ESSAYS ON BANKING AND FINANCIAL STABILITY
A Dissertation Presented to the Faculty of the Graduate School of Cornell University in Partial Fulﬁllment of the
Requirements for the Degree of Doctor of Philosophy
Sonali Sanghita Das August 2012

c 2012 Sonali Sanghita Das

ABSTRACT ESSAYS ON BANKING AND FINANCIAL STABILITY
Sonali Sanghita Das, Ph.D. Cornell University 2012
Financial crises have occurred repeatedly throughout history in both high and middle-to-low income countries. This dissertation studies how the interactions of ﬁnancial market participants aﬀect ﬁnancial stability. In the ﬁrst part of the dissertation, I analyze sales of assets between ﬁnancial institutions in the United States and ﬁnd evidence consistent with the theory that credit-constraints aﬀect the demand for, and price of, assets sold in ﬁre-sales. In the second part, I document the empirical regularity that the correlation of banks’ stock return – a measure of the interconnectedness of banks – increases in the run up to banking crises and thus helps predict crises. The third part ﬁnds that the main measure of asset risk-exposure that banks report to regulators are thought to be credible by equity investors, but less so in countries where regulators have allowed banks more discretion over the calculation of the measure.

iii
BIOGRAPHICAL SKETCH
Sonali Das is a PhD candidate in the Department of Economics at Cornell University, with research interests in international ﬁnance and banking. During the course of her doctoral studies, she has interned at the International Monetary Fund, the Federal Reserve Bank of San Francisco, and the Brookings Institution. Sonali holds an MA from McGill University (2005, Economics) and a BSc from the University of Toronto (2004, Economics and Statistics). Sonali was born and raised in Moncton, Canada.

iv
ACKNOWLEDGMENTS
A dissertation is far from a solo venture and I would like to thank the numerous individuals without whom its completion would not have been possible. First, I feel fortunate to have had Assaf Razin as my advisor and am deeply thankful for his guidance and insight. I am grateful to my other committee members, Eswar Prasad and Matthew Freedman, who provided invaluable guidance and support. Thanks also to Levon Barseghyan for his help and encouragement throughout my studies.
Several economists have contributed to my training as an economist outside of an academic setting. Kemal Dervi¸s has been a teacher and mentor. Simon Kwan, along with the AEA/CSWEP committee, provided me with the opportunity to participate in the research environment of the Federal Reserve Bank of San Francisco. Amadou N.R. Sy was my ﬁrst manager at the IMF. It was great to work with him and I hope my future managers will be as open and encouraging.
The helpful suggestions of George Jakubson, Andrew Karolyi, and Edith Liu on my dissertation papers are much appreciated, and thanks also to Reuven Glick, Galina Hale, and Jose Lopez for their interest and encouragement. Many Cornell staﬀ have provided assistance regularly. Thank you Shelly Hall, Eric Humerez, Ulrike Kroeller, Eric Maroney, Amy Moesch, Darrie O’Connell, and Sheri Van Deusen.
Among the many friends that have made this stage of life memorable, I am especially glad for the time spent with Karen Brummund, Peter Brummund, Brian Dillon, and Gabor Pinter. I am grateful to Audri for his constant support and counsel, and also for making life more interesting. Finally, the gratitude owed to my parents is immeasurable – thank you Ma and Baba.

CONTENTS

v

Contents

1 Introduction

1

2 The Eﬀect of Leverage on Asset Sales between Financial In-

stitutions: Deal Level Evidence

4

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2.2 Access to funding within the ﬁnancial sector . . . . . . . . . . 7

2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.4 Modeling the determinants of asset transaction values . . . . . 10

2.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2.7 Summary and conclusion . . . . . . . . . . . . . . . . . . . . . 24

3 Interconnections in Banking, Systemic Risk, and Crisis

25

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3.2 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . 29

3.2.1 Literature on banking crises . . . . . . . . . . . . . . . 29

3.2.2 Measuring systemic risk . . . . . . . . . . . . . . . . . 31

3.3 Measure of interconnectedness . . . . . . . . . . . . . . . . . . 34

3.4 Data and empirical method . . . . . . . . . . . . . . . . . . . 38

3.4.1 Probability of a banking crisis . . . . . . . . . . . . . . 38

3.4.2 Probability of n bank failures during a crisis . . . . . . 40

3.4.3 Bank failures and share of failed assets in all years -

large versus small banks . . . . . . . . . . . . . . . . . 42

3.4.4 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.5.1 Probability of a banking crisis . . . . . . . . . . . . . . 47

3.5.2 Probability of n bank failures during a crisis . . . . . . 47

CONTENTS

vi

3.5.3 Bank failures and share of failed assets in all years large versus small banks . . . . . . . . . . . . . . . . . 51
3.6 Robustness . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 3.6.1 Instrumental variables estimation . . . . . . . . . . . . 55 3.6.2 Additional controls . . . . . . . . . . . . . . . . . . . . 57
3.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 3.8 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4 How Risky are Banks’ Risk-Weighted Assets? Evidence from

the Financial Crisis

68

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.2 Stylized facts . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

4.3 Risk-weighted assets: Basel II Standardized and IRB Ap-

proaches . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

4.4 Related literature . . . . . . . . . . . . . . . . . . . . . . . . . 76

4.5 Data and Methodology . . . . . . . . . . . . . . . . . . . . . . 79

4.6 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . 86

4.6.1 Determinants of stock returns . . . . . . . . . . . . . . 86

4.6.2 Market risk and balance-sheet measures of risk exposure 95

4.7 Performance during the Eurozone debt crisis . . . . . . . . . . 99

4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

4.9 Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104

5 References

106

LIST OF TABLES

vii

List of Tables
2.1 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . 13 2.2 Determinants of value of asset sale – OLS . . . . . . . . . . . 14 2.3 Determinants of deal probability and deal value . . . . . . . . 16 2.4 Determinants of deal probability and deal value – Deposit-
taking institutions versus other ﬁnancial institutions . . . . . . 18 2.5 Number of transactions between ﬁnancial sectors . . . . . . . 19 2.6 Buyer capital by ﬁnancial sector . . . . . . . . . . . . . . . . . 19 2.7 Determinants of deal probability and deal value – sectoral de-
composition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 2.8 Determinants of deal value – controlling for type of asset . . . 22 2.9 Determinants of deal value – US sales to buyers in all countries 23
3.1 Number of banking crises and time in crisis by interconnectedness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
3.2 Descriptive statistics . . . . . . . . . . . . . . . . . . . . . . . 46 3.3 Determinants of banking crises – benchmark results . . . . . . 48 3.4 Probability of n bank failures during a crisis . . . . . . . . . . 50 3.5 Probability of bank failures in all years – country-size-year panel 53 3.6 Assets of failed banks in all years – country-size-year panel . . 54 3.7 Determinants of banking crises – IV estimation . . . . . . . . 56 3.8 Determinants of banking crises – overall stock market movements 58 3.9 Determinants of banking crises – concentration and govern-
ment ownership of banks . . . . . . . . . . . . . . . . . . . . . 59 3.10 Determinants of banking crises – balance-sheet risk exposures 62 3.11 Banking crises 1993-2009 . . . . . . . . . . . . . . . . . . . . . 65 3.12 Bank failures by country and size – median classiﬁcation . . . 66 3.13 Bank failures by country and size – 10B US$ classiﬁcation . . 67

LIST OF TABLES

viii

4.1 Descriptive statistics – bank stock returns over periods of crisis 80 4.2 Descriptive statistics – explanatory variables . . . . . . . . . . 84 4.3 Determinants of returns – Do risk-weighted assets aﬀect stock
returns? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 4.4 Determinants of returns – Do risk-weighted assets aﬀect stock
returns of large banks? . . . . . . . . . . . . . . . . . . . . . . 89 4.5 Determinants of returns – Is there a capital-funding trade-oﬀ? 92 4.6 Basel II implementation schedules – credit risk measurement . 93 4.7 Determinants of returns – Basel I versus Basel II standardized
approach to measuring RWA . . . . . . . . . . . . . . . . . . . 94 4.8 Market risk and balance-sheet measures of risk exposure . . . 97 4.9 Market measures of risk and balance-sheet measures of risk
exposure – have the relationships changed since the crisis? . . 98 4.10 Determinants of returns – performance during the European
sovereign debt crisis . . . . . . . . . . . . . . . . . . . . . . . . 100 4.11 Determinants of returns during the European sovereign debt
crisis – Basel I versus Basel II approaches to measuring RWA . 101 4.12 List of countries – full sample . . . . . . . . . . . . . . . . . . 104 4.13 Descriptive statistics – correlations of explanatory variables . . 105

LIST OF FIGURES

ix

List of Figures
3.1 Interconnectedness of largest 5 banks and interconnectedness of largest 5 non-ﬁnancial ﬁrms in the US . . . . . . . . . . . . 35
3.2 Interconnectedness of large banks and interconnectedness of small banks in the US . . . . . . . . . . . . . . . . . . . . . . 37
3.3 Interconnectedness of largest 5 ﬁrms by sector – United States 64
4.1 Decrease in risk-weighted assets at US BHCs . . . . . . . . . . 72 4.2 Quality of regulatory capital at US BHCs . . . . . . . . . . . . 73 4.3 Ratio of RWA to total assets in Asia, Europe and North Amer-
ica (2002-2010) . . . . . . . . . . . . . . . . . . . . . . . . . . 74

CHAPTER 1. INTRODUCTION

1

Chapter 1
Introduction
Economists have long struggled with the question of how to structure ﬁnancial systems to achieve an eﬃcient allocation of resources and allow agents to share risk. They key impediment lies in the fact that ﬁnancial markets are prone to panics that can be self-fulﬁlling and contagious, magnifying the consequences of shocks that hit few institutions or markets. Over the last two centuries, there has been only a single year in which no country was experiencing a ﬁnancial crisis.1 This dissertation studies a component of ﬁnancial markets often overlooked in economics – how ﬁnancial market participants interact with each other – and its implications for ﬁnancial stability.
In the second chapter, I analyze how the equity capital position of ﬁnancial institutions aﬀects their demand for assets and the resulting value of
1The year being 1823. According to Reinhart and Rogoﬀ’s (2009) history of ﬁnancial crises for 69 countries, several countries have had a ﬁnancial crisis (banking crisis, currency crisis, and/or external debt crisis) in each year since 1800. The average number of countries in crisis in each year of this period is 14.

CHAPTER 1. INTRODUCTION

2

transactions between ﬁnancial institutions. When intermediaries are creditconstrained and have a sudden need for liquidity, they are forced to sell assets to other institutions for cash. My results show a positive relationship between buyer capital and the likelihood of buying assets, and between buyer capital and the value of the deal. That is, those institutions that are the least constrained in their ability to raise funding are those that demand assets and pay more for them. This result does not hold, however, for deposit-taking institutions that had access to several government programs designed to improve their funding situation during the crisis of 2008.
The third chapter takes a broad view and documents a new empirical regularity in the run up to banking crises. In a sample of 45 countries, covering the period from 1993 to 2009, the correlation of banks’ stock returns increases before the onset of a crisis. The increase in the correlation measure is not driven by an overall increase in the national stock market – i.e. it does not simply capture ‘boom’ periods – and there is no signiﬁcant relationship between the correlation of non-ﬁnancial ﬁrms and crisis. Thus the stock return correlation can be seen as a simple measure of interconnectedness among banks that helps predict banking crises.
The fourth chapter, written jointly with Amadou N.R. Sy, focuses on banks – regulated, deposit-taking institutions – and studies market perceptions of the riskiness of their risk-weighted assets (RWA) by examining the determinants of the stock returns of an international panel of banks. Banks are required to hold capital as a percentage of RWA and the rules used to

CHAPTER 1. INTRODUCTION

3

calculate RWA have changed over time, allowing banks more discretion over the calculation. We ﬁnd a negative relationship between RWA and stock returns over periods of ﬁnancial crisis, suggesting that investors believe RWA are an indication of bank portfolio risk. This relationship is weaker in countries that had implemented Basel II before the onset of the crisis, allowing banks to use internal risk models to assess credit risks.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

4

Chapter 2
The Eﬀect of Leverage on Asset Sales between Financial Institutions: Deal Level Evidence
2.1 Introduction
In times of crisis, ﬁnancial institutions are often forced to sell assets in order to stay solvent. While this may be a necessary strategy for an individual institution facing borrowing constraints, the action of selling under duress can in fact drive down the price of assets and deepen the crisis. There is a growing literature on these ﬁre-sales of assets, and their role in deepening

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

5

ﬁnancial instability (Shleifer and Vishny 2011). Shleifer and Vishny (1992) ﬁrst formalized the idea of the ‘ﬁre-sale’ pointing out that asset liquidation often happens when the best users are (also) credit constrained, leading to a lower liquidation value.
This paper provides new evidence on asset transactions between ﬁnancial institutions. I ﬁrst examine how the capital position of potential buyers of assets aﬀects both their decision to purchase and the value of transactions themselves. There are two main features to recent theoretical contributions that model ﬁre-sales as an amplifying mechanism of liquidity crises (Brunnermeier and Pederson (2009), Krishnamurthy (2010), and Fostel and Geanakoplos (2008) in a multiple asset setting). First, the amount of funding available to ﬁnancial intermediaries is a function of their equity capital, up to a maximum amount; and second, the demand for assets is a function of the total funds available to intermediaries. These models focus on the constraints of the sellers and sales are assumed to be absorbed by agents who have lower valuations of the assets. The ‘cash-in-the-market’ pricing models of Allen and Gale (1998, 2005) explicitly model the buyers, however, and show that an asset’s sale price will be determined by the limited amount of cash, or liquidity, held by the surviving ﬁnancial intermediaries, since they are the marginal buyers. I ﬁnd that the capital to assets ratio of ﬁnancial institutions is positively related to both their decision to purchase assets and to the value of the transaction.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

6

Second, I analyze whether there are sectoral1 diﬀerences in how buyer
capital aﬀects the demand for assets and the deal value. Based on their re-
lationships before the subprime crisis, and government policies implemented
during the crisis, diﬀerent types of ﬁnancial institutions have had varying
degrees of access to funding over the past few years. He, Khang, and Kr-
ishnamurthy (2010) carefully measure how securitized assets shifted across
sectors in the United States during the 2008 crisis. They ﬁnd that the hedge
fund and broker-dealer sectors, sectors that rely on repo ﬁnancing, reduced
asset holdings and the commercial banking sector, which had access to more
stable funding sources, increased asset holdings. Their evidence suggests that
certain groups of ﬁnancial institutions can step in to ease liquidity problems
during ﬁnancial crises, but also that government liquidity policies imple-
mented to encourage commercial banks to lend to the real sector may have
unintended eﬀects. By disaggregating to the institution level in this paper,
I am able to shed light on the extent to which credit constraints aﬀect asset demand and price across institutions that are similarly aﬀected by policy.2
I focus on the potential buyers of assets, as opposed to the sellers, for two
reasons. First, distressed ﬁnancial institutions sell assets for reasons that are
fairly well understood. When in need of liquidity, they have three options:
1By ‘sectoral’ I mean sub-sectors within the ﬁnancial sector. Broadly: deposit-taking institutions (commercial and savings banks), investment banks, broker-dealers, hedgefunds, and real estate and insurance companies.
2He, Khang, and Krishnamurthy (2010) study securitized assets while I study assets such as property and actual loan portfolios held by ﬁnancial institutions. More detail is provided in the data description in section 2.3.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

7

raise equity capital, raise debt, or sell assets for cash. Raising equity capital is thought to be costly due to debt overhang (Myers 1977) or adverse selection problems facing the potential equity investors (Myers and Majluf 1984) even in good times, so it is likely to be especially diﬃcult or costly in times of ﬁnancial distress. Acharya, Gujral, and Shin (2009) ﬁnd that ﬁnancial intermediaries did raise new capital in 2008, both from private investors and from government-funded capital injections, but it was predominantly in the form of hybrid claims such as preferred equity and subordinated debt. That is, claims that are debt-like and cannot be thought of as equity capital. Second, the information on sellers of assets in the database used is less detailed and complete than that on the buyers, making an analysis of the sellers’ balance-sheets diﬃcult.3 The next section describes diﬀerences in access to funding across ﬁnancial sectors, sections 2.3 and 2.4 describe the data used in the analysis and the estimation strategy, sections 2.5 and 2.6 present the results and robustness checks, and section 2.7 concludes.

2.2 Access to funding within the ﬁnancial sector
Funding composition diﬀers across diﬀerent types of ﬁnancial institutions. The ﬁrst distinction, in all periods and not just during crises, is that commercial and savings banks raise (partially) insured deposits, which are considered to be a relatively stable and cheap form of borrowing. Runs on banks
3For example, when a real-estate property is being sold the database often lists the name of the selling company as simply the address of the property being sold. Identifying information for the buyers is properly recorded, however.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

8

by depositors have been relatively rare in the United States since the creation of the Federal Deposit Insurance Corporation (FDIC) in 1934.4 Addition-
ally, coverage limits – the amount per depositor that is insured by the FDIC
– were increased from $100,000 to $250,000 in 2008. A second policy af-
fecting commercial banks was the Temporary Liquidity Guarantee Program
(TLGP), in place from October of 2008 to December of 2010. It allowed
deposit-taking institutions to issue senior unsecured debt with a maximum
three year term, with the FDIC insuring default on these bonds for a fee of
25 to 50 basis points. Finally, the Fed cut the discount rate for commercial
banks several times beginning in August 2007.
The Fed also allowed investments banks to begin borrowing directly from
the discount window in March of 2008, using a broad range of debt securities
as collateral. Hedge funds and broker-dealers, on the other hand, did not
have access to government support and traditionally raise debt mostly in the
form of repo ﬁnancing. These diﬀerences in funding sources suggest that
deposit-taking institutions had greater access to, or a lower cost of, funding,
followed by investment banks, and then hedge-funds and broker-dealers.
4Prior to the crisis of 2008, some academic economists declared depositor bank runs to be dead after the implementation of deposit insurance, while others pointed to the runs that took place in emerging market economies in which there was deposit insurance in place. This crisis saw a resurgence of bank runs – ﬁrst with Northern Rock in the UK and then BearStearns and IndyMac in the United States, in September 2007, March 2008, and July of 2008.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

9

2.3 Data

The main dataset used in the analysis is the Thomson Reuters SDC Platinum
M&A database, which contains data on mergers and acquisitions of ﬁrms, as
well as on sales of assets. The assets traded are primarily real estate portfolios
(apartment buildings, oﬃce buildings, etc.), loan portfolios, bank branches or
units of ﬁnancial institutions, and there are a few observations on other assets
such as equity investment portfolios, asset-backed securities, and IT systems.
The type of asset is not given by data providers, unfortunately, so these were coded from a text description of the deal where possible.5 I analyze deals
between ﬁnancial institutions located in the United States between 2005 and
2011, where the buyer is a publicly-traded company. Approximately 85% of
the assets sold by US institutions to other ﬁnancial institutions in this time period are to other US institutions.6
To estimate a model that controls for sample selection bias arising from
the possibility that a ﬁrm’s characteristics aﬀects its decision to buy assets, I
ﬁrst start with the universe of publicly-traded ﬁnancial ﬁrms in United States
that are contained in the Worldscope/Datastream database. Financial ﬁrms
are those with a Standard Industrial Classiﬁcation code beginning with the
5The description of the deal was searched for strings such as: “home loan portfolio”; “real estate portfolio”, “acquired” and “bank” and “branches”; “asset backed securities”, etc. covering all the possible types.
6Another 6% are sold to Australian and Canadian ﬁrms, and the remaining deals are made with the following 13 countries: Belgium, France, Germany, Ireland, Israel, Japan, Mexico, Netherlands, South Korea, Spain, Switzerland, United Kingdom, and United Arab Emirates.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

10

digit 6 (division H). The focus on publicly-traded ﬁrms is driven primarily by data constraints, as the Worldscope database only contains balance-sheet data for public ﬁrms. The sample of potential buyers consists of 1116 ﬁnancial ﬁrms. This sample of potential buyers is then merged to the deals with publicly-traded buyers in the deals database. The resulting sample is of 402 deals, representing 402 ‘sellers’ and 183 unique buyers.

2.4 Modeling the determinants of asset transaction val-
ues
The main hypothesis being tested in this paper is that there is a positive relationship between the capital to assets ratio of the buyer, and the value of an asset sale.7 There are two reasons to expect this. First, an institution’s cash is counted in its capital measure, so ﬁrms with higher capital may simply have higher cash on hand with which to purchase assets and may have a higher willingness to pay for assets. Second, since capital ratios are often seen as a measure of health for ﬁnancial institutions, those with more capital should be able to borrow on better terms. The leverage constraint theories of Brunnemeier and Pederson (2009) and Fostel and Geanokoplos take the maximum debt ﬁnancing available to an intermediary to be proportional to its equity capital. A second hypothesis concerns the intensity of the relationship between ﬁrm capital and deal value. As leverage constraints are more
7By which I mean the price at which the asset is sold. I use the term ‘deal value’ instead of ‘price’ simply to make clear that the units of the assets being sold are not standardized.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

11

important for non-deposit taking institutions, we expect the positive relationship between buyer capital and deal value to be greater for non-deposit taking institutions.
To test these hypotheses, I use a Heckman selection model (Heckman 1979) to estimate the eﬀect of buyer capital on the value of an asset sale. Under the assumption that any unobservable characteristics that aﬀect a ﬁnancial ﬁrm’s decision to buy assets are uncorrelated with unobservable characteristics that aﬀect the value of the deal itself, ordinary least squares would produce unbiased estimates of the eﬀect of buyer capital on the value of an asset sale. This is too strong an assumption to make, however, as one can imagine the preferences of a manager inclined to expand during crisis times to aﬀect his approach to bargaining on price. In addition, the eﬀect of a buyer’s characteristics on its propensity to purchase an asset is interesting in itself.
Let i=seller, j=buyer, and Ejt = 1 if institution j buys an asset in year t. The ﬁrst stage selection equation is a Probit estimation of the probability that a buyer purchases an asset in a given year of the sample.

Pr(Ejt = 1) = F (δ1(capitaljt/Ajt)+δ2 log Ajt+δ3Agrowthjt+δ4 log mrkttobookjt+vt) (2.1)
where the buyer’s capital to assets ratio, capitaljt/Ajt, and size, given by the log of assets, are variables of interest in both the selection equation and the main equation. Two other buyer characteristics, asset growth and

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

12

the log of the market-to-book ratio, are included to identify the selection equation. Agrowthjt is the buyer’s percentage increase in total assets over the previous year and mrkttobookjt is the market value of a ﬁrm’s assets divided by the book value of its assets.8 The ﬁrst variable captures whether the ﬁrm has been expanding and the second the ﬁrm’s potential to grow. Year dummies are included to control for macroeconomic shocks that aﬀect all ﬁnancial institutions alike. The expected signs of the coeﬃcients are δ1 > 0, δ2 > 0, δ3 > 0, and δ4 > 0. That is, institutions with more capital, larger institutions, institutions that have been expanding, and institutions with a higher Tobin’s q, are expected to be more likely to purchase assets.
The estimated coeﬃcients from equation 2.1 are used to calculate the inverse Mills ratio, φ(δ)/Φ(δ), which is then included in the main equation to correct for potential sample selection bias. The equation estimating the determinants of the value of asset sales is:

log yijs = β1(capitaljt/Ajt)+β2 log Ajt+θ1Xij+θ2M arketRs+λ(φ(δ)/Φ(δ)jt)+uijt (2.2)
The dependent variable log yijs is the log of the value of a transaction between seller i and buyer j that takes place on day s of year t, in millions of US dollars. The selection equation 2.1 is estimated using a buyer-year panel, while equation 2.2 is estimated on a pooled sample of deals with selection bias correction at the buyer-year level. The main explanatory variable of interest
8Calculated as (market value of equity + book value of liabilities)/book value of assets. This is standard practice in the corporate ﬁnance literature.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

13

is the capital ratio, and the hypothesis is that β1 > 0. We also expect β2 > 0 as larger ﬁrms are likely to buy larger assets. The control variables included in Xij are indicator variables that denote whether the seller and buyer are in the same US city, the same US state, and the same sector. M arketRs is the stock market return over the month prior to the day the deal is announced, included to control for macroeconomic shocks.
Table 2.1 provides descriptive statistics corresponding to both the full sample of equation 2.1 and the deal (censored) observations. We see that institutions that buy assets have a high capital ratio on average, at 71 percent, compared to 31 percent for all ﬁrms. The ﬁrms that buy assets are also larger.

Table 2.1: Descriptive statistics

Descriptive statistics
Deal value, millions of US dollars Buyer capital/assets Buyer assets, millions of US dollars Asset growth (%), previous year Buyer market to book US stock market return, previous month Same city Same state Same sub-sector

Uncensored (5823 obs)

Mean

Std Dev

31.28 20,600
10.63 0.56

26.52 145,000
22.63 1.52

Censored/deal (402 obs)

Mean

Std Dev

350.75

1551.44

71.02

31.11

67,700

295,000

17.49

27.12

1.08 0.73

0.54 4.61

0.04 0.20 0.28 0.45

0.27 0.02

2.5 Results
I ﬁnd evidence in support of the hypothesis that the value of a deal is increasing in the buyer’s capital ratio. Table 2.2 presents the estimation of

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

14

equation 2.2 using OLS (not including the sample selection correction term), for the sake of comparison with the Heckman selection model.

Table 2.2: Determinants of value of asset sale – OLS

Dependant variable: log(Value of asset sale) (1)
log(deal value)

Buyer capital/assets Buyer log(assets) Same city Same state Same sub-sector US stock market return

0.016*** (0.003) 0.578*** (0.037) 0.223 (0.354) 0.187 (0.154) 0.888*** (0.204) 0.004 (0.014)

Observations Adj R-squared

402 0.390

This table presents the estimation of the deal value equation using OLS. The dependent variable is the log of the value of the transaction, in millions of US dollars. The explanatory variables are the buyer’s total capital to assets ratio, the buyer’s size given by the log of total assets, indicator variables for whether the seller and buyer are based in the same city, same state, and whether they belong to the same sub-sector, and the US stock market return over the month prior to the day the deal is announced. Standard errors are in parentheses below the coeﬃcient estimates. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

The coeﬃcient on the capital ratio is the estimated semi-elasticity of the deal value with respect to the capital ratio. The estimate of 0.016 indicates that a 1 percentage point increase in the capital ratio is associated with a 1.6 percent increase in deal value. The coeﬃcient on the size variable indicates

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

15

that a 1 percent increase in the size (assets) of the buyer is associated with a 0.6 percent increase in the deal value, on average. Whether the seller and buyer are located in the same US city or state seem not to aﬀect the value of the deal, but deals between ﬁrms in the same ﬁnancial sector have a higher value, with the coeﬃcient of 0.888 indicating they are priced higher by 2.43 million US dollars on average.
Table 2.3 shows the benchmark estimation results of the Heckman selection model. The estimated coeﬃcients in column (1) show that the capital ratio, size, and asset growth of the buyer are positively related to its propensity to buy assets. The marginal eﬀects are as follows: a 1 percentage point increase in the capital ratio (from the mean) increases the probability of a deal by 0.8 percent, a 1 percent increase in size increases the probability of a deal by 0.05 percent, and a 1 percentage point increase asset growth increases the probability of a deal by 0.06 percent.
The positive coeﬃcient lambda in column (2) indicates that unobservables in equations 2.1 and 2.2 are positively correlated.9 That is, unobserved characteristics that increase a ﬁnancial institution’s likelihood of buying assets also increase the value of the deal. This coeﬃcient is statistically signiﬁcant at the 5 percent level, indicating that there is indeed a sample selection eﬀect and OLS is not an appropriate method to estimate equation 2.2. Once we correct for the selection bias, the eﬀect of capital on deal value is higher: 3.2
9The correlation between the error terms is in fact ρ where ρ = λ/σ and σ is the variance of the error term in equation 2.2, uijt.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

16

Table 2.3: Determinants of deal probability and deal value

Heckman selection model

(1) (2) probability(deal) log(deal value)

Buyer capital/assets Buyer log(assets)

0.026*** (0.002) 0.178*** (0.014)

0.032*** (0.008) 0.695*** (0.062)

Same city Same state Same sub-sector US stock market return

0.157 (0.351) 0.180 (0.152) 0.872*** (0.199) 0.007 (0.014)

lambda Buyer asset growth Buyer log(market to book)

0.002* (0.001) -0.032 (0.057)

0.778** (0.337)

Observations

5823

402

This table presents the results of estimating the Heckman selection model. Column (1) shows the selection equation. The explanatory variables are the buyer’s total capital to assets ratio, the buyer’s size given by the log of total assets, the US stock market return over the month prior to the day the deal is announced, the buyer’s growth in assets in the year prior to the deal, and the log of the buyer’s market value of assets to book value of assets. Column (2) shows the deal value equation. The explanatory variables are the buyer’s total capital to assets ratio, the buyer’s size, indicator variables for whether the seller and buyer are based in the same city, state, whether they belong to the same sub-sector, the selection correction terms, and year dummies (not shown).

percent for a 1 percentage point increase in capital/assets compared to the 1.6 percent for the OLS results in Table 2.2.
Next, I group the sample (buyers) into deposit-taking institutions and

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

17

‘non-deposit-taking’ institutions, and include interaction terms for the nondeposit-taking institutions in the estimation. Table 2.4 shows the results. We see that it is the non-deposit-taking institutions that account for the positive relationship between buyer capital and both the likelihood of buying assets and the deal value. This is consistent with the hypothesis that the capital position of deposit-taking institutions should not aﬀect their asset purchases as much as other institutions, as the deposit-taking ones had better access to or cheaper funding during the crisis.
The next speciﬁcation digs further into the diﬀerences between diﬀerent ﬁnancial sub-sectors. Tables 2.5 and 2.6 provide further descriptive statistics on the types of ﬁnancial institutions. Table 2.7 shows the estimation results including interaction eﬀects for each group of non-deposit-taking institutions: investment banks and other credit institutions, hedge funds and broker-dealers, and insurance and real estate. The results show no relationship between capital, the likelihood of making a purchase, and the deal value for deposit-taking institutions. There is a signiﬁcant and positive relationship between capital and the probability of making a purchase for each other type of potential buyer, however. Column (2) shows no relationship between capital and deal value for each group, and no signiﬁcant diﬀerences across sectors.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

18

Table 2.4: Determinants of deal probability and deal value – Deposit-taking

institutions versus other ﬁnancial institutions

Heckman selection model

(1) (2)

probability(deal) log(deal value)

Buyer capital/assets

-0.014*

-0.005

Buyer capital/assets * non-deposit taking Buyer log(assets)

(0.008) 0.042*** (0.009) 0.198*** (0.028)

(0.014) 0.033*** (0.011) 0.718*** (0.062)

Buyer log(assets) * non-deposit taking

0.004

-0.001

(0.004)

(0.003)

Same city

0.153

Same state Same sub-sector

(0.347) 0.205 (0.151) 0.884***

(0.197)

US stock market return

0.006

(0.014)

lambda

0.840**

Buyer asset growth Buyer log(market to book)

-0.002 (0.003) 0.345*

(0.331)

(0.177)

Observations

5823

402

This table presents the results of estimating the Heckman selection model, including an interaction term for ﬁnancial institutions that do not raise deposits (indicated by ‘* non-deposit taking’). The other explanatory variables are as in Table 2.3.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

19

Table 2.5: Number of transactions between ﬁnancial sectors

Sellers Depositary institutions Inv bank & other credit Hedge fund & broker-dealers Insurance & real estate Total

Buyers Deposit inst 51 10 4 2 67

Inv bank & oth 2 18 9 3 32

HF & BD 1 6 15
255 277

Ins & real est 0 0 0 26 26

Total 54 34 28 286 402

This table shows the number of deals that took place between each ﬁnancial sector included in the deal sample.

Table 2.6: Buyer capital by ﬁnancial sector

Depositary institutions Inv bank & other credit Hedge fund & broker-dealers Insurance & real estate

Uncensored (5823 obs) obs Mean Std Dev

4017 494 445 851

18.17 63.42 85.04 45.86

8.74 29.12 14.68 26.32

Censored/deal (402 obs) obs Mean Std Dev

67 18.22

8.44

32 48.33 29.94

277 88.54 10.98

26 48.35 28.74

This table shows the average capital to assets ratios of each ﬁnancial sector, for the whole sample of potential buyers and also for the sample of buyers that purchased assets.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

20

Table 2.7: Determinants of deal probability and deal value – sectoral decom-

position

Heckman selection model

(1) (2) probability(deal) log(deal value)

Buyer capital/assets Buyer capital/assets * Inv bank & other credit Buyer capital/assets * Hedge fund & broker-dealers Buyer capital/assets * Insurance & real estate Buyer log(assets) Buyer log(assets) * Inv bank & other credit Buyer log(assets) * Hedge fund & broker-dealers Buyer log(assets) * Insurance & real estate

-0.009 (0.007) 0.016** (0.007) 0.040*** (0.007) 0.023*** (0.007) 0.167*** (0.022) 0.019 (0.024) 0.005 (0.032) -0.066*** (0.022)

0.001 (0.018) -0.004 (0.019) 0.008 (0.021) 0.015 (0.019) 0.516*** (0.067) 0.138*** (0.053) 0.221*** (0.062) 0.093 (0.069)

Same city Same state Same sub-sector US stock market return

0.213 (0.347) 0.160 (0.152) 0.743** (0.302) 0.007 (0.014)

lambda Buyer asset growth Buyer log(market to book)

0.001 (0.001) -0.163*** (0.063)

0.329 (0.356)

Observations

5823

402

This table presents the results of estimating the Heckman selection model, including interaction terms for the ﬁnancial sub-sector of the buyer. Indicator variables for the sub-sector of both the seller and buyer are included as well (not shown) and the category left out is deposit-taking banks. Other explanatory variables are as in Table 2.3.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

21

2.6 Robustness

I perform two robustness exercises in this section. First, I estimate the
selection model using the subset of deals for which the type of assets (e.g.
building versus loan portfolio) was coded and control for the asset type, to
ensure the results are not driven by a relationship between buyer capital and
the class of assets purchased. In these 252 deals, 88 percent are properties,
7 percent are loan portfolios, and 5 percent are units or branches of banks.
The results are shown in Table 2.8.
The coeﬃcient on the buyer’s capital ratio and the other explanatory
variables are very close to the benchmark estimation. The dummy variables
for asset type indicate that loan portfolios sell for 2.24 million more than
properties, on average.
Finally, Table 2.9 shows the results of estimating equation 2.2 on the full
sample of transactions involving US sellers – those bought by US ﬁrms as well as the other 15 countries listed in the data section.10 The estimated co-
eﬃcients are close to the estimates from the original OLS estimation shown
in Table 2.2. The increase in value when a deal occurs between two ﬁrms in
the same sector is now smaller, however, by about 1 million US dollars. In
column (2), I include an interaction term for buyers that are from countries
10I do not estimate a selection model here because (1) identifying the universe of potential buyers is less straightforward than for the United States and (2), even having done so, the number of ‘zeros’ (non-purchases) in the selection equation would be problematic for the estimation. Just around 15 percent of sales from US ﬁrms had foreign buyers. A potentially fruitful future project would be to study deals between a broader set of countries.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

22

Table 2.8: Determinants of deal value – controlling for type of asset

Heckman selection model
Buyer capital/assets Buyer log(assets) Same city Same state Same sub-sector US stock market return Asset type: loan portfolio Asset type: bank branch/unit lambda Buyer asset growth Buyer log(market to book) Observations

(1) probability(deal)
0.036*** (0.002) 0.199*** (0.019)
0.002 (0.002) -0.127* (0.074)
5673

(2) log(deal value)
0.066*** (0.015) 0.767*** (0.091)
-0.141 (0.395) 0.456*** (0.170) 0.993*** (0.346) -0.005 (0.016) 0.810* (0.442) 0.572 (0.642) 1.193*** (0.426)
252

This table shows the results of estimating the two-stage model on a subsample of data for which the type of asset – be it real estate property, a loan portfolio, or a unit or branch of a bank – was coded. The category ‘property assets’ is left out, while indicators for ‘loan portfolio’ and ‘bank branch/unit’ are included. The variables are described in Table 2.3.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

23

that were in ﬁnancial crisis starting in 2007 or 2008, according to the classiﬁcation by Laeven and Valencia (2008). This includes the United States. There does not appear to be a diﬀerent eﬀect of capital on deal value for countries in crisis. The positive eﬀect of buyer size on deal value is larger when the buyer is from a crisis country, however.

Table 2.9: Determinants of deal value – US sales to buyers in all countries

OLS Buyer country in crisis Buyer capital/assets Buyer capital/assets * buyer country crisis Buyer log(assets) Buyer log(assets) * buyer country crisis Same city Same sub-sector (Buyer - US) stock market return Observations Adj R squared

(1)
0.015*** (0.003)
0.563*** (0.039)
0.062 (0.291) 0.447** (0.194) 0.028 (0.072)
591
0.361

(2) -1.956* (1.016) 0.015*** (0.003) -0.007 (0.006) 0.528*** (0.042) 0.328*** (0.083) 0.128 (0.284) 0.443** (0.190) 0.037 (0.073)
591 0.377

This table presents the estimation of the deal value equation using OLS, including institutions in all countries that purchased an asset from the United States. buyer country crisis’ is an indicator variable for whether the buyer’s country is in a ﬁnancial crisis, starting in 2007/2008, according to the Laeven and Valencia (2008) classiﬁcation. (Buyer – US) stock return is the diﬀerence the in the stock market return of the buying institutions’ country and the US return. The other variables are as in Table 2.3.

CHAPTER 2. LEVERAGE AND FINANCIAL ASSET SALES

24

2.7 Summary and conclusion

I study asset transactions between ﬁnancial ﬁrms in the United States and ﬁnd that greater capital, or lower leverage, increases the probability that a potential buyer will purchase an asset and also increases the value of the deal that takes place. This does not hold for deposit-taking institutions, however, who had greater access to or cheaper funding during the ﬁnancial crisis. These results are consistent with theories that posit that demand for assets is a function of funding availability, and show that credit-constraints of ﬁnancial intermediaries can interact to reduce asset prices and deepen liquidity crises. It would be interesting to see whether these results hold for assets transaction in other countries, where government policies providing support to the ﬁnancial sector diﬀered.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

25

Chapter 3
Interconnections in Banking,
Systemic Risk, and Crisis
3.1 Introduction
The recent global ﬁnancial crisis has prompted a renewed interest in the factors that lead to fragility in the ﬁnancial sector. Waves of panics or failures of banks have been common throughout history in both high and middle-to-low income countries (Reinhart and Rogoﬀ 2009) and, despite an increase in the number of government policy interventions designed to stabilize the banking system, recent decades have not seen a decrease in the frequency of banking crises (Calomiris 2009). In addition to being a recurring phenomenon, banking crises are costly for the economies involved, through their role in triggering or deepening recessions and increasing the debt of governments

CHAPTER 3. INTERCONNECTIONS AND CRISIS

26

who engage in bank bailouts.1 Recent policy discussions on preventing future banking crises focus on
banks that are too big, or too interconnected, to fail. The idea being that the failure of large or interconnected banks has serious negative eﬀects for non-ﬁnancial sectors or contagion risks to other banks. The data on banks’ exposures needed to measure interconnectedness may be reported in part to national banking regulators in some countries, but is not currently publicly available. This paper proposes a novel market-based measure of banking sector interconnectedness which can be calculated for a large group of countries – the correlation of banks’ stock returns – and asks whether interconnected banking sectors are more likely to end up in crisis.
The theory on individual bank fragility is well established, traditionally focusing on the maturity mismatch between bank liabilities and assets. The creation of long-term assets from short-term deposits leaves banks susceptible to panic-based runs on their liabilities (Bryant 1980, Diamond and Dybvig 1983). A number of extensions of the Diamond-Dybvig model relate bank runs to sunspots, and Goldstein and Pauzner (2005) link the probability of a bank run to a signal about the overall state of the economy. Models that incorporate spillovers between banks, which lead a failure to, in fact, become a crisis, have been put forth more recently. Chen (1999) extends
1Considering 42 of the systemic banking crises that occurred between 1970 and 2007, Laeven and Valencia (2008) calculate an average ﬁscal cost associated with crisis management of 13.3% of GDP, and output losses (measured as deviations from trend) averaging about 20% of GDP during the ﬁrst four years of crisis.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

27

Diamond-Dybvig to a set up with multiple banks and interim revelation of information about the performance of some banks. With depositors who update according to Bayes rule, a suﬃcient number of interim bank failures results in pessimistic expectations about the general state of the economy and leads to runs on the remaining banks. In practice, multiple bank failures are likely to be a more precise signal to depositors about the health of the banking industry than economy-wide activity. In the model of Acharya and Yorulmazer (2008), depositors have information on the extent to which banks lend to the same risk-type of borrowers (borrowers who will invest in the same industries, for example). The information spillover from one bank failure then shows up in increased borrowing rates for remaining banks and potentially also in bank failures (if the increased rates are high enough).
I use the correlation of banks’ stock returns as a proxy for interconnectedness to study the relationship between interconnectedness and banking system stability. De Nicol`o and Kwast (2001) and Billio, Getmansky, Lo, and Pelizzon (2010) also use stock returns, or indexes of stock returns, to measure linkages between ﬁnancial institutions. Observing that bond prices reﬂect individual default risk while credit default swap contracts also reﬂect counterparty risk, Giglio (2010) uses the information content of bond and credit default swap prices for 15 large US and European ﬁnancial institutions to create bounds on the probability of joint failures. My measure of interconnectedness can be calculated from data that is readily available for banks in a large number of countries, which makes it ideal for a cross-country

CHAPTER 3. INTERCONNECTIONS AND CRISIS

28

comparison of banking sector distress. First, I estimate a binary Logit probability model to see whether banking crises, as classiﬁed by Laeven and Valencia (2008, 2010) are more likely in interconnected banking sectors. After performing the benchmark estimation, I use an instrumental variables estimation procedure to ensure the results are not driven by endogeneity of banks’ stock return correlation. Second, I estimate the relationship between interconnectedness and the number of bank failures in times of crisis, which is indicative of its severity, using an ordered Logit model. Boyd, De Nicolo`o, and Loukoianova (2009) suggest that since banking crisis indicators are constructed in large part using information on government actions undertaken in response to bank distress, they do not accurately measure the start of a banking crisis but instead capture a variety of government policy responses that occur after the onset of crisis. Third, in light of these concerns about the accuracy of the crisis indicators used in the previous empirical literature, I also consider the number of bank failures that occur in each year, whether a crisis period or not. I estimate the probability of bank failures in all years, distinguishing between large banks and small banks, to see whether the relationship between interconnectedness holds in all periods and similarly for large and small banks.
For a sample of 45 countries from 1993 to 2009, I ﬁnd that both the probability of a banking crisis and the probability of a greater number of large bank failures during times of crisis is increasing in banking sector interconnectedness. I also ﬁnd that the probability of large bank failures is

CHAPTER 3. INTERCONNECTIONS AND CRISIS

29

higher for more connected banking sectors in all years. No signiﬁcant relationship between the interconnectedness of small banks and the probability of small bank failures is found. The paper proceeds as follows: section 3.2 reviews the related literature, section 3.3 describes the measure of interconnectedness, section 3.4 describes the data and empirical method, section 3.5 discusses the results, section 3.6 includes robustness tests, and section 3.7 concludes.

3.2 Related literature
3.2.1 Literature on banking crises
The ﬁrst empirical studies of banking crises found that the probability of crisis was higher in countries with a weak prior macroeconomic environment and a weak institutional environment.2 The literature then expanded into empirical studies of two types: (i) studies of how the economic structure of a banking sector aﬀects its likelihood of having a crisis and (ii) an early-warning systems literature whose goal is to predict the onset of crises. The main ﬁndings from the ﬁrst set of studies are that the probability of banking crisis is decreasing in the concentration or competitiveness of a banking system (Beck, Demirgu¨c¸-Kunt, and Levine 2006; Schaeck, Cihak, and Wolfe 2009,
2Speciﬁcally, the factors that have been found to be positively related to the likelihood of crisis are: low real GDP growth, high inﬂation, high real interest rates, ﬁnancial liberalization, lending books, asset price declines, weak law and order, weak accounting standards, and weak legal enforcement. See Demirgu¨c¸-Kunt and Detragiache (1997, 1998), Hardy and Pazarbasioglu (1999), and Hutchison and McDill (1999).

CHAPTER 3. INTERCONNECTIONS AND CRISIS

30

among others) and, while most cross-country diﬀerences in banking regula-
tion and supervision do not have signiﬁcant eﬀects, the stringency of oﬃcial
capital requirements decreases the probability of banking crises (Barth et al
2004).3 Demirgu¨¸c-Kunt and Detragiache (2002) ﬁnd that the existence of an
explicit deposit insurance scheme increases a country’s probability of having
a banking crisis. The authors suggest that the positive relationship may be
the result of a moral hazard eﬀect of deposit insurance, whereby deposit-
taking institutions with limited liability increase their risk-taking. The early
warning systems literature is growing (see Frankel and Saravelos (2010) and
chapter 3 of the IMF’s September 2011 Global Financial Stability Report).
Rose and Spiegel (2009), however, estimate a multiple-indicator multiple-
cause model for 107 countries to examine sixty-ﬁve potential causes of the
2008 crisis and ﬁnd few factors that are robustly linked to the incidence of
crises across countries. They warn that this bodes poorly for early warnings
models, which would have to predict the timing of crises out-of-sample in
3Policy studies of the overall economic eﬀect of increasing banks’ capital requirements have recently been conducted by the Basel Committee on Banking Supervision (BCBS) and economists at the Bank of Canada and the Bank of Japan. Expecting a negative relationship between capital adequacy and the likelihood of a banking crisis, these papers estimate probability of crisis models to help quantify the expected beneﬁt of higher capital requirements. Perhaps as a result of having a very speciﬁc goal in mind, these studies each focus on a small group of countries and use few explanatory variables in their Logit or Probit probability of crisis estimations. As expected, they ﬁnd that a higher capital ratio in the banking sector reduces the probability of a banking crisis. Speciﬁcally, the Bank of Canada report (August 2010) ﬁnds that an increase of 2 percentage points in the aggregate capital to assets ratio decrease the probability of a banking crises by between 0.8 and 2.6 percentage points. The BCBS/FSB Long Term Economic Impact report found that an increase of 2 percentage points in bank capital ratios reduced the probability of a ﬁnancial crisis by 2.9 percentage points.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

31

addition to successfully predicting crises in the cross-section. Instead of relying on the standard events-based indicators of banking
crises,4 Von Hagen and Ho (2007) develop an index of money market pressure and identify banking crises as periods in which there is excessive demand for liquidity in the money market. Deﬁning crisis episodes in this way, they ﬁnd evidence for macroeconomic factors that precede banking crises that are consistent with prior studies.5 In addition, several papers study the interplay between banking crises and currency crises or sovereign debt crises (Kaminsky and Reinhart (1999), Glick and Hutchinson (2000), Reinhart and Rogoﬀ (2011)).

3.2.2 Measuring systemic risk
Systemic risk is the risk of joint defaults in the ﬁnancial system. That is, the
risk of a banking crisis occurring. De Bandt and Hartmann (2000) provide
precise deﬁnitions of systemic risk and crises.6 Banking crises result from
4Demirgu¨¸c-Kunt and Detragiache (2002, 2005), Caprio et al. (2005), Reinhart and Rogoﬀ (2008b), and Laeven and Valencia (2008, 2010). These banking crisis indicators build on the classiﬁcation ﬁrst compiled by Caprio and Klingebiel (1996, 1999).
5That is, that a slowdown of real GDP, lower real interest rates, extremely high inﬂation, large ﬁscal deﬁcits, and over-valued exchange rates tend to precede banking crises in 47 countries from 1980 to 1996.
6A narrow systemic event is one in which the release of bad news about a ﬁnancial institution, or its failure, leads in a sequential fashion to considerable adverse eﬀects on one or several other ﬁnancial institutions. That is, a case where a failure of one bank due to an idiosyncratic shock spreads to another bank or several other banks. A broad systemic event includes not only the events described above but also simultaneous adverse eﬀects on a large number of institutions or markets as a consequence of a severe and widespread (systematic) shock. Next, a systemic event is categorized as strong if the institution(s) aﬀected in the second round or later actually fail as a consequence of the initial shock, although they were solvent ex ante. Putting these together, their deﬁnition

CHAPTER 3. INTERCONNECTIONS AND CRISIS

32

two types of events: (1) an idiosyncratic shock that causes one ﬁnancial institution to fail, with this failure spreading to other institutions, or (2) a systematic shock that aﬀects multiple ﬁnancial institutions causing a joint default. Most crises will in fact lie in between these two extreme types but, for a given set of external shocks, banking systems in which banks are more connected, whether through similar distributions of claims on non-ﬁnancial ﬁrms or signiﬁcant claims on each other, will be more likely to end up in crisis. A more interconnected banking system may have a more severe crisis in response to a systematic shock. In practice, the onset of banking crises are dated based on a combination of: (i) observing a large number of defaults in a particular country, (ii) large changes in the aggregate balance-sheet of the banking sector that indicate distress,7 and (iii) some judgment about the seriousness of the events. When identifying banking crises, of course, it is diﬃcult to diﬀerentiate between crises that result from the propagation of an idiosyncratic shock versus crises that are due to systematic shocks.
Recent work on systemic risk focuses on measuring the individual contributions to systemic risk of particular ﬁnancial institutions. Acharya, Pedersen, Philippon, and Richardson (2010) measure a ﬁnancial institution’s contribution to systemic risk (the systemic expected shortfall) as its propensity to be undercapitalized when the system as a whole is undercapitalized.
of a systemic crisis is a “systemic event (narrow or broad) that aﬀects a considerable number of ﬁnancial institutions or markets in a strong sense, thereby severely impairing the general well-functioning of the ﬁnancial system.”
7For example, an exhaustion of aggregate banking system capital, or sharp increases in non-performing loans.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

33

The CoVaR measure of Adrian and Brunnermeier (2009) captures the Valueat-Risk of ﬁnancial institutions conditional on other institutions being in distress. These analyses aim to measure contributions to systemic risk by conditioning on manifestations of systemic risk – that is, on ﬁnancial crises – with the goal of improving the regulation and supervision of individual banks, which is currently based on individual risk measures. The goal of this paper is to use variation in the interconnectedness of banking sectors across countries to determine if greater interconnectedness in the sector increases the probability that a crisis will occur.
An analysis of systemic risk in the Unites States that is closest in spirit to my measure of interconnectedness is that of Billio, Getmansky, Lo, and Pelizzon (2010). They calculate Granger causality networks among the returns of indexes of the one-hundred largest banks, brokers, hedge funds, and insurance companies in the United States for ﬁve diﬀerent sub-periods: 1994-1996, 1996-1998, 1999-2001, 2002-2004, and 2006-2008. They ﬁnd that connections increase before ﬁnancial crises (the crisis triggered by the collapse of Long Term Capital Management in 1998, and the subprime crisis) and during crises. Observing that bond prices reﬂect individual default risk while credit default swap contracts also reﬂect counterparty risk, Giglio (2010) uses the information content of bond and credit default swap prices for 15 large US and European ﬁnancial institutions to create bounds on the probability of joint failures. My measure of interconnectedness can be calculated from data that is readily available for banks in a large number of countries, which makes

CHAPTER 3. INTERCONNECTIONS AND CRISIS

34

it ideal for a cross-country comparison of banking sector distress and allows it to shed light on the relationship between interconnectedness before a crisis and the manifestation of crises.

3.3 Measure of interconnectedness
To measure interconnectedness in the banking sector, I calculate the correlation of the quarterly stock returns of the largest ten banks in each country, over the previous 3 years. That is, I calculate the rolling three-year (12 quarter) correlation between each pair of banks and take the simple average.8 In the absence of data on the returns on banks’ loan and securities portfolios, the correlation of stock returns serves as a proxy for the correlation of their portfolio returns. There is a positive relationship between the accounting return on assets (net income divided by assets) of banks and stock returns in all but one country in the sample. Market returns reﬂect information more rapidly than measures based on accounting variables and the frequency of stock price data allows us to measure the correlation more precisely than could be done using information from annual accounting data.
Two prior studies have used stock returns to measure interconnections between banks. As previously mentioned, Billio et. al. (2010) use the correlations of returns of indexes of US ﬁnancial institutions to measure the linkages component of systemic risk9 and, in a study of whether the consoli-
8The results are robust to using the correlation of the returns of the largest ﬁve banks, by assets, and also to using ﬁve-year (20 quarter) rolling windows instead.
9Their framework is meant to cover the“four L’s” of systemic risk: liquidity, leverage,

CHAPTER 3. INTERCONNECTIONS AND CRISIS

35

dation of ﬁnancial ﬁrms in the United States throughout the 1990s increased systemic risk, De Nicol`o and Kwast (2001) use the correlation of the stock returns of large and complex banks as a measure of their interdependencies.
Figure 3.1 shows the correlation measure calculated for the largest 5 nonﬁnancial ﬁrms in the United States, and Figure 3.3 of the appendix shows the correlation measure calculated for the largest 5 ﬁrms in several diﬀerent sectors in the United States. There is no comparable increase in the correlation measure prior to the crisis for the non-ﬁnancial sectors.

Figure 3.1: Interconnectedness of largest 5 banks and interconnectedness of
largest 5 non-ﬁnancial ﬁrms in the US
Correlation of stock returns
United States

0 .2 .4 .6 .8
199409 199512 199703 199806 199909 200012 200203 200306 200409 200512 200703 200806 200909

Largest 5 non-financial firms linkages, and losses.

Largest 5 banks

CHAPTER 3. INTERCONNECTIONS AND CRISIS

36

Table 3.1 shows the number of banking crises in the sample and the time in crisis as the correlation measures increases. There is a positive relationship between crises and the correlation measure: in the lowest tertile of the correlation measure, 6.6% of the sample is in a banking crisis, while in the highest tertile it increases to 22% in crisis.

Table 3.1: Number of banking crises and time in crisis by interconnectedness

Mean

Correlation of Bank Stock Returns (top 10)

Tertile 1

Tertile 2

Tertile 3

0.072

0.590

0.904

Crisis Years (%) No. Crises

6.57 3

9.90

21.97

Chi-squared: 16.4***

3 17 Chi-squared: 18.8***

I also split the sample into the largest and smallest banks in each country and calculate the measures of interconnectedness using the correlation of the stock returns of the largest 5 banks and of the smallest 5 publicly-traded banks, respectively. Figure 3.2 shows the interconnectedness of large banks and of small banks in the United States over time. The interconnectedness of both the largest banks and smallest banks increased in the years prior to the subprime crises, but the increase was steadier for the largest 5 banks.
Several potential problems with the measure of interconnectedness must be addressed. First, do changes in the stock return correlation simply reﬂect changes in the overall stock market? In order to ensure this is not the case, I scale the stock price of each bank by the price of a national market index for the country in which the bank is located and calculate correlations using the

CHAPTER 3. INTERCONNECTIONS AND CRISIS

37

Figure 3.2: Interconnectedness of large banks and interconnectedness of small banks in the US
Interconnectedness - Largest and smallest banks in the United States

199703 -.2 0 .2 .4 .6
199806 199909 200012 200203 200306 200409 200512 200703 200806 200909 201012

Smallest 5 banks

Largest 5 banks

stock returns based on these relative prices.10 Second, do cross-country differences in the stock return correlation reﬂect national factors that determine stock prices? The estimation strategy described in the next section includes country factors and the instrumental variables estimation in the robustness section shows that the main results are not driven by simultaneity.
10I also control for market returns and volatility in the robustness section.

CHAPTER 3. INTERCONNECTIONS AND CRISIS
3.4 Data and empirical method

38

3.4.1 Probability of a banking crisis
To study the relationship between banking sector interconnectedness and the probability of a crisis, I conduct a panel estimation using a Logit probability model. In the benchmark speciﬁcation, an indicator variable equal to one when country i is in a banking crisis in year t is regressed on the variable of interest – the measure of interconnectedness – and control variables Z, both lagged by one year, country ﬁxed factors αi and year eﬀects vt:

Pr(Yit = 1) = Λ(αi + βInterconnectednessi,t−1 + Zi,t−1γ1 + vt) (3.1)

where Λ(βX) = exp(βX)/ exp(1 + βX) is the logistic function. A positive and signiﬁcant estimate of β indicates that banking sectors that are more interconnected are more likely to end up in crisis. I use the dates of banking crises provided by Laeven and Valencia (2008, 2010), discussed further in section 3.4.4, to create the banking crisis indicator.11 Using this method, twenty-three crises occur between 1993 and 2009 in the 45 country sample. Table 3.11 of the appendix lists the countries and banking crises in the sample. Interconnectednessi,t−1 is the 12 quarter (three-year) rolling correlation of the quarterly stock returns of country i’s largest ten banks discussed in
11The majority of the previous literature on the probability of banking crises is based on these indicators of banking crises. Von Hagen and Ho (2007) is an exception.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

39

the previous section.12
Turning to the control variables, as in prior studies of banking crises, real
GDP growth is included to capture macroeconomic developments that may
aﬀect the quality of bank’s assets. A time-variant dummy variable equal to
one in the presence of an explicit deposit insurance policy and the Ilzetzki,
Reinhart and Rogoﬀ (2008) classiﬁcation of the exchange rate regime, which
is increasing in ﬂexibility, are also included as controls. The share of ﬁnancial
sector assets that is foreign is included to control for the banks’ susceptibility
to foreign shocks, and the log of per capita GDP in US dollars is a proxy
for diﬀerences in the institutional environments of each country. Country
ﬁxed-eﬀects are included in all speciﬁcations, to control for any inter-country
variation that is constant over the sample period. For example, one would
expect variation in the strength of banking regulation and supervision across countries to aﬀect the probability of banking crises.13 Time dummies are
included in all speciﬁcations, to capture time-dependent shocks to the world
economy. All of the explanatory variables are lagged by one year to minimize
concerns about possible simultaneity. In addition, as in the previous litera-
ture, I keep only the ﬁrst year of a banking crisis and the years in which there
12The results are very similar using the three-year rolling correlations of the top ﬁve banks, by assets, as well as the ﬁve-year rolling correlations of the top 10 banks or top 5 banks.
13Barth, et al (2004), however, create indexes of various components of banking regulations and supervision for a cross-section of countries in 1999, and ﬁnd they have little signiﬁcance in explaining banking crises after including country indexes of the quality of institutions in the analysis. Also, since regulation varies little over time and, speciﬁcally, because the indexes are time-invariant, it is not possible to include both the indexes of regulation and country ﬁxed eﬀects in the estimation.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

40

was no crisis in my sample to minimize concerns about possible endogeneity between the explanatory variables and the occurrence of crises. That is, the subsequent years of each banking crises are not included in the estimation.14 Standard errors are adjusted for heteroskedasticity following Huber (1967) and White (1980), generalized to allow for correlation within countries.

3.4.2 Probability of n bank failures during a crisis
Next, I identify the number of banks that failed during each crisis and esti-
mate the relationship between interconnectedness and the number of bank
failures using an ordered Logit model. Systemic risk is deﬁned as the risk of
joint failures, but the banking crisis indicators used in the literature use a
variety of data to determine whether a banking crisis has occurred. Even af-
ter the establishment of deposit insurance funds to protect depositors in the
case of failures, many governments have continued to rescue distressed banks
precisely due to concerns about contagion. The number of bank failures is
an indication of the severity of a banking crisis, although the importance
of the particular banks that failed must of course be taken into account. I
distinguish between the number of large and small bank failures, where large
14The discrepancies across diﬀerent classiﬁcations of crises are greater for the end dates than the starting dates. For example, for only the start year of the crisis, the discrepancy between the Laeven and Valencia classiﬁcation and that of Reinhart and Rogoﬀ is equal to 1.7 percent of common country-years, but when considering all crisis country-years, the discrepancy is 7 percent. So for robustness, I also drop years two and three of each banking crises, instead of each year until the end given in the Laeven and Valencia classiﬁcation, following Dell’ Ariccia, Detragiache, and Ragan (2008) who take the length of banking crises to be three years, and Demirgu¨¸c-Kunt et al. (2006) who ﬁnd that GDP growth returns to its pre-crisis level in the fourth year of a crisis.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

41

banks are those that have assets greater than the median bank in the sector. Ordered Logit models can be derived from an underlying latent variable model just as in the case of binary dependent variables.15 We assume that the latent variable y∗ is determined by

y∗it = βInterconnectednessi,t−1 + Zi,t−1γ1 + vt + it

(3.2)

where it ∼ Λ(βInterconnectednessi,t−1 + Zi,t−1γ1). The dependent vari-
able is now the number of bank failures in country i in year t. Instead of
considering each possible integer value of the number of bank failures, I let
y = 3 when there are three or more bank failures. This reduces the number
of threshold parameters to be estimated and does not aﬀect the remaining
parameter estimates which are used to interpret the partial eﬀects of the
explanatory variables. The explanatory variables are as in the binary Logit
model of the previous section.
15Let X = (Interconnectedness, Z) and θ = (β, γ1) and deﬁne threshold parameters (α1, α2, α3) such that: y = 0 if y∗ ≤ α1, y = 1 if α1 < y∗ ≤ α2, y = 2 if α2 < y∗ ≤ α3, and y = 3 if y∗ > α3. Then the probabilities of one, two, or three failures are given by:
Pr(y = 0|X) = Λ(α1 − Xθ) Pr(y = 1|X) = Λ(α2 − Xθ) − Λ(α1 − Xθ) Pr(y = 2|X) = Λ(α3 − Xθ) − Λ(α2 − Xθ) Pr(y = 3|X) = 1 − Λ(α3 − Xθ)
and the parameters (α1, α2, α3, θ) can be estimated by maximum likelihood.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

42

3.4.3 Bank failures and share of failed assets in all
years - large versus small banks
Regulators take diﬀerent approaches to the rescue or resolution of large banks and small banks and the business models of the two groups vary. The literatures on banking crises and systemic risk are primarily concerned with the failure of large ﬁnancial institutions, as these failures are likely to have the most severe eﬀects – both in terms of contagion to other banks and on other sectors of the economy. The failures of even small banks can have detrimental eﬀects on non-ﬁnancial economic activity, however. Ashcraft (2003) ﬁnds that the failure of relatively small US banks in 1998 and 1992 permanently reduced local real county income by 3 percent, through a decline in bank lending.
In order to examine whether interconnectedness has the same eﬀect on the failure of both large and small banks, I divide each country into two regions: the “large” banking sector and the “small” banking sector. The large banking sector is categorized in one of two ways. It includes (i) all banks that are larger than the median bank in each country-year, by assets, or (ii) all banks that have assets greater than 10 billion US dollars.16 The small banking sector for each country-year consists of the remaining banks.
16I present the results using the median classiﬁcation ﬁrst, and results using the banks larger and smaller than 10 billion US dollars second. Initial considerations of systemically important ﬁnancial institutions (SIFIs) deﬁned them to be institutions with assets greater than 50 billion US dollars. These are primarily focused on banks in wealthier countries in Europe and the United States. I use a threshold of 10 billion US dollars to include large institutions in emerging markets.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

43

This leaves us with a country-sector-year panel. I then estimate the probability of a bank failure occurring in each country-
sector-year using a Logit model, similar to before, except I include a (i) dummy variable for the large sector, (ii) the interaction between interconnectedness and the large sector, and (iii) the share of banking system assets in each country-sector-year observation to control for the relative sizes of the large and small sectors. The model is:

y∗ijt = α + β1Interconnectednessi,t−1 + β2Dj + β3(Interconnectednessi,t−1 × Dj)

+ β4Shareijt + Zi,t−1γ1 + vt + ijt

(3.3)

In this speciﬁcation the dependent variable is (i) an indicator variable equal to one if there was a bank failure in sector j of country i in year t and (ii) the share of assets of the failed banks in sector j of country i in year t. A positive and signiﬁcant β1 indicates that interconnectedness among small banks increases the probability of small bank failures and a positive and signiﬁcant β3 indicates that interconnectedness among large banks increases the probability of large bank failures.

3.4.4 Data
I assemble an international panel of 45 high and middle income countries, over the period from 1993 to 2009. The dataset has four main components: the indicators of banking crises from Laeven and Valencia (2008, 2010), bank

CHAPTER 3. INTERCONNECTIONS AND CRISIS

44

histories from the Bankscope database to determine bank failures, bank stock price data from Datastream to calculate the measure of banking system interconnectedness, and the control variables.
The most up-to-date classiﬁcations of banking crises are provided by Laeven and Valencia, henceforth LV, and Reinhart and Rogoﬀ (2009), henceforth RR. In this paper, I use the classiﬁcation of LV as they study a smaller set of countries and appear to have more precise dates.17 See Table 1 of Boyd, De Nicol´o, and Loukoianova (2009) for a comparison of four classiﬁcations of banking crises, including LV and RR. LV deﬁne a systemic banking crisis as a situation in which a country’s corporate and ﬁnancial sectors experience a large number of defaults and ﬁnancial institutions and corporations face great diﬃculties repaying contracts on time. Using this broad deﬁnition of crisis, and combining quantitative data with some subjective assessment of the situation, they identify the starting year of systemic banking crises around the world since 1970. They deﬁne the end of a crisis as the year before two conditions hold: real GDP growth and real credit growth are positive for at least two consecutive years. In case there is growth in real GDP and real credit in the ﬁrst two years of a crisis, the crisis is dated to end in the same year that it starts.
Information on bank failures in the United States is from the FDIC’s Historical Statistics on Banking. Bank failures in other countries are identi-
17For example, LV dates the banking crisis in Japan from 1997 to 2001 while RR dates it from 1992 to 2000, RR do not count the 1988 US savings and loan crisis whereas LV do, and a few mistakes were noted in the dates RR compiled from other sources.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

45

ﬁed from the Bankscope dataset, which contains paragraph long histories of banks. Speciﬁcally, in the case of banks that have left the Bankscope dataset, the history identiﬁes the year in which the bank has become inactive and a short description of the circumstances. I count banks that have been listed as bankrupt, liquidated, or had their banking licenses revoked as failed banks. In times of ﬁnancial distress, mergers are often encouraged or forced by national authorities. The data does not allow me to distinguish between these “bad” mergers and voluntary mergers, so I do not count banks that have ceased to exist due to mergers as failed banks. The number of bank failures thus identiﬁed can be seen as an underestimate of distress in the banking sector. This is a contribution to the data on ﬁnancial crises, as reliable data on bank failures does not exist for many countries. Tables 3.12 and 3.13 of the appendix show the number of bank failures, both large and small, and their mean size for each country in the sample.
The data on real GDP growth comes from the World Bank’s World Development indicators and information on deposit insurance schemes comes from the World Bank’s Deposit Insurance around the World dataset, which contains federal deposit insurance policies and characteristics for a large sample of countries since its inception in the United States in 1933.18 The classiﬁcation of exchange rate regimes is the ﬁne classiﬁcation from Ilzetzki, Reinhart
18The dataset ends in 2003, so I update the key variable, an indicator for whether there is an explicit deposit insurance scheme in place in a given country in a given year to the present and correct a mistake in the database regarding the years during which Argentina has had explicit deposit insurance. Almost all explicit deposit insurance schemes are mandatory, requiring all banks that collect time or savings deposits to join.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

46

and Rogoﬀ (2008). To maximize sample size, I use an unbalanced panel in which some country-year observations are missing. I exclude, however, countries for which there are less than six subsequent years of data available. Constraints on the availability of stock price data leave us with a sample of 45 countries, in which 23 banking crises took place, from 1993 to 2010, after excluding 1 percent of outliers on either tail of the distributions. Table 3.2 provides descriptive statistics for the explanatory variables.

Table 3.2: Descriptive statistics

Interconnectedness - top 10 Interconnectedness - top 5 Interconnectedness - bottom 5

Obs 678 678 500

Mean 0.50 0.57 0.46

StdDev 0.23 0.25 0.27

RealGDPgrowth Deposit Insurance ExchangeRateRegime ForeignAssets/Assets Concentration (top 3 banks) Share maj government owned Market index quarterly return Market index volatility

720 720 716 720 670 572 516 488

3.54 0.79 7.57 17.63 63.53 21.44 4.69 11.29

3.43 0.40 4.32 16.13 18.95 25.21 16.24 7.90

Min -0.51 -0.67 -0.65
-10.89 0 1
0.27 20.09
0 -66.89
0.32

Max 0.99 0.99 0.98
18.29 1
15 74.53 96.40
81 92.51 51.92

CHAPTER 3. INTERCONNECTIONS AND CRISIS
3.5 Results

47

3.5.1 Probability of a banking crisis
The results of the benchmark estimation support the hypothesis that higher interconnectedness increases the probability of a banking crisis occurring. The measure of interconnectedness, the return correlation of the largest ten banks, has a positive and signiﬁcant eﬀect on the probability of crisis (see Table 3.3). The average marginal eﬀect corresponding to the coeﬃcient of 3.71 indicates that a 1 percent increase in the return correlation increases the probability of a crisis by 5.8 percentage points. This is a large eﬀect and greater in magnitude than the eﬀect of real GDP growth by 2 percentage points. Consistent with the previous literature, prior real GDP growth decreases the probability of crisis and the coeﬃcients on the deposit insurance and exchange rate regime variables are negative, as expected, but not signiﬁcant.

3.5.2 Probability of n bank failures during a crisis
While the probability of a banking crisis is increasing in the correlation measure, the results of the estimation of the probability of the number of bank failures indicate that higher correlation does not necessarily predict more bank failures, when all bank failures are counted. The estimated coeﬃcient of interconnectedness on the probability of n bank failures, where n is 1, 2, or more than 3, is not signiﬁcant when all bank failures are counted. When

CHAPTER 3. INTERCONNECTIONS AND CRISIS

48

Table 3.3: Determinants of banking crises – benchmark results

Dependent variable: Banking Crisis Indicator

Logit

Coefficient Coefficient

(1) (2)

Interconnectedness (t-1)

3.709**

(1.516)

RealGDPgrowth (t-1) Deposit Insurance (t-1) Exchange Rate Regime (t-1) No. obs

-0.351* (0.200) -2.708 (1.797) -0.136 (0.143)
612

-0.341* (0.183) -2.649 (1.969) -0.143 (0.155)
612

eF/eX (2)
1.686
-1.267
-1.978
-1.028

Pseudo-R2 No. crises correctly predicted % crises + % non-crises correct

0.51 18/23 137.7

0.52 19/23 149.0

The table presents panel regressions for 45 countries over the 1993–2009 period. The dependent variable is the banking crisis indicator, a discrete variable equal to 1 if there is a banking crisis in year t and zero otherwise. The independent variables are the rolling correlation of the largest 10 banks’ quarterly stock returns, from year t-1 to t-4, real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. Standard errors (in parentheses below the coeﬃcient estimates) are adjusted for heteroskedasticity following Huber (1967) and White (1980), generalized to allow for correlation within countries. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

49

restricting consideration to only large bank failures, however, the probability of the number of failures is increasing in interconnectedness. A one percent increase in the correlation results in a 1.4 percentage point increase in the probability of two large bank failures, and a 1.1 percentage point increase in the probability that three or more banks will fail (see Table 3.4).
When counting only the number of small banks failures, I ﬁnd no signiﬁcant relationship between interconnectedness and the number of failures. This could be because a larger share of the deposits of small banks will be insured deposits, making them less susceptible to failure through runs on deposits. Bologna (2011) ﬁnds indications that banks in the United States with a higher share of deposits above the level covered by deposit insurance19 had a higher probability of failure between 2007 and 2009.

19That is, the share of deposits with denominations greater than $100,000. The FDIC insured deposits up to the amount of $100,000 at the start of the sample, although this was increased to $250,000 in 2008.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

Table 3.4: Probability of n bank failures during a crisis

Dependent variable: Number of bank failures Ordered Logit

All banks

Large banks

Small banks

Interconnectedness
RealGDPgrowth
No. obs Pseudo-R2

-0.703 (0.668) -0.106** (0.042)
612 0.34

1.902* (1.105) -0.160** (0.070)
612 0.39

-1.791 (0.834) 0.046 (0.082)
419 0.33

d{Prob 1 failure}/e(Interconnectedness) d{Prob 2 failures}/e(Interconnectedness) d{Prob 3+ failures}/e(Interconnectedness)

-0.02 -0.009 -0.016

0.026 0.014 0.011

-0.036 -0.012 -0.014

The table presents panel regressions for 45 countries over the 1993–2009 period. The dependent variable is a discrete variable equal to 1 if there is one bank failure in country i in year t, equal to 2 if there are two failures in country i in year t, and equal to 3 if there are three or more bank failures. Banks are classiﬁed as large if they have assets greater than the median bank in each country-year observation. The independent variables are the rolling correlation of the largest 10 banks’ quarterly stock returns (the correlation of the smallest 5 banks in the third column) from year t-1 to t-4, real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

50

CHAPTER 3. INTERCONNECTIONS AND CRISIS

51

3.5.3 Bank failures and share of failed assets in all
years - large versus small banks
To further study the relationship between interconnectedness and bank failures in all years, we turn now to the country-sector-year panel, where the sector denotes either the large banks in a country’s banking sector or the smaller banks. We again see a diﬀerent eﬀect of interconnectedness on large banks and on small banks. The estimated relationship between interconnectedness and the probability of large bank failures (table 3.5) is consistent with the results on banking crises. A one percent increase in the correlation measure is associated with an approximately 5 percentage point increase in the probability of large bank failures. The coeﬃcient on the dummy variable for the large sector shows that the probability of failure of large banks is lower, by 7 percentage points when classifying large banks as those with assets greater than 10 billion US dollars.
Finally, corresponding to the bank failures in Table 3.5, Table 3.6 shows the results of the estimation of model 3.3 with the share of assets of failed banks to all banks as the dependent variable. These results are, as they should be, consistent with the results of the probabilities of failures presented in Table 3.5. Higher interconnectedness of large banks is associated with a higher share of failed assets, 0.63 percentage points higher for a one percent increase in correlation in the median classiﬁcation. For the small banking sector, however, the estimated coeﬃcient on interconnectedness is

CHAPTER 3. INTERCONNECTIONS AND CRISIS

52

again negative but not signiﬁcant. Interconnectedness has no signiﬁcant effect on the probability of small bank failures, or the share of assets made up of small banks that fail.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

Table 3.5: Probability of bank failures in all years – country-size-year panel

Dependent variable: Bank failure indicator

Logit

Median classification

10B US$ classification

Coefficient dF/eX

Coefficient dF/eX

Interconnectedness (t-1)

-2.602*** -0.074

-2.022*

-0.034

(0.972)

(1.124)

Large

-3.241*

-0.126

-4.840*** -0.073

(1.697)

(1.554)

Large*Interconnectedness (t-1)

2.892**

0.065

5.472**

0.054

(1.410)

(2.406)

RealGDPgrowth (t-1)

-0.095

-0.012

-0.102

-0.008

(0.089)

(0.121)

Share sector (t-1)

0.028

0.106

0.006

0.010

(0.018)

(0.004)

No. obs

982 982

Pseudo-R2

0.42 0.47

No. failure cases correctly predicted

42/99

18/56

% crises + % non-crises correct

140.4

131.2

The table presents panel regressions for 45 countries over the 1993–2009 period. The dependent variable is a discrete variable equal to 1 if there are bank failure(s) in country i in the large or small section of the ﬁnancial sector j in year t. The independent variables are the rolling correlation of banks’ quarterly stock returns from year t-1 to t-4, real GDP growth, and the share of ﬁnancial sector assets made up of banks in group j, all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

53

CHAPTER 3. INTERCONNECTIONS AND CRISIS

Table 3.6: Assets of failed banks in all years – country-size-year panel

Dependent variable: Log share of failed assets

Logit

Median classification

OLS Tobit

log Interconnectedness

-0.240

-0.573*

(0.210)

(0.341)

Large

0.281

-0.899

(0.603)

(1.131)

log Large*Interconnectedness

0.630**

1.746*

(0.303)

(1.033)

RealGDPgrowth

0.068*

-0.027

(0.038)

(0.068)

log Share sector

-0.120

0.052

(0.141)

(0.208)

No. obs

303 303

R2 0.38

Pseudo R2

0.14

10B US$ classification

OLS Tobit

-0.339

-0.755**

(0.203)

(0.303)

0.155

-1.338

(0.285)

(0.864)

0.473*

1.035

(0.260)

(1.194)

0.045

-0.041

(0.038)

(0.070)

-0.213** -0.464***

(0.101)

(0.177)

311 311

0.37

0.20

The table presents panel regressions for 26 countries over the 1995–2009 period. The dependent variable is log (assets of failed banks in ijt/total assets in ijt). The independent variables are the rolling correlation of banks’ quarterly stock returns from year t-1 to t-4, real GDP growth, and the share of ﬁnancial sector assets made up of banks in group j, all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

54

CHAPTER 3. INTERCONNECTIONS AND CRISIS
3.6 Robustness

55

3.6.1 Instrumental variables estimation
The results of the previous section indicate that the probability of a banking crisis, and similarly of the failure of large banks, is increasing in the level of interconnectedness of the banks prior to crisis. In order to ensure the results are not driven by increasing correlation in times of crises, or by factors that increase both the correlation of banks’ stock returns and the probability of crises, I conduct instrumental variables estimation. The instrument for the banks’ stock return correlation is the ratio of the value of total shares traded on the national stock market to real market capitalization. Although stock market capitalization is known to fall during most banking crises, there is no clear relationship expected between crises and the total value of shares traded. While the value of certain stocks may fall, turnover may increase. At the same time, increased turnover is likely to have a positive eﬀect on the return correlation. Panel B of Table 3.7 shows the ﬁrst stage regression, where the t-test that the instrument is not related to interconnectedness can be rejected at a 5% level of conﬁdence. Panel A shows the second stage results. The 2SLS result corresponds to a linear probability model, with the standard errors corrected for heteroskedasticity, and the second stage Logit coeﬃcient is consistent with the benchmark results. Both show that an increase in interconnectedness increases the probability of a banking crisis.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

56

Table 3.7: Determinants of banking crises – IV estimation

Panel A:

Probability of Crisis

Dependent variable: Banking Crisis Indicator
Interconnectedness (t-1) RealGDPgrowth (t-1) No. obs R2 Pseudo-R2

(1) 2SLS 1.518** (0.741) 0.008 (0.009) 572 0.22

(2) 2nd-stage logit
37.958* (22.024) -0.863 (1.374)
572
0.78

Panel B:

First Stage Estimation

Dependent variable: Interconnectedness
StockMarketValueTraded
No. obs F-stat R-squared

(1) 0.053** (0.021)
572 20.25 0.57

ThReobtuasbtlestapnredsaerndtesrrinosrstriunmpaernetnatlhveaserisa, b*l*e*s pe<s0ti.m01a,t*i*onp<r0e.s0u5l,ts* fpo<r0.4150 countries over the 1993–2009 period. Panel B present the ﬁrst stage estimation, where the dependent variable is the rolling correlation of banks’ quarterly stock returns and the instrument is the total value of shares traded in the stock market for country i. Panel A presents the second stage estimations, using both 2SLS (corresponding to a linear probability model) and a Logit probability model. The dependent variable is a discrete variable equal to 1 if there are bank failure(s) in country i in the large or small section of the ﬁnancial sector j in year t. The independent variables are the (instrumented) rolling correlation of banks’ quarterly stock returns from year t-1 to t-4, real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

57

3.6.2 Additional controls
To check for the possibility that the observed relationship between banking crises and interconnectedness is driven by stock market movements that aﬀect the stock return correlation, I next control for the return on a national stock market index, as well as the volatility of the index. The positive eﬀect of interconnectedness on the probability of banking crisis is robust to these inclusions (Table 3.8).
Past studies have found mixed results concerning the eﬀect of competition on banking crises. I ﬁnd concentration, measured as the share of assets held by the three largest banks, to have no signiﬁcant eﬀect on the probability of banking crises (Table 3.9). Karolyi and Taboada (2010) calculate the share of banks that is majority owned by the government for 1995, 2000, and 2005. In a reduced sample20 with these three years, I check whether the share of banks that are majority government owned decreases the probability of crises, as would be expected. I ﬁnd no signiﬁcant eﬀect of government ownership on the probability of crisis.
20The sample is reduced as the share of government ownership is calculated only for 1995, 2000, and 2005.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

58

Table 3.8: Determinants of banking crises – overall stock market movements

Dependent variable: Banking Crisis Indicator

Logit

(1) (2)

Interconnectedness (t-1)

5.258**

8.410**

(2.306)

(4.194)

Market index return (t-1)

-0.008

(0.023)

Market index volatility (t-1)

0.076

(0.074)

RealGDPgrowth (t-1)

-0.427*

-0.950***

(0.219)

(0.315)

Deposit Insurance (t-1)

Foreign asset share (t-1)
No. obs
Pseudo-R2 No. crises correctly predicted % crises + % non-crises correct

0.034 (0.065)
488
0.57 14/20 155.2

0.254*** (0.091)
465
0.71 12/20 179.5

The table presents panel regressions for 45 countries over the 1995–2009 period. The dependent variable is the banking crisis indicator, a discrete variable that equals one if there is a banking crisis in year t and zero otherwise. The independent variables are the rolling correlation of the largest 10 banks’ stock returns, from t1 to t-4, the return on a national market index in the last quarter of the prior year, the rolling standard deviation of the market quarterly returns, from t-1 to t-4, real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

59

Table 3.9: Determinants of banking crises – concentration and government

ownership of banks

Dependent variable: Banking Crisis Indicator

Logit

(1)

Interconnectedness (t-1)

4.633***

(1.662)

Concentration (t-1)

-0.019

(0.038)

Share of Government

Owned Banks (t-1)

RealGDPgrowth (t-1)

-0.365**

(0.186)

No. obs

609

(2) 4.378* (2.493)
0.024 (0.022) -0.482 (0.339)
119

Pseudo-R2 No. crises correctly predicted % crises + % non-crises correct

0.55 19/23 149.0

0.18 0/3 100.0

The table presents panel regressions for 45 countries over the 1993–2009 period in column (1) and for 1993, 2000, and 2005 in column (2). The dependent variable is the banking crisis indicator, a discrete variable that equals one if there is a banking crisis in year t and zero otherwise. The independent variables are the rolling correlation of the largest 10 banks’ stock returns, from t-1 to t-4, the concentration of the banking sector (measured as the share of assets in the largest three banks), the share of banks that are majority government owned from Karolyi and Taboada (2010), real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

60

Following Allen, et al. (2002), several measures of balance-sheet risk exposure were investigated but not found to have a signiﬁcant eﬀect on the probability of banking crises. They lay out a framework for understanding crises based on the examination of stock variables in the aggregate balancesheet of a country and the balance-sheets of its main sectors. Four types of balance-sheet mismatches help to determine a country’s ability to service debt in the face of shocks: (i) maturity mismatches, where a gap between liabilities due in the short term and liquid assets leaves a sector unable to honor its contractual commitments if the market declines to roll over debt,21 or creates exposure to the risk that interest rates will rise; (ii) currency mismatches, where a change in the exchange rate leads to a capital loss; (iii) capital structure problems, where a heavy reliance on debt rather than equity ﬁnancing leaves a ﬁrm or bank less able to weather revenue shocks; and (iv)solvency problems, where assets are insuﬃcient to cover liabilities. They note that maturity mismatches, currency mismatches, and a poor capital structure can all contribute to solvency risk, but solvency risk can also arise from simply borrowing too much or from investing in low-yielding assets.
The currency composition of bank’s assets and liabilities is unfortunately not available in the Worldscope database, so currency mismatches cannot be measured. The speciﬁc measures used to address the other three risks men-
21The classic model of Diamond and Dybvig (1983) shows how this maturity mismatch makes banks susceptible to panic-based runs by depositors. The idea, of course, extends to other debtors of banks. Gorton (2009) illustrates how the recent ﬁnancial crisis started as a run in repurchase markets (short term, or overnight, lending between ﬁnancial ﬁrms).

CHAPTER 3. INTERCONNECTIONS AND CRISIS

61

tioned above are created as follows. First, the aggregate maturity mismatch is the ratio of the total short-term and current portion of long-term debt in the banking sector to the total value of liquid assets. The current portion of long-term debt is that which is due within a year, and liquid assets are composed of cash and investments in government securities. Exposure to risk is increasing in this measure. Second, capital structure problems are measured simply as the ratio of total capital to total assets. Exposure to risk is decreasing in this measure as capital is widely the main buﬀer between banks and failure, and capital adequacy is the key target of banking supervisors and regulators. Estrella et al (2000) ﬁnd that capital ratios are a good predictor of bank failure in the United States from 1998 to 1992 and that, somewhat surprisingly, a risk-weighted ratio does not consistently outperform the simpler ratios, particularly with short horizons. Third, solvency risk exposure is calculated as the net worth (assets plus capital minus liabilities) of the banking sector, scaled by assets.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

62

Table 3.10: Determinants of banking crises – balance-sheet risk exposures

Dependent variable: Banking Crisis Indicator

Logit

(1)

Interconnectedness (t-1)

4.597**

(1.962)

MaturityMismatch (t-1)

0.008

(0.008)

Capital/Assets (t-1)

NetWorth/Assets (t-1)
RealGDPgrowth (t-1) No. obs Pseudo-R2 % crises + % non-crises correct

-0.641*** (0.196) 570 0.60 155.9

(2) 4.643** (2.097)
-0.172 (0.108)
-0.358** (0.150)
534 0.55 162.3

(3) 4.679** (1.974)
-0.087 (0.106) -0.709*** (0.246)
571 0.60 156.1

The table presents panel regressions for 43 countries over the 1994–2009 period. The dependent variable is the banking crisis indicator, a discrete variable that equals one if there is a banking crisis in year t and zero otherwise. The independent variables are the rolling correlation of the largest 10 banks’ stock returns, from t-1 to t-4, the asset-weighted mean ratio of short-term and current portion of longterm debt to liquid assets at public banks, the asset-weighted mean capital to assets ratios of public banks, the asset-weighted mean ratio of the net worth to assets of public banks, real GDP growth, an indicator variable for the presence of an explicit deposit insurance scheme, the classiﬁcation of the exchange rate regime, which is increasing in ﬂexibility, the share of ﬁnancial sector foreign assets (insigniﬁcant), and the log of per capita GDP in US dollars (insigniﬁcant), all lagged by one year. Country ﬁxed eﬀects and year eﬀects are included. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

63

3.7 Conclusion

I study the eﬀect of interconnectedness among banks within a banking sector on banking system stability and ﬁnd that (i) the probability of a banking crisis is increasing in interconnectedness, (ii) the probability of a greater number of large bank failures during crisis periods is increasing in interconnectedness, and (iii) the probability of one or more large bank failures is higher when banks are more interconnected, in all years. I interpret these results as indicative of contagion within banking sectors. When banks are more connected, the failure of one bank has adverse aﬀect on other banks, resulting in subsequent bank failures. Interestingly, I ﬁnd no signiﬁcant relationship between the interconnectedness of smaller banks and the probability of small bank failures. This could be due to the fact that a larger share of the deposits of small banks are insured, making them less susceptible to failure through runs on deposits.
This paper adds to the growing literature that suggests that banking regulators could gain from incorporating stock price information into their assessments of bank risk. The focus of the paper is on linkages within banking sectors and the potential for contagion within a country. Future research on the transmission mechanisms through which ﬁnancial crises spillover to other countries would be valuable.

CHAPTER 3. INTERCONNECTIONS AND CRISIS
3.8 Appendix

64

Figure 3.3: Interconnectedness of largest 5 ﬁrms by sector – United States

Correlation of firm returns by sector

United States

Agriculture,Forestry,Fishing

Construction

Manufacturing

-.5 0 .5 1

Largest 5 firms

Mining

Services

Trade

112222211112222299000009999000009910000999900010361863186386311810010101011010012662626262262662 -.5 0 .5 1
111222229990000099900010386316181101000122626662

CHAPTER 3. INTERCONNECTIONS AND CRISIS

65

Table 3.11: Banking crises 1993-2009
Emerging and developing economies

Country Argentina Brazil Colombia Ecuador Malaysia Philippines Thailand Turkey

Crisis years 2001-2003 1994-1998 1998-2000 1998-2002 1997-1999 1997-2001 1997-2000 2000-2001

Non-crisis countries

Bangladesh

Pakistan

Chile

Peru

China

South Africa

Egypt

SriLanka

Kenya

Venezuela

Lebanon

Mexico

Morocco

Advanced Economies

Country Austria Belgium Denmark France Germany Greece Ireland Japan Korea (South) Netherlands Spain Sweden Switzerland United Kingdom United States

Crisis years 20082008200820082008200820081997-2001 1997-1998 200820082008200820072007-

Non-crisis countries Australia Canada Finland Hong Kong Israel Italy Norway Poland Singapore

CHAPTER 3. INTERCONNECTIONS AND CRISIS

66

Table 3.12: Bank failures by country and size – median classiﬁcation

Country
Argentina Australia Austria Bangladesh Belgium Brazil Canada Chile China Colombia Denmark Ecuador Egypt Finland France Germany Greece HongKong India Ireland Israel Italy Japan Kenya Korea Malaysia Mexico Morocco Netherlands Norway Pakistan Peru Philippines Poland Singapore SouthAfrica Spain SriLanka Sweden Switzerland Thailand Turkey UK USA Venezuela

Total failures
27 10 9 0 15 21 9 0 0 16 3 3 0 0 38 31 0 10 0 9 3 14 14 4 12 3 3 0 10 1 1 5 1 2 7 13 4 0 1 33 3 2 37 288 1

Small banks Number
27 1 7 0 15 12 9 0 0 14 0 2 0 0 32 14 0 10 0 3 1 14 5 4 0 2 2 0 10 0 1 4 1 2 0 11 4 0 1 28 0 2 20 175 0

Mean size (millions US$)
390 280 195
313 53 102
167
64
183 129
612
2,093 65 136
1,434 37
206 241
925
28 90 331 227
309 504
3 138
1,052 320 85

Large banks Number
0 3 2 0 0 9 0 0 0 2 3 1 0 0 6 17 0 0 0 1 0 0 9 0 12 1 1 0 0 1 0 1 0 0 0 2 0 0 0 5 3 0 17 126 1

Mean size (millions US$)
5,403 30,489
1,042
2,734 517 863
2,824 1,054
11,608
22,253 2,954 30,044 3,466
3,270 279
2,429
15,566 1,832 9,805 29,954 1,128

Large banks are those with assets greater than the median bank in each countryyear.

CHAPTER 3. INTERCONNECTIONS AND CRISIS

67

Table 3.13: Bank failures by country and size – 10B US$ classiﬁcation

Country
Argentina Australia Austria Bangladesh Belgium Brazil Canada Chile China Colombia Denmark Ecuador Egypt Finland France Germany Greece HongKong India Ireland Israel Italy Japan Kenya Korea Malaysia Mexico Morocco Netherlands Norway Pakistan Peru Philippines Poland Singapore SouthAfrica Spain SriLanka Sweden Switzerland Thailand Turkey UK USA Venezuela

Total failures
27 10
9 0 15 21 9 0 0 16 3 3 0 0 38 31 0 10 0 9 3 14 14 4 12 3 3 0 10 1 1 5 1 2 7 13 4 0 1 33 3 2 37 288 1

Small banks Number Mean size
(millions US$) 27 390
9 1,369 7 195 0 15 313 21 477 9 102 0 0 16 488 3 517 3 330 0 0 37 308 31 636 0 10 612 0 8 1,592 3 1,023 14 136 8 2,896 4 37 12 2,954 2 206 3 1,316 0 10 925 1 3,270 1 28 5 128 1 331 2 227 7 529 13 635 4 504 0 13 31 207 3 1,832 2 1,052 32 1,055 280 502 1 1,128

Large banks No. Mean size
0 1 13,787 2 30,489 0 0 0 0 0 0 0 0 0 0 0 1 11,410 0 0 0 0 1 11,608 0 0 6 30,714 0 0 1 30,044 0 0 0 0 0 0 0 0 0 0 0 0 0 2 37,641 0 0 5 27,868 8 53,108 0

Large banks are those with assets greater than 10 billion US$.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

68

Chapter 4
How Risky are Banks’ Risk-Weighted Assets? Evidence from the Financial Crisis
“The leverage ratio - a simple ratio of capital to balance-sheet assets - and the more complex risk-based requirements work well together. The leverage requirement provides a baseline level of capital to protect the safety net, while the risk-based requirement can capture additional risks that are not covered by the leverage framework. The more advanced and complex the models become, the greater the need for such a baseline. The leverage ratio ensures that a capital backstop remains even if model errors or other miscalculations

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

69

impair the reliability of risk-based capital. This is a crucial consideration - particularly as we work through the implementation of Basel II standard. By restraining balance-sheet growth, the leverage ratio promotes stability and resilience during diﬃcult economic periods.” - Remarks by Sheila Bair, Chairman, Federal Deposit Insurance Corporation before the Basel Committee on Banking Supervision, Merida, Mexico, October 4, 2006.

4.1 Introduction
The ﬁnancial crisis that begain in 2007 has exposed a number of weaknesses in banking regulation. Capital requirements are the primary tool of bank capital regulation, making how to appropriately determine the riskiness of bank’s portfolios a key challenge for banks, regulators, and investors. Risk-weighted assets (RWA) are an important element of risk-based capital ratios as banks can increase their capital adequacy ratios in two ways: (i) by increasing the amount of regulatory capital held, which boosts the numerator of the ratio, or (ii) by decreasing risk-weighted assets, which is the denominator of the regulatory ratio. The principle that regulatory capital requirements should be tied to the risks taken by banks was accepted internationally and formalized with the Basel I accord in 1988, and the deﬁnition of acceptable forms of capital and guidelines for measuring risk have undergone several revisions since that time. The second Basel accord, published in 2004, recommended

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

70

that banks hold total regulatory capital equal to at least 8 percent of their risk-weighted assets. The recently updated Basel III guidelines recommend increasing the amount of capital that should be held and emphasize higher quality forms of capital, but continue the approach of calculating capital requirements relative to their risk-weighted assets.1 Basel III also proposes a non-risk-weighted leverage ratio as a complementary measure.2
A key concern about current methods of calculating risk-weighted assets is that they leave room for individual banks to “optimize” capital requirements by underestimating their risks and thus being permitted to hold lower capital. Jones (2000) discusses techniques banks can use to engage in regulatory capital arbitrage and provides evidence on the magnitude of these activities in the Unites States. Merton (1995) provides an example in which, in place of a portfolio of mortgages, a bank can hold the economic equivalent of that portfolio at a risk-weight one-eighth as large. Innovations in ﬁnancial products since the ﬁrst Basel accord have also likely made it easier for ﬁnancial institutions to manipulate their regulatory risk measure. Acharya, Schnabl, and Suarez (forthcoming) analyze asset-backed commercial paper and ﬁnd evidence suggesting that banks used this form of securitization to concentrate, rather than disperse, ﬁnancial risks in the banking sector while reducing bank capital requirements. In addition to concerns about underestimating
1Basel III capital standards require minimum common equity, Tier 1 capital, and total capital equal to 4.5, 6, and 8 percent of risk-weighted assets, respectively. In addition, the standards recommend an additional capital conservation buﬀer of 2.5 percent of riskweighted assets. Supervisors can add a countercyclical buﬀer in the range of 0-2.5 percent.
2The speciﬁc deﬁnition and enforcement of the leverage ratio under Basel III is pending.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

71

the riskiness of assets, there are diﬀerences in calculation of risk-weighted assets across countries that may have unintended eﬀects on ﬁnancial stability. Lord Adair Turner, chairman of the UK Financial Services Authority, warned last June that international diﬀerences in the calculation of risk-weighted assets could undermine Basel III3 and Sheila Bair, former chairman of the US Federal Deposit Insurance Corporation, added her concern that Europe and the US may be diverging in their calculation of their risk-weighted assets: “The risk weightings are highly variable in Europe and have led to continuing declines in capital levels, even in the recession. There’s pretty strong evidence that the RWA calculation isn’t working as it’s supposed to.”4
In this paper, we ask whether equity investors ﬁnd banks’ reported riskweighted assets to be a credible measure of risk. First, did banks with lower risk-weighted assets have higher stock returns during the recent ﬁnancial crisis? Second, do measures of risk based on equity market information correspond to risk-weighted assets? Demirguc¸-Kunt, Detragiache, and Merrouche (2010) and Beltratti and Stulz (forthcoming) also study banks’ stock return performance during the ﬁnancial crisis, focusing primarily on the eﬀect of diﬀerent measures of capital and bank governance, respectively. Our paper studies whether markets price bank risk as measured by RWA, to inform the debate on how best to measure the risks embedded in banks’ portfolios.
3Risk Magazine, June 24, 2011, “FSA’s Turner: RWA divergence would undermine Basel III” www.risk.net/risk-magazine/news/2081533/fsas-turner-rwa-divergenceundermine-basel-iii
4Risk Magazine, June 24, 2011, “Europe lax on RWA calculations, says Bair” www.risk.net/risk-magazine/news/2081139/europe-lax-rwa-calculations-bair

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS
Figure 4.1: Decrease in risk-weighted assets at US BHCs
Risk Weighted Assets - US BHCs

72

% 60 65 70 75 80

1990

1995

2000 year

2005

RWA/Tangible Assets
US BHCs, Data source: SNL Financial

2010

The ratio of aggregate risk-weighted assets to tangible assets at all bank holding companies (BHC) in the United States. The tangible assets of a bank are equal to total assets minus intangibles and goodwill.

4.2 Stylized facts

The past decade has seen a decrease in the measured risk-weighted assets of banks, as well as a decrease in the quality of capital held by banks. For instance, there is a downward trend in the aggregate ratio of risk-weighted assets to tangible assets for US bank holding companies since the late 1990s, shown in Figure 4.1.
Figure 4.2 shows the downward trend in three measures of capital as a percentage of risk-weighted assets at bank holding companies in the United States in the years preceding the crisis. This decline is sharpest in tangible common equity, which has the greatest loss-absorbing capacity, and less

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS
Figure 4.2: Quality of regulatory capital at US BHCs
Capital Ratios - US BHCs

73

% 10 15

5

1990

1995

2000 year

Total Capital/RWA TCE/RWA
US BHCs, Data source: SNL Financial

2005

2010

Tier1 Capital/RWA

Tangible common equity (TCE) is equal to common equity minus intangibles and goodwill. Tier 1 capital includes shareholders equity, both common equity and non-cumulative preferred stock, and retained earnings. Tier 2 capital includes subordinated debt, hybrid instruments - such as perpetual preferred stock, revaluation reserves, general provisions, and undisclosed reserved are also permitted by regulators in some countries. Total capital is the sum of Tier 1 capital and Tier 2 capital.

pronounced in total regulatory capital. There are also diﬀerences in RWA between geographical regions. On
average, European banks have much lower RWA as a share of their total assets than banks in North America or Asia (see Figure 4.3).

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

74

Figure 4.3: Ratio of RWA to total assets in Asia, Europe and North America (2002-2010)

Asia

Europe

North America

70

58%

60

50 51%
40

30 36%

20

10

0 2002 2003 2004 2005 2006 2007 2008 2009 2010

Source: Bloomberg, SNL Financial. Aggregate ratios of RWA to total assets
for banks in Asia, Europe, and North America. Data are for a sample of the largest banks in each region.

4.3 Risk-weighted assets: Basel II Standardized and IRB Approaches

The 1998 Basel I Accord proposed a simple framework for measuring risk based on risk buckets’ for four broad categories of claims: sovereigns, banks, mortgages, and corporate. In contrast, Basel II aimed at improving the risk sensitivity of capital requirements. A prelude to the Basel II internal ratingsbased (IRB) approach was the adoption by the Basel Committee in 1996 of the value-at-risk (VaR) approach to determine capital requirements for market risk (subsequently integrated into the Basel II framework). The 2004 Basel II accord extended a similar approach to credit risk. It opened the way

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

75

for banks to determine risk-weights using external credit rating agencies or their own internal ratings systems based on historical default data, after supervisory validation, and to calculate the parameters of a uniform regulatory formula.
For banks not (yet) deemed by supervisors to be able to implement modelbased approaches, Basel II contemplated simpliﬁed approaches for each of the risk categories it covered. In particular, the Standardized Approach (SA) for credit risk provided a much more diﬀerentiated treatment of exposures than Basel I, while allowing risk-weights for each exposure to vary according to ratings issued by external credit rating agencies. However, although external ratings could drive risk weights signiﬁcantly higher than Basel I’s highest weight of 100 percent, they could also drive them much lower. Moreover, in the important case of home mortgage loans, the risk weight was reduced from 50 to 35 percent.
The Basel II IRB approach is built on three risk parameters: (i) the probability of default (P D), which describes the likelihood that an obligor will default, (ii) the exposure at default (EAD), and (iii) the loss given default (LGD), which describes the loss rate on the exposure in the event of default.5 Indeed, the level of regulatory capital that a bank should hold depends on the amount of loss it is expected to exceed with a small, predeﬁned probability – the Value-at-Risk (VaR). Capital is set according to the
5The Basel II IRB formula is based on an asymptotic single-risk factor (ASRF) model. See Basel Committee on Banking Supervision (2005).

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

76

bank’s unexpected loss (U L) which is the gap between the bank’s expected loss (EL) and Value-at-Risk at a certain conﬁdence level, over a one-year horizon. The expected loss is calculated as follows: EL = P D × EAD × LGD or, in percentage of EAD, EL% = P D × LGD.
The supervisory capital charge as a percentage of the exposure (K) is set using a Merton model which depends on P D and LGD. Risk-weighted assets (RWA) are then expressed as a function of the capital requirement and the minimum capital ratio of 8 percent (or its reciprocal 12.5):
RW A = 12.5 × K × EAD.
It is important to note that under the Basel II IRB approaches, banks have considerable discretion in reporting their own average P D, and also EAD and LGD under the advanced IRB.
Basel III improves risk weights on exposures to market risk but leaves Basel II standardized approach (SA) risk weights on credit risk exposures unchanged. However, banks are now expected to use higher than externalratings-based risk weights if their own risk assessment so warrants.

4.4 Related literature
This study expands on two main strands of literature. First, this paper is related to studies of bank resilience during the recent ﬁnancial crisis. Demirgu¸cKunt, Detragiache, and Merrouche (2010) ﬁnd that capital was positively

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

77

related to banks’ stock returns during the crisis6 and Beltratti and Stulz (forthcoming) ﬁnd that large banks with more capital7 and higher reliance
on deposits for short-term funding in 2006 had higher stock returns dur-
ing the crisis, but these factors did not have a robust impact on bank risk,
as measured by the bank’s idiosyncratic volatility and distance-to-default.
In earlier works, Kim and Santomero (1988) use a mean-variance model to
evaluate the eﬀectiveness of both a uniform capital ratio requirement and a
risk-related capital plan in controlling bank risk and derive theoretically cor-
rect risk-weights to maintain banking system soundness, and Berger, Herring
and Szego (1995) discuss the diﬃculty of developing an accurate measure of
risk exposure that is reasonably simple and can be uniformly applied across
banks.
Second, we build on the empirical literature that studies bank risk-taking
and regulatory risk in the United States. Avery and Berger (1991b) and
Bradley et al (1991) ﬁnd that RWA for banks and for thrifts, respectively, are
positively related to the bank or thrift’s probability of failure and accounting
measures of risk, but that these relationships are fairly weak. Cordell and
King (1995) compare stock market measures of risk to regulatory risk-based
capital measures for banks and thrifts in the United States in 1990. Their
6Speciﬁcally, they ﬁnd that during the crisis, a stronger capital position was associated with better stock market performance, most markedly for larger banks, and the relationship between stock returns and capital is stronger when capital is measured by the leverage ratio (capital to total assets) rather than the risk-adjusted capital ratio.
7They test both the ratio of Tier1 capital to RWA and tangible equity to total assets. The regulatory ratio is found to be statistically signiﬁcant in most regression speciﬁcations.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

78

results suggest that market and risk-based capital standards agree to some extent on the adequacy of institutions’ current capital, but the measures of asset risk are not positively correlated.
Most importantly, we shed light on the credibility of RWA as currently measured. The debate on the appropriate capital requirements for banks has reached consensus on higher capital requirements, but the discussion of appropriate measurement of asset-risk has come to the forefront only recently. Acharya (2011) notes that the importance of residential housing as an asset class increased endogenously in response to the low risk weights on residential mortgage backed securities in capital requirements and, although the subprime crisis showed that banks were clearly not holding enough capital against these assets, the relatively low risk weight on this asset class has continued. The signiﬁcance of the problem is also clear in the current Eurozone sovereign debt crisis. The zero risk weights on European banks’ holdings of the debt of their sovereigns are clearly not in line with current assessments of their riskiness.
In studying the eﬀects of government bailouts, Duchin and Ross (2012) ﬁnd evidence that bailed-out banks approve riskier loans and shift investment portfolios toward riskier securities. This shift in risk occurs mostly within the same asset class and therefore has little eﬀect on the closely monitored capitalization levels. Consequently, bailed-out banks appear safer according to the capitalization requirements, but show a signiﬁcant increase in marketbased measures of risk. They conclude that those banks’ responses to capital

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

79

requirements may erode their eﬃcacy in risk regulation. To what extent are diﬀerences between regulatory and market assess-
ments of risk problematic? Market discipline requires investors to both (i) monitor the condition of banks and incorporate those assessments into their stock prices, and (ii) inﬂuence the managers of banks through these changes in stock prices (Bliss and Flannery 2001). This study does not delve into whether market participants are able to inﬂuence banks managers but instead focuses on the ﬁrst, necessary, component of discipline. That is, how does the market assess the riskiness of banks based on their regulatory reports, balance-sheet, and income statements? Even if markets are not able to inﬂuence manager’s actions,8 an understanding of their assessments is important in that bank regulators and supervisors can incorporate this information into their assessments and actions.

4.5 Data and Methodology
The sample consists of 804 publicly-listed deposit-taking institutions9 in 32
countries, spanning North America, Europe, Asia, and Australia. These are
listed in Table 4.12 of the appendix. The balance-sheet and income statement
8There is some support for market inﬂuence in banking. Flannery and Rangan (2002) document a build-up of regulatory and market equity capital in large U.S. bank holding companies from 1986 and 2000, beyond levels necessary to meet regulatory standards, and attribute this increase in capital to enhanced market incentives to monitor and price large banks’ default risk. Barrios and Blanco (2003) argue that Spanish banks’ capital ratios over the period 1985–1991 were primarily driven by the pressure of market forces rather than regulatory constraints.
9Institutions with a deposit/asset ratio above 20 percent and a loan/asset ratio above 10 percent, as in Beltratti and Stulz (forthcoming).

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

80

data are from the Bankscope database, which has good coverage from 2004 onwards, and the stock return data is from Datastream. We discard obvious mistakes in the data as well as outliers at the 1 percent and 99 percent levels.
Banks performed poorly over the crisis, with an average stock return of −42 percent from June 30, 2007 to December 30, 2008, as shown in Table 4.1. The recent ﬁnancial instability originating from concerns about the sovereign debt of Eurozone countries has also been accompanied by poor performance of banks. Their average stock return was −13 percent over the three month period from June 30, 2011 to September 30, 2011. In both cases, the standard deviations of the returns are quite large, showing a large variation in bank performance during periods of instability.

Table 4.1: Descriptive statistics – bank stock returns over periods of crisis

Mean Std. Dev. Min Max

June 2007 to Dec 2008 (769 obs)
-41.67 31.09 -99.97 62.59

June 2011 to Sep 2011 (804 obs)
-13.15 19.59 -75.28 192.36

Table 1 shows summary statistics for the stock returns of the banks in the sample for two periods: (1) June 30, 2007 to December 30, 2008 corresponding to the subprime crisis, and (2) June 30, 2011 to September 30, 2011 corresponding to the Eurozone sovereign debt crisis.
Our ﬁrst hypothesis is that banks with higher risk-weighted assets will
perform worse during a period of crisis. This would be an indication that

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

81

markets do give credence to the regulatory measure of asset-risk. In other words, investors expect banks with higher RWA to have larger LGDs during the ﬁnancial crisis. We separate banks’ regulatory capital ratios into simple leverage ratios and RWA scaled by tangible assets, to determine the eﬀects of RWA on stock returns while controlling for the capital of the bank.10
H1 Banks with lower RWA performed better during the crisis.

Second, we expect a positive relationship between capital and stock returns, since capital functions as a buﬀer against adverse shocks by providing loss absorbency beyond provisions and other expected-loss buﬀers. A higher share of customer deposits in funding decreases the susceptibility of banks to runs, so we expect banks with more stable funding to perform better during periods of crisis. There may also be a trade-oﬀ between these two factors, in that banks with higher capital are better able to withstand funding problems.

H2 Banks with higher capital ratios will had higher stock returns during the crisis and banks with more stable funding had higher stock returns during the crisis.

H2a Banks with more stable funding did not receive as high a reward for higher capital, compared to banks with less stable funding.

Third, if RWA are a good measure of asset-risk, we expect they will be

positively related to market-based measures of risk. The relationship between

10For example:

T ier1Capital RW A

=

T ier1Capital T angibleAssets

×

T

angibleAssets RW A

.

Previous studies include

either the regulatory capital ratio or the simple leverage ratio, which does not allow for a

direct test of the eﬀect of RWA on performance.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

82

the two, if found, could change in either direction after the crisis. The onset of the crisis could render RWA less credible to investors, or the increase in risk-aversion associated with crisis could result in increased sensitivity to any available information on asset-risk.
H3 Market measures of bank risk will be positively correlated with RWA.
We estimate two models to test these hypotheses. In the ﬁrst, we focus on the eﬀects of risk-weighted assets and capital adequacy on the cross-section of banks’ stock returns, while controlling for other balance-sheet measures of risk exposure. We perform OLS estimation of equation 4.1:

ri = θ0 + θ1(RW A/T A)i,t−1 + θ2(Capital/T A) + Xi,t−1γ1 + νt (4.1)
where the dependent variable ri is the real stock return in US dollars from June 30, 2007 to December 30, 2008 for bank i. This is the period over which there was a sharp decline in aggregate stock market indexes.11 We
also study stock returns over the three-month period from June 30, 2011 to
Sep 30, 2011 to compare market reactions in the recent European sovereign
debt crisis to the ﬁrst phase of the global ﬁnancial crisis. In both cases, all
of the explanatory variables are lagged by one year.
11The U.S. Department of the Treasury established the Troubled Assets Relief Program, in which they infused capital into qualifying ﬁnancial institutions, in October 2008. This may have had an eﬀect on bank’s stock returns unrelated to underlying soundness of banks. For this reason we estimate equation (1) using the period preceding TARP, from June 20, 2007 to September 30, 2008, as a robustness check and ﬁnd the results to be quantitatively similar.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

83

The main explanatory variables of interest are the ratio of risk-weighted
assets to tangible assets and the capital ratio.12 We use tangible common
equity (total equity minus preferred shares, intangibles, and goodwill) as our
measure of bank capital in most speciﬁcations as one can expect holders of
non-TCE capital instruments to have had weaker incentives to monitor bank
risk-taking than common equity shareholders in the run up to the recent
ﬁnancial crisis.
The other explanatory variables, in Xi,t−1, are:
i Stability of funding, measured as the share of customer deposits in total deposits and short-term funding13
ii The ratio of securities to assets
iii The share of non-performing loans
iv Return on average assets, a measure of the bank proﬁtability
v The log of assets, to proxy for bank size
vi Country dummies, to control for diﬀerences in the institutional and regulatory environments across countries
12Three diﬀerent measures of capital are (i) tangible common equity, (ii) Tier 1 regulatory capital, and (iii) Total regulatory capital. Tier 1 capital consists of shareholder’s funds, perpetual non-cumulative preference share, and retained earnings. Tier 2 capital includes hybrid capital, subordinated debt, loan loss reserves, and valuation reserves. Total regulatory capital is equal to the sum of Tier 1 capital and Tier 2 capital. After comparing three diﬀerent measures of capital in our ﬁrst speciﬁcation, we proceed to use tangible common equity (total equity minus preferred shares, intangibles, and goodwill from total equity, where the data is available) as our measure of bank capital in the rest of the paper.
13This is equal to 100 − x where x is the funding fragility measure of Demirgu¸c-Kunt and Huizinga (2009): deposits from other banks, other deposits, and short-term borrowing as a fraction of total deposits plus money market funding.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

84

vii Dummy variables to control for the business model of the bank14

Table 4.2 reports descriptive statistics for the explanatory variables in 2006 and 2010.

Table 4.2: Descriptive statistics – explanatory variables

RWA/TangibleAssets TCE/TangibleAssets Tier 1 Capital/TangibleAssets Total Capital/TangibleAssets CustDeposits Securities/Assets NPL/Loans ROAA Assets - millions USD Beta

2006

(769 obs)

Mean

Std. Dev.

41.89

31.67

7.92 5.41

9.01 5.01

6.31 6.60

90.44

12.05

20.26

11.86

1.66 2.72

0.88 0.85

55,900

228,000

0.24 1.06

2010

(804 obs)

Mean Std. Dev.

66.38

14.87

7.04 3.93

8.69 3.49

9.93 3.57

91.15

11.31

22.20

11.88

4.41 4.78

0.37 1.17

83,764 316,632

1.00 1.33

The variables are the ratio of risk-weighted assets to tangible assets, capital ratios (tangible common equity (TCE)), Tier1 Capital, and Total regulatory capital, all divided by tangible assets (TA)), the share of stable (customer) deposits, the share of securities in the bank’s assets, the share of non-performing loans (NPL), the return on assets, assets, and the bank’s stock’s beta with a national market index.
The share of customer deposits in total deposits is expected to increase a
bank’s stability since these deposits are less likely to be withdrawn in a bank
run, due to deposit insurance.15 Gorton (2010) and others have described
14Bank holding companies, commercial banks, cooperative banks, investment banks, real estate and mortgage banks, and savings banks
15Our measure is equal to 100 – x where x is the funding fragility measure of Demirgu¸c-

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

85

the recent crisis as a run in non-retail funding markets. We expect a positive relationship between the stability of funding and banks’ stock returns.
A second speciﬁcation of this model includes an interaction term between the capital ratio and funding stability. This is to test the notion that there is a trade-oﬀ in the capital adequacy and funding required to satisfy markets of a bank’s health. A negative coeﬃcient on the interaction term between funding stability and the capital ratio suggests that banks with more stable funding are not required to hold as much capital in order to receive the same stock return, compared to a banks with fragile funding.
The second model is a panel estimation of how the explanatory variables described above are related to banks’ systematic risk – beta from a Capital Asset Pricing Model (CAPM). We perform a panel estimation of equation 4.2 over the period from 2004 to 2010:

Riski = µj + δ1(RW A/T A)i,t−1 + δ2(Capital/T A) + Xi,t−1γ1 + νi,t (4.2)

The explanatory variables are as in equation 4.1 and we alternately include country ﬁxed eﬀects, to control for time-invariant country-speciﬁc factors that may aﬀect bank systematic risk, and bank ﬁxed eﬀects, to study the dynamic relationship between bank systematic risk and RWA. We also include year dummies to control for macroeconomic shocks that may aﬀect bank stock returns. The dependent variable, Riski,t, is the coeﬃcient β from
Kunt and Huizinga (2009).

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS the following model:

86

rit − rft = αi + β(rMt − rft) + βW (rW t − rft) + it

(4.3)

estimated over the previous 60 month period. Nominal stock prices in US dollars are deﬂated by the CPI to obtain real returns, rf,t is the 3-month US Treasury bill rate adjusted to a one month risk-free rate, rM,t is the MSCI national market index for the country M in which bank i is headquartered, and rW,t is the MSCI World index. That is, for each bank-year observation in model 4.3, β is estimated using monthly, US-dollar, real excess returns over the last 5 years.

4.6 Results and Discussion
4.6.1 Determinants of stock returns
Market perceptions of risk-weighted assets Table 4.3 presents the results of our benchmark estimation of equation 4.1,
the determinants of stock returns over 2007-2008 crisis period. As expected, we ﬁnd that stock returns are lower for banks with higher risk-weighted assets. The estimated coeﬃcient in column (1) indicated that banks with a one percentage point higher RWA to tangible assets ratio have a stock return that is 0.075 percent lower. The estimated coeﬃcients on the three diﬀerent capital ratios are not statistically signiﬁcant but their magnitudes are consistent

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

87

with the ﬁnding of Demirgu¸c-Kunt, Detragiache, and Merrouche (2010) that certain types of capital matter more in explaining stock returns. The signs of the coeﬃcients on the additional explanatory variables are as expected. Higher stock returns are associated with more stable funding, more securities holdings, a lower share of non-performing loans, and a higher accounting return on assets. However, only the share of securities to assets, and the accounting return on assets have a strong statistical relationship with the stock returns.
Table 4.4 shows the same estimation on the sample restricted to larger banks, with assets greater than 50 billion US dollars in 2006. The coeﬃcient on RWA is also negative and statistically signiﬁcant in this sample. Large banks with a 1 percent point higher RWA to tangible assets ratio have a 0.39 percent lower stock return, on average. Large banks are diﬀerent than the rest of the sample in that access to stable short term funding seems to be more important for them than securities holdings. The coeﬃcient on funding stability is now statistically signiﬁcant while the coeﬃcient on the ratio of securities in assets is not. A one percentage point increase in the share of stable funding at a large bank is associated with a stock return that is 0.63 percent higher, on average. As with the whole sample, higher net income as indicated by a higher ROAA, is rewarded with a higher stock return.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

88

Table 4.3: Determinants of returns – Do risk-weighted assets aﬀect stock

returns?

Dependent variable: Crisis stock return

(1)

RWA/TangibleAssets

-0.075***

(0.027)

TCE/TangibleAssets

0.066

(0.195)

Tier1capital/TangibleAssets

TotalCapital/TangibleAssets

CustDeposits Securities/Assets NPL/Loans ROAA log(Assets) Beta Observations Adj R-squared

0.122 (0.087) 0.759*** (0.116) -0.338 (0.216) 6.052*** (0.707) -0.032 (1.277) -0.069 (0.805)
769 0.210

(2) -0.054 (0.033)
-0.307 (0.209)
0.101 (0.093) 0.783*** (0.106) -0.359 (0.226) 5.542*** (0.678) -0.510 (1.306) 0.018 (0.796)
762 0.209

(3) -0.065** (0.027)
-0.065 (0.184) 0.126 (0.089) 0.763*** (0.120) -0.349 (0.226) 5.815*** (0.678) -0.120 (1.139) -0.031 (0.786)
769 0.210

The table presents regressions for banks in 32 countries. The dependent variable is the bank’s stock return over the period from June 30, 2007 to December 30, 2008. The independent variables, all values for 2006, are the ratio of risk-weighted assets to tangible assets, capital ratios (tangible common equity (TCE)), Tier1 Capital, and Total regulatory capital, all divided by tangible assets), the share of stable deposits, the share of securities in the bank’s assets, the share of non-performing loans, the return on assets, the log of assets, and the stock’s beta with a national market index. Country dummy variables and dummy variables representing the bank’s business model are included. Standard errors (in parentheses below the coeﬃcient estimates) are adjusted for heteroskedasticity following Huber (1967) and White (1980), and generalized for clustering at the country level. ***, **, and * indicate signiﬁcance at the 1, 5, and 10 percent conﬁdence levels, respectively.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

89

Table 4.4: Determinants of returns – Do risk-weighted assets aﬀect stock

returns of large banks?

Large banks: Assets>50 Billion US$

Dependent variable: Crisis stock return

(1)

RWA/TangibleAssets

-0.388***

(0.108)

TCE/RWAfloor

-0.559

(1.430)

Tier1capital/RWAfloor

TotalCapital/RWAfloor

CustomerDeposits Securities/Assets NPL/Loans ROAA log(Assets) Beta Observations Adj R-squared

0.625** (0.272) 0.100 (0.303) 0.692 (1.516) 21.200*** (6.519) -0.627 (2.760) 0.175 (3.451)
90 0.595

(2) -0.448***
(0.127)
1.666 (2.463)
0.532** (0.217) 0.090 (0.304) 1.161 (1.654) 18.980*** (6.567) -0.189 (2.879) -0.221 (3.472)
90 0.597

(3) -0.678* (0.353)
2.264 (2.670) 0.578** (0.223) 0.019 (0.312) 0.936 (1.466) 18.323*** (6.539) -0.222 (2.851) -0.441 (3.415)
90 0.598

The table presents regressions for large banks in 32 countries – those with total assets greater than 50 billion US dollars in 2006. The dependent variable is the bank’s stock return over the period from June 30, 2007 to December 30, 2008. The independent variables are as in Table 4.3.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

90

Is there a capital-funding trade-oﬀ ? We ﬁnd a trade-oﬀ between capital and funding in terms of their positive eﬀects on bank stock returns. Table 4.5 presents the results of the estimation of model 4.1 in which an interaction term between capital and funding stability is included. The negative coeﬃcient on the interaction term in column (2) shows that the more stable a bank’s funding, the less positive the eﬀect of higher capital on its stock return. Column (3) indicates that this trade-oﬀ exists for large banks as well. Diﬀerences by Basel credit risk measurement method At the time of the subprime crisis, countries had made diﬀerent degrees of progress towards becoming following Basel II guidelines. The EU Capital Requirements Directive required that countries in the European Union implement the Basel II guidelines by the time of the crisis, generally initially following the standardized approach to measuring credit risk, while many countries were still following Basel I guidelines. Table 4.6 shows the dates of implementation of the Basel II approaches (standardized and advanced approaches) in diﬀerent countries. We investigate whether the relationship between stock returns and RWA is the same for banks in Basel I and Basel II countries by adding a countrylevel indicator variable for counties that had moved to the Basel II Standardized Approach by the time of the crisis, as well as interaction between the indicator and RWA/TA, to equation (1) 4.1. We also include dummy variables for each region, North America, Europe, and Asia, in these spec-

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

91

iﬁcations and, in order to ensure that the results are not being driven by diﬀerences in accounting methods, we also include indicator variables to control for the accounting regime being followed by each bank.16
The results are presented in Table 4.7. Column (2) shows that for the whole sample, there is no signiﬁcantly diﬀerent eﬀects of RWA/TA on returns for banks in countries that were Basel II compliant. For large banks, however, the negative relationship between RWA and returns is signiﬁcantly smaller when banks are in Basel II countries.
The results presented in this section are robust to using the stock return from June 30, 2007 to September 30, 2008, the phase of the ﬁnancial crisis before the beginning of government capital purchase programs, as the dependent variable.

16Le Lesl´e and Avramova (2012) note that a key diﬀerence between the International Financial Reporting Standards (IFRS) and the US Generally Accepted Accounting Principles (GAAP) is that the relevant oﬀ-balance-sheet assets can be calculated net of derivatives in the GAAP, suggesting RWA would be lower under GAAP than under IFRS, all else equal. They ﬁnd, however, that RWA do not appear to be diﬀerent across diﬀerent accounting methods in a sample of ﬁfty internationally active banks in 25 countries.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

92

Table 4.5: Determinants of returns – Is there a capital-funding trade-oﬀ?

Dependent variable: Crisis stock return
RWA/TangibleAssets TCE/TangibleAssets CustDeposits (TCE/TangibleAssets)*CustDeposits Securities/Assets NPL/Loans ROAA log(Assets) Beta Observations Adj R-squared

(1) All banks -0.075*** (0.027)
0.066 (0.195) 0.122 (0.087)
0.759*** (0.116) -0.338 (0.216) 6.052*** (0.707) -0.032 (1.277) -0.069 (0.805)
769 0.210

(2) All banks -0.088** (0.034) 3.710* (2.124) 0.395** (0.170) -0.039* (0.021) 0.739*** (0.120)
-0.322 (0.201) 4.947*** (0.671) 0.366 (1.262) 0.049 (0.692)
769 0.213

(3) Large banks -0.339***
(0.116) 4.636* (2.497) 0.878*** (0.288) -0.072** (0.035) 0.165 (0.307) 2.421 (1.896) 19.127*** (5.890) -1.301 (2.623) 0.851 (3.411)
90 0.609

The table presents regressions for banks in 32 countries. The dependent variable is the bank’s stock return over the period from June 30, 2007 to December 30, 2008. Columns (1) and (2) present the whole sample, and column (3) present results for the sample of large banks – banks with assets greater than 50 billion US dollars in 2006. The independent variables are as in Table table:rwareg1 and additionally include an interaction term between the capital ratio (TCE/tangible assets) and stable deposits.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

93

Table 4.6: Basel II implementation schedules – credit risk measurement

Standardized approach Advanced approaches

Australia

Jan 2008

Jan 2008

Austria

Jan 2007

Jan 2008

Belgium

Jan 2007

Jan 2008

Canada

Nov 2007

Nov 2007

China

NA 2011-2013

Denmark

Jan 2007

Jan 2008

Finland

Jan 2007

Jan 2008

France

Jan 2007

Jan 2008

Germany

Jan 2007

Jan 2008

Greece

Jan 2007

Jan 2008

Hong Kong

Jan 2007

Jan 2007

India

April 2007

post Dec 2011

Indonesia

Jan 2009

Oct 2010

Italy

Jan 2007

Jan 2008

Japan

March 2007

March 2008

Korea

Dec 2007

Dec 2007

Malaysia

Jan 2008

Jan 2010

Norway

Jan 2007

Jan 2008

Philippines

Jul 2007

post 2010

Poland

Jan 2007

Jan 2008

Russian Federation

Jul 2012

NA

Singapore

March 2007

Jan 2008

Spain

Jan 2007

Jan 2008

Sri Lanka

Jan 2008

post 2010

Sweden

Jan 2007

Jan 2008

Switzerland

Jan 2007

Jan 2008

Taiwan

end 2006

NA

Thailand

Jan 2009

Jan 2009

Turkey

June 2011

NA

Ukraine NA NA

UK

Jan 2008

Jan 2008

USA NA mid-2009 for 'core banks'

NA=not announced

Sources: Supervisory agency websites and surveys,and IMF Financial

Stability Assessment Program Reports.

‘Core banks’ are those that have consolidated total assets ≥ $250 billion, consolidated on-balance-sheet foreign exposure ≥ $10 billion, or are subsidiaries of a core bank.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

94

Table 4.7: Determinants of returns – Basel I versus Basel II standardized

approach to measuring RWA

Dependent variable: Crisis stock return (1) (2) (3)

Basel II indicator RWA/TA

All banks -0.077**

All banks 2.946 (8.461)
-0.108***

Large banks -29.371** (12.939) -0.435***

Basel II * RWA/TA

(0.029)

(0.024) 0.067

(0.089) 0.430**

TCE/TA

-0.175 (0.347)

(0.075) -0.110 (0.278)

(0.198) -0.214 (1.274)

CustDeposits Securities/Assets

0.255* (0.147) 0.661*** (0.179)

0.257* (0.135) 0.634*** (0.197)

0.527*** (0.193) 0.187 (0.204)

NPL/Loans ROAA log(Assets)

-0.336 (0.361) 4.048** (1.852) -0.274 (1.183)

-0.213 (0.288) 4.268** (1.621) -0.204 (1.222)

-0.787 (1.110) 10.970* (5.889) -0.818 (2.239)

Beta Observations

-0.453 (1.018)
769

-0.371 (0.909)
769

-2.014 (2.644)
90

Adj R-squared

0.151

0.153

0.422

The table presents regressions for banks in 32 countries. The dependent variable is the bank’s stock return over the period from June 30, 2007 to December 30, 2008. Columns (1) and (2) present the whole sample, and column (3) present results for the sample of large banks – banks with assets greater than 50 billion US dollars in 2006. The independent variables are an indicator variable for countries that use the Basel II Standardized Approach to calculating RWA, and an interaction with the bank’s RWA to tangible assets, the capital ratios (tangible Common Equity (TCE) divided by tangible assets (TA)), the share of stable deposits, the share of securities in the bank’s assets, the share of non-performing loans, and the return on assets. The log of assets, the stock’s beta with a national market index, regional dummies, and dummy variables representing the bank’s business model are included in each speciﬁcation.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

95

4.6.2 Market risk and balance-sheet measures of risk
exposure
This section presents the results of estimating equation (3.2), to study the relationship between a bank systematic risk and RWA. Column (1) of Table 4.8 presents the estimation results for the period from 2004 to 2010, including country ﬁxed eﬀects. There is no signiﬁcant relationship between systematic risk and RWA. After including bank ﬁxed eﬀects in column (2), however, the estimated coeﬃcient on RWA/TA is positive and signiﬁcant. In the ﬁrst instance we controlled for time-invariant country-speciﬁc unobservables, while the speciﬁcation with bank ﬁxed eﬀects controls instead for any banklevel unobserved variables that do not vary over time. Thus, it appears there is no static relationship where banks with higher RWA have greater systematic risk, but instead, banks with higher RWA have greater systematic risk over time. The coeﬃcient of 0.005 in column (2) suggests that a one standard deviation increase corresponds to a 0.12 standard deviation increase in systematic risk. Turning to the coeﬃcients on the control variables, we see that the factors that are positively related to stock return performance are negatively related to systematic risk, as expected.
We next split the sample into two periods, the three years prior to the start of the crisis from 2004 to 2006 and three after the onset from 2008 to 2010, and estimate model (3.2) on both samples to investigate whether there is a change in the factors that aﬀect systematic risk. Shown in Table

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

96

4.9, the Chow test rejects the hypothesis that the relationship is the same before and since the crisis for several explanatory variables: the RWA ratio, the share of securities in assets, non-performing loans, and return on assets. The relationship between RWA/TA and systematic risk is positive before the crisis, but negative since the crisis.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

Table 4.8: Market risk and balance-sheet measures of risk exposure

Panel 2004-2010

Dependent variable: Systematic Risk (\beta)

(1) (2) (3) (4)

All banks

All banks

Large banks

Large banks

RWA/TangibleAssets (t-1)

0.001

0.005***

0.001

0.003

(0.017)

(0.119)

(0.031)

(0.077)

TCE/TangibleAssets (t-1)

0.009

0.023*

-0.052**

-0.109**

(0.037)

(0.096)

(-0.170)

(-0.360)

CustDeposits (t-1)

-0.001

-0.003

-0.004

0.005

(-0.012)

(-0.028)

(-0.091)

(0.109)

Securities/Assets (t-1)

-0.009***

-0.000

0.001

0.014*

(-0.088)

(-0.005)

(0.010)

(0.260)

NPL/Loans (t-1)

0.026***

0.032***

0.041**

0.106***

(0.063)

(0.079)

(0.114)

(0.295)

ROAA (t-1)

-0.066***

-0.070***

-0.022

-0.030

(-0.059)

(-0.062)

(-0.023)

(-0.031)

log(Assets) (t-1)

0.274***

0.157

0.071

-0.653**

(0.502)

(0.287)

(0.113)

(-1.039)

Year dummies

Yes Yes Yes Yes

Country fixed effects

Yes No Yes No

Bank fixed effects

No Yes No Yes

Observations

4280

4280

563

563

Adj R-squared

0.217

0.184

0.176

0.149

97

The table presents panel regressions for banks in 32 countries from 2004 to 2010. The dependent variable is the stock’s beta with a national market index estimated from a CAPM model. The independent variables, all lagged by one year, are as in Table 4.3. Additionally, each speciﬁcation includes dummy variables for each year. Columns (1) and (3) include country ﬁxed eﬀects and dummy variables representing the banks’ business model, and columns (2) and (4) include bank-level ﬁxed eﬀects.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

Table 4.9: Market measures of risk and balance-sheet measures of risk exposure – have the relationships changed since the crisis?

Risk pre and post-crisis

Dependent variable: Systematic Risk (\beta)

(1)

2004-2006

RWA/TangibleAssets (t-1)

0.006***

(0.116)

TCE/TA (t-1)

0.016***

(0.068)

CustDeposits (t-1)

-0.004

(-0.126)

Securities/Assets (t-1)

-0.003***

(-0.029)

NPL/Loans (t-1)

0.031

(0.050)

ROAA (t-1)

0.015

(0.008)

log(Assets) (t-1)

0.148***

(0.836)

(2) 2008-2010 -0.002***
(-0.067) -0.001 (-0.003) 0.005 (0.193) -0.015*** (-0.150) -0.026** (-0.053) -0.107*** (-0.073) 0.212 (1.296)

p value for test: coeff (1) = coeff (2)
0.000
0.140
0.034
0.001
0.072
0.000
0.630

Constant
Observations Adj R-squared

-1.934*** (-0.721)

4280 0.328

-2.106 (-0.830)

0.927

98

Standardized coeﬃcients are reported in parentheses. The table presents panel regressions for banks in 32 countries from 2004-2006 in Column (1) and 2008-2010 in Column (2). The dependent variable is the stock’s beta with a national market index estimated from a CAPM model. The independent variables, all lagged by one year, are as in 4.3. Country ﬁxed eﬀects and dummy variables representing the bank’s business model are included.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

99

4.7 Performance during the Eurozone debt crisis

We check the robustness of the negative relationship between bank stock returns and RWA in this section, by estimating equation 4.1 for a recent period of ﬁnancial instability – the Eurozone sovereign debt crisis. We ﬁnd that the negative relationship between RWA and stock returns is robust to the diﬀerent time period.
Table 4.10 shows the results of model (3.1) estimated for the period from June 30, 2011 to September 30, 2011. The signs of the coeﬃcients are the same as for model 4.1 estimated over 2007-2008 crisis period, however only RWA/TA, return on assets, and bank size had a statistically measurable eﬀect on the stock returns in this period.
Next, we diﬀerentiate between the method used to calculate credit risk, similar to Table 4.7 presented above for the subprime crisis. By 2010, banks in several countries had moved towards using one of the Basel II advanced approaches to measuring credit risk (FIRB or AIRB), while many were using the Basel II standardized approach (SA), and some were still following Basel I guidelines. The estimated coeﬃcients on Basel II SA*RWA/TA in Table 4.11 suggests that the relationship between RWA and returns is less negative for banks using the Basel II SA, compared to those using Basel I. The relationship is not signiﬁcantly diﬀerent for banks following one of the advanced approaches.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

100

Table 4.10: Determinants of returns – performance during the European

sovereign debt crisis

Dependent variable: Stock return from June 2011 to Sep 2011

(1) (2) (3)

All banks Excluding US Large banks

RWA/TangibleAssets

-0.095+

-0.243***

-0.026

(0.057)

(0.068)

(0.166)

TCE/TangibleAssets

-0.021

0.424

0.759

(0.070)

(0.359)

(1.491)

CustDeposits

0.039

0.024

-0.064

(0.046)

(0.071)

(0.112)

Securities/Assets

0.052

-0.031

0.018

(0.035)

(0.076)

(0.145)

NPL/Loans

-0.254

-0.053

-0.221

(0.187)

(0.179)

(0.737)

ROAA

2.584***

1.101

6.305*

(0.284)

(0.937)

(3.146)

log(Assets)

-2.977***

-1.704**

-4.024*

(0.440)

(0.624)

(2.359)

Beta

-0.254

-2.456

-1.409

(0.204)

(3.069)

(1.626)

Observations

804 304 129

Adj R-squared

0.363

0.722

0.707

The table presents regressions for banks in 32 countries. The dependent variable is the bank’s stock return over the period from June 30, 2011 to September 30, 2011. The independent variables, all values for 2010, are the ratio of risk-weighted assets to tangible assets, the capital ratio (tangible common equity (TCE) to tangible assets (TA)), the share of stable deposits, the share of securities in the bank’s assets, the share of non-performing loans, the return on assets, the log of assets, and the stock’s beta with a national market index. Country dummies and dummy variables representing the bank’s business model are included in each speciﬁcation. Column (1) presents the whole sample of banks, column (2) the sample excluding banks based in the United States, and column (3) the banks with total assets greater than 50 billion US dollars in 2010.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

101

Table 4.11: Determinants of returns during the European sovereign debt

crisis – Basel I versus Basel II approaches to measuring RWA

Dependent variable: Stock return from June 2011 to Sep 2011

(1) (1) (2)

All banks

All banks Excluding US

Basel II SA indicator

-28.565** -37.732**

(11.211)

(15.532)

Basel II Advanced indicator

14.742

2.630

(11.506)

(11.950)

RWA/TA

-0.217**

-0.216+

-0.533***

(0.101)

(0.129)

(0.162)

Basel II SA * RWA/TA

0.228*

0.406*

(0.130)

(0.213)

Basel II Advanced * RWA/TA

-0.193

-0.076

(0.152)

(0.176)

TCE/TA

-0.369***

-0.020

-0.170

(0.122)

(0.104)

(0.351)

CustDeposits

0.268**

0.194***

0.592***

(0.102)

(0.062)

(0.079)

Securities/Assets

0.094*

0.106**

0.039

(0.051)

(0.051)

(0.098)

NPL/Loans

-0.301

-0.227

-0.199

(0.188)

(0.161)

(0.141)

ROAA

1.706**

2.625***

2.592

(0.633)

(0.256)

(1.596)

log(Assets)

-2.759*** -3.985*** -4.245***

(0.580)

(0.483)

(0.725)

Beta

0.020

0.141

0.693

(0.145)

(0.121)

(2.095)

Observations

767 767 304

Adj R-squared

0.249

0.314

0.423

The table presents regressions for banks in 32 countries. The dependent variable is the bank’s stock return over the period from June 30, 2011 to September 30, 2011. The independent variables include: an indicator variable representing countries where banks predominantly use the Basel II Standardized Approach to calculating RWA, an indicator variable for countries where banks predominantly use one of the Basel II Advanced IRB approaches, and interactions between these indicators and RWA/TA. The remaining independent variables are as in Table table:euroreg1and each speciﬁcation includes regional dummies (for North America, Europe, or Asia), dummies representing bank business model, and dummy variables representing the accounting method.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

102

4.8 Conclusion

There has been a steady decline in the measure of asset-risk that banks report to regulators – risk-weighted assets (RWA) – over the last decade. In light of this trend and other indications that banks may under-report RWA in an attempt to minimize the amount of capital they must hold, we study how equity market investors account for the riskiness of RWA by examining the determinants of stock returns and a stock-market measure of risk of an international panel of banks.
Regarding banking stock returns, we ﬁnd a negative relationship between RWA and stock returns over periods of ﬁnancial crisis, suggesting that investors use RWA as an indicator of bank portfolio risk. Banks with higher risk-weighted assets performed worse both during the subprime crisis and a recent period of stock market decline accompanying the Eurozone sovereign debt crisis. Comparing regions with diﬀerent regulatory structures, we ﬁnd evidence that the relationship between stock returns and RWA is weaker in countries where banks have more discretion in the calculation of RWA. Speciﬁcally, in countries that had implemented Basel II before the onset of the recent ﬁnancial crisis, primarily following the standardized approach to measuring credit risk, the relationship between stock returns and RWA is less negative or even positive. In addition, we ﬁnd a trade-oﬀ between capital and funding in terms of their positive eﬀects on bank stock returns. The more stable a bank’s funding, the less positive the eﬀect of capital on its

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

103

stock return. We also study a market measure of risk, the bank’s systematic risk, or
beta, from 2004 to 2010. We ﬁnd no static relationship between RWA and systematic risk across banks, however, we ﬁnd evidence of a dynamic eﬀect where systematic risk increases for those banks whose RWA increases. Finally, there is evidence of a change in the relationship between systematic risk and RWA since the start of the crisis. The positive relationship between RWA and market risk in the three years prior to the crisis, from 2004 to 2006, becomes negative, albeit small in magnitude, after the crisis.
In light of increasing risk-aversion in markets during times of crisis, the question of how market assessments of risk should be incorporated into banking regulation and supervision remains. Indeed, the asymmetry of information between banks, supervisors, and market participants regarding how risky RWA are can lead to increased uncertainty about the adequacy of bank capital which, during a ﬁnancial crisis, can have damaging eﬀects for ﬁnancial stability.

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS
4.9 Appendix

104

Country Australia Austria Belgium Canada China Denmark Finland France Germany Greece Hong Kong India Indonesia Italy Japan Korea Malaysia

Table 4.12: List of countries – full sample

# banks 6 4 2 10 11 11 2 4 5 8 6 12 8 20 81 4 9

% Sample 0.75 0.5 0.25 1.24 1.37 1.37 0.25 0.5 0.62 1 0.75 1.49 1 2.49 10.07 0.5 1.12

Country Norway Philippines Poland Russian Federation Singapore Spain Sri Lanka Sweden Switzerland Taiwan Thailand Turkey Ukraine UK USA

# banks 14 11 10 8 4 8 5 4 3 9 6 10 2 7 500

% Sample 1.74 1.37 1.24 1 0.5 1 0.62 0.5 0.37 1.12 0.75 1.24 0.25 0.87 62.19

CHAPTER 4. BANKS’ RISK-WEIGHTED ASSETS

Table 4.13: Descriptive statistics – correlations of explanatory variables

Correlations of explanatory variables in 2006 (762 observations)

RWA/TA TCE/TA

Tier1/TA TotalCap/TA

RWA/TA

1

TCE/TA

0.141

1

Tier 1 Capital/TA

0.084

0.948

1

Total Capital/TA

0.742

0.689

0.671

1

CustDeposits

-0.064

0.268

0.249

0.055

Securities/Assets

-0.097

-0.130

-0.161

-0.036

NPL/Loans

0.240

-0.217

-0.229

0.088

ROAA

-0.127

-0.292

-0.309

-0.367

log(Assets)

0.153

-0.581

-0.592

-0.169

Beta

0.173

-0.173

-0.182

0.058

CustDeps Sec/Assets NPL/Loans

1 -0.227 -0.060 -0.096 -0.494 -0.206

1 0.160 -0.082 0.313 0.088

1 -0.073 0.292 0.218

ROAA log(Assets)

1 0.152 0.050

1 0.3955

Beta 1

Correlations of explanatory variables in 2010 (804 observations)

RWA/TA TCE/TA

Tier1/TA TotalCap/TA

RWA/TA

1

TCE/TA

0.241

1

Tier 1 Capital/TA

0.495

0.765

1

Total Capital/TA

0.572

0.724

0.961

1

CustDeposits

0.337

0.159

0.246

0.210

Securities/Assets

-0.568

-0.045

-0.161

-0.194

NPL/Loans

0.138

-0.026

0.045

0.079

ROAA

-0.076

0.224

0.046

0.054

log(Assets)

-0.478

-0.360

-0.496

-0.426

Beta

0.018

-0.122

-0.044

-0.026

CustDeps Sec/Assets NPL/Loans

1 -0.317 -0.061 -0.080 -0.508 -0.043

1 -0.179 0.200 0.299 0.024

1 -0.392 -0.107 0.009

ROAA log(Assets)

1 0.1702 -0.063

1 0.2672

Beta 1

105

106
REFERENCES
Acharya, Viral V. (2011), “Ring-fencing is good, but no panacea,” in Thorsten Beck, ed.: The Future of Banking (A VoxEU.org eBook), Centre for Economic Policy Research, 35.
Acharya, Viral V., Gujral, Irvind and Hyun Song Shin (March 18, 2009), “Dividends and Bank Capital in the Financial Crisis of 2007-2009”
Acharya, Viral V., Pedersen, Lasse Heje, Philippon, Thomas and Matthew P. Richardson (April 23, 2010), “Measuring Systemic Risk,” FRB of Cleveland Working Paper No. 10-02.
Acharya, Viral V., Philipp Schnabl and Gustavo Suarez (2010), “Securitization without risk transfer,” Journal of Financial Economics, forthcoming.
Acharya, Viral V. and Tanju Yorulmazer (2007), “Too many to fail An analysis of time-inconsistency in bank closure policies,” Journal of Financial Intermediation 16, pp 1-31.
Acharya, Viral V. and Tanju Yorulmazer (2008), “Information Contagion and Bank Herding,” Journal of Money, Credit and Banking 40(1), pp 215-231.
Adrian, Tobias and Markus K. Brunnermeier (August 27, 2009), “CoVaR”, FRB of New York Staﬀ Report No. 348.
Allen, Franklin and Douglas Gale (1998), “Optimal Financial Crises,” The Journal of Finance 53(4), pp 1245-1284.
Allen, Franklin and Douglas Gale (2000), “Financial Contagion,” Journal of Political Economy 8(1), pp 1-33.
Allen, Franklin and Douglas Gale (2005), “From Cash-in-the-Market Pricing to Financial Fragility,” Journal of the European Economic Association 3(2-3), pp535-546.
Allen, Mark, Christoph Rosenberg, Christian Keller, Brad Setser, and Nouriel Roubini (2002), “A Balance sheet approach to ﬁnancial crisis,” IMF Working Paper No. 210.
Aggarwal, Raj and Kevin T. Jacques (2001), “The impact of FDICIA and prompt corrective action on bank capital and risk: Estimates using a simultaneous equations model,” Journal of Banking and Finance 25, pp 11391160.
Ashcraft, Adam B. (2005), “Are Banks Really Special? New Evidence from the FDIC-Induced Failure of Healthy Banks,” The American Economic Review 95(5), pp 1712-1730.
Avery, R.B., and A.N. Berger (1991), “Risk-based capital and deposit insurance reform,” Journal of Banking and Finance 15, pp 847-874.
Bank of Canada (August 2010), “Strengthening International Capital and Liquidity Standards: A Macroeconomic Impact Assessment for Canada.
Bank for International Settlements, Basel Committee on Banking Supervision (August 2010), “An Assessment of the Long-Term Economic Impact of the New Regulatory Framework”
Barth, James R., Gerard Caprio Jr., and Ross Levine (2004), “Bank Regulation and supervision: what works best?” Journal of Financial Intermediation 14, pp 205-248.

107
Basel Committee on Banking Supervision (2005), “An Explanatory Note on the Basel II IRB Risk Weight Functions,” Bank for International Settlements.
Beck, Thortsen and Asli Demirgu¨¸c-Kunt (2009), “Financial Institutions and Markets Across Countries and over Time: Data and Analysis,”World Bank Policy Research Working Paper No. 4943.
Beltratti, Andrea and Ren´e M. Stulz (2011), “The Credit Crisis Around the Globe: Why Did Some Banks Perform Better?” Journal of Financial Economics, forthcoming.
Berger, Allen N., Herring, Richard J. and Giorgio P. Szego (1995), “The role of capital in ﬁnancial institutions,” Journal of Banking and Finance 19, pp 393-430.
Bhattacharya, S. and D. Gale (1987), “Preference Shocks, Liquidity, and Central Bank Policy,” in W.A. Barnett and K.J. Singleton (eds.), New Approaches to Monetary Economics, Cambridge University Press.
Bliss, Robert R. and Mark Flannery (2001), “Market Discipline in the Governance of U.S. Bank Holding Companies, Monitoring versus Inﬂuencing” in Frederic S. Mishkin, ed: Prudential Supervision: What Works and What Doesn’t (University of Chicago Press).
Bradley, M.G., C.A. Wambeke, and D.A. Whidbee (1991), “Risk weights, riskbased capital and deposit insurance,” Journal of Banking and Finance 15, pp 875-893.
Bologna, Pierluigi (2011), “Is There a Role for Funding in Explaining Recent U.S. Banks Failures?” IMF Working Paper WP/11/180.
Boyd, John, Gianni De Nicol´o and Elena Loukoianova (2009), “Banking Crises and Crisis Dating: Theory and Evidence,” IMF Working Paper No. 141.
Bryant, John (1980), “A model of reserves, bank runs, and deposit insurance,” Journal of Banking and Finance 4(4), pp 335-344.
Brunnermeier, Markus, and Lesse Pedersen (2009), “Market Liquidity and Funding Liquidity.” Review of Financial Studies 22(6), pp 2201-2238.
Calomiris, Charles (2009), “Banking Crises and the Rules of the Game,” National Bureau of Economic Research Working Paper No. 15403.
Chen, Yenhing (1999), “Banking Panics: The Role of the First-Come, FirstServed Rule and Information Externalities,” Journal of Political Economy 107, pp 946-68.
Cordell, L.R., and Kathleen.K. King (1995), “A market evaluation of the riskbased capital standards for the U.S. ﬁnancial system,” Journal of Banking and Finance 19, pp 531-562.
Danielsson, Jon and Jean-Pierre Zigrand (2008), “Equilibrium Asset Pricing with Systemic Risk,” Economic Theory 35, pp 293-319.
De Bandt, Olivier and Philipp Hartmann (November 2000), “Systemic Risk: A Survey,” ECB Working Paper No. 35.
Demirgu¨¸c-Kunt, Asli, Baybars Karacaovali and Luc Laeven (2005), “Deposit Insurance around the World: A Comprehensive Database,” Policy Research Working Paper Series 3628, Washington, DC: World Bank.
Demirgu¨¸c-Kunt, Asli, and Enrica Detragiache (October 2002), “Does deposit insurance increase banking system stability? An empirical investigation,”

108
Journal of Monetary Economics 49(7), pp 1373-1406. Demirgu¨¸c-Kunt, Asli and Detragiache, Enrica and Ouarda Merrouche (2010),
“Bank capital: lessons from the ﬁnancial crisis,” World Bank Policy Research Working Paper Series 5473, Washington, DC: World Bank. Demirgu¨¸c-Kunt, Asli and Harry Huizinga (2009), “Bank activity and funding strategies: The impact on risk and return,” European Banking Center Working Paper, Tilburg University. Diamond, D. W., and P.H. Dybvig (1983), “Bank runs, deposit insurance, and liquidity,” Journal of Political Economy 91(3), pp 401-419. Douglas W. Diamond and Raghuram G. Rajan (2005), “Liquidity Shortages and Banking Crises,” The Journal of Finance 60(2), pp 615-647. Duchin, Ran and Denis Sosyura (2012), “Safer Ratios, Riskier Portfolios: Banks Response to Government Aid,” University of Michigan Ross School of Business Working Paper No. 1165. Flannery, Mark J. and Kasturi P. Rangan (2002), “Market forces at work in the banking industry: evidence from the capital buildup of the 1990s,” Federal Reserve Bank of Chicago Proceedings, May. Fostel, Ana and John Geanakoplos (2008), “Leverage Cycles and the Anxious Economy,” The American Economic Review 98(4), pp 1211-1244. Frankel, Jeﬀrey and George Saravelos (2010), “Are Leading Indicators of Financial Crises Useful for Assessing Country Vulnerability? Evidence from the 2008-09 Global Crisis,” NBER Working Paper 16047. Glick, Reuven and Michael Hutchison (2001), “Banking and Currency Crises: How Common Are Twins? In Financial Crises in Emerging Markets,” ed. by Glick, Moreno, and Spiegel: New York: Cambridge University Press, 2001. Giglio, Stefano (November 2010), “CDS Spreads and Systemic Financial Risk” Goldstein, Itay and Ady Pauzner (June 2005), “Demand Deposit Contracts and the Probability of Bank Runs,” The Journal of Finance 60(3), pp 1293-1327. Gorton, Gary B. (2010), “Slapped by the Invisible Hand: The Panic of 2007” (Oxford University Press). Hasan, I. and G. Dwyer (1994) “Bank Runs in the Free Banking Period,” Journal of Money, Credit and Banking 26, pp 271-288. Heckman, James J. (1979), “Sample Selection Bias as a Speciﬁcation Error,” Econometrica 47(1), pp 153-61. Huber, P. J. (1967), “The behavior of maximum likelihood estimates under nonstandard conditions,” In Proceedings of the Fifth Berkeley Symposium on Mathematical Statistics and Probability. Berkeley, CA: University of California Press, Vol. 1, pp 221-233. Jones, D.S., and Kathleen. K. King (1995), “The implementation of prompt corrective action,” Journal of Banking and Finance 19, pp 491-510. Kaminsky, G. L., and Reinhart, C. M. (1999), “The twin crises: The causes of banking and balance-of-payments problems,” The American Economic Review 89(3), pp 473-500. Kim, Daesik, and Anthony M. Santomero (1988), “Risk in Banking and Capital Regulation,” The Journal of Finance 43, pp 1219-1233.

109
Kodres, L. and M. Pritsker (2002), “A Rational Expectations Model of Financial Contagion,” The Journal of Finance 57(2), pp 769-799.
Krishnamurthy, Arvind (2010), “Ampliﬁcation Mechanisms in Liquidity Crises,” American Economic Association Journals - Macroeconomics 2(3), pp 1-30.
Laeven, Luc and Fabian Valencia (2008), “Systemic Banking Crises: A New Database,” IMF Working Paper WP/08/224.
Laeven, Luc and Fabian Valencia (2010), “Resolution of Banking Crises: The Good, the Bad, and the Ugly,” IMF Working Paper WP/10/146.
McIntosh, C.S. and J.F. Dorfman (1992), “Qualitative Forecast Evaluation: A Comparison of Two Performance Measures,” American Journal of Agricultural Economics 74, pp 209-214.
Le Lesl´e, Vanessa and Soﬁya Avramova (2012), “Revisiting Risk-Weighted Assets,” IMF Working Paper WP/12/90.
Merton, R.C. (1995), “Financial innovation and the management and regulation of ﬁnancial institutions,” Journal of Banking, and Finance 19, pp 461-481.
Myers, S. (1977), “Determinants of Corporate Borrowing,” Journal of Financial Economics 5, pp 147-75.
Myers, S., Majluf, N. (1984), “Corporate ﬁnancing and investment decisions when ﬁrms have information that investors do not have,” Journal of Financial Economics 13, pp 187- 221.
Randall J. Pozdena (1991), “Is Banking Really Prone to Panics?” FRBSF Weekly Letter, No. 91-35 (October 11).
Reinhart, Carmen M. and Kenneth S. Rogoﬀ (2009), This Time is Diﬀerent: Eight Centuries of Financial Folly (Princeton University Press).
Reinhart, Carmen M. and Kenneth S. Rogoﬀ (August 2011), “From Financial Crash to Debt Crisis,” The American Economic Review 101, pp 1676-1706.
Rose, Andrew K. and Mark M. Spiegel (2009), “Cross-Country Causes and Consequences of the 2008 Crisis: Early Warning,”CEPR Discussion Paper 7354.
Schaeck, Klaus and Martin Cihak and Simon Wolfe (2009), “Are Competitive Banking Systems More Stable?” Journal of Money, Credit and Banking 41(4), pp 711-734.
White, H. (1980), “A heteroskedasticity-consistent covariance matrix estimator and a direct test for heteroskedasticity,” Econometrica 48, pp 817-830.
Wooldridge, J (2002) Econometric Analysis of Cross Section and Panel Data. Cambridge: MIT Press.
Von Hagen, Ju¨rgen and Tai-Kuang Ho (August 2007), “Money Market Pressure and the Determinants of Banking Crises,” Journal of Money, Credit and Banking 39(5), pp 1037-1066.