tilburg university essays on banking and regulation ... · radomir ivanov todorov geboren op 8...

Tilburg University

Essays on banking and regulation

Todorov, R.I.

Document version:Publisher's PDF, also known as Version of record

Publication date:2013

Link to publication

Citation for published version (APA):Todorov, R. I. (2013). Essays on banking and regulation. Tilburg: CentER, Center for Economic Research.

General rightsCopyright and moral rights for the publications made accessible in the public portal are retained by the authors and/or other copyright ownersand it is a condition of accessing publications that users recognise and abide by the legal requirements associated with these rights.

- Users may download and print one copy of any publication from the public portal for the purpose of private study or research - You may not further distribute the material or use it for any profit-making activity or commercial gain - You may freely distribute the URL identifying the publication in the public portal

Take down policyIf you believe that this document breaches copyright, please contact us providing details, and we will remove access to the work immediatelyand investigate your claim.

Download date: 09. Feb. 2019

https://research.tilburguniversity.edu/en/publications/d0f202a7-ae6f-43d5-91b4-3ee4640d9b91

Essays on Banking and Regulation

Essays on Banking and Regulation

Proefschrift ter verkrijging van de graad van doctor aan Tilburg

University op gezag van de rector magnificus, prof. dr. Ph. Eijlander, in

het openbaar te verdedigen ten overstaan van een door het college voor

promoties aangewezen commissie in de aula van de Universiteit

op woensdag 5 juni 2013 om 10.15 uur

door

Radomir Ivanov Todorov

geboren op 8 september 1982 te Dobrich, Bulgarije.

Promotiecommissie:

Promotor:

prof.dr. Thorsten Beck

Copromotor:

dr. Olivier De Jonghe

Overige Leden:

prof. dr. Franklin Allen

prof. dr. Hans Degryse

prof. dr. Vasso Ioannidou

dr. Fabio Feriozzi

dr. Maria Fabiana Penas

iv

Acknowledgements

I am grateful to the many people that have supported me in writing this

thesis. I thank my supervisors for their invaluable support and advice, and I hope

to continue working with them in the future. I also thank the members of

my PhD committee, which comments have greatly benefitted my work. The

Department of Finance at CentER and the Wharton Financial Institutions

Center in Philadelphia have been great hosts for my research, and my spe-

cial thanks go to Frank de Jong, Joost Driessen, and Franklin Allen. I also

want to take this opportunity to thank my fellow PhD students, colleagues,

and friends for the good laughs and moments we had in the past years.

Above all, I thank my family for being a constant source of unconditional

support, encouragement, and inspiration over the years. Unfortunately, my

father has passed away a month before my defense in June 2013. My dad

has been a great source of inspiration and support, and he would have been

very happy if he could attend my defense. I am immensely grateful to my

parents for standing by me throughout my life and the past years, and I

gratefully dedicate this thesis to them.

Radomir Todorov

Frankfurt (Main), May 2013

v

Table of contents

Preface…………………………………………………………………………….. 1

Chapter 1. The Effect of Peers on Bank Capital………………………………….. 4

Chapter 2. Supervising Cross-Border Banks: Theory, Evidence and Policy…….. 54

Chapter 3. Banks and Monetary Policy in Africa: Evidence from Uganda……… 95

vi

Preface

This Ph.D. dissertation consists of three essays on multinational bank supervision, bank

capital, and monetary policy. The purpose of these essays is to explore (i) the influence of peer

banks on bank financing decisions, (ii) the leniency of national supervisors in the intervention

of banks with cross-border activities, and (iii) the impact of monetary policy and international

credit market conditions on the provision of credit in a developing country. Below I present a

brief overview of the three chapters of the thesis.

Chapter 1: The Effect of Peers on Bank Capital

This chapter studies the impact of peer banks on a bank choice of capital. Previous studies

suggest that banks herd in many areas of their business, which may have aggravated the extent of

the global financial crisis. However, little is known whether bank capital decisions are influenced

by mimicking the capital choice of peers. The question is important because regulating capital

is an essential part of any debate on bank regulation. If peer banks influence bank capital, this

can also lead to externalities with a significant impact on the financial system.

The study adds to the discussion on optimal bank capital level. In the banking literature,

there is no clear consensus yet about the optimal range for capital, but a common agreement

that bank value decreases when capital is either very little or too much. In either case, capital

sends a warning sign to investors about the degree of agency conflicts or asymmetric information

in the bank. Since market investors have to make a guess about the permissible level of capital

for any bank in the group, the capital holdings of peer banks can provide a clue about the

admissible range of capital. For this reason, a bank has an incentive to adjust its capital ratio to

the capital level in the peer group; otherwise, it can face a higher cost of equity or debt financing

in the markets.

In line with these arguments, I show empirically in the chapter that peer bank choices on

capital have a statistically and economically significant impact on bank capital. The evidence

from data for publicly listed commercial banks in the US between 1971 and 2010 suggests that

publicly listed banks follow the capital choices of their peers when they decide on changes in

bank capital, or in other areas of financial policy such as the reliance on nondeposit liabili-

ties. Furthermore, the evidence suggest that the impact of peers intensifies with an increase in

competition and changes in the banking environment.

Chapter 2: Supervising Cross-Border Banks: Theory, Evidence and Policy (with Thorsten

1

2 Essays on Banking and Regulation

Beck and Wolf Wagner)

Cross-border banking has gained importance across Europe in recent decades, as a part of

a larger globalization wave in financial services. The problematic resolution of failing cross-

border banks in Europe during the current crisis has focused attention on the discrepancy in

geographic boundaries of bank activities and their supervision. The failure of the Icelandic

banks, for example, or the Fortis bank have come with wide-ranging economic and political

consequences. In this chapter, we study the distortions of national supervision of international

banks by comparing the trade-offs of national versus supra-national resolution authority.

In a theoretical model, we demonstrate first the distorted incentives of national supervisors

when deciding to intervene in failing banks with cross-border activities. We show that national

supervisors’ incentives to intervene in a timely manner in a failing bank increase in the bank

foreign equity share and decrease in the bank share of foreign deposits and assets.

In a second step, we provide empirical evidence consistent with the model using a sample

of intervened banks during the crisis of 2007-2009. We find that banks with a higher share of

foreign assets or deposits are subject to interventions at a higher level of fragility. Banks with

a higher share of foreign equity are subject to less lenient regulatory decisions.

Finally, we add to our analysis a discussion of the current regulatory arrangements for cross-

border banking in Europe and recent reforms. For example, a supranational supervisor could

always improve welfare because of taking into account the effects that materialise outside of

each country. However, supra-national supervision might itself also be subject to imperfections

and we discuss in the paper cases when the supranational supervisors may fail.

Chapter 3: Banks and Monetary Policy in Africa: Evidence from Uganda (with Thorsten

Beck)

This chapter studies how changes in monetary policy and foreign interest rates affect bank

credit provision in Uganda between 1999 and 2005. Like in many other developing countries, the

country faces conditions for the conduct of monetary policy that are uncommon for developed

countries. There is a small banking sector that is dominated by foreign-owned banks. A signifi-

cant share of assets and deposits at some banks are in foreign currency. Further, the institutions

necessary for the fluent conduct of monetary policy are at an early stage of development.

Under these conditions, also unique for many other developing countries, there is limited

empirical evidence of how bank lending reacts to monetary tightening. On the example of

Uganda, we study the mechanism of monetary policy transmission and whether banks have

Preface 3

particular ways of mitigating the impact of monetary policy shocks. We try to find answers to

these questions using a unique quarterly dataset on commercial banks coming from the central

bank in Uganda.

Our analysis gives no evidence at the aggregated data level of bank loans or bank assets

being sensitive to monetary policy in Uganda. In contrast, we find preliminary evidence for

the transmission of foreign financial shocks to the Ugandan economy. We discuss and test in

a following step for the existence of a bank lending channel of monetary policy under three

alternative views. Under the conventional credit view, the rise in the policy interest rate induces

a decline in bank liquidity. When banks cannot get additional funds, they have to cut the

supply of loans. Another view considers the role of foreign-owned banks in transmitting foreign

monetary shocks in the host country. Monetary tightening in a parent bank’s country can

negatively affect the credit supply in the economy hosting its subsidiary. Our empirical analysis

let us conclude there is no support for the conventional view of a bank lending channel in Uganda,

while foreign monetary policy can have real effects on the economy of the country. Changes in

the foreign interest rates are found to inhibit loan growth in domestic currency and increase the

growth in foreign assets.

A third view of the bank lending channel stresses the role of financial dollarization in bank

deposits. Under this view, banks can offset any liquidity shortage during monetary tightening

by converting foreign currency deposits in the domestic currency. Being less affected by mone-

tary shocks, those banks should have a higher loan growth than other banks during monetary

tightening. We find support for the role of deposit dollarization in the transmission of monetary

policy shocks in the banking system. We also provide evidence of reserve requirements having

a significant impact on bank asset composition.

Finally, we draw and discuss several policy conclusions. Among others, our findings imply

that central banks in developing countries should take into account the possible impact of foreign

currency deposits on monetary policy and a county’s exposure to foreign monetary shocks.

Chapter 1The Effect of Peers on Bank Capital

Abstract: I analyze if bank peers influenced bank decisions on financial policy and Tier 1

capital among publicly listed US commercial banks between 1971 and 2010. The results show

that capital decisions of peer banks are an economically significant determinant of capital. Banks

adjust capital ratios in response to capital decisions of other banks in their peer group. Similarly,

banks make adjustments to Tier 1 capital and nondeposit liabilities depending on the choice

made by their peers. There is also evidence suggesting that the impact of peer banks on bank

capital decisions is stronger when competition among banks intensifies.

1.1 Introduction

This paper studies the impact of peer banks on a bank choice of capital.1 Previous studies

suggest that banks herd in many areas of their business, which may have aggravated the extent

of the global financial crisis (Freixas (2010), Farhi and Tirole (2012), Bonfim and Kim (2012)).

However, little is known whether bank capital decisions are influenced by the capital choice of

peers. The question is important because regulating capital is an essential part of any debate

on bank regulation. If peer banks influence bank capital, this can also lead to externalities with

a significant impact on the financial system. Therefore, better understanding of peer effects

among banks is necessary for an effective bank capital regulation.

The study also adds to the discussion on optimal bank capital level. In the banking literature,

there is no clear consensus about the optimal range for capital. However, there is a common

agreement that bank value decreases when capital is either very little or too much. In either

case, capital sends a warning sign to investors about the degree of agency conflicts or asymmetric

information in the bank (e.g. Berger et al. (1995), Acharya et al. (2013)). Since market investors

have to make a guess about the permissible level of capital for any bank in the group, the capital

holdings of peer banks can provide a clue about the admissible range of capital. Therefore, a

bank has an incentive to adjust its capital ratio to the capital level in the peer group; otherwise,

it can face a higher cost of equity or debt financing in the markets.

1 In the area of research in banking, the capital ratio traditionally refers to the ratio of equity to assets.

4

The Effect of Peers on Bank Capital 5

In addition to bank capital, banks could also adjust the mix of debt financing or regulatory capital

in response to peer levels. Market investors could get additional information from these variables about

bank financial health. Holding capital above the minimum level can be costly except out of precautionary

motives not to violate the minimum capital requirement. The same line of reasoning can apply to the

ability of banks to rely on non-deposit funding as well. Deposits offer a cheap source of financing to

banks, also subject to deposit insurance. Because bank ability to attract non-deposit debt relative to its

peers depends on financial health and soundness, a high share of non-deposit debt can lower the cost of

debt financing.

In the analysis of bank peer interactions, I use data for publicly listed commercial banks in the US

between 1971 and 2010. A challenge comes from disentangling the simultaneity between actions of an

individual and the actions of peers in the peer group. This type of simultaneity is commonly known

as the reflection problem as discussed by Manski (1993, 2000). An instrumental variable approach can

help in resolving the problem. It requires a variable that is correlated with peer characteristics but not

with the characteristics of any individual member of the group. For isolating the impact of peer financial

policies on bank capital, I use the idiosyncratic shocks to stock returns of peer banks in an IV analysis.

My estimations show that the peer equity shock has a significant impact on peer capital, but it is unlikely

to explain the capital of a bank excluded from the peer group. The approach follows Leary and Roberts

(forthcoming), which evidence suggest that this instrument can adequately address the endogeneity issues

in a peer effects analysis on publicly listed firms.

I estimate models on bank capital, Tier 1 capital, and the mix of deposit versus nondeposit financing.

Because the data in the sample spans over 40 years, I can also compare how the impact of peers on

bank capital varies with changes in the degree of bank regulation, the intensity of bank competition,

or changes in market structure. The estimated models show that peer bank choices on capital have a

statistically and economically significant impact on bank capital. The evidence suggests that publicly

listed banks follow the capital choices of their peers when they decide on changes in bank capital, or

in other areas of financial policy such as the reliance on nondeposit liabilities. On average, a bank

increases its book capital and market capital ratios by 0.8 percent and 1.8 percent respectively after a

standard deviation increase in the peer capital ratios, holding all other factors constant. In addition, a

bank increases its risk-adjusted capital ratio on average by 0.23 percent for a standard deviation increase

in the Tier 1 capital ratio of peer banks. The results give additional evidence of the impact of peer bank

characteristics on bank capital. A bank of any size increases its capital and Tier 1 capital ratio if peers

increase in size (or become more profitable), but not when they become smaller or have lower profits.


In addition, the paper brings into discussion what should be the proper definition of a bank peer group.

The theoretical literature predicts that bank size and business focus are a joint measure of bank success

(Strahan (2008)). Forming peer groups in the paper follows this prediction reflecting a standard approach

among practitioners and supervisors when comparing peer banks (FFIEC (2008)). First, in every time

period banks are sorted in two peer groups having a diversified or regional focus. The banks in each

peer group are ranked according to total assets and assigned to their quartile of the asset distribution. In

a robustness check, I also follow a less general approach. I assign banks according to their geographic

location in one of the 12 Federal Reserve Bank districts in the first step. Next, I repeat the same procedure

in forming the peer groups. The estimated models yield results that are similar in economic and statistical

significance under both definitions of the peer groups. The findings support the view of banks being

perceived by investors as a homogeneous group nationwide among investors while bank location seems

to be of secondary importance.

Finally, the results in the paper are of relevance to bank regulation. Mimicking peer capital poli-

cies can have negative externalities when many banks engage (Acharya and Yorulmazer (2008), Freixas

(2010), Wagner (2011), Bonfim and Kim (2012), Farhi and Tirole (2012)). Peer influence can weaken

financial stability and heighten the spread of contagion. For example, systemically important banks may

have an incentive to adopt similar business strategies. If they fail, this behavior increases the chance

of a collective bailout. Allen et al. (2011) show that similar portfolio decisions by banks can lead to a

higher systemic risk by making bank defaults more correlated. Similarly, when many banks engage in

following peer capital decisions, systemic risk can increase if it leads to reducing capital buffers. While

microprudential policy can address this practice at individual banks, the possible engagement of many

banks simultaneously creates concern at a macroprudential level.2 However, the possible impact of peers

on bank capital remains untouched by any regulatory policy.

To the best of my knowledge, my empirical study is the first one providing evidence of interactions

among banks and their peers having an effect on bank capital holding. The study is close to Leary

and Roberts (forthcoming), but both studies differ along several dimensions. While Leary and Roberts

(forthcoming) analyzes peer capital structure decisions as a determinant of firm capital structure, this

study places a focus on analyzing peer effects in a regulated industry. Further, the empirical analysis

takes into account the specific nature of bank business developing empirical hypotheses in line with the

existing theoretical and empirical models of bank capital (Berger et al. (1995), Heider and Gropp (2010),

Acharya et al. (2013)). Different factors explain bank capital structure and firm capital structure. Finally,

2Bonfim and Kim (2012) provide similar arguments for bank herding in the management of liquidity risks.


Leary and Roberts (forthcoming) uses firm industry as the criterion for assigning peer groups, while this

study uses size and degree of diversification as within-industry criteria for peer group formation.

The paper continues as follows. The next two sections include a discussion on why peer effects could

have an impact on bank capital and offer a brief review of the literature. Section 1.4 presents the data and

methodology. It also describes the peer group formation, and addresses the identification of peer effects.

Section 1.5 presents the estimated models, and section 1.6 contains the conclusion.

1.2 Context

There are several explanations in the literature on herding in bank business. One view explains it with

managers’ concerns for a reputation, while the other with the tension between bank managers and bank

shareholders (e.g. Rajan (1995), Zwiebel (1995)). Scharfstein and Stein (1990) offer a model explaining

the behavior of mimicking others with the goal of attaining a good image. The incentive follows from

a belief that a reputation improves by mimicking peers whose better payoffs are the result of a better

business status or high financial performance. Freixas (2010) argues that bank managers can be under

pressure by bank owners to follow the trends in a particular market if this will boost short-run profits.

The tension can motivate managers to follow market trends even if their view is pessimistic. A long-run

investment focus can lead to a more stable return on equity, but it could also decrease bank profits in

comparison to competitors. Therefore, bank owners can hold managers liable for a weak performance.

Meanwhile, they are less likely to be liable for a focus on short-term profits if a market-wide shock

causes bank losses. In this case, bank owners and markets will be more forgiving since aggregate shocks

are largely unavoidable by any bank.

In the empirical research on bank capital, there is no evidence of peer effects. Theory, however,

suggests that markets can use the capital of a bank to assess how it fares with asset management and

financial health (Flannery and Rangan (2008)). Bank capital can also give information about the degree

of agency conflicts between shareholders and managers within the bank, or between shareholders and

creditors. An ambiguity about the optimal bank capital level still prevails in the literature. However,

there is a consensus that bank value decreases when capital is very little or too much (Dewatripont and

Tirole (1994), Berger et al. (1995), Flannery and Rangan (2008), Acharya et al. (2013)).3 In this case,

one could argue that peer bank actions can influence bank capital decisions in more than one way. Facing

3For example, Dewatripont and Tirole (1994)’s model features excessive creditor intervention when there is too much

debt and managerial shirking with too much equity. Note also there is no benchmark result about optimal bank capital in a

Modigliani-Miller (1958) model or about the irrelevance of bank capital structure. In a frictionless world, there is no rationale

for financial institutions to exist.


uncertainty about the admissible long-term capital ratio of any bank, investors can use the capital of peer

banks to make a guess. Deviations from the peer average capital could be a useful criterion for judging

the impact of stakeholders’ conflicts on future bank prospects. The credit market - a bank’s major line of

business, for example - features a high degree of asymmetric information letting investors extract limited

information about bank loan quality. Thus, banks have an incentive to conform to peer discipline getting

capital ratios closer to the average capital in their group. Otherwise, they may face a higher cost of debt

or equity financing. This hypothesis supports the common view of banks facing two capital requirements

- one by the market and one by the regulator (Berger et al. (1995), Flannery and Rangan (2008)).4 It

should be also stressed that banks can adjust capital ratios without issuing equity. For example, when

banks experience a period of high profits but bank management does not increase dividend payouts or

share repurchases proportionally, the level of capital can increase. In a similar way, bank capital can

increase when the share price appreciates.5

The importance of peer banks in the choice on bank capital is also consistent with the existing theories

about bank capital structure in a frictional world. Take first the tax advantage of debt and the expected

costs of financial distress, which is among the overriding departures of the Modigliani and Miller (1958)

propositions. Under this assumption, the market capital "requirement" is the capital ratio at which the tax

disadvantage of debt offsets the expected costs of financial distress. Bank owners prefer debt financing

because of the tax deductibility of interest payments. However, higher leverage reduces capital and

creditors can count on fewer funds from shareholders in case of an insolvency. With a decrease in capital,

the expected costs of financial distress, such as the cost of bankruptcy and loss in bank value, rise. In

turn, the capital requirement imposed by the market rises, too. Banks can anticipate that peer average

capital helps the market in assessing how they fare compared with peers because of the opaqueness of

bank activities. Thus, banks have incentives to adjust capital ratios considering peer bank characteristics

along with bank ones. In this way, peer capital can influence the long run capital target of a bank.6

4Berger et al. (1995, p. 3) broadly define the market capital requirement as the capital ratio that maximizes the value of the

bank in the absence of regulatory capital requirements, but in the presence of the rest of the regulatory structure (e.g. safety net,

etc.). The market capital requirement embeds the notion of an optimal market capital ratio as a target for each bank in the long

run in the absence of regulatory capital requirements.5Flannery and Rangan (2008, p. 5) discuss in a greater detail how other bank capital adjustments can take place: “. . . For

example, many BHCs sold their headquarters building in the late 1980s, booked a capital gain, and then leased it back from the

purchaser. A bank can also “cherry-pick” its securities portfolio, realizing the gains on appreciated securities while postponing

recognition of unrealized losses. Loan provisioning provides another (notorious) avenue for troubled banking firms to boost

their book capital.”6The market capital requirement does not refer to stock market participants only. Shareholders share the costs of financial

distress with debt holders. Because creditors have a limited ability to estimate the costs of financial distress at debt issuance,

they can make a guess about the probability of incurring these costs by noting how bank capital deviates from the peer average

capital. Under risk neutrality, they will shift all expected costs to shareholders by adjusting the interest rates they demand for

bank debt.


Peer capital can also have a role in resolving asymmetric information problems. Agency conflicts be-

tween creditors and shareholders set an upper limit on the admissible bank capital (Berger et al. (1995)).

These conflicts can shift wealth from creditors to bank owners by investing in excessively risky projects

with negative net present values. Creditors adjust interest rates to account for this possibility.7 Publicly

available information can help creditors assess the risk of an asset portfolio, but bank balance sheets

provide limited asset quality information (Merton (1977)). Thus, banks have an incentive of committing

to market discipline by raising capital ratios to signal a better asset quality. Supporting evidence by

Flannery and Rangan (2008) shows indeed that market discipline lowers the probability of default. It

also becomes more important for stock market investors when regulatory support is withdrawn.8 While

raising capital ratios signals bank soundness, the deviation from the peer level can additionally show how

closely aligned are the interests of debt holders and shareholders in the bank compared to other banks.

Furthermore, agency conflicts between shareholders and managers impose a lower limit on admissi-

ble capital level. Bank owners cannot oversee manager’ actions effectively but they can use leverage as

a motivation for bank management to increase bank value ((Jensen and Meckling (1976), Calomiris and

Kahn (1991), Diamond and Rajan (2000)). More leverage exerts a pressure on bank managers to make

better investments and try to avoid bankruptcy. Deviations from peer capital can signal how a bank fares

in resolving such conflicts compared to other banks. Peer capital therefore can serve as a benchmark for

the lower limit of admissible capital levels to shareholders and managers.

1.3 Related literature

This paper adds to the literature on capital structure, bank capital, and social interactions among eco-

nomic agents. Gropp and Heider (2010) find similarities in the capital structures of large non-financial

firms and banks suggesting they have common determinants. The paper also challenges the common

assumption of bank regulation as the overriding departure from the Modigliani-Miller (1958) irrelevance

propositions. Its results are in line with model predictions that capital requirements are not necessarily

binding (Flannery (1994), Myers and Rajan (1998), Diamond and Rajan (2000), Allen et al. (2011)).

The literature has aimed to explain why banks have positive levels of capital. Some studies point to

precautionary motives to avoid liquidity shortages ((Blum and Hellwig (1995), Ayuso et al. (2004), Barth

et al. (2005), Bolton and Freixas (2006), Peura and Keppo (2006), Peura and Keppo (2006), Van den

7See Jensen and Meckling (1976) for general results and Acharya et al. (2012) for recent empirical evidence about risk

shifting in the case of banks during the global financial crisis.8Flannery and Rangan (2008) find that market discipline for US banks increased in importance to banks once implicit

government guarantees were withdrawn in the late 80’s.


Heuvel (2008), Berger et al. (2008), Brewer et al. (2008)). Other studies stress on the positive impact

of bank competition on capital buildups (for example, Barrios and Blanco (2003), Cihak and Schaeck

(2008)). Mehran and Thakor (2009) and Allen et al. (2011) explain positive capital levels by market

discipline coming from the asset side of the balance sheet. Flannery and Rangan (2008) identify either

regulatory pressure, periods of unusually high profitability, or market discipline as determinants of the

capital buildups at US banks.

There are a growing number of studies on the dynamic behavior of bank capital as well. These

studies assume that banks, like any other firms, maintain an optimal target ratio for capital in a world

without frictions. In the presence of frictions, shocks create deviations from the desirable target levels,

and banks face adjustment costs to reach their desired state of capital holdings. Flannery and Rangan

(2006) use a partial-adjustment model of firm leverage to show that firms have target capital structures.

They also find that the typical firm closes about one-third of the gap between its actual and its target debt

ratios each year. Most studies in banking show that banks close to the regulatory minimum requirements

exhibit a faster adjustment (Jacques and Nigro (1997), Rime (2001), Berger et al. (2008), Memmel

and Raupach (2010), Jokipii and Milne (2011)). De Jonghe and Oztekin (2012) expand this analysis in

a thorough cross-country study to show that a bank’s ability to adjust its capital ratio is influenced by

macroeconomic conditions, herding behavior and bank regulatory and institutional characteristics. They

find that banks can make faster capital structure adjustments in countries with better supervision, stricter

capital requirements, more developed capital markets and high inflation. The results suggest that in

times of crises, banks adjust their capital structure significantly faster, while in normal times, adjustment

is substantially slower if banks must increase their capital ratio.

Hirshleifer and Teoh (2009) provide arguments why social interactions matter for rational decision-

making. Economic agents herd or mimic peers because they often have to deal with information they

infer from the behavior of other agents. This form of social influence is defined as any likeness in behav-

ior that results from a direct or indirect interaction between agents. There are several reasons justifying

the existence and relevance of social influence for economic decisions. For example, an agent can follow

peers if she believes the information held by others is more accurate than hers, social behavior known

as an informational cascade (see Banerjee (1992) and Bikhchandani, Hirshleifer, and Welch (1992)). As

a result, she ignores her own information set and acts according to the information set of other agents.

Scharfstein and Stein’s (1990) model shows how concerns for a reputation can give agents an incentive

to mimic other agents. The model features the desire for a good reputation as an underlying motive. It

shows that a reputation increases by mimicking agents whose payoffs are determined by a better image


or performance. Zwiebel (1995) and a survey among CEOs by Graham and Harvey (2001) support these

findings.9 There is also evidence of how social interactions can lead to financial runs. In the models

of Diamond and Dybvig (1983) and Bernardo and Welch (2004), an action by an agent can affect the

payoffs to other agents by the action itself. In a bank run, for example, a payoff externality arises since

any depositor will have an incentive to withdraw deposits if she expects others to do it as well.

There are also models on bank herding. Acharya and Yorulmazer (2008) show that profit maximizing

bank owners have an incentive to herd with other banks because of the risk of information contagion.

Bad news about a given bank such as increases in loan-loss reserves can exert an impact on the cost of

borrowing of other banks when their loan portfolios share a common factor. Bank herding by investing

in correlated investments offers a way to minimize the impact of bad news about other banks on the

cost of borrowing and future profits. Rajan (1995) motivates herding among banks by short-termism

and reputational concerns. Bank managers in the model are rational but aim to maximize short-run

profits. They are also concerned about the market view of their abilities. Managers anticipate that the

stock market is more forgiving for bank losses during a banking crisis and the consequences for a bank

reputation are milder when all banks perform poorly. Therefore, it is optimal for managers to coordinate

jointly on investment decisions and herd by lending to negative NPV projects. Recently, Wagner (2011)

and Farhi and Tirole (2012) also point to the bank benefits of adopting the same risky strategies as other

banks. Banks anticipate that authorities have a little choice but to intervene if their collective actions lead

to a financial crisis.

The empirical evidence on bank peer effects and herding is scarce. A focus of previous research is

on bank business activities and risk management.10 Jain and Gupta (1987), Barron and Valev (2000)

and Uchida and Nakagawa (2007) find evidence of herding among banks in the US and Japanese loan

markets. Recently, Bonfim and Kim (2012) provide evidence of herding in managing liquidity risks

among the largest banks in the US and Europe. Jain and Gupta (1987)’s study uses data from the US

to test if bank herding was a main trigger of the debt crisis between 1982 and 1984. Many loans to

governments of developing countries precede this crisis and anecdotal evidence suggests that smaller

banks may have followed larger peer banks in making the same decisions of lending to certain countries.

9In their survey, Graham and Harvey (2001) report that about one third of CEOs consider the behavior of competitors to be

an important factor in financial decision making.10There is large literature on herding in the area of economics and corporate finance. There is evidence of an impact of

herding: on market pricing as a result of herding among mutual funds (Wermers (1999), Brown, Wei, and Wermers (2012)), on

international portfolios flows (Froot, O’Connell and Seasholes (2001)), on stock returns or prices in the short term as a result of

herding among institutional investors (see Sias (2004), Dasgupta, Prat, and Verardo (2011) and Choi and Sias (2009)). Leary

and Roberts (2013) find evidence of a high degree of interdependency of firm financial policies across corporate firms in the

US: a standard deviation change in the average leverage of peer corporate firms in the US is associated with an 11% change in

a firm’s own leverage.


Jain and Gupta (1987) find weak evidence of herding as a cause of the crisis. Barron and Valev (2000)

use a theoretical model to examine if it is optimal for investors to wait before investing until they receive

information about the investments of others. According to the model, less wealthy investors gain the

least from using only current information, so it is optimal to wait until the wealthier make a first move.

Using data for the 1982-1984 debt crisis, Barron and Valev (2000) find empirical support for the tendency

of smaller banks (followers) to follow larger banks (leaders) in granting credit to foreign governments.

Uchida and Nakagawa (2007) give evidence of herding in lending among Japanese banks that can explain

the increasing number of nonperforming loans after the bust of the asset bubble in the early 1990s. The

paper also shows that herding varies in degree with the highest peak during the bubble period in 1987.

Banks can also herd in areas such as risk management. They can engage in risk taking strategies

jointly instead of optimizing liquidity risks individually because of a higher chance of a collective bailout.

Bonfim and Kim (2012) find such evidence for the largest banks in the US and Europe. The authors argue

that this herding could have a negative impact on the banking system by increasing systemic risk. While

liquidity risks are addressed usually at a microprudential level, the collective actions by banks change

it to a macroprudential one. By herding, banks obviously ignore the socially optimal level. Further,

Ratnovski (2009) shows that banks choose suboptimal liquidity in risk management by herding as long

as they expect that other banks will do the same. In this way, herding contributes to a new component of

liquidity risk that is currently overseen by regulation.

1.4 Data and methodology

1.4.1 Methodology

The model follows Leary and Roberts (forthcoming) in taking a general form as:

yijt = β0 + βy−ijt−1 + γ′X−ijt−1 + λ

′Xijt−1 + δ′µj + φ

′ci + θ′ct + uijt. (1)

for a bank i in peer group j at time t. ci , ct and µj denote respectively bank level fixed effects, time fixed

effects, and peer fixed effects. The model assumes correlation in the residuals uijt of a given bank. Due

to the presence of heteroscedasticity, the model uses robust standard errors clustered at the bank level.

The simultaneity between yijt and y−ijt−1 is known as the reflection problem (Manski (1993)). Section

1.4.4 describes in details the instrument variable and its use for resolving the reflection bias.

To consider the nature of bank business, I use a standard set of bank capital determinants used in


previous studies (Gropp and Heider (2010)).11 The vector yijt includes the book capital ratio and the

market capital ratio as alternative measures of bank financial policy. The book capital ratio is of relevance

to regulators since there are capital requirements on book capital only. Capital market participants use

the market capital ratio to assess the prospects of the bank including the probability of default and the

expected costs of financial distress. As a result, bank managers have to manage jointly the levels of

book and market capital so they can comply with both requirements.12 On the left-hand side, I also

include either the bank Tier 1 capital ratio as a proxy of risk-weighted capital, or nondeposit liabilities as

a measure of bank debt composition.

Every peer variable is the average value among all banks in a quarter within a peer group excluding

bank i ’s observation. y−ijt is the measure of bank peers’ financial policies defined as the average effect

of a peer group j’s financial decisions on bank i in period t. Averaging of the variables aggregates the

most relevant information in peer characteristics. By that, the possible impact of nonlinearities in peer

interactions on the empirical results is minimized. The channels by which peers can exert an impact

on bank financial policy include either peer capital levels or balance sheet characteristics, or both. For

example, it could be the profits of bank A or its bank capital that can influence the financial policy of

bank B. Therefore, the set of control variables includes measures of bank profitability, market-to-book

ratio, collateral, and bank size for an individual bank (Xijt−1) , and those of bank peers (X−ijt−1). A

dummy variable controls for dividend payouts. It takes the value of one if a bank has paid a dividend

in a previous quarter, and zero otherwise. The model also includes the ratio of loan loss provisions to

total assets as a proxy for asset risk. Appendix A lists definitions for the key variables in the empirical

analysis.

1.4.2 Data

The data come from two sources. Bank balance sheet data for publicly listed US commercial banks

are obtained from the Standard & Poor’s (S&P) Compustat database. Next, the dataset is merged with

the stock price database of the Center for Research in Security Prices (CRSP). The data for financial

firms other than commercial banks are filtered from the combined dataset. All variable distributions

are winsorized two-side at the one percentile level to minimize the possible impact of outliers. Any

11Gropp and Heider (2010) find evidence of similarities in the set of cross-sectional, time-varying determinants of the capital

structure of non-financial firms and large, publicly listed banks.12See Berger et al. (1995) and Flannery and Rangan (2008) for a discussion on these two points. The reason for use of book

capital for regulatory purposes is the fact that not all banks are publicly listed. While the value of bank capital listed on the

stock exchange can be easily estimated by using market values, the value of closely held banks is technically more challenging.

The market value of capital for banks without traded shares can be estimated by finding the discounted net present value of

expected cash flows to shareholders.


missing observations for all variables in the model are also removed. Additionally, banks are classified

as either diversified or regional according to the Global Industry Classification Standard system (GICS).

Appendix B lists the GICS criteria that determine if a bank is diversified or regional according to the focus

of its business activities. The final sample contains 36, 741 bank-quarter observations for 909 publicly

listed banks during the period 1971 to 2010. Capital regulation is introduced in the US in the 90’s, so

estimations on regulatory capital use observations from a subsample of the data. The subsample contains

22, 233 bank-quarter observations for 805 banks between 1990 and 2010. To assure consistency, the

estimations of the empirical models are performed on the full sample of the data and on the subsample.

In both cases, the regressions yield results that are qualitatively similar.

Table 1 lists summary statistics for the variables in the empirical model. A typical bank in the cross-

section has a book capital ratio of 8 percent and market capital ratio of 12 percent. Out of total debt,

nondeposit liabilities make up a share of only 15 percent. Banks keep a buffer of excess capital that is on

average 6 percent above the permissible level. They are frequent dividend payers and the payout takes

place in 83 percent of all cases. Besides, the typical bank in the sample gives out 30 percent of its net

income as a dividend.

In addition, Table 1 documents wide variations in bank size and other characteristics in the cross sec-

tion. The market capital ratio is between 1.2 percent and 30 percent. Bank holdings of Tier 1 regulatory

capital vary between 5.6 percent and 25.1 percent. The variation across banks in the holdings of nonde-

posit liabilities is large as well. While a bank has on average about 15 percent of nondeposit liabilities,

nondebt holdings vary between zero and 43 percent among banks. The largest difference across banks

is in size. On average, a bank has assets of about 9 billion. The largest one in the dataset has more than

1300 billion dollars in a given quarter, and the smallest one, about 10 million.

1.4.3 Which are the peers of banks?

Defining the peer group for a given population of agents is of key relevance to any analysis of peer effects

(Manski (1993, 2000)). Using a wrong definition of the peer group can lead to a bias in the results. The

peer groups are formed according to bank size and a focus of activities in every time period. Banks

are classified as diversified or regional banks depending on their mix of business according to the GICS

criteria. Historically, industry classification has been an essential part of investment decision making.

The GICS has been commonly accepted as an industry analysis framework for investment research,

portfolio management and asset allocation by market participants. Similar criteria for bank peers are

used by practitioners (e.g. Bankscope, SNL) and US regulators.


The peer group definition is in line with the criteria of the Federal Financial Institutions Examination

Council in the United States (FFIEC) distinguishing among banks according to asset size.13 Established

in 1979, the FFIEC main task is to develop and apply uniform standards, and report forms for the federal

examination of financial institutions. It is an interagency developing those standards on bank analysis

for the Board of Governors of the Federal Reserve System (FRB), the Federal Deposit Insurance Cor-

poration (FDIC), the National Credit Union Administration (NCUA), the Office of the Comptroller of

the Currency (OCC), and others. The agency reports publicly available data on bank peer groups once

their call reports become known. Market investors with interest in banks have access to this data and

information about the bank peer groups’ definitions. Every bank has also access to the information about

its peer group and its peers.

The use of bank size and degree of diversification as peer criteria follows predictions by theory and

empirical studies such as Jain and Gupta (1987). Theory predicts that banks become larger and more

diversified in the long run (Diamond (1984), Strahan (2008)). Larger banks are of systemic importance

and face a higher probability of a (collective) bailout during a financial crisis. Similarly, larger and

more diversified banks tend to fare better during a financial crisis relative to smaller and regional banks.

Therefore, size and the degree of diversification exert an impact on the probability of a bank default;

which further impacts the cost of bank capital and debt.

The within-group variation according to bank size also assures that the orthogonality condition be-

tween the instrument and any bank specific peer average of the right-hand side variables in the first stage

of 2SLS regressions is satisfied. At this stage, linear dependence can arise because of the joint presence

of peer average variables and time fixed effects.

When forming peer groups, banks are sorted first as diversified or regional. Next, banks are ranked

within each peer group according to size and then banks are split into subgroups according to the quartiles

of the asset distribution. In total, a bank can be in one of eight groups. Peer groups are newly formed in

every time period so a bank can change a peer group from one period to the other, for example because

of a bank merger or acquisition. On average, a bank in the sample has 40 peers in any time period, while

the number of peers in a given quarter varies between 2 and 159. Table C1 in Appendix C provides

statistics on the dispersion in bank capital within the peer groups. The average standard deviation of the

book capital ratio in the peer groups is about 1.5 percent, and 4.5 percent for the market capital ratio.

13For example, the so-called Uniform Bank Performance Reporting (UBPR) System maintained by the Federal Financial

Institutions Examination Council (FFIEC) classifies banks according to asset size. It splits banks into more than 10 different

groups (FFIEC, 2008) as a basis for a comparative analysis of peer banks. As stated by FFIEC, an UBPR is produced for each

commercial bank in the United States that is supervised by the Board of Governors of the Federal Reserve System, Federal

Deposit Insurance Corporation, or the Office of the Comptroller of the Currency.


In addition, the Panels of Table C1 show that regional banks outnumber diversified banks across groups.

Further, there is less dispersion of book capital compared to market capital within and across peer groups.

The standard deviation of book capital increases with the quartiles of the asset distribution, while there

is a varying trend for market capital.

1.4.4 Identification

The key empirical challenge in any analysis of peer effects is about identification. Banks can simulta-

neously influence one another. Endogeneity between the dependent variable yijt and the peer variable

y−ijt−1 causes a bias broadly known as the "reflection problem" (Manski (1993, 2000)). The bias re-

quires a proper instrument to isolate the peer group impact on any member from the individual impact

of a member on the entire group. Leary and Roberts (forthcoming) use the idiosyncratic component of

stock returns when analyzing the impact of peer corporate firms on the capital structure of firms.14 Peer

equity shocks can be a valid instrument because it is unique for every firm. It is unshared with a firm

excluded from the group and uncorrelated with its characteristics.

I use the same instrument in the analysis of peer effects among banks. I estimate average monthly

idiosyncratic returns as the difference between realized and expected returns estimated by a Fama-French

(1993) factor-pricing model. The model conditions bank monthly total return Rit on a standard set of

factor loadings:

Rit = α0 + βMit (RMt −RFt) + βSMB

it SMBt + βHMLit HMLt + uit. (2)

The set of factors includes the excess market return on the broad market index ( RMt − RFt ), the

small minus big portfolio return ( SMBt ), and the high minus low portfolio return (HMLt).15

I run rolling regressions over a window that requires at least 24 time periods to generate time varying

betas for every bank in each period t. After estimating predicted returns, the idiosyncratic stock return

for a bank i in period t is extracted as:

uit = Rit − Rit. (3)

14Leary and Roberts (2013) study the instrument properties and find evidence that the exclusion condition holds and the

relevance condition is satisfied.15The value factor HML is constructed as the difference in stock returns of firms with a high book-to-market ratio and firms

with a lower one. The size factor SMB is constructed as the difference in stock returns of firms with a large capitalisation and

firms with a small capitalisation.


I annualize the equity shock by a factor of 12 to come to the average annual return.16 In regressions,

the equity shock is taken in a second lag. Table 1 shows that the equity shock as a residual of a regression

is on average only close to zero. It is expected since the estimate is a conditional average.

The exclusion condition may not be necessarily satisfied because it implies that the instrument affects

the capital of bank i only via its effect on the capital levels of peer banks. Since banks are subject

to systemic risk in contrast to non-financial firms, a liquidity shock to a systemically important bank

can easily span over other banks via contagion triggering a major repercussion in the banking system.

A shared unobserved common factor among banks can be eliminated by including time fixed effects.

However, financial systems are marked with a high degree of interdependence (Allen and Babus (2009)).

It is likely that bank stock returns can have a common systemic determinant linked to their asset portfolios

as well (Acharya and Yorulmazer (2008)). If this common factor is not eliminated by constructing bank

idiosyncratic stock returns, the exclusion condition can be violated. The instrument will directly affect

not only peer banks but also bank i.

Table 2 presents results from a test of validity for the instrument. The peer average equity shock is

fit on the set of control variables used in regressions in order to gauge the degree of statistical correlation

between the covariates and the instrument. In Column 1, Table 2, the instrument is regressed on the

set of control variables used in the baseline model. None of the control variables except for dividends

and asset risk is found to contain information about the peer average of the idiosyncratic stock return.

However, the coefficient estimates of those variables are of a very low economic significance despite

being statistically significant. A standard deviation increase in asset risk, for example, is associated with

an increase in the peer average of idiosyncratic stock returns by about half a basis point. Finally, there

is no information that the instrument carries any information about future values of control variables in

Column 2 except for the asset risk variable. However, in comparison to its estimate in column 1, it further

decreases in statistical and economic significance. Overall, the results in Table 2 suggest that the peer

average idiosyncratic stock return passes the test of a weak instrument and can be used in IV analysis.17

The feature of opaqueness in bank balance sheets can further stress the usefulness of peer equity shocks

16The use of compounding instead of annualizing yields qualitatively similar results as well.

17I test if the bank asset-pricing model adequately corrects for the presence of a common factor in bank stock returns.

Running estimations after including the return on equally weighted bank sector portfolio excluding bank i in the Fama-French

model in a set of robustness checks yields similar results. Additionally, unreported tests with pairwise correlation estimates

between the instrument and book leverage show absence of any conditional correlation. Its conditional correlation with the

market ratios ratio is -0.13; about 0.3 with the Tier1 risk adjusted ratio and -0.4 with non-deposit liabilities. These correlations

are statistically significant at the 1 percent level though the economic significance of the estimates is very low. Further, Cragg-

Donald Wald tests confirm the weak validity of the instrument in the 2SLS estimations in the following sections (see Stock and

Yogo (2005)).


as an instrument. Unlike the case of corporate firms, it is difficult for the market to assess the risk profile

of every bank from its balance sheet. Therefore, the equity shocks to peers are likely to contain less

information about any bank excluded from the peer group.

1.5 Results

1.5.1 A basic model of bank capital structure

The results in Table 3 offer insights about the determinants of bank capital for publicly listed US commer-

cial banks. The capital ratios are fit on a standard set of bank characteristics used in previous empirical

research (e.g. Gropp and Heider (2010)). These results also offer a test of the validity of the Modigliani-

Miller (1958) propositions.18 There is a view in the banking literature that deposit insurance is the main

overriding departure of the Modigliani-Miller (1958) propositions about the irrelevance of capital struc-

ture in the case for banks. In this case, the control variables for bank characteristics should enter the

model with either statistically or economically insignificant coefficient estimates. Otherwise, it can be

concluded that deposit insurance may not be the first-order determinant of bank capital structure relative

to other determinants.19 Appendix D summarizes expectations about the sign on control variables that

enter the models in Table 3. Following Gropp and Heider (2010), their predicted effects differ under

two alternative views of bank capital structure. The first column in the table in Appendix D shows the

expected signs of the control variables based on documented evidence for non-financial firms in the exist-

ing corporate finance literature, or the so called market view (e.g. Rajan and Zingales (1995), Frank and

Goyal (2004)). The second column lists expectations under an alternative view of banks as risk averse

firms aiming to build a reserve of funds so they can prevent a raise in equity on short notice.20 Under the

buffer view, banks that have either lower market-to-book ratios, pay less dividends, or are less profitable

have higher book or market capital ratios in consequent periods. Besides, banks increase the share of

nondeposit liabilities when they face a lower cost of raising debt under both views.

The first column in Table 3 presents the estimated model on bank book capital ratio and its determi-

nants. The control variables that enter regressions except for the categorical dummies are scaled to have

a mean with the value of zero and a standard deviation with the value of one so their economic effect

18The test is possible since the capital ratio is an inverse function of bank leverage, i.e., the ratio of bank debt to total assets.19See Gropp and Heider (2010) for a more thorough discussion on this point. One can argue that the determinants of bank

capital structure correspond to the same determinants of non-financial firms with the only difference that they reflect the unique

nature of banks. If regulation is the overriding departure of the Modigliani and Miller (1958) propositions, one should expect

these control variables to have no explanatory power.20For example, such need for issuance in equity can arise due to unexpected liquidity shortages driven by business cycle

conditions (Ayuso et al. (2004); Peura and Keppo (2006)). The risk aversion hypothesis is also consistent with the model

predictions by Myers and Majluf (1984) that the degree of asymmetric information determines the cost of raising equity.


on the dependent variable can be easily compared across table columns. All explanatory variables are

standardized to compare their economic effect on the dependent variable in a percentage change.21 The

market-to-book ratio enters the model with insignificant coefficient estimate. It suggests that the growth

opportunities of publicly listed commercial banks have no impact on the book levels of bank capital.

The results show that larger banks tend to keep a higher leverage, and tend to rely less on equity capital.

Similarly, increases in collateral levels have a negative effect on the book capital ratio. Holding all other

factors constant, the book capital also increases with bank profits. It is positively associated with regular

dividend payouts, but negatively associated with changes in asset risk, as proxied by loan loss provisions.

An increase in the asset portfolio risk by one standard deviation leads to a decrease of 0.06 percent in the

quarterly bank capital. Column 2 in Table 3 presents estimates of the model on the market capital ratio.

The results are similar to the ones in the model on the book capital ratio. In addition, the coefficient

estimate of the market-to-book ratio enters statistically significant at the one percent level. Holding other

factors constant, a standard deviation increase in the market-to-book ratio leads to an increase by 3.7

percent in the market capital ratio.

The standard determinants of bank capital structure can also explain the relative composition of bank

debt. Column 3, Table 3 lists estimates of a model about the bank holdings of nondeposit liabilities. In

contrast to non-financial firms, bank liabilities include deposits as well as short-term or long-term debt.

Because larger banks can access easier debt financing at a lower cost, results show that those banks tend

to have a higher share of nondeposit liabilities. Additionally, more profitable banks and banks with lower

market-to-book ratios tend to increase their nondeposit holdings in consequent quarters, ceteris paribus.

Further, neither asset risk, nor the regular payment of dividends, is found to be a significant determinant

of the relative composition of bank debt. The model on the Tier 1 regulatory capital is presented in the

last column of Table 3. More profitable banks, regular dividend payers or banks with higher levels of

collateral are found to increase the level of regulatory capital while it tends to decrease at larger banks.

Also, the market-to-book ratio and asset risk are found to have no impact on the regulatory capital.

The model on the market capital ratio has the best fit among the models in Table 3 with an adjusted

R-squared of about 0.80, while the model on regulatory capital has the lowest explanatory power. In

addition, the estimates in columns 1 and 2 have a greater match with the market view on leverage rather

than with the buffer view. These results are similar to the findings of Gropp and Heider (2010) for the

largest US and EU banks. The current study, however, uses the sample of all publicly listed banks in the

US and covers a much larger sample period. There is also support to the findings by Gropp and Heider

21The change is obtained by multiplying the coefficient estimate of a given variable with its standard deviation reported in

the summary statistics given in Table 1.


(2010) on the secondary importance of regulation as the overriding departure from the Modigliani-Miller

(1958) proposition. While most control variables in the models of book and market capital in Table 3

enter with statistically significant coefficient estimates, there are large differences in economic signifi-

cance. A standard deviation increase in profits, for example, is associated with a 0.06 percent increase

in the book capital ratio, ceteris paribus, while one standard deviation of collateral decreases the capital

ratio by 0.60 percent. Additionally, a standard deviation in log (size) leads to a decrease in the book ratio

about 0.39 percent. The difference in economic significance points out that bank regulation may not be

the only overriding departure for the case of banks.

1.5.2 A reduced-form peer model

Panel A of Table 4 shows the estimates of the baseline peer model in its reduced form. Each column

corresponds to a model that aims to explain a different measure of bank capital or bank debt composi-

tion. Across columns, the variable in use is indicated on the top of each column. Other estimates from

regressions such as the F-statistics, the number of observations, or the number of banks are given in the

bottom of each column. The F-statistics corresponds to the Cragg-Donald Wald test statistic for a weak

instrument with its range of critical values being estimated by Stock and Yogo (2005). The table also

reports the adjusted R-squared for brevity though it provides no useful information about the fit of a

2SLS model (see Wooldridge (2010)). The regressions are run under the assumption that the dependent

variable in each regression is explained by a set of bank level characteristics and a set of characteristics

of bank peers. The empirical model also includes panel fixed effects and time fixed effects. The time

dummy variables control for unspecified macroeconomic and financial market factors with an impact on

bank capital, while bank fixed effects have been shown to be important in previous studies (Flannery and

Rangan (2006), Lemmon, Roberts and Zender (2008), Gropp and Heider (2010)). The definitions for

explained and explanatory variables can be found in appendix A.

These results are not robust yet to endogeneity such as unobserved heterogeneous factors affecting

all banks in the peer group. However, despite the presence of the reflection problem, they can give a first

insight on the effect of peers on bank financing decisions. In Panel A of Table 4, the peer average level

of the RHS variables is listed first followed by the bank-level characteristics and the characteristics of

peer banks.

Since the coefficient estimates of the bank-level control variables are similar to those obtained by

the models in Table 3 in economic and statistical significance, the focus is on the estimates of the peer

variables. In columns 1 and 2 in Panel A of Table 4, the peer average capital ratio is found to exert


a positive effect on the bank capital ratios. A standard deviation increase in its value is associated on

average with about 0.47 percent increase in the book capital, and about 1.5 percent increase in the market

capital level. Similarly, the relative debt composition of peer banks as well as the peer average regulatory

capital exert a positive impact on bank choices on nondeposit financing and the build-up of regulatory

capital. A standard deviation increase in the nondeposit liabilities of peer banks leads to a 0.40 percent

increase in the nondeposit debt of a representative bank in the population, and 0.18 percent in bank

regulatory capital.

Panel A of Table 4 shows that average size and profits of peers can influence bank financing decisions,

too. Banks tend to keep more capital when peer banks are larger, while peer profits tend to have a

negative impact on capital levels. This finding could indicate that peer pressure can exert a positive

impact on bank managers to maximize the return on bank assets. Banks with more profitable peers could

raise bank profitability by using more capital to invest in high-returns projects. Further, banks operating

in an environment with smaller or less profitable peers is associated with a decrease in the share of

nondeposit bank liabilities. Because creditors determine the cost of bank debt by comparing how a bank

fares relative to its peers, this finding is consistent with the idea that smaller and less profitable banks

are less likely to obtain cheaper debt funding. Finally, column 4 in Panel A of Table 4 shows that the

bank Tier 1 capital ratio is negatively associated with the profitability of bank peers. This finding is

again consistent with the hypothesis that banks seem to deplete regulatory capital when facing fiercer

competition. As a robustness test, the lagged dependent variable is also included in the model to check if

the impact of peers retains statistical significance. In Panel B of Table 4, estimations yield a statistically

significant coefficient estimate of the peer book average variable, thus suggesting the absence of a lagged

dependent variable qualitative effect on the peer average. Its economic effect is tenfold smaller compared

to the estimate of the lagged book ratio. In addition, the coefficient estimate of the market capital peer

variable becomes statistically insignificant. It should be noted that the model in this form can only offer

a basic insight. The model with the lagged dependent variable is misspecified because the error term is

correlated with the dependent variable. As a result, the regression model can no longer yield consistent

estimators for the regression parameters.

1.5.3 IV Analysis

1.5.3.1 IV Regressions

The results from the baseline 2SLS model are given in Panel A of Table 5. Columns 1 and 2 report

the results for the models of book and market capital ratios, respectively, while column 3 - for the bank


share of nondeposit liabilities. The first stage regressions are not reported except for the coefficient

estimates obtained for the instrument, and the estimate of the F-statistics of the Cragg-Donald Wald test

in the bottom of the table. The F-statistics across all models in columns 1-5 are well above the minimum

threshold value of 10, so they pass the weak instrument test (Stock and Yogo (2005)).

The first stage regression estimates indicate that on average banks tend to increase the holdings of

capital in response to uncertainty as being embedded by equity shocks to their stock. The results reported

on the bottom of Panel A of Table 5 show that the average equity shock of peer banks is found to have

a positive and statistically significant effect on the average book and market capital ratios of peer banks,

and a negative one on the peer average holdings of nondeposit liabilities. The coefficient estimates are

also economically significant. One standard deviation in peer banks’ average equity shock is associated

with a 0.53 percent increase in the peer banks’ average book capital ratio, and about 3.4 percent increase

in the peer average market ratio. Equity shocks seem to increase the cost of nondeposit funding to banks

so consequently they rely more on deposit funds, and less so on nondeposit liabilities. The corresponding

quarterly decrease in nondeposit liabilities is by 0.43 percent. Similarly, banks also tend to increase the

risk adjusted capital. A standard deviation increase in the average equity shock leads to about 1.30

percent increase in the Tier 1 holdings of peer banks.

The second stage results in Panel A of Table 5 show that peer effects are an economically significant

determinant of bank capital. In response to a standard deviation increase in peer banks’ average book

and market capital ratios, the book capital ratio increases by about 0.77 percent (Column 1), and the

market capital ratio by about 1.78 percent (Column 2), ceteris paribus. Similarly, a given bank actively

increases its risk-adjusted capital ratio by approximately 0.23 percent for a standard deviation increase

in the risk-adjusted ratios of peer banks (Column 4). Banks also actively adjust their debt composition,

in particular their reliance on nondeposit funding in response to peers.

Table 5 also shows that the financial policy of peer banks is of high relevance to bank financial policy.

Across all columns, the average peer ratio is among the most important determinants of book and market

capital ratios. In comparison, other bank capital determinants are of lower economic significance. The

coefficient estimates of bank-specific control variables have the expected signs shown in previous studies

(Flannery and Rangan (2008), Gropp and Heider (2010)). Profits have a positive effect on bank capital,

while larger banks tend to have less equity. Most of these estimates also retain significance in the 2SLS

estimation alike in the OLS regressions in Table 3; therefore, the instrument adds sufficient exogenous

variation to the empirical model.

Panel A of Table 5 also gives some evidence on the "dynamics" of bank interactions. The results


highlight the importance for a bank on how it fares in comparison to rival banks. Peer size and prof-

itability seem to be most important to a bank for capital decisions. A standard deviation increase in the

average profits of bank peers is associated with a decrease in the quarterly book ratio by about 0.045

percent, and about 0.06 percent in bank regulatory capital. These findings suggest that banks exhaust

capital levels for new investments when peers are more profitable. Further, the results suggest that the

bank capital is used to compensate for a smaller size relative to larger peer banks in assessing the market

cost of bank equity. Since bank size is also a measure of systemic importance, this finding is at edge with

the key goal of regulation that aims at having risk-adjusted capital increase with bank size independent of

the peers’ total assets. Finally, a bank of any size has higher non-deposit debt when its peer competitors

are smaller in size but not when they are larger.

Peer pressure could take the form of deviations from the peer average. Banks have an incentive not

to deviate much of the peer level, otherwise they may face higher costs of equity or debt. Moreover,

peer pressure could be asymmetric. Banks below the peer average can differ in the speed of "catching

up with the Joneses" compared to banks above the peer average level. For this analysis, I create two

data subsamples. I estimate the deviation of bank book capital from the peer average sorting banks as

being either above the average peer level, or banks that are below the peer average level in every time

period. In a following step, I run the baseline empirical model. Results are reported in Panels B &

C of Table 5. There is strong evidence that banks below the peer average level tend to increase book

capital consequently to the average level. In contrast, banks above the peer average level tend to increase

book capital but not as much. The coefficient estimate is statistically significant at the 10 percent level

only. The economic effect is twice smaller compared to the one observed in the sample below the peer

average. Therefore, there is evidence of asymmetry in bank responses to peer average levels. Results are

similar for market capital and non-deposit liabilities. For example, the coefficient estimate of the peer

average is statistically insignificant in the subsample of banks above the peer average market capital, but

economically and statistically significant at the one percent level for the other subsample. The economic

effects in the models of Tier 1 regulatory capital are of a similar magnitude.

In section 1.2, it was discussed that agency conflicts between shareholders and creditors can set the

upper limit on the admissible bank capital, while conflicts between shareholders and managers can set

the lower one. The results from Panels B & C of Table 5 suggest an interpretation in line with the existing

theories. Banks have an incentive to signal a low degree of shareholder - credtors conflicts, meaning also

a lower risk of wealth shifting from creditors to bank owners. In this case, banks having capital above the

peer average level will be willing to decrease capital holdings to the level of their peers as shown by the


results in Panel B. In addition, shareholder-manager conflicts give a strong incentive to banks in adjusting

capital upward to the peer average level according to Panel C. This outcome can mean that banks do not

use leverage as a motivation for managers to increase bank value. Rather than that, it implies that high

leverage can signal a tension between shareholders and managers questioning the effective monitoring of

bank business. It should be noted that the economic effect of regulatory capital is of a similar magnitude

in both tables, thus there is no evidence of an asymmetry in bank responses in the model.

In Panels D and E of Table 5, the baseline model is subject to robustness tests. Placebo tests replacing

the peer average book and market capital values with the average capital values across randomly assigned

peer groups tend to show that peer groups in the empirical analysis are well-defined. The analysis

includes a simulation of 240 times of the model to find the simulated distribution of book capital and

market capital. The F-statistics of the estimations reported in Panel D of Table 5 are on average below

the value of 10. Thus, results do not pass the weak validity test for the instrument in majority and provide

support that the peer groups are well defined.

Panel E of Table 5 addresses concerns with identification coming from the construction of the instru-

ment. There are several assumptions for addressing adequately the identification problem. The results

from Table 2 in the paper show that the instrument carries no significant current and future information

about bank characteristics. The first-stage F statistics in the 2SLS analysis help judge whether the in-

strument passes the weak validity tests. Further, the use of time fixed effects controls for unobserved

common factors in any given period of time.

An omitted common factor in the Fama-French model however can have an impact on the estimated

models. Panel E of Table 5 reports estimated models after including an additional factor loading in the

model and requiring at least 36 time periods of available data. The unweighted average return on the

banking sector excluding bank i in a given period t, βBNKit (R−it − RFt) is considered to absorb any

remaining commonality between bank stock returns. The estimated models in Panel E yield qualitatively

similar results. The analysis gives weak evidence of omitted variables as a determinant of the estimated

effects.

1.5.3.2 Peer effects and regulatory minimum

Table 6 documents the degree to which banks mimic peers when they approach either the minimum

required regulatory capital, or the threshold under the market "requirement". Banks with capital levels

close to these minima are expected to focus more on stabilizing their capital position in the first place,

and less inclined to herd with their peers. For example, banks that are close to the regulatory threshold

will face penalties upon crossing it such as constraints on their business. Besides, the cost of bank equity


for those banks increases since the market and the regulator anticipate that the probability of default, and

by that the expected costs of financial distress, get higher, too.

Two dummy variables are constructed therefore and interacted with the variables of interest. The

dummy Close takes the value of one if a bank has Tier 1 capital that is below 6.5 percent, and zero

otherwise. The second dummy variable, Close 80s, defines if a bank is close to the stock market "re-

quirement." It takes the value of one if the bank is below 7 percent in holdings of primary capital (i.e.

at most 1.5 percent above the minimum of 5.5 percent), and zero otherwise.22 Estimations are carried

out on a subsample of the dataset starting in 1991 when Tier 1 capital requirements were introduced in

the US. The estimates in column 1 of Table 6 give no evidence that the peer average book capital level

has any effect on the book value of equity when a bank approaches the minimum capital threshold. The

results in Column 2, however, show that those banks still actively manage their market capital ratios.

Ceteris paribus, a standard deviation increase in the average market equity of peers leads to about 0.82

percent increase in the bank market capital ratio. Finally, Panel C shows that the variables of interest

retain statistical significance when estimation is applied on the same subsample but using the baseline

model from Table 5 without interactions with the categorical variables.

The findings suggest that adjusting book and market capital ratios by mimicking other banks is of

different importance to banks when they are close to crossing the regulatory and market thresholds. On

the one hand, supervisors in the US use book values to judge the financial condition of a given bank

because of the need to apply uniform criteria for publicly traded and privately held banks (Berger et

al. (1995), Flannery and Rangan (2008)). On the other hand, stock market investors determine the cost

of equity using market equity, and therefore banks are subject to two constraints. When being close to

exhausting bank capital, the results show it is still important for banks to adjust the market capital ratio

so it does not deviate much from the peer level. In contrast, adjusting the book capital ratio to the peer

level tends to be of secondary importance.

1.5.3.3 Peer Effects and Changes in the Banking Environment

The results in Panel A of Table 5 and Table 6 show that peer capital decisions influence bank choice

of capital and non-deposit debt. We should also expect mimicking of peer capital to differ across periods

22Following Flannery and Rangan (2008), a bank is classified as being close to the regulatory minimum if it is at most by

1.5 percent above the Tier 1 regulatory threshold in a previous quarter. Flannery and Rangan (2008) point out further that the

Federal Deposit Insurance Corporation Improvement Act (FDICIA) from the beginning of the 1990s defines a bank to be well

capitalized if its Tier 1 ratio is above 5 percent. Before the 1990s, bank primary capital was in common use as an adequacy

measure. Primary capital is defined as the sum of loan loss provisions and equity over total assets. The minimum threshold

for primary capital has been set at 5.5 percent. By assumption, this capital adequacy measure can also serve as a proxy for the

stock market requirement on bank capitalization. See Berger et al. (1995) for a discussion on book capital and market capital

requirements.


of crises, changes in market structure, or recessions. All these events can have an impact on bank

performance and competition as well as on peer interactions. The models in Table 7 analyze if indeed

mimicking of peers vary significantly with environmental changes along these dimensions.

In Panel A of Table 7, the peer average capital variable in the estimated models is additionally

interacted with a categorical variable controlling for the event of a bank crisis or a market crisis occurring

in the US. The information on crises comes from Berger and Bouwman (2010). The bank crisis periods

include the credit crunch from early 1990s and the subprime mortgage banking crisis from 2007 to 2009.

There are also three market-related crises. The first one was in 1987 when the stock market crashed,

while the second one was in 1998 when the Russian government defaulted on its debt. The third market

crisis occurred in 2001 with the dot.com bubble burst and September 11 terrorist attack. Appendix A

provides definitions and sources of all event variables.

The results in Panel A give limited evidence that peer effects vary differently in crisis and non-crisis

periods. Only the models of market capital and Tier 1 capital yield first stage F-statistics estimates above

the value of 10, thus passing the instrument validity test. Among models, only the coefficient estimate of

peer market capital enters statistically significant (Column 2 of Panel A). This lets us conclude that peers’

impact on bank market capital is much lower during crisis periods than non-crisis ones. The findings are

consistent with the idea of banks preferring to focus in the first place on stabilizing their balance sheets

and capital positions in bad times. In Panels B and C, the estimated models make a further distinction

between stock market or banking crisis periods. The results from these models give no evidence that the

impact of peers is significantly different in stock crisis periods compared to other times. However, peer

banks continue exerting an impact on bank market capital and nondeposit liabilities during a banking

crisis. Yet, their impact is of a lower degree letting us conclude that mimicking becomes less important

when the banking sector is hit by aggregate negative shocks.

In Table 8, the peer average variables of interest are additionally interacted with categorical variables

controlling either for the effect of intrastate branch deregulation across different US states, or for the

impact of recessions. The start of intrastate branch deregulation across states in the 70s reshapes the

market structure of the banking sector, making competition among banks within states stronger and

increasing bank efficiency and performance (Flannery (1984), Jayaratne and Strahan (1998)). There

is evidence showing that the removal of branching restrictions has also led to an increased entry into

local banking markets (Amel and Liang (1992)). With increased rivalry, the performance of a bank

compared to that of bank peers should have a much stronger impact on the market cost of bank equity.

Following this presumption, banks should have more incentives to adjust capital levels to that of their


peers. Thus, we expect to find that the effect of peers on bank financing decisions vary with the degree of

competition. The results in Panel A show that intrastate branch deregulation is found to have a positive

impact on the book and market capital of banks compared to the periods when branch restrictions are

still in place. Further, the impact of intrastate deregulation on regulatory capital and the holdings of

nondeposit liabilities is a positive one as well. Peer capital seems also to have a differential impact on

bank book and market capital in recessions. According to the estimates in Panel B of Table 8, peer capital

has a positive impact on book capital during economic slowdowns but a negative one on market capital.

This finding is consistent with the conjecture that book capital is a relevant measure of bank soundness

during recessions for the stock market.

In Panel A of Table 9, the peer average variables are interacted with a dummy variable controlling

for the passing of the Gramm–Leach–Bliley Act (GLB) in 1999. The GLB Act repeals parts of the

Glass-Steagall Act from 1933 not allowing any financial institution to have activities in the areas of

banking, insurance and investment banking. The results from these models show that the enactment of

the GLB Act had no impact on book capital adjustments in response to peer banks. However, peer effects

in market capital and non-deposit liabilities are found to be not as strong following the enactment. In

Panel B of Table 9, the peer variables are also interacted with a dummy controlling for the interstate

branch deregulation that came into effect in 1994 with the Riegle-Neal Interstate Banking and Branching

Efficiency Act. The results from these models show that the peer effects on book capital become stronger,

and the peer effects on market capital become weaker in the period after interstate deregulation. However,

the event has had no impact on the mimicking of peers in the area of Tier 1 regulatory capital and non-

deposit liabilities. It could be argued that both laws have removed the constraints to banks in terms of

geographic and business-type of diversification, which seems to be of relevance to stock markets. The

removal of constraints among those dimensions seems to have decreased the incentives of banks to adjust

market capital levels to the average level of their peers.

1.5.3.4 Leaders vs. Followers

The theory on herding suggests that managers tend to adopt the business strategies of their more

successful peers. Success can be measured by a better business image that results from a history of

higher payoffs (Scharfstein and Stein (1990)). To analyze if there are leader banks in the sample that

are followed by the rest in making capital choices, the model is extended to differentiate between banks

with a leading position in the market for deposits or loans, and other banks.23 More precisely, the deposit

and loan shares of all banks in a given peer group are estimated in the first step. Second, the banks with

23Leary and Roberts (forthcoming) conduct a similar analysis for non-financial firms finding evidence of a leader effect.


market shares above the 75th percentile of the loan or deposit distribution are defined as leaders, and

the rest as followers. In this empirical model, the capital decisions of follower banks are fit on the peer

average capital variable of leader banks.

The results are listed in Table 10 and give evidence of a leader effect in the making of capital de-

cisions by follower banks. The peer average bank capital at leader banks enters with statistically and

economically significant coefficient estimates in all estimated models. In response to increase in the

book capital of leader banks by one standard deviation, follower banks increase their book capital by

0.55 percent, ceteris paribus. Similarly, the market capital ratio of a follower bank increases by 0.34

percent, nondeposit liabilities by 0.70 percent, and Tier 1 capital by 0.33 percent, respectively. Results

are consistent with the theoretical predictions of observational learning models (Banerjee (1992) and

Bikhchandani, Hirshleifer and Welch (1992)). In those models, agents assign a positive weight on past

decisions of others believing these agents have more expertise or information. At the same time, they

ignore and do not count on their information. In unreported results, I also estimate models where the

peer average of follower banks is fit on the sample of leader banks as a robustness check. The models

return statistically insignificant coefficient estimates, therefore giving additional support to the presence

of a leader effect.

1.5.3.5 Peer effects and geographic location

The empirical models in previous sections assume so far that regulators and markets consider the

group of all publicly listed banks as a homogeneous one on a national scale. Size and focus of business

activities serve as the only main criteria by investors in the peer analysis for two reasons. Since all

banks are under nationwide supervision by the Federal Reserve and other agencies, they all conform to

same regulatory set, irrelevant of geographic location. Besides, commercial banks comply with identical

SEC requirements for getting listed on a stock exchange. Therefore, two main measures of the cost of

equity when it deviates from the peer average capital level and balance sheets are very opaque can be:

1) bank size as a measure of systemic importance and soundness, as well as 2) focus of activities as a

measure of the degree of diversification. Market efficiency ensures then that all banks with the same

bank characteristics have the same cost of equity.

These assumptions for peer group formation do not take bank geographic location into account. For

instance, the same peer group in any time period can include a bank located in Florida on the East

Coast and another one located in Los Angeles. It could be argued that banks mimic peers only in the

geographical region where they are active. Table 11 lists the estimates of empirical models where peer

groups are formed according to size and focus of activities in every period but within the districts of the


12 Federal Reserve Banks. The results give evidence of local mimicking in peer capital decisions and

nondeposit funding. Banks have higher book and market capital ratios following increases in the average

capital ratios of peers within their Federal Reserve district. We also see from Table 11 that the results

about regulatory capital are inconclusive since the model of Tier 1 capital achieves low F-statistics. The

coefficient estimates are similar in economic significance to those of the baseline model in Table 5.

Thus, they show that bank geographic location is of secondary importance in forming peer groups across

publicly listed US banks. Peer groups under this definition seem to be subsets of the peer groups under

the broader definition on group formation, thus explaining the similarities in coefficient estimates.

1.6 Conclusion

This paper studies the impact of peer banks and their characteristics on the choice of bank capital fi-

nancing. The analysis uses data for a sample of publicly listed commercial banks in the US from 1971

through 2010. The evidence suggests that banks manage capital ratios by observing how capital of peer

banks changes. Banks similarly adjust the levels of regulatory capital and nondeposit liabilities following

changes at their peers.

The mimicking of peers in bank capital decisions suggests important implications for bank regula-

tion. It remains largely unaddressed by policy makers in the current regulatory framework. Because bank

capital is a basic element of bank regulation, further research is necessary on the impact of peer effects on

stability in the banking system. It can have adverse outcomes for the financial system such as contagion

if it causes capital depletion at systemically important banks. Thus, bank peer effects may originate on

a micro-level but they would need a macroprudential response that is currently missing. Banks should

mimic peers in other dimensions such as on the asset or liability side of the balance sheet as well. While

such study can lead to testing interesting empirical hypotheses about interdependencies in bank funding

activities, it is beyond the scope of the current study and is an area for future research.


References

[1] Acharya, V. and T. Yorulmazer, 2008. Information Contagion and Bank Herding. Journal of Money,

Credit and Banking 40(1): 215-231.

[2] Acharya, V., I. Gujral, N. Kulkarni, and H. S. Shin, 2012. Dividends and Bank Capital in the

Financial Crisis of 2007-2009. C.E.P.R. Discussion Papers.

[3] Acharya, V., H. Mehran, and A. Thakor, 2013. Caught Between Scylla and Charybdis? Regulating

Bank Leverage When There is Rent-Seeking and Risk-Shifting. mimeo

[4] Allen, F. and A. Babus, 2009. Networks in Finance, The Network Challenge , edited by Paul Klein-

dorfer and Jerry Wind, Wharton School Publishing, 367-382. Wharton School Publishing.

[5] Allen, F., E. Carletti, and R. Marquez, 2011. Credit Market Competition and Capital Regulation.

Review of Financial Studies 24, no. 4: 983-1018.

[6] Amel, D. and N. Liang, 1992. The relationship between entry into banking markets and changes in

legal restrictions on entry. The Antitrust Bulletin37, 631-49.

[7] Ayuso, J., D. Perez, and J. Saurina, 2004. Are capital buffers pro-cyclical? Evidence from Spanish

panel data. Journal of Financial Intermediation 13, 249-264.

[8] Banerjee, A. 1992. A Simple Model of Herd Behavior. Quarterly Journal of Economics, Vol.107,

No.3, pp. 797-817.

[9] Barrios, V. and J. Blanco, 2003. The Effectiveness of Bank Capital Adequacy Regulation: A Theo-

retical and Empirical Approach. Journal of Banking and Finance 27:1935—1958.

[10] Barron, J. M. and N. T. Valev, 2000. International Lending by U.S. Banks. Journal of Money, Credit,

and Banking, Vol.32, No.3, pp. 357-381.

[11] Barth, J., G. Caprio, and R. Levine, 2006. Rethinking Bank Regulation: Till Angels Govern. Cam-

bridge University Press, March 2006.

[12] Berger, A.N., R. DeYoung, M. Flannery, D. Lee, and O. Oztekin, 2008. How Do Large Banking

Organizations Manage Their Capital Ratios? Journal of Financial Services Research 34, 123-149.

[13] Berger, A.N. and C.H.S. Bouwman, 2010. Bank Liquidity Creation, Monetary Policy, and Financial

Crises. mimeo.

[14] Berger, A.N., R.J. Herring, and G.P. Szego, 1995. The Role of Capital in Financial Institutions.

Journal of Banking and Finance, 19, 393-430.

[15] Bernardo, A. E. and I. Welch, 2004. Liquidity and Financial Market Runs. Quarterly Journal of

Economics, Vol.119, No.1, pp. 135-158.

[16] Bikhchandani, S., D. Hirshleifer, and I. Welch. 1992. A Theory of Fads, Fashion, Custom, and

Cultural Change as Informational Cascades. Journal of Political Economy, Vol.100, No.5, pp. 992-

10.

[17] Blum, J. and M. Hellwig, 1995. The Macroeconomic Implications of Capital Adequacy Require-

ments for Banks. European Economic Review, 39, 739-749.


[18] Bolton, P. and X. Freixas, 2006. Corporate Finance and the Monetary Transmission Mechanism.

Review of Financial Studies, vol. 19(3), pages 829-870.

[19] Bonfim, D. and M. Kim, 2012. Liquidity Risk in Banking: Is There Herding? European Banking

Center Discussion Paper No. 2012-024.

[20] Brewer, E., G. Kaufman, and L. Wall. 2008 Bank capital ratios across countries: Why do they vary?

Journal of Financial Services Research34, 177-201.

[21] Brown, N. C., K. D. Wei, and R. R. Wermers. 2012. Analyst recommendations,

mutual fund herding, and overreaction in stock prices. Working paper, available at:

http://ssrn.com/abstract=1363837.

[22] Calomiris, C. and C.M. Kahn. 1991. The Role of Demandable Debt in Structuring Optimal Banking

Arrangements. American Economic Review 81:497-513.

[23] Choi, N. and R. W. Sias, 2009. Institutional Industry Herding. Journal of Financial Economics,

Vol.94, Is. 3, pp. 469-491.

[24] Cihak, M. and K. Schaeck, 2008. Banking Competition and Capital Ratios. Working paper, Inter-

national Monetary Fund, Washington.

[25] Dasgupta, A., A. Prat, and M. Verardo, 2011. Institutional Trade Persistence and Long-Term Equity

Returns. Journal of Finance, Volume 66, Issue 2, pages 635–653, April 2011.

[26] De Jonge, O. and O. Oztekin, 2012. How Do Banks Adjust Their Capital Structure? International

Evidence, 2012, mimeo.

[27] Dewatripont M. and J. Tirole, 1994. The Prudential Regulation of Banks. the MIT Press.

[28] Diamond, D., 1984. Financial Intermediation and Delegated Monitoring. Review of Economic

Studies 51, 393–414.

[29] Diamond, D. W. and P. Dybvig, 1983. Bank Runs, Deposit Insurance, and Liquidity. Journal of

Political Economy, Vol.91, No.3, pp. 401-419.

[30] Diamond, D. and R. Rajan, 2000. A Theory of Bank Capital. Journal of Finance, 55, 2431-2465.

[31] Fama, E. F. and K. R. French, 1993. Common Risk Factors in the Returns on Stocks and Bonds.

Journal of Financial Economics 33, no. 1: 3-56.

[32] Farhi, E. and J. Tirole, 2012. Collective Moral Hazard, Maturity Mismatch, and Systemic Bailouts.

American Economic Review, 102(1): 60–93.Federal Financial Institutions Examination Coun-

cil (FFIEC), 2008. A User’s Guide for the Uniform Bank Performance Report. available at

http://www.ffiec.gov/.

[33] Flannery, M. J., 1984. The social costs of unit banking restrictions. Journal of Monetary Eco-

nomics13, 237-49.

[34] Flannery, Mark J. 1994. Debt Maturity Structure and the Deadweight Cost of Leverage: Optimally

Financing Banking Firms. American Economic Review, pp. 320-331.

[35] Flannery, M. J. and K. P. Rangan, 2006. Partial adjustment toward target capital structures. Journal

of Financial Economics 79, 469-506.


[36] Flannery, M. J. and K. P. Rangan, 2008. What Caused the Bank Capital Build-Up of the 1990s?

Review of Finance, Vol. 12, Issue 2, pp. 391-429, 2008.

[37] Frank, M. and V. Goyal, 2009. Capital structure decisions: Which factors are reliably important?

Financial Management, Financial Management 38(1): 1-37.

[38] Freixas, X. 2010. Post-crisis challenges to bank regulation. Economic Policy 25(62): 375-399.

[39] Froot, K. A., P. G. J. O’Connell, and M. S. Seasholes, 2001. The Portfolio Flows of International

Investors. Journal of Financial Economics, Vol.59, Iss. 2, pp.151-193.

[40] Graham, J. and C. Harvey, 2001. The practice of corporate finance: Evidence from the field.

Journal of Financial Economics, 60(2/3): 186-243

[41] Gropp, R. and F. Heider, 2010. The Determinants of Bank Capital Structure. Review of Finance,

Vol. 14, pp. 587-622.

[42] Hirshleifer, D. A. and S. H. Teoh, 2009. Thought and Behavior Contagion in Capital Markets.

Handbook of Financial Markets: Dynamics and Evolution, 2009.

[43] Holmstrom, B. and J. Tirole, 1997. Financial Intermediation, Loanable Funds, and the Real Sector.

Quarterly Journal of Economics, 112, 663-691.

[44] Jacques, K., and P. Nigro, 1997. Risk-based capital, portfolio risk, and bank capital: A simultaneous

equations approach. Journal of Economics and Business 49, 533-547.

[45] Jain, A. K. and S. Gupta, 1987. Some Evidence on `Herding’ Behavior of U.S. Banks. Journal of

Money, Credit and Banking, Vol.19, No.1, pp. 78-89.Jokipii, T., and A. Milne, 2011. Bank capital

buffer and risk adjustment decisions. Journal of Financial Stability 7, 165-178.

[46] Jayaratne, J. and P. E. Strahan, 1998. Entry restrictions, industry evolution and dynamic efficiency:

Evidence from commercial banking. Journal of Law and Economics41, 239-74.

[47] Leary M. and M. Roberts, 2013. Do Peer Firms Affect Corporate Financial Policy? Journal of

Finance, forthcoming.

[48] Lemmon, M. L., M. R. Roberts, and J. Zender, 2008. Back to the Beginning: Persistence and the

Cross-Section of Corporate Capital Structure. Journal of Finance 63(4): 1575-1608.

[49] Manski, C. F. 1993. Identification of Endogenous Social Effects: The Reflection Problem. The

Review of Economic Studies 60(3): 531-542.

[50] Manski, C. F. 2000. Economic Analysis of Social Interactions. The Journal of Economic Perspec-

tives 14(3): 115-136.

[51] Jensen, M. C. and W. H. Meckling, 1976. Theory of the Firm: Managerial Behavior, Agency

Costs and Ownership Structure. Journal of Financial Economics, Elsevier, vol. 3(4), pages 305-

360, October.

[52] Mehran, H. and A. Thakor, 2011. Bank Capital and Value in the Cross-Section, Review of Financial

Studies 24 (4):1019-1067.

[53] Memmel, C., and P. Raupach, 2010. How do banks adjust their capital ratios? Journal of Financial

Intermediation 19, 509-528.


[54] Merton, R. C. 1977. An Analytic Derivation of the Cost of Deposit Insurance Loan Guarantees.

Journal of Banking and Finance 2(1), 3–11.

[55] Modigliani, F. and M. Miller, 1958. The cost of capital, corporation finance and the theory of

investment, American Economic Review 48, 261-297.

[56] Myers, S. and N. Majluf, 1984. Corporate financing and investment decisions when firms have

information that investors do not have. Journal of Financial Economics 13, 187-221.

[57] Myers, S. C. and R. G. Rajan, 1998. The Paradox of Liquidity. Quarterly Journal of Economics,

113(3), pp. 733-771.

[58] Peura, S. and J. Keppo, 2006. Optimal Bank Capital with Costly Recapitalization. The Journal of

Business 79(4):2163-2202.

[59] Rajan, R. G. 1995. Why Bank Credit Policies Fluctuate: A Theory and Some Evidence. The Quar-

terly Journal of Economics 109(2): 399-441.

[60] Rajan, R. and L. Zingales, 1995. What do we know about capital structure? Some evidence from

international data. Journal of Finance 50, 1421-1460.

[61] Ratnovski, L. 2009. Bank liquidity regulation and the lender of last resort. Journal of Financial

Intermediation 18(4): 541-558.

[62] Rime, B., 2001. Capital requirements and bank behaviour: Empirical evidence for Switzerland,

Journal of Banking and Finance 25, 789-805.

[63] Sias, R. W. 2004. Institutional Herding. Review of Financial Studies, Vol.17, No.1, pp. 165-206.

[64] Scharfstein, D. S. and J. C. Stein, 1990. Herd Behavior and Investment. American Economic Re-

view, Vol.80, No.3, pp. 465-479.

[65] Stock, J.H. and M. Yogo, 2005. Testing for Weak Instruments in Linear IV Regression. In D.W.K.

Andrews and J.H. Stock, eds. Identification and Inference for Econometric Models: Essays in

Honor of Thomas Rothenberg. Cambridge:Cambridge U y Press, 2005, pp. 80-108.

[66] Strahan, P. E. 2008. Bank structure and lending: what we do and do not know, in Handbook of

Financial Intermediation and Banking, edited by Arnoud Boot and Anjan Thakor, Elsevier.

[67] Uchida, H. and R. Nakagawa, 2007. Herd Behavior in the Japanese Loan Market: Evidence from

Bank Panel Data. Journal of Financial Intermediation, Vol.16, No.4, pp. 555-583.

[68] Van den Heuvel, S. 2008. The Welfare Cost of Bank Capital Requirements. Journal of Monetary

Economics 55:298-320.

[69] Wagner, W. 2011. Systemic Liquidation Risk and the Diversity-Diversification Trade-Off. Journal

of Finance 66(4): 1141-1175.

[70] Wermers, R. 1999. Mutual Fund Herding and the Impact on Stock Prices. Journal of Finance,

Vol.54, No. 2, pp. 581-622.

[71] Wooldridge, J.M. 2010. Econometric Analysis of Cross Section and Panel Data. Cambridge, MA:

MIT Press.

[72] Zwiebel, J. 1995. Corporate Conservatism and Relative Compensation. Journal of Political Econ-

omy 103(1): 1-25.


Name Defintion

Book leverage 1- Total equity / Total book assets.

Book equity ratio Total equity / Total book assets.

Tier 1 ratio Total equity / Total risk weighted assets.

Non-deposit liabilities Book leverage - Total deposits.

PayoutCash Dividends Declared on Common Stock / Income Before Extraordinary

Items.

Market-to-Book Ratio Market value of assets / Total book assets.

Profits (Pre-tax profit + interest expenses) / book value of assets.

ln(Total Assets) Natural log of total bank assets.

Collateral(Total Investment Securities+Property&Buildings+Cash & Due from other

banks) / Total book assets.

DividendsA categorical dummy variable that takes the value of 1 if the bank pays a

dividend in a given year, and the value of 0 otherwise.

Payout Dividends / Total earnings.

Asset risk Loan loss provisions / Total assets.

Instrument Average peer equity shock excluding the equity shock of bank i.

CrisisA categorical dummy variable that takes the value of 1 in years with a bank or

stock crisis defined above, and the value of 0 otherwise.

Bank CrisisA categorical dummy variable that takes the value of 1 in years 1990, 1992,

1991, 2007, 2008, 2009, and the value of 0 otherwise.

Stock crisisA categorical dummy variable that takes the value of 1 in years 1987, 1998,

2000, 2001, 2002, and the value of 0 otherwise.

RecessionA categorical dummy variable that takes the value of 1 in years 1980, 1981,

1982, 1990, 1991, 2001, 2007, 2008, 2009, and the value of 0 otherwise.

Basel I IntroductionA categorical dummy variable that takes the value of 1 starting year 1991, and

the value of 0 otherwise.

DeregulationA categorical dummy variable that takes the value of 1 starting the year when

a given state has introduced deregulation, and the value of 0 otherwise.

Appendix A: Definition variables

This table provides an overview of the key variables' definitions. The sample consists of all regional and

diversified publicly listed commercial banks on the intersection of the Compustat and CRSP databases

between 1971 and 2010 with nonmissing data for all analysis variables.


Diversified Banks

Regional Banks

Appendix B: GICS definitions of diversified and regional banks

Commercial banks whose businesses are derived primarily from

commercial lending operations and have significant business

activity in retail banking and small and medium corporate lending.

Excludes banks classified in the Regional Banks and Thrifts &

Mortgage Finance sub-industries. Also excludes investment banks

classified in the Investment Banking & Brokerage Sub-Industry.

Commercial banks whose businesses are derived primarily from

commercial lending operations and have significant business

activity in retail banking and small and medium corporate lending.

Regional banks tend to operate in limited geographic regions.Excludes companies classified in the Diversified Banks sub-

industry.

This table shows the definions of a diversified and a regional bank according to the MSCI's

Global Industry Classification Standard (GICS), available online at msci.org.


Panel A: Summary statistics of peers book capital ratio

Peer

groupN Mean Median St. Dev. Min. Max.

1 2,080 0.086 0.09 0.011 0.043 0.103

2 1,831 0.08 0.083 0.011 0.049 0.099

3 3,802 0.065 0.06 0.011 0.045 0.102

4 4,627 0.06 0.057 0.012 0.039 0.106

1 6,511 0.094 0.095 0.005 0.054 0.105

2 7,292 0.085 0.087 0.006 0.05 0.101

3 5,727 0.082 0.083 0.008 0.052 0.1

4 4,871 0.085 0.084 0.01 0.052 0.105

Total 36,741 0.081 0.085 0.014 0.039 0.106

Panel B: Summary statistics of peers market capital ratio

Peer

groupN Mean Median St. Dev. Min. Max.

1 2,080 0.119 0.115 0.032 0.035 0.201

2 1,831 0.11 0.12 0.05 0.036 0.218

3 3,802 0.074 0.06 0.039 0.034 0.232

4 4,627 0.082 0.068 0.044 0.029 0.225

1 6,511 0.132 0.135 0.031 0.042 0.212

2 7,292 0.126 0.135 0.038 0.036 0.212

3 5,727 0.126 0.135 0.042 0.036 0.221

4 4,871 0.142 0.153 0.039 0.04 0.231Total 36,741 0.117 0.122 0.045 0.029 0.232

The sample consists of all publicly listed commercial banks on the intersection of the Compustat

and CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables.

Appendix A lists variable definitions and data sources for the book capital and market capital

ratio, and Appendix B lists definitions of a diversified and regional bank. The table presents the

distribution of book capital of diversified banks in the cross-section. Banks are allocated in four

peer groups according to the quartiles of the asset distributions.

Appendix C: Summary statistics on peer groups

Table C1: Peer group summary statistics

Diversified banks

Regional banks

Diversified banks

Regional banks


market view buffer view

Market-to-book ratio + -

Profits + -

Log (Size) - +/-

Collateral - 0

Dividends + -

Risk + +

Appendix D: Predicted effects under the market and buffer view

This table replicates Table 4 from Gropp and Heider (2010). It shows the predicted effects of

control variables under the market and buffer view of bank capital structure.

Predicted effects


Variable Name N Mean Median Standard Deviation Minimum Maximum

Book capital ratio 36,741 0.081 0.079 0.024 0.029 0.173Market capital ratio 36,741 0.116 0.112 0.061 0.012 0.294Non-deposit liabilities 22,233 0.150 0.134 0.097 0.006 0.433Tier 1 capital 22,233 0.119 0.114 0.032 0.056 0.251Excess capital 22,233 0.059 0.054 0.032 -0.004 0.191Payout 22,233 0.311 0.310 0.261 0 0.920Market-to-Book Ratio 36,741 1.042 1.031 0.063 0.932 1.263Profits 36,741 0.011 0.011 0.005 -0.009 0.030Total Assets (in bln.) 36,741 8.814 1.774 37.257 0.027 1309.639Collateral 36,741 0.295 0.292 0.106 0.075 0.605Dividends 36,741 0.833 1.000 0.373 0 1.000Asset risk 36,741 0.010 0.009 0.005 0 0.110Bank equity shock 36,741 0.029 0.029 0.163 -0.471 0.501

Peer bank averages

Book capital ratio 36,741 0.081 0.085 0.014 0.039 0.106

Market capital ratio 36,741 0.117 0.122 0.045 0.029 0.232

Non-deposit liabilities 22,233 0.149 0.143 0.050 0.022 0.400

Tier 1 22,233 0.119 0.119 0.011 0.074 0.177

Excess capital 22,233 0.059 0.059 0.011 0.012 0.117

Payout 22,233 0.308 0.308 0.080 0.000 0.920

Market-to-Book Ratio 36,741 0.117 0.122 0.045 0.029 0.232

Profits 36,741 0.011 0.011 0.004 -0.003 0.029

Total Assets (in bln.) 36,741 7.337 2.517 12.156 0.160 121.689

Collateral 36,741 0.295 0.293 0.045 0.138 0.522

Asset risk 36,741 0.002 0.006 0.018 -0.632 0.016

Instrument 36,741 0.030 0.034 0.100 -0.470 0.362

Total number of banks 909Average quarterly number of banks

in a group40

Table 1: Summary statistics

The sample consists of all regional and diversified publicly listed commercial banks on the intersection of the Compustat and

CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A lists variable definitions

and data sources. The table presents summary statistics after all variables have been winsorized at the 1 percent level on both

tails of the distribution, except for the variable bank size which is reported before being winsorized and log-transformed. The

peer bank variables represent the average value for all banks in a given peer group, excluding the i observation.


Dep. VariableInstrument

Instrument

(1 Period Lead Ind. Variables)

1 2

Market-to-Book Ratio 0.003 -0.002 0.005 0.005Profits -0.002 -0.004 0.006 0.006ln(Total Assets) -0.002 -0.012 0.015 0.014Collateral 0.005 0.002 0.004 0.005Dividends 0.026** 0.010

0.012 0.012Asset risk 0.012*** 0.010**

0.004 0.004Peers variables YES YESObservations 36,741 35,450Number of banks 909 907Bank FE YES YESYear FE YES YES

Table 2: Test on the properties of the average equity shock of peer banks as an instrument

The sample consists of all regional and diversified publicly listed commercial banks on the intersection of the

Compustat and CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A

lists variable definitions and data sources. The table presents estimated coeffcients of the instrument (the average

idiosyncratic stock component of peer banks) on the set of control variables that enter in the 2SLS models. All

variables are scaled by the corresponding variable's standard deviation with robust standard errors clustered at the

bank level. The dependent variable is specified in the header of each column. All independent variables are in levels.

Bank specific factors denote variables corresponding to bank i's value in quarter t-1. Significance levels notation: ***

p<0.01, ** p<0.05, * p<0.1.


Dep. Variable Book ratio Market ratio Non-deposit liabilities Tier 1 ratio

1 2 3 4

Market-to-Book Ratio -0.028 0.590*** -0.067*** -0.027

0.024 0.011 0.018 0.023

Profits 0.120*** 0.087*** 0.074*** 0.119***

0.018 0.011 0.016 0.025

ln(Total Assets) -0.258*** -0.167*** 0.582*** -0.300***

0.076 0.037 0.080 0.107

Collateral -0.057** -0.010 0.098*** 0.308***

0.023 0.010 0.026 0.032

Dividends 0.204*** 0.106*** -0.032 0.164***

0.043 0.020 0.039 0.055

Asset risk -0.115*** -0.063*** -0.009 -0.033

0.023 0.010 0.015 0.029

Constant -0.881*** -0.434*** -0.971*** -0.106

0.171 0.093 0.277 0.113

Observations 36,741 36,741 36,741 22,233

Adj. R² 0.189 0.790 0.277 0.156

Number of banks 909 909 909 805

Bank FE YES YES YES YES

Year FE YES YES YES YES

Table 3: A basic model of bank capital structure


Compustat and CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix

A lists variable definitions and data sources. The table presents estimated coeffcients scaled by the corresponding

variable's standard deviation with robust standard errors clustered at the bank level. The dependent variable is

specified in the header of each column. All independent variables are in levels. Bank specific factors denote

variables corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01, ** p<0.05, *

p<0.1.



1 2 3 4

Peer average of the dependent

variable 0.339*** 0.340*** 0.078** 0.166***

0.035 0.036 0.036 0.028


0.024 0.011 0.018 0.024

Profits 0.119*** 0.088*** 0.077*** 0.117***

0.018 0.011 0.016 0.025

ln(Total Assets) -0.253*** -0.161*** 0.613*** -0.296***

0.076 0.038 0.084 0.109

Collateral -0.057*** -0.010 0.103*** 0.307***

0.022 0.010 0.025 0.032

Dividends 0.184*** 0.101*** -0.026 0.144***

0.041 0.020 0.038 0.054

Asset risk -0.118*** -0.064*** -0.010 -0.038

0.021 0.010 0.015 0.028

Peers variables

Market-to-Book Ratio -0.048 -0.267*** 0.131*** -0.037

0.040 0.031 0.041 0.046

Profits -0.119*** -0.040** -0.115*** -0.156***

0.031 0.016 0.035 0.038

ln(Total Assets) 0.153*** 0.033* -0.139*** 0.182***

0.038 0.017 0.048 0.056

Collateral 0.001 -0.011 -0.034 -0.102**

0.030 0.013 0.031 0.044

Asset risk 0.011 0.004 0.001 -0.010

0.011 0.005 0.014 0.014

Constant -0.284 -0.269*** -0.715** -0.265*

0.184 0.093 0.289 0.145

Observations 36,741 36,741 36,741 22,233

Adj. R² 0.223 0.794 0.282 0.170




Table 4: Panel A- Reduced form regressions

The sample consists of all regional and diversified publicly listed commercial banks on the intersection

of the Compustat and CRSP databases between 1971 and 2010 with nonmissing data for all analysis

variables. Appendix A lists variable definitions and data sources. The table presents estimated coeffcients

scaled by the corresponding variable's standard deviation with robust standard errors clustered at the

bank level. The dependent variable is specified in the header of each column. All independent variables

are in levels. Bank specific factors denote variables corresponding to bank i's value in quarter t-1.

Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1.


Dep. Variable Book ratio Market ratio

1 2

Peer average 0.027*** 0.013

0.006 0.016

Lagged dependent variable 0.909*** 0.875***

0.005 0.013

First Stage F statistics 101.8 199.5

Bank FE YES YES

Time FE YES YES

Table 4: Panel B - Lagged dependent variable inclusion

The sample consists of all regional and diversified publicly listed commercial banks on the

intersection of the Compustat and CRSP databases between 1971 and 2010 with

nonmissing data for all analysis variables. Appendix A lists variable definitions and data

sources. The table presents estimates of the model in Panel A of Table 5 when the lagged

dependent variable is added. The dependent variable is specified in the header of each

column. All independent variables are in levels. Bank specific factors denote variables

corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01, **

p<0.05, * p<0.1.



1 2 3 4

Peer average of the dependent variable 0.547*** 0.396*** 0.605*** 0.208***

0.122 0.114 0.152 0.064


0.024 0.011 0.018 0.024

Profits 0.118*** 0.088*** 0.080*** 0.117***

0.018 0.011 0.016 0.025

ln(Total Assets) -0.238*** -0.161*** 0.549*** -0.289***

0.077 0.038 0.087 0.11

Collateral -0.055** -0.01 0.112*** 0.307***

0.021 0.01 0.027 0.032

Dividends 0.176*** 0.100*** -0.019 0.139**

0.041 0.02 0.038 0.054

Asset risk -0.119*** -0.064*** -0.009 -0.039

0.021 0.01 0.015 0.028

Bank Peer specific factors

Market-to-Book Ratio -0.008 -0.302*** 0.047 -0.028

0.042 0.075 0.043 0.043

Profits -0.112*** -0.039** -0.098*** -0.156***

0.032 0.016 0.035 0.038

ln(Total Assets) 0.192*** 0.036** -0.482*** 0.209***

0.044 0.017 0.113 0.073

Collateral -0.012 -0.012 -0.025 -0.120**

0.032 0.013 0.036 0.052

Asset risk 0.019 0.005 0.026 -0.01

0.012 0.005 0.017 0.014

Observations 36,741 36,741 36,741 22,233

R² 0.189 0.788 0.192 0.137

Number of gvkey 909 909 909 805



First Stage Instrument 0.053*** 0.339*** -0.0448*** 0.129***

0.0053 0.002 0.004 0.006

First Stage F statistics 98.61*** 201.8*** 135.1*** 406.6***

Table 5. Panel A: 2SLS Regressions


Compustat and CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix

A lists variable definitions and data sources. The table presents estimated coeffcients of the baseline 2SLS model.

All variables are scaled by the corresponding variable's standard deviation. The dependent variable is specified in

the header of each column. All independent variables are in levels. Bank specific factors denote variables

corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1.



1 2 3 4

Peer average 0.309* 0.13 0.459*** 0.234**

0.173 0.173 0.172 0.105

First Stage F statistics 67 142.4 63.58 174.4


Time FE YES YES YES YESObservations 17111 17111 17111 10131


1 2 3 4

Peer average 0.739*** 0.609*** 0.819*** 0.178**

0.16 0.175 0.264 0.089

First Stage F statistics 23.43 65.28 65.01 141.4Bank FE YES YES YES YES

Time FE YES YES YES YESObservations 19,565 19,565 19,565 12,033

Table 5. Panels B & C: 2SLS Regressions: Deviations from the peer average


intersection of the Compustat and CRSP databases between 1971 and 2010 with nonmissing data

for all analysis variables. Appendix A lists variable definitions and data sources. The tables

presents estimated coeffcients of the baseline model on two subsamples. The subsample in Panel

B contains all observations for banks, which bank capital is above the peer average level. The

subsample in Panel C contains all observations for banks, which bank capital is below the peer

average level. All variables are scaled by the corresponding variable's standard deviation with

robust standard errors clustered at the bank level. The dependent variable is specified in the

header of each column. All independent variables are in levels. Bank specific factors denote

variables corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01,

** p<0.05, * p<0.1.

Panel B: Sub-sample of banks above the peer average level of capital and non-deposit debt

Panel C: Sub-sample of banks below the peer average level of capital and non-deposit debt


Mean Minimum Maximum

Book ratio

First Stage F statistics 1.07 0 10.163

Market ratioFirst Stage F statistics 1.37 0 8.948


intersection of the Compustat and CRSP databases between 1971 and 2010 with nonmissing data for

all analysis variables. Appendix A lists variable definitions and data sources. The table presents the

first-stage F-statistics estimated by a simulation of the peer groups' construction. Banks are randomly

assigned to one of the 8 groups. The simulation exercise is run 240 times. Significance levels notation:

*** p<0.01, ** p<0.05, * p<0.1.

Table 5. Panel D: Placebo tests


Dep. Variable Book ratio Market ratio

1 2

Peer average 0.429*** 0.425***

0.129 0.130

First Stage F statistics 107.4 202.2

Bank FE YES YES

Time FE YES YESObservations 35,509 35,509

The sample consists of all regional and diversified publicly listed commercial banks on

the intersection of the Compustat and CRSP databases between 1971 and 2010 with

nonmissing data for all analysis variables. Appendix A lists variable definitions and data

sources. The instrument is estimated by a Fama-French model including also the

unweighted average return on the banking sector excluding bank i in a given period. The

table presents estimates of the baseline model under the new Fama-French model for the

instrument. All variables are scaled by the corresponding variable's standard deviation

with robust standard errors clustered at the bank level. The dependent variable is

specified in the header of each column. All independent variables are in levels. Bank

specific factors denote variables corresponding to bank i's value in quarter t-1.

Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1.

Table 5. Panel E: 2SLS Regressions: Robustness check


Panel A: Banks close to the Tier 1 Regulatory threshold


1 2 3 4

Peer average * Close 0.347 0.182** -3.097 0.271

0.494 0.090 5.342 0.417

Close -0.788*** -0.326*** -0.363 -1.104***

0.168 0.038 0.659 0.099

Observations 23,961 23,961 23,961 21,190




First Stage F statistics 102.3 217.6 0.181 2.326

Panel B: Banks close to the Market Capital Adequacy Requirement


1 2 3 4

Peer average * Close 80s 0.093 0.121* 0.171 0.385

0.309 0.071 0.379 0.253

Close 80s -0.871*** -0.343*** 0.155 -0.584***

0.121 0.029 0.109 0.107

Observations 23,961 23,961 23,961 21,190





Panel C: The 2SLS baseline model over same sample period


1 2 3 4

Peer average 0.650*** 0.254** 0.502*** 0.223***

0.121 0.119 0.109 0.063

Observations 23,961 23,961 23,961 21,190





Table 6. 2SLS Regressions with banks close to the minimum regulatory and market requirements

The sample consists of all regional and diversified publicly listed commercial banks on the intersection of the Compustat and CRSP

databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A lists variable definitions and data

sources. All variables are scaled by the corresponding variable's standard deviation with robust standard errors clustered at the banklevel. The dependent variable is specified in the header of each column. All independent variables are in levels. Bank specific factors

denote variables corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1. The tablesummarizes the main findings of regressions assessing how peer effects across banks vary in intensity depending on how close they

are to the regulatory minimum (6 percent).


Panel A: Peer effects over crisis periods

Dep. Variable Book ratio Market ratio Nondeposit Liabilities Tier 1 Capital

1 2 3 4

Peer average * Crisis -0.042 -0.530*** 2.775 -0.169

1.577 0.161 6.831 0.163

Crisis 1.143 0.660*** 0.684 -0.608***

0.920 0.113 0.702 0.086



Time FE YES YES YES YES

Observations 36,741 36,741 36,741 22,233

Panel B: Peer effects over bank crisis periods


1 2 3 4

Peer average * Bank Crisis 0.274 -0.495*** -0.188*** 2.118

0.349 0.103 0.070 1.635

Bank Crisis 0.920** 0.145 1.378*** -1.040

0.457 0.128 0.346 0.823




Observations 36,741 36,741 36,741 22,233

Panel C: Peer effects over stock crisis periods


1 2 3 4

Peer average * Stock Crisis 25.137 0.521 -0.033 0.227

386.676 0.546 0.106 0.228

Stock Crisis -17.122 0.088 0.299** -0.088

265.562 0.076 0.117 0.161




Observations 36,741 36,741 36,741 22,233

Table 7. 2SLS Regressions with periods of bank and stock market crises

The sample consists of all regional and diversified publicly listed commercial banks on the intersection of the Compustat and CRSP databases

between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A lists variable definitions and data sources. All variables

are scaled by the corresponding variable's standard deviation with robust standard errors clustered at the bank level. The dependent variable is

specified in the header of each column. All independent variables are in levels. Bank specific factors denote variables corresponding to bank i's

value in quarter t-1.Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1. The table summarizes the main findings of regressions

assessing how peer effects across banks vary in intensity with periods of facing different shocks: crisis, bank crisis, stock crisis.


Panel A: Peer effects following deregulation

Dep. VariableBook

ratio

Market

ratio

Nondeposit

Liabilities

Tier 1

Capital

1 2 3 4

Peer average * Deregulation 0.488*** 0.157** 0.822*** 0.207***

0.139 0.070 0.223 0.063

Deregulation 0.671*** 0.203** -0.505*** -0.691***

0.209 0.086 0.144 0.157




Observations 36,741 36,741 36,741 22,233

Panel B: Peer effects following recessions

Dep. VariableBook

ratio

Market

ratio

Nondeposit

Liabilities

Tier 1

Capital

1 2 3 4

Peer average * Recession 0.345* -0.286*** -0.078 0.149

0.203 0.061 0.064 0.127

Recession 0.744*** -0.153 1.083*** -0.147

0.262 0.144 0.335 0.225




Observations 36,741 36,741 36,741 22,233



variables. Appendix A lists variable definitions and data sources. All variables are scaled by the

corresponding variable's standard deviation with robust standard errors clustered at the bank level. The

dependent variable is specified in the header of each column. All independent variables are in levels.

Bank specific factors denote variables corresponding to bank i's value in quarter t-1. Significance levels

notation: *** p<0.01, ** p<0.05, * p<0.1. The table summarizes the main findings of regressions

assessing how peer effects across banks vary in intensity with periods of facing different shocks: branch

interstate deregulation or economic recession.

Table 8. 2SLS Regressions with periods of interstate deregulation and recessions


Panel A: Peer effects after the Gramm–Leach–Bliley Act in 1999

Dep. VariableBook

ratio

Market

ratio

Nondeposit

Liabilities

Tier 1

Capital

1 2 3 4

Peer average * GLB -0.214 -0.380*** -0.796* -0.008

1.243 0.129 0.409 0.170

GLB -0.573 -0.071 -0.665 -0.088

2.456 0.206 0.741 0.343

First Stage F statistics 0.345 39.31 2.508 8.291Bank FE YES YES YES YES

Time FE YES YES YES YESObservations 36,741 36,741 36,741 22,233

Panel B: Peer effects following intrastate branch deregulation in 1994

Dep. VariableBook

ratio

Market

ratio

Nondeposit

Liabilities

Tier 1

Capital

1 2 3 4

Peer average * Intrastate 0.345* -0.286*** -0.078 0.149

0.203 0.061 0.064 0.127

Intrastate 0.744*** -0.153 1.083*** -0.147

0.262 0.144 0.335 0.225




Observations 36,741 36,741 36,741 22,233

Table 9. 2SLS Regressions with periods of intrastate deregulation and post-GLB period



variables. Appendix A lists variable definitions and data sources. GLB (Gramm–Leach–Bliley Act) is a

dummy variable that takes the value of one with the introduction of the GLB Act in 1999, and zero

otherwise. Intrastate is a dummy variable that takes the value of one following the intrastate branch

deregulation act in 1994, and zero otherwise. All variables scaled by the corresponding variable's

standard deviation with robust standard errors clustered at the bank level. The dependent variable is

specified in the header of each column. All independent variables are in levels. Bank specific factors

denote variables corresponding to bank i's value in quarter t-1.Significance levels notation: *** p<0.01,

** p<0.05, * p<0.1. The table summarizes the main findings of regressions assessing how peer effects

across banks vary in intensity with periods of facing different shocks: branch intrastate deregulation or

the repeal of parts of the Glass-Steagal Act.



1 2 3 4

Peer average of the dependent variable 0.396*** 0.246* 0.500*** 0.236**

0.136 0.139 0.122 0.115

Market-to-Book Ratio -0.014 0.581*** -0.070*** 0.009

0.028 0.014 0.020 0.029

Profits 0.085*** 0.070*** 0.065*** 0.079**

0.025 0.016 0.018 0.033

ln(Total Assets) -0.297*** -0.180*** 0.468*** -0.378**

0.102 0.052 0.097 0.148

Collateral -0.061** -0.010 0.123*** 0.309***

0.028 0.013 0.028 0.037

Dividends 0.128** 0.081*** -0.023 0.053

0.055 0.027 0.046 0.073

Asset risk -0.091*** -0.047*** -0.001 0.010

0.028 0.013 0.018 0.037


Market-to-Book Ratio -0.119*** -0.183** 0.110*** -0.122**

0.045 0.089 0.029 0.048

Profits -0.141*** -0.043*** 0.035 -0.171***

0.031 0.016 0.031 0.045

ln(Total Assets) 0.126** -0.007 -0.422*** 0.147**

0.057 0.020 0.103 0.074

Collateral -0.028 -0.023** 0.060** -0.110**

0.025 0.011 0.027 0.049

Asset risk 0.002 -0.003 0.067*** 0.027

0.036 0.018 0.022 0.057

Observations 24,634 24,634 24,634 14,812

R² 0.113 0.765 0.272 0.114





Table 10. 2SLS Regressions of followers and leaders across banks


Compustat and CRSP databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A

lists variable definitions and data sources. All variables are scaled by the corresponding variable's standard deviation

with robust standard errors clustered at the bank level. The dependent variable is specified in the header of each

column. All independent variables are in levels. Bank specific factors denote variables corresponding to bank i's value

in quarter t-1. Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1. This table asks the question if there are

followers and leaders among banks. The capital structure variables of peer follower banks is regressed on the peer

average of leader banks in their groups. In this table, a leader bank is classified as a bank with a share of deposits

above the 75th percentile in its peer group.



1 2 3 4

Peer average of the dependent variable 0.593** 0.258** 0.549** 0.085

0.299 0.130 0.255 0.325


0.026 0.012 0.019 0.027

Profits 0.122*** 0.090*** 0.069*** 0.125***

0.020 0.012 0.015 0.026

ln(Total Assets) -0.284*** -0.201*** 0.495*** -0.346***

0.095 0.043 0.094 0.123

Collateral -0.043* -0.005 0.132*** 0.303***

0.025 0.011 0.027 0.032

Dividends 0.148*** 0.093*** -0.025 0.159***

0.047 0.022 0.040 0.057

Asset risk -0.103*** -0.056*** -0.036** -0.021

0.023 0.011 0.017 0.031


Market-to-Book Ratio -0.012 -0.149 0.101** -0.008

0.025 0.095 0.042 0.027

Profits -0.088* -0.001 -0.106*** -0.078*

0.045 0.013 0.041 0.041

ln(Total Assets) 0.191** 0.004 -0.352** 0.086

0.079 0.019 0.169 0.172

Collateral 0.014 0.001 -0.048** -0.015

0.023 0.009 0.024 0.130

Asset risk 0.078*** 0.025*** -0.007 0.036

0.015 0.006 0.014 0.030

Observations 31,805 31,805 31,805 20,284

R² 0.0688 0.770 0.148 0.135





Table 11. 2SLS Regressions - Geographical location by Federal Reserve district location

The sample consists of all regional and diversified publicly listed commercial banks on the intersection of the Compustat and CRSP

databases between 1971 and 2010 with nonmissing data for all analysis variables. Appendix A lists variable definitions and data

sources. All variables are scaled by the corresponding variable's standard deviation with robust standard errors clustered at the bank

level. The dependent variable is specified in the header of each column. All independent variables are in levels. Bank specific

factors denote variables corresponding to bank i's value in quarter t-1. Significance levels notation: *** p<0.01, ** p<0.05, * p<0.1.

Peer averages in these models, excluding the bank i observation, are formed according to bank location in one of the 12 districts of

Federal Reserve banks in the USA. Within each district, banks are assigned according to their focus of activities into a group of

diversifed or regional banks. Next, within each of these groups, a bank is assigned to one of 4 groups according to its size as

measured by bank assets. The four size groups correspond to the quartiles of the bank asset distribution.

Chapter 2

Supervising Cross-Border Banks: Theory, Evidence and Policy

Abstract: This paper analyzes the distortions that banks’ cross-border activities, such as foreign assets,

deposits and equity, can introduce into regulatory interventions. We find that while each individual

dimension of cross-border activities distorts the incentives of a domestic regulator, a balanced amount of

cross-border activities does not necessarily cause inefficiencies, as the various distortions can offset each

other. Empirical analysis using bank-level data from the recent crisis provide support to our theoretical

findings. Specifically, banks with a higher share of foreign deposits and assets and a lower foreign equity

share were intervened at a more fragile state, reflecting the distorted incentives of national regulators.

We discuss several implications for the supervision of cross-border banks in Europe.

2.1 Introduction

The problematic resolution of failing cross-border banks in Europe during the current crisis has focused

academics’ and policy makers’ attention alike on the misalignment of geographic boundaries of banks

and their supervision. The resolution of Fortis on the national level, undertaken separately by Dutch,

Belgian and Luxembourg authorities has confirmed Charles Goodhart’s and Mervyn King’s point that

"banks are international in life and national in death." The failure of the Icelandic banks, with wide-

ranging economic and political repercussions has shed doubts on the viability of large multinational

banks in small countries. The recent reform debate has - among other items - focused on issues of

national versus supra-national supervision, the responsibility, obligation and capacity of home country

supervisor relative to host country supervisor to resolve large cross-border banks and, in general, the

need to coordinate the resolution of large international banks across borders. On the political level,

arguments over national sovereignty and the role of European institutions are being used to argue in

favor or against the establishment of a European-level bank supervisory authority. But what are the

distortions of national supervision of international banks? What are the rationales behind national and

supra-national supervisors; what are the trade-offs of national versus supra-national resolution authority?

54

Supervising Cross-Border Banks: Theory, Evidence and Policy 55

This paper first presents a simple theoretical model that demonstrates the distorted incentives that

national supervisors face when deciding to intervene in failing banks with activity outside their borders.

Specifically, we show that national supervisors’ incentives to intervene in a timely manner into a weak

bank increase in the foreign equity share and decrease in the share of foreign deposits and assets. The

intuition for this result is, chiefly, that the gains from letting a weak bank continue mainly accrue to

equity, while the costs accrue to debt holders and other stakeholders in the economy. The result is

robust to variations in the utility function of the regulator, endogenizing risk choice by banks and type of

intervention (bank closure or bailout). Second, we provide empirical evidence consistent with the model

using a sample of intervened banks during the crisis of 2007-2009. Taking their CDS spread at the time

of intervention as a measure of regulatory lenience, we find that higher foreign asset and deposit shares

and a lower foreign equity share are associated with more lenient regulatory decisions. These findings

are robust to including an array of bank-level and country-level control variables, testing for anticipation

effects and controlling for selection bias. We link our theoretical and empirical analysis to a discussion

of the current regulatory arrangements for cross-border banking in Europe and recent reform discussions.

Our paper is related to a small but growing theoretical literature on the regulation of cross-border

banks.1 Loranth and Morrison (2007) discuss the implications of capital requirements and deposit in-

surance for cross-border banks and show that capital requirements set at a level to off-set the safety

net subsidy of the deposit insurance result in too little risk-taking in the case of multinational banks.

Dell’Arricia and Marquez (2006) show that competition between national regulators can lead to lower

capital adequacy standards, since national regulators do not take into account the external benefits of

higher capital adequacy standards in terms of higher stability in other countries. A supra-national central

regulator is therefore more likely to emerge among more homogeneous countries if it sets regulatory stan-

dards higher than those of the country with the highest individual standards. Acharya (2003), however,

shows that coordinating capital adequacy ratios across countries without coordinating on other dimen-

sions of the regulatory framework, such as resolution policies, can have detrimental effects. Freixas

(2003) and Goodhart and Schoenmaker (2009) show that ex-post negotiations on recapitalization of fail-

ing cross-border banks can lead to underprovision of the necessary resources and prove the advantage

of ex-ante burden sharing agreements in helping overcome coordination problems between regulators.

Holthausen and Ronde (2002) consider cooperation between home and host country supervisor on the

1For an early discussion, see White (1994).


intervention decision for a multinational bank. Given that national regulators represent national interests,

a misalignment of interests leads to suboptimal exchange of information and distorted intervention deci-

sions. Our paper is most closely related to Calzolari and Loranth (2011) who show how the organization

structure of multi-national banks can influence regulatory behavior. Specifically, organization of foreign

presence through branches leads to higher incentives to intervene as the home country regulator can draw

on all assets, while at the same time it reduces the incentives if the regulator is responsible for repaying

all deposits, including in foreign branches. There is also a more institutionally oriented literature on

legal differences across countries in the treatment of domestic and foreign creditors (e.g. Krimminger,

2007). Osterloo and Schoenmaker (2007) and Schoenmaker (2010) discuss the importance of regulation

of cross-border banks within Europe. Allen et al. (2011) discuss policy options for the regulation of

cross-border banks in the European Union. The contribution of our paper is to show that different di-

mensions of cross-border activities can result in different distortions of national regulators when deciding

to intervene in a weak cross-border bank. Unlike previous work, we therefore distinguish specifically

between the equity, deposit and asset dimensions of cross-border bank activities and show their different

impact on intervention decisions by national regulators. Unlike previous papers, we explicitly test our

theoretical predictions, using data from the recent crisis. Finally, we embed both theoretical model and

the empirical findings into the current discussion on the revamping of the European financial safety net.

The resolution of cross-border banks during the recent crisis is consistent with our theoretical analy-

sis. Intervention into large cross-border banks came often at a late stage and often with conflicts between

home and host country supervisors. While the lack of an effective bank resolution framework in most

European countries was certainly an important factor in explaining the late and uncoordinated interven-

tion into failing bank, incentives for domestic regulators facing weak international banks have played an

important role as well (Claessens et al., 2010). This became most obvious in the case of Fortis, a Benelux

based cross-border bank that was split up along national lines for resolution purposes. Similarly, the late

intervention into the Icelandic banks likely reflects not only lack of regulatory strength and competence

but also distorted incentives of national regulators.2

This paper contributes to the literature on regulation of cross-border banks by focusing on one spe-

cific aspect, the intervention decision of supervisors. While the previous literature has focused on capital

regulations and deposit insurance across borders, this paper analyzes the implications of cross-border

2For an in-depth discussion of the recent Icelandic crisis, see Benediktsdottir and Zoega (2011).


banking for the intervention into failing banks. In focusing on this specific aspect, we hold constant

other elements, such as capital requirements and deposit insurance. We also abstract from market dis-

cipline to focus exclusively on supervisory discipline. While the basic set-up of this model is similar

to Calzolari and Loranth (2011), we differ along several dimensions, including the distinction between

cross-border activities in terms of assets, deposits and equity. In addition, we provide empirical evidence

on regulatory bias in intervention decisions during the recent crisis.

The remainder of the paper is organized as follows. The next section discusses trends in cross-border

banking across Europe and the development of the corresponding regulatory frameworks. Section 2.3

presents the theoretical model. Section 2.4 contains the empirical analysis. The final section derives

policy conclusions.

2.2 Cross-border banking in Europe - trends and regulations

Cross-border banking has gained importance across Europe in recent decades, as part of a larger global-

ization wave in financial services. Even more impressive has been the transformation of many banking

systems in Central and Eastern Europe, which went from state-owned mono-bank systems to foreign-

bank dominated systems within a period of 10 years. Figure 1 illustrates this trend towards increasing

importance of cross-border banks across European financial systems. Hand in hand with an increase

in cross-border banking went an increase in bank concentration. As reported by Cihak and Decressin

(2007), before the crisis 16 large cross-border financial institutions accounted for about one third of EU

banking assets. The trend towards cross-border banking can also be illustrated for individual banks in

Europe. The percentage of foreign assets in total assets is 82 percent for Deutsche Bank, 64 percent for

Santander, 62 percent for UniCredit, 41 percent for BNP Paribas and 29 percent for Societe Generale

(Allen et al., 2011).

The patterns of cross-border banking, however, are very different in the “old” EU member countries

of Western Europe and the new EU member countries of Central and Eastern Europe. Financial inte-

gration in Western Europe has been a more gradual process though at accelerated speed over the past

decade. The introduction of the Single Banking License in 1989 through the Second Banking Directive

was a decisive step towards a Unified European Financial Market, which subsequently led to a conver-

gence in financial legislation and regulation across member countries. The introduction of the Euro in

1999 eliminated currency risk and provided a further push for financial integration (Kalemli-Ozcan et


al., 2010). In addition to cross-border lending, the increase in financial integration also came in the form

of cross-border mergers and acquisitions. Among the most high-profile cases were the take-over of Hy-

pobank in Germany by the Italian Unicredito, the takeover of the British Abbey National by the Spanish

Santander and the takeover of the Dutch ABN Amro Bank in 2007 by Fortis (Belgium), Royal Bank of

Scotland and Banco Santander.

While Western European countries have been both home and host of large cross-border banks, Cen-

tral and Eastern Europe has been exclusively host of such banks. The ownership transformation of the

banking system, from a state-owned mono-bank system towards a privately owned market-based finan-

cial system was key to achieving macroeconomic stability in the late 1990s, with countries finalizing the

ownership transformation process the fastest also being the first ones to successfully emerge out of the

systemic banking crises of the 1990s.

The expansion of banks across national borders has raised the issue of regulatory and supervisory re-

sponsibility. Across the world and in Europe, primary responsibility for bank regulation and supervision

lies with national authorities, even after the recent crisis. In the context of European financial integration,

the Second Banking Directive of 1993 introduced home-country control and mutual recognition, result-

ing in a “single passport” for branching across the EU: any bank licensed in an EU country is allowed

to open branches in other EU countries provided it meets some common, minimum standards. In spite

of this, however, many banks still choose to establish subsidiaries with separate capital, which might

be driven by bank specific factors and country circumstances (Cerrutti et al., 2007). On the supervisory

level, Memorandums of Understanding and colleges of supervisors for specific banks have become a

common tool of cooperation across borders. However, in spite of regulatory convergence and increased

supervisory cooperation in the early 2000s, important differences remained across EU countries in regu-

latory frameworks, supervisory standards and especially in bank resolution frameworks. Most countries

did not have bank-specific insolvency frameworks, which limited the options towards failing banks dur-

ing the crisis to either liquidation in regular bankruptcy procedures with all the interruptions that such a

long drawn-out process would bring with it or bailout with tax-payer money.

The crisis of 2008 has clearly shown the deficiencies of both national resolution frameworks, but

especially of cross-border resolution frameworks. The Dutch-Belgian Fortis bank is a good example for

that. In 2007, the Belgian Fortis was allowed to participate in the take-over of Dutch ABN Amro in spite

of already facing solvency problems, which points to both regulatory capture in Belgium and the lack of


information exchange between Belgian and Dutch supervisors. The conflict between Belgian and Dutch

supervisors following the take-over about who would be lead supervisor of Fortis made cooperation dur-

ing the subsequent crisis in 2008 difficult. Initial coordinated recapitalization failed to calm the markets,

which resulted in each national government taking their own actions, ultimately not only nationalizing

resolution of the individual bank pieces in Belgium, Luxembourg and Netherlands, but nationalizing the

banks themselves. Insiders stress that cooperation ultimately broke down when the Ministers of Finance

got involved.

Another, very different, example are the Icelandic banks. The late intervention by the Icelandic super-

visors is consistent with the theoretical model presented below and can be explained by the high shares

of foreign assets and deposits that Icelandic banks were holding, while equity was almost exclusively

held by domestic agents. The fact that a large share of deposits were collected through branches rather

than subsidiaries exacerbated the situation for host country supervisors as they had little information and

even less power to intervene in time.

In the wake of the crisis, attempts have been made to address these deficiencies both on the national

but also on the European level. The De Larosière Report (2009) acknowledged the need for better coordi-

nation among Member States, in order to allow for a well-functioning Single Market in banking, without

recommending, however, full centralization of EU regulation and supervision. Following the report,

the European Banking Authority (EBA) was established to more intensively coordinate micro-regulation

issues, while the European Systemic Risk Board (ESRB) is in charge of addressing macro-prudential

issues. Among other tasks, the EBA is charged to facilitate agreement between national supervisory au-

thorities, where necessary settling any disagreements, including within colleges of supervisors, to ensure

supervisors take a more coordinated approach. On the other hand, the EBA does not have any direct

supervisory power over banks and cannot legally force the intervention into weak banks or specific res-

olution techniques.

Further reaching reform suggestions, such as creating a European-level supervisor with intervention

powers or a European deposit insurance fund with resolution powers modeled after the U.S. FDIC or the

Canadian CDIC were rejected. Explanations are manifold; one institutional explanation is the principle

of subsidiarity which states that policy areas can only be transferred to the European level if they cannot

be undertaken at the national level. There is also a political economy argument that banking is considered

too critical a sector for individual countries to delegate regulatory powers to a supranational level. In


addition, resolution of large banks often involves taxpayer money; transferring the authority to commit

taxpayers’ money to the European level is seen as violating national sovereignty. Finally, there is the legal

challenge that a new institution cannot be created without a new European treaty, which is politically

difficult. While these political and constitutional constraints continue to weight against a supranational

supervisor on the European level, the Eurozone’s reaction to the ongoing sovereign debt crisis with more

fiscal centralization on the European level suggests that these constraints might lose weight over time.

2.3 Theoretical analysis

We present a simple model of bank supervision, with three periods, 0, 1 and 2. For ease of analysis, we

assume that the discount factor and the deposit interest rate are zero.3 There is a single representative

bank whose balance sheet is normalized to 1 and that issues debt d and equity k, so that d + k = 1. In

period 0, the bank invests its resources into an investment project whose success is random and outside

the control of the bank. Specifically, with probability λ (λ ∈ (0, 1)), the investment succeeds and yields

a return R>1 in period 2, and with probability 1 − λ, the project fails and yields zero gross return in

period 2.

While the supervisor has imperfect information about λ at date 0,4 λ becomes known at date 1. Based

on this information, a supervisor can decide whether to intervene in the bank or to allow it to continue.

If the supervisor decides to intervene in the bank, she can recover the initial investment of one. This

intervention can take different forms, ranging from liquidation to a purchase and assumption operation

involving another bank. If the supervisor decides to not intervene and allows the bank to continue to

period 2, with probability λ , the bank will be successful and be able to repay its debt and equity holders.

With probability 1− λ, the bank will fail and there are external costs of c.5

We assume that the supervisor maximizes domestic welfare, consisting of the returns to domestic

debt, equity minus domestic external costs (we will relax this assumption below). In the case of a purely

domestic bank, her intervention decision will hence coincide with the one that maximizes (world) wel-

fare. The intervention threshold is given by the λ which equates the expected returns from continuation

3The deposit rate itself does not matter. What matters though is that it is risk-insensitive, for example, because there is

deposit insurance with a flat premium.4Since the supervisor takes actions at date 1 only, the nature of date-0 uncertainty is not important.5For a discussion on the external costs that bank failure can impose on the remaining financial system and the real economy,

see Beck (2011). In principle, intervention at date 1 may also incur some costs, however, we would think that such costs are of

orders lowers than the ones arising from bank failure at date 2. Including external costs in period 1 does not change the main

conclusions of our model under reasonable parameterizations.


with the return from immediate liquidation. We have

λR− (1− λ)c = 1. (1)

Solving for λ gives

λ∗ =1 + c

R+ c. (2)

Quite intuitively, we can see that intervention becomes more likely when bank failure costs, c, increase

(the latter follows from λ∗′(c) > 0 for R > 1). By contrast, a higher return R reduces the intervention

probability.

While we assume throughout the paper that λ becomes perfectly known at date 1, we can easily

introduce a noisy signal on λ. As long as the signal is symmetrically distributed around the true λ, the

intervention threshold is the same. This can be seen by noting that the costs from intervening are linear

in λ (left hand side of equation 1).

2.3.1 The Incentives of a National Supervisor with Cross-Border Banking

We now introduce cross-border banking into our model. For this we allow the bank to be partially fi-

nanced by foreign deposits and foreign equity, as well as having asset holdings abroad. More specifically

we denote with βD the foreign share of deposits, with βE the foreign share of equity and with βA the

share of foreign firms (assets) financed by the bank.

The introduction of cross-border banking obviously does not modify the efficient intervention thresh-

old as it does not affect total payoffs in the world economy (thus including foreigners). It only affects

the share of the payoffs that accrue to domestic agents. As national supervisors only care about domes-

tic payoffs, this can change the intervention incentives for the domestic supervisor and drive a wedge

between the socially efficient and the domestic intervention point.

The domestic intervention point can be derived as follows. As before, if the domestic regulator

intervenes at the intermediate date, the bank will be liquidated. Total (world) proceeds from this are 1.

Domestic depositors obtain (1− βD)d and domestic equity obtains (1− βE)(1− d) of these proceeds.

Total payoff in the domestic economy is thus (1−βD)d+(1−βE)(1−d). In case there is no intervention

the bank succeeds with probability λ. In this case domestic depositors obtain (1 − βD)d, while equity

obtains (1−βE)(R−d). With probability 1−λ the bank fails. In this case both equity and debt holders


do not obtain any return and the country in addition suffers (1 − βA)c due to bank failure costs. Total

expected domestic payoff is hence λ((1−βD)d+(1−βE)(R− d))− (1−λ)(1−βA)c. The domestic

intervention threshold is defined by

λ((1− βD)d+ (1− βE)(R− d))− (1− λ)(1− βA)c = (1− βD)d+ (1− βE)(1− d). (3)

Rearranging for λ gives

λ =(1− βD)d+ (1− βE)(1− d) + (1− βA)c(1− βD)d+ (1− βE)(R− d) + (1− βA)c

. (4)

Note that for βD = βE = βA we obtain λ = λ∗. That is, when the resident balance sheet is propor-

tional to the entire balance sheet of the bank, liquidation decisions are efficient. Thus, if the cross-border

ownership share equals the other two cross-border shares, the domestic regulator always takes efficient

decisions regardless of the overall level of cross-border activities. The intuition for this is straightfor-

ward: if cross-border engagement is the same along all three dimensions, the domestic regulator will

simply perceive a fraction of both benefits and costs of intervention. Since this fraction is the same for

the costs and benefits, her decision will not be distorted.

We next derive comparative statics for the intervention threshold with respect to the various domestic

shares.

Proposition 1 The intervention threshold of the domestic supervisor, λD, is

i) decreasing in the share of foreign deposits βD,

ii) increasing in the share of foreign equity βE ,

iii) decreasing in the share of foreign assets βA.

Proof. Follows from taking the derivative of the intervention threshold λ with respect to βD,βE and

βA.

The intuition behind these results is as follows.

Deposits. Since the national regulator only cares about domestic depositors, a higher share of foreign

deposits will reduce the costs for her in period 2 and thus make intervention in period 1 less likely. A

higher share of domestic deposits, in turn, makes the domestic regulator less inclined to gamble on bank

success in the second period. Hence, with a higher share of domestic deposits, the domestic regulator


becomes more likely to intervene, that is, the range of λ′s where intervention takes place increases.

Equity. Shareholders have a relatively higher interest in continuing the bank due to the standard

risk-shifting problems (the costs of bank failure are partly borne by debt holders and firms). A higher

share of domestic shareholders aligns the interests of the regulator more with the one of shareholders.

This makes interventions less likely, that is, the threshold decreases. If, on the other hand, the share of

foreign equity holders is higher, the regulator is more likely to intervene in period 1.

Assets. When a higher share of bank assets is domestically invested, this raises the domestic external

costs of bank failure. This, in turn, makes the domestic regulator more averse to continuation. As a

result, she becomes stricter at date 1 (the minimum required success probability increases). On the other

hand, a higher share of foreign assets involves that a higher share of external costs in period 2 are being

borne by agents outside the home economy, which makes the regulator more reluctant to intervene in

period 1.

Note that the comparative statics are a direct consequence of the fact that in our model the gains from

continuation accrue at the margin to equity at the cost of the other two stakeholders. If this feature is

modified, the comparative statics may change. For example, there might be a cost of liquidation also at

date 1. If this cost is mainly borne by debtors (or affects primarily the holders of assets), equity may (in

relative terms) lose when the bank is continued. Similarly, continuation incentives may be reversed in

the presence of a (large) interest rate on debt that only accrues at date 2 (thus only when the bank is not

liquidated at date 1). In such situations our comparative static results may no longer hold.

Proposition 1 has straightforward welfare implications. We know that for βD = βE = βA =

1, domestic and efficient liquidation thresholds coincide. Since we also know, for example, that the

domestic liquidation threshold is increasing in the share of domestic deposits, it follows that when βD >

0 and βE = βA = 0 we have λ < λ∗. This implies that there is a range of λ (λ ∈ [λ, λ∗]) where it is

efficient to liquidate but the domestic supervisor decides to let the bank continue to operate (the domestic

regulator is then too lenient).

The following corollary summarizes this welfare result, alongside with the corresponding ones for

foreign equity and assets.

Corollary 2 When there is cross-border banking, domestic and efficient interventions generally do not

coincide. In particular we have:


i) If cross-border banking takes place only via deposits (βD > 0 and βE = βA = 0): there are

ranges for λ where the domestic regulator lets the bank continue even though this is inefficient (the

domestic regulator is too lenient);

i) If cross-border banking takes place only via equity (βE > 0 and βD = βA = 0): there are

ranges for λ where the domestic regulator liquidates the bank even though this is inefficient (the domestic

regulator is too strict)

iii) If cross-border banking takes place only via assets (βA > 0 and βD = βE = 0): there are ranges

for λ where the domestic regulator lets the bank continue even though this is inefficient (the domestic

regulator is too lenient)

Proof. Follows directly from Proposition 1 and λ = λ∗ for βD = βE = βA = 0.

If cross-border banking takes place through more than one channel, the welfare results obviously

depend on the strength of each channel. For example, if there are both cross-ownership of deposits and

equity, the biases created by each channel go in opposite directions and hence tend to offset each other. If

there is mainly foreign deposit-taking but little foreign ownership, we are then likely to end up with a too

lenient domestic regulator, and vice versa. This implies that in order to evaluate the efficiency properties

of cross-border banking, one has to look at all aspects of cross-border banking jointly, and not only at

one channel in isolation.

2.3.2 Discussion

Our analysis focuses exclusively on cross-border activities as the source of inefficient liquidation deci-

sions. It is for this reason that we have assumed that intervention decisions maximize domestic welfare.

There are several reasons why even in the domestic case regulatory interventions are not efficient. In

their presence, the distortions induced by cross-border banking have to be evaluated relative to these

inefficiencies.

Objectives of the Regulator We have assumed that the domestic authority responsible for the interven-

tion decision maximizes the returns to all domestic stakeholders. If this is not the case, other distortions

can arise, which can either strengthen or weaken the initial bias. If, for example, the authority which

decides on interventions is the (domestic) deposit insurance fund, intervention behavior will tend to be

tougher as the deposit insurer will try to maximize returns to domestic depositors rather than domestic


equity holders. If the central bank is in charge of intervention decisions, there may likewise be a tendency

towards strict interventions, if the central bank primarily cares about external failure costs. On the other

hand, if intervention decisions are taken by an independent supervisor, interventions may be relatively

lenient, in case of regulatory capture by domestic equity holders.

However, for our comparative static results it is not important that the domestic regulator maximizes

welfare. Proposition 1 continues to hold as long as the regulator puts a (positive) weight on the returns

of each of the stakeholders. The latter assumption seems plausible since many stakeholders are involved

in the resolution of large cross-border banks and one can hence expect their interests to be reflected to at

least some degree. Formally, suppose the regulator puts weights of φD, φE , φA on domestic debt, equity,

and asset holders. Condition (3) in the baseline model then becomes

λ((1−βD)φDd+(1−βE)φE(R−d))−(1−λ)(1−βA)φAc = (1−βD)φDd+(1−βE)φE(1−d). (5)

Rearranging for λ gives

λ =(1− βD)φDd+ (1− βE)φE(1− d) + (1− βA)φAc(1− βD)φDd+ (1− βE)φE(R− d) + (1− βA)φAc

. (6)

Note that for γ′i := γiφi (i ∈ {D,E,A}) this expression is the same as in the baseline model (equation

(4)). It hence follows that the (signs of the) comparative statics for λ with respect to γi are the same as

long as∂γ′i∂γi

> 0, that is, when φi > 0.

Endogenous Risk Taking Our baseline model has abstracted from a risk choice of the bank. Such a

risk choice can be introduced as follows. Assume that the probability of success λ (which is learned at

date 1) is distributed on [0, 1]. At date 0 the manager of the bank can make an effort choice e which

(stochastically) affects this success probability λ. Effort leads to private costs m(e) and we assume that

∂E[λ(e)]∂e > 0,

∂2E[λ(e)]∂2e

< 0 and m′(e),m′′(e) > 0. We assume that the supervisor cannot commit to a

date-1 intervention schedule and hence has to decide on the intervention in a time-consistent manner.

At date 1, the supervisor observes the realization of λ. Since effort at this point has been already

chosen, the regulator faces precisely the same situation as in the baseline model: taking λ as given, he

compares the pay-off from continuation with the pay-off from intervention. In particular, condition (1)

readily applies and the interventions decisions are thus the same as in the baseline model. Hence the


comparative statics of Proposition 1 on which we base our empirical analysis, continue to apply.

However, the time-consistent intervention no longer leads to welfare maximization – even in the

domestic case. This is because the effort choice will generally be inefficient. As we show in the appendix,

effort can then be either under- or overprovided.

Bailouts In our model intervention in the bank takes the form of a liquidation of the bank. An alterna-

tive form of intervention, which has been often used during the crisis of 2007-2009, is a bailout.6

Bailouts can be introduced in our model as follows. Suppose that instead of having the option to

liquidate the bank at date 1, the supervisor can decide to inject equity in the bank. Assume that if a

bailout takes place, the supervisor injects an amount of equity d in the bank, which is just sufficient to

avoid bank failure in case the project is not successful at date 2. We also assume that bailouts incur some

efficiency losses K > 0 to the economy, for example, because of the cost of public funds or because

bailouts distort financing decisions in the economy (in the absence of such costs it would always be

optimal to bailout a purely domestic bank as bailouts have no costs but can avoid the external cost of

bank failures c). In return for the equity injection the supervisor takes an equity stake of α in the bank.7

The intervention decision in a cross-border bank is now subject to the following considerations. In

case the supervisor does not intervene, domestic pay-offs are the same as in the baseline model: λ((1−

βD)d+(1−βE)(R−d))− (1−λ)(1−βA)c. When the supervisor intervenes, he incurs costs of d+K

at date 1. If the project succeeds, pay-offs for debt and equity are d andR at date 2 and if the project fails

debt receives d and equity zero. Hence total domestic pay-offs at date 1 and 2, including the return on the

supervisor’s stake in the bank, are−(d+K)+λ((1−βD)d+(1−βE)(1−α)R+αR)+(1−λ)(1−βD)d.

The intervention decision is thus given by

λ((1−βD)d+(1−βE)(R−d))−(1−λ)(1−βA)c <

−(d+K)+λ((1−βD)d+(1−βE)(1−α)R+αR)+(1−λ)(1−βD)d. (7)

Rearranging gives

K + d < λ(1− βE)(d− αR) + λαR+ (1− λ)(1− βA)c+ (1− βD)d. (8)

6For an evaluation of bank recapitalization during the recent crisis, see Mariathasan and Merrouche (2012).7Governments may also interfere with bank operations after a bailout in order to scale down bank risk. This can be inter-

preted as a partial liquidation of the project in the baseline model and would hence lead to the same comparative statics.


The left hand side of the equation is the cost of bailouts. The right hand side gives us the domestic

benefits from bailouts. They arise because of higher pay-offs for shareholders, λ(1− βE)(d− αR), the

government’s return on its stake, λαR, lower external failure costs accruing domestically, (1 − λ)(1 −

βA)c, and higher payouts to domestic debtors, (1− βD)d.

We assume that the bailout stake α is set such that the supervisor breaks even in expectation.8 Given

that the expected return on his equity stake is λαR, the breakeven stake is determined by d = λαR.

Rearranging gives

α =d

λR. (9)

Inserting into the equation for the liquidation threshold we obtain

K < (1− λ)((1− βA)c− (1− βE)d) + (1− βD)d. (10)

Solving for the critical λ gives us:

λ = 1− K − (1− βD)d(1− βA)c− (1− βE)d

. (11)

Noting that we need K− (1−βD)d > 0 and (1−βA)c− (1−βE)d > 0 for an interior maximum for λ

(λ < 1), it follows that λ′((1− βD)) > 0, λ

′((1− βE)) < 0 and λ

′((1− βA)) > 0. Hence, Proposition

1 continues to hold. Bailouts thus lead to the same comparative statics as liquidation decisions. The

efficiency implications in the case of bailouts are also the same as in the baseline model – which can

be appreciated from the fact that in contrast to the two previous extensions of the model, there are no

distortions in the absence of cross-border banking.

2.3.3 Numerical analysis

The analysis suggests that the intervention decision of a domestically-oriented supervisor depends in

principle on the various dimensions of a bank’s cross-border activities. An interesting question is whether

the cross-border activities can also be quantitatively important for the intervention decision. Only if this

is the case, can we expect significant welfare losses to result from cross-border activities.

This subsection contains a simple numerical exercise, with the aim of providing some sense of the

8Including also the efficiency loss K for this will not affect the results.


potential quantitative implications of cross-border activities. For this we parameterize the baseline model

and analyze the intervention threshold for different assumptions on cross-border activities. We assume

a return on investment R in period 2 of 1.085 (thus a net return of 8.5% conditional on success of

the project) and a debt share d of 0.9. We take external failure costs in period 2 as c = 0.5. In Box

1 we report the resulting intervention thresholds λ (calculated from equation (4)) for different foreign

activity levels. We consider four cases: a purely domestic bank and banks that have respectively 50

percent of either foreign assets, deposits or equity. Next to the intervention threshold we also calculate

the implied CDS spread at the time of intervention.9 We can see that under the chosen parameters the

critical intervention threshold for a domestic bank is 0.946. This translates into a CDS spread of 536bps

(by means of comparison, the average spread of 55 intervened banks considered in the empirical analysis

of Section 2.4 is 417bps). A bank with 50% foreign deposits sees a higher critical CDS spread of 749bps,

as predicted by Proposition 1 that the regulator is then more lenient. Next, we see that a bank with 50%

foreign assets has a critical spread of about 637bps. Finally, we see that a bank that has 50% foreign

equity has a critical CDS spread of only 284bps, suggesting a much stricter regulator.

A key implication of our analysis is that a bank with substantial foreign activities still can be subject

to efficient regulatory treatment – as long as its activities are balanced along the various dimensions. In

order to better understand the trade-offs involved in achieving balance, Figure 2 shows the combinations

of the cross-border shares for which the regulatory intervention is efficient (the x-axis is domestic de-

posits, the y-axis is domestic assets and the z-axis (vertical axis) is domestic equity). Above the surface

the regulator is too lenient (domestic equity is too high given the banks cross-border mix of deposits and

assets), while below the surface the regulator is too strict (domestic equity is too low for efficiency). We

can see that the trade-offs between the various shares are fairly linear and reasonable. For example, for

most combinations of domestic assets and deposits, there exists a domestic (equity) ownership share that

avoids regulatory inefficiency. This implies that achieving balance is feasible for banks regardless of how

international they are. We can also see that none of the three activity-shares is dominating, that is, each

activity share can be offset by appropriate shares along the other two dimensions.

9Noting that the expected loss at the critical value is (1− λ) · 1 and assuming that the CDS premium reflects the expected

loss on the underlying asset, we obtain a corresponding CDS spread of CDS = 1− λ.


2.3.4 Branches versus subsidiaries

Our model can be used to discuss regulatory implications of different organizational forms for inter-

national banks in establishing their presence in host markets. Banks have two main ways to undertake

foreign operations: through branches or by establishing a foreign subsidiary. The key difference between

branches and subsidiaries is that in the case of a branch the supervisor in the country of the parent bank

has responsibility (home supervisor), while in the case of a subsidiary it is the regulator in the coun-

try where the supervisor is located (host supervisor). Our model can be used to understand the relative

regulatory attractiveness of either mode of foreign entry in terms of their welfare properties.

Consider first the case of a subsidiary. From the perspective of the host country regulator, the sub-

sidiary has a large share of foreign equity as the profits of the subsidiary will return to the parent company

(high βE). Since the subsidiary will typically lend largely domestically in the host economy, the share of

domestic assets is, however, large (low βA). In addition, the subsidiary might also source deposits largely

locally (low βD). Thus, applying our model, from the perspective of the host supervisor, cross-border

banking largely takes place through foreign equity ownership. Corollary 2 tells us that regulation and

supervision will hence tend to be too strict.

Consider next branching. Under branching, the home country supervisor has responsibility for su-

pervision and the intervention decision. This supervisor can decide to intervene in the foreign branch but

only jointly with intervention at the parent bank. We distinguish in the following between two cases: i)

the size of the branch is small relative to the parent bank and, ii), the size of the branch is large relative to

the parent bank. To focus ideas, we also assume that the health of the parent and the foreign branch are

fully correlated (in terms of the model: both have the same realization of λ at date 1), an assumption we

relax below. In the case of the parent bank having more than one foreign operation (possibly in differ-

ent countries), relative size is defined as the combined size of all foreign branches relative to the parent

company.

Consider first the case where the foreign operations are small. From the perspective of the home

regulator there is hence effectively no cross-border banking. Her liquidation decision is hence unbiased

and efficient. In the case of large foreign operations, things play out as follows. Due to presence of

foreign lending by the foreign branches, there is a substantial part of foreign assets (βA high). In addition,

there are also foreign deposits (βD high), while there is no foreign equity (βE low). Using Corollary 2


we thus obtain that the domestic supervisor is too lenient.

What does this imply for the regulatory desirability of branching versus representation through a

subsidiary? In the case of a small foreign operation, branching is preferred as this leads to unbiased in-

tervention decisions. When the foreign operation is large, there is a trade-off. In the case of a subsidiary,

intervention in the foreign operation might be too strict, especially in countries with effective resolution

frameworks. In the case of a branch, intervention is too lenient. In either case, this leads to inefficient

liquidation decisions both domestically and abroad.

When (total) foreign operations are small relative to the size of the parent bank, cross-border banking

should take place through branching. When (total) foreign operations are large relative to the size of the

parent bank, either branches or subsidiaries may be preferred to obtain an efficient intervention decision.

Relaxing the assumption of perfect correlation between λ in the home and the host countries com-

plicates things somewhat in the case of large cross-border activity. If (1− (1− βA))λF + (1− βA)λD

< λ, where F denotes foreign and D domestic, the home country supervisor will intervene. If the two

λs are sufficiently different, this might imply that external costs of failure resolution are imposed on a

country where the banking operation is perfectly healthy (i.e. high λ). As the home country supervisor

internalizes only (1− βA)c , the supervisor is more lenient towards negative signals from the host coun-

tries. This can be further complicated if the home country supervisor receives only a noisy signal about

λ in the host countries. While not affecting the intervention threshold, it will increase both Type 1 and

Type 2 errors and thus reduce welfare.

Comparing the regulatory effects of branch versus subsidiary structure with the actual decision of

international banks shows that banks with cross-border retail operations prefer indeed subsidiaries, while

banks with small cross-border operations prefer branches (Cerutti et al., 2007). The recent expansion of

some European banks (e.g. Icelandic banks and Nordea) in the form of branches, however, provides

serious regulatory challenges, as we have shown in this sub-section.

In this respect it is interesting to note that our model suggests that banks may use branching exces-

sively from a welfare perspective. This is because branching leads to lenient intervention, which benefits

equity holders. In case where a subsidiary is the desirable form of organization from a welfare perspec-

tive, there is thus a potential inefficiency as banks may have an incentive to form foreign operations in

the form of branches.


2.4 Empirical analysis

The failure and intervention of Icelandic banks provide an illustrative example for our theoretical model.

The late intervention by the Icelandic supervisors can be explained by the high shares of both foreign

assets and deposits that Icelandic banks were holding, while equity was almost exclusively held by

domestic agents. The fact that a large share of deposits were collected through branches rather than

subsidiaries exacerbated the situation for host country supervisors as they had little information and even

less power to intervene in time.

In the following, we subject our theoretical model to a formal empirical test by exploring a sample

of 55 banks across 15 countries that failed and were intervened between 2007 and 2009. Specifically, we

use the CDS spread at the time of intervention as indicator of regulatory lenience or strictness and relate

it to the mix of foreign equity, assets and deposits of these banks, controlling for an array of other bank

characteristics. We first explain the methodology, before presenting the data and discussing the results.

2.4.1 Methodology

In our model, information about bank health (λ) is realized at a single point in time (date 1). This

means that the regulator intervenes whenever the realization of λ is anywhere below the critical λ. In

reality, bank health will rather evolve in a more continuous fashion. This suggests that regulators will

intervene precisely when the health has deteriorated to the degree that the critical λ is reached – at least

if the regulator does not perceive an option value of not closing down the bank. As a consequence, the

CDS spread at the time of intervention would be a good indicator of regulatory lenience or strictness. An

option effect may arise from the fact that a regulator may prefer not to close down a bank that has reached

the critical threshold because there is the chance that the bank will recover in the future and end up above

the threshold. Such a consideration would simply serve to reduce the critical threshold at which the

regulator intervenes – but the threshold would still depend on the various dimensions of “foreignness”

as outlined in the analysis of the previous section. Hence, it remains appropriate to study how bank

health at the time of intervention (as a measure of regulatory lenience) depends on foreignness. Figure 3

shows the evolution of CDS spreads for the 55 banks in our sample over the 90 days before intervention,

normalized by 100 for the day of intervention. Consistent with our discussion, the spread increases over

these 90 days, with the increases accelerating in the weeks before intervention.


Proposition 1 (which was derived assuming that the continuation value of the bank mainly accrues to

equity) and the discussion above suggest the following testable hypothesis:

The CDS-spread at the time of intervention i) decreases in the share of foreign equity, ii) increases

in the share of foreign assets, and iii) increases in the share of foreign deposits.

We test this hypothesis with the following empirical specification:

yi = α+ β · Fi + θ · Zi + εi. (12)

where Fi is a vector of cross-border activities (share of foreign deposits, assets and equity) and Zi is

a vector of control variables. The dependent variable yi is the log of the CDS spread (CDS) or the

difference between log of CDS spread and the log of CDS index for the region where the bank is located

(CDS spread relative to index). This second variable allows for the possibility that regulatory lenience

not only depends on a bank’s financial health but also on the general situation in the banking sector.

The set of control variables includes different bank characteristics that can possibly explain the timing

of regulatory intervention, but are outside our theoretical model. We include bank size, defined as the

log of bank total assets, as there might be a regulatory bias towards intervening large banks too late, a

phenomenon known as too-big- or too-complex-to-fail. In robustness tests, we also use the share of bank

assets in total assets of the banking system instead of its size. We include the logs of the tier-1 capital

ratio, as regulators face higher pressure to intervene undercapitalized banks, while lower liquidity, as

measured by the ratio of liquid assets to total assets, might provide an additional indication of fragility

and thus trigger regulatory intervention.10 As an alternative to those bank balance variables we also use

the CDS spread of a bank 12 months preceding its intervention for robustness tests. We take this CDS

spread as a proxy of the bank’s historical health.

In addition, we control for government ownership and timing of intervention. The considerations

when intervening in banks with an equity stake owned by the government are presumably different ones

and this is may be reflected in bank CDS spreads. For this reason, we use a dummy (State ownership) in

all specifications which indicates that the government has a stake of more than 5 percent in the bank. We

also include a crisis dummy (Post-Lehman period) in our empirical model to isolate the effect of Lehman

Brothers’ collapse on CDS spreads – as this event has arguably increased pressure on regulators to

10Some of these variables might be endogenous although we measure them at the end of the year before intervention.


intervene weak banks. The bailout dummy takes the value of 0 prior to September 2008 and 1 afterwards.

We expect to find a negative association of the Post-Lehman period dummy and the State-Dummy with

the intervention threshold.

Moreover, we include several measures gauging the regulatory structure of countries. The dummy

Central Bank takes the value of one if the central bank is involved in the supervision of banks and the

dummy variable Single Agency indicates whether more than one agency is involved in the supervision

of banks. As discussed above, we expect central banks to be more stringent supervisors when being

granted with such supervisory function. In addition, we expect that intervention decisions are more

lenient in the case of multiple supervisors, as coordination problems may make it more difficult to agree

on intervention. CDS spreads at intervention should hence be higher. In robustness tests, we also use

a variable that measures deposit insurance coverage relative to GDP per capita as a proxy for the costs

faced by national supervisors.

Finally, in unreported robustness tests, we also control for an alternative explanation for cross-bank

variation in CDS spreads. We use a measure of CDS liquidity defined as the difference between the bid

and ask for debt insurance of a particular bank (Bid Ask spread) normalized by the CDS spread. It is

expected that lower levels of liquidity will lead to higher CDS spreads. As a result, the CDS spreads

that we observe can also reflect a higher compensation for liquidity risk rather than information about

deterioration of bank financial health.

The table in the appendix lists all variables used in regressions as well as their definitions and data

sources.

2.4.2 Data

Our analysis is based on a unique hand-collected bank-level dataset, which contains information on

cross-border activities of European and U.S. banks that were intervened during the financial crisis in

the period between 2007 and 2009. Our main sources of information on foreign assets and deposits are

annual reports and the accompanying notes to bank financial statements from the fiscal year preceding

bank intervention. When data on foreign assets are missing, we use the share of foreign loans or deposits

instead. In a similar way, we use the available data on foreign assets and loans as a complement for


missing shares of foreign deposits.11 In the case of foreign equity, we collect data from Bankscope

on ownership by foreign shareholders. Since equity shares are likely to change quickly within a year,

the share of foreign ownership is taken at the last available time period prior to the month of bank

intervention.

Intervention dates are taken from the dataset compiled by Laeven and Valencia (2010), comple-

mented with own collected data. Table 1 reports the intervention dates. We measure regulatory lenience

or strictness by the CDS spread of the bank at the time of intervention. The idea is that a higher CDS

spread at intervention reflects that the regulator has waited for bank health to deteriorate significantly be-

fore intervening, that is, her critical λ is low.12 Such a regulator is lenient in the language of our model.

Conversely, a regulator who tends to intervene already at low CDS spreads is considered a stricter regu-

lator. We collect daily observations on five year senior debt CDS spreads from Datastream before bank

intervention. Ideally we would like to use the values of the CDS spreads immediately prior to the first re-

lease of a public announcement on bank intervention. However, the CDS spread at the day of intervention

(or the previous day) may already reflect intervention expectations. In order to mitigate this problem, we

use CDS spreads 3 days prior to intervention. As a bank’s CDS spread might partly be driven by overall

market movements, we use the difference between the log of a bank CDS spread and the log of a CDS

index as alternative indicator. A CDS index from Datastream about the European bank sector is used if

the bank is located in Europe and the Datastream CDS index pertaining to the US banking sector if the

bank is located in the US. In addition, looking at the difference to the CDS index will also help to con-

trol for non-bank-specific risk factors in CDS spreads, such as economy-wide risk and liquidity premia.

These premia have been shown to be an important part of CDS prices (see Amato (2005) and Bongaerts,

de Jong and Driessen (2008)) but should in principle not affect regulatory intervention decisions. The

CDS spreads and bank-level variables in all regressions are winsorized at the 5 percent level and bank

size is also taken in logs.

Table 2 provides summary statistics for the variables included in our baseline analysis. On average,

the cross-border activities of banks in our sample seem balanced - the mean share of bank foreign equity

is 35 percent and the average share of foreign assets and deposits is 33 and 32 percent, respectively.

11The missing share of foreign assets is replaced with data on foreign loans and deposits in 3 and 4 cases, respectively.

Replacement of missing data on foreign deposits with available data on foreign loans or assets occurs in 8 cases.12Recalling that λ is the likelihood of project success and that the LGD in our model is 100%, the relationship between CDS

spread and threshold lambda is: CDS = 1− λ.


There is a large variation in CDS spreads three days before intervention across banks. While the mean

(unwinsorized) CDS spread is approximately 417 bps around the time of intervention, it varies between

52 and 3626 basis points across banks. The three major Icelandic banks and the U.S.-based bank Wash-

ington Mutual Inc. have the highest CDS spread at time of intervention in the dataset. On the other tail

of the distribution, we have BNP Paribas and Credit Agricole SA with the lowest CDS spreads at the

time of intervention. On average, bank total assets are over 450 billion Euros. The tier 1 ratio is, on

average, eight percent and the share of liquid assets 22 percent. 16 percent of banks had state ownership

before intervention and 67 percent of banks in our sample were intervened after the Lehman Brothers

failure. About half of intervened banks are located in countries where a single agency is in charge of

bank supervision and about 60 percent in a country where the central bank has a supervisory function.

Table 3 reports correlations among the key variables used in regressions. The pair-wise correlations

between the CDS-based measures of lenience and the shares of foreign bank activities have the expected

signs but are not significant at the 10 percent level. The CDS spread for example is decreasing in foreign

equity and increasing in the share of foreign assets and deposits. The insignificance of the correlation

estimates may reflect that we need to control for various other factors that influence the intervention

decision. Next we look at the control variables and their correlation with our measures of lenience. Bank

size and the liquid asset ratio have a statistically significant and negative correlation with the CDS-based

measures and the expected signs, while the Tier 1 ratio has a statistically significant and a negative one.

There is a high correlation between foreign assets and deposits suggesting that their joint inclusion in

regressions can lead to multicollinearity. Thus in regressions we either use their average or include them

in separate specifications.

2.4.3 Results

The results in Table 4 provide evidence consistent with the hypotheses derived from our model. Here, we

regress the CDS spread three days before intervention on the various foreign activity shares, controlling

for other bank and country variables. In column 1 of Table 4 we fit a model that includes foreign equity

and foreign deposits while in column 2 the share of foreign deposits is replaced with the share of foreign

assets. Due to high correlation between foreign assets and deposits, we include the average of both in

column 3. In column 4, we fit the same model as in column 3 but also include our set of additional


controls.13 The coefficient estimates of the variables of interest have signs consistent with the theoretical

model. Banks’ foreign equity share is negatively associated with the CDS spread at time of intervention,

with coefficient estimates that are statistically significant either at the one or the five percent level. An

increase in the share of bank foreign equity by one percentage point is associated with a decrease in CDS

spreads around the period of intervention between 0.86 and 0.98 percent, ceteris paribus. Similarly, the

coefficient estimates of foreign assets and deposits are significant and have the expected positive sign and

this is also the case when the average share of foreign assets and deposits is included. One percentage

point increase in those shares is associated with an increase in CDS spreads between 0.78 and 1.3 percent.

In columns 5 to 8 of Table 4 we replace the log of bank CDS spread as the dependent variable with the

log of the CDS spread relative to the log of the regional CDS index as a measure of relative lenience. We

confirm our results both in statistical as in economic significance. We note that a better fit (as indicated

by the R-squared) is obtained for the regressions with the relative CDS spreads, i.e. our variables explain

a larger share of the variation of relative rather than absolute CDS spreads at time of intervention.

Turning to the control variables, we find that bank size is negatively and significantly associated with

the CDS spread at time of intervention, suggesting that regulators intervene earlier into big banks. Banks

that were intervened after September 2008 were intervened at lower CDS spreads, suggesting a stricter

regulatory approach after this event. Similarly, the negative and significant sign on state ownership

suggests a stricter approach of regulators towards these banks. In columns (4) and (8), we also find that

a higher level of bank liquidity is associated with earlier intervention. We do not find evidence that the

institutional structure of supervision matters for intervention thresholds. Neither the involvement of the

central bank nor the existence of multiple supervisors is consistently significantly associated with the

CDS spread at time of intervention.

In unreported robustness tests (available on request) we included several other bank- and country-

variables that might be related to the CDS spread at intervention. First, we include a measure of deposit

insurance coverage relative to GDP per capita as a proxy of the costs that a regulator will anticipate upon

bank failure. The main message is confirmed (though we lose significance for foreign asset and deposits

in the regressions with the relative CDS index), while deposit insurance coverage enters negatively and

significantly only when using the CDS spread relative to the index as dependent variable, suggesting

earlier intervention in countries where deposit insurance coverage is higher. Second, we control for the

13We include some of the variables only in column 4 as they are not available for all banks and thus reduce the sample size.


relative importance of a bank to national economies and replace bank size by the share of bank assets

in total assets of the national banking system. Our cross-border variables continue to enter significantly,

while the market share variable enters significantly in the majority of specifications, with a negative sign

when using the CDS spread and a positive sign when using the CDS spread relative to the index. Third,

we include our measure of CDS liquidity (bid-ask spread) in the regression. While this variable does not

enter significantly, our main findings continue to hold, though sometimes at lower significance levels.

Finally, we also estimate our model using as dependent variable CDS spreads taken at different points in

times before the intervention date. When using the CDS coefficient two weeks before intervention, the

coefficient estimates of the variables of interest are of lower statistical significance, with higher standard

errors. This is to be expected since CDS spreads may then not fully reflect the deterioration in bank

health that is very likely to take place just prior to the intervention. As we move farther away from the

intervention date, the significance of the coefficient estimates becomes lower.

Next, we address the concern that the market may partly anticipate interventions. CDS spreads

before the intervention date may then not only reflect bank health but also the effect of the intervention

itself. This may bias our results if the anticipation effect is systemically related to the foreign shares. In

Table 5 we report results of regressions where we regress the difference between CDS spreads at time

of intervention and four weeks prior to intervention (the anticipation effect) on the foreign shares and

others controls. All the foreign shares are insignificant at the 5% level. Thus, any anticipation effect that

may be present is unrelated to the foreign shares and hence there is no reason to expect that our results

might be biased.

We also test whether our results are driven by selection bias as our sample so far contains only

intervened banks. We use a Heckman model to control for possible sample selection, where the first

stage explains the probability of a bank to be intervened or not. For this we include next to our sample

of failed banks also 40 banks that were not intervened during the sample period. We use the same

controls as in the baseline model plus the loan-asset ratio and the ratio of non-performing loans in the

first stage; these asset risk proxies should be good predictors of whether a bank will face problems and

hence is likely to be intervened. At the same time, there is no reason to believe that asset risk affects

the threshold at which a regulator intervenes in a bank, that is, the second stage.14 Table 6 reports the

14The intervention threshold (right hand side of equation (4)) depends only on the foreign activity shares, debt and failure

costs but not on asset risk.


first and second stages of the Heckman regression. The first-stage results suggest that a higher share

of foreign equity and a lower share of foreign assets and deposits make an intervention more likely,

while banks were more likely to be intervened after September 2008, and if they had higher loan-asset

ratios. Critically, the second-stage results are consistent with our previous findings. All cross-border

variables enter significantly, with coefficients of similar size as in Table 4. Bank size, the Post-Lehman

period dummy and the state ownership dummy also enter negatively and significantly, as in Table 4.

The inverse of the Mills-ratio, lambda, is insignificant suggesting absence of sample selection problems.

Overall, the Table 6 results show that our findings are not driven by selection bias.

In a further robustness test (available on request) we test the validity of our intervention model for

a sample of 25 banks that were not intervened during the financial crisis and for which we have also

CDS spread data available. If our model provides a reasonable description of intervention thresholds, the

spreads of non-intervened banks should tend to be below their predicted intervention threshold. We test

this by calculating the predicted intervention thresholds for non-intervened banks (using the estimated

coefficients from regression 3 in Table 3). We then compare these predicted spreads to the actual spreads

of these banks during the sample period. On average, 98 percent of daily spreads are below predicted

spreads before Lehman Brothers’ collapse and about 87 percent afterwards. For many banks, 100 percent

of CDS spreads are below the predicted intervention threshold, with a few outliers, such as the Spanish

Banco Popular, Barclays and the US financial conglomerate MetLife Inc. that have a substantial period

with CDS spreads above the predicted threshold after Lehman Brothers’ failure. Overall, this lends

support to our empirical specification of the intervention decision since the model (when estimated on

intervened banks) tends to predict that non-intervened banks’ CDS spreads are below the intervention

threshold.

2.5 Conclusions and Policy Implications

This paper uses a simple model to illustrate the trade-offs involved when intervening in cross-border

banks. We show that foreign assets and deposits, on the one hand, and foreign equity, on the other hand,

have different implications for the intervention decisions of home country regulators. Critically, a mix of

the three can lead to the same intervention threshold as a purely domestic bank. Our model can inform

both the discussion on national versus supra-national bank supervision and the discussion on the optimal

organization of cross-border activity from the regulator’s viewpoint. While regulators may not want to


directly control imbalances in cross-border activities of individual banks, it should be the task of the

European Systemic Risk Board to monitor such imbalances since they lead to inefficient supervision of

banks.

Our empirical analysis using a sample of intervened banks during the recent crisis confirms the pre-

dictions of the model. Banks with a higher share of foreign equity were intervened relatively early as

their financial health deteriorated, while banks with a high share of foreign deposits and assets were in-

tervened relatively late. These results clearly support the prediction of the theoretical model that national

regulators have biased incentives when dealing with cross-border banks. The message that emerges has

obvious implications for the ongoing debate on the reform of resolution regimes around the world.

A supra-national supervisor could, in principle, always improve welfare because this supervisor

would also take into account the effects that materialize outside the country. However, supra-national

supervision might itself also be subject to imperfections. First, a global supervisor may have imperfect

knowledge about the success probability at date 1, receiving only a noisy signal.15 This means that the

supra-national supervisor, even though having the correct incentives, will make sometimes wrong deci-

sions due to imperfect knowledge of the success probability. The benefits from delegation to a supra-

national supervisor (arising because it avoids the distorted incentives of domestic supervision) thus have

to be weighed against the costs arising because the global supervisor has an informational disadvantage.

Second, the external costs from failure in period 2 might be higher for the affected economies under

supra-national supervision, as intervening and resolving a bank that is present in markets with different

legal frameworks can result in lengthy and costly resolution. Again, this presents an additional cost to

supranational supervision.

Our theoretical and empirical analyses can also be seen in the broader context of a trilemma of

financial integration (Schoenmaker, 2010) that states that financial integration, financial stability and

national sovereignty in bank regulation cannot be achieved simultaneously and one has to give. We have

shown – both theoretically and empirically – that this is the case due to imbalances in multinational

banks. Presuming that one wants to maintain financial stability, the options are hence either a move

towards national banking systems, with stand-alone, fire-walled subsidiaries or a move towards supra-

national supervision, with the caveats mentioned above. Such a regime can in principle improve the

failure resolution for imbalanced banks.

15See Holthausen and Ronde (2002) for a similar argumentation.


However, one condition for this is that the jurisdiction of the supra-national supervisor corresponds

to the geographical area of bank activities. As shown by Osterloo and Schoenmaker (2007) and Schoen-

maker (2010), the largest 25 European have, on average, 25% of their assets outside their home country

in other European countries. This share ranges from two percent in the case of BBVA (which has 31% of

assets outside Europe) to the Nordea Group, with 74% of assets outside its home countries in other Euro-

pean countries (and no assets outside Europe). A European-level supervisor can only alleviate distortions

to the extent that they arise from imbalances within the Europe – but not the ones arising from external

imbalances. In addition, such a supervisor can only improve on a purely national resolution framework

if equipped with the necessary means and resources to resolve a bank efficiently. The resolution powers

also have to come with the necessary supervision and monitoring tools; a close relationship with national

supervisors is therefore critical.

Different institutional options have been discussed for a European-level bank supervisor. Allen et al.

(2011) suggest to combine the functions of deposit insurance and resolution for large European cross-

border banks within a European equivalent of the Federal Deposit Insurance Corporation (FDIC). Such

an institution could operate either in parallel to the European Banking Authority (EBA) or be merged

with it. An alternative suggestion would be a choice-based model where EU member states can opt to

delegate supervision of their largest banks to the ECB. Such a model allows for more flexibility and is

likely to face less political resistance than the creation of a pan-European institution (Hertig et al. 2010).


References

[1] Acharya, Viral (2003), Is the international convergence of capital adequacy regulation desirable?,

Journal of Finance 58, 2745–2781.

[2] Allen, Franklin, Thorsten Beck, Elena Carletti, Philip Lane, Dirk Schoenmaker and Wolf Wagner

(2011), Cross-Border Banking In Europe: Implications for Financial Stability and Macroeconomic

Policies, CEPR Policy Report.

[3] Amato, Jeffery D. (2005), Risk Aversion and Risk Premia in the CDS Market, BIS Quarterly Re-

view, December, 55–68.

[4] Barth, James R., Gerard Caprio and Ross Levine (2001), The regulation and supervision of banks

around the world - a new database, World Bank Policy Research Working Paper No. 2725.

[5] Beck, Thorsten (2011), Bank failure resolution: A conceptual framework, In: Panagiotis Delimat-

sis and Nils Herger (Eds.): Financial Regulation at the Crossroads: Implications for Supervision,

Institutional Design and Trade, 53-71.

[6] Benediktsdottir, Sigridur, Jon Danielsson and Gylfi Zoega (2011) Lessons from a Collapse of a

Financial System. Economic Policy 66, 183-235.

[7] Bongaerts, Dion, Frank De Jong and Joost Driessen (2011), Derivative Pricing with Liquidity Risk:

Theory and Evidence from the Credit Default Swap Market, Journal of Finance 66, 203-40.

[8] Calzolari, Giacomo and Gyongyi Loranth (2011), Regulation of multinational banks: A theoretical

inquiry, Journal of Financial Intermediation 20, 178-98.

[9] Cihak, Martin and Decressin, Joerg (2007), The Case for a European Banking Charter, IMF Work-

ing Paper 07/173.

[10] Claessens, Stijn, Richard Herring, and Dirk Schoenmaker (2010), A Safer World Financial System:

Improving the Resolution of Systemic Institutions, Geneva Report on the World Economy 12.

[11] Claessens, Stijn, and Neeltje van Horen (2010), Foreign Banks: Trends, Impact and Financial

Stability, IMF Working Paper WP/12/10.

[12] Cerutti, Eugenio, Giovanni Dell ’Ariccia and Maria Soledad Martinez Peria (2007), How banks go

abroad: Branches or subsidiaries?, Journal of Banking and Finance 31, 1669-92.

[13] De Larosière (2009), Report of the High-Level Group on Financial Supervision in the EU, Brussels:

European Commission.

[14] Dell’Ariccia, Giovanni and Robert Marquez (2006): Competition Among Regulators and Credit

Market Integration, Journal of Financial Economics 79, 401-30.

[15] European Commission (2010), Communication from the Commission to the European Parliament,

the Council, the European Economic And Social Committee and the European Central Bank - Bank

resolution funds.

[16] Fonteyne, Wim, Wouter Bossu, Luis Cortavarria-Checkley, Alessandro Giustiniani, Alessandro

Gullo, Daniel Hardy and Sean Kerr (2010), Crisis Management and Resolution for a European

Banking System, IMF Working Paper10/70.


[17] Freixas, Xavier (2003), Crisis Management in Europe. In Financial Supervision in Europe, ed. J.

Kremers, D. Schoenmaker, and P. Wierts, 102–19. Cheltenham: Edward Elgar.

[18] Goodhart, Charles and Dirk Schoenmaker (2009), Fiscal Burden Sharing in Cross-Border Banking

Crises, International Journal of Central Banking 5, 141-65.

[19] Hertig, Gerard, Ruben Lee and Joe McCahery (2010), Empowering the ECB to Supervise Banks:

A Choice-Based Approach" (with Ruben Lee and Joseph A. McCahery), European Company and

Financial Law Review 8, 169-213.

[20] Holthausen, Cornelia and Thomas Ronde (2002), Cooperation in international banking supervision:

a political economy approach, ECB Working Paper 315, European Central Bank.

[21] Kalemli-Ozcan, Sebnem, Elias Papaioannou, José-Luis Peydró (2010). What Lies Beneath the

Euro’s Effect on Financial Integration: Currency Risk, Legal Harmonization, or Trade?, Journal

of International Economics 81, 75–88.

[22] Krimminger, Michael (2007), Banking in a Changing World: Issues and Questions in the Reso-

lution of Cross-Border Banks, In: Douglas Evanoff, George Kaufman and John LaBrosse (Eds.):

International Financial Instability. Global Banking and National Regulation.

[23] Laeven, Luc and Fabian Valencia (2010), Resolution of Banking Crises: The Good, the Bad, and

the Ugly, IMF working paper 10/146, June 2010.

[24] Loranth, Gyonguyi and Alan Morrison (2007), Deposit Insurance, Capital Regulations and Finan-

cial Contagion in Multinational Banks, Journal of Business Finance & Accounting 34, 917-49.

[25] Mariathasan, Mike and Ouarda Merrouche (2012) Recapitalization, Credit and Liquidity, Economic

Policy, forthcoming.

[26] Osterloo, Sander and Dirk Schoenmaker (2007), Financial Supervision in an Integrating Europe:

Measuring Cross-Border Externalities, International Finance 8, 1-27.

[27] Schoenmaker, Dirk (2010), Burden Sharing for Cross-Border Banks, Revista de Estabilidad Fi-

nanciera, 18 , Bank of Spain, 31-47.

[28] Wagner, Wolf (2010), In the Quest of Systemic Externalities: A Review of the Literature, CesIfo

Economic Studies 56, 96-111.

[29] White, Lawrence (1994), On the International Harmonization of Bank Regulation, Oxford Review

of Economic Policy 10, 94-105.

[30] World Bank (2008), World Bank Survey on Bank Regulation and Supervision


Appendix A: Efficiency of Effort Provision

In this appendix we analyze the bank’s effort choice that arises when regulators act in a time-consistent

manner at date 1 and compare this to the (socially) efficient effort choice.

Denote with λ∗ the intervention decision of the regulator in a domestic bank (as we have just argued,

this threshold is identical to equation (2) of the baseline model). Bank effort is determined as follows.

The pay-off to bank equity is 1− d in the case of intervention and R − d in the case of project success.

In the case of failure the pay-off is zero. We hence have for bank effort:

e∗B = argmaxeWB = (1− d)E[λ(e) < λ∗] + E[λ(e)(R− d) | λ(e) ≥ λ∗] (13)

The first order condition for bank effort is

WB′(e) = (E[λ(e) | λ(e) ≥ λ∗](R− d)− (1− d))∂E[λ(e) ≥ λ∗]

∂e

+ (R− d)∂E[λ(e) | λ(e) ≥ λ∗]

∂e−m′(e) = 0.

The first term is the benefit from higher effort because the bank can now continue in more states of

the world (this effect arises because the bank takes into account that its own effort choice affects the

regulatory decision). The second term is the benefit arising because when the bank can continue, average

λ will be higher. The third terms are the effort costs.

By contrast, efficient effort in the economy maximizes welfare. Given that payoffs are the same as in

the baseline model – except that we also have to take into account the effort costs m(e)) – efficient effort

solves

e∗ = argmaxeW = E[λ(e) < λ∗] + E[λ(e)R− (1− λ(e))c | λ(e) ≥ λ∗]−m(e). (14)

The efficiency condition is hence

W ′(e) = (E[λ(e) | λ(e) ≥ λ∗](R+ c)− c− 1)∂E[λ(e) ≥ λ∗]

∂e

+∂E[λ(e) | λ(e) ≥ λ∗]

∂e(R+ c)−m′(e) = 0.

The first term is the effect that arises because a change in effort affects also the optimal liquidation

threshold. This effect is zero due to the envelope theorem. The condition hence simplifies to

W ′(e) =∂E[λ(e) | λ(e) ≥ λ∗]

∂e(R+ c)−m′(e) = 0. (15)

Note that W ′(e) is larger than the term∂E[λ(e)|λ(e)≥λ∗]

∂e (R − d) in WB′(e). Intuitively, this is because a

bank does not take into account that higher effort has benefits for depositors and asset holders. However,

the condition for private optimality has also the term, (E[λ(e) | λ(e) ≥ λ∗](R−d)−(1−d))∂E[λ(e)≥λ∗]

∂e .

This term is positive because higher effort increases the likelihood that the bank can continue, which

increases bank pay-off (but not welfare since at the λ∗ the economy is indifferent to continuation). There

are thus two offsetting effects. Effort can hence be either under- or overprovided even in an economy

without cross-border banking.


Appendix B: Variable Definitions

Variable name Description Source

CDS spread CDS spread of a bank 3 days before in-

tervention; in logs and winsorized at the 5

percent level

Datastream

CDS spread relative to index Difference between bank CDS spread 3

days before intervention and CDS spread

index about the region where the bank is

located; in logs and winsorized at the 5 per-

cent level

Datastream

Foreign ownership Share of bank foreign ownership BankscopeForeign assets Share of bank foreign assets Annual reports and authors’

own collected dataForeign deposits Share of bank foreign deposits Annual reports and authors’

own collected dataSize (in mil. EUR) Total assets; in logs and winsorized at the

5 percent level

Bankscope

Liquid assets Ratio of liquid assets over total assets win-

sorized at the 5 percent level

Bankscope

Tier 1 ratio Tier 1 capital ratio winsorized at the 5 per-

cent level

Bankscope

Post-Lehman period A monthly dummy that takes the value of

1 after Lehman Brother’s collapse in Sep-

tember 2008, and 0 otherwise

Authors ’ calculations

State ownership A dummy that takes the value of 1 if a gov-

ernment has a stake in a bank of more than

5 percent, and 0 otherwise

Bankscope

Central bank A dummy that takes the value of 1 if the

central bank supervises banks, and 0 other-

wise

World Bank’s Banking Reg-

ulation Survey (June 2008),

question 12.1.1Single agency A dummy that takes the value of 1 if there

is a single agency to supervise banks & fi-

nancial institutions in a given country, and

0 otherwise

World Bank’s Banking Reg-

ulation Survey (June 2008),

question 12.1.4

Note: This table provides an overview of definitions and sources of all variables used in the empirical analysis. Bank balance

sheet variables are from the last fiscal period prior to bank intervention. BankScope denotes Bureau van Dijk’s BankScope

database and Datastream - Thomson Reuters Datastream.


Box 1: Implied Intervention Thresholds

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

Fully domestic 50% For. deposits 50% For. assets 50% For. equity

Foreign deposits 1 0.5 1 1

Foreign assets 1 1 0.5 1

Foreign equity 1 1 1 0.5

Lambda 0.946 0.925 0.936 0.972

CDS spread (in basis points) 536 749 637 285

Note: Box 1 reports the intervention threshold for bank with different degrees of cross-border activities.


2030

4050

6070

Shar

e of

For

eign

Ban

ks

1995 1997 1999 2001 2003 2005 2007 2009

Year

EU economies Transition economies in the EU

Nontransition economies in the EU

Figure 1: Cross-Border Banking in European Union

Note: This figure shows the share of foreign banks among total banks in the European Union between 1995 and

2009. Source: Claessens and van Horen (2011)


Figure 2: Efficient Regulatory Intervention

Note: This graph shows the combination of the three domestic shares (given by 1 minus the foreign share) for which regulatory

intervention is efficient. The estimation is based on the parameter values listed in Box 1. X-axis is domestic deposits, y-axis

is domestic assets, z-axis (verical axis) is domestic equity. Above the surface the regulator is too lenient (domestic equity is

too high given the bank’s cross-border mix of deposits and assets), while below the surface the regulator is too strict (domestic

equity is too low for efficiency).


Figure 3: Average CDS Spread before Intervention

Note: This figure shows the average daily CDS spread of the 55 failed banks in our database up

to 90 days before being intervened. The average CDS spread at any given day is scaled relative

to the average CDS spread at the time of intervention. To minimize anticipation effects in

building this index, we use CDS spreads 3 days before intervention as the benchmark. Source:

Thomson Reuters DataStream and authors’ calculations


Table 1: List of Intervened Banks

Bank Intervention date Bank Intervention date

ABN AMRO NV 29-Sep-08 HSH Nordbank AG 3-Apr-09

AEGON Bank NV 12-Nov-08 IKB Deutsche Industriebank AG 1-Aug-07

Alliance & Leicester Plc 14-Jul-08 ING Bank NV 20-Oct-08

Allied Irish Banks plc 12-Feb-09 Intesa Sanpaolo 20-Mar-09

Anglo Irish Bank Plc 19-Dec-08 JP Morgan Chase & Co. 13-Oct-08

BNP Paribas 20-Oct-08 KBC Bank NV 1-Dec-08

Banca Monte dei Paschi di Siena 27-Mar-09 Kaupthing Bank hf 9-Oct-08

Banca Popolare di Milano SCaRL 25-Mar-09 Landesbank Baden-Wuerttemberg 15-May-09

Banco Popolare 19-Jun-09 Landsbanki Islands hf 7-Oct-08

Bank of America Corporation 30-Oct-08 Lehman Brothers Holdings Inc. 15-Sep-08

Bank of Ireland 11-Mar-09 Lloyds Banking Group Plc 20-Oct-08

Bayerische Landesbank 4-Dec-08 Merrill Lynch & Co., Inc. 15-Sep-08

Bear Stearns Companies LLC 14-Mar-08 Morgan Stanley 27-Oct-08

Bradford & Bingley Plc 29-Sep-08 Natixis 31-Jul-09

Caixa Geral de Depositos 17-Dec-08 Norddeutsche Landesbank 18-Dec-08

Citibank NA 14-Oct-08 Northern Rock Plc 18-Feb-08

Commerzbank AG 3-Nov-08 Raiffeisen Zentralbank AG 30-Jan-09

Countrywide Financial Corp. 11-Jan-08 Royal Bank of Scotland Group 20-Oct-08

Credit Agricole CIB 20-Oct-08 SNS Bank N.V. 12-Nov-08

Danske Bank A/S 1-May-09 Société Générale 20-Oct-08

Dexia 30-Sep-08 Swedbank AB 4-Nov-08

Dexia Crédit Local SA 30-Sep-08 UBS AG 16-Oct-08

Erste Group Bank AG 30-Oct-08 US Bancorp 3-Nov-08

Fortis Bank Nederland N.V. 29-Sep-08 UniCredit SpA 18-Mar-09

Fortis Bank SA/ NV 29-Sep-08 Wachovia Corporation 29-Sep-08

Glitnir Bank 29-Sep-08 Washington Mutual Inc. 25-Sep-08

HBOS Plc 18-Sep-08 Wells Fargo & Company 28-Oct-08

Note: This table lists the banks in our dataset. The sample consists of 55 banks that have been intervened between 2007

and 2009 in Western Europe and in the USA.

Source: Laeven and Valencia (2010) and author collected data.


Table 2: Summary Statistics

Variable Observations Mean Median St. Dev. Min. Max. Min. Max.

(unwinsorized)

CDS spread 55 338.74 201.6 351.40 51.7 1331 51.7 3626

CDS spread relative to index 55 0.31 0.24 0.80 -0.79 2.01 -.79 3.03

Foreign ownership 55 0.35 0.29 0.25 0 1 0 1

Foreign assets 55 0.33 0.33 0.24 0 0.90 0 0.90

Foreign deposits 55 0.32 0.27 0.25 0 1 0 1

Size (in mil. EUR) 55 456,541 309,476 389,052 5,528 1,306,283 5,528 2,586,701

Single agency 55 0.47 0 0.50 0 1 0 1

Central bank 55 0.58 1 0.50 0 1 0 1

Tier 1 ratio 49 .08 .08 .01 .06 .12 0.05 0.21

Liquid assets 55 .21 .20 .12 .04 .40 0.01 0.57

Post-Lehman period 55 0.67 1 0.47 0 1 0 1

State ownership 55 0.16 0 0.37 0 1 0 1

Note: This table lists summary statistics of the key variables used in regressions. Definitions and sources of variables are listed

in Appendix B. Bank level variables are reported after being winsorized at the 5 percent level on both tails of the distribution

and before being taken in logs. The last two columns of the table show the minimum and maximum values of each variable

before being winsorized.


Tab

le3:

Corr

elat

ion

Mat

rix

Var

iable

s[1

][2

][3

][4

][5

][6

][7

][8

]

[1]

CD

Ssp

read

1

[2]

CD

Ssp

read

rela

tive

toin

dex

0.8

51***

1

(0.0

00

)

[3]

Fore

ign

ow

ner

ship

-0.1

64

-0.1

56

1

(0.2

31

)(0

.25

6)

[4]

Fore

ign

asse

ts0.0

33

0.0

37

0.2

42*

1

(0.8

14

)(0

.78

7)

(0.0

75

)

[5]

Fore

ign

dep

osi

ts0.1

46

0.1

28

0.1

71

0.6

91***

1

(0.2

88

)(0

.35

3)

(0.2

11

)(0

.00

0)

[6]

Siz

e-0

.419***

-0.5

38***

0.0

98

0.1

26

0.0

25

1

(0.0

01

)(0

.00

0)

(0.4

78

)(0

.35

9)

(0.8

59

)

[7]

Tie

r1

rati

o0

.24

7*

0.2

83

**

0.3

81

**

*0

.31

3*

*0

.20

2-0

.206

1.0

00

(0.0

87

)(0

.04

8)

(0.0

07

)(0

.02

8)

(0.1

65

)(0

.156)

[8]

Liq

uid

asse

ts-0

.25

1*

-0.3

06

**

-0.0

82

0.3

79

**

*0

.24

8*

0.4

19***

0.0

85

1.0

00

(0.0

65

)(0

.02

3)

(0.5

50

)(0

.00

4)

(0.0

68

)(0

.001)

(0.5

61)

No

te:

Th

ista

ble

list

sth

ep

airw

ise

corr

elat

ion

so

fse

lect

edvar

iab

les

use

din

reg

ress

ion

s.T

he

sam

ple

con

sist

sof

55

ban

ks

that

hav

ebee

n

inte

rven

edb

etw

een

20

07

and

20

09

inth

eU

.S.an

dW

este

rnE

uro

pe.

Defi

nit

ion

san

dso

urc

eso

fvar

iab

les

are

list

edin

Appen

dix

B.N

um

ber

sin

bra

cket

sin

dic

ate

p-

val

ues

and

**

*,

**

,*

corr

esp

on

dto

the

on

e,fi

ve

and

ten

per

cen

tle

vel

of

sig

nifi

can

ce.

92 Essays on Banking and RegulationT

able

4:

Reg

ula

tory

Len

ience

and

Cro

ss-B

ord

erA

ctiv

itie

s-

Bas

elin

eM

odel

log(C

DS

spre

ad)

log(C

DS

spre

ad)

-log(C

DS

index

)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

Fore

ign

ow

ner

ship

-0.9

19***

-0.9

81***

-0.9

83***

-0.8

42**

-0.7

44***

-0.8

13**

-0.8

00***

-0.9

23***

0.2

79

0.3

37

0.3

02

0.4

16

0.2

68

0.3

08

0.2

85

0.3

12

Fore

ign

dep

osi

ts0.9

31**

0.6

30*

0.3

63

0.3

35

Fore

ign

asse

ts0.7

80*

0.6

47*

0.4

25

0.3

61

Aver

age

fore

ign

asse

tan

ddep

osi

tsh

are

1.0

15**

1.3

30***

0.7

58**

1.1

31***

0.4

10

0.3

95

0.3

59

0.3

15

Siz

e-0

.293***

-0.3

13***

-0.3

05***

-0.2

14**

-0.3

45***

-0.3

61***

-0.3

54***

-0.2

96***

0.0

735

0.0

830

0.0

797

0.0

909

0.0

632

0.0

663

0.0

648

0.0

729

Post

-Leh

man

per

iod

-0.8

88***

-0.8

41***

-0.8

76***

-0.6

29*

-0.8

28***

-0.8

04***

-0.8

26***

-0.8

72***

0.1

97

0.2

05

0.2

02

0.3

16

0.1

79

0.1

76

0.1

78

0.2

18

Sta

teow

ner

ship

-0.6

51**

-0.7

02**

-0.6

97**

-0.4

59

-0.3

36**

-0.3

88**

-0.3

74**

-0.3

26*

0.2

73

0.2

96

0.2

80

0.2

97

0.1

59

0.1

81

0.1

66

0.1

83

Sin

gle

Agen

cy-0

.00760

-0.0

292

-0.0

134

0.0

479

0.0

964

0.0

849

0.0

950

0.1

17

0.1

84

0.1

96

0.1

86

0.1

85

0.1

42

0.1

47

0.1

42

0.1

55

Cen

tral

ban

k0.0

0876

-0.1

18

-0.0

349

-0.1

00

-0.2

15

-0.2

90**

-0.2

34

-0.2

21

0.2

09

0.2

13

0.2

10

0.1

83

0.1

46

0.1

42

0.1

42

0.1

42

Tie

r1

rati

o3.0

31

2.7

26

7.0

69

6.1

28

Liq

uid

asse

tsh

are

-2.2

27**

-1.2

85**

0.8

46

0.5

56

Const

ant

9.8

39***

10.2

0***

10.0

0***

8.6

56***

5.3

87***

5.6

31***

5.4

84***

4.7

26***

0.9

83

1.0

80

1.0

44

1.5

31

0.8

67

0.9

00

0.8

91

1.2

68

Obse

rvat

ions

55

55

55

49

55

55

55

49

R-s

quar

ed0.5

01

0.4

83

0.5

01

0.4

85

0.6

01

0.6

03

0.6

08

0.6

74

No

te:

Inth

ista

ble

,th

ed

epen

den

tvar

iab

les

are

the

log

of

ab

ank

CD

Ssp

read

thre

ed

ays

bef

ore

inte

rven

tio

nan

dth

elo

go

fa

ban

kC

DS

spre

adta

ken

rela

tive

toth

elo

g

of

ab

ank

CD

Sin

dex

for

the

reg

ion

wh

ere

the

ban

kis

loca

ted

.D

efin

itio

ns

and

sou

rces

of

var

iab

les

are

list

edin

Ap

pen

dix

B.

Ban

kbal

ance

shee

tre

gre

ssors

are

bas

edon

dat

afr

om

the

last

fisc

aly

ear

bef

ore

inte

rven

tio

n.

All

mo

del

sre

po

rtO

LS

esti

mat

esw

ith

robu

stst

and

ard

erro

rsan

d*

**

,*

*,

*co

rres

pond

toth

eone,

five

and

ten

per

cent

level

of

sig

nifi

can

ce.


Table 5: Anticipation Effect

4CDS spread

(1) (2) (3)

Foreign ownership 0.399 0.379 0.403

0.266 0.257 0.260

Foreign deposits -0.343

0.254

Foreign assets -0.110

0.300

Average foreign asset and deposit share -0.269

0.307

Size 0.139 0.143 0.143

0.0852 0.0860 0.0869

Post-Lehman period 0.632*** 0.603*** 0.619***

0.123 0.119 0.121

State ownership -0.0705 -0.0777 -0.0650

0.128 0.133 0.128

Single agency -0.0480 -0.0354 -0.0421

0.116 0.118 0.116

Central bank -0.100 -0.0377 -0.0682

0.101 0.106 0.105

Constant -2.242** -2.375** -2.323**

1.069 1.068 1.074

Observations 55 55 55

R-squared 0.473 0.454 0.463

Note: In this table, the dependent variables is the difference between bank CDS spread 3 days before

intervention and its value four weeks in advance. Definitions and sources of variables are listed in Ap-

pendix B. Bank balance sheet regressors are based on data from the last fiscal year before intervention.

All models report OLS estimates with robust standard errors and ***, **, * correspond to the one, five

and ten percent level of significance


able

6:

Sam

ple

Sel

ecti

on

Anal

ysi

s

CD

Ssp

read

(1)

(2)

(3)

Fir

stst

age

Sec

ond

stag

eF

irst

stag

eS

econd

stag

eF

irst

stag

eS

econd

stag

e

Fore

ign

ow

ner

ship

3.1

53***

-1.0

55***

2.9

56***

-1.0

20**

3.3

08***

-1.0

95***

1.0

98

0.4

08

0.9

79

0.4

53

1.1

11

0.4

24

Fore

ign

dep

osi

ts-3

.596***

1.1

50***

1.3

33

0.3

67

Fore

ign

asse

ts-3

.248*

0.6

81*

1.8

13

0.4

14

Aver

age

fore

ign

asse

tan

ddep

osi

tsh

are

-4.0

87**

1.0

56**

1.6

10

0.4

15

Siz

e1.0

99***

-0.3

13***

1.0

64***

-0.3

21***

1.1

31***

-0.3

21***

0.2

85

0.0

766

0.3

61

0.0

856

0.3

04

0.0

827

Post

-Leh

man

per

iod

11.3

0***

-1.0

29**

12.4

2***

-0.9

66**

12.2

6***

-1.0

17**

1.2

02

0.4

29

1.8

75

0.4

61

1.5

09

0.4

19

Sta

teow

ner

ship

2.6

80***

-0.5

80**

2.4

20***

-0.6

08**

2.6

10***

-0.6

21**

0.8

66

0.2

63

0.6

76

0.2

74

0.7

42

0.2

67

Sin

gle

agen

cy0.4

82

0.0

585

0.5

58

0.0

210

0.4

52

0.0

396

1.0

17

0.1

70

0.8

87

0.1

83

0.9

49

0.1

72

Cen

tral

ban

k0.6

06

0.0

329

0.3

10

-0.1

27

0.4

49

-0.0

420

0.8

29

0.2

07

0.7

43

0.2

05

0.7

87

0.2

03

Loan

sto

Ass

ets

rati

o1.8

21*

1.3

51*

1.4

42*

0.9

57

0.7

32

0.8

17

Non-p

erfo

rmin

glo

ans

0.1

96

0.4

19

0.3

13

0.3

36

0.3

51

0.3

40

Const

ant

-13.2

5***

10.1

0***

-11.8

9***

10.4

2***

-13.1

3***

10.3

1***

3.8

15

1.2

36

3.8

82

1.3

87

3.5

85

1.2

95

Lam

bda

0.0

0397

-0.1

02

-0.0

631

0.2

71

0.2

95

0.2

50

Obse

rvat

ions

95

95

95

95

95

95

No

te:

Th

ista

ble

rep

ort

sre

sult

sfr

om

the

firs

tan

dse

con

dst

age

of

aH

eck

man

-ty

pe

of

reg

ress

ion

wit

hm

axim

um

likel

ihood

esti

mat

ion.

The

dep

end

ent

var

iab

leis

the

log

of

ab

ank

CD

Ssp

read

thre

ed

ays

bef

ore

inte

rven

tio

n.

Lam

bd

ais

the

esti

mat

eo

fth

esa

mple

corr

ecti

on

bia

s,re

port

ed

her

ew

ith

its

stan

dar

der

ror.

Th

elo

an-t

o-a

sset

rati

oan

dth

era

tio

of

no

n-p

erfo

rmin

glo

ans

tog

ross

loan

sar

ein

clu

ded

asad

dit

ional

contr

ol

var

iable

s

inth

efi

rst

stag

e(p

rob

it)

mo

del

of

esti

mat

ion

.D

efin

itio

ns

and

sou

rces

of

var

iab

les

are

list

edin

Ap

pen

dix

B.

Ban

kbal

ance

shee

tre

gre

ssors

are

bas

edo

nd

ata

fro

mth

ela

stfi

scal

yea

rb

efo

rein

terv

enti

on

.A

llm

od

els

rep

ort

OL

Ses

tim

ates

wit

hro

bu

stst

and

ard

erro

rsan

d***,

**,

*co

rres

pond

toth

eon

e,fi

ve

and

ten

per

cen

tle

vel

of

sig

nifi

can

ce.

Chapter 3

Banks and Monetary Policy in Africa: Evidence from Uganda

Abstract: We analyze the impact of monetary policy on bank lending activities and bank asset portfolios

in Uganda. We find no evidence of a bank lending channel under the conventional credit channel view.

There is evidence of asset substitution that depends on credit market conditions abroad: changes in

foreign interest rates dictate allocations of funds among bank assets such as government securities or

foreign assets at the expense of domestic loans. Our findings have policy implications for the conduct of

monetary policy in developing countries.

3.1 Introduction

In this paper, we study how changes in monetary policy and foreign interest rates affect bank credit

provision in Uganda between 1999 and 2005. Like in many other developing countries, the central

bank in Uganda faces conditions for the conduct of monetary policy that are uncommon for developed

countries. According to an IMF report, the economy has a low degree of monetization with financial

depth (M3/GDP) of 30 percent (IMF (2003)). Moreover, the institutions necessary for the fluent conduct

of monetary policy are at an early stage of development. There is a small banking sector that is dominated

by foreign-owned banks. Further, a significant share of assets and deposits at some banks are in foreign

currency. Under these conditions, also unique for many other developing countries, there is limited

empirical evidence of how bank lending reacts to monetary policy. Is the mechanism of monetary policy

transmission the same as for developed countries? Do banks have particular ways of mitigating the

impact of monetary policy shocks? In addition, the topic of how banks adjust asset portfolios in response

to financial shocks is still not well explored. Can bank portfolio adjustments further limit the future

provision of credit? Honohan and Shi (2002) suggests that banks in emerging economies can make

loanable funds permanently unavailable to domestic borrowers by exporting these funds abroad.

While most studies on developing countries use cross-country data at annual frequency, we try to

find answers to these questions using a unique quarterly dataset on Ugandan commercial banks. The

dataset comes from the central bank in Uganda. It contains data on bank balance sheet statements and

income statements over five years. As a proxy of monetary policy shocks in Uganda, we use the return on

3-month Ugandan Treasury bonds. The Ugandan T-bill interest rate serves as the main policy indicator

of the Bank of Uganda during the sample period (IMF (2003)). Since we also need a proxy of foreign

95


interest rate shocks, we take the return on 3-month British Treasury bonds because British banks have a

historical presence in the Ugandan banking system.

We get insights about the sensitivity of bank lending in Uganda to changes in monetary policy at an

aggregated level first. An increase in the policy interest rate (monetary tightening) should decrease bank

credit in the economy. However, we find no evidence at the aggregated data level of bank loans or other

bank assets being sensitive to monetary policy in Uganda. Changes in the policy interest rate have an

insignificant impact on loan growth. Further, the policy interest rate also has an insignificant impact on

the growth in government securities or foreign assets. In contrast to these findings, we find that growth

in bank loans or other bank assets are sensitive to changes in the British T-bill rate. These results give

preliminary evidence for the transmission of foreign financial shocks to the Ugandan economy.

Because of the unique conditions for the conduct of monetary policy, we discuss and test several

empirical hypotheses in the paper under three alternative views of the bank lending channel of monetary

policy. Under the conventional credit view, the rise in the policy interest rate induces a decline in bank

liquidity (Bernanke and Blinder (1992), Kashyap and Stein (1995, 2000)). When banks cannot get

additional funds from the capital markets compensating for this liquidity shortage, they have to cut the

supply of loans. The decline, however, among banks will vary because of market frictions. Banks

that obtain additional funds at a higher cost are more financially constrained. Financial constraints are

generally unobserved but the literature has come to several proxies with bank size being the standard

one. Because smaller banks are more financially constrained, they will have to cut lending more than

larger banks after a monetary tightening. This view of the bank lending channel is the prevailing one in

developed countries.

To analyze how loan growth reacts to a monetary tightening among banks under the conventional

view, we interact changes in the policy interest rate with bank size as our measure of financial constraints.

However, the results from this empirical analysis give us weak evidence, if any, for a bank lending

channel in Uganda. Additionally, we interact changes in the policy interest rate with bank dummy

variables. This method allows us to measure the partial effect of monetary policy on loan growth at

every bank. We rank banks according to the average value of total assets during the sample period. We

compare next the coefficient estimate for every bank with its asset ranking position. Our expectation is

that we should find that loan growth at larger banks is less sensitive to changes in monetary policy than

at smaller banks. However, only at a few banks loan growth is affected by changes in the policy interest

rate, but the results are controversial. Our findings let us conclude there is no support for the conventional

view of a bank lending channel in the country at the aggregated and disaggregated data level.

Banks and Monetary Policy in Africa: Evidence from Uganda 97

Another view considers the role of foreign-owned banks in transmitting foreign monetary shocks

in the host country. Uganda is a benchmark case for the most common challenges faced by monetary

authorities in the developing world. Located in East Africa, it is a low-income open economy. The

worldwide trend of a fast increasing foreign bank dominance in the 1990s and 2000s has also affected

the country. At the end of the 1990s, 13 out of 17 banks in Uganda were foreign-owned according to our

data. Yet, there is empirical evidence showing that the dominance of foreign banks in a banking sector

can cause a disadvantage for credit supply. Cetorelli and Goldberg (2012) find evidence of a cross-border

transmission of monetary shocks. A monetary tightening in a parent bank’s country can negatively affect

the credit supply in the economy hosting its subsidiary. This is due to the existence of internal capital

markets between parent banks and their foreign subsidiaries. In a similar way, host economies can also

become vulnerable to other financial shocks from abroad (Peek and Rosengren (1997, 1999)).

In our analysis, we follow this line of research assessing whether bank access to headquarters and

foreign markets negatively affects bank credit provision in Uganda. Banks engaging in foreign activities

should be more sensitive to foreign financial shocks. In line with this hypothesis, we document an

economically significant impact of the foreign interest rate on bank loan growth when we interact bank

dummy variables with the British T-bill rate or analyze bank asset substitution. Therefore, we find

support for the cross-border transmission of foreign shocks having influence on bank lending in the

Ugandan economy. According to our results, while monetary policy by the central bank can prove to be

ineffective in controlling the supply of loans, foreign monetary policy can have real effects in Uganda as

the host economy.

Another view of the bank lending channel stresses the role of financial dollarization in bank deposits.

This view seems more suitable for the conditions faced by the central bank in Uganda. Like in other

developing countries as shown by Baliño et al. (1999), roughly one third of assets and liabilities in the

Ugandan banking system are denominated in foreign currency according to our data. However, the partial

dollarization of bank assets and liabilities can be a challenge for monetary stability (Ize and Levy-Yeyati

(2003), Levy-Yeyati (2006)). The demand and supply of foreign currency assets and deposits depends

partly on credit conditions in foreign markets. Thus, these processes are beyond the entire control of the

central bank. We follow Mora (2008) in developing and testing empirical hypotheses about the impact of

partial dollarization on monetary policy transmission. Mora (2008) shows that foreign currency deposits

can have an impact on bank liquidity in dollarized economies in the example of Mexico. Under this

view, banks can offset any liquidity shortage during a monetary tightening by converting foreign currency

deposits in the domestic currency. A bank lending channel comes into existence because the financial


constraints among banks are determined by their initial share of foreign currency deposits. Banks with

a higher share of foreign currency deposits face a lower cost of currency exchange. Being less affected

by monetary shocks, those banks should have a higher loan growth than other banks during a monetary

tightening.

In contrast to Mora (2008), we can test for the existence of this bank lending channel under conditions

of underdeveloped capital and bond markets. In Uganda, there are few companies listed on the exchange.

Auctions for the secondary trade of government treasury bills are also introduced first in the second half

of our sample period. In our results, we find support for this view of the bank lending channel. We

interact bank share of foreign currency deposits with changes in the policy interest rate following Mora

(2008)’s model. The coefficient estimate of the interaction enters statistically significant at standard sig-

nificance levels. The finding implies that central banks in developing countries should take into account

the possible impact of foreign currency deposits on the transmission of monetary policy. Banks have

a differential response to monetary policy depending upon their holdings of foreign currency deposits.

This response can lead to unpredictable outcomes for the real economy if ignored by the central bank.

Additionally, we assess the effectiveness of using changes in reserve requirements by the central

bank in influencing loan supply. Next to the policy interest rate, reserve requirements are an alternative

tool available to central banks for signaling the monetary policy stance. We consider the impact of three

different types of changes to bank reserves on lending during the sample period. In the results, we find

that changes in reserve requirements seem to be more effective as a monetary policy tool in the country

compared to using T-bill auctions in open market operations. The use of reserve requirements seems to

be useful in regulating excess liquidity in the banking sector, too.

In a final step, we analyze if banks could overcome the impact of a monetary tightening on bank

lending by using asset substitution. Excess liquidity often prevails in the Ugandan banking sector due

to donor subsidies (IMF (2003), Musinguzi and Katarikawe (2004)). If banks tend to invest in assets

abroad, those funds can become unavailable for future lending in the local economy. We thus document

bank choice on allocating excess liquidity among assets during a monetary tightening. The results of the

estimated models reveal that the foreign interest rate has a significant impact on bank portfolio composi-

tion. The share of foreign assets, i.e. assets that banks export abroad, increases at the expense of the share

of bank loans when the foreign interest rate increases. The introduction of capital controls in Uganda

can be a useful policy tool for controlling this process. Our results support a recent recommendation by

the IMF on the benefits of capital controls for the developing world (Cordero and Montecino (2010)).

The paper continues as follows. Section 3.2 documents some major developments of the Ugandan


financial system and national economy, also giving a brief review of monetary policy in Uganda. Sec-

tion 3.3 tracks briefly prior literature on the bank lending channel. Section 3.4 describes the data and

methodology, while results are given in Section 3.5. Section 3.6 summarizes the main conclusions and

comments on possible implications for policy.

3.2 Economic development and monetary policy in Uganda

Uganda is a small open economy located in East Africa, bordering Kenya, Tanzania, Rwanda, DR Congo,

and Sudan.1 Over the last two decades, the country has experienced a strong economic growth following

a period of political instability and civil conflicts throughout the 1970s and 1980s. Between 1999 and

2003, average annual GDP growth was positive though it fluctuates widely as Figure B.1 in Appendix

B shows. The agricultural sector is the largest one in the country employing about 80 percent of the

population. Figure B.2 in the appendix shows the relative contribution of the main industry sectors

to GDP in 2004. The services sector has the largest contribution to GDP with a share of 42 percent.

Infrastructure developments and the privatizations of several communication companies in the beginning

of the 2000s contribute to the expansion. The share of the agricultural sector in GDP is 34 percent,

followed by the industrial sector with a share in GDP of roughly 19 percent. A main driver of growth

in the agricultural sector is the production of coffee, which is susceptible to trade shocks. The exports

of coffee, tea and other commodities from Uganda to other countries can fluctuate widely with foreign

demand.

The government has introduced effectively programs for sustaining economic growth, fiscal con-

solidation, and poverty reduction. Foreign assistance by international organizations and donor funds

have largely contributed to this success. Macroeconomic stability and low inflation have prevailed in

the economy since 1994. While inflation in 1992 was at 26 percent, it has remained in single digits

since 1995, falling to 5 percent in 2000. Like in other African countries, poverty is widespread. Despite

higher economic growth and macroeconomic stability, Figure B.3 shows that GDP per capita in Uganda

is moderately low compared to the rest of Africa. Population was about 27 million in 2005 growing at

an average rate of about 3 percent a year, while GDP amounted to 8 billion dollars.2

The financial sector in Uganda developed fast over the last two decades as well. Between 1990 and

2004, financial depth (M3) as a percentage of GDP increased from 8 percent to 30 percent, thus pointing

1This section uses information from sources coming from the OECD (2005), IMF (2003), the World Bank (2007), and

Musinguzi and Katarikawe (2004).2The decline in population living in poverty over time, however, is notable. While 56 percent of the population was poor in

1992, the share fell to 35 percent in 2000. The country also has one of the highest population growths in the world. Statistics

show that women in Uganda on average give birth to seven children in a lifetime.


to a fast monetization of the economy. The share of gross domestic savings increased from 2 percent

in 1990 to 7.1 percent in 2003. Capital markets are also developing, though at a slower pace. Debt

and stock issuance continue to be a rare source of firm financing in the country. In 2002, for example,

only five Ugandan companies were listed on the Ugandan stock exchange. The most widespread asset

investment is in short-term government treasury bills. The absence of developed capital market rules has

an important role for banks in firm financing. Like in other developing countries, Ugandan firms rely

heavily on bank loans, thus stressing further the importance of understanding how monetary policy can

influence bank lending supply in the country.

The Bank of Uganda (BoU) has been entirely in charge of the conduct of monetary policy since 1993

when it became independent in this area from the Ugandan Ministry of Finance. The main objectives of

the bank are monetary stability and economic growth. BoU is also responsible for the supervision and

regulation of financial institutions. The central bank in Uganda is quite successful in controlling inflation

and the exchange rate within reasonable limits (IMF (2003), Musinguzi and Katarikawe (2004)). A main

challenge for preventing high inflation is the excess liquidity in the market resulting from frequent donor

funds to the government. The sterilization of excess liquidity often affects the exchange rate float and

the level of exports.

The money market instruments of the central bank include Treasury bill auctions, repos, and foreign

exchange sales.3 The 3-month Ugandan T-bills serve as a monetary policy indicator. The bank discount

rate fixes the cost of borrowing by banks from the central bank. Figure B.4 shows that the T-bill rate and

bank discount rate are, on average, high. The interest rates fluctuate widely between 1999 and 2005 in a

range of 10 to 20 percent. The central bank manages liquidity in the banking system by trading treasury

bills. Until 2004, there were weekly actions for treasury bills with a short-term maturity of up to 364

days only. The introduction of long-term government bills with a maturity of two, three or five years took

place in 2004. Secondary markets for trading in government securities were not fully developed during

the sample period.

The central bank uses bank reserve requirements as an additional policy tool, in particular for control-

ling the excess liquidity in the banking system. Until August 2000, bank reserves only covered deposits

in the domestic currency. In September 2000, the central bank added foreign currency deposits to the

reserve base of banks. Deposits in other currency are subject to the same reserve requirement for deposits

in Ugandan schillings. All reserves have to be held in the domestic currency, which are not remunerated

3Next to interventions in the exchange rate markets, the Lombard facility is another monetary policy tool. The Lombard

facility is the last-resort window of the central bank. Banks have access to this short-term lending facility from the central bank

with the right to borrow cash up to five percent of required reserves.


by the central bank. In December 2000, the central bank raised reserve requirements by one percentage

point for all types of deposits. The new reserve requirement on demand deposits was 10 percent, while

term and savings deposits were subject to a requirement of 9 percent. In June 2004, the central bank

made a further change to bank reserve requirements. The requirements on demand deposits and term and

savings deposits were equated to 9.5 percent.

In the last two decades, the Ugandan banking system has also made substantial progress. Until the

end of the 1990s, the largest bank in Uganda was the Ugandan Commercial Bank (UCB) owned by the

government. Among foreign banks, British banks have been historically present in Uganda since the

1910s. Subsidiaries of Barclays Bank and Standard Chartered Bank continued to have major market

shares in the 1990s and 2000s. Since 1987, there are no restrictions on foreign-owned banks opening

branches in the country. Banks from abroad achieved dominance in the banking sector in 1998 and 1999

when the central bank closed four domestic banks because of business violations.4 The foreign-owned

bank Stanbic acquired UCB in 2002, thus becoming the largest bank in the country.

Table B.1 compares the banking sector in Uganda with neighboring countries and other developing

countries. The banking sector in Uganda is small. The share of bank liquid liabilities in GDP is about 19

percent in 2004. The ratio of private credit to GDP is below 6 percent, which is the lowest level among

other low-income countries listed in the table. The banking sector is also marked with a high market

concentration. At the end of 2004, the four largest foreign banks accounted for about 70 percent of total

deposits in the banking system, and 60 percent of total loans.

Despite advances in the economy and the banking sector, several structural deficiencies impeded

credit provision in the country between 1999 and 2005, especially to small and medium enterprises.

There is no credit registry or a system for credit reference and national identification of borrowers in

Uganda. Moreover, commercial banks continue to keep remarkably high interest rate spreads and mar-

gins over time (Beck and Hesse (2009)).5 The net interest margin in Uganda was 13.8 percent in 2004.

Table B.1 shows that it is the highest one among the margins in neighbouring countries and other low

income countries. Even though loan contracts usually have a short-term maturity, bank credit continues

to be an important source of funding in the economy. Figure B.5 shows the distribution of lending to dif-

ferent industrial sectors of the economy between 1999 and 2005. Trade & Commerce and Manufacturing

4Two of the closed banks (Cooperative Bank and Greenland) were at that time the second and fifth largest banks in terms of

total deposits. In the beginning of 1999, the government fully compensated depositors of these banks beyond the legal insured

minimum of $2000. The privatization of UCB also took place during this time but soon after it was renationalized again because

of bad management.5Beck and Hesse (2009) identify bank level characteristics such as bank size or operating costs as determinants of the high

spreads and margins in addition to time invariant bank-level fixed effects.


sectors were among the largest users of bank credit, followed by Agriculture and Building & Construc-

tion. Since 1998, there have also been microfinance activities in the country. While introduced with the

support of donor funds, microfinance activities are not widespread. Table B.2 reports estimates showing

that the share of microfinance lending in total financial assets is less than one percent. In contrast, the

share of assets held by commercial banks in total financial assets is about 82 percent, and the share of

commercial bank assets in GDP is 24 percent.

3.3 Context

There is a vast literature on the bank lending channel under the credit view of monetary policy trans-

mission, but the evidence for developing countries is limited. Under the credit view of monetary policy

transmission, a rise in the policy interest rate (tightening) reduces the liquid funds available to banks for

lending (Bernanke and Blinder (1988, 1992)). The existence of a bank lending channel requires three

conditions (Kashyap and Stein (1995)). First, bank loans and borrowing from capital markets should

be imperfect substitutes for firms in need of external financing. When this assumption does not hold,

firms will be unaffected from a decline in loan supply. Consequently, a monetary tightening will have no

real effects on the economy. The notion of imperfect capital markets, as a departure from a Modigliani-

Miller world, is also relevant for the second requirement. Under this assumption, information asymme-

tries create an external information premium for financially constrained banks. When capital markets are

imperfect, these banks cannot offset negative shocks to deposits with other sources of financing because

the financial constraints lead to a higher external premium of borrowing. Consequently, more financially

constrained banks have to respond by cutting the supply of loans during a monetary tightening. The last

condition necessary for the existence of a bank lending channel is common to any model on monetary

policy. Prices in the economy have to adjust imperfectly on an aggregate level. If this is not the case,

any changes in nominal reserves and prices will be equiproportional. Without this assumption, monetary

policy will have no effect on real investment.

These assumptions provide the basis for testing empirical hypotheses about the existence of a bank

lending channel. During a monetary tightening, any increase in the policy rate can decrease the demand

and supply of loans at the same time. For this reason, model estimations require adequately controlling

for the demand of loans. The key identification challenge is finding an appropriate "curve shifter" in

the data. Such variable can help isolate the effect of monetary policy on loan supply from that of de-

mand. The need of financial constraints for the existence of the bank lending channel under the three

assumptions implies observing a differential response among banks to monetary policy. More financially


constrained banks should have higher cuts in the supply of loans compared to less financially constrained

banks. However, financial constraints are unobserved. Employing an appropriate proxy of bank financial

constraints can help solve the identification problem. Appropriate measures of financial constraints can

come from balance sheet data or natural experiments.6 Three standard proxies commonly used in the

literature are bank size (e.g. Kashyap and Stein (1995)), liquidity (e.g. Kashyap and Stein (2000)), and

capitalization (e.g. Peek and Rosengren (1995)). Thus, less liquid, smaller, or less capitalized banks

should be observed cutting loan supply more than other banks under this empirical hypothesis.

Evidence for the existence of a bank lending channel is found in the US based on macro datasets on

bank lending (e.g. Bernanke and Blinder (1992)) or using disaggregated data (e.g. Kashyap and Stein

(1995, 2000)). However, many other studies find that the existence and strength of the bank lending

channel varies greatly across member countries of the EU, or other developed countries (e.g. Angeloni

et al. (2003), Ehrmann et al. (2003)). In the case of developing countries, there is a recent resurgence of

interest about the existence of a bank lending channel. The broad consensus among policy makers is that

the bank lending channel of monetary transmission could take a different path for countries at an earlier

stage of development (BIS (1998)). Because of limits to data availability, however, there have only been

a few studies.

Foreign-owned banks in developing countries can face additional financial constraints arising due

to their relationships with parent banks. Cetorelli and Goldberg (2012) give evidence of a cross-border

transmission of monetary shocks affecting parent banks. Peek and Rosengren (1997, 2001) provide

similar evidence for other financial shocks. In a cross-country study on emerging economies from Asia

and Latin America, Arena et al. (2007) uses bank ownership as the proxy for bank financial constraints.

In contrast to domestic banks, foreign owned banks should be less financially constrained because of

access to financing by parent banks via internal capital markets. Therefore, foreign banks should be less

sensitive to monetary policy in host economies. The sample in Arena et al. (2007) covers 20 emerging

countries between 1989 and 2001. The cross-country study, however, gives weak evidence of a bank

lending channel under this empirical hypothesis. The results show no large differences in the lending

response among foreign and domestic banks during a monetary tightening.

A significant share of bank assets and deposits are in foreign currency in developing countries, thus

likely to influence the conduct of monetary policy in partially dollarized economies. According to esti-

mates of Baliño et al. (1999) and Levy-Yeyati (2006) for 18 emerging economies, the share of foreign

currency deposits is roughly 30% on average. As Mora (2008) shows, foreign currency deposits can

6For example, Khwaja and Mian (2008) use an event of unanticipated nuclear tests in Pakistan in 1998 to model a natural

experiment and isolate supply-driven effects of bank liquidity shocks.


indeed serve as a measure of bank financial constraints by affecting bank liquidity differentially. Some

evidence suggests that financial dollarization can also encourage the exports of capital abroad. According

to Honohan and Shi (2002), banks can make use of the differences between domestic and foreign interest

rates to maximize profits. Under the uncovered interest rate parity, a higher interest rate differential can

increase the return on foreign investments compared to the return on domestic ones. In support to this

asset substitution hypothesis, banks are found to shift a sizeable fraction of foreign currency deposits as

foreign assets in offshore accounts instead of offering domestic loans.

3.4 Data and Methodology

We use data by the central bank in Uganda on all commercial banks in the country. The dataset available

to us covers in a quarterly frequency the balance sheets and income statements of all banks over the

period between the first quarter of 1999 and the second quarter of 2005. There are in total 17 banks

during the sample period. Four of the banks are domestic, and the remaining ones are foreign ones.

We base our analysis on the following general empirical model::

4yi,t = αi + δ4rt−1 + βXt−1 + ui,t. (1)

Time is indexed by t = 1, . . . , T and banks are indexed by i = 1, . . . , N . αi stands for bank level

fixed effects. yi,t is the vector of dependent variables, including domestic currency loans, total loans,

government securities, foreign assets, and net foreign assets. Total loans include loans in domestic and

other currency. In the regression analysis, we use either bank asset shares, or asset growth rates as

dependent variables.7 The sources and definitions of all variables can be found in the appendix. The

vector ∆rt−1 in Equation 1 includes the 3-month Ugandan T-bill rate as the monetary policy indicator of

the Ugandan central bank, and the 3 month British T-bill rate as a proxy of foreign interest rates. We use

the British T-bill rate because some of the large foreign banks in Uganda are from the United Kingdom.

The interest rates in the model are taken in the first difference to resolve issues with the presence of unit

roots. The interest rates are also lagged once to alleviate the possibility of a simultaneity bias.8

In the empirical analysis, we use the three changes in reserve requirements occurring during the

sample period. We use a "Reserve increase" dummy variable that takes the value of one when the central

bank increased reserve requirements by one percentage point in December 2000, and zero otherwise. A

7Growth variables are defined as the first difference in logs. In addition, the growth variables of the assets are subject to an

outlier removal where all observations exceeding 100 percent are filtered out.8Unreported results from unit root tests confirm that the time series of the interest rates are stationary (available upon

request).


second dummy variable, "Reserve equating" takes the value of one starting with June 2004, when the

central bank equated the reserve requirements on different types of deposits to be 9.5 percent, and zero

otherwise. We also use a dummy variable for imposing bank reserve requirements on foreign currency

deposits in year 2000. The dummy "Inclusion foreign deposits" takes the value of one with the beginning

of September 2000, and zero otherwise.

The vectorXt−1 includes GDP growth, inflation, and depreciation in the exchange rate. GDP growth

controls for loan demand and business cycle fluctuations, while the other two, for macroeconomic un-

certainty and revaluation effects. The model includes quarterly and yearly dummy variables controlling

for seasonal and year fixed effects. Finally, we also control for the acquisition of the UCB bank by the

Stanbic bank in the third quarter of 2002. A "Post-merger period" dummy takes the value of one follow-

ing the merger, and zero otherwise.9 We also control for the break in Stanbic bank balance sheet data

after the acquisition. The "Stanbic merger" dummy variable takes the value of one following the event,

and zero otherwise.

We use in the analysis Driscoll and Kraay’s (1998) robust standard errors correcting for autocorre-

lation and heteroscedasticity. The standard errors also correct for spatial correlation, which is largely

ignored by other statistical methods.10 Bank mergers can lead to spatial dependence by making ob-

servations of the merging panels correlated across time. Two of the largest banks merge in 2002, thus

suggesting residuals are no longer independently distributed. Additional unreported tests available upon

request confirm the presence of spatial correlation in the data.

Table 1 shows summary statistics for the variables used in the empirical analysis. Quarterly loan

growth is about 4 percent on average, and domestic currency loan growth is about 3 percent. The standard

deviation is about 14 percent, thus loan growth varies widely across banks and over time. Quarterly

growths in foreign assets and government securities are on average about 5 and 6 percent, respectively,

thus higher compared to loan growth. The standard deviations of both growth series are high even after

the removal of outliers, thus indicating large differences among banks.11

Table 1 reports further statistics on the asset shares of banks. The asset shares vary largely among

banks though loans are on average the dominating asset in bank portfolios (34 percent). The mean share

9For details of UCB-Stanbic merger, see Clarke et al. (2007).10The Driscoll and Kraay’s method (1998) extends the Newey and West’s nonparametric variance-covariance estimator by

correcting the series of cross-sectional averages of the moment conditions. The adjustment assures the consistency of the

covariance matrix estimator by making it independent of the cross-sectional dimension N. The Driscoll and Kraay’s (1998)

method places no limiting restrictions on the number of panels. In this way, the method offers much more flexibility in small

sample estimations such as the current sample than other covariance estimators.11The similarity in the summary statistics for Foreign assets and Net foreign assets is explained by the presence of zero

foreign liabilities for most banks over the sample period.


of domestic currency loans is about 26 percent.12 Compared to this, the mean share of foreign currency

loans is significantly smaller at a level of about 9 percent. Further, the average share of foreign currency

deposits is 33 percent. Thus, banks seem unwilling to channel foreign currency deposits into foreign

currency loans, possibly anticipating the higher risk of borrower default with exchange rate depreciation.

A large devaluation of the currency can make the repaying of loans in other currency costly for domestic

borrowers.

The summary statistics also reveal that the share of government securities is second largest on average

after the share of loans in bank asset portfolios. Bank holdings of government securities are about 24

percent, which is close to the level of domestic currency loans. The average share of bank assets held

abroad is about 18 percent. Additional inspection of the data reveals no large difference in the holdings

of foreign assets among domestic and foreign banks. Further, assets due from abroad dominate over

liabilities due to abroad in the data on bank foreign balances. Most banks have zero foreign liabilities

over long periods of time. This fact explains the similarity in summary statistics for foreign assets and net

foreign assets. Because foreign liabilities are close to zero, net foreign assets are on average 17 percent,

thus close to the level of foreign assets. Furthermore, Table 1 reveals that the representative bank in

the sample tends to have on average similar shares of loans, government securities, or foreign assets.

This asset composition is commonly observed for other countries in the Sub-Sahara region. However,

it contrasts findings about the asset composition of banks in other regions (Beck and Honohan (2007,

Figure 2.7)).13 The asset composition suggests that banks in Uganda tend to balance between different

types of assets as a means of diversifying risk exposure. Among foreign assets and government securities,

lending tends to be the riskiest asset in bank portfolios. Lending rates and interest margins remain much

higher over time compared to the return on government securities or foreign assets.

Table 1 shows as well that the change of rate in the policy interest rate fluctuates widely over the

sample period. The median value of a quarterly change in the Ugandan T-bill rate is about one percent

with a standard deviation close to five percent. A closer inspection of the data shows high jumps in the

policy rate series over time. The highest peak of change in two quarters is 13.6 percent. This variation in

the changes of the policy rate poses challenges for its use of signaling the stance of monetary policy. The

excessive volatility in interest rates generally obscures the information content of interest rate changes for

monetary policy purposes. Furthermore, the macroeconomic indicators fluctuate in a reasonable range.

12The data reveals that the lowest share of domestic currency loans is held by the government-owned UCB at one point,

reaching six percent. On the other tail of the distribution is TransBank with a share of domestic currency loans of 91 percent.13Beck and Honohan (2005, Figure 2.7) reports that the share of loans in bank asset portfolios in 2005 exceeds 50 percent on

average in high-income and low-income countries from Asia, Europe, and America. The loan share for the Sub-Sahara region

is around 30 percent.


Average quarterly GDP growth is about one percent showing an economic expansion of the economy

during the sample period. Average inflation is about six percent, which is well below two-digits levels.

The exchange rate depreciation is also in a reasonable range indicating a stable currency float.

Table 2 lists correlations among the key variables used in the analysis. Most correlation estimates are

statistically insignificant, highlighting the importance of controlling for other factors in a multivariate

regression analysis to estimate the partial effects. There is no correlation between loan growth and

changes in the policy interest rate. The correlation estimate of exchange rate depreciation and loan

growth is negative and statistically significant at the ten percent level. The correlation can be explained

by the presence of foreign currency loans as a component of total loans. When the domestic currency

depreciates, foreign currency loans become costlier to borrowers and riskier investments for banks. Thus,

swings in the exchange rate can have an inhibiting effect on the demand and supply of assets in other

currency. Table 2 also shows a positive correlation between growth in foreign assets and exchange rate

depreciation. This finding can be explained by the fact that foreign assets can act as a shield against

exchange rate risk. Moreover, there is a revaluation effect because the value of foreign assets in Ugandan

Shillings increases when the currency depreciates.

Further, the correlation estimates hint on the substitutability among bank assets. There is a negative

correlation between loan growth and growth in net foreign assets. The correlation estimate of loan growth

and growth in government securities is negative as well. Table 2 also shows a negative and statistically

significant correlation between domestic currency loans and GDP growth. This finding is, however,

contrary to the view that credit supply should increase during economic booms and decrease during

recessions. The use of a multivariate regression analysis in the following section helps isolate the partial

effect of GDP growth on domestic currency loans while controlling for other factors.

3.5 Results

We test several hypotheses using our baseline model in the empirical analysis. The first set of estimated

models in Table 3 assesses at an aggregated level the sensitivity of bank assets to changes in the policy

interest rate and foreign shocks. Following a monetary tightening, the higher interest rate should induce

a decline in bank lending. The two models of loan growth in Column 1 and Column 2 provide first

evidence about the sensitivity of bank lending to changes in the Ugandan T-bill rate. The two models

allow testing if there is a cross-border transmission of monetary policy shocks as well. A higher foreign

interest rate should decrease liquidity at the parent bank. Because the funding available to subsidiaries


abroad via internal capital markets declines, loan growth in the host country should decrease.14 Under

this empirical hypothesis, the British T-bill rate should have a negative impact on loan growth. We also

gauge how growth in government securities, foreign assets, and net foreign assets respond to changes

in the Ugandan or British T-bill rate. Following movements in interest rates, banks have to outweigh

the risks in lending to domestic borrowers with minimizing the risk exposure to other assets. When the

policy interest rate increases, the return on government securities increases. Therefore, the Ugandan

T-bill rate should have a positive impact on the growth in government securities.

Turning to the results in Table 3, we find no evidence of a credit channel in Uganda. Under our

expectation, loan growth should decelerate with a rise in the policy interest rate, no matter if caused by

lower credit supply or demand. Contrary to this expectation, the coefficient estimate of the Ugandan

T-bill rate enters statistically insignificant in the estimated models in Column 1 and Column 2 of Table

3. Bank loan growth is insensitive to changes in the policy interest rate, thus suggesting the absence of

any transmission of monetary shocks.15 On the contrary, there is a statistically significant response of

bank loan growth to changes in the foreign interest rate. The coefficient estimate of the British T-bill rate

is statistically significant and has a negative sign in both models. A one percent increase in the rate of

change in the British T-bill leads to a decline of 4.7 percent in the growth of domestic currency loans.

Similarly, total loan growth decreases by 8 percent in Column 2 of Table 3. Therefore, we find evidence

of the cross-border transmission of foreign shocks.

Further, the results show that total loan growth in Uganda is sensitive to changes in inflation and

the exchange rate. For example, when the exchange rate depreciates by one percent, total loan growth

decelerates by 0.46 percent, ceteris paribus. Neither the exchange rate depreciation, nor inflation is found

to have any impact on domestic currency loans. In contrast to the correlation estimate, GDP growth is

found to exhibit a positive effect on the growth in domestic currency loans, as expected.

We use the results from the remaining three models in Table 3 to draw inference whether substitution

among non-loan assets explains why loan growth is found to be insensitive to changes in the policy

interest rate. Banks may have an incentive to reshuffle assets between government securities and foreign

assets when the central bank policy regime changes. The estimated model on government securities in

Column 3 of Table 3 finds no association between banks’ holding of Ugandan treasury bonds and the

change in the Ugandan T-bill rate. However, there is a statistically significant response of government

14Note also that a large fraction of the foreign assets of Ugandan banks are deposits with parent or correspondent banks

abroad (IMF (2003)), thus making up an additional source of liquidity. When banks export these deposited funds abroad, they

move from the liability side of the bank balance sheet to the asset side.15Additional robustness tests provide further support to those findings. Similar findings are obtained when using feasible

GLS-type of estimation assuming about autocorrelation and heteroscedasticity in errors and panels.


securities holdings to changes in the British T-bill rate. The coefficient estimate of the foreign interest

rate is statistically significant at the 5 percent level, and economically significant. When the British T-bill

rate changes up by one percent, the quarterly growth in government securities held by banks accelerates

on average by 13 percent in the following quarter, ceteris paribus. Therefore, the findings in Columns

1-3 of Table 3 suggest that independent of the policy interest rate, banks prefer to invest in short-term

government securities when the foreign interest rate rises. In addition, banks cut the provision of credit.

The estimated model in Column 3 gives some further evidence that the growth in government se-

curities decreases during economic booms. The coefficient estimate of GDP growth enters statistically

significant at the 10 percent level. It shows that when GDP growth increases by 1 percent, the growth

in government securities held by banks decreases on average by 0.85 percent. This finding is expected

since the number of positive NPV projects increases during booms of the business cycle. Moreover,

government securities can be a safe buffer of excess liquid funds during recessions. When the economic

climate improves, it is reduced by funding other bank activities.

While banks may invest in government securities at the expense of loans, they also have a choice to invest

in foreign assets. The estimated models in Column 4 and Column 5 of Table 3 regress the growth in bank

foreign assets and net foreign assets on changes in interest rates. Because of similarities in model esti-

mates, we only comment on the regression for growth in foreign assets given in Column 4. The results

provide some evidence that higher policy interest rates accelerate the growth of banks’ foreign assets.

The coefficient estimate of the Ugandan T-bill rate enters the estimated model in Column 4 statistically

significant at the five percent level. When the change of rate in the domestic policy interest rate increases

by one percent, growth in foreign assets increases by 0.30 percent, ceteris paribus. This effect is econom-

ically small. Even though, it shows that banks can export funds abroad despite the shortage of liquidity

resulting from monetary policy. Further, the coefficient estimate of the British T-bill rate is statistically

insignificant at standard significance levels. Overall, changes in interest rates do not seem to be a major

determinant of growth in bank foreign assets according to these findings.

GDP growth, inflation, and depreciation in the exchange rate that control for macroeconomic condi-

tions have a much stronger impact on bank asset portfolios. According to the estimated model in Column

4, one percent increase in GDP growth is associated on average with a decrease in the growth of bank

foreign assets by 0.77 percent, ceteris paribus. Thus, like government securities, foreign assets can serve

a role as a liquidity buffer to banks in recessions. Banks can use this buffer for funding domestic in-

vestment projects when the economic conditions in the Ugandan economy improve. Depreciation in the

exchange rate is found to have the largest effect on foreign assets, which can be explained by a revalua-


tion effect. Its coefficient estimate enters statistically significant at the one percent level. The estimated

model shows that when the local currency depreciates by 1 percent, ceteris paribus, the growth in bank

foreign assets accelerates by 1 percent on average.

Table C1 in Appendix C reports results for the estimated models in Table 3 with standard errors

clustered at the year level. Clustering at the year level is an alternative method to using Driscoll and

Kraay’s (1998) robust standard errors. Year clustering can control for spatial correlation but not for

serial correlation. The results in Table 3 and Table C1 are qualitatively similar for the variables of

interest. Further, Table C2 in the appendix reports model estimations when GDP growth is excluded

from the set of control variables. The results in the table suggest that using seasonal and year dummy

variables is sufficient to control for loan demand. However, a comparison of Table C2 and Table 3 shows

that including GDP growth leads to a modest increase in the adjusted R-squared, thus improving its fit.

In addition, the economic significance of the estimates for the foreign interest rate also decreases in the

models in Table C2. Overall, Tables C1 and C2 show that the results in Table 3 are robust to alternative

assumptions about the standard errors or using GDP Growth as a proxy of loan demand in a model with

seasonable and year dummy variables.

Given our findings in Table 3, we analyze next which of the three views of a bank lending channel

tend to hold for the case of Uganda. Mora (2008) argues that banks with a higher share of foreign cur-

rency deposits will be less affected by a monetary tightening than others facing a lower cost of currency

exchange. Converting deposits in foreign currency into the domestic currency can help compensate for

any shortage in liquidity because of a monetary tightening. This can possibly explain why we find no

response of bank credit to changes in the rate of the Ugandan T-bill rate at the aggregated data level. To

assess the role of foreign currency deposits in the transmission of monetary policy shocks in the Ugandan

banking system, we follow Mora (2008)’s empirical model. We include the share of foreign currency

deposits in the baseline model adding also its interaction with the rate of change in the policy interest

rate. We expect to find that banks with a higher share of foreign currency deposits will be less sensitive

to changes in the policy interest rate.

The estimated models of domestic currency loans and total loans in Column 1 and Column 2 of Ta-

ble 4 give evidence supporting this empirical hypothesis. The interaction of the foreign currency deposit

share with the Ugandan T-bill rate enters statistically significant in the model of domestic currency lend-

ing in Column 1 with a negative sign. Thus, the holdings of foreign currency deposits tend to reduce the

effect of the policy interest rate on bank liquidity. Furthermore, the partial effects of the policy interest

rate and the interaction with the deposit share tend to offset each other. Mora (2008)’s findings for Mex-


ican banks are similar suggesting that banks initially holding a high share of foreign deposits during a

monetary tightening experience a stronger loan growth. The models also yield similar results after ac-

counting for the changes in reserve requirements as shown by Column 3 and Column 4 of Table 4. Table

C3 in Appendix C reports further how the marginal effect of monetary policy on bank lending in Column

1 varies with the bank share of foreign currency deposits. The marginal effect is statistically significant

at the one percent level and at the highest value when the deposit share is close to zero. The economic

significance of the estimate decreases at the mean value of foreign currency deposits at Ugandan banks,

which is 33 percent according to the summary statistics in Table 1. Any further increase in foreign

currency deposits leads to a lower marginal effect of monetary policy on loan growth in economic and

statistical significance. The marginal effect is weakly significant when the share of foreign currency de-

posits is 50 percent. The estimate loses statistical significance when the share reaches 70 percent, which

is the maximum value of the range for foreign currency deposits according to Table 1. In addition, Table

C4 reports estimates of models including an interaction between the share of foreign currency deposits

and changes in the foreign interest rate. The results for the policy interest rate are qualitatively similar

in Table 4 and Table C4. The interaction between the foreign interest rate and share of foreign currency

deposits enters with a statistically insignificant coefficient estimate in all models.

Overall, the results from the two estimated models in Table 4 suggest that the bank lending channel

with foreign currency deposits works by inhibiting domestic currency loan growth only. In the model

of total loans in Column 2 of Table 4, the coefficient estimate of the interaction between the foreign

currency deposit share and the change of rate in the policy interest rate enters statistically insignificant.

It should be noted that total loans also include loans in foreign currency, which supply is independent of

fluctuations in the domestic policy interest rate. This could explain why the estimated model achieves a

statistically insignificant coefficient estimate.

We proceed to analyze whether we can find support for the conventional view of monetary policy in

Uganda. Under the credit view, more financially constrained banks experience lower loan growth during

a monetary tightening. Our proxy of financial constraints is bank size. Under this hypothesis, larger

banks should be able to withstand better any liquidity shortage induced by monetary shocks. To test this

hypothesis, we interact changes in the policy interest rate with bank size and report the results in Table 5.

The coefficient estimates of bank size and the interaction are statistically insignificant in the models of

domestic currency loan growth in column 1 and total loan growth in Column 2 of the table. Thus, there

is no evidence of a bank lending channel at the aggregated level.

The results in Table 5 can be statistically insignificant because of a non-linear relationship between


bank size and the policy interest rate. We extend the model in Table 5 by interacting changes in the policy

interest rate with three indicators of bank size in Table C5 in Appendix C. The indicators of bank size

are constructed to take the value of one if a bank is in a given tercile of the asset distribution in every

time period, and zero otherwise. Banks in lower terciles of the asset distribution are expected to be more

sensitive to changes in the policy interest rate under the conventional view of a bank lending channel.

The results in Table C5 however show no evidence supporting the conventional view. The interactions

of bank size with changes in the policy interest rate enter the models statistically insignificant, and the

results for the interactions of bank size with the foreign interest rate are similar.

Responses of individual banks can differ from the response of the average bank in a systematic way

as discussed by Ehrmann et al. (2003). We extend the model on loan growth in Table 3 by adding

interactions between bank level fixed effects and the rate of change in the policy interest rate in Table C6

in Appendix C. This method allows us to measure the partial effect of monetary policy on loan growth

at every bank and test for the existence of a bank lending channel at an individual bank level. We order

banks in the first column to the left in Table C6 according to the average of total assets for every bank

over the sample period. The ranking is in an increasing order starting with the smallest banks and ending

with the largest ones. Next to the ranking, we report the estimates when interest rates are interacted

with bank dummy variables.16 We compare the size of a bank according to the asset ranking with the

coefficient estimate of the variables interaction. We expect to find that loan growth at the larger banks

should be insensitive to the policy interest rate.

The results in Table C6 gauge the variation of monetary policy effectiveness across banks of different

sizes. The first two columns in Table C6 list models of domestic currency loan growth with the Ugandan

T-bill rate (Column 1) and British T-bill rate (Column 2). The next two columns give estimates of the

models of total loan growth. The results in Column 1 and Column 3 of Table C6 provide weak evidence

of a transmission of domestic monetary shocks to domestic currency lending and total lending. Loan

growth at most banks is insensitive to changes in the rate of the Ugandan T-bill rate. The coefficient

estimate is statistically significant at four banks but of different size. Loan growth decelerates at one

bank during a monetary tightening that is below the median in asset holdings. This is the only finding

consistent with the conventional bank lending channel view. The coefficient estimates for the other three

banks have a positive sign opposite to the view of financially constrained banks. Similarly, there are no

16We also assume that the possible impact of loan demand is sufficiently well captured by the use of GDP growth and the

other macroeconomic variables in the empirical model. Under this assumption, the interactions of bank dummies with domestic

interest rates can be indicative of supply-driven changes in loan growth. We omit the coefficient estimates for the other control

variables in Apendix Table C#. The complete regression output is available upon request.


significant reactions of total loan growth given the estimates in Column 3 of Table C6. On the whole,

the lack of many responses to changes in the policy interest rate suggests the absence of a strong bank

lending channel of monetary policy, if any.

We contrast these results with the response of bank loan growth to foreign interest rates in columns 2

and 4 in Table C6. Most banks are insensitive to changes in the British T-bill rate. But its impact on loan

growth at the few banks with statistically significant coefficient estimates is economically significant.

The growth in total loans at three of the largest commercial banks in the banking system decelerates by

more than 1 percent following a change in the British TBill rate by 0.10 percent. For example, a change

in the British T-bill rate by 0.10 percent decelerates loan growth in domestic currency at Bank 13 by 2.8

percent, ceteris paribus, and its total loan growth by 2.3 percent. Thus, credit market conditions abroad

tend to dictate credit supply in the Ugandan economy, further supporting the view of a cross-border

transmission of monetary shocks. While the model estimates in Table 4 stress the need to consider the

impact of foreign currency deposits on monetary policy, the estimated models in Table 5 and Appendix

Tables C5 and C6 give no evidence of a bank lending channel under the conventional credit market view.

In addition to analyzing the impact of monetary policy on bank lending, we study next how changes in

interest rates affect the composition of bank asset portfolios. Honohan and Shi (2002) argue that loanable

funds can become unavailable for lending in the local economy permanently, once being exported out

of the country. Because this can hinder the future supply of loans in the economy, we study the impact

of monetary policy on asset substitution in Table 6. We report estimated models of the growth in asset

shares on the policy interest rate under this hypothesis. We use seemingly unrelated regression methods

taking into account the accounting identity of the dependent variables. Because monetary policy changes

the return on domestic assets, we expect to find a positive effect of the policy rate on the growth in foreign

assets share, but a negative one in the model of loan share growth.

The reported results in Columns 1-3 of Table 6 show that the estimates for the policy interest rate

is statistically insignificant in the models of loan growth and government securities, although having

an expected sign. Further, growth in the shares of foreign assets and net foreign assets are found to

be sensitive to changes in the policy interest rate as given by Columns 4-5 of Table 6. Foreign assets

growth accelerates by 0.10 percent when the rate of change in the Ugandan T-bill rate increases by one

percent, ceteris paribus. Moreover, there is a support for the impact of the foreign interest rate on bank

asset shares and asset portfolio composition. The British T-bill rate is found to exert a positive effect on

government securities and foreign assets. A one-percent increase in the change of the British T-bill rate

leads to a growth in the share of government securities by 4 percent, and foreign assets, by 3 percent.


These findings suggest that the foreign interest rate changes the relative cost and risky prospects of

loans compared to other assets. With a tightening in foreign monetary conditions, banks seem to prefer

either investing abroad, or making investments in short term government securities. Thus, credit market

conditions from abroad can dictate the substitutability among bank assets and can constrain the future

provision of credit.

While we find some evidence suggesting asset substitution at the expense of loans following a mon-

etary tightening, we proceed by using models of bank asset shares to judge the degree of substitutability.

Further, we compare how changes in reserve requirements fare next to the policy interest rate as a mone-

tary policy tool by adding the event dummy variables to the models. We drop the year dummy variables

from the model with the inclusion of these variables.

The reported results in Table 7 give evidence of reserve requirements having a significant impact on

bank asset composition. The rise in reserve requirements by one percentage point has decreased the share

of domestic currency loans by about 10 percent compared to the period preceding the policy change. The

share of government securities has increased by 13.2 percent, ceteris paribus, according to the model in

Column 4 of Table 7. It has also decreased the share of foreign assets by about 4 percent according

to the model in Column 4. Further, equating the reserve requirements on different types of deposits at

commercial banks in 2004 had only an impact on total loans and foreign assets. The coefficient estimate

of the "Reserve Equating" dummy variable enters statistically significant in the estimated models at the

one or five percent level showing that the share of total loans has increased by about 2 percent in the

period after the change (Column 2), and the share of foreign assets has decreased by about 4 percent

(Column 4).

The event of subjecting foreign currency deposits to a reserve requirement has influenced the substi-

tutability of domestic loans with foreign assets as well. The reserve requirement considers the domestic

currency equivalent value of these deposits that also applies to deposits in Ugandan Shillings. Therefore,

swings in the exchange rate determine what fraction of foreign currency deposits should be added in the

reserve base every month. The "Inclusion foreign deposits" dummy variable in Table 7 enters with a

coefficient estimate that is statistically significant at the one percent level in all estimated models except

the one on government securities. The loan shares decline by 3 and 4 percent compared to the period

before the event according to Columns 1 and 2. Similarly, the shares on foreign and net foreign assets

increase by approximately 4 percent in Columns 4 and 5. Also, we find that stronger currency depreci-

ation leads to credit cuts and more investment in government securities as short-term liquid assets. The

results in Table 7 support the previous findings about the impact of interest rates on asset allocations.


In addition, changes in the British T-bill rate have a negative impact on the shares of domestic currency

loans and total loans while having a positive impact on the shares of government securities and foreign

assets. In contrast to the foreign interest rate, the estimates of the policy interest rate are statistically or

economically insignificant in all estimated models.

3.6 Conclusions

Our paper is among the first to analyze in detail the interaction between domestic bank credit and mon-

etary policy in a country from the Sub-Saharan Africa region. Monetary policy in Uganda as proxied

by changes in the Ugandan 3 Month T-Bill rate is found to have an either statistically insignificant or

economically negligible impact on loan growth under the conventional bank lending channel view. In

contrast to this, we find that conditions on the international credit market as proxied by the British T-Bill

rate seem to exert significant influence on the provision of credit in Uganda. There is also a support for

the existence of a bank lending channel that depends on the degree of deposit dollarization. Consistent

with Mora (2008), we find that the impact of monetary policy shocks on domestic currency loan growth

is larger for banks with a smaller share of foreign currency deposits.

The results have policy implications for the conduct of monetary policy in developing countries. Our

findings suggest that central banks have to take into account the degree of dollarization of the banking

system in the conduct of monetary policy. As already stated, the effect of monetary policy varies at each

bank according to the share of foreign currency deposits. A bank with a higher share of foreign deposits

is better able to withstand monetary policy shocks relative to other banks.This distributional effect on

loan growth can have important consequences for the real economy if not adequately addressed by the

central bank.

The evidence of a cross-border transmission of foreign monetary shocks and asset substitution with a

negative impact on loans tend to support the view of the International Monetary Fund about the benefits of

capital controls (Cordero and Montecino (2010)). Since developing countries depend largely on capital

flows such as donor subsidies from abroad, they are sensitive to any short term fluctuations in capital

movements. A policy of capital controls can stabilize the real exchange rate and price levels by limiting

capital movements, and they can also foster competitiveness and economic growth. We point to an

additional benefit of capital controls. The restrictions in capital movements may also motivate banks to

channel more liquid funds in the form of credit to local borrowers. Our results suggest that when banks

export loanable funds abroad, they may become permanently unavailable for lending in future periods.

To the extent that capital controls constrain banks’ ability of investing in other markets, they can also


prevent future declines in the supply of credit. Since we also find that not only foreign assets but also

government securities and loans seem to act as substitutes of each other, capital controls can generally

decrease the substitutability among all bank assets. In turn, when bank assets act as imperfect substitutes,

the overall effectiveness of monetary policy should improve.


References

[1] Angeloni, I ., Kashyap, A. and B. Mojon 2003, Monetary Policy Transmission in the Euro Area. A

Study by the Euro system Monetary Transmission Network, Cambridge University Press, 514 pp.

[2] Arena, M., C. Reinhart, and F. Vazquez, 2007. The Lending Channel in Emerging Economies: Are

Foreign Banks Different?. IMF Working Papers, Vol. , pp. 1-52, 2007.

[3] Baliño, T., A. Bennett, and E. Borensztein, 1999. Monetary Policy in Dollarized Economies. Inter-

national Monetary Fund Occasional Paper No. 171.

[4] Bernanke, B. and A. Blinder, 1988. Credit, Money, and Aggregate Demand. American Economic

Review, May 1988 (Papers and Proceedings), 78(2), pp. 435–39.

[5] Bernanke, B. and A. Blinder, 1992. The Federal Funds Rate and the Channels of Monetary Trans-

mission. American Economic Review, September 1992, 82(4), pp. 901–21.

[6] Beck T. and H. Hesse, 2009. Why are interest spreads so high in Uganda? Journal of Development

Economics, Volume 88, Issue 2, March 2009, Pages 192-204.

[7] Beck T. and P. Honohan, 2007. Making Finance Work for Africa. World Bank, Washington, Febru-

ary 2007.

[8] BIS Report, The Transmission of Monetary Policy in Emerging Market Economies. Policy Papers

No. 3. 1998

[9] Cetorelli, N. and L.Goldberg, 2012. Banking Globalization and Monetary Transmission. Journal of

Finance, pp. 1811-1843.

[10] Clarke, G. R.G., Cull, R., and M. Fuchs, 2007. Bank Privatization in Sub-Saharan Africa: The Case

of Uganda Commercial Bank, Policy Research Working Paper 4407, The World Bank Development

Research Group.

[11] Cordero, J. A. and J. A. Montecino, 2010. Capital Controls and Monetary Policy in Developing

Countries. CEPR Reports and Issue Briefs 2010-10, Center for Economic and Policy Research

(CEPR).

[12] Driscoll, J. C., and A. C. Kraay. 1998, Consistent Covariance Matrix Estimation with Spatially

Dependent Panel Data.Review of Economics and Statistics 80: 549–560.

[13] Ehrmann, M., L. Gambacorta, J. Martínez-Pagés, P. Sevestre, and A. Worms, 2003. Financial sys-

tems and the role of banks in monetary policy transmission in the Euro Area. In: Angeloni, Ignazio,

Kashyap, Anil, Mojon, Benoit K. (Eds.), Monetary Policy in the Euro Area. Cambridge University

Press, pp. 235–269.

[14] Honohan, P. and A. Shi, 2002. Deposit dollarization and the financial sector in emerging economies.

Policy Research Working Paper Series 2748, The World Bank.

[15] IMF, 2003. Uganda: Financial System Stability Assessment Report. IMF country Report No. 03/97.

[16] Ize, A. and Levy-Yeyati, E., 2003. Financial Dollarization. Journal of International Economics, Vol.

59, pp. 323-347, 2003.


[17] Kashyap, A. and J.C.Stein, 1995. The Impact of Monetary Policy on Bank Balance Sheets.

Carnegie-Rochester Conference Series on Public Policy, June 1995, 42, pp. 151–95.

[18] Kashyap, A. and J.C.Stein, 2000. What Do a Million Observations on Banks Say about the Trans-

mission of Monetary Policy?American Economic Review, 90(3): 407–428.

[19] Khwaja, A. I. and A. Mian, 2008. Tracing the Impact of Bank Liquidity Shocks: Evidence from an

Emerging Market. American Economic Review, vol. 98(4), pages 1413-42, September.

[20] Levy-Yeyati, E., 2006. Financial dollarization: evaluating the consequences. Economic Policy, 21:

61–118.

[21] Musinguzi, Polycarp and Katarikawe, Mary. "Monetary Policy Frameworks in Africa: The Case of

Uganda", Bank of Uganda Working Paper, 2004.

[22] Mora, N., 2008. The Bank Lending Channel in a Partially Dollarized Economy: The Case of Mex-

ico, Working paper

[23] OECD, 2005. African Economic Outlook 2004/2005: Uganda. African Economic Outlook 2005,

OECD Publishing.

[24] Peek, J. and E.Rosengren, 1997. The International Transmission of Financial Shocks: The Case of

Japan.American Economic Review, September 1997, 87(4), pp. 495–505.

[25] Peek, J. and E.Rosengren, 2000. Collateral Damage: Effects of the Japanese Bank Crisis on Real

Activity in the United States. American Economic Review 90, no. 1 (March): 30-45.

[26] World Bank, 2007. Uganda: Moving Beyond Recovery: Investment & Behavior Change, For

Growth, Report No. 39221-UG.


Appendix A: Variable Definitions

Variable Description Source

Loans (in LCU) Log difference in bank loans (in dom. currency). BoU; authors’ calculations

Loans Log difference in bank total loans. BoU; authors’ calculations

Government Securities Log difference in bank holdings of govern-

ment securities.

BoU; authors’ calculations

Foreign Assets Log difference in bank foreign assets. BoU; authors’ calculations

Net Foreign Assets Log difference in bank net foreign assets. BoU; authors’ calculations

DL/TA Share of bank domestic currency loans in

total assets.


TL/TA Share of bank total loans in total assets. BoU; authors’ calculations

FA/TA Share of bank foreign assets in total assets. BoU; authors’ calculations

NFA/TA Share of bank net foreign assets in total as-

sets.


4rd Return on 3 month Ugandan Treasury bills;

difference in logs.


4rf Return on 3 month UK Treasury bills; dif-

ference in logs.

IMF statistics

Post-Merger period A dummy that takes the value of one after

the merger of UCB and Stanbic bank in the

third quarter of 2002, and zero otherwise.

authors’ calculations

Stanbic Merger A dummy that takes the value of one for

Stanbic bank after its merger with UCB in

the third quarter of 2002, and zero other-

wise.

authors’ calculations

GDP growth Log difference in GDP index. BoU; authors’ calculations

Inflation Log difference in CPI index. BoU; authors’ calculations

Reserve Increase A dummy that takes the value of one start-

ing from December 2000, and zero other-

wise.


Reserve Equating A dummy that takes the value of one start-

ing from June 2004, and zero otherwise.


Inclusion Foreign Deposits A dummy that takes the value of one start-

ing from September 2000, and zero other-

wise.


Note: This table provides an overview of sources and definitions of the key variables used in the analysis. BoU indicates Bank

of Uganda.


Appendix B: Figures and Tables on Economic Development of Uganda

10

50

510

GD

P G

row

th

1999q1 2000q3 2002q1 2003q3 2005q1

Figure B.1: GDP growth in Uganda (1999-2005)

Note: This figure shows growth in real GDP in Uganda between 1999 and 2005.

Source: BoU and author’s calculations.

Figure B.2: GDP by Sector in 2003/2004

Note: This figure reproduces Figure 3 from OECD (2005) showing the contri-

bution of industry sectors in Uganda to GDP. Source: OECD (2005).


Figure B.3: GDP per capita in Africa and Uganda (current $)

Note: This figure reproduces Figure 2 from OECD (2005) showing the GDP

per capita in Africa and Uganda. Source: OECD (2005).

Figure B.4: Monetary Policy Indicators in Uganda (1999-2005)

Note: This figure shows the nominal T-bill rate and discount rate between 1999

and 2005. Source: BoU and author’s calculations.


Figure B.5: Sectoral Distribution of Lending in 1999/2005

Note: This figure shows the distribution of lending to different industrial sectors

between 1999 and 2005. Source: BoU and author’s calculations.


Table B1: Financial Intermediation across countries in 2004

Variable Private Credit/ Liquid Liabilities/ Bank deposits/ Loan/ Net Interest Overhead

GDP GDP GDP Deposit Ratio Margin Costs

Uganda 5.90% 19.00% 14.80% 39.90% 13.40% 9.10%

Kenya 25.30% 39.80% 32.90% 73.20% 6.70% 5.70%

Tanzania 7.80% 22.10% 16.70% 46.70% 7.70% 6.40%

Sub-Saharan Africa 17.80% 30.80% 24.20% 66.00% 8.30% 6.70%

Low income countries 14.70% 29.40% 22.10% 65.60% 7.50% 6.20%

Note: This table reproduces Table 1 from Beck and Hesse (2008). Private Credit/GDP is total claims of financial institutions on

the domestic private non-financial sector as share of GDP. Bank deposits/GDP is total deposits in deposit money banks as share of

GDP. Liquid Liabilities are liquid liabilities of the financial system (currency plus demand and interest-bearing liabilities of banks and

nonbank financial intermediaries) as a share of GDP. Bank deposits/GDP is the ratio of demand, time and savings deposits in money

banks to GDP. Loan-deposit ratio is the aggregate ratio of lending to the private sector to total deposits for deposit money banks. The

data is from the International Financial Statistics (IFS)). Overhead costs are banks’ operating costs relative to total earning assets.

Interest margin is the net interest revenue relative to total earning assets. Source: Beck and Hesse (2008).

Table B2: Financial System Assets as of June 2002

Variable Assets % of % of

(U. Sh. billions) Assets GDP

Commercial banks 2450 81.80% 24.10%

NBFI 200 6.70% 2.00%

Pension 254 8.50% 2.50%

Insurance 60 2.00% 0.60%

Microfinance 30 1.00% 0.30

Total assets of the financial system 2994 100% 29.50%

Note: This table reproduces Table 2 from IMF (2003). Source: IMF (2003).


Table 1: Summary Statistics

Variable Observations Mean Median St. Dev. Min. Max.

Loans (in LCU) 396 0.028 0.030 0.149 -0.733 0.632

Loans 396 0.039 0.034 0.139 -0.615 0.938

Government Securities 358 0.057 0.038 0.286 -0.999 1

Foreign Assets 365 0.046 0.033 0.319 -0.773 0.958

Net Foreign Assets 359 0.049 0.037 0.323 -0.837 0.958

Share Foreign Deposits 396 0.326 0.349 0.151 0.007 0.692

DL/TA 396 0.256 0.248 0.102 0.057 0.669

TL/TA 396 0.342 0.323 0.151 0.057 0.914

GS/TA 396 0.237 0.219 0.145 0 0.642

FA/TA 396 0.180 0.174 0.111 0 0.468

NFA/TA 396 0.173 0.173 0.110 -0.162 0.468

4rd 396 0.089 1.100 5.216 -13.600 9.600

4rf 396 -0.039 0.010 0.326 -0.760 0.600

GDP Growth 396 0.012 0.009 0.044 -0.081 0.075

Exchange Rate Depreciation 396 0.010 0.011 0.040 -0.074 0.146

Inflation 396 0.006 0.005 0.026 -0.062 0.061

Note: This table lists summary statistics of the key variables used in regressions. Definitions and sources of

variables are listed in Appendix A. The asset growth variables are reported after the removal of outliers.

Banks and Monetary Policy in Africa: Evidence from Uganda 125T

able

2:

Corr

elat

ion

Mat

rix

Var

iable

s[1

][2

][3

][4

][5

][6

][7

][8

][9

][1

0]

[1]

GD

PG

row

th1.0

00

[2]

Exch

.R

ate

Dep

reci

atio

n-0

.288***

1.0

00

(0.0

00)

[3]

Infl

atio

n0.4

20***

-0.1

61***

1.0

00

(0.0

00)

(0.0

01)

[4]

4r d

-0.1

44***

0.2

69***

0.1

44***

1.0

00

(0.0

04)

(0.0

00)

(0.0

04)

[5]

4r f

0.0

97*

-0.1

83***

0.3

04***

-0.1

69***

1.0

00

(0.0

54)

(0.0

00)

(0.0

00)

(0.0

01)

[6]

4L

oan

Gro

wth

(in

LC

U)

-0.0

85*

-0.0

51

-0.0

26

0.0

11

-0.0

06

1.0

00

(0.0

93)

(0.3

16)

(0.6

08)

(0.8

22)

(0.9

07)

[7]

4T

ota

lL

oan

Gro

wth

-0.0

42

-0.0

93*

0.0

02

0.0

09

-0.0

12

0.8

20***

1.0

00

(0.4

08)

(0.0

65)

(0.9

62)

(0.8

52)

(0.8

06)

(0.0

00)

[8]

4G

ov.

Sec

uri

ties

Gro

wth

0.0

52

-0.0

33

0.0

07

-0.0

53

-0.0

51

-0.1

92***

-0.1

56***

1.0

00

(0.3

30)

(0.5

30)

(0.8

91)

(0.3

16)

(0.3

37)

(0.0

00)

(0.0

03)

[9]

4F

ore

ign

Ass

ets

Gro

wth

-0.0

37

0.1

17**

-0.0

65

0.0

18

-0.0

37

0.0

19

-0.0

93*

0.0

34

1.0

00

(0.4

81)

(0.0

26)

(0.2

16)

(0.7

26)

(0.4

86)

(0.7

16)

(0.0

76)

(0.5

33)

[10]

4N

etF

ore

ign

Ass

ets

Gro

wth

-0.0

85

0.1

25**

-0.1

32**

0.0

12

-0.0

72

-0.0

07

-0.1

25**

0.0

46

0.9

44***

1.0

00

(0.1

09)

(0.0

18)

(0.0

13)

(0.8

28)

(0.1

76)

(0.9

01)

(0.0

17)

(0.3

99)

(0.0

00)

No

te:

Th

ista

ble

list

sth

ep

airw

ise

corr

elat

ion

so

fse

lect

edvar

iab

les

use

din

reg

ress

ion

s.D

efin

itio

ns

and

sou

rces

of

var

iab

les

are

list

edin

Appen

dix

A.N

um

ber

sin

bra

cket

s

ind

icat

ep

-val

ues

and

**

*,

**

,*

corr

esp

on

dto

the

on

e,fi

ve

and

ten

per

cen

tle

vel

of

sig

nifi

can

ce.


able

3:

Ass

ets

Gro

wth

and

Mac

roec

onom

icC

ondit

ions

4L

oan

Gro

wth

4T

ota

lL

oan

4G

ov.

Sec

uri

ties

4F

ore

ign

Ass

ets

4N

etF

ore

ign

(in

LC

U)

Gro

wth

Gro

wth

Gro

wth

Ass

ets

Gro

wth

(1)

(2)

(3)

(4)

(5)

GD

PG

row

th0.2

57*

0.1

44

-0.8

46*

-0.7

65***

-0.6

33**

0.1

36

0.1

17

0.4

52

0.2

45

0.2

53

Exch

.R

ate

Dep

reci

atio

n-0

.077

-0.4

62***

0.8

43

1.3

64***

1.1

11***

0.1

49

0.1

15

0.5

53

0.1

80

0.3

51

Infl

atio

n0.4

16

0.6

85***

1.1

51

-0.9

65*

-1.9

24**

0.2

75

0.2

24

1.0

74

0.4

69

0.7

16

4r d

-0.0

0001

-0.0

01

0.0

02

0.0

03**

0.0

02

0.0

01

0.0

01

0.0

03

0.0

02

0.0

02

4r f

-0.0

47**

-0.0

79***

0.1

31**

0.0

29

-0.0

10

0.0

19

0.0

13

0.0

59

0.0

34

0.0

54

Sta

nbic

Mer

ger

0.0

89

0.0

87

0.0

41

0.1

03

0.1

68

0.0

66

0.0

50

0.1

07

0.0

95

0.1

12

Post

Mer

ger

Per

iod

0.1

24***

0.0

88***

-0.1

13**

-0.1

02**

-0.0

80

0.0

14

0.0

11

0.0

42

0.0

36

0.0

49

Const

ant

-0.0

56***

-0.0

49***

0.0

04

0.0

46*

0.1

87***

0.0

12

0.0

12

0.0

59

0.0

23

0.0

44

Obse

rvat

ions

394

394

357

364

358

R2

0.1

00

0.1

01

0.0

776

0.0

866

0.1

17

Num

ber

of

ban

ks

17

17

17

17

17

No

te:

All

reg

ress

ion

sin

clu

de

ban

k-l

evel

fixed

effe

cts

and

seas

on

alan

dy

ear

du

mm

yvar

iab

les

no

tre

po

rted

her

e.T

he

sam

ple

cover

s

the

per

iod

of

19

99

q2

till

20

05

q2

.A

llm

od

els

rep

ort

OL

Ses

tim

ates

wit

hD

risc

oll

-Kra

ayst

and

ard

erro

rs.

Defi

nit

ions

and

sourc

esof

var

iab

les

are

list

edin

Ap

pen

dix

A.

***

,*

*,*

corr

esp

on

dto

the

on

e,fi

ve

and

ten

per

cen

tle

vel

of

sig

nifi

can

ce.


Table 4: Loan Growth and Foreign Currency Deposits

4Loan Growth 4Total Loan 4Loan Growth 4Total Loan

(in LCU) Growth (in LCU) Growth

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

GDP Growth 0.192 0.096 0.217 0.108

0.111 0.102 0.139 0.125

Exch. Rate Depreciation 0.264* -0.298* 0.128 -0.459*

0.137 0.161 0.262 0.219

Inflation -0.027 0.418 0.035 0.554

0.257 0.281 0.356 0.408

4rd -0.005*** -0.003** -0.006*** -0.004**

0.001 0.001 0.001 0.002

4rf -0.028* -0.073*** -0.027 -0.076***

0.016 0.014 0.021 0.018

Share Foreign Deposits 0.140 0.229** 0.142 0.232**

0.083 0.081 0.084 0.081

Foreign Deposits*4rd 0.006** 0.002 0.007*** 0.003

0.002 0.003 0.002 0.003

Inclusion Foreign Deposits 0.031 0.033

0.037 0.032

Reserve Equating 0.000 -0.013

0.020 0.025

Reserve Increase 0.004 -0.024

0.048 0.059

Stanbic Merger 0.120 0.138** 0.121 0.138**

0.073 0.056 0.073 0.056

Post Merger Period 0.120*** 0.086*** 0.126*** 0.091***

0.011 0.013 0.014 0.015

Constant -0.053* -0.091*** -0.044 -0.084**

0.027 0.029 0.029 0.032

Observations 394 394 394 394

R2 0.113 0.116 0.113 0.117

Number of Banks 17 17 17 17

Note: The sample covers the period of 1999q2 till 2005q2. The model is extended to include the share of foreign

currency deposits in previous quarter and its interaction with the rate of change in domestic interest rates.

Definitions and sources of variables are listed in Appendix A. All regressions include bank-level fixed effects

and seasonal and year dummy variables not reported here. All models report OLS estimates with Driscoll-Kraay

robust standard errors. ***, **, * correspond to the one, five and ten percent level of significance.


Table 5: Loan growth and bank size

4Loan Growth (in LCU) 4Total Loan Growth

(1) (2)

GDP Growth 0.238 0.129

0.151 0.120

Exch. Rate Depreciation -0.049 -0.429***

0.166 0.117

Inflation 0.367 0.656**

0.302 0.230

4rd -0.012 -0.016

0.012 0.014

Bank Size*4rd 0.001 0.001

0.001 0.001

4rf 0.125 -0.154

0.332 0.311

Bank Size*4rf -0.010 0.004

0.018 0.017

Bank Size -0.025 -0.050

0.043 0.038

Stanbic Merger 0.106 0.114*

0.079 0.064

Post Merger Period 0.126*** 0.092***

0.014 0.011

Constant 0.382 0.811

0.751 0.667

Observations 394 395

R2 0.104 0.107

Number of Banks 17 17

Note: The table reports estimates of the models after including bank size and interac-

tions of bank size with changes in the interest rates. Size is defined as the log of total

assets of a given bank taken in the first lag. All regressions include bank-level fixed

effects and seasonal and year dummy variables not reported here. Appendix A pro-

vides definitions and sources of variables. The sample covers the period of 1999q2 till

2005q2. All models report OLS estimates with Driscoll-Kraay robust standard errors.

***, **, * correspond to the one, five and ten percent level of significance. significance.


Table 6: Bank assets growth and open market operations

4DL/TA 4TL/TA 4GS/TA 4FA/TA 4NFA/TA

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

GDP Growth 0.156* 0.160 -0.331*** -0.296*** -0.302***

0.081 0.116 0.106 0.095 0.098

Exch. Rate Depreciation -0.035 -0.203 0.079 0.092 0.024

0.104 0.149 0.135 0.122 0.125

Inflation 0.053 0.200 -0.344 -0.430** -0.421*

0.178 0.256 0.232 0.209 0.215

4rd -0.0001 -0.0001 0.002** 0.001** 0.001**

0.001 0.001 0.001 0.001 0.001

4rf -0.007 -0.020 0.040*** 0.031** 0.027*

0.012 0.017 0.015 0.014 0.014

Stanbic Merger 0.003 0.002 0.045* 0.030 0.038*

0.018 0.026 0.023 0.021 0.022

Post Merger Period 0.045*** 0.063*** -0.032* -0.017 -0.017

0.013 0.018 0.016 0.015 0.015

Constant -0.010 -0.004 -0.001 0.000 0.016

0.015 0.021 0.019 0.017 0.018

Observations 394 394 394 394 394

R2 0.1606 0.1256 0.1205 0.2697 0.2645

Number of Banks 17 17 17 17 17

Note: All regressions report estimates of the variables from the baseline model including bank-level

fixed effects and seasonal and year dummy variables not reported here. The sample covers the period

of 1999q2 till 2005q2. All models report estimates of seemingly unrelated regressions (SUR) with

robust standard errors and small sample adjustment. Definitions and sources of variables are listed in

Appendix A. ***, **, * correspond to the one, five and ten percent level of significance.


Table 7: Bank asset shares and changes to reserve requirements

DL/TA TL/TA GS/TA FA/TA NFA/TA

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

GDP Growth 0.158 0.175 -0.127 -0.0786 -0.0830

0.105 0.120 0.107 0.065 0.071

Exch. Rate Depreciation -0.369*** -0.452*** 0.421** -0.1946** -0.2213***

0.112 0.138 0.144 0.069 0.076

Inflation 0.552** 0.659** -0.695** 0.0978 0.1557

0.230 0.252 0.296 0.116 0.112

4rd -0.001 -0.001 0.001* 0.0002 0.0005

0.001 0.001 0.001 0.000 0.000

4rf -0.055*** -0.058*** 0.055*** 0.0115* 0.0190**

0.015 0.016 0.018 0.006 0.007

Inclusion Foreign Deposits -0.042*** -0.032*** 0.015 0.0429*** 0.0455***

0.010 0.009 0.014 0.005 0.004

Reserve Equating 0.004 0.023** 0.002 -0.0392*** -0.0415***

0.006 0.008 0.009 0.010 0.010

Reserve Increase -0.099*** -0.111*** 0.132*** -0.0389*** -0.0376***

0.019 0.022 0.024 0.012 0.012

Stanbic Merger -0.150*** -0.150*** 0.276*** -0.1987*** -0.1483***

0.016 0.016 0.022 0.018 0.023

Post Merger Period 0.039*** 0.033*** -0.008 -0.0204** -0.0236**

0.009 0.008 0.013 0.009 0.009

Constant 0.328*** 0.414*** 0.144*** 0.1864*** 0.1760***

0.011 0.010 0.013 0.005 0.005

Observations 394 394 394 394 394

R2 0.374 0.231 0.401 0.269 0.199

Number of Banks 17 17 17 17 17

Note: All regressions include bank-level fixed effects and seasonal time dummy variables not reported here.

The sample covers the period of 1999q2 till 2005q2. All models report OLS estimates with Driscoll-Kraay

standard errors. Definitions and sources of variables are listed in Appendix A. ***, **, * correspond to the one,

five and ten percent level of significance.

Banks and Monetary Policy in Africa: Evidence from Uganda 131A

ppen

dix

C

Tab

leC

1:

Ass

ets

gro

wth

and

mac

roec

onom

icco

ndit

ions

-cl

ust

erin

gby

yea

r

4L

oan

Gro

wth

4T

ota

lL

oan

4G

ov.

Sec

uri

ties

4F

ore

ign

Ass

ets

4N

etF

ore

ign

(in

LC

U)

Gro

wth

Gro

wth

Gro

wth

Ass

ets

Gro

wth

(1)

(2)

(3)

(4)

(5)

GD

PG

row

th0.2

57*

0.1

44

-0.8

46*

-0.7

65***

-0.6

33**

0.1

36

0.1

17

0.4

52

0.2

45

0.2

53

Exch

.R

ate

Dep

reci

atio

n-0

.077

-0.4

62***

0.8

43

1.3

64***

1.1

11***

0.1

49

0.1

15

0.5

53

0.1

80

0.3

51

Infl

atio

n0.4

16

0.6

85***

1.1

51

-0.9

65*

-1.9

24**

0.2

75

0.2

24

1.0

74

0.4

69

0.7

16

4r d

-0.0

001

-0.0

01

0.0

02

0.0

03**

0.0

02

0.0

01

0.0

01

0.0

03

0.0

02

0.0

02

4r f

-0.0

47**

-0.0

79***

0.1

31**

0.0

29

-0.0

10

0.0

19

0.0

13

0.0

59

0.0

34

0.0

54

Sta

nbic

Mer

ger

0.0

89

0.0

87

0.0

41

0.1

03

0.1

68

0.0

66

0.0

50

0.1

07

0.0

95

0.1

12

Post

Mer

ger

Per

iod

0.1

24***

0.0

88***

-0.1

13**

-0.1

02**

-0.0

80

0.0

14

0.0

11

0.0

42

0.0

36

0.0

49

Const

ant

-0.0

56***

-0.0

49***

0.0

04

0.0

46*

0.1

87***

0.0

12

0.0

12

0.0

59

0.0

23

0.0

44

Obse

rvat

ions

394

395

357

364

358

R-s

quar

ed0.1

00

0.1

01

0.0

776

0.0

866

0.1

17

Num

ber

of

ban

ks

17

17

17

17

17

No

te:

Th

eta

ble

rep

ort

ses

tim

ates

of

the

mo

del

sin

Tab

le3

afte

rcl

ust

ein

gat

ay

ear

level

inst

ead

of

usi

ng

Dri

sco

ll-K

raay

robust

stan

dar

der

rors

.

All

reg

ress

ion

sin

clu

de

ban

k-l

evel

fixed

effe

cts

and

seas

on

alan

dy

ear

du

mm

yvar

iab

les

no

tre

po

rted

her

e.T

he

sam

ple

cover

sth

eper

iod

of

19

99

q2

till

20

05

q2

.A

pp

end

ixA

pro

vid

esd

efin

itio

ns

and

sou

rces

of

var

iab

les.

All

mo

del

sre

po

rtO

LS

esti

mat

esw

ith

robust

stan

dar

der

rors

clu

ster

edb

yy

ear.

**

*,

**

,*

corr

esp

on

dto

the

on

e,fi

ve

and

ten

per

cen

tle

vel

of

sig

nifi

can

ce.


Tab

leC

2:

Ass

ets

gro

wth

and

mac

roec

onom

icco

ndit

ions

-ex

clusi

on

of

GD

Pgro

wth

4L

oan

Gro

wth\

4T

ota

lL

oan

4G

ov.

Sec

uri

ties

4F

ore

ign

Ass

ets

4N

etF

ore

ign

(in

LC

U)

Gro

wth

Gro

wth

Gro

wth

Ass

ets

Gro

wth

(1)

(2)

(3)

(4)

(5)

Exch

.R

ate

Dep

reci

atio

n0.0

71

-0.3

79***

0.3

65

0.9

05***

0.7

27**

0.0

97

0.0

60

0.4

94

0.1

98

0.2

87

Infl

atio

n0.2

19

0.5

75**

1.7

69

-0.3

65

-1.4

22

0.2

88

0.2

38

1.0

68

0.5

60

0.8

36

4r d

0.0

00

-0.0

00

-0.0

01

0.0

01

0.0

00

0.0

01

0.0

01

0.0

04

0.0

02

0.0

02

4r f

-0.0

35**

-0.0

72***

0.0

92*

-0.0

07

-0.0

40

0.0

14

0.0

07

0.0

52

0.0

45

0.0

58

Sta

nbic

Mer

ger

0.0

89

0.0

87

0.0

41

0.1

03

0.1

68

0.0

66

0.0

50

0.1

07

0.0

95

0.1

12

Post

Mer

ger

Per

iod

0.1

18***

0.0

85***

-0.0

94**

-0.0

84**

-0.0

66

0.0

15

0.0

12

0.0

40

0.0

38

0.0

46

Const

ant

-0.0

58***

-0.0

51***

0.0

13

0.0

54

0.1

94***

0.0

10

0.0

11

0.0

58

0.0

32

0.0

49

Obse

rvat

ions

394

395

357

364

358

R-s

quar

ed0.0

980

0.0

997

0.0

713

0.0

824

0.1

14

Num

ber

of

ban

ks

17

17

17

17

17

No

te:

Th

eta

ble

rep

ort

ses

tim

ates

of

the

mo

del

sin

Tab

le3

afte

rex

clu

din

gG

DP

gro

wth

fro

mth

ese

to

fco

ntr

olvar

iable

s.A

llre

gre

ssio

ns

incl

ude

ban

k-l

evel

fixed

effe

cts

and

seas

on

alan

dy

ear

du

mm

yvar

iab

les

no

tre

po

rted

her

e.T

he

sam

ple

cover

sth

ep

erio

dof

1999q2

till

2005q2.

All

mo

del

sre

po

rtO

LS

esti

mat

esw

ith

Dri

sco

ll-K

raay

robu

stst

and

ard

erro

rs.

Ap

pen

dix

Ap

rov

ides

defi

nit

ion

san

dso

urc

esof

var

iable

s.***,

**,

*

corr

esp

on

dto

the

on

e,fi

ve

and

ten

per

cent

level

of

sig

nifi

can

ce.


Table C3: Marginal effects of monetary policy in the presence of foreign currency deposits

Share For. Curr. Deposits Marginal effect Confidence intervals

0.01 -0.005*** (-0.007, -0.004)

0.33 -0.003*** (-0.005, -0.001)

0.50 -0.002* (-0.005, 0.0002)

0.69 -0.001 (-0.004, 0.002)

Note: The table shows how the marginal effect of monetary policy on domestic

currency loan growth varies for a given share of foreign currency deposits. The

estimates are based on the model on domestic currency loan growth reported in

Column 1, Table 4. ***, **, * correspond to the one, five and ten percent level

of significance.


Table C4: Foreign currency deposits and the foreign interest rate

4Loan Growth 4Total Loan 4Loan Growth 4Total Loan

(in LCU) Growth (in LCU) Growth

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

GDP Growth 0.201* 0.103 0.227* 0.116

0.101 0.099 0.126 0.119

Exch. Rate Depreciation 0.246* -0.313* 0.103 -0.479**

0.127 0.166 0.243 0.210

Inflation -0.022 0.421 0.045 0.562

0.248 0.279 0.348 0.406

4rd -0.005*** -0.003** -0.006*** -0.004**

0.001 0.001 0.001 0.002

4rf 0.008 -0.044 0.009 -0.047

0.046 0.039 0.047 0.039

Share foreign currency deposits 0.138 0.227** 0.140 0.230**

0.081 0.081 0.083 0.081

Share for. Deposits*4rd 0.006** 0.002 0.006** 0.003

0.003 0.003 0.002 0.003

Share for. Deposits*4rf -0.122 -0.096 -0.124 -0.098

0.143 0.137 0.142 0.136

Inclusion foreign deposits 0.032 0.034

0.035 0.031

Reserve equating -0.000 -0.013

0.019 0.025

Reserve increase -0.003 -0.027

0.043 0.055

Stanbic Merger 0.119 0.137** 0.120 0.138**

0.073 0.056 0.073 0.056

Post Merger Period 0.121*** 0.086*** 0.128*** 0.092***

0.011 0.013 0.014 0.014

Constant -0.051* -0.089*** -0.042 -0.082**

0.027 0.028 0.027 0.030


R-squared 0.114 0.117 0.115 0.118


Note: The table reports estimates of the models in Table 4 after including the interaction of the foreign interest

rate with the share of foreign currency deposits. The sample covers the period of 1999q2 till 2005q2. Appendix

A provides definitions and sources of variables. All regressions include bank-level fixed effects and seasonal

and year dummy variables not reported here. All models report OLS estimates with Driscoll-Kraay robust

standard errors. ***, **, * correspond to the one, five and ten percent level of significance.


Table C5: Size indicators and the policy interest rate.

4Loan Growth (in LCU) 4Total Loan Growth

(1) (2)

GDP Growth 0.252 0.143

0.146 0.114

Exch. Rate Depreciation -0.057 -0.438***

0.165 0.116

Inflation 0.380 0.661**

0.301 0.233

4rd -0.001 -0.001

0.002 0.002

Size indicator 2*4rd 0.001 0.000

0.002 0.003

Size indicator 3*4rd 0.001 0.001

0.002 0.002

4rf -0.051 -0.108***

0.037 0.033

Size indicator 2*4rf 0.031 0.068

0.044 0.053

Size indicator 3*4rf -0.016 0.029

0.063 0.062

Bank Size -0.025 -0.049

0.043 0.039

Stanbic Merger 0.108 0.117*

0.080 0.065

0.014 0.011

Constant 0.373 0.808

0.754 0.682

Observations 394 395

R-squared 0.104 0.110

Number of banks 17 17

Note: The table reports estimates of the models in Table 5 after including interactions of size

indicators with the interest rates. Size is defined as the log of total assets of a given bank

taken in the first lag. Size indicators are defined to take the value of one if bank assets are in

a given tercile of the size distribution, and zero otherwise. Size indicator 1 corresponds to the

lowest tercile of the asset distribution, and Size indicator 3 - to the highest one. In addition,

Size indicator 1 serves as the base one. All regressions include bank-level fixed effects and

seasonal and year dummy variables not reported here. Appendix A provides definitions and

sources of variables. The sample covers the period of 1999q2 till 2005q2. All models report

OLS estimates with Driscoll-Kraay robust standard errors. ***, **, * correspond to the one,

five and ten percent level of significance.


Table C6: Monetary policy and loan growth at a bank level

Loan growth

(in LCU)

Total loan

growth

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

Bank Ranking MP=4rd MP=4rf MP=4rd MP=4rf

below or at 25th

percentile

Bank1*MP -0.006 -0.060 0.002 -0.131

0.005 0.090 0.005 0.105

Bank2*MP 0.014 0.117 0.003 -0.101

0.014 0.090 0.008 0.064

Bank3*MP -0.0004 -0.064 -0.001 -0.097*

0.002 0.058 0.003 0.0539

Bank4*MP -0.005 -0.064 -0.005 -0.082

0.00327 0.074 0.004 0.079

Bank5*MP -0.007* -0.030 -0.007** -0.047

0.00384 0.060 0.003 0.053

below or at the

median

Bank6*MP -0.010* -0.067 -0.007 -0.144

0.005 0.148 0.005 0.152

Bank7*MP 0.008** 0.086 0.005 0.070

0.003 0.094 0.003 0.083

Bank8*MP 0.006** -0.189** 0.004 -0.174*

0.002 0.068 0.005 0.083

Bank9*MP -0.007 0.024 -0.010** 0.055

0.004 0.056 0.004 0.048

below 75th

percentile

Bank10*MP -0.002 0.042 0.001 -0.031

0.004 0.035 0.004 0.038

Bank11*MP 3.30e-05 -0.090** 0.0002 -0.112***

0.003 0.040 0.003 0.035

Bank12*MP 0.005 -0.004 0.005 -0.041

0.006 0.062 0.006 0.054

Bank13*MP -0.004 -0.280** -0.008** -0.227**

0.004 0.120 0.003 0.089

above the 75th

percentile

Bank14*MP 0.002 -0.101 0.003 -0.123*

0.003 0.075 0.002 0.0679

Bank15*MP 0.007** -0.099 0.006* -0.110*

0.003 0.066 0.003 0.057

Bank16*MP -0.003 -0.130 -0.002 -0.128

0.004 0.097 0.005 0.078

Bank17*MP 0.002 0.078 0.003 0.017

0.003 0.086 0.004 0.083


R2 0.148 0.142 0.138 0.132

Number of Banks 17 17 17 17

Note: The table reports estimates of the models in Table 3 after including bank dummy variables interacted with changes

in the monetary policy indicator or the foreign interest rate (MP). Bank*MP shows the overall response of loan growth

at a given bank to changes in the given interest rate. The table only reports the coefficient estimates of the responses

omitting the remaining regression results (available upon request). The first column to the left shows the relative ranking

of banks in terms of average total assets over the sample period. Banks in the highest percentiles, i.e. above the 75

percentile of the bank asset distribution can be found in the bottom of the table. LCU is abbreviation for local currency

units. All regressions include bank-level fixed effects and seasonal and year dummies not reported here. Appendix

A provides definitions and sources of variables. The OLS estimates have Driscoll-Kraay standard errors. ***, **, *

correspond to the one, five and ten percent level of significance.

tilburg university essays on banking and regulation ... · radomir ivanov todorov geboren op 8...

Documents