest that indeed corruption has a role in the determining
firms’ access to credit. In particular, the number of inspections and its quadratic term are
highly significant and support the existence of the non—linear relationship. In addition, the

coefficient on the perception of corruption being a problem to firm’s functionality is also

 

2" Note that if the fixed effects procedure is appropriate, we could also use first differencing to get rid of the
Ci since in a two-period panel the two are equivalent.

71

significant at the 5% level. The impact turns negative for more than 2.34 inspections while
about 19% of responding firms undergo more than the 2.3 inspections. This is a significant
population of the firms and suggests that indeed beyond the threshold, the impact is actually
negative. The average partial effects range form 0.02 for the perception of corruption to 0.06
for the number of inspections. As for the other measures of bribery, the coefficients are not
statistically significant. However, the joint significance test shows that as a group, corruption
measures do have an effect on firms’ ability to borrow from banks. In terms of the
economic significance, it appears most of the impact is explained by country effects.

The specification that uses frequency of bribes (larlzfmq) supports the view that
corruption is a factor in firms’ ability to raise external finance. In this case, the main variable,
bribe frequency, is not significant but both the perception of corruption and the size of the
bribe reduce the firm’s likelihood of receiving a loan. The coefficients on these variables are
significant at the 5% level. Moreover, we have a positive impact on the possibility of getting
service from an uncorrupt official, further reinforcing the importance of corruption in
determining firms’ access to credit. The APES are very small with the highest at 0.011 for the
perception of corruption. This could imply that even though the coefficients are statistically
significant, they may not be as important in economic terms as other country-level
characteristics.

In both specifications, the proxy for the quality of physical infrastructure is
significant at the 1% level and firm-specific characteristics matter for banks’ propensity to
lend. Specifically, small firms are associated with lower probabilities of getting a loan from
commercial banks. This result is consistent with the literature on borrowing constraints (see
for example Beck, Demirguc-kunt and Maksimiv 2002 and Gertler and Gilchrist 1994).

Government and foreign firms also have a lower likelihood of obtaining bank loans. W’ith

72

both, this could be a demand effect in that both are likely to have alternative sources of
funding that are a good subsritute for bank lending. In particular, foreign firms may be able
to borrow from their own countries and thus require less domestic loans. On the supply
side, it could be that foreign firms are associated with higher levels of risk as they have less
difficulty absconding with loans pending by simply leaving the country. Noteworthy is the
fact that unlike the APES for the corruption measures, the APES for the firm-specific
characteristics are considerably larger with government ownership at 0.127 followed by being
a small firm at 0.107. The other APES ranging from 0.042 for being a foreign firm to 0.164
for being a manufacturing firm are still significantly larger than the APES for corruption
variables.

Other empirical work on firms’ access to finance include as explanatory variables,
bank concentration and sales, as a proxy for firm size (Beck et a1, 2002, 2006). Unfortunately,
the BEEPS do not include information on sales; yet, we control for firm size using the
number of employees. As for banks concentration, its effect should be captured by the

country fixed effect.

(ii) What Factors are Important for the Determination of the Terms

Attached to Loans?

Having found some evidence suggesting corruption has a negative impact on firms’
access to credit, we now proceed to investigate its impact on the terms of the loans (i.e.
interest rate, collateral requirements, and maturity).

Recall first that because the question on the number of inspections only appears in
the last survey, we are not ale to use it for the estimations that requires demeaning or
differencing. Thus, in what follows, we only use the specification with the frequency of

bribes together with the other measures of corruption.

73

Table 2-7 presents the regression results for the impact of corruption on the interest
rate banks charge firms. These results suggest again that corruption in the system has an
impact on firms’ borrowing conditions. In particular, interest rates are increasing in the
commonality of paying bribes to obtain service and the size of the bribe. These coefficients
are significant at the 5% level and the coefficients are relatively large at .70 and .33
respectively. Interestingly, the advance knowledge of the bribe and its magnitude reduces the
interest rate. With respect to public services, the firms’ use of ICT services strongly lowers
the interest rate and so does the quality of the legal system. Firm specific characteristics
appear to be quite important in the interest rate that a firm faces. With the exception of size,
all variables are significant at the 5% level. Firms in the manufacturing sector pay higher
interest rates. Foreign firms also pay higher interest rates, which may suggest banks impose a
higher rate due to the relative easiness pull up stakes and leave the host country. Finally, the
interest rate is decreasing in the age of the firm.

Turning to maturity of loans, there is some indication of the influence of corruption
(see Table 2-9). Both the perception of corruption and advance knowledge of the bribe are
significant, but at the 10% significance level. Economically the impact is significant at 1.23
and 1.42, respectively. Yet again the quality of public services increases the maturity of the
loan, perhaps reflecting the positive impact the effectiveness of these services have on ability
and willingness of agents to make long term plans. Firms’ specific characteristics do not have
such a large impact as they have on the interest rate faced by the firms. Only being in the
manufacturing sector and being foreign are significant and both have a negative impact on
the loan term.

Finally, with respect to determinants of collateral requirements, it appears that we are

perhaps facing an issue of banks preferring credit rationing to asking for high levels of

74

collateral. Only the perception that uncertainty is a problem to firms’ operations is
significant. The results are presented in Table 2—8. None of the corruption measures affect
the level of collateral.

However, consider a scenario in which corruption has permeated the court system.
Hence, a firm with sufficient power can force the enforcing official to drag their feet about
seizure of the pledged property. Knowing this, collateral is not useful to lenders and they opt

not to make the investments.

5 Conclusion

The primary objective of this study was to test whether corruption in an economy,
specifically bribery to government officials in different sectors, affects firms’ access to and
the cost of external finance through cash flow effects. The key assumption for this
proposition is the fact that lending conditions depend largely on assessed default risk, which
in turn is a function of profitability of financed projects.

The empirical analysis has demonstrated that indeed, corruption does influence
lending outcomes. The results are particularly strong with firms’ ability to access credit in the
presence of bribery. This relationship between lending and bribery suggests that corruption
has potentially wider implications and adds another level towards understanding the costs of
corruption to firm performance and growth. Clearly, there are policy implications but more
research is needed to be able to inform the design of policy to deal with this problem. The
increasing quantity and quality of firm-level data will go a long way towards making this

possible.

75

REFERENCES

Beck, T., Demirguc-Kunt, A. and R. Levine (2005), “Law and Firmr’ Accer: to Finance”,
American Law and Economic: Review 7(1), pp. 211-52

Beck, Demirguc-Kunt and Maksimovic (2003), “Bank Competition, Financing Obstacles and
Access to Credit”, Unpublished Paper

Beck, Demirguc-Kunt, Laeven and Levine (2007), “The Determinants of Financing
Obstacles”, journal of International 114 one}! 2’)” Finance, forthcoming

Beck, T., Demirgiic—Kunt, A. and R. Levine (2005), “Bank Supervision and Corruption in
Lending”, Unpublished Manuscript

Beck, T., Demirgiig-Kunt, A. and V Maksimovic (2004) “Financing Patterns around the
World: Are Small Firms Different?” Unpublished Manuscript

Bester, H. (1985), “Screening v s. Rationing in Credit Markets with Imperfect Information”,
T/Je American Economic Review, 75(4) pp.850-855

Cetorelli, N. and P. Strahan (2004), “Finance as a Barrier to Entry: Bank Competition and
Industry Structure in Local US Markets,” NBER Working Paper 10832

Ciocchini, F., Durbin, E., and D. T. Ng, (2002), “Does Corruption Increase Emerging
Market Bond Spreads?”, manuscript

Demirgiic-Kunt, A. and V Maksimovic, (1998), “Law, Finance, and Firm Growth”, Tbe
journal of Finance, 53(6) pp. 2107-2137

Denizer, C., Iyugun, C. and A. Owen (2000), Finance and Macroeconomic Volatility
Depken, C., LaFountain, C. and R Butters (2006), “Corruption and Creditworthiness:
Evidence from Sovereign Credit Ratings”, Manuscript

Djumashev, R. (2007), “Corruption, Uncertainty and Growth”, Department of Economics
Discussion Paper 3716, Monash University

Freixas, X. andJ. Rochet (1997), “Microeconomics of Banking,” The MIT Press, Cambridge,
Massachusetts

Garmaise, M. and J. Liu (2005), “Corruption, Firm Governance and the Cost of Capital,”
Anderson Graduate School of Management, UCLA

Gaviria, A. (2000), “Assessing the Effects of Corruption and Crime on Firm Performance”,
Manuscript, Fedesarrollo, Bogota, Colombia

76

Gertler, M and S. Gilchrist, (1994), “Monetary Policy, Business Cycles and the Behavior of
Small Manufacturing Firms”, Quarter/J journal of Economics 109 (2), 309-340

Gilchrist, S. and E. Zakrajsek (1995), “The Importance of Credit for Macroeconomic
Activity: Identification through Heterogeneity,” in Is Bank Lending Important for
the Transmission of Monetary Policy? Eds. J. Peck and E. Rosenberg

Honohan, P and T. Beck (2007), “Making finance work for Africa” World Bank,
Washington, DC

Hunt,J and Sonia Laszo (2005) “Bribery: Who Pays, Who Refuses, What are the Pay-offs?”
The William Davidson Institute, Working Paper No. 752, University of Michigan

Kamel, S. (2005), “The Use of Information Technology to Transform the Banking Sector in
Developing Nations,” Information Technology for Development, 11(4), pp. 305-312

Kuncoro, A. (2004), “Bribery in Indonesia: Some Evidence from Micro-Level Data”, Bulletin
ofIna’oneria Economic Studies, Vol. 40, No. 3, pp 329-54.

Lee, C. and D. Ng (2002), “Corruption and International Valuations: Does Virtue Pay?”
Mimeo, Cornell University

Love, I. (2003), “Financial Development and Financing Constraints: International Evidence
from the Structural Investment Model,” Review of Financial 5 tudier, 16(3), pp.765-791

Mauro, P. (1995), “Corruption and Growth”, Quarterl ' journal of Economics 110, 681-712.

Mocan, N.H. (2007), ‘What Determines Corruption? International Evidence from Micro
Data,” NBER Working Paper W10460

Polinsky, A. and S. Shavell (2001), “Corruption and Optimal Law Enforcement”, journal of
Public Economics, vol. 81, pp. 1-24

Reinika, R. and J. Svensson, (2002), “Measuring and Understanding Corruption at the Micro
Level”, Unpublished Manuscript

Sherif, K. (2005), “Making it Easier to Do Business in South Europe”, in EU Accession —
Financial Sector Opportunities and Challenges for Southeast Europe, Springer Berlin
Heidelberg

Stiglitz, J. E. and A. Weiss (1981), “Credit Rationing in Markets with Imperfect
Information,” Toe American Economic Review, 71(3), pp.393-410

Sullivan, J. and A. Shkolnikov (2004), “Combating Corruption: Private Sector Perspectives

and Solutions,” Center for International Private Enterprise, Economic Reform Issue
Paper No. 0409

77

Tanzi V., and H. Davoodi (1997), “Corruption, Public Investment and Growth”, UHF
Working Paper 97 / 739

Tanzi, V. (1998), “Corruption around the World: Causes, Consequences, Scope and Cures”,
IMF Working Paper 98 / 6 3

Wei, S. (1997), “How Taxing is Corruption on International Investors?” NBER Working
Paper 6030

Wooldridge, J..M (2001), “Econometric Anabuir of Crow Section and Panel Data ”, The MIT Press.
Cambridge, Massachusetts

Wooldridge, J..M (2001), “Introductory Economic: A Modern Approacli”, The MIT Press.
Cambridge, Massachusetts

Wu, X. (2005), “Firm Accounting Practices, Accounting Reform and Corruption in Asia”,
Poligl and Sociegl, 24(3), pp. 53—78

78

Table 2—1 a

Summary Statistics: All Countries

 

Lending Outcomes
bnklns

interest

collat

1nterm

Corruption Measures

brbfreq

corrup

nobnbe

brbpay

advance

brbsize

inspecs

Public Infrastructure
serviceq

commtech

legal

Firm Characteristics
manufac

small

govern

foreign

Age

Mean

0.18
17.13
149.65
26.89

1.71
2.18
3.00
2.51
2.55
1.30
1.30

7.26
-0.13
3.00

0.39
0.69
0.12
0.34
15.11

N

13169
4906
4168
5066

10448
12651
12473
12333
11211
11952
7123

12909
13528
10525

13528
13528
13530
13530
13521

79

Appendix 2-1
Variable Definition

 

 

 

 

 

Variable Definition Source
Probability that a firm can obtain financing BEEPS
bnklns from domestic commercial banks
The interest rate that banks charge banks for
interest loans
The amount of collateral firms have to pledge
col/at in order to get credit from banks
The time period over which a loan has to be
1nterm repaid

 

frequency of bribes (Inn/reg)

Number of times a firm paid a bribe to obtain
service

 

number of inspections (num_imp5)

Number of inspections a firm is subjected to
by government agents in different sectors

 

Pervasiveness of bribery (brbpay)

How common is it for firms in a particular
line of business to have to pay bribes to get
things done?

 

corruption perception (con-up)

The probability that firms consider
corruption an obstacle to their performance
and growth

 

uncorrupt official (nobrilre)

The likelihood that a firm faced with a bribe
demand can obtain service from another
officer without paying the bribe

 

advance

Do firms know the magnitude of the
unofficial payment in advance

 

bribe size (bronze)

The amount firms pay in a given period in
bribes measured as a proportion of revenues

 

comm. technology

Firms’ use of different forms communication
and information technology

 

The quality of the legal systems in terms of

 

legal system affordabiliifairness, efficiency
Perception of respondents that uncertainty is
uncertainty a constraint to their operations

 

foreign loans

Proportion of operations financed with
foreign loans

 

development Bank Loans

Share of firm operations financed with loans
from development banks

 

equity

Percentage of operations financed with equity

 

Is the firm’s primary activity manufacturing:
takes the value of 1 if this is the case and 0

 

manufacturing otherwise
Is the firm classified as small: measured by
small the number of employees?

 

government-owned

Takes the value of 1 if the firm is primarily
owned by government and 0, otherwise

 

 

foreign-owned

 

Is the majority shareholding in he hands of
foreigners?

 

 

80

 

 

Appendix 1 cont.

 

Number of years since the firm was

 

 

age established
GDP per Capita Growth WDI
inflation WDI

 

 

 

 

81

 

 

oood- mood- oood oood- omod wNod- mood Cod owod mood- 595w
Sod- oood- m ood m 5d- mmod mood- mood- Sod- m Nod- oood- emote-d
oood owod N- Sd oood womd —mod No — .o- 0 5d mood Nmod- Sesame
omod oood oood- oood NmNd- o_od- mood o_od wNod omod- =oEm
oood- om _ .o- wmod- om mo- wood- Nmod- om od 5- «d- owod- owod CE
mm _ .o ww _.o m ood- _ 5d Hm # d wood owmd- om od mwmd o 5d- amounted
_m _.o oomd wid- Nowd wofid oomd oood- oood am no mood :85.
mmod mmod _ood owod Rod- omod mood oood- no mo- N 5d 59858
mono w_ _d Food mm _.o Nwfid mm _.o wwod momd oood- N 5d- ouomtuo
oomd- owmd- omod m Nod- wmod- momd- mood mood- oood- oood- Eowfl
womd m owd ow _ d- momd Nood wmmd wNod- oood oood oood 033.5
oood mwod m m mo- owmd omod Emd oo—d- oood owfio omod- 85261.
oood m 3d- womd o_ yd oood Fw od- wo_d wofid Nwod- .9255
oood \- _Nd- m wod- om Fd- omod NNod- mood- ded- 2:50:
oood wood nowd oood- wood NH «d m-odd morocco
oood m m _.o w~ fid- 3 mo ow _d wood- “8%.:
ood; m ood- wood mood wood v2.35
ooo.— mood- Sum-d- 5 5d- 88::
oood S «d Hmod 3:8
oood omod- $235
ood; 82.3

3526a lowdown ohm-£0: mac-So teamed won-33 €85 8:8 5835 32:3

 

goofing-BU venue—um

N-N nigh.

82

 

 

ooo._ oooo- oooo oomo- :oo oooo omoo- owoo owoo- omoo Eoo- 53%
83 oooo oo_.o- Nooo oooo wooo o:.o mmoo- mooo omoo cote
83 go- Soo oooo Boo wwoo oooo zoo- owoo Essa:
oood zoo- o-So- Boo- Smo- oooo Soo- oowo =2...
83 oooo Eoo- Soo omoo- mooo wooo- o5
oood mzo omoo 5o- o_o.o- Nwoo tomato
oood wooo- _o_.o 32o- oooo :85
oood mEo- oooo wooo- £3858
oood aoo- wowo Esta
oood 9.3. is
oood 835
530w ammo-ood oooscoE :55 GE ammo-96m :85. 58888 duodenum Sowe— vain—.5

 

mucus—850 veto—om

60:52.00 N-N uEuH

83

Impact of Corruption on Lending Conditions

Table 2-4

Dependent Variable: Bank Loans

 

Standard Logit

Logit - Country Dummies

 

 

Coefficient SE Coefficient SE APES
number of inspections 0.306** 0.052 0.332** 0.057 0.064
inspsq -0.080** 0.019 -0.072** 0.020 -0.013
corruption perception -0.009* 0.005 -0.009* 0.005 -0.002
uncorrupt official 0.003 0.037 0.022 0.039 0.004
advance —0.014 0.036 0.037 0.038 0.006
bribe size -0.011 0.023 -0.018 0.024 -0.004
comm. technology 0.262** 0.044 0.232** 0.048 0.043
legal system -0.024 0.033 -0.010 0.035 -0.003
uncertainty 0.072 0.051 0.059 0.055 0.013
foreign loans -0.010 0.058 0.031 0.061 0.007
development bank loans -0.016** 0.006 ~0.017** 0.006 «0.003
equity -0.008** 0.003 -0.008** 0.003 -0.002
manufacturing 0.081 0.109 0.153 0.114 0.026
small -0.292** 0.126 -0.304** 0.132 -0.091
government-owned -1.280** 0.241 ~1.340** 0.246 -0.163
foreign-owned -0.226 0.166 -0.306* 0.172 -0.057
age -0.004 0.003 -0.007** 0.003 -0.002
GDP per Capita Growth -0.032** 0.012 0.075** 0.037 0.005
inflation -0.036** 0.013 -0.159** 0.053 -0.020
Number of Observations 2410 2410
Log likelihood 4120.753 -1056.91
Wald Chi_sq 167.8 295.490

 

** Significant at 5%, * significant at 10%

SE: Standard Error

84

Impact of Corruption on Lending Conditions

Table 2-5

Dependent Variable: Bank Loans - Frequency of Bribes

 

Standard Logit

Logit - Country Dummies

 

 

 

Coefficient SE Coefficient SE APES
brbfreq 0.027 0.020 0.021 0.020 0.004
corrup -0.010** 0.004 -0.011** 0.004 0.004
brbpay —0.014 0.046 0.032 0.048 ~0.002
nobribe -0.022 0.030 0.011* 0.006 -0.002
advance 0.015 0.040 0.024 0.041 0.005
brbsize -0.040** 0.018 -0.043** 0.018 —0.006
commtech 0.250** 0.034 0.223** 0.036 0.034
legalq -0.014 0.027 0.002 0.028 -0.001
uncert 0.048 0.041 0.016 0.043 0.007
frnlns 0.024 0.046 0.029 0.048 0.005
devlns -0.010** 0.004 -0.011** 0.004 -0.002
equity -0.007** 0.002 -0.006** 0.003 -0.001
manufac 0.052 0.087 0.111 0.091 -0.112
small -0.519** 0.096 -0.538** 0.100 -0.021
govern -1.302** 0.181 -1.351 0.185 -0.130
foreign -0.175* 0.092 -0.209** 0.096 —0.039
age 0.002 0.002 0.000 0.003 -0.095
percapgr -0.016** 0.004 0.007 0.008 0.002
infl -0.026 0.011 -0.048 0.036 -0.012
Number of Observations 3666
Log Likelihood —1687.725
Wald Chi_sq 387.940
** significant at 5%
*significant at 10%

SE: Standard Error

85

Table 2-6

Firms' Access to External Finance

 

Percentage Loans in

 

 

 

Financing Probability of Obtaining Credit
Tobit Estimation Logit Estimation
Coefficient SE APE Coefficient SE APE
num_insps 11.107** 1.881 0.064 0.332** 0.057 0.058
inspsq ~2.428** 0.649 -0.013 —0.073** 0.020 -0.012
corrup -0.395** 0.185 —0.002 -0.010** 0.005 0.002
nobribe 1.235 1.311 0.004 0.022 0.039 0.006
advance 1.105 1.273 0.006 0.037 0.038 0.004
brbsize -0.183 0.778 ~0.004 -0.018 0.024 -0.001
commuse 6.882** 1.580 0.043 0.232** 0.048 0.034
legalq 0.648 1.161 -0.003 -0.010 0.035 -0.004
uncert 2.501 1.826 0.013 0.059 0.055 0.014
foreign 0.5128 2.057 0.007 0.031 0.061 0.004
devlns -0.642** 0.201 -0.003 -0.017** 0.006 -0.003
equity -0.334** 0.118 —0.002 -0.008** 0.003 -0.002
manufac 2.815 3.806 0.026 0.153 0.114 0.012
small -9.843** 4.430 —0.091 -0.304** 0.132 -0.073
govern -44.621** 7.795 -0.163 -1.340** 0.246 -0.066
foreign -9.918* 5.774 0.057 -0.305* 0.172 -0.041
age -0.189* 0.1 13 -0.002 -0.007** 0.003 -0.001
percapgr 1.963 1.191 0.005 0.075** 0.037 0.001
infl -4.805** 1.280 -0.020 -0.159** 0.053 -0.018
Number of
Observations 2140 2410
Log Likelihood 3944.369 —1056.91
Wald Chi_sq 321.4 295.490
** Significant at 5%

*significant at 10%
SE: Standard Error

86

Table 2-7
Impact of Corruption on Lending Outcomes
Dependent Variable: Interest Rate

 

 

 

 

 

Pooled OLS Random Effects Fixed Effecrs
Coefficient SE Coefficient SE Coefficient SE
brbfreq -0.100 0.119 -0.048 0.109 -0.117 0.117
corrup 0.115 0.271 -0.130 0.256 -0.004 0.273
brbpay 0.740** 0.263 0.346 0.256 0.695** 0.271
nobribe 0.027 0.163 0.092 0.180 0.130 0.193
advance -0.477** 0.212 -0.463** 0.216 -0.572** 0.230
brbsize 0.307** 0.150 0.149** 0.085 0.325** 0.090
commtech —1.568** 0.251 -0.847** 0.202 -1.430** 0.208
legalq -0.218 0.155 —0.287* 0.156 -0.279* 0.165
uncert 0.676** 0.235 0.447* 0.234 0.672** 0.245
frnlns -0.029** 0.014 -0.033** 0.013 -0.029** 0.014
devlns -0.008 0.010 -0.012 0.015 -0.007 0.016
equity -0.007 0.009 0.004 0.014 -0.015 0.014
manufac 1.593** 0.487 1.409** 0.491 1.725** 0.525
small -0.101 0.575 0.564 0.542 0.281 0.578
govern 4.269** 1.222 4.366** 0.960 4.561** 1.034
foreign 1.957** 0.519 2.739** 0.530 2.068** 0.558
age -0.032** 0.013 —0.030** 0.015 -0.032** 0.015
gdpcapgr 0.096 0.070 0.371 * 0.190 0.005** 0.079
inflation 0.686** 0.044 0.560** 0.040 0.673** 0.024
Number of
Observations 1578 1439 1439
Groups 21 21
R2 0.44 0.43
** Significant at 5%
*significant at 10%

SE: Standard Error

87

Table 28
Impact of Corruption on Lending Conditions
Dependent Variable: Collateral Requirements

 

Pooled OLS Random Effects Fixed Effects

 

 

 

Coefficient SE Coefficient SE Coefficient SE

 

 

brbfreq 0.656 1.085 -0.111 0.987 0.362 1.028
corrup -1.331 2.311 -0.156 2.308 -1.038 2.395
brbpay 2.340 2.433 3.139 2.344 2.127 2.407
nobribe 0.957 1.772 3.188** 1.623 1.507** 1.690
advance 1.185 1.956 0.689 1.959 1.507 2.027
brbsize -0.071 0.845 -0.777 0.770 0.070 0.794
commtech -2.052 1.802 -0.571 1.837 -1.567 1.809
legalq -0.918 1.429 -0.934 1.419 -0.756 1.447
uncert 5.623** 2.094 3.356 2.140 4.723** 2.173
frnlns —0.167* 0.100 -0.125 0.114 -0.178 0.118
devlns 0.018 0.130 -0.078 0.146 -0.070 0.151
equity -0.143 0.111 0.052 0.129 -0.166 0.127
manufac 4.874 4.634 3.424 4.454 4.400 4.611
small 5.744 4.994 5.136 4.910 5.400 5.083
govern 3.558 9.190 -0.668 8.932 0.901 9.355
foreign -6.496 4.871 -7.801 4.766 -7.678 4.878
age —0.028 0.110 ~0.142 0.134 -0.019 0.138
gdpcapgr —0.344 0.667 -1.610 1.685 -0.609 0.712
inflation -1.606** 0.177 -0.851** 0.362 -1.622 0.211
Number of

Observations 1356 1277 1277

Groups 21 21

R2 0.08 0.11

** Significant at 5%

*significant at 10%
SE: Standard Error

88

Impact of Corruption on Lending Outcomes

Table 2-9

Dependent Variable: Loan Term

 

 

 

Pooled OLS Random Effects Fixed Effects
Coefficient SE Coefficient SE Coefficient SE
brbfreq 0.253 0.332 0.229 0.314 0.263 0.324
corrup 1.025 0.760 1.075 0.746 1.425* 0.766
brbpay -0.393 0.857 -0.004 0.748 -0.531 0.758
nobribe 0.670 0.556 0.675 0.524 0.548 0.539
advance —1.572** 0.576 -1.192** 0.631 -0.226** 0.648
brbsize -0.214 0.218 -0.185 0.248 -O.242 0.254
commtech 2.684** 0.526 1.498** 0.588 2.281 ** 0.580
legalq 1.039** 0.430 0.860** 0.456 1.099** 0.461
uncert -0.138 0.759 0.605 0.692 0.025 0.695
frnlns 0.113** 0.038 0.041 0.036 0063* 0.038
devlns 0.078** 0.041 0.055 0.045 0.051 0.047
equity 0.020 0.042 0.029 0.041 0.033 0.039
manufac -5.626** 1.442 -4.477** 1.436 -5.431** 1.474
small 3.972** 1.656 1.324 1.590 1.855 1.630
govern 0.369 3.522 0.563 2.766 0.242 2.862
foreign -5.103** 1.488 -6.373** 1.552 -5.762** 1.575
age -0.004 0.043 —0.070 0.043 -0.037 0.043
gdpcapgr -0.182 0.317 1.135** 0.531 0.024 0.225
inflation -0.435** 0.042 0.089 0.1 1 1 -0.405** 0.067
Number of
Observations 1602 1461 1461
Groups 21 21
R2 0.1 0.12

 

** Significant at 5" 0
*significant at 10%
SE: Standard Error

89

CHAPTER 3. ARE BANKS IMPORTANT IN TRANSMITTING MONETARY
POLICY SHOCKS TO THE REAL ECONOMY?

1 Introduction

How are monetary policy changes transmitted to the real economy? This is an old
question in macroeconomics, but one that remains a subject of active debate.24 Although
there is a general consensus that monetary policy does affect economic activity, there has
been substantial disagreement about the channels through which this effect is transmitted.
There is even more disagreement about the role and importance of the banking sector,
(Romer and Romer, 1990; Christiano, Eichenbaum and Evans, 2000; Ramey, 1993; and,
Rudebush and Oliner, 1994). Nevertheless, recent literature has provided some evidence
that banks do transmit monetary policy shocks to the real sector. But the importance of this
mechanism has yet to be established convincingly.

Recent studies have analyzed the issue of transmission mechanisms through the
banking sector, stressing the importance of financial frictions in the process.25 Banks are
considered to have some specific characteristics such as size and balance sheet strength that
introduce heterogeneity in their ability to raise external finance with which to smooth their
lending during periods of monetary contractions. Because of these differences, faced with a
policy induced deposit shortfall, some banks will be forced to cut back on their lending
volume, thus affecting investment and production activity. Thus exogenous shocks will have
different effects on the lending of banks with different characteristics. Empirical studies

have appealed to some aspect of bank financial strength and bank size as sources of

 

2" See for example, Bemanke and Blinder, 1988, 1992; Gertler and Gilchrist, 1996; Rudebush and Oliner,
1994; Romer and Romer,1990; Stein, 1998; Kashyap, Stein and Wilcox, 1993; Bemanke and Gertler, 1995;
Jarayante and Morgan, 2000; Nilsen, 1999; Van den Heuvel, 2001; Opiela and Kishan, 2000; Driscoll,
2003; and Kashyap and Stein, 1993, 2000.

25 See for example, Kashyap and Stein , 2000; Nilsen; 2002; and Bichsel and Perrez, 2004

90

heterogeneity and their findings support the existence of differential response to shocks
(Kishan and Opiela, 2000).

In this paper, we consider the importance of bank financial soundness in the
transmission mechanism of monetary policy shocks. Our work differs from existing studies
in that we appeal to a broader measure of bank soundness as opposed to using just one
measure of financial strength.26 Specifically, we use CAMEL-type variables as a measure of
overall financial strength and add bank size for a more complete measure of heterogeneity.

The main aim of our paper is to test whether overall financial soundness is an
important factor in determining bank-lending response to policy shocks. We appeal to bank
heterogeneity to uncover differential response functions for sound and ‘unsound’ banks, and
ultimately to uncover the role of banks in the transmission of monetary policy. Our initial
hypothesis is that, given two banks of the same size, a bank with a stronger financial position
should have better ability to mitigate the effects of a policy-induced deposit outflow.

Thus, this paper adds to the existing literature by re-examining the bank-lending
channel using a panel of quarterly disaggregated data on U.S. commercial banks. To identify
loan supply response to monetary policy, we appeal to specific characteristics that
differentiate banks, namely bank size and financial soundness of the institutions. The
obvious source of heterogeneity appears to be bank size (measured in terms of total assets).
Stylized facts suggest that based on size alone, behavior of banks differ significantly and the
constraints that banks face also vary according to whether the bank is large or small, see for
example Gilchrist and Zakrajsek, 1995. For example, small banks suffer from a

disproportionately large share of information asymmetry problems and therefore, typically

 

2’ There are two specific papers that do this. The first is the Kashyap and Stein which uses liquidity and
bank size as sources of heterogeneity and the other is by Kishan and Opie1a(2000). This study uses bank
capital as a gauge of financial strength. It also uses bank size as an additional source of heterogeneity.

91

encounter more difficulty when trying to raise external finance. Conversely, large banks are
generally presumed to be sound and safe and this feeds into the flight-to-quality
phenomenon whereby during downturns, firms and individuals move their financial assets
from small to large institutions. Also, because information problems are less severe for large
banks, these banks can raise managed liabilities more easily. Because of these differences,
Kashyap and Stein (2000), henceforth KS and Kishan and Opiela (2000), henceforth KO,
among others have shown that banks of different sizes react differently to shocks since they
face different constraints. As a result, if we pooled the sample for all banks, the estimated
long-run responses to policy shocks may not be informative with respect to the effect of
monetary policy on loan growth. To this end, in testing for the existence of loan supply
effects, we first separate banks into different size groups.

We find that overall financial soundness matters in how bank lending responds to
policy shocks. In particular, most of the individual ratios also indicate that banks have a non-
trivial role in this process. However, most important and more interesting is the finding that
even for the largest banks, which have so far been assumed completely insulated from policy
shocks, overall financial health plays a role in how lending is affected by policy shocks.

There are two papers that are closely related to ours. The first is by Kashyap and
Stein (2000), hereafter KS. This paper has provided the most convincing evidence for the
existence of the bank-lending channel by appealing to agent heterogeneity to uncover
differential responses to policy shocks. Their intuition is that if the observed decline in bank
lending after a monetary policy shock was the result of changes in demand patterns, then all
banks should be equally affected. However, if the decline in loans was the result of
constrained lending capacity on the part of the banks, then some banks should experience a

larger drop in lending activity. Thus using bank level data on U.S. banks covering the period

92

1976 to 1993, the authors analyze the effects of monetary policy on banks of different sizes
(measured in total assets) and liquidity. Liquidity is used as a proxy for the strength of the
bank balance sheet. They find that small less liquid banks respond more strongly to
monetary policy contractions because of their relative inability to raise funds from uninsured
sources.

However, while we recognize that the size of a bank is important in how the bank
reacts to policy changes, we believe that using only liquidity to measure balance sheet
strength may not be enough. There are several reasons why a bank might hold a large stock
of liquid assets. For instance, banks that are subject to severe information asymmetries may
choose to hold large stocks of securities as a cushion against any disruptions in deposit
flows. Also, as KS point our, banks that lend to firms that are prone to cyclical fluctuations
may hold liquid assets that reflect this aspect of their business. Large holdings of liquid assets
could also be an outcome of an excessively risk averse institution and this would be reflected
in tighter lending practices. Thus, liquidity may not necessarily be an adequate indicator of
financial strength. Further, there are other variables that measure financial strength, perhaps
better than the liquidity, and leaving them out of the estimation could result in a significant
bias.

The other study is by Kishan and Opiela (2000) who use bank capitalization to proxy
for bank strength and this is based on the fact that capital can mitigate agency problems in
the market for external finance and determines the capacity of banks to issue liabilities. Thus,
according to the authors, a high level of capital reflects a healthy balance sheet. However,
this is not necessarily the case. On the one hand, a highly capitalized bank has a better ability
to absorb losses and hence, has a small probability of becoming insolvent. Holding capital is

at the same time costly to banks in terms of foregone high return investments. As a result,

93

banks will opt to hold the lowest possible level of capital, and observing a bank holding a
high level of capital can potentially be an indication of high risk of the underlying assets.
Hence, on its own, capital may not be an adequate measure of bank financial soundness and
interpretation of results based on capital as it appears on balance sheets is unclear.

Our paper differs from these studies in two fundamental respects. First, to address
the shortcomings of using a single indicator of bank financial strength, we use the full set of
CAMEL ratings. Secondly, we use estimation techniques that allow us to control for specific
features of the data and the structure of the banking market. Specifically, our estimation
approach allows us to control for bank specific fixed effects which KS explicitly do not
control for because with their approach, they would lose a significant part of their
observations. Finally, one feature of the U.S. banking industry that stands out is the declining
number of banks. This attrition could be potentially problematic for empirical analysis if it
results in the sample changing in a non-random way. In this paper, we explicitly test for
effects of sample attrition. ’

The CAMEL ratings are the most commonly used and accepted measure of financial
soundness for banks.27 The Fed uses a mixture of quantitative and qualitative measures to
determine the composite CAMEL rating for each bank. Because we only have information
on quantitative measures of soundness, we compute the ratios included in the CAMEL
calculation from the banks’ balance sheets and income statements. In order to avoid
imposing ad-hoc weights on the ratios, we include each of these ratios as a separate
explanatory variable in the regressions. High ratios are assumed to indicate overall financial

strength while banks with low ratios are considered weak or unsound. Intuitively, investors

 

27 CAMEL is an acronym for capitalization, asset quality, management, earnings, liquidity and sensitivity
to market risk. These are the six components used for determining bank safety and soundness. In this paper
we assume low ratios to imply weakness or a lack of financial soundness. The higher the ratios are the
stronger or the more sound the bank is assumed to be.

94

should care about the examiner’s information about a bank when making a decision to lend
to a particular bank. Thus CAMEL ratings should play a role in whether the bank can issue
external debt. De Young et. a1. (2000) find evidence that the bond market is more sensitive
to examination information about institutions with low CAMEL ratings as they are
perceived as being troubled with a high potential for default, than they are to institutions
with a high rating.

One well-established theory is that banks keep liquid assets as a buffer stock, so that
whenever they have a shortfall in deposits, they can draw down their stock of securities to
continue their lending (Lucas and Macdonald, 1992 and, Alger and Alger, 2000). Thus banks
with ample holdings of securities will be in a better position to smooth the effects of
monetary contractions. We are, therefore, also interested in the role of liquid assets in this
process. Hence, we also investigate the behavior of banks with respect to maintaining a
given level of securities.

This paper uses quarterly data from the Call Report on all insured commercial banks
in the US from the first quarter of 1991 to the last quarter of 2000. Our results lend support
to the existence of a bank-lending channel. The estimation shows that bank loan supply is
indeed sensitive to differentiating characteristics, and most importantly the results support
the contention that strong and weak banks respond differently to policy changes. Consistent
with the findings of Kishan and Opiela (2000) and KS (2000), the effects appear to be
strongest for small banks.

The rest of the paper is organized as follows. In section 2, we lay out some
theoretical underpinnings for the existence of the bank-lending channel as well as existing
evidence. Section 3 presents the data with the details of how the variables of interest were

constructed. In section 4 we describe the estimation methodology with a description of

95

problems that the model addresses. we also discuss the results in this section. W’e conclude

with section 5.

2 The Bank Lending Channel: Theoretical Framework and Evidence

2.1 Theoretical Underpinnings

Two major views of the monetary transmission mechanism are the money View and
the credit view. The money view is based on the notion that reductions in outside money
cause interest rates thus increase. In the traditional view, bank loans and bonds are assumed
to be perfect substitutes. Thus, banks do not have a special role in monetary policy
transmission. \When policymakers change bank reserves thereby changing the quantity of
money, they affect the nominal interest rates, which in turn, causes changes in the real
interest rate and consequently affect aggregate demand. This view, however, fails to account
for the observed timing or the persistence or the distributional effects of monetary policy
shocks. In an attempt to resolve these puzzles, researchers have turned to theories of
imperfections in financial markets and their impact on the cost and availability of credit. The
effects arising from frictions in the credit markets reinforce the operations of the interest
rate channel and amplify the effects.28 This ‘alternative’ literature identifies two related ways
in which the financial sector can propagate and magnify the effects of monetary policy
shocks: the bank-lending channel and the broad credit channel. Both these channels rely on
information asymmetries between borrowers and lenders, which, in general, influence the
cost of funds to the borrower. Further, these channels are not mutually exclusive; rather

each channel reinforces the effects of the other.

 

28 This is well articulated in Bemanke and Blinder (1988) which extends the conventional lS-LM model to
show that contractions in the monetary base shift the supply of credit thereby imposing a greater restraint
on real activity than that implied by the conventional money view alone.

96

The first of these theories, the bank lending channel, argues that monetary policy
shocks affect real spending through availability of credit. Thus, the essential feature of this
channel is that by altering the quantity of base money, monetary policy actions should have a
direct impact on the supply of loans. Furthermore, for banks to be effective in policy
transmission there must be some borrowers who depend on banks for their external
finance.29 This bank-dependency can emanate from borrowers facing different costs of
external finance because of information problems and thus, having to rely on intermediated
finance to reduce agency costs. Hence the importance of bank credit depends on the severity
of information costs with the role for financial intermediaries being more pronounced when
high information costs create a significant gap between the cost of internal and external
finance. As a result, any shift in the ability or willingness of banks to supply credit will affect
these borrowers directly, generating bank loan supply effects. In these situations, banks are
better placed to provide credit because they specialize in creating covenants on loan
contracts as well as establishing special relationships with their clients. These characteristics
of financial contracting make bank credit an imperfect substitute for non-bank credit.
Consequently, this imperfect substitution between bank and non-bank credit contributes to
the bank-lending channel of monetary policy transmission.

The second theory is the broad credit channel. This channel focuses on the nature of
relationships between borrowers and lenders, and consequently on borrowing terms.
Specifically, information asymmetries between contracting parties create a wedge between
the cost of internal and external finance. Without information asymmetries, the interest rate

charged on a loan should reflect the cost of funds and the risk characteristics of the

 

2’ This dependence can be justified on the basis of there being fixed costs to direct participation in financial
markets. Because of their ability to monitor borrowers at least cost, banks are better placed to economize on
these fixed costs making them the natural lenders to borrowers that might have characteristics, e. g. size,
that prohibits them from issuing securities directly to the market.

97

borrower. However, information and incentive problems will affect this cost. Borrowers’ net
worth is also an important component of this channel since it can mitigate the agency costs
emanating from these imperfections. In particular, the external finance premium -the gap
between the cost of internal and external finance- varies inversely with net worth (See among
others Kiyotaki and Moore, 1997; Bester, 1995; Bernanke, Gertler and Gilchrist, 1996).
When borrowers’ net worth is low, borrowers have less collateral with which to back their
loans. Furthermore, low net worth increases moral hazard problems since the borrowers
have a smaller stake and would have more incentive to engage in risky projects. Since low
levels of net worth imply a higher probability of loan default, it will lead to a reduction in
lending and hence in real spending. An adverse shock that affects the borrowers’ net worth
increases the cost of external finance and restricts borrowers’ access to finance. Thus the
effects are more intense during monetary contractions since the resulting high interest rates
increase debt service and reduce the value of collateralizable net worth and consequently the

marginal cost of external finance increases.

2.2 Empirical Studies

In addition to these theoretical underpinnings, there is a substantial volume of
empirical work seeking to demonstrate the relevance of the financial system with some
studies devoted exclusively to the importance of financial intermediaries.

Earlier studies (Bernanke and Blinder, 1992; Ramey, 1993; Oliner and Rudebush,
1994) utilized aggregate data on variables such as total bank loans, the mix between
commercial paper and bank loans, output, interest rate spreads and other economy-wide
variables. These studies, e.g. Bernanke and Blinder, have shown that bank loans declined
after monetary tightening and have argued that this provided evidence for bank-lending

effects. Because monetary policy shocks affect loan demand as well as loan supply,

98

identification requires adequate separation of shifts in loan supply from movements that
originate on the demand side. Identification has, therefore, been a major problem with these
attempts because aggregate data does not adequately control for loan demand effects, and as
such, makes it difficult to identify effects emanating from credit availability. Hence, even
though the results of these studies provide evidence consistent with the existence of loan
supply effects, the interest rate channel could also yield similar results. Thus, evidence from
these efforts has been mostly inconclusive. Kashyap, Stein and W’ilcox (1993), hereafter
KSW, tackle this identification difficulty by investigating movements between bank loans
and non-bank financing for which they use commercial paper as a proxy. Using a simple
model of firms’ capital structure choices to highlight conditions under which distinctions
between loans and securities are irrelevant, they explicitly test for two conditions: 1) whether
on the asset side of the banks’ balance sheets, loans and securities are considered imperfect
substitutes which is necessary for monetary policy to affect loan supply; and 2) whether for
firms’ financing there is an imperfect substitutability of loans and alternative finance so that
reduced loan supply can generate real effeCts. Their intuition is that shifts in substitutes for
bank financing should reveal information about the demand for loans. If there are loan
supply effects, then there should be a negative co-movement between loans and commercial
paper. The study finds evidence for the existence of an active bank-lending channel.

Following a policy shock, firms’ issuance of commercial paper increases and loans decline,

albeit at a slower rate. The slow response of loans is attributed to prior commitments that

banks are obliged to fulfill.

The existence of the bank-lending channel relies on lending responses emanating
from loan supply effects. In addition, these effects should be disproportionately large for less

creditworthy borrowers, which should have difficulty substituting lost deposits with external

99

forms of finance. In order to identify differential. responses to shocks, current research
appeals to heterogeneity of agents through the use of microeconomic data. Gertler and
Gilchrist (1994) use firm level data on U.S. manufacturing firms to test whether the response
of small firms to monetary policy shocks differs from that of large firms. Their intuition is
that because small firms are more likely to rely on banks for their external finance, loan
supply effects from policy tightening should show up in small firms having to reduce their
borrowing more sharply. The study finds that indeed, the impact of policy is larger for
smaller firms.

Along the same line of thinking, Kashyap and Stein, 1995 and 2000, focus on cross-
sectional differences in financing and lending decisions of banks of different sizes and
balance sheet positions. In this study, the authors are interested in how banks of different
sizes and different liquid asset holdings react to policy changes. They argue first that less

liquid banks will reduce their loans if they are not able to replenish lost insured deposits with
other forms of finance and they specifically expect to see 02Li t /aBi t 8M it < 0for less

liquid banks. HereB stands for liquidity,M is the monetary policy indicator andL stands for
lending while iand tare indexes for bank and time, respectively. They test this hypothesis by
focusing on how small and large banks differ in their ability to issue managed liabilities. They
find that, indeed, the effect of monetary policy is strongest for the smallest banks and that
large banks are not sensitive to contractionary policy shocks because they have access to the
market for uninsured funds.

Based on alternative theoretical bases for the existence of capital market
imperfections, other studies following the same line of inquiry have used capitalization to

measure balance sheet strength and hence banks’ access to other forms of finance. Theory

100

suggests that bank leverage contributes to the existence and severity of capital market
imperfections. Furthermore, a bank’s capital determines the ability to issue liabilities and also
determines the amount of loans and securities it can hold. Equity can mitigate information
asymmetry problems in the market for managed liabilities. For instance, the assets of an
undercapitalized bank will be considered riskier than identical assets of a well-capitalized
bank because the latter has more ability to absorb futures losses, should any result. Further,
capital constraints limit the bank’s ability to finance profitable projects. Thus several studies
have attempted to uncover loan supply effects using bank capital as a discriminating
characteristic. Scholars have argued that information problems are more severe for small
banks, thus their ability to raise funds in the capital market will be more constrained relative
to large banks. Following on this line of thought and the capitalization argument, Kishan and
Opiela (2000) sought to establish loan supply shifts in response to monetary policy by
dividing banks according to size (based on total assets) and their capital ratio.30 This study
emphasizes the importance of bank capitalization in explaining the effect of policy shocks on
loan growth. In particular, the authors aim to show that small-undercapitalized banks are
unable to raise alternative funds to continue their lending activity during a monetary
contraction. Their findings support the existence of loan supply effects. Lending of
undercapitalized banks is more sensitive to monetary policy. Further, this finding appears to
be more relevant for small banks.

Van den Heuvel (2001) also focuses on bank capital as a source of differential
response but looks more at the regulatory framework. He emphasizes the role of banks’
capital structure in influencing the response to policy—induced changes in interest rates and

other relevant variables. His view is that bank equity can affect the bank-lending channel by

 

3° For another interesting study on the relationship between bank capitalization and credit market
imperfections, see Jarayatne and Morgan, 2000.

101

its relationship with adverse selection and moral hazard problems. Effectively, a bank that
has low equity suffers more from information asymmetry problems than a well-capitalized
bank. Thus the level of capital can have a mitigating effect on these problems. He further
shows that there is more risk attached to liabilities of under-capitalized banks. Thus for these
banks, it is optimal to issue less debt and reduce lending following a policy shock.
Consequently, he sees the interaction between capital adequacy requirements and monetary

policy as an important cause of loan supply shifts.

2.3 Changing Structure of the Banking Industry and the Bank-Lending

Channel

During the past two decades, financial markets have undergone a major
transformation driven mainly by rapid growth and diffusion of new technology that has
resulted in lower transaction costs and reduced information asymmetries. There has been a
pronounced trend towards market-based financing with a tremendous growth of capital
market activity that is evident from the decline in bank assets as a share of total financial
assets, D’Astria (2000).

These developments have led to intensified competition in financial markets that
have resulted in narrowed margins for traditional banking activities. Access to market
finance has become easier and asset substitutability among financial markets has improved.
Banks have, therefore, had to reposition themselves in order to remain profitable by greatly
expanding the scope of their activities. An important strategic development in this regard has
been the move by banks into new areas of off-balance sheet activities such as derivatives and

securitization of loans”. The process of securitization presents several advantages to the

 

’ ' Securitization is a process by which pools of similar assets are sold in the form of tradable securities.

102

securitizing banks. By facilitating unbundling and trading of risks, securitization has
improved efficiency and profitability of banks’ operations (Mishkin & Strahan, 1999).
Securitization also provides banks with additional funding since banks can originate loans
and securitize them immediately, thus obviating the need to fund them on an on—going basis.
Thus banks are able to originate more loans than they would if they could not sell some of
these loans. Phillips (1996) argues that many of the loans that banks securitize would not
have been originated if banks had been required to keep them on their balance sheets. Also
the ability to engage in off-balance sheet activities is shown to permit banks to invest in
loans with positive net present values that they would pass up if restricted to deposit
financing. Finally by removing some assets from their balance sheets, banks are able to lower
the burden of capital requirements.

What are the implications of these developments for monetary policy transmission,
in particular for the bank-lending channel? We have argued that the existence of the bank-
lending channel is based on credit market imperfections that are caused by asymmetric
information. Additional requirements for the existence of this channel are that there should
be imperfect substitutability between retail deposits and wholesale deposits so that a fall in
retail deposits induced by monetary contraction is followed by a decline in loans rather than
an increase in wholesale deposits. Further, some borrowers should be bank dependent in the
sense that they cannot easily switch to alternative forms of external finance. Thus, at a glance
it would appear that the developments in the financial sector negate the very basis of the
bank-lending channel as they brought about a reduction in information asymmetries,
improved funding so that the dependence on insured deposits is reduced and an
improvement in the substitutability of assets and liabilities. In fact, Estrella (2000)

hypothesizes that by reducing credit market imperfections these developments must have

103

weakened the credit channel and the bank-lending channel. \X’e argue differently and instead
see these developments, particularly the loan sale process, as additional sources of the bank-
lending channel. Existing studies seek to establish the existence of the bank lending channel
by showing that different classes of agents are affected differently by monetary policy. In
particular, it has been shown that small banks, especially when they are not adequately
capitalized or liquid are affected more by shocks (Kishan and Opiela). In this paper, we
hypothesize that the soundness of a bank will affect the way a bank reacts to monetary
contractions. Thus to see whether securitization affects the bank lending channel, we need to
know what type of banks securitize. If all banks can securitize their assets, then deposit
shortfalls induced by policy shocks would be offset by funds raised through loan sales.
Therefore, banks would not have to reduce their lending. Hence, these changes would
significantly reduce the potency of the bank-lending channel of monetary policy only if they
are distributed equally among banks. However, not all banks securitize. W’hen banks sell
loans, they are mostly selling their reputations. Specifically, it is likely that large banks are the
main participants in these activities. Existing studies, e.g. Garrido (2000), show that even
though all banks have the possibility of originating assets for securitization, small banks lack
the necessary resources to be effective participants. These banks can only securitize if they
combine their resources to deliver large pools of assets, which could be sold. Also with
securitizing, the risk exposure is a main consideration because even though investors
consider securitized assets safer, buyers are not able to assess the quality of the underlying
credits. Hence it is conceivable that buyers of these assets will tend to prefer the credits from
large banks because of perception of size being equivalent to good quality. Thus, our
contention that instead of reducing the potency of the bank-lending channel, these changes

add additional sources through which this channel manifests.

104

2.4 Some Stylized Facts

The analysis in this paper relies on banks’ ability to shield their loan supply from the
effects of a contractionary policy change. W’e want to test whether actions of the Federal
Reserve do affect the loan supply as suggested by KS, BB, KSW’ and others. If this is the
case, a positive policy innovation should lead to a decline in deposits, reducing the banks’
lending capacity. To counteract the effects of this decline, banks will issue external debt or
sell their securities so that they can maintain their portfolios rather than reduce their lending
activities. However, it has been shown that banks (De Young et. al, 2000) do not have equal
access to markets for debt instruments, some classes of borrowers encounter difficulties
issuing debt. This is because, unlike retail deposits, the liabilities in external markets are not
covered by federal deposit insurance. They are therefore, subject to considerations of
information asymmetries whereby lenders are concerned with the perceived creditworthiness
of the borrowers and also with the knowledge that borrowers have private information
about the quality of their assets. Thus some banks will obtain funds from these sources while
other banks might not be able to, even at a higher cost of borrowing. These facts form the
basis of market frictions, which are the root of the existence of the bank-lending channel.

The presence of frictions in the external market for funds imposes costs on banks.
In addition to the marginal cost of borrowing there is also an external finance premium,
which reflects the severity of the constraints that each bank faces and consequently the
lenders’ assessment of the default risk. The effects of frictions will tend to intensify during
periods of tight money. The rise in interest rates that accompanies a policy tightening affects
the profitability of funded projects. As a result, the likelihood of default increases, adversely
affecting the asset quality for banks and possibly making moral hazard more pervasive. Thus,

information asymmetries are likely to worsen and we can therefore infer that the external

105

finance premium is higher during these periods as well. Hence, a larger proportion of banks
will not be able to borrow from the external market for funds.

KSW, 1993; Nilsen, 1999; and, Bernanke and Gertler, 1995, show that different
classes of borrowers are subject to differing degrees of information asymmetries. In
particular, small borrowers tend to be associated with more severe information problems,
making it harder for them to obtain alternative funding to maintain their lending activity.
Thus, we do the analysis controlling for size.

Banks keep stocks of securities as a buffer against periods of low levels of deposits
and unpredictable withdrawals. Banks also need to guard against other factors that can affect
the flow of deposits in and out of the bank, particularly changes in monetary policy that
would induce reductions in deposits. In this respect, banks are assumed to invest in securities
mainly for precautionary motives. Thus if this were the only motive for holding securities,
for all banks, investment in liquid assets would be proportional to the level of deposits. This
would be the outcome if there were no constraints to obtaining alternative funding and
banks had equal access to non-deposit finance. However, the amount of liquid assets held
depends on the ability of banks to issue managed liabilities when they experience a shortage
of liquidity. A bank that can easily obtain funds from alternative sources will tend to hold
less liquid assets than a bank with limited access to these funds. As stated earlier, some
classes of borrowers have limited access to capital market financing. In particular large banks
have access to uninsured debt instruments and thus they tend to hold relatively low levels of
liquid assets whereas small banks seem to encounter some difficulties and will therefore tend
to hold a larger proportion of their assets in liquid form. Furthermore, it has been shown
that banks with different characteristics, size and, balance sheet and financial strength,

exhibit different behavior with respect to the amount of liquid assets they hold. Table 1b

106

supports this View as the ratio of assets held in the form of liquid assets declines with the
size of the bank. In particular, the level of liquid assets held by small banks is strongly
influenced by the relative strength of their balance sheets, such that small banks with strong
financial positions hold a larger stock of securities relative to their weaker counterparts (See
Lucas and MacDonald, 1992).

The existence of imperfections makes investment in liquid assets a function of
factors other than the level of deposits and expectations of withdrawals. These include
factors such as the ability of a bank to quickly issue non-deposit debt instruments. Kashyap
and Stein (1993, 2000), Worms (1999), Stein (1995), Alger and Alger (1998), Kashyap, Rajan
and Stein (2001), among others, have shown that because large banks have relatively easier
access to non—deposit instruments they tend to keep a low level of liquid assets, relative to
their deposit base. This is because large banks, like large firms face less severe information
problems and therefore their borrowing from external markets attracts low agency costs.
Small banks, on the other hand, seem to encounter some difficulties obtaining funding on
the financial or inter-bank markets. This implies that in general small banks will hold a
higher proportion of liquid assets, but there is a further distinction that needs to be made.
Investing in liquid assets represents a cost to banks because they pay a low rate of return.
Therefore, although prudence requires banks to hold liquid assets as a buffer against deposit
shortfalls, for banks with weak balance sheets, the opportunity cost of investing in these
assets may be too high. This follows from the fact that in general, weak banks only have a
positive net present value if they get the highest return possible on a unit of investment.
Consequently, these banks tend to hold a much lower level of securities. Thus depending on
the financial conditions of banks, there should be a difference in the level of securities held.

In addition, small banks are associated with more severe information problems and if they

107

are perceived to have private information about the quality of their assets, their ability to
raise external finance will be affected. This is because, unlike deposits, external finance is not
insured and as such is sensitive to information about the asset quality of a bank. This is in
direct contrast to the argument by Romer and Romer (1990) that all banks can costlessly

issue large denomination CDs to counter low levels of deposits induced by contractionary

o

‘—

monetary policy.3

The intuition for this expectation is that weak banks need a high return for each unit
of investment they make. This can only be realized by investing in illiquid loans because they
offer the highest return. Strong banks can afford to hold some level of low yielding securities
with the intention of drawing on them when there is a policy change that reduces their
deposit base. As a result, we should also see a different response in lending activities within
the same class of banks when there is a monetary policy induced reduction in deposits. Weak
banks should, thus, reduce their lending activity more sharply.

From the above framework, we can empirically test the following hypotheses for the

existence of the lending channel:

Hypotberir 7: Given a level of deposits, within the same size group, financially sound banks
should hold higher levels of securities relative to banks with a weak financial performance.
Therefore, when there is a monetary policy induced deposit shortfall, banks with a weak
financial condition will experience sharper declines in lending activity than sound banks.

Further, given that large banks face fewer constraints, the effect should be more pronounced

for small banks.

 

’2 Bernanke and Gertler (1995) found that interest rates on CDs increased significantly more than T-bill
rates during periods of monetary contraction, providing further evidence against the Romer and Romer’s
argument that demand for managed liabilities was perfectly elastic.

108

Hypotberir 2: Given a level of financial health, sound banks should be able to raise larger
amounts of external funding during periods of tight money. Furthermore, large banks should
be able to solve their funding problem by securitizing some of their loans. Thus, sound

banks should not have to reduce their lending drastically.

H fiat/78513 3: When faced with a reduction in insured deposits, banks that are able to will
partially offset the effects of low deposits by liquidating their securities. Thus we should

observe a reduction of securities during periods of tight money.

3 DATA

The empirical work in this paper uses quarterly data from the Call Report of Income
and Condition that provides bank level data on all insured commercial banks. This data set is
made available by the Federal Reserve Bank of Chicago and covers the period 1976 to 2000.
Over this period the number of banks declined from over 15,000 in 1976 to just over 9,000
by the end of 2000. There are several reasons for this decline with the main ones being
consolidations and failures. Each of these factors has potentially interesting implications and
their importance to the analysis depends on whether these disappearances are systematic.
This attrition, as well as generally missing data, has presented some challenges in how to
handle the estimation. We test of the importance of this sample attrition. The data set
provides information on several components of bank balance sheets. Because of changing
reporting standards over the years, the way the variables of interest are reported has changed
considerably. Hence, we have had to rely on the notes provided by Kashyap and Stein to
form consistent time series.

Following on the influential paper by Bernanke and Blinder (1988), we use the

federal funds rate (FFR) as an indicator of the stance of monetary policy. The data are

109

provided on a monthly basis and since our main dataset is quarterly, we use the last month
of each quarter. In this paper, we have argued that the lending channel operates through
changes in the level of insured deposits as a result of contractionary policy. Thus, to test the
appropriateness of the FFR as a measure of policy stance, we have conducted a simple test
that shows a statistically significant negative correlation of 0.39 between the federal funds
rate and deposits. Because we have also argued that banks rely mostly on insured deposits
for their lending activities since these are not sensitive to information asymmetries, we need
to find a reasonable measure of insured deposits. We have adopted the method used by
Ashcraft (1999) and use core deposits for a measure of insured deposits. This variable is

constructed as the difference between total. deposits and large time deposits.

3.1 Measures of Financial Soundness

The main thrust of this paper is that overall financial soundness of a bank, and not
only size, may be important for the bank’s ability to shield its loan supply from disruptions
induced by policy innovations. As a proxy for financial soundness, we use a measure based
on the CAMEL ratings. These ratings, which focus on six components of bank safety and
soundness, namely, capitalization, asset quality, management, earning capability, liquidity and
sensitivity to market risk, are a universally accepted measure of bank financial health and
should therefore, be appropriate for our analysis. Although the ratings are normally used as
an index giving a score between 1 and 5, in this work, we compute the relevant ratios using
data on both the balance sheets and the income statements and enter them separately in the
regression equations. Banks with low ratios are considered unsound or weak.

Table 1b presents a summary of the ratios for each size category. From this summary
certain characteristics are evident. The data show small banks to be better capitalized than

large banks, as the mean is highest for the smallest banks and lowest for the largest banks.

110

For the capital ratio (equity), we have a mean of .097 for banks in the smallest size category
and this ratio declines to .078 for the largest banks. This picture is not surprising since
capitalization of a bank reflects its ability to absorb losses and large banks generally remove
some of the risky assets from their balance sheets through loan sales which effectively
reduces the capital they have to maintain.

The liquidity component of the CAMEL measures the bank’s ability to fund loan
commitments or withdrawal demands at a reasonable cost. Therefore, a higher ratio indicates
a greater ability to meet unexpected liquidity needs. However, banks’ ability to meet funding
needs also depends on their reliance on non-core funding. Large banks can easily raise non-
core funding and thus do not have to maintain a large stock of liquid assets. The data
supports this view in that the ratio gets smaller as the size category gets larger. For the banks
in the 75th percentile, the ratio is .347 and banks above the 95th percentile have a ratio of
.166.

The earnings capability is measured as the ratio of net income to total assets (ROA).
We are interested in this ratio because earnings are an important source of capital for banks
and as such provide an additional cushion for withstanding negative shocks. Thus a bank is
better off the higher this ratio is. Like the capital ratio, for banks in the small size group, this
ratio is substantially higher at .036 than it is for the three larger categories, for which the
ratio is about .007.

Authors seem to use different measures for asset quality. Some authors use the ratio
of loan loss reserve to non—performing loans, which is Qua/2'97 in the table, (e.g. Shen et al.,
1999) whereas others use the ratio of non—performing loans to total loans, Qua/1'92 (e.g.
Siems, 2002). We have used both ratios because we believe both to be important as they

provide complimentary information, and separately they may not give a complete picture of

111

the bank’s performance in the criteria. The ratio of loan loss reserves to non—performing
loans provides an indication of a bank’s ability to absorb losses from bad loans. \While this is
important in its own right, it does not tell us anything about the extent to which banks’
portfolios are impaired and we get this information from ,eralig’Z. We have also included a
third measure, which we call Qua/2'91}. This is constructed as the ratio of non-performing
loans to total assets. This ratio gives an indication of balance sheet impairment instead of
restricting the measure to loan portfolios. All ratios increase with size, which may not be
surprising since large banks can limit their exposure to bad loans through securitization. At
the same time, large banks may also have more financial resources enabling them to provide
adequately for non-performing loans.

The data suggest that management quality is higher at the larger banks. To measure
management capability we use the banks’ relative productive efficiency, calculated as the
ratio of non-interest expense to total assets as a proxy. This suggests that the banks with a
better ability to control their expenses will generally fair better than the others. A bank is
more efficient when this ratio is low and from the summary we see that the ratio is higher
for banks in the smallest size group indicating that management capability improves with

size.

3.2 Other Descriptive Statistics

Table 1a presents some key statistics on the structure of the U.S. banking sector.
From this table, it is evident that an overwhelming majority of banks (over 70%) are
considered small with median assets of $44,000. On the other extreme, the largest banks,
those whose size is above the 95th percentile of total assets, compose about 7% of the total
number of banks. However, even though they are few of them, on average, these banks

control around 80% of the total assets.

112

This trend is also evident in the lending aspect. Despite their comparatively small
number, on average, these banks account for about 85% of total lending while the banks in
the smallest size group account for only 6% of the total loans despite their multitude. Also,
small banks are the most soundly capitalized but have the lowest growth rate of either total
loans or commercial and industrial loans while the largest banks have the highest loan
growth rate.

On the liability side, the figures also support the stylized facts; mainly large banks are
net borrowers in the federal funds market. The statistics show that the amount of federal
funds issued increase with bank size with the smallest banks issuing only 1% of their total
liabilities while the largest banks have an average of 9%. The banks in the smallest size
groups on average have the highest level of insured deposits with the ratio smallest for the
largest banks while for large deposits we have the largest banks holding the highest level of

these liabilities.

3.3 Potential Problems

3.3.1 Sample Attrition: Tests and Correction

The structure of the U.S. banking system has undergone some major changes over
the last three decades. Most notable of these changes is the decline in the number of banks
through bank failures and consolidations. For the sample period we are using, the number of
banks declined from 13,418 in 1991 to 9,261 by the end of 2000. This reduction represents a
significant amount of sample attrition and is thus potentially problematic for inference about

- 33
parameters of interest.

 

’3 In our case the attrition is an absorbing state, i.e. once a bank drops out of the sample it stays out
permanently.

113

The main concern with the attrition is selection bias, which would result if attrition
were non-random. Thus, we are concerned with whether the banks that are in the sample for
the entire period differ in any significant manner from those that attrite. If they do, then we
have a selection on observables and / or selection on unobservables, which may generate a
selection bias and consequently a distortion of the estimation. Suppose we are interested in

estimating a linear unobserved effects model of the following form:

yit = xitfl+ci +uit
Because of selection issues we have to assume that (xi t , yi t) is observed for all iat t = land
if Si t is the selection indicator for each time period, then Sit = lif (in , yit ) are observed
and Si t = 0 if (x1. t , yi t) are not observed. We first difference the model to remove the fixed

effects and the equation reduces to

Ayit = Axitfl+Auit t= 2,....T

The presence of attrition implies that this equation can only be estimated for banks that are

in the sample in period t , or only when Sit = 1 and yi t is observed.

From this model, suppose we are interested in the conditional density of

f ( y t I x I) but we observe g( y t | x t’Si t = 1). We need some restrictions or additional

information that will allow us to use the data we observe so that we can

infer f C“) from g(*). Fitzgerald et a1 (1998) suggest obtaining these restrictions from the

probability function of attrition, pit (Sit = l | yit ,in ’Zit) , where xit is the matrix of our

main regressors and 21. t is the matrix of variables that are important in predicting selection

114

and are not included in xi t . This could include lags of both the dependent and independent

variables.
We can state the attrition as follows to determine if the selection we have is

potentially problematic:

a:

s. =a +a x. +0: 2. +12.
1t 0 lzt 21t 1

t
1'f * < o
s. = 1 s. _
It It
. at:
S. = 01f s. > 0
It It
* I I I I v I I v
Here Si t 18 a latent index and attrition occurs if this variable equals zero and vitls a zero-

mean random influence on the probability of attrition. Our first question is whether the

attrition exhibits selection on observables and/ or on unobservables.

We have selection on observables lfpit(sit =1 l yit’xit’zit) = pit(sit =1 l Zit)’
t =1,....T . Thus conditional on yit and zit , the attrition probability is independent of yl. t

and of the unobserved factors in the error term of our primary model of estimation.

Sl tinon nbservbls xitsif. .= . . . ¢ . . =12. .Inth
eec o u o a e e s p1t(slt llyzt’xzt’zlt) p"(sn, I It) e
parametric model, selection on unobservables occurs if vi t is not independent of the error

term conditional onx. If there is selection, then our next concern is the ignorability

property. If selection is ignorable, the estimation of our parametric linear model leads to

unbiased estimates of fl,. Selection would be ignorable if yit and 21. t were independent
conditional on xit and sit = land we can reduce the probability function

to pl. t (Si t = I I xit ’zit) = pr(sit = I | X]. t) that ts the probability of attrition 18 independent

115

of zit . If these conditions are met, then the estimation of the linear equation yields

consistent estimates of the coefficients. However, if these conditions do not hold, then
selection is non-ignorable and we need to account for the attrition in the analysis.
To identify a model with selection on unobservables, the usual approach is to use

some exclusion restrictions, which requires the use of some variable, q that could predict

attrition but was independent of u t | x t , as well as being included in xi t . The consensus

appears to be that it is extremely difficult to find such a variable, because there are few
variables that can both predict non-response and be convincingly excluded from the main

equation for y . However, while selection on unobservables might remain a problem,

correcting for selection on observables by using all the available information will reduce the
amount of residuals and unexplained variation due to non-response. Thus controlling for
selection on observables will also reduce biases due to selection on unobservables.”

With respect to selection on observables, as an initial analysis, we have tested the
means of our important variables for the attritors (those banks that drop out of the sample
at some time) and those banks that stay in the sample throughout the entire period, which
we refer to as the stayers. We find that for most variables the means are significantly
different indicating that we might have selection on observables (see table 3-2) ’5. The issue is
whether the attrition will yield a significant selection bias of the coefficients, i.e. would the
attrition yield a significant selection bias in estimates?

With respect to selection on observables, as an initial analysis, we have tested the

means of our important variables for the attritors (those banks that drop out of the sample

 

3" See Alderman, Behrman, Kohler, Maluccio and Watkins (200]).

3’ The difference in means could also be attributable to a host of other factors that are not related to
attrition, e. g. changes in operation procedures over the intervening period could result in means of the
variables being significantly different.

116

at some time) and those banks that stay in the sample throughout the entire period, which
we refer to as the stayers. We find that for most variables the means are significantly
different indicating that we might have selection on observables (see table 3-2)”. The issue is
whether the attrition will significantly bias the coefficient estimates.

There are a number of tests that are commonly used to determine the importance of
attrition. None of these are perfect tests but two approaches appear to be more suitable for
our purposes given the structure of our dataset. One is the test that entails comparison of
the weighted and ZSLS regressions. Similarity of the estimates suggests that attrition may not
impose a significant bias on the estimate. The other test we use is suggested by Wooldridge
(2001) and involves including a lead of the selection indicator in a simple ZSLS regression. If
the coefficient on this variable is not significant, then we can still obtain reasonable inference
from the estimates. Thus estimation of our structural equation based on the observed data

for non-attritors will yield consistent coefficients of the ,6 . The results of these tests are

presented in table 3-4 to 3-6 and suggest that attrition will not largely bias our estimates.
With the caveat that the tests are not error—proof, the results suggest that it is valid to use the
usual estimation procedures with either an unbalanced panel or a balanced sub-panel for

estimation.

3.3.2 Endogeneity

Most of the right-hand-side variables of interest, especially the ratios that measure
bank financial soundness are derived from the banks’ balance sheets and the income
statements. Generally, these variables are not determined outside the model and as such we

have to deal with the potential estimation problems presented by endogeneity.

 

3’ The difference in means could also be attributable to a host of other factors that are not related to
attrition, e. g. changes in operation procedures over the intervening period could result in means of the
variables being significantly different.

117

Liquidity: The decision of banks to hold a certain level of liquid assets is a function of a
number of factors including market frictions. The stock of securities banks hold is a function
of information asymmetries and whether the core borrowers of a bank are prone to cyclical
fluctuations may also influence the choice of liquid assets that a bank holds. Finally, we
cannot exclude the possibility that holding large stocks of liquid assets may be a reflection of
a relatively intense risk aversion on the part of the bank. If this is the case, then the banks in
questions are likely to have more stringent lending conditions. Otherwise, the possibility that
there were differences in loan demand between risky and less risky firms in response to a
monetary policy shock would make liquidity a poor measure for identifying loan supply
effects.

Large deposits could also be endogenous. Suppose that lenders are sensitive to the financial
health of the banks they deposit their funds with. One possible outcome of tight monetary
policy is that the soundness of some banks may be compromised. Holders of uninsured
deposits may want to withdraw their funds from these banks. At the same time, because
contractionary policy reduces core deposits, banks’ demand for large deposits will be
increasing. Consequently, banks may try to induce inflow of these deposits by offering
higher interest rates. Thus we have a possible joint determination with respect to large
deposits.

Return-on-Assets: Banks’ ability to lend is influenced by its ability to generate income but
the lending a bank does also determines how much income the bank is able to generate.
Thus the ROA is also potentially endogenous.

Asset quality (measured as the ratio of loan loss reserve to non-performing loans) is

another suspect variable in that it could be a reflection of a bank with a solid balance sheet

118

and an ability to absorb problem loans. However, a high ratio could also hide the fact that
the bank had a risky portfolio and was anticipating a relatively large incidence of default.
Equity: High capitalization is usually associated with banks being less prone to asymmetric
information problems vis-a-vis their lenders. Thus a high capital ratio should entail a lower
external finance premium relative to that of poorly capitalized banks. However, bank
capitalization may reflect the riskiness of its loan portfolio if banks held high levels of capital

because they were more risky.

3.3.3 Correlation between fixed efi’ects and explanatory variables

A related problem arises when we have bank specific fixed effects that are correlated
with some or all of the RHS variables.

A bank’s attitude towards risk determines the portfolio choice of that bank. Thus
there is a correlation between this fixed effect and the liquidity ratio. Other than portfolio
choice, this effect will also influence the banks’ actions with respect to other operational
aspects such as providing for loan losses and earning capacity.

Geographical location is another time invariant effect that is correlated with some
explanatory variables. For example, most banks, especially small ones, tend to limit their
operations to the states or cities where they are located. Banks with geographically
concentrated activities are potentially vulnerable to local economic conditions because of an

inherent concentration of loan and deposit customers.

4 Estimation and Results

4.1 Estimation

119

The strate r for the em irical anal rsis is to test for a differential res onse of bank
g) P i P
lending to changes in monetary policy across banks, depending on their financial position.

For analysis, we use panel data techniques to estimate an equation of the form:

_ 4 -
yit=ayit—1+Z fijflrt—j +7in +¢Zir
F4 (1)

 

 

4
+ Z lajffrt—jzit +(ogt +77, “‘1?
J:—

— d

where yit represents real loans, X it is a matrix of bank specific variables that capture

determinants of loan supply shifts that are not caused by monetary policy changes; Zit is a

matrix of the components of the CAMEL ratios, which are allowed to vary across banks and

time; ”i represents a bank’s unobserved time-invariant characteristics and uit is the

idios ncratic error term. With the exce tion of the interest rates, we have deflated all
Y P

variables by the CPI. The variable gt (real GDP) is a proxy for the overall health of the

economy. We estimate this equation with real loans as the dependent variable. We want to
determine whether policy changes affect lending activities for banks.
We use first differencing to remove the fixed effects, thus we are effectively

estimating an equation of the following form:

1

F

4
Ayir =aAyit_j +j§lfljAfi9t_j +71:in +¢AZit
(2)

 

 

4
+216jflrI—jzit+mgt+Auit
_ l=- _

Since banks of different sizes appear to react differently to shocks, our estimation also takes

size effects into consideration. To do this we have divided banks into four size groups based

120

on total assets. We have banks with assets below the 75Lh percentile, the 90‘h percentile, the
95‘h percentile and above the 95’h percentile. \X’e estimate the regression equation separately

for each size group. Our monetary policy indicator is the federal funds rate, thus we test the

hypothesis that fl]. < 0 as well as 20]. < 0 . Since we are also concerned with whether

financial strength affects the way banks respond to policy changes we are also interested in

20}. for each ratio as well as the overall significance.

4.2 Estimation Methodology

We estimate the equation using the Generalized Methods of Moments (GMM) for
dynamic panel data models; specifically we use the Arellano-Bond (1991) approach. This
approach is appropriate for several reasons. It is possible that the choice of current period
loans is influenced by past values. Thus we would want to allow for the dynamic nature of
the model by including lags of the dependent variable and for this purpose, most standard
GMM estimators would suffice. We also want to control for possible simultaneity and
reverse causality of the explanatory variables and the dependent variable. This we can
achieve by relaxing the assumption of strict exogeneity of the explanatory variables and
instead assuming weak exogeneity (where the error term can be correlated with the leads but
not the lags) and then using instrumental variables. This approach would also provide
consistent estimators for endogenous independent variables. Because we do not have
outside instruments, we estimate the model using lagged independent variables as
instruments and we show the conditions under which these are valid instruments.

However, we still need to account for the existence of unobserved individual effects
resulting from substantial differences across banks. Estimating the equation using the

standard GMM alone would not be adequate because the unobserved heterogeneity would

121

lead to inconsistent coefficient estimates. We first difference the equation to eliminate the
fixed effects. However, because we have a relatively short panel, this technique may not
work very well and would still yield inconsistent estimators.
Consider our model in this simplified form:
4

.=..+. .+'x.+.+u. 3

yzt aylt—j ’6].Zjfit—j ylt 77' (I <)
J=-1

where

E(uit Ixi0""'xiT ,0!) = 0

After first differencing the equation, we get

4
Ayit = aAyit __ j + ,6]. 1'21 Afli‘t _ j + yAxit + Aui t (4)

The differencing also induces a new source of correlation between the differenced error

term and the differenced lagged dependent variable because by construction, Ayit—t is
correlated with Aui t . These correlations require the use of valid instruments for consistency.
Under the assumption that “it is not serially correlated and that the explanatory variables are
weakly exogenous, we get the following moment conditions for the explanatory variables:
E(yit_SAuit)=0 (5)

E(x A

it—s “it)=0 (6)

When these moment conditions apply then

(in _ 2 ,in _ 3 ,....y1 ) and (in _ 2 ,in _ 3,....x1) are valid instruments. Arellano and Bond

(1991) have shown that the estimator from the above model would be consistent and

efficient. We use the Sargan test for over-identifying restrictions to check the validity of the

122

instruments used, and the presence of second—order serial correlation of the error term. The

second test is necessary because the consisrency of the estimators relies on the condition

that E[Aui [Aui t _ 2] = 0, i.e. we should not have second order serial correlation in the

errors. Violation of this condition invalidates the instruments as they would be correlated

with the error term.

4.3 Empirical Results

Our goal is to test the hypothesis that the response of bank loan supply to monetary
policy shocks is a function of banks’ financial strength and that it could also depend on the
size of the bank. For this purpose, we need to use a specification that relies on interactions
between monetary policy shocks and measures of bank strength. The stratification of the
sample by asset levels takes care of the size effect. However, direct relationships that might
exist between bank strength, size and lending activity should provide useful initial insights
into the importance of these characterizations of banks for the transmission of monetary
policy. We start with these and discuss their importance briefly without much detail besides

whether or not they matter for lending.

a) Bank lending sensitivity to bank financial strength

In tables 3-4 to 3-9 we report the results of the regressions for the impact of
monetary policy shocks given bank strength separately for each of the four size groups
These results for all panels are considerably different indicating that the changing structure
of the banking industry may have some important implications for banks’ response to
shocks.

For the variable of interest, the federal funds rate, we included one lead and four lags

and we are interested in the cumulative effect. Because a lot of decisions concerning future

123

lending are made well in advance of changes in monetary policy, loans react to policy shocks
with a considerable lag, (Bernanke & Blinder, 1992). However, since 1994 the Fed has been
announcing the target level for the federal funds rate. Rationally, banks should react to the
announcement and start adjusting their portfolios appropriately. Thus, we need to allow for
some announcement effect, hence the inclusion of the lead in the regression.

Looking only at the federal funds rate, it appears that size is the only factor that
matters in how monetary contractions affect bank lending. The cumulative effect is negative
and significant for the banks in the 75'’1 and the 90th percentile only. For the two larger size
categories, none of the coefficients are statistically significant.

We have, however, made a case for the possibility that the overall financial strength
of banks might play a role in mitigating the effects of monetary contractions on their lending
and we have appealed to components of CAMEL ratings for heterogeneity or differences in
financial strength. Thus for a basic understanding of how these variables could potentially
help banks withstand policy-induced deposit falls, we should analyze their impact on bank
lending. These are presented in Table 3-7. In addition we have looked at how monetary
policy changes affect some of these ratios, especially the liquidity to determine whether
indeed some banks fall back on their security holdings to maintain the levels of their lending.
The results of the impact of shocks on net liquid assets show a coefficient esfimate that is
negative and strongly significant for all banks except those in the largest size group. This
result implies that in periods of tight monetary policy, lending activity will be more restrictive
for less liquid banks. Banks with adequate liquidity will offset, at least partially, the effects of
policy-induced deposit shortfalls by selling their liquid assets. Thus we should observe a
reduction in liquid assets when there is a monetary tightening. The lack of a significant

response on the part of the largest banks supports the hypothesis that because of access to

124

alternative sources of funding, these banks do not need to maintain a high liquidity base to
meet funding needs. Thus while liquid assets are important in helping banks in lower asset
groups to smooth their lending they are not so important in the highest asset group.

For the influence of ratios on lending in the level estimations, it appears that the
importance of individual ratios to bank lending activity varies by size. However, most ratios
are significant for all size groups except the largest banks.

The estimates for the ROA ratio indicate that this measure of financial strength is an
important determinant of lending activities for banks in the smallest and largest size groups.
For the small banks the coefficient is positive and significant. If monetary policy affects bank
loan supply and if this ratio is important to the banks’ ability to maintain its lending, then
this ratio being positive would support the prediction that lending at strong banks would
decline by a lesser proportion. However, for the largest banks the coefficient estimate is
negative and significant. There might be a simple explanation for this outcome. The negative
coefficient could be capturing the fact that large banks do not rely that much on traditional
loan activities for income generation. They have access to an extensive base of non-lending
activities and fee-based operations, which have become very important sources of income
for large banks. Thus, during periods of tight money when firms would not be very
profitable, large banks may choose to substitute out of the traditional loan market into other
income generating activities that do not entail as much risk.

Our first measure of asset quality is the ratio of loan loss provision to non-
perforrning loans. This ratio can be interpreted in two ways. First, the high ratio can be
thought to indicate prudent behavior on the part of banks, which could then be taken to
imply some positive correlation with financial strength. If this view holds, a high level of

provisions relative to bad loans should be reflected in a high level of lending or a smaller

125

decline during periods of tight money. Thus the estimates should be positive to show banks’
ability to weather effects of restrictive policy actions. Shen (1999) argues that in fact an
increase in the ratio or a high level of provisions could actually indicate that the bank was
expecting a higher proportion of loans to default. If there is an expectation of an increase in
non-performing loans, then we should observe larger declines in lending activity. Our results
would seem to support this latter argument, as the coefficient estimates are negative for all
but the largest banks.

The second measure of asset quality is the ratio of non-performing loans to total
loans, which measures the extent of portfolio impairment. One of our hypotheses is that
during periods of tight money, sound banks have a better ability of shielding their loan
supply through their ability to raise external finance. It is reasonable to expect that the
markets will consider the banks’ asset quality as an indicator of default possibility. If this is
true, then banks with better loan portfolios should not have to drastically reduce their
lending following a monetary contraction and it is also likely that those banks with poor loan
portfolios may encounter more difficulties when attempting to borrow from external
markets. Our findings appear to support this argument in that in all cases, the coefficients
are significant and positive indicating that asset quality is important in how banks react to
shocks. Further, these results indicate that banks with high asset quality portfolios are able to
smooth out the effects of policy shocks. The coefficients are significant and positive in all
cases.

In the next stage, we introduce interaction terms to the specification described
above. We still want to address the question of whether the behavior of bank lending is
influenced by differences in banks’ financial positions. Interacting the monetary policy

indicator with the measures of financial strength should provide some additional information

126

on how individual characteristics of banks influence their response to policy shocks.
Specifically, we have interacted each CAMEL ratio with the lead and four lags of the federal
funds rate and we estimate the regressions separately for each size group. Our discussion of
the results is based on the cumulative effects of all the lags.

For statistical significance of the explanatory variables, we use the joint significance
tests presented at the bottom of table 3-8. Note however, that because of the substantial
difference in the number of observations across the different bank sizes, we cannot make
valid comparisons on the statistical importance of the right hand side variables. Thus we
restrict analysis of statistical significance within each size group. In addition to statistical
significance, economic importance measured by the size of the effect is also important. For
this purpose, we use standardized coefficients and since for each size group we are
effectively moving the variable over the same distance regardless of the sample size, we can
determine the relative importance of variables across different size groups.

Once again we see a vast difference in the importance of financial soundness for
banks in different size groups. The results are suggestive of financial ratios playing a role in
how bank lending reacts to monetary contractions. The results show that all variables are
jointly significant for all bank sizes expect for those in the 95‘h percentile group. However,
the cumulative impact of the individual measures of bank strength varies significantly across
the different size categories. It appears that banks in different size groups respond differently
to different ratios, i.e. a ratio that is significant to a bank in the 90th percentile may not be
significant for banks in the 95th percentile and vice versa. This would suggest that perhaps
factors that determine financial strength for large banks are different from those factors that
determine financial health for small banks, possibly because of different approaches to

generating income.

127

Again there is variability in how banks in different size groups react to policy shocks
and in the importance of individual variables process. The liquidity term has a cumulative
effect that is significant for all size groups However, for the large banks, this effect is only
significant at the 10% level. Perhaps more interesting is the economic importance of these
effects. The magnitude of the impacts clearly reflects the importance of bank size, especially
that of large banks, for economic activity. The results show that an increase of one standard
deviation in net liquid assets reduces lending activity by 0.8 percentage points and 2
percentage points for banks in the 75'h and 90th percentile groups while the impact is over 20
percentage points for the large banks with the largest banks having more than twice as large
an impact.

Also, the effect is positive for the largest banks indicating that these banks actually
increase the volume of lending following a monetary policy contraction. This could be in
support of the flight-to-quality argument whereby financial assets gravitate towards large
institutions during periods of economic downturns.

Notably, the interaction term for capital is not significant for any of the groups.
Does this imply that bank capitalization is not important in determining bank financial
soundness? Obviously not since capital requirements are one of the major ratios that central
banks consider when examining bank soundness. The results we obtain could be reflecting
the use of total book-value capital that is not risk-weighted. Unfortunately, data on risk-
adjusted capital, which would, perhaps, be more informative on how markets respond to
bank capitalization, are not available. Notwithstanding the size of the cumulative effects
gives reason to pause and reconsider the importance of the implied influence. The impact of
a unit increase in the standard deviation of equity reduces lending for all but the largest

banks -- by from 1 percentage point for the smallest banks to 28 percentage points for banks

128

in the 95‘h percentile group. This more than likely, reflects the fact that when banks increase
their capital holdings, for whatever reason, they are effectively reducing their base for
alternative investments. However, for banks in the largest size category, the impact is
positive and relative to the other sizes, very large. The positive impact could be a reflection
of the ability of these banks to resell loans thus reducing the need for the capital base from
their balance sheets.

The return on assets measure is statistically significant only for the largest banks
again with a large disparity in the economic significance across the sizes. For example,
increasing the variable by one standard deviation increases lending of the two smallest
groups by about $0.02 compared to $0.36 increase in the lending of the largest banks.

The other variables that are important for banks’ ability to maintain their lending
activity following a shock are efficiency, which matters only for the smallest and largest
banks, the ability of banks to absorb losses from bad loans (qua/1'94), which is important for
banks of all sizes and the measure of the impairment of portfolios (qua/1'92), which is
significant for that smallest and largest banks.

Again the impact is suggestive of bank size determined implications for economic
activity. In particular, the magnitudes of the cumulative effects show that for efficiency, a
unit increase in the standard deviation, increases lending by roughly $1.50 for the banks in
the two smallest size groups relative to about $5.00 for banks in the upper percentile
categories.

The importance of asset quality across all groups suggests firstly that markets are
sensitive not only to banks’ ability to absorb losses from bad loans and therefore remain

solvent. Second, relative to the amount of total loans, the impact of improving asset quality

129

is larger for small banks which could reflect the possibility that they are associated with

higher perceived default risk compared to large banks,.

b) Does bank’s financial position matter for its ability to raise external

finance?

We have argued in the main body of the paper that part of the reason why banks
have different abilities to smooth out the effects of contractionary shocks emanates from
having differential access to markets for external finance depending on their individual
characteristics. We have hypothesized that financially sound banks should be able to obtain
larger amounts of external funding. We test this hypothesis using the same estimation
technique as we did for loan response and the same ratios but with the federal funds
purchased as a dependent variable.

Since smaller banks are net suppliers of federal funds and we are interested in ability
to borrow from the market, we limit the estimations to large banks. The results show that
only the return on assets (ROA) has a statistically significant coefficient. The second measure
of asset quality and the liquidity ratios are both marginally significant at the 10% level.37
jointly, all the measures of soundness are significant at the 5% level. Thus, the results
somewhat support the hypothesis that bank financial strength improves the ability of banks

to issue federal funds. However, most of this process could be driven by size since the data

show that large banks are the only effective borrowers in this market.

5 Conclusion

 

37 These results have to be interpreted with caution, as the model is potentially misspecified. Ideally, this
regression should be run using a tobit model since our dependent variable has a significant pile-up at zero
(26%) and a continuous distribution to the right. However, most of our variables of interest are potentially
endogenous. We do not have outside instruments and ivtobit cannot be run with lags. Thus we have had to
sacrifice specification for consistency.

130

In this paper we have investigated the importance of banks’ financial position in
their ability to smooth the effects of monetary contractions on their lending activity. We
specifically analyzed ways in which imperfections in financial markets can help propagate
and magnify the effects of shocks.

Using data on commercial banks in the U.S. we have tested for the importance of
banks’ financial position in mitigating the adverse effects of policy tightening. Because of the
existence of financial frictions that make it difficult for some firms and banks to obtain
alternative funding, this is an appropriate way of testing for differential responses to
monetary policy. We have found that, indeed banks’ financial positions matter in how banks
respond to policy actions, although we find that the response is not uniform among banks of
different sizes. Furthermore, unlike the existing literature that finds no loan supply effects
for large banks, we find that some measures of bank strength are important in banks’
reaction to monetary contractions. We also find that banks may rely to some extent on their
liquid assets to partially offset reduction of their lending activity. The financial frictions and
their asymmetric nature can potentially introduce uncertainty into the size and timing of the
response of the economy to monetary policy innovations. Thus, it is important that policy
makers understand these and should incorporate the assumption of credit constraints in

policy design.

131

REFERENCES

Ahn, S.C. and HR. Moon (2001), “Large-N and Large-T Properties of Panel Data
Estimators and the Hausman Test”, Working Paper

Alger, G. and I. Alger (1999), “Liquid Assets in Banks: Theory and Evidence”, Unpublished
Working Paper.

Alvarez, J. and M. Arellano (1998), “The Time Series and Cross-Section Asymptotics of
Dynamic Panel Data Estimators”, CEMFI Working Paper No. 9808

Arellano and Bover (1995), “Another Look at the Instrumental Variables Estimation of
Error Components Models”, journal of Econometrics, vol. 68 pp. 29 — 51.

Arellano, M and S. Bond (1991), “Some Tests of Specification for Panel Data: Monte Carlo
Evidence and an Application to Employment Equations”, Tbe Review of Economic
Studies, vol. 58, pp. 277 - 297

Ashcarft, A.B. “New Evidence on the Lending Channel”, Federal Reserve Bank of New
York, Sept. 2001

Balgati, H. Badi (2001), “Econometric Anabrrir of Panel Datcf ’, John Wiley & Sons

Benston, G. and G. Kaufman (1988), “Risk and Solvency Regulation of Depository
Institutions”, Monograph Series in Finance and Economic

Bernanke, BS. and AS. Blinder (1988), “Credit, Money, and Aggregate Demand”, American
Economic Review, vol. 78, pp. 435-9

Bernanke, BS. and AS. Blinder (1992), “The Federal Funds Rate and the Channels of
Monetary Transmission”, American Economic Review, vol. 82, pp. 901-21

Bernanke, BS. and M. Gertler (1995), “Inside the Black Box: The Credit Channel of
Monetary Policy Transmission Mechanism”, Toe journal of Economic Penpective

Brissirnis S.N., Kamberoglou, N.C. and G. T. Simigiannis (2001), “Is there a Bank Lending
Channel of Monetary Policy in Greece? Evidence from Bank Level Data”, European
Central Bank, Working Paper No. 104

D’Arista,J (2002) “Rebuilding the Transmission System for Monetary Policy”, Financial
Markets and Society

De Young, R., M. Flannery, W., Lang and S. Sorescu (2001). “The Information Content on
Bank Exam Ratings and Subordinated Debt Prices”, Forthcoming in the journal of
Credit, M any and Banking

Estrella, A (1999), “Securitization and the Efficacy of Monetary Policy”, Economic Policy
Review, Federal Reserve Bank of New York

132

Farinha, L and CR. Marques (2001), “The Bank Lending Channel of Monetary Policy:
Identification and Estimation Using Portuguese Micro Bank Data”, European
Central Bank Working Paper No. 102

Favero, C., Giavazzi, F. and L. Flabi (1999), “The Transmission Mechanism of
Monetary Policy in Europe: Evidence from Banks’ Balance Sheets”, NBER Working Paper
No. 7231

Freixas, X. and J. Rochet (1997), “Microeconomicr of Banking ’, The MIT Press, 1997

Gertler, M. and S. Gilchrist (1994), “Monetary Policy, Business Cycles, and the Behavior of
Small Manufacturing Firms”, Quarterbl journal ofEconomicr, 109(2), pp. 309-340

Gilchrist, S. and E. Zakrajsek (1995), “The Importance of Credit for Macroeconomic
Activity: Identification through Heterogeneity,” in Is Bank Lending Important for
the Transmission of Monetary Policy? Eds. J. Peck and E. Rosenberg

Holmstrom, B. andJ. Tirole (1997), “Financial Intermediation, Loanable Funds and the Real
Sector”, Quarterbl journal of Economics, Vol. 112, pp. 663-691

Jayaratne,J and D. Morgan (2000), “Capital Market Frictions and Deposit Constraints at
Banks”, journal of Credit; Mongi and Banking, Vol. 32, pp. 74-92

Kashyap, A. K., Rajan, R., and J.C. Stein (1999), “Banks as liquidity Providers: An
Explanation for the Coexistence of Lending and Deposit-Taking”, NBER working
paper no. 6962

Kashyap, A.K. and J.C. Stein (1995), “What Do a Million Banks Have to Say About
Monetary Policy?” NBER Working Paper No. 6056,June 1995

Kashyap, A.K. and J.C. Stein (2000), “What Do a Million Observations on Banks Have to
Say About the Transmission of Monetary Policy?” American Economic Review, vol. 90,
No.3, pp. 407-428.

Kashyap, A.K., J. Stein and Wilcox (1993), “Monetary Policy and Credit Conditions:
Evidence from the Composition of External Finance”, American Economic Review, Vol.
83, pp. 78-98

Kendall, L. T. and Fishman M.J, (1996), “A Primer on Securitization“, The MIT Press

Kishan, RF. and T. P. Opiela (2000), “Bank Size, Bank Capital, and the Bank Lending
Channel”, journal ofMongr Credit, and Banking, Vol. 32, No. 1

Nielsen,J (2002), “Trade Credit and the Bank Lending Channel”. journal of Credit, Mong' and
Banking, Vol. 34, pp. 226-253

133

Perasan, M.H. and Y. Shin, (1997), “Estimating Long—run Relationships from Dynamic
Heterogeneous Panels’, journal of Econometricr, vol. 68, pp.79-113

Romer, CD. and DH. Romer (1990), “New Evidence on the Monetary Transmission
Mechanism,” Brookings Papers on Economic Activity, 1990, Issue 1 pp. 149-148

Ronen,J, A. Saunders and A. C. Sondhi (1990), “Off-Balance Sheet Activities”, The Vincent
Ross Institute of Accounting Research

Rudebush, G and SD. Oliner, (1995), “Is there a Bank Lending Channel for Monetary
Policy?” Federal Reserve Bank of San Francisco Working Paper 2

Shen, C and M. Hsien (2003), “Prediction of Bank Failures Using Combined Micro and
Macro Data”, Unpublished Working Paper

Stein,Jeremy (1995), “An Adverse selection model of Bank Asset management and Liability

Management with Implications for the Transmission Mechanism of Monetary
Policy”, NBER Working Paper No. 5217

Van den Heuvel, Skander (1999), “The Bank Capital Channel of Monetary Policy” Wharton
School Working Paper

Wooldridge, Jeffrey M (1995) “Selection Correction for Panel Data Models under
Conditional Mean Independence Assumption”, journal of Econometrics, Vol. 68 pp.
115-132

Wooldridge, Jeffrey M (1999), “Introductoo/ Econometrics: A Modern Approaclr,” South Western
College Publishing, 1999.

Wooldridge, Jeffrey M (2001), “Econometric Anab'rir of Cross Section and Panel Data’ ’, The MIT

Press

Worms, Andreas (2001), “The Reaction of Bank Lending Policy Measures in Germany,”
European Central Bank Working Paper No. 96

134

Table 3-1

Summary Statistics

 

All Banks

Variable
assets
loans
Loans ratio
Commercial
Real estate
household
agriculture
cashrat
liabilities
deposits
core

large

loan growth

Median
Assets

1991q1

Variable
assets
loans
Loans ratio
Commercial
Real estate
household
agriculture
cashrat
liabilities
deposits
core

large
loan growth

All

Mean
474888
250459
0.553
0.173
0.557
0.170
0.105
0.062
4389200
3024848
0.419
0.112
0.879
83779

ALL
Mean
316610
171562
0.527
0.196
0.506
0.195
0.108
0.069
297611.5
2200293
0.379
0.116

75th
Mean
52979
29605.3
0.539
0.168
0.539
0.171
0.130
0.062
47891.1
46076.6
0.424
0.109
0.650
44249

75th
Mean
47492.1
25081.1
0.509
0.190
0.496
0.195
0.129
0.070
43440.4
42092.4
0.383
0.1 12
37890.5

90th
Mean
234026
138774
0.595
0.178
0.640
0.155
0.026
0.051
2139298
1965269
0.408
0.117
5.011
216281

90th
Mean
232290
137201
0.592
0.212
0.576
0.191
0.019
0.053
2156593
2011929
0.357
0.125
2115540

135

95th
Mean
615487
373213
0.610
0.187
0.622
0.170
0.013
0.062
5692699
4944060
0.388
0.124
14.158
575412

95th
Mean
615162
383071
0.625
0.219
0.569
0.186
0.008
0.074
5739387
5064558
0.357
0.142
5753280

>95th
Mean
7.39E+06
4.73E+06
0.613
0.228
0.513
0.207
0.008
0.081
6.87E+06
5.37E+06
0.400
0.136
18.105
2.58E+06

>95th
Mean
5.43E+06
3.64E+06
0.624
0.264
0.468
0.205
0.006
0.098
5.1E+06
4.3E+06
0.375
0.155
2.7E+06

Table 3-1 Continued

2000q4

Variable
assets
loans

Loans ratio
Commercial
Real estate

household
agriculture

cashrat

liabilities

deposits
core

large

loan growth

ALL
Mean
817061
444513
0.611
0.171
0.607
0.135
0.085
0.055
7473086
4949346
0.410
0.156

75th
Mean
60185.7
37410.4
0.597
0.168
0.579
0.141
0.1 13
0.058
53812.7
50954.6
0.433
0.153
53146

90th
Mean
234974
151471
0.645
0.169
0.688

0.1 12
0.029
0.045
2129026
1931101
0.395
0.156
218726

136

95th
Mean
609434
388332
0.642
0.175
0.670
0.123
0.018
0.051
5559563
4739542
0.310
0.163
574229

>95th
Mean
1.16E+07
7.24E+06
0.626
0.212
0.572
0.166
0.009
0.055
1.07E+07
7.78E+06
0.276
0.188
3.E+06

Table 3-2

 

Summary Statistics for Ratios

Percentiles of total asset distribution

 

Ratio 75.1. 90.1. 95.1. >93h
Liquidity 0.356 0.310 0.282 0.213
Net Liquid 0.347 0.285 0.235 0.167
Equity 0.097 0.092 0.090 0.078
ROA 0.036 0.007 0.007 0.007
Qualityl 11.473 13.387 14.147 15.094
Quality2 0.015 0.014 0.017 0.019
Quality3 0.008 0.008 0.009 0.012
Efficiency 0.036 0.021 0.021 0.017

 

137

Appendix 3-1—Definition of Ratios

Net liquid: The ratio of net liquid assets to total assets. Net liquid assets are constructed as
Securities + Federal Funds Sold - Federal Funds Issued.

Liquidigrz This ratio is constructed as (securities + federal funds sold) / total assets. Note that
this measure does not account for borrowing. ROA: Return on Assets calculated as the ratio
of net income to total assets.

Equig: Bank capitalization calculated as the ratio of total book-value capital to total assets.
Efiia'eng: Proxy for bank management constructed as

Qua/1'94: The measure of asset quality is calculated as the ratio of loan loss allowance to non-
performing loans.

Qualig/Z: The second measure of asset quality based on the ratio of non-performing loans to

total loans.

138

Table 3-3
Descriptive Statistics by Attrition

 

 

Original Attritioned
Sample Sample (1)-(2)
-1 -2 p—value

assets 2722666 430126

20331.76 55275.5 0.002
Loans 1146.92 1341.41

105.46 205.94 0.362
Fed Funds 153.99 233.77

13.31 36.37 0.020
Large Deposits 230.12 360.95

15.107 42.542 0.001
Core Deposits. 72.55 33.1

9.144 9.144 0
Total Deposits 16.54 63.5

9.3188 15.7502 0
liquidity 0.3644 0.3181

0.0013 0.0013 0
Net Liquid Assets 0.3542 0.0014

0.0014 0.0014 0
Equity 0.0894 0.0807

0.0006 0.0280 0.710
ROA 0.0015 0.0068

0.0003 0.0006 0
efficiency 0.0090 0.0343

0.0001 0.0009 0
Qualityl 7.7941 9.2638

0.2695 0.6551 0.021
Quality2 0.0205 0.0222

0.0003 0.0002 0
Quality3 0.0110 0.0128

0.0001 0.0001 0

 

Table 3-4

 

Test with SeleCtion Lead
Dependent Variable: Loans

 

Coefficient z-stat

 

Loans_l 0.11 0.83
Loans_2 0.043 0.25
Large Deposits 5.389 1.82*
Netliquid -0.012 2.02**
Equity 69.531 0.8
ROA 27.685 0.93
Efficiency -98.064 1.01
Qualityl 55.062 0.97
Quality2 -53.582 0.42
Quality3 -566 0.72
ffr -60.536 0.18
ffr_l -18.221 1.98**
ffr_2 2.938 0.04
ffr_3 -2.807 0.78
ffr_4 23.624 0.78
Core Deposits 0.311 2.04**
RGDP 4.522 1.02
Real Assets 0.408 2.26**
Spread -1 1.492 0.82
Selection Indicator -20.947 0.07

 

140

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$93.5 $4938 $288.va ipaose 1&3.va 1485.3 3.2338 1:835
wast: 883.. $3.2- madfi- 928.: $32 25.2- $3.:- at
$2838 1.88% 34386 1.58.5 $3- .228an £63.85 1235:
an”? 4893 83.2- mmmnfim- 3.6-3.7 8%?- Nswqfl- 85.8.3- 8.
1.19.33 stages 419:.va thmmtmo 39838 :8me 3333 4:95.43
383.- SEN- N33- 88me- azme- 3:38- $83- Enga- have
83- Ed- Sod- 598$ caravan 4...:va ream-g $33.8
o o o o o o o o Eda
summon m8;- moa- gas: tact-Hg *tfimav lies .....*§_.o
383. 883- 88%. 5256 883 383 883. 883 was:
Ed- 33- :3- Load #5. Ed- ragged .4988
o o o o o o §8¢ o mucus:
$3- $3- m3- 338.9 $8.9 Sto- tvamoa $93.0
o o o o o o .886- Sago 3:81
352$ ........§no tamed 1268.9 tLNwoo $495.3 Rao- .1983
883- 583. 580.c- 583. 883- «885- _88.o- ~885- 28%
when than 938 mafia Yong. muse. «bum 38w ES don—

 

 

 

 

 

 

 

 

 

 

bafinem 88.52 "ma «3&-

 

 

141

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

888 $88 888 888 888 888 888 888 808580
1.13m _ .8 ”11.80 F .vv ***Amww.vv ***AQON.$ ***Avom.mv 1.23-v.8 ***Amwh.mv ***Aoom.¢v
88.8 .888 88.8 88.8 88.8 888.8 88.8 «888 88.
188.9 tLES 1.38.3 3.98.8 33.288 .4888 88.8- «:8.
o o o o o o o o 8888
158.8 1.224.; £4388: 3.8me .3286 :me3; £28.33 4......38; c
883 5: 3 88-2 883: 883 48.8.2 82.48 38.8 258.0
3.2888 128.0 12.8.8.3 1.88.8 .4536 :28me 8.3.8.3 4.5.38.3
88.8 588 58.8 888.8 RS Emu-no 882.8 884.8 8.8%
.128an 1.28.8 3.38.3 .4888 188.3 188.9 1......an 1.88.8
A88.8. 888.8- 2 8.8- 988.8. 888- 888.8. 8888- 8m88- :88

 

 

326:8 3 oz..."

 

 

142

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

82- £3- £3th $2- 398% .39de .1.meth
$33. $85 $85 2E3 83.3 323 £33 N326
:2- 23. ES- :afid :avvé ...LONNNV timovev
8256 8086 880.? $25.9 $88.? 383- £83- 3:25
..ioxwé 1.68.3 .1...e§.oe 1:820 .3...“ch $35.3 :zmowés
£83 83$- aammvfiw M5: 333 $337 282:- £9
23.33% $82.3 #2233 ...Lvoqi ..ismqmmv Emma 5&8sz
383. 8337 Sean..- 85.2 $8: wwmwfl- ammowfi- «8
1.83.3 $38.9 $283: 32.63% gaFa 333a :...§~.§
amend- $93- $3.3- $83- 82- $93.7 Soon? big...
83. $2. $8- :Awfid ism: ikvwmd mad-
o o o o o o o v.52
1250b ***Amv_.8 ***Ava.mv ***Awhw.mv ***Gmo.b .IiA—Nfib ***ovmm.vv
N586- :83- $98.? 883- :83- 28%. 883- man:
52.: $3- $6.? $3. 3:- 055. Sod-
583. o o o o o o N188:
2d- ......_.§~.$ $2. $3. «$6. .........ao_.e $8wa
o 583- o o o 5°86- 582- 352.
3:- ...sowd sad- 83- $3. itfig iafié
o 883 o o o 883 “88.9 881
was: 3?: madam Nlfiw 3.33. 0:33 99%

 

 

 

 

 

 

 

 

 

$053500 m6. u—aah.

 

 

143

 

 

 

 

 

 

 

 

mmwmom mmwmom mmwmom $38 $38 mmmmom mmwmom maouatufio
inane geese 1...?qu 1465.3 3886 $986 .........Em.$
583 8085 883 583 880.0 N885 883 808
ESQ 3:. ...ane 3.9.8.3 $3- $3- :afid
o o o o o o o 383
***Cov.ov ***Amo¢.mv ww fie- ***Awwa.vv 3.23328 ***Awmv. y C ***Amww.mv
59.3- 333- $25. $890 826m 8%? SS3 259%

 

 

 

 

 

 

 

 

 

‘05:“:00 mun Duh—N.H.

 

 

144

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

3.25.3 1.98.3 :8. 83. 288. 3 83- ........98.3 888.
8838- 88:88- 5.88 88:8 882. 8888- $388- :888- «8838
2833 83 8- 88- 88. v .2333 888- 88- 5.8-
8888- 8888 5888 2888- 8888 8888 8 8888- 8.838
$282.83 £2833 328383 1.2833 $288.3 $258.83 128883 $288.83
88.: 83.38 88:83 8888- 8888 588.8- 88882- 3.882 289
£288.83 $288.83 £2833 #28883 1.2823 $28883 $228.83 128883
882.: 8888 8838. 88.91 888.... 888.:- 8888.3- 3.83: 8.
1288.3 .3288. H 3 1.28833 8 8- .1283 _ 3 $2828 3 83 _- 1288.3
8888 8:..— 8888 8388 8.82 88:: $838- 883- have
.2883 88- .2833 838- NE. 7 SN. _- 888- .2883
8 8 8 8 8 8 8 8 888
$37 288. 3 .2833 £283 1.28 33 3.28.3 $288.3 $2883
8888- S888- 8888- 8888- 8888- 8888- 8888- 8888- was:
83. .2833 83.8- 888- 288.3 1288.3 1.2—8.3 83-
8 8 8 8 8 8 8888- 8 818.81
888. 88- 12888.3 888- 88- 85.8. 68.7 82-
8 8 8 8 8 8 8 8 3:81
3.8- 82- ...2883 82- 88. 888- 288.3 888-
8 8 8 8 8 8 8 8 88:
vmafim nabs“ mm??- Hmkgm a ism w T258 3 18m 3 T288

 

 

 

 

 

 

 

 

 

 

82.5.50 3 oz“..-

 

 

145

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

888888 888888 888888 888888 888888 888888 888888 888888 8882380
882- 8888- 882- 2888.83 1288.83 32888.3 8.28883 $2888.83

8 8 8 8 8 5888.8 8888.8 8888.8 252
888.8- 888.8- 888.8- 888.8- 238.3 88.1- 2888. 3 $2888.83

8 8 8 8 8 8 8 8 3322
*28883 888.8- 12888.83 1288.83 2888.83 8888- 3288883 1.288883

88888.8- 88888.8- 8888.8. 3888.8. 8288.8- 88888.8- 88888.8 88888.8- 88838

 

 

50:58:00 mam 03.88

 

 

146

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

$2888.83 12888.3 1.2533 12888.83 $2888.83 $2888.83 $2888.83 $2888.83
88888.8- 88888.8- 88888.8- 88888.8- 58888- 88888.8- 88888.8- 88888.8- 282
$2888.83 $2888.83 $288.83 32888.83 12888.83 3288383 12888.3 888._-
888888- 8 888888. 5888.8- 8888.8- 5888.8- _8888.8- 8 3383
1.2888. 3 $2888.83 $2888.83 12888.8 3 .128 88.83 1.22883 1.28888 3 8.28888 3
888 _ .88 88888.88 8 8888.88 8888.88 88 888.88 8 888.88 8888.88 88888.88 8888888
$2888.83 $2888.83 12888. 883 1.288883 12888.8 3 $2888.83 3.28 8883 12888.83
8888.88. 8888.88- 88888.88- 888.88- 8888388- 888388- 82:88. 8888.88- 8888.38
8888- 88:- 28883 888.8- 888.3 88.8- 8838- 888.8-
8888.8- 88888.8 88888.8 88888.8 :83. 5888.8- 5888.8- _8888.8- 1888.88
3.28838 3 $2888.83 12888883 12888.83 12888.8 3 $2888.83 $2888.83 82888883
88888.8 888888 8888888. 88888.88 8- 88888.8 88888.8 88888.8 8- 88888.88 _- 8888
12888883 3.2888883 3.28883 32888.8 3 12888888 12888.3 12888.8 3 1.28888 3
88 8888 88888.8 88888.8 7 88888.88. 88888.88 88888.8 8 8888.88- 8888.88- 88
$2888.83 $2888.83 $2888.83 :88- 8888- $2888.83 $2888.83 3.288883
8888; 88888.8 8888.8 8888.8 88888.8 88888.8 88888.8 888:: 88388
82888.3 88. 8- 32888.83 82888.83 $2888.83 $2888.83 1.28883 2888. 3
8 8 8 8 8 8 8 8 862
$2888.83 £28888 3 12888.83 12888.83 12888.83 2.28883 $2888.83 $2888.83
88888.8 88888.8 88888.8 88888.8 88888.8 88888.8 88888.8 88888.8 888:
888.8- 888.8- 888.8- *9 88.3 88288883 8888- 828. 8.28883
8 8 8 8 8 8 8 8 8.381
888.8- 882- 8888- 888.8- 888.8- 888.8- 88.8- 2888.3
8 8 8 8 8 8 8 8 3588
3.288383 $2888.83 12888.83 12888.83 8.288883 12888.88 1288.3 ...28883
88888.8 88888.8 88888.8 88888.8 88888.8 88888.8 88888.8 8 88:
9:888 098888 83288 R2888 3288 8.8.8.258 89858 32888

 

 

 

 

 

 

 

 

 

 

floaéaou mam 0.7—«H.

 

 

147

Table 3-6

Correction for Attrition
Dependent Variable: Real Loans

 

 

Weighted Instrumental
ZSLS Variable
Loans_l 0.901 -O.394
(2.075)** (4.826)**
Loans_2 0 0
0 (.>
Large Deposits 8.369 7.032
(3.963) ** (17.074)**
Liquidity -0.055 -0.007
(2.533)** (5.452)**
Equity 2375.785 -15185.313
—0.057 -0.714
ROA -74646.461 35605.798
-0.184 -0.390
Efficiency 31567.059 -22131.475
-O.245 -0.587
Qualityl -13.358 -11.522
-0.584 —1.102
QualityZ -20266.748 -56046.266
-0.327 -1.255
Quality3 16526.100 45349.763
-0.064 -0.394
ffr —140.036 —44.81 5
-0.645 -0.589
ffr_l 52.806 21.584
-0.878 -1.179
ffr_2 35.512 0.803
-0.688 -0.042
ffr_3 16.053 8.742
-0.423 -0.575
ffr_4 -0.542 -6.417
-0.019 —0.541
Core Deposits -0.011 -0.008
-0.331 -0.485
rgdp -0.576 0.612
-0.363 -0.748
Assets 0.002 0.002
-0.288 -0.447
Spread 106.603 36.080
-0.562 -0.560
Observations 1 19097 119097

 

148

Table 3-7
Loan Response to Policy Change - Levels

Dependent Variable: Real Loans

 

 

 

Percentile
75th 90th 95th >95th
Loans_l 0.583 0.333 0.466 0.137
(8.590)** (3.713)** (4.158)** -1.292
Loans_2 0.027 0.078 0.037 0.188
-1.171 (2.912)** (1.647)* (1.693)*
Loans_3 0.1 10 0.098 0.044 0.547
(5.478)** (3.063)** (2.513)** (5.403)”
netliquid 0.428 -0.002 1 1.673 0.002
(2.217)** (6.626)** (2.842)** -0.496
equity 5.746 6.544 -0.001 -8604.010
-0.185 -0.571 —0.663 (2.210)**
ROA 0.000 -12.919 35.004 -55880.477
-0.117 -0.205 -1.415 (1.986)**
efficiency -0.002 -17.783 0.024 4.563
-0.713 -1.046 -0.080 (1 .702)*
Qualityl 2.756 -0.001 -164.543 286.828
(14.730)** -0.934 -0.387 -0221
Quality2 6.581 -13.034 122.515 -3398.442
(2.159)** -0.926 -0.447 (1 .742)*
cffr 1.323 8.585 25.979 1619.882
(4.694)** (3.741)** (2.205)** (2.165)**
cffi'_1 —0.793 -5.049 -7.187 -1115.531
(1 .749)* (1 .925)* -O.569 -1.209
cffr_2 3.663 9.345 16.797 972.248
(10.188)** (3.915)** -1.297 -1.429
cffr_3 0.674 5.738 -11.080 1576.207
-1.501 (2.244 ** -0.776 -0.975
cffr_4 -2.860 -11.364 -15.982 -1810.052
(10.078 ** (5.589)** (1 .752)* (2.497)“:
crcore 0.211 0.196 0.244 1.648
(11.518 ** (7 .423)** (5.477)** (4.285)**
cspread -0.945 -4.165 2.540 -1 163.240
(2.830 ** (1 .947)* -O.232 ~0.829
crgdp 0.007 0.135 0.221 19.825
(2.025)M (5.440)** (2.127)** -1.256
Observations 37991 10348 2873 31 18
Number of entity 5595 2065 642 551

 

Robust 2 statistics in parentheses

* significant at 10%; ** significant at 5%

149

 

Table 8

Loan Response to Monetary Contraction - Regressions with Interactions

Dependent Variable: Real Loans

 

 

 

Percentiles
75‘h 90‘h 95th >95‘h
Loans_l 0.649 0.283 0.471 0.139
(8.119)** (2.238 ** (5.922 ** -1.394
Loans_2 -0.003 0.072 0.016 0.147
-O.1 18 (2.268)** -0.539 -1.140
Loans_3 0.104 0.120 0.055 0.542
(4.965)** (2.746)** (2.268)** (5.189)**
LeadLiq 64.848 -15.909 491.231 24.018
(2.252)** -1.150 -0.818 ~0.812
Intnetliq -13.779 69.251 -0.166 -11478.199
-0.064 -1.517 -0.144 -0.840
Intnetliql 0.004 -26.342 -58.378 7.467
(4.289)** (1 .960)** (2.773)** -0.131
Intnetliq2 12.317 -16.873 1507.141 140525.944
(1 .944)* (1 .978)** -0.027 (1 .786)*
Intnetliq3 -8.827 -13.730 -543.878 -27724.594
-0.065 -0.648 -0.652 -1.205
Intnetliq4 11.779 339.714 140.280 9586.346
(2.737)** (2.905)** -1.320 -0.743
Leaquuity -28.969 -8.014 —594.878 -158242.000
-1.158 -0.230 -0.913 -0.093
Intequity 39.068 -24.084 -45.971 ~22156.397
-0.212 -0.487 -1.101 (2.163)**
Intequityl -0.011 26.957 51.300 3934.247
-1.188 -0.566 -0.127 -0.250
Intequity2 -1.643 98.717 134.771 7.014
(1 .813)* (2.560 ** —0.666 (2.445)**
Intequity3 -0.496 -51.833 384.050 -352533.000
-1.399 —1 .615 (1 .662)* -0.237
Intequity4 -10.719 123.643 69.722 7.519
-1.375 -0.586 (1.891)* (3.208)**
LeadEffy 18.410 -262.868 —871.247 33437.769
(2.515)** -1.252 ~0.270 (1 .957)*
IntEffy 12.708 155.137 -32.331 ~2963.531
-0.974 —0.728 -0.660 -1.572
IntEffyl -4.929 -1.364 -150.724 -73839.827
(2.909)** -0.005 -1.458 (1.972 **
IntEffyZ -4.135 -145.435 254.713 13099.041
-0.808 -0.731 -0.288 (2.078)**
IntEffy3 0.000 117.007 -1 109.591 -5.752
-0.786 -0.548 -0.140 -0.647
IntEffy4 -2.358 -15.331 854.062 415290.550
-0.479 -0.069 (2.867)** (1 .932)*

150

Table 8 continued
LeadROA
IntROA
IntROAl
IntROAZ
IntROA3
IntROA4
LeadQualityl
IntQuality1
IntQuality1_1
IntQuality1_2
IntQuality1_3
IntQuality1_4
LeadQualityZ
IntQuality2
IntQuality2_1
IntQuality2_2
IntQuality2_3
IntQuality2_4
cffr

cffr_1

cffr_2

cffi'_3

cffrlead

13.601
-0774
-61.225
4.334
0.002
-O.598
-6.288
-0.1 13
0.001
(3.926)**
0.000
(3.207)**
43.659
4.440
-7.767
-0721
4.400
-O.276
23.033
-0.548
40.494
(2.233)**
40.719
4.375
41.067
(2.463)**
45.126
(3.863)**
12.419
-0.180
4.286
(1.802)*
81.259
(1.775)*
42.018
(2.725)**
4.835
(9.238)**
-3.148
(5.357)**
3.808
(7.783)**
-3029
(6.508)**
-3753
(8.702)**

-158.040
4.335
403.844
(2.867)**
61.333
-0541
63.383
-0542
496.028
(1.984)**
49.995
(1.841)*
-0007
-0445
0.023
-O.864
-0.046
(1.913)*
0.024
(2.100)**
0.002
(1.966)**
123.643
-0.586
-67.546
4.277
-59774
4.026
-59.621
4.155
-54.256
(1.965)**
69.901
(1.957)*
-27430
-0.607
21.097
(5.456)**
41.214
(3.764)**
9.833
(3.498 **
40.578
(3.706)**
41.458
(3.869)**

151

-0030
(2.769)**
790.810
4.372
80.557
-0914
0.084
4.121
50.922
-O.S36
33.672
4.159
4.399
-0.206
326.879
-0941
253.390
-O.698
480.448
4.157
470.166
-0.248
69.722
(1.891)*
106.241
-O.582
-0023
-0914
791.031
4.173
0.070
-0917
-60.280
-0950
28.097
4.231
296.344
-0721
333.137
-0319
936.937
4.040
500.699
-0.873
-278.214
-0.676

202542.082
(1.982)**
-4240.855
(1.895)*
495.115
4.331
233857.1 10
(1.871)*
15222014
(1.668)*
8180.833
(2.045)**
-34619.967
4.359
255066.668
4.179
-849.379
(1.983)**
-85342.591
-0.080
3969.323
4.347
7.519
(3.208)**
72375.328
4.483
47.824
-0.306
5933.934
(1.752)*
-7536.938
-0.600
4387.982
(2.002 **
-9.693
(2.032)**
2976.018
(1.920)*
4909.088
(1.656)*
2118.603
(2.929 **
4523.900
.0.709
4777.282
(1.648)*

Table 8 continued

 

 

crcore 0.085 0.124 0.178 1.759
(5.076)** (5.153)** (4.479)** (4.398)**
cspread 1.976 6.088 14.058 -333.408
(4.346)** (2.157)** -1.000 -0.172
crgdp 0.004 0.122 0.227 18.240
—1.073 (4.259)** (2.009)** -1.077
Observations 39284 10024 2737 3024
Number of entity 5632 2020 619 531
Test For Joint Significance of the Lags
F-Stat (All Variables) 4.000 3.970 2.460 3.490
P-value 0.0454** 0.0463** 0.1 17 0.0617*
F-Stat (Liquidity) 6.280 5.860 0.710 2.930
P-value 0.0122** 0.0155** 0.0582* 0.0881 *
F-Stat (Equity) 1.740 1.500 2.650 2.350
P-value 0.679 0.220 0.104 0.125
F-Stat (ROA) 1.650 0.630 1.040 2.710
P-value 0.199 0.426 0.308 0.0996*
F-Stat (Efficiency) 2.760 1.830 0.690 4.010
P-value 0.0965* 0.176 0.407 0.0454**
F-Stat (Qualityl) 5.900 3.320 1.160 3.730
P-value 0.0151** 0.0687* 0.282 0.0533*
F-Stat (Quality2) 2.910 2.650 1.650 2.710
P-value 0.0879* 0.104 0.197 .1000*
F-Stat (FFR) 6.610 4.470 0.250 1.230
0.0102** 0.0432** 0.615 0.267

 

152

Table 9
Ability to Issue External Debt

Dependent Variable: Federal Funds Purchased

 

 

Coefficient z-stat
Funds_1 0.116 -0.745
Liquid Ratio -386.221 (1.776)*
Equity 1636.881 -1.114
ROA 8912.664 (2.013)**
Efficiency -303.177 -0.188
Quality1 0.741 (1 .824)*
Quality2 208.507 -0.867
Quality3 -2531.828 -1.35
cffr 48.128 -0.401
cffr_1 -38.720 -1.154
cffr__2 -19.223 -0.583
cffr_3 -1.197 -0.044
cffr_4 -4.871 -0.23
crcore —0.01 9 -0.681
cspread -233.007 (2.519)**
crgdp 1.541 —0.55
Observations 7123
Entities 1251

 

153

    
 

EEEEEE

111111111111111(11113111111111