I N T E R N A T I O N A L M O N E T A R Y F U N D

ESSAYS ON THE GLOBAL

FINANCIAL CRISIS
Edited by Heiko Hesse
To Evi
About the Author

Heiko Hesse is an Economist at the International Monetary Fund and is currently on

secondment at the European Commission. Since joining the IMF in 2007, Mr. Hesse's work
has focused on financial stability risks (on the GFSR), the Middle East, banking sector issues
in Europe as well as sovereign debt (restructuring).
His IMF country experiences include e.g. Bulgaria, Cyprus, the Euro area, Lebanon,
Romania, Spain and Turkey.

Prior to the IMF, he was an Economist at the World Bank on the Commission on Growth and
Development as well as a Visiting Scholar at Yale University.

He also worked at McKinsey and NERA Economic Consulting. Mr. Hesse obtained his PhD
in Economics from the University of Oxford and his B.Sc. in Financial Economics from the
University of Essex.
 Essays on the Global Financial Crisis

Table of Content

Preface ___________________________________________________________________8

Introduction _______________________________________________________________9

Origins of the Global Financial Crisis and Policy Response

I. The Transmission of Liquidity Shocks during the Crisis, with Nathaniel Frank and
Brenda González-Hermosillo, 2008 ___________________________________________14
A. Introduction ____________________________________________________________14
B. Transmission of Spillovers during the Subprime Crisis___________________________16
C. Data __________________________________________________________________20
D. Methodology ___________________________________________________________21
E. Results ________________________________________________________________23
F. Conclusion _____________________________________________________________28
References ________________________________________________________________29

II. The Effectiveness of Central Bank Interventions During the First Phase of the
Subprime Crisis, with Nathaniel Frank, 2009 __________________________________33
A. Introduction ____________________________________________________________33
B. Review of Developments and Policy Interventions ______________________________35
C. Empirical Analysis _______________________________________________________40
D. Bivariate GARCH Framework______________________________________________55
E. Policy Implications and Conclusions _________________________________________57
References ________________________________________________________________60

Spillovers and Contagion

III. Financial Spillovers to Emerging Markets during the Global Financial Crisis, with
Nathaniel Frank, 2009 _____________________________________________________62
A. Introduction ____________________________________________________________62
B. Transmission of Spillovers to EM Countries During the Subprime Crisis: A Qualitative
Overview _________________________________________________________________64
C. Data __________________________________________________________________68
D. Methodology ___________________________________________________________72
E. Results ________________________________________________________________73
F. Conclusion _____________________________________________________________78
References ________________________________________________________________79
 2

IV. Global Market Conditions and Systemic Risk, with Brenda González-Hermosillo,
2009_____________________________________________________________________81
A. Introduction ____________________________________________________________81
B. Overview of Systemic Risk ________________________________________________84
C. Global Market Conditions and Systemic Risk: A Qualitative View _________________86
D. Markov-Regime Switching Analysis _________________________________________89
E. Results During the Peak of the Crisis _________________________________________90
F. Results After Massive Government Programs in 2009 to Address the Global Crisis ____93
G. Conclusion _____________________________________________________________97
References ________________________________________________________________99

Case Studies

V. What do Sovereign Wealth Funds Imply for Financial Stability?, with Tao Sun, 2009102
A. Introduction ___________________________________________________________102
B. Literature Review _______________________________________________________104
C. Data and Methodology ___________________________________________________105
D. Data _________________________________________________________________106
E. Methodology___________________________________________________________109
F. Empirical Results _______________________________________________________110
G. Conclusion ____________________________________________________________114
References _______________________________________________________________116

VI. Recent Credit Stagnation in the MENA Region: What to Expect? What Can Be
Done? with Adolfo Barajas, Ralph Chami and Raphael Espinoza, 2010 ___________118
A. Introduction ___________________________________________________________118
B. The Recent Credit Cycle in Historical and International Perspective _______________122
C. Anatomy of the MENA Credit Slowdown ____________________________________126
D. Econometric Analysis of Bank-Level Credit Growth ___________________________129
E. Conclusion ____________________________________________________________133
References _______________________________________________________________136

VII. Financial Spillovers and Deleveraging: The Case of Romania, 2012___________138

A. Introduction ___________________________________________________________138
B. Foreign Bank Deleveraging _______________________________________________140
C. Financial Spillover Analysis ______________________________________________143
D. Conclusion ____________________________________________________________148
References _______________________________________________________________150

VIII. Progress with Bank Restructuring and Resolution in Europe, with Nadege
Jassaud, 2013 ___________________________________________________________154
A. Executive Summary _____________________________________________________154
B. Introduction ___________________________________________________________155
C. Recent Developments ____________________________________________________155
 3

D. Crisis Response ________________________________________________________156

E. On-Going Challenges ____________________________________________________160
F. Resolution and Restructuring Framework ____________________________________162
G. Resolution Framework for Problem Banks ___________________________________163
H. Disclosure_____________________________________________________________168
References _______________________________________________________________173

Stress Testing Issues

IX. Next Generation System-Wide Liquidity Stress Testing, with Christian Schmieder,
Benjamin Neuendorfer, Claus Puhr and Stefan Schmitz, 2012,___________________175
A. Introduction ___________________________________________________________175
B. Review of General Concepts to Assess Liquidity Risks _________________________179
C. Methodological Aspects __________________________________________________182
D. Framework of Next Generation Liquidity Stress Tests __________________________186
E. Design of Stress Scenarios ________________________________________________190
F. Run-off Rates for Different Funding Sources _________________________________191
G. Asset side: Fire Sales & Rollover __________________________________________193
H. Link Between Liquidity and Solvency_______________________________________196
I. Liquidity Stress Tests in Recent FSAPs and Benchmark Scenarios _________________197
J. Case Study _____________________________________________________________201
K. Case Study Fully Fledged Cash Flow Analysis ________________________________203
L. Conclusion ____________________________________________________________204
References _______________________________________________________________230

X. European FSAP: Technical Note on Stress Testing of Banks, with Daniel Hardy,
2013____________________________________________________________________234
A. Executive Summary _____________________________________________________234
B. Introduction ___________________________________________________________236
C. Background ___________________________________________________________237
D. The 2013 Bank Solvency Stress Testing Exercise ______________________________239
E. Publication and Transparency _____________________________________________239
F. Consistency and Quality Control Mechanisms_________________________________240
G. Input Data Review ______________________________________________________241
H. Refinement of Satellite Models ____________________________________________244
I. Achieving Supervisory Orientation __________________________________________246
J. Future Priorities _________________________________________________________249
K. Liquidity Stress Testing __________________________________________________251
A. Literature Review _______________________________________________________254
B. Integrating Liquidity and Solvency Risks and Bank Reactions in Stress Tests ________255
C. Liquidity Risks Analysis by Authorities _____________________________________256
D. Basel III and Liquidity Stress Testing _______________________________________257

XI. How to Capture Macro-Financial Spillover Effects in Stress Tests?, with Ferhan
Salman and Christian Schmieder, 2014 ______________________________________262
A. Introduction ___________________________________________________________262
 4

B. Financial Spillovers from the Euro periphery to the Rest of the World _____________266
C. DCC GARCH Approach _________________________________________________268
D. Liquidity and Solvency Stress Testing_______________________________________272
E. Integration of the Financial Spillover Analysis with the Stress Testing Approach _____274
F. Conclusion ____________________________________________________________279
References _______________________________________________________________291

Debt Sustainability and Sovereign Debt Restructuring

XII. Reprofiling and Domestic Financial Stability: Recent Experiences, 2014_______294

References _______________________________________________________________304

XIII. Is Banks’ Home Bias Good or Bad for Debt Sustainability? with Tamon Asonuma
and Said Bakhache, 2015 __________________________________________________305
A. Introduction ___________________________________________________________305
B. Literature Review _______________________________________________________308
C. Empirical Analysis on Home Bias __________________________________________309
D. Borrowing Costs of Sovereigns ____________________________________________311
E. Public Debt ____________________________________________________________316
F. Primary Balance Adjustments _____________________________________________318
G. Debt under Distress _____________________________________________________321
H. Robustness Tests _______________________________________________________322
I. Other Home Bias Issues __________________________________________________323
J. Conclusion _____________________________________________________________325
References _______________________________________________________________338

BOXES
1. Proposed Resolution Directive––Risks and Areas for Enhancements1 ______________164
2. Capital Outcome of the 2011 Stress Test and Recapitalization Exercises ____________238
3. EBA Stress Tests and Bank Funding Costs ___________________________________245
4. Principles for Macro-financial Stress Testing__________________________________250
5. Asset Encumbrance and Liquidity Risk Assessments ___________________________252
6. Integrating Liqudity and Solvency Risks and Bank Reaction in Stress Tests _________273
7. Impact of Sovereign Debt Maturity Extensions on Domestic Bank’s Balance Sheets___299
8. Central Bank Liquidity Provision in Past Reprofiling Cases ______________________300
9. Accounting Treatment of Bank Holdings of Government Bonds __________________301
10. Banking Sector Developments in Greece during the Crisis ______________________303

FIGURES
1. Selected Conditional Correlations ___________________________________________23
2. Conditional Correlations from Modified DCC Model ____________________________25
3. U.S., U.K., and Euro area Libor-OIS Spreads __________________________________39
4. Decomposition of U.S. and Euro area Libor-OIS Spreads. ________________________42
5. Decomposition of Libor-OIS Spreads ________________________________________43
 5

6. Markov Switching Mean-variance Model for Euro area and U.S. Libor-OIS __________48
7. Markov Switching ARCH Model for Euro area and U.S. Libor-OIS Spreads __________50
8. Impulse Reponse Functions ________________________________________________54
9. U.S. and EM Financial Variables ____________________________________________70
10. U.S. and EM Financial Variables ___________________________________________72
11. Implied Correlations between U.S. and EM Financial Variables ___________________75
12. Implied Correlations between U.S. and EM Financial Variables ___________________77
13. Euro-dollar Forex Swap __________________________________________________91
14. Markov-Switching ARCH Model of VIX ____________________________________92
15. Markov-Switching ARCH Model of TED Spread ______________________________93
16. Euro-Dollar Forex Swap __________________________________________________94
17a. Markov-Switching ARCH Model of VIX____________________________________95
17b. Markov-Switching ARCH Model of VIX ___________________________________95
18a. Markov-Switching ARCH Model of TED Spread _____________________________96
18b. Markov-Switching ARCH Model of TED Spread _____________________________97
19. Ratios of SWF Investments and Divestments _________________________________108
20. Recent Declines in Real Credit Growth _____________________________________119
21a.. Credib Boom Events in the Last Expansion ________________________________123
21b. Regional Average Differences from Trend__________________________________124
22. Frequency of Credit Booms throughout the World, 1983–2008 __________________125
23. Boom Frequency over Time ______________________________________________125
24. MENA: Credit Behavior Surrounding Booms ________________________________126
25. Decomposition of the Credit Slowdown in Selected MENA Countries _____________128
26. Loan-Deposit Ratios in Selected MENAP Countries ___________________________129
27. Drivers of Lending Growth in MENA Banks _________________________________132
28. Banks’ External Positions ________________________________________________139
29. CDS and EMBIG Developments __________________________________________139
30. Romanian Banks’ Parent Funding _________________________________________141
31. Parent Funding by Maturity ______________________________________________141
32. CESEE Foreign Bank Funding ____________________________________________142
33. DCC GARCH Equity Market Mode ________________________________________145
34. DCC GARCH CDS Model _______________________________________________146
35. DCC GARCH EMBIG Model ____________________________________________147
36. ARCH Markov Switching Models _________________________________________148
37. Assets of EU and U.S. Banking Groups _____________________________________156
38. EU: ECB Monetary Financing Operations vis a vis Euro Area Banks______________158
39. Deleveraging/Restructuring Plans _________________________________________158
40. EU: Tier 1 Ratio of EU Banks 2008–12 _____________________________________159
41. EU Banks NPLs to Total Loans ___________________________________________161
42. EU: NPLs to Total Loans ________________________________________________161
43. Overview on Liquidity Risk Framework ____________________________________188
44. Outcome of Implied Cash Flow Stress Tests for Stylized Banks __________________203
 6

45. Stylised Design of Stress Tests ____________________________________________264

46. Estimated GARCH Correlations GIIPS with European Countries _________________270
47. Estimated GARCH Correlations GIIPS with Non-European Countries_____________270
48. Estimated GARCH Correlations GIIPS with EM Countries and Korea_____________271
49. Estimated GARCH Correlations GIIPS with Germany and the U.S: _______________271
50. Overview of the concept to simulate stress at the bank level _____________________275
51. Outcome of solvency tests _______________________________________________277
52. Outcome of liquidity tests ________________________________________________279
53. Banks’ Domestic Sovereign Holdings/Total Bank Assets and Public Debt __________306
54. Average Public Debt (2007) and Home Bias (average, 2005–07) in AM and EM ____311
55. Bond Spreads and Home Bias in AMs ______________________________________314
56. EM Sovereigns Borrowing Costs in the Domestic Market _______________________315
57. Estimated GARCH Correlations with VIX___________________________________316
58. Public Debt in GDP and Home Bias (Average, 2005–07) _______________________318
59. Fiscal Reactions and Home Bias___________________________________________320
60. Debt in Distress and Home Bias ___________________________________________321
61. European Banks’ Domestic Holdings of Sovereign Debt________________________325

TABLES
1. Markow Switching Parameters for Leveles and Volatility Models __________________46
2. Bivariate VAR Model _____________________________________________________52
3. Impact of Central Bank Interventions on LIBOR-OIS Spreads _____________________57
4. Country of Target Firms __________________________________________________107
5. Acquiring SWFs ________________________________________________________108
6. Stock Market Reactions to Announcements of SWF Investments and Divestments ____112
7. Stock Market Reactions to Announcements of SWF Investments and Divestments ____115
8. Balance Sheet Decomposition of Changes in Credit Growth in the MENA Region ____134
9. MENA Countries-Regressions for Bank-Level Loan Growth _____________________135
10. External Positions of BIS-reporting Banks vis-à-vis CESEE _____________________144
11. EU: Public Interventions in the EU Banking Sector: 2008–11____________________157
12. Comparison of Pros and Cons of Balance Sheet Type TD and BU Liquidity Stress
Tests ___________________________________________________________________186
13. Overview on the Main Elements of Three Liquidity Tests_______________________189
14. Magnitude of Runs on Funding—Empirical Evidence and Stress Test Assumptions __192
15. Supervisory Haircuts Based on Solvency Regime and Liquidity Regime ___________195
16. Benchmark Scenarios ___________________________________________________200
17. Implied Cash Flow Case Study—Sample Banks ______________________________201
18. Outcome of Fully Fledged Cash Flow Stress Tests for Stylized Banks _____________204
19. Indicators of Fundamentals and Policy Track Record __________________________296
20. Summary of Home Bias Indicators (average, 2005–07 and 2009–11) ______________310
21. Average EM and AM Estimated GARCH Correlations with VIX _________________316
22. Estimated Fiscal Policy Reactions _________________________________________320
 7

ANNEXES
1. Figures_________________________________________________________________31
2. DCC GARCH Methodology _______________________________________________151
3. Markov-Regime Switching Analysis ________________________________________153
4. Experience with Asset Quality Reviews ______________________________________170
5. Experience with Asset Management Companies in Crisis Countries ________________172
6. Reviewing Liquidity Issues during the Financial Crisis __________________________206
7. Cross-Country Funding Pattern ____________________________________________211
8. Details on all Modules of the Stress Testing Framework _________________________213
9. Additional Information on Scenario Specification ______________________________225
10. Link Between Solvency and Liquidity ______________________________________229
11. Approaches to Liquidity Stress Testing _____________________________________254
12. Outcome of Panel Regressions Assessing Spillover Risks _______________________281
13. Outline of the DCC GARCH Method_______________________________________284
14. Benchmark Stress Scenarios ______________________________________________285
15. Illustrative Example for the Solvency Test ___________________________________286
16 Illustrative Example for Liquidity __________________________________________289
17. Computations of Home Bias Indicators _____________________________________327
18. Details and Sources of Macroeconomic Variables _____________________________329
19. Outline of the DCC GARCH Method_______________________________________330
20. Home Bias Regression Tables ____________________________________________331
 PREFACE

After working for a year on the Commission on Growth and Development at the World
Bank, I had the fortune of joining the IMF in September 2007 at the very beginning of the
financial crisis that had its origins in the United States. In my first assignment in the Global
Financial Crisis Division and on the GFSR, I was immediately drawn into crisis-related
research and policy work such as the transmission of liquidity shocks, central bank
interventions, systemic risks, emerging market spillovers or contagion. It has been an
exciting time ever since on my different IMF assignments.

This open access book “Essays on the Global Financial Crisis” brings together research and
policy work that I have worked on over the last nine years at the IMF. In predominantly joint
work with my co-authors, the book covers a wide range of issues from the origins of the
financial crisis, the policy response, spillovers and contagion, case studies, bank stress testing
to debt sustainability and sovereign debt restructuring.

The book chapters are mainly of empirical nature, while also distilling relevant policy
conclusions. All individual chapters are published as IMF working papers or part of IMF
country reports or policy papers. A number of the chapters have been also published in peer-
reviewed journals, e.g. World Economics, International Finance Review, Journal of
Emerging Market Finance, Journal of Financial Perspectives, Czech Journal of Economics
and Finance or Central Banking. Many of the chapters appeared as shorter blog versions on
VOX and EconoMonitor. As a matter of fact, I have been a strong supporter over the years in
making my research and policy work accessible to the wider public with shorter and non-
technical blog publications. This open access book continues this tradition by combining in
one monograph thirteen chapters on diverse but also common issues on the Global Financial
Crisis.

The work would not have been possible without the excellent collaborations of my numerous
co-authors over the years. My deepest thanks go out to Nathaniel Frank, Brenda Gonzalez-
Hermosillo, Christian Schmieder, Tamon Asonuma, Adolfo Barajas, Said Bakhache, Ralph
Chami, Raphael Espinoza, Daniel Hardy, Nadege Jassaud, Benjamin Neuendorfer, Claus
Puhr, Ferhan Salman, Stefan Schmitz and Tao Sun. I am also very grateful for the guidance
and support of my IMF supervisors in the different book chapters: Charles Enoch, Daniel
Hardy, Laura Kodres, Said Bakhache, Reza Baqir, Hugh Bredenkamp, Ralph Chami, Mark
Flanagan, Marina Moretti, Genevieve Verdier and Erik de Vrijer.

The usual disclaimer applies: The views expressed in this open access book are my views
and those of the co-authors and should not be attributed to the IMF, its Executive Board, or
its management. Any errors and omissions are the solely my responsibility and those of the
co-authors.
 9

INTRODUCTION

The Global Financial Crisis has been a watershed event not only for the United States and
many advanced economies but also emerging markets (EM) around the world. The subprime
crisis that began in the summer of 2007 was triggered by deteriorating quality of
U.S. subprime mortgages. This rapidly propagated across different asset classes and financial
markets. Increased delinquencies on subprime mortgages, driven by rising interest rates for
refinancing and falling house prices, resulted in uncertainty surrounding the value of a
number of structured credit products which had these assets in their underlying portfolios. As
a result, rating agencies downgraded many of the related securities and announced changes in
their methodologies for rating such products. Meanwhile, structured credit mortgage-backed
instruments saw rapid price declines, and the liquidity for initially tradable securities in their
respective secondary markets evaporated.

The losses, downgrades, and changes in methodologies shattered investors' confidence in the
rating agencies' abilities to evaluate risks of complex securities, a result of which, investors
pulled back from structured products in general. With interbank markets across various
advanced economies becoming dysfunctional in early August 2007, there was clear evidence
of a run for “quality” by investors. Financial markets more generally showed signs of stress,
as investor preference moved away from complex structured products in a flight to quality
and liquidity, and global investors' risk appetite sharply decreased due to a widespread re-
pricing of risk.

It soon became apparent that a wide range of different financial institutions had exposures to
many of these mortgage-backed securities, often off-balance sheet entities such as conduits
or structured investment vehicles (SIVs). Due to the increasing uncertainty with regard to
their exposure to and the value of the underlying mortgage-backed securities, investors
became unwilling to roll over the corresponding asset- backed commercial paper. As the
problems with SIVs and conduits deepened, banks came under increasing pressure to rescue
those that they had sponsored by providing liquidity or by taking their respective assets onto
their own balance sheets. As a result, the balance sheets of those financial institutions were
particularly strained by this re-absorption, which in addition was amplified by losses due to
declining asset values.

Many European banks that had large exposures to US asset-backed securities had difficulties
accessing wholesale funding, inducing subsequent market illiquidity in different market
segments. The US subprime crisis increasingly became one of insolvency, as banks such as
Northern Rock, IKB, and Bear Stearns had to be rescued. Due to the major importance of the
interbank money market, central banks in turn intervened by reducing interest rates and
providing additional liquidity to the markets in order to reduce pressures.
 10

The Lehman Brothers collapse in September 2008 was the watershed event that unleashed a
full-blown systemic crisis with global risk aversion dramatically increasing, asset markets
across countries and regions plunging and the unwinding of carry trades that saw high-
yielding EM currencies sharply depreciate within a short period of time. The interbank
market became even more exposed to counterparty and liquidity risk, leading market
participants to globally withdraw from these market segments.

Volatility also spilled over into the foreign currency markets with the carry trades starting to
rapidly unwind at the end of September 2008. High-yielding and previous investment
currencies saw large depreciations against the U.S. dollar, while funding currencies such as
the Japanese yen benefited by a repatriation of funds into Japan. There was a scramble for
U.S. dollars.

EM countries were less affected during the initial stages of the subprime crisis than advanced
economies, as for example EM equity markets peaked in November 2007. But the
persistence of the market dislocations, the deterioration of economic fundamentals in
advanced economies and rising global risk aversion significantly affected EM countries by
late 2008 after the Lehman Brothers collapse. Even EM countries with sound
macroeconomic and financial pre-conditions, built-up over the previous years, have been
strongly affected by the financial contagion that in late 2008 spilled over to the real sector
with export and GDP growth rates plunging and trade finance sharply contracting across the
world.

Against the backdrop of the above short narrative of the beginning of the crisis (see chapter
III for more details), the first four chapters of the open access book focus on the origins and
early phase of the Global Financial Crisis as well as spillovers and contagion:

Chapter I (published in 2008) with Nathaniel Frank and Brenda Gonzalez-Hermosillo

examines the linkages between market and funding liquidity pressures, as well as their
interaction with solvency issues surrounding key financial institutions during the 2007
subprime crisis. Chapter II (published in 2009) with Nathaniel Frank examines the
effectiveness of central bank interventions during the early phase of the crisis by focusing on
the Federal Reserve (Fed) and the European Central Bank (ECB). It provides evidence that
central bank interventions had a statistically significant impact on easing stress in unsecured
interbank markets during the first phase of the subprime crisis which began in July 2007. But
the economic magnitudes of the central bank interventions have overall not been very large.

Chapters III and IV focus on spillovers and contagion of the Global Financial Crisis.
Specifically, chapter III (first published in 2009) with Nathaniel Frank looks at the financial
spillovers to EMs by examining potential financial linkages between liquidity and bank
solvency measures in advanced economies and EM bond and stock markets. The findings at
the time indicate that the notion of possible de-coupling (in the financial markets) had been
 11

misplaced. While EM stock markets reached their peak in the last quarter of 2007,
interlinkages between funding stress and equity markets in advanced economies and EM
financial indicators were highly correlated and have seen sharp increases during specific
crisis moments then. Following that, chapter IV (first published in 2009) with Brenda
Gonzalez-Hermosillo examines several key global market conditions, such as a proxy for
market uncertainty and measures of interbank funding stress, to assess financial volatility and
the likelihood of crisis. Using Markov regime-switching techniques, it shows that the
Lehman Brothers failure was a watershed event in the crisis, although signs of heightened
systemic risk could be detected as early as February 2007. In addition, the chapter analyzes
the role of global market conditions to help determine when governments should begin to
exit their extraordinary public support measures.

The next book chapters narrow down on a number of (geographical) case studies during and
after the Global Financial Crisis:

Chapter V (first published in 2009) with Tao Sun examines financial stability issues that
arise from the increased presence of sovereign wealth funds (SWFs) in global financial
markets by assessing whether and how stock markets react to the announcements of
investments and divestments to firms by SWFs using an event study approach. Based on 166
publicly traceable events collected on investments and divestments by major SWFs during
the period from 1990 to 2009, results show that there was no significant destabilizing effect
of SWFs on equity markets, which is consistent with anecdotal evidence.

Chapter VI (first published in 2010) with Adolfo Barajas, Ralph Chami and Raphael
Espinoza focuses on the credit stagnation during the Global Financial Crisis in the Middle
Eastern and North African (MENA) countries from three analytical angles. First, it finds that,
similar to other regions and to its past history, a credit boom preceded the current slowdown,
and that a protracted period of sluggish growth was likely going forward. Second, it uncovers
a key role played by bank funding (deposit growth and external borrowing slowed
considerably) but whose effect was frequently dampened by expansionary monetary policy.
Third, bank-level fundamentals—capitalization and loan quality—helped to explain
differences in credit growth across banks and countries. The chapter concludes on what
policy measures could have been taken to revive credit growth in the MENA region.

Chapter VII (published in 2012) examines the case of Romania, which had been
significantly impacted by the Global Financial Crisis. The chapter looks at foreign bank
deleveraging and examines how Romania’s asset prices have been impacted from European
crisis spillovers.

With European banks at the centre during the Global Financial Crisis, chapter VIII
(published in 2013) with Nadege Jassaud looks at the balance sheet repair of European banks
and the progress with bank restructuring at that time. The chapter specifically discusses the
 12

hurdles that impair restructuring and resolution that were present during the 2012/ 2013
European FSAP.

The next three chapters focus on the important aspect of bank stress testing. The Global
Financial Crisis has clearly shown the shortcomings of stress testing but also its importance
in examining financial stability and helping form policy recommendations.

Neglecting liquidity risks has come at a substantial price during the Global Financial Crisis.
Over the last decade, large banks became increasingly reliant on short-term wholesale
funding to finance their rapid asset growth. At the same time, funding from non-deposit
sources (such as commercial paper placed with money-market mutual funds) soared. With
the unfolding of the Global Financial Crisis, when uncertainties about the solvency of certain
banks emerged, various types of wholesale funding market segments froze, resulting in
funding or liquidity challenges for many banks. In the light of this experience, there is a
widespread consensus that banks’ extensive reliance on deep and broad unsecured money
markets is to be avoided. In this regard, chapter IX (published in 2012) with Christian
Schmieder, Benjamin Neuendorfer, Claus Puhr and Stefan Schmitz presents a framework to
run system-wide, balance sheet data–based liquidity stress tests. The liquidity framework
includes a module to simulate the impact of bank-run type scenarios, a module to assess risks
arising from maturity transformation and rollover risks, and a framework to link liquidity and
solvency risks.

Stress testing has become an essential and very prominent tool in the analysis of financial-
sector stability and the development of financial-sector policy, but in itself can have only a
limited impact unless it is tied to action. Stress testing and related simulations can serve
various functions, such as the calibration of the relative importance of various risk factors,
and the assessment of banks’ capital needs when they are already under stress. The
publication of stress-test results with enough supporting material (including on the initial
condition of banks) can be helpful in reducing uncertainty; even banks that are revealed to be
relatively weak may benefit if the market paralysis engendered by great uncertainty is
relieved. But stress tests are of value mainly when they are followed up by concrete and swift
actions by the authorities (supervisory and others) and by bank managers that improve the
condition of banks and of banks’ clients. In this regard, chapter X (published in 2013) with
Daniel Hardy discusses the role of European-wide stress tests.

One of the challenges of financial stability analysis and bank stress testing is how to establish
scenarios with meaningful macro-financial linkages, i.e., taking into account spillover effects
and other forms of contagion. Chapter XI (first published in 2014) with Ferhan Salman and
Christian Schmieder presents an analytical approach to simulate the potential impact of
spillover effects based on the “traditional” design of macro-economic stress tests.
Specifically, the chapter examines spillover effects observed during the financial crisis and
simulates their impact on banks’ liquidity and capital positions. The outcome suggests that
 13

spillover effects have a highly non-linear impact on bank soundness, both in terms of
liquidity and solvency.

The final two chapters XII and XIII examine some aspects of debt sustainability and
sovereign debt restructuring, namely the domestic financial stability implications from a
face-value preserving maturity extension, so-called reprofiling as well as the role of banks’
home bias for sovereign debt sustainability:

Chapter XII (published in 2014) analyses the experience with past cases of reprofiling to
assess whether they had destabilizing effects on the domestic banking system. It examines
several past maturity extensions (Cyprus, Jamaica, Pakistan, and Uruguay) and finds that
destabilizing effects did not materialize. Several factors contributed to the generally
successful outcomes under maturity extensions: financial stability concerns were taken into
account in the design of the restructuring and program strategy; banks mainly held their
sovereign assets as held-to-maturity (HTM); a reprofiling was not assessed to be an
impairment event requiring a write down of these assets (e.g., Cyprus, Jamaica); regulatory
incentives for banks were provided (e.g., Jamaica or Uruguay); capital and liquidity support
mechanisms were established (e.g., Jamaica) or were present (Cyprus); the amount of bank
holdings of sovereign bonds in most cases was not very large; and some forbearance was
used. The Jamaican case illustrates how a restructuring was designed to be light in order to
ensure a limited impact on the financial system. The chapter also proposes possible measures
that could help protect the banking system during a reprofiling and encourage participation
by domestic banks in the exchange. Finally, the chapter examines financial stability
implications of a creditor bailout. Although a reprofiling may have some disruptive effects, a
bailout does not necessarily insulate the domestic financial system, as the Greek experience
demonstrates.

Motivated by the recent increase in domestic banks’ holdings of domestic sovereign debt
(i.e., home bias) in the European periphery, chapter XIII (published in 2015) with Tamon
Asonuma and Said Bakhache analyzes implications of banks’ home bias for the sovereign’s
debt sustainability. The main findings, based on a sample of advanced (AM) and EM
economies, suggest that home bias generally reduces the cost of borrowing for AMs and EMs
when debt levels are moderate to high. A worsening of market sentiments appears to dimish
the favorable impact of home bias on cost of borrowing particularly for EMs. In addition, for
AMs and EMs, higher home bias is associated with higher debt levels, and less responsive
fiscal policy. The findings suggest that home bias indeed matters for debt sustainability:
Home bias may provide fiscal breathing space, but delays in fiscal consolidation may
actually delay problems until debt reaches dangerously high levels.
 14

I. THE TRANSMISSION OF LIQUIDITY SHOCKS DURING THE CRISIS, WITH NATHANIEL

FRANK AND BRENDA GONZALEZ-HERMOSILLO, 20081

We examine the linkages between market and funding liquidity pressures, as well as their
interaction with solvency issues surrounding key financial institutions during the 2007
subprime crisis. A multivariate GARCH model is estimated in order to test for the
transmission of liquidity shocks across U.S. financial markets. It is found that the interaction
between market and funding illiquidity increases sharply during the recent period of
financial turbulence, and that bank solvency becomes important.

A. Introduction

The rapid transmission of the U. S. subprime mortgage crisis to other financial markets in the
United States and abroad during the second half of 2007 raises some important questions.2 In
particular, through which mechanisms were the liquidity shocks transmitted across U.S.
financial markets during this period? What was the relative strength of these potential
linkages? Did the episode of funding illiquidity in structured investment vehicles (SIVs) and
conduits turn into an issue of bank insolvency? Conceptually, a number of transmission
mechanisms are likely to have been established during the recent period of turbulence, either
through increased market illiquidity, funding illiquidity, or even default risks. The relative
strength of the interaction among these factors during the subprime crisis of 2007 is an
empirical question, which is analyzed below.

In general, the mechanisms through which liquidity shocks influence various markets may
operate through different channels during normal times than in the midst of an episode of
financial stress. During tranquil periods, market illiquidity shocks are typically short-lived, as
they create opportunities for traders to profit and, in doing so, provide liquidity and
contribute to the price-discovery process. However, during periods of crisis, several
mechanisms may amplify and propagate liquidity shocks across financial markets, creating
systemic risks. These mechanisms can operate through direct linkages between the balance
sheets of financial institutions, but also indirectly through asset prices. The existing literature
examining these connections include Adrian and Shin (2007), Cifuentes, Ferrucci, and Shin
(2005), and Brunnermeier and Pedersen (2008). Leverage in the models presented in these
studies is procyclical and can amplify the financial cycle. Specifically, asset price movements
are set in motion when financial institutions face marked-to-market price declines. As a
consequence, positions are deleveraged, and if the value of the corresponding assets is

1
This chapter is based on Frank, González-Hermosillo and Hesse (2008).
2
The events that led to the U.S. subprime crisis are discussed in the IMF Global Financial Stability Report
(2008).
 15

significantly affected, the creditworthiness of the respective institutions will deteriorate due
to rising risk of default.

This paper extends and discusses in more detail the results presented in Chapter 3 of the IMF
Global Financial Stability Report (2008). In particular, some methodological refinements are
introduced, which produce more accurate estimates of the transmission of liquidity shocks
during the subprime crisis in 2007. The estimation is conducted by applying a multivariate
GARCH specification, whereby the Dynamic Conditional Correlation (DCC) model
developed by Engle (2002) is adopted. This allows us to evaluate the transmission of the
liquidity shocks that spread from U.S. conduits and banks’ off-balance sheet SIVs to other
credit and equity markets in the United States. Furthermore, this GARCH framework allows
for the modeling of the heteroscedasticity exhibited by the data, in addition to interpreting the
conditional variance as a time-varying risk measure. Following Cappiello, Engle and
Sheppard (2006), the DCC specification is modified to account for possible structural breaks
in the unconditional correlations amongst the variables. The spillovers of U.S. liquidity
shocks to international money markets and emerging market countries, using the same
techniques, is discussed in Frank, González –Hermosillo and Hesse (2008).

The findings suggest that during the recent crisis period the interaction between market and
funding liquidity sharply increases in U.S. markets while a proxy for bank solvency issues
become important. In contrast, these transmission mechanisms were largely absent before the
onset of financial turbulences in July 2007. The introduction of the structural break in the
long-run mean of the conditional correlations between the liquidity and other financial
market variables is statistically significant and further strengthens these conclusions. Finally,
we quantify the estimation uncertainty surrounding the correlation processes by providing
estimates of their respective confidence intervals.

This paper makes several important contributions to the emerging literature on liquidity
shocks during the recent subprime crisis. First, as far as we can tell, this is the first attempt to
model empirically the transmission of liquidity shocks across U.S. financial markets during
the recent period of financial stress. Second, the GARCH model explicitly addresses the links
between market and funding liquidity effects and the dynamics of bank insolvency pressures
among the largest complex financial institutions. This connection is of critical importance
since this latest crisis, which in its early stages was perceived as a temporary liquidity
episode, eventually metastasized into one of solvency for a number of major global banks.
Indeed, the subsequent write-downs and losses emanating from structured financial products,
required that banks raised significant amounts of new capital from other investors such as
sovereign wealth funds. Third, we argue that the DCC model by Engle (2002) can potentially
lead to an understatement of the duration and severity of the period of market stress. This is
because the autoregressive model parameterization implies that the conditional correlations
are mean reverting to their constant long-run unconditional average. Using the DCC
specification by Cappiello, Engle and Sheppard (2006) allows us to explicitly model the
 16

subprime crisis as a structural break in the data generating processes, rather than a transitory
shock.

The rest of the paper is organized as follows. Section B reviews the salient features of the
recent turmoil in global financial markets for clues as to how the liquidity shocks may have
transmitted across differing asset classes. Section C details the data selection and Section D
discusses the empirical methodology. Section E examines the main results. Finally, Section F
concludes.

B. Transmission of Spillovers during the Subprime Crisis

This section gives a brief overview how the U.S. financial market segments that are relevant
for the empirical analysis of the transmission of spillovers were affected during the recent
subprime crisis. But before focusing attention on the mechanisms through which liquidity
shocks were actually transmitted, market and funding liquidity need to be defined. Consistent
with the existing literature, market liquidity is an asset-specific characteristic measuring the
ease with which positions may be traded without significantly affecting their corresponding
asset price. In contrast, funding liquidity refers to the availability of funds such that a solvent
agent is able to borrow in the market in order to service his obligations.

An important determinant of market liquidity is the completeness and the symmetry of

information with regard to the underlying asset. Other factors include the trading venue and
the characteristics of the mechanisms for exchange. Thus, for example, securities traded in
over-the-counter (OTC) markets may be subject to market illiquidity because of the absence
of market-makers and a central clearing house, potentially impairing the price discovery
process by limiting the potential matching of buyers and sellers. Also important in this
respect is the absorptive capacity of market-makers, and the depth of secondary markets. In
this context, many complex structured products are typically custom-designed. Thus, high
issuance of these heterogeneous OTC-traded assets does not necessarily imply abundant
resale possibilities in secondary markets.

Funding liquidity risk, as the risk that funds may not be available to a solvent agent, is
implicitly embedded in many forms of financial intermediation, but is of limited relevance
during times of tranquility. In contrast, during periods of crisis, vulnerabilities to these risks
increase significantly as outlined below.

The most recent episode of turbulence, beginning in the summer of 2007, started with
deteriorating quality of U.S. subprime mortgages, a credit, rather than a liquidity event.3 We

3
See Kiff and Mills (2008) and Dell’Ariccia, Igan and Laeven (2008) for details on the structure of the U.S.
subprime mortgage market and the deterioration of lending standards.
 17

argue that its propagation across different asset classes and financial markets is attributable to
an amplification mechanism due to asymmetric information resulting from the complexity of
the structured mortgage products and, subsequently, as a result of a more widespread re-
pricing of risk which may have taken the form of a decrease in global investors’ risk appetite
(see Gonzalez-Hermosillo (2008)).

Increased delinquencies on subprime mortgages, driven by rising interest rates and falling
house prices, resulted in uncertainty surrounding the value of a number of structured credit
products which had these assets in their underlying portfolios. As a result, rating agencies
downgraded many of the related securities and announced changes in their methodologies for
rating such products, first in mid-July but then again in mid-August and in mid-October.
Meanwhile, structured credit mortgage-backed instruments measured by the ABS indices
(ABX) saw rapid declines, and the liquidity for initially tradable securities in their respective
secondary markets evaporated. The losses, downgrades, and changes in methodologies
shattered investors’ confidence in the rating agencies’ abilities to evaluate risks of complex
securities, a result of which, investors pulled back from structured products in general.

It soon became apparent that a wide range of different financial institutions had exposures to
many of these mortgage-backed securities, often off-balance sheet entities such as conduits
or structured investment vehicles (SIVs). The SIVs or conduits were funded through the
issuance of short-term asset-backed commercial paper (ABCP) in order to take advantage of
a yield differential resulting in a maturity mismatch. Due to the increasing uncertainty with
regard to their exposure to and the value of the underlying mortgage-backed securities,
investors became unwilling to roll over the corresponding ABCP.

As with most other OTC products, measures of market liquidity of these assets are difficult to
obtain due to the lack of a centralized exchange which publishes prices and trading volumes.
In this context, Caruana and Kodres (2008) point out that the average maturity of outstanding
ABCP shortened from 24 to 18 days during the summer of 2007, and that the amount of
outstanding ABCP declined by approximately $300bn between early August and early
November in the U.S. market alone, suggesting the ABCP market became less liquid.

As the problems with SIVs and conduits deepened, banks came under increasing pressure to
rescue those that they had sponsored by providing liquidity or by taking their respective
assets onto their own balance sheets. As a result, the balance sheets of those financial
institutions were particularly strained by this reabsorption, which in addition was amplified
due to declining asset values. A further strain on banks’ balance sheets came from
warehousing a higher than expected amount of mortgages and leveraged loans, the latter
usually passed on to investors in order to fund the highly leveraged debt deals of private
equity firms. Both the market for mortgages and leveraged loans dried up from the collapse
of transactions in the mortgage-related securitization market and collaterized loan obligations
(CLOs). Banks also felt obliged to honor liquidity commitments to alternative market
 18

participants, such as hedge funds and other financial institutions, that also suffered from the
drain of liquidity. With regard to alternative channels of liquidity provision, stress in the FX
swap markets and the negative reputational signal resulting from using the Fed discount
window limited options further.

Consequently, the level of interbank lending declined both for reasons of liquidity and credit
risk. The former is based on a prudency motive whereby banks hoarded liquid assets in order
to insure themselves again contingent liabilities. In contrast, the latter was due to uncertainty
with regard to the mortgage exposure of counterparties and the inability to value their
respective assets. Subsequently, money markets were affected especially in advanced
countries in the form of a widening of the Libor–overnight index swap (OIS) spreads, which
in turn led to increased funding costs.

As turbulence related to the U.S. subprime mortgages heightened, financial markets more
generally showed signs of stress, as investor preference moved away from complex
structured products in a flight to liquidity. Subsequently, positions were shifted in order to
invest in only the safest and most liquid of all assets, such as U.S. Treasury bonds.
Furthermore, hedge funds that held asset backed securities and other structured products
were burdened by increased margin requirements, driven in turn by greater market volatility.
As a consequence, they attempted to offload the more liquid parts of their portfolios in order
to meet these margin calls and also respond to redemptions by investors. As argued by
Khadani and Lo (2007), quantitatively driven hedge funds were especially engaged in
liquidation sales across different asset classes, thus leading to a transmission of market stress
As a result, trading volumes and numbers of trades in both the bond and the stock markets in
the developed and emerging countries increased markedly, whilst the liquidity surrounding
structured investments evaporated.

Finally, the evident deterioration of market and funding liquidity conditions had implications
with regard to the solvency position of banks for several reasons. First, financial institutions
saw a decline in the values of the securitized mortgages and structured securities on their
balance sheets, which in turn resulted in extensive write-downs. Second, funding liquidity
pressures forced rapid deleveraging during this period, further depressing asset prices. Third,
funding costs increased due to rising money market spreads, which was amplified by the fact
that many financial institutions had become increasingly reliant on funding from wholesale
money markets. Jointly, these pressures resulted in a decline in the capital ratios throughout
the banking sector, and as a result of which credit default swap (CDS) spreads increased
significantly across the industry during the crisis.

The transmission mechanisms of liquidity shocks across differing U.S. financial markets
outlined so far have been described as being unidirectional and sequential. But during periods
 19

of financial stress, as witnessed during the subprime crisis, re-enforcing liquidity spirals may
be observed.4 As discussed above, market illiquidity can turn into funding illiquidity, as
banks are forced to reabsorb their SIVs onto their balance sheets. Alternatively, infrequent
trading and a limited price discovery process can cause increased volatility, which in turn
will raise margins and needed collateral. Thus, this reduces the leverage and the funding
possibilities which are open to traders. Furthermore, market illiquidity in complex structured
products could lead to the inability of market participants to assess the fair value of assets,
such as when the French bank BNP Paribas announced in August 2007 it would refuse to
accept withdrawals from three of its investment funds.

Funding illiquidity can also lead to market illiquidity, whereby the former forces financial
agents to sell securities at fire-sale prices, resulting in a sharp decline in asset prices and
further deleveraging (Bernardo and Welch, 2004). Subsequently, the absorptive capacity and
liquidity of secondary markets, especially if the assets are complex securities which are only
sold over-the-counter (OTC), may become exhausted. In addition, financial institutions that
operate across multiple markets could be affected when stress in specific funding markets
spills over to market illiquidity in related areas. One example is when European banks in late
2007 required dollar funding through foreign exchange swaps, but due concerns over
counterparty credit risk, liquidity, typically obtained in the underlying swap market dried up.

It has been argued that these spillovers has been facilitated by recent structural changes in the
financial markets and by financial innovation. In this context, banks have become
increasingly reliant on wholesale funding and short term liquidity lines. Also, increased
complexity of securities has led to great information asymmetries among market participants.
Favorable macroeconomic conditions, especially low interest rates in recent years, have
increased investors’ risk appetite and the demand for high yield products in order to satisfy
profit margins. Finally, increased correlations between returns of differing asset classes due
to algorithmic trading, such as by quantitative hedge funds, has heightened the vulnerability
with regard to the transmission of illiquidity.

The possibility of re-enforcing liquidity spirals, in addition to the operation of spillovers

across the five markets outlined above, is important for the model selection in the empirical
analysis which is set out in section D. The presence of potential multi-directionality of the
propagation motivates us to conduct estimation using a multivariate DCC GARCH
specification. This allows us to model the correlation dynamics between asset classes such
that we can evaluate whether different markets co-moved to a greater extent during the
subprime crisis of 2007.

4
These are discussed in more detail in the Global Financial Stability Report (2008) as well as Brunnermeier and
Pedersen (2008).
 20

C. Data

The econometric methodology, which is discussed in more detail in section D, is chosen to

shed light on the transmission of recent liquidity shocks across the five U.S. financial
markets introduced above. The model uses a system of five corresponding variables to
summarize key linkages, which act as proxies for overall market liquidity, funding liquidity,
default risk and volatility.

Firstly, funding liquidity conditions in the ABCP market segment are modeled by the spread
between the yield of 3-month ABCP and that of 3-month U.S. Treasury bills. The second
variable examined in the system is the spread between the 3-month U.S. interbank Libor rate
and the overnight index swap (OIS) which measures bank funding liquidity pressures.
Thirdly, S&P 500 stock market returns are included into the reduced form model, whereby in
its second moment it serves as a proxy for market volatility. Furthermore, the spread between
the 2-year on-the-run and the off-the-run U.S. Treasury bond yields captures overall market
liquidity conditions.5 Finally, default risk of banks is modeled by the credit default swap
spreads of 12 large complex financial institutions.6

The data analyzed in this paper constitute a simplification of the dynamics that may occur
during periods of stress. For example, in practice, the widening of the ABCP and LIBOR-
OIS spreads could also potentially reflect an unobserved component that represents changes
in the perceived credit risk of the collateral backing ABCP, and in that of banks. Similarly,
CDS prices and the credit premia implicit in LIBOR rates may also partly reflect additional
compensation for market participants’ risk appetite and overall uncertainty in the markets.
Disentangling these components is difficult, since they are not directly observable and can be
time-varying. Michaud and Upper (2008) find that credit risk measures have little
explanatory power for the day-to-day fluctuations in the LIBOR-OIS spread. However, the
Bank of England (2007) notes that credit concerns since October 2007 appear to account for

5
The “on-the-run” Treasury note is usually the most recently issued of a particularly liquid maturity and is used
for pricing other assets. An on-the-run Treasury bill becomes “off-the-run” when a new note is issued in that
maturity bracket. Other alternative measures of overall market liquidity were also examined, including the
spread between the 10-year and the 5-year on-the-run and off-the-run U.S. Treasury securities, and the spread
between the 10-year U.S. Treasury bond and other less liquid maturities. Overall, the findings were broadly in
line with the 2-year on-the-run spread. It should be noted though, as pointed out by Fleming (2003), that the
various measures are imperfect proxies of U.S. Treasury market liquidity but that the 5-year and the 2-year note
spreads showed the biggest increase during the 1998 LTCM crisis in response to a desire for investors to move
to the most liquid assets. The high demand for 5 and 2-year Treasury notes for potential repurchases suggests
this variable may capture some funding as well as market liquidity.
6
This variable was created by taking the unweighted daily average of the 5-year credit default swaps for the
following institutions: Morgan Stanley, Merrill Lynch, Goldman Sachs, Lehman Brothers, JP Morgan, Deutsche
Bank, Bank of America, Citigroup, Barclays, Credit Suisse, UBS, and Bear Sterns.
 21

a more significant portion of the LIBOR spreads. Frank, Hesse and Klueh (2008) decompose
the LIBOR spread into a liquidity and credit component and using a Markov-Switching
framework find that for both the US and Euribor LIBOR spread, the credit component
becomes more prevalent during the latter stages of the crisis.

The data sample encompasses January 3rd 2003 until January 9th 2008. We conduct unit root
tests for the crisis period and formally identify non-stationarity in the data. Thus, we first
difference the spreads, so that they can be applied to the estimation framework set out below.

In the Annex 1, Figure A1.1 illustrates the historical spreads for ABCP, CDS and Libor-OIS.
Between 2003 and the summer 2007, these exhibit approximate constancy. The LIBOR
spread remained at about 10 basis points, whereas both the ABCP and CDS stayed below 50
bp. Following the beginning of the crisis in July 2007, all spreads subsequently jumped up
and have remained broadly at these elevated levels. Annex Figure A1.2 represents the on-the-
run/off-the-run 5-year U.S. Treasury bond spread. Again, liquidity pressures became
apparent during the second half of 2007. Annex Figures A1.3 and A1.4 show the first
differences of selected variables. Whereas during the pre-crisis period the processes are
approximately constant, a structural break in their respective data generating processes is
evident at the onset of the recent financial turbulence.

D. Methodology

The estimation presented below is conducted within a multivariate GARCH framework,

which takes the heteroskedasticity exhibited by the data into account, in addition to providing
the natural interpretation of the conditional variance as a time-varying risk measure. In this
context, the Dynamic Conditional Correlation (DCC) specification by Engle (2002) is
adopted, which provides a generalization of the Constant Conditional Correlation (CCC)
model by Bollerslev (1990).7 These econometric techniques allow us to analyze the co-
movement of markets by inferring the correlations of the changes in the spreads discussed
above, which in turn is essential in understanding whether the recent episode of financial
distress has become systemic.

First, in our estimation, a VAR(1) filter is proposed in order to pre-whiten the returns series
Xt.

7
We initially estimated the CCC model as well but the assumption of constant conditional correlation among
the variables of interest is not very realistic especially in times of stress where correlations can rapidly change.
Therefore, the DCC model is a better choice since correlations are time-varying.
 22

The potential presence of common shocks motivates this pre-filtering as we are able to
condition on S&P 500 stock market returns, which we regard as an appropriate proxy for
changes in global risk appetite.

Second, the DCC model is estimated in a three-stage procedure. From the estimation above,
let rt denote an n x 1 vector of pre-whitened asset returns, exhibiting a mean of zero and the
following time-varying covariance:

Here, Rt is made up from the time dependent correlations and Dt is defined as a diagonal
matrix comprised of the standard deviations implied by the estimation of univariate GARCH
models, which are computed separately, whereby the ith element is denoted as hit . In other
words in this first stage of the DCC estimation, we fit univariate GARCH models for each of
the five variables in the specification. In the second stage, the intercept parameters are
obtained from the transformed asset returns and finally in the third stage, the coefficients
governing the dynamics of the conditional correlations are estimated. Overall, the DCC
model is characterized by the following set of equations (see Engle, 2002, for details):

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset
returns rt. As defined above, Rt is the time varying correlation matrix and is a function of Qt,
which is the covariance matrix. In the matrix Qt, ι denotes a vector of ones, A and B are
square, symmetric and  is the Hadamard product. Finally, λi is a weight parameter with the
contributions of Dt21 declining over time, while κi is the parameter associated with the
squared lagged asset returns.

In order to quantify the estimation uncertainty of the results provided below, confidence
bands around the conditional correlations are reported. In this context, two techniques are
available. First, Monte Carlo Methods may be applied, whereby the first two moments of the
original DCC parameters are used to simulate their respective empirical distribution, under
the assumption of joint normality. Secondly, non-parametric bootstrapping is conceptually
similar in obtaining the required parameter distribution. Due to the time dependence in the
data, a circular block bootstrap, as proposed by Politis and Romano (1992), is to be
implemented for resampling, whereby the trade-off between the approximation of the
observed data characteristics and the randomness of the replication mechanism is taken into
 23

account in the selection of the optimal block length. Finally, with both approaches, the 5th
and 95th percentiles are reported in order to make inference with regard to the parameter
uncertainty.

E. Results

As discussed above, the estimation is conducted using the DCC GARCH framework. Figure
1 provides selected implied conditional correlations from this model.8 The confidence bands,
which are derived using Monte Carlo simulations, are reported for a 95% level of
significance, and indicate a low degree of estimation uncertainty. There is clear evidence of
increased correlations across all spreads during the summer of 2007, implying heightened
interaction between market and funding liquidity, as well as solvency aspects becoming
important.

Figure 1. Selected Conditional Correlations

Correlation (ABCP, Libor) Correlation (ABCP, CDS)
0.7 0.3

0.6 0.25

0.5 0.2

0.4 0.15

0.3 0.1

0.2 0.05

0.1 0

0 -0.05
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Libor, CDS) Correlation (Returns, CDS)

0.35 0.05

0.25
-0.05

0.15

-0.15

0.05

-0.25
-0.05

-0.15 -0.35
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

8
These are also presented in chapter 3 of the IMF Global Financial Stability Report (2008) on the subprime and
credit crisis. In this paper, we use the on-the-run/off-the-run spread for the 2-year U.S. Treasury Bond, rather
than the 5-year equivalent.
 24

Two further points of interest are to be noted in figure 1. Firstly, there is evidence of strong
reversion by the correlations to their respective long run means, such as in the panels for the
ABCP/Libor and the ABCP/CDS spreads. Secondly, a large jump on July 24th 2007 is
observed simultaneously across all markets. These raise the question of whether the data
generating process underwent an unobserved structural shift in the levels of the correlations
during the US subprime crisis. This is of great importance as failing to model this break
would imply that the mean reverting drift is potentially spurious, as convergence would be
occurring towards an incorrect long run average. As a result, the DCC specification is
modified in order to account for a structural break in the unconditional correlations on this
date, as proposed in similar work by Cappiello, Engle and Sheppard (2006):

where Qt is the modified covariance matrix that governs the dynamics of the time- varying
correlation matrix Rt in the above standard DCC model, and both S1 and S2 are the new
correlation matrices of the residuals εt. The dummy variable function dt may take on differing
forms, depending upon the assumptions one is willing to make with regard to the transition
from a pre-crisis to a crisis period. For example, a step function with a break at τ may be
parameterized, or alternatively, a more gradual change such as in a linear form or one based
on a cumulative density function around the hypothesized break point can be specified. In
this paper we re-estimate the model under the following assumption using the linear
approach:9

We find that the null hypothesis of the constancy of the unconditional correlation is rejected
at below the 0.001% significance using a likelihood ratio test with k(k-1)/2 degrees of
freedom. As a result, we adopt the model as described in equation (4). It should be pointed
out that this is not inconsistent with the results from the Global Financial Stability Report
(2008). Rather, the methodological refinements, by allowing for structural breaks in the mean
of the conditional correlations, strengthen previous findings.

The introduction of the dummy variable has an effect on both the empirical results and on
their subsequent interpretation. As illustrated in Figure 1, even substantial shocks to the

9
Since we do not want to impose a specific date for the hypothesized break at the end of July 2008, the linear
approach allows us to capture a larger window of days for the structural break.
 25

conditional correlations seem to be transitory in the initial DCC specification. Following the
subprime crisis, interlinkages across markets increase, but due to the autoregressive nature of
the model parameterization, the respective correlations are pulled back to their long-run
unconditional means. As a result, there is the distinct possibility of spurious reversion and
understatement of the duration and the severity of the periods of market distress. In the
modified DCC model, we introduce a break in the unconditional correlations, which are
based on the standardized residuals of the respective sub-samples. Thus, mean reversion
occurs to a different level during the crisis following the structural shift in the data generating
process.

Figure 2. Conditional Correlations from Modified DCC Model

Correlation (ABCP, Libor) Correlation (ABCP, Returns)
0.7 0.2

0.6
0.1

0.5
0

0.4
-0.1
0.3

-0.2
0.2

-0.3
0.1

0 -0.4
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (ABCP, CDS) Correlation (ABCP, Two)

0.3 0.25

0.25

0.2 0.15

0.15

0.1 0.05

0.05

0 -0.05

-0.05

-0.1 -0.15
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008
 26

Figure 2. Conditional Correlations from Modified DCC Model (concluded)

Correlation (Libor, Returns) Correlation (Libor, CDS)
0.2 0.4

0.1
0.3

0
0.2

-0.1
0.1
-0.2

0
-0.3

-0.1
-0.4

-0.5 -0.2
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Libor, Two) Correlation (Returns, CDS)

0.25 0.1

0.2 0

0.15 -0.1

0.1 -0.2

0.05 -0.3

0 -0.4

-0.05 -0.5

-0.1 -0.6
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Returns, Two) Correlation (CDS, Two)

0.1 0.25

0
0.15

-0.1

0.05

-0.2

-0.05
-0.3

-0.4 -0.15
03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008 03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

In Figure 2 the results of the revised DCC specification are presented. Consistent with
previous findings, there is strong evidence of increased interaction between the proxies for
market and funding liquidity. In this context the implied correlations between the ABCP and
Libor spreads rise from a pre-crisis average of approximately 0.3 to above 0.5, a level at
which they remain. Furthermore, the linkages between these two funding liquidity measures
and the 2-year on-the-run/off-the-run spread jump from around zero to 0.2. This implies that
before the period of financial distress these markets did not move together in a statistically
significant way, unlike during the subprime crisis.

Furthermore, stronger interactions between the market liquidity in the bond market and the
stock market return volatility are evident with S&P 500 returns and the two year on-the-run
spread becoming more highly correlated among each other, as well as with all other
 27

variables. Finally, the co-movement between liquidity and solvency is sharply increased.
Before the hypothesized break date, changes in the CDS spreads remain approximately
uncorrelated with all other measures, whereby the magnitude of these correlations increase to
between 0.25 and 0.5 in absolute value.

Interestingly, the implied conditional correlations remain constant around their new mean for
most series during the crisis period, and exhibit only limited variation in the form of negative
shocks. Thus, this provides further support for the choice of the mean switching model
specification adopted here, as a potential spurious reversion to an incorrect long-run average
is avoided.

In summary, the five markets using Libor and ABCP spreads as proxies for funding liquidity,
the 2-year on-the-run spread capturing market liquidity and finally stock market returns and
the CDS spreads as proxies for stock market volatility and bank default risk, respectively,
have exhibited extraordinary co-movement during the US subprime crisis. While implied
correlations had been fairly small in the pre-crisis period, the results presented here suggest
that new channels of transmission of liquidity shocks were established during the second half
of 2007.

Furthermore, the results of a very pronounced interaction between market and funding
liquidity are consistent with the emergence of re-enforcing liquidity spirals during the crisis
period. Although the conditional correlations from the DCC GARCH model do not imply
directionality or causality, they leave open the possibility that the interrelationship between
both market and funding liquidity is dynamic and interdependent. Indeed, the events that
followed the onset of the crisis in late July are consistent with the view that a re-enforcing
liquidity spiral emerged.

On the one side of this liquidity spiral, financial institutions were exposed to refinancing
needs in the form of issuing ABCP, a situation where market illiquidity in complex
structured products led to funding illiquidity. In this regard, the results also show that
increased correlations between the ABCP and Libor spreads reduced the possibilities of
funding from the interbank money market, thus highlighting systemic risks. Though not
shown explicitly in the paper, on the other side of this spiral, many European banks that had
large exposures to U.S. asset-backed securities had difficulties accessing wholesale funding,
so that subsequent market illiquidity in different market segments was caused. Due to the
major importance of the interbank money market, central banks in turn intervened by
reducing interest rates and providing additional liquidity to the markets in order to reduce
pressures.10

10
Frank, Hesse and Klueh (2008) discuss this.
 28

In addition to the described period of illiquidity, the U.S. subprime crisis increasingly
became one of insolvency, as banks such as Northern Rock, IKB and Bear Stearns had to be
rescued. This is captured by the implied correlations between the CDS and other variables in
the DCC GARCH model, which show clear signs of a structural break during the crisis
period. Furthermore, these correlations have remained at elevated levels since then,
suggesting that solvency concerns remain an issue.

Finally, it is also shown that risk in U.S. stock markets, proxied by volatility in the S&P
index and market liquidity in U.S Treasury bonds were affected during these times of severe
stress. These transmission mechanisms were not restricted to the U.S. financial markets, but
were also observed across other advanced and key emerging market economies. In particular,
many of these markets abroad were also subject to heightened implied correlations between
funding and market liquidity, and their respective domestic stock and bond markets.11

F. Conclusion

This paper examined empirically the linkages between market and funding liquidity
pressures, as well as their interaction with solvency issues surrounding key financial
institutions during the 2007 subprime crisis. A multivariate DCC GARCH model was
estimated in order to test for the transmission of liquidity shocks across U.S. financial
markets. It is found that the interaction between market and funding illiquidity increased
sharply during the recent period of financial turbulence, and that bank solvency became
important. In contrast, many of the transmission channels were not present before the onset
of the crisis.

The DCC GARCH specification which was adopted is based on the novel approach by
Cappiello, Engle and Sheppard (2006) which allows us to model the subprime crisis
explicitly as a structural break in the data generating processes for the time-varying
covariances across markets, rather than a transitory shock.

The financial turbulence that originated in U.S. financial markets has so far been very
protracted. What started out as a liquidity crisis, turned into a solvency issue. Indeed, a
number of major central banks have intervened heavily in order to maintain the stability of
the global financial system. Many of the largest complex financial institutions have had to
strengthen their balance sheet positions through capital injections from other investors. The
analysis presented here suggests that increasing financial integration and innovation can
make market and funding liquidity pressures readily turn into issues of insolvency.

11
These linkages are examined in Frank, Gonzalez-Hermosillo and Hesse (2008).
 29

References

Adrian, Tobias and Hyun Song Shin, 2007, "Liquidity and Financial Contagion," Financial
Stability Review - Special issue on liquidity, Banque de France, No. 11, December
2007.

Bank of England, 2007, “Markets and Operations,” Bank of England Quarterly Bulletin - Q4,
Vol. 47, No. 4.
Bernardo, Antonio E., and Ivo Welch, 2004, “Liquidity and Financial Market Runs,”
Quarterly Journal of Economics, Vol. 119 (February), pp. 135–58.

Bollershev, Tim, 1990, “Modelling the Coherence in Short-run Nominal Exchange Rates: a
Multivariate Generalized ARCH Approach,” Review of Economics and Statistics,
Vol. 72, pp.498–505.

Brunnermeier, Markus and Lasse Pedersen, 2008, “Market Liquidity and Funding Liquidity,”
Review of Financial Studies (forthcoming).

Capiello, Lorenzo, Robert F. Engle and Kevin Sheppard, 2006, “Asymmetric Dynamics in
the Correlations of Global Equity and Bond Returns,” Journal of Financial
Econometrics, Vol. 4, No. 4, pp. 537–72.

Caruana, Jaime and Laura Kodres, 2008, “Liquidity in Global Markets”, Financial Stability
Review - Special issue on liquidity, Banque de France, No. 11, December 2007.

Cifuentes, Rodrigo, Gianluigy Ferrucci and Hyun Song Shin, 2004, "Liquidity Risk and
Contagion", mimeograph, January 2004.

Dell’Ariccia, Giovanni, Deniz Igan and Luc Laeven, 2008, “Credit Booms and Lending
Standards: Evidence from the Subprime Mortgage Market,” mimeo (Washington:
International Monetary Fund)

Engle, Robert 2002, "Dynamic Conditional Correlation: A Simple Class of Multivariate

Generalized Autoregressive Conditional Heteroskedasticity Models", Journal of
Business & Economic Statistics, Vol. 20, pp. 339–50.

Fleming, Michael, 2003, “Measuring Treasury Market Liquidity,” Federal Reserve Bank of
New York Economic Policy Review, Vol. 9 (September), pp. 83–108.

Frank, Nathaniel, Brenda González-Hermosillo, and Heiko Hesse, 2008, “Transmission of

Liquidity Shocks: Evidence from the 2007 Subprime Crisis,” IMF Working Paper
2008/200 (Washington: International Monetary Fund). A shorter version is also
published in the Central Banking Journal, 2008, Vol. 19 (1), and VOX.
 30

Frank, Nathaniel, Heiko Hesse, and Ulrich Klueh, 2008, “Term Funding Stress and Central
Bank Interventions During the 2007 Subprime Crisis,” (Washington: International
Monetary Fund).

González-Hermosillo, Brenda, 2008, “Investor’s Risk Appetite and Global Financial Market
Conditions”, (Washington: International Monetary Fund).

International Monetary Fund, 2008, Global Financial Stability Report, World Economic and
Financial Surveys (Washington, April).

Khandani, Amir E., and Andrew Lo, 2007, “What Happened to the Quants in August 2007?”
MIT Working Paper (Cambridge, Massachusetts: Massachusetts Institute of
Technology,

Kiff, John and Paul Mills, 2008, “Money for Nothing and Checks for Free: Recent
Developments in U.S. Subprime Mortgage Markets,” IMF Working Paper 07/188
(Washington: International Monetary Fund).

Michaud, François-Louis, and Christian Upper, 2008, “What Drives Interbank Rates?
Evidence from the LIBOR Panel,” Bank for International Settlements Quarterly
Review (March), pp. 47–57.

Politis, D. N. and Romano, J. P. (1992): “A Circular Block Re-sampling Procedure for

Stationary Data”, In: LePage, R., Billard, L.: “Exploring the Limits of the Bootstrap”,
Wiley & Sons, New York, pp. 263–70.
 31

Annex 1. Figures

Figure A1.1. Aggregate Bank Credit Default Swap Rate and Selected Spreads
Aggregate Bank Credit Default Swap Rate and Selected Spreads
(In basis points) (In basis points)
350 350

300 300

250 Asset-backed commercial 250

paper spread1

200 200

150 150

LIBOR spread3
100 100

50 Credit default swap2 50

0 0
1/1/2003 7/1/2003 1/1/2004 7/1/2004 1/1/2005 7/1/2005 1/1/2006 7/1/2006 1/1/2007 7/1/2007

Sources: Bloomberg L.P.; and IMF staff estimates.

1
Spread between yields on 90-day U.S. asset-backed commercial paper and on three-month U.S. Treasury bill.
2
The unweighted daily average of the five-year credit default swaps for the following institutions: Morgan Stanley, Merrill Lynch, Goldman
Sachs, Lehman Brothers, JPMorgan, Deutsche Bank, Bank of America, Citigroup, Barclays, Credit Suisse, UBS, and Bear Stearns.
3
Spread between yields on three-month U.S. LIBOR and on three-month U.S. overnight index swap.

Figure A1.2. On-the-Run/Off-the-Run

On-the-Run/Off-the-Run Five-Year
Five-Year U.S. Treasury U.S.
Bond Spread
1 Treasury Bond Spread1
0.10 0.10

0.08 0.08

0.06 0.06

0.04 0.04

0.02 0.02

0.00 0.00

-0.02 -0.02

-0.04 -0.04
1/2/2006 4/2/2006 7/2/2006 10/2/2006 1/2/2007 4/2/2007 7/2/2007 10/2/2007 1/2/2008

Source: Bloomberg L.P.

1
Spread between yields on five-year off-the-run bond and on five-year on-the-run bond.
 32

Figure Spreads
United States: Selected A1.3. United States: Selected Spreads
(First difference; in basis(First
points)difference; in basis points)
80 80
Asset-backed commercial paper spread 1/
60 60
LIBOR spread 2/
40 40

20 20

0 0

-20 -20

-40 -40

-60 -60
1/3/2006 3/28/2006 6/20/2006 9/12/2006 12/5/2006 2/27/2007 5/22/2007 8/14/2007 11/6/2007

Sources: Bloomberg L.P.; and IMF staff estimates.

1
Spread between yields on 90-day U.S. asset-backed commercial paper and on three-month U.S. treasury bill.
2
Spread between yields on three-month U.S. LIBOR and on three-month U.S. overnight index swap.

FigureUnited
A1.4.States:
United
S&P States: S&P
500 Stock 500 Stock
Market Market
Returns Returns
and Credit and
Default Credit
Swap Default
(CDS)
1 Swap (CDS)1
(First difference; in basis points)
(First difference; in basis points)
15 15
Stock market returns
10 CDS 10

5 5

0 0

-5 -5

-10 -10

-15 -15
1/3/2006 3/23/2006 6/12/2006 8/30/2006 11/17/2006 2/6/2007 4/26/2007 7/16/2007 10/3/2007 12/21/2007
Sources: Bloomberg L.P.; and IMF staff estimates.
1
CDS is calculated as an unweighted daily average of the five-year credit default swaps for the following institutions:
Morgan Stanley, Merrill Lynch, Goldman Sachs, Lehman Brothers, JP Morgan, Deutsche Bank, Bank of America, Citigroup,
Barclays, Credit Suisse, UBS, and Bear Sterns.
 33

II. THE EFFECTIVENESS OF CENTRAL BANK INTERVENTIONS DURING THE FIRST PHASE
OF THE SUBPRIME CRISIS, WITH NATHANIEL FRANK, 200912

This paper provides evidence that central bank interventions had a statistically significant
impact on easing stress in unsecured interbank markets during the first phase of the
subprime crisis which began in July 2007. Extraordinary liquidity provisions, such as the
Term Auction Facility by the Federal Reserve, are analyzed. First a decomposition of the
Libor OIS spread indicates that credit premia increased in importance as the crisis
deepened. Second, using Markov switching models, central bank operations are then
graphically associated with reductions in term funding stress. Finally, bivariate VAR and
GARCH models are adopted to econometrically quantified these impacts. While helpful in
compressing Libor spreads, the economic magnitudes of central interventions have overall
not been very large.

A. Introduction

Following the onset of the subprime crisis in July 2007, central banks have been at the center
of the subsequent policy response to alleviate market dislocations due to the financial
turbulence. Whilst the origins of the crisis can be traced back to a combination of imbalances
in the global economy, structural weaknesses in the financial system and a severe relaxation
of lending standards in the presence of over-abundant macro-liquidity, one of its main
manifestations has been the partial dysfunction of the interbank money markets (see Frank
and others (2008) for further discussion). Due to this unprecedented period of stress central
banks have engaged in equally unprecedented liquidity providing operations, the nature and
effectiveness of which are the focus of this paper. Within this framework of liquidity
management one of the most important indicators of funding pressures has been the spread
between lending in unsecured interbank markets over expected overnight rates.

In some recent work, Taylor and Williams (2008) propose a model for interest rate
determination, and they hypothesize that central bank policies such as the Term Auction
Facility (TAF) by the Federal Reserve will not materially reduce stress levels as measured
through Libor spreads, as only a net injection of liquidity in the form of total supply of
central bank reserves could potentially have an effect. Furthermore, they conduct a series of
econometric tests and find that the TAF has indeed been largely ineffective in reducing term
funding pressures. In contrast, Michaud and Upper (2008) reach different conclusions. Based
on a comparison of the timing of central bank actions and major market moves, they show
that extraordinary liquidity operations have contributed to a substantial compression of Libor
spreads, whilst credit default swap (CDS) premia for banks did not appear to react in

12
This chapter is based on Frank and Hesse (2009).
 34

systematic ways. In addition, Aït-Sahalia and others (2009) find for a number of advanced
economies and using an event study methodology that government interventions had a
significant impact on the Libor spreads but this effect become smaller the more prolonged the
crisis became.13

This paper builds on our initial work in IMF (2008) and extends it and the existing literature
along several dimensions. Firstly, we decompose the Libor-Overnight Index Swap (OIS)
spread into a liquidity component and into one reflecting credit or counterparty risks.14 This
is of importance as we believe that central bank interventions are more effective in
addressing the former rather than the latter, a distinction which is not made by Taylor and
Williams (2008). Secondly, we adopt a Markov switching framework in order to identify
periods of differing levels of stress in the interbank money markets. Subsequently, the
corresponding dates of regime transitions are related to major central bank announcements
and policy implementation. Thirdly, we improve the univariate analysis by Taylor and
Williams (2008) by explicitly taking into account the partial co-movement between rates in
different currencies. A bivariate VAR econometric model of the Euro and U.S. dollar Libor-
OIS spreads is specified in order to test whether central bank operations have lowered stress
in term funding markets. Finally, we adopt bivariate GARCH models in order to examine the
impact of central banks’ interventions on the volatility of the Libor spreads besides the
level-effect.

The motivation for including the bivariate analysis is due to the fact that Libor fixings in
Euros and U.S. dollars display substantial interdependence, as funding conditions are
increasingly of a global nature, in particular during periods of crisis. Similarly,
extraordinary changes to central bank liquidity operations in one currency have the
potential to change funding conditions in another, as they are transmitted through the
default probability of counterparties, conditions in foreign exchange markets and changes
in overall risk aversion. Finally, some central bank measures put in place during the
recent turmoil have explicitly targeted frictions in global liquidity allocation, most
notably the extension of the TAF by the Federal Reserve through swap arrangements
with the European Central Bank (ECB) and the Swiss National Bank (SNB).

In the empirical analysis of this paper much focus is placed on the effectiveness of two
specific policy tools. Firstly, liquidity injections by the ECB through supplementary 90-day

13
Further discussion and findings in this area are provided by IMF (2008), where the underlying
dynamics of the volatility of Euro and U.S. term spreads are modeled by employing univariate
GARCH specifications. Focus is placed on intervention instruments already at the disposal to central
banks during the onset of the crisis, whereby liquidity injections over and above the neutral level
needed to just fulfill reserve requirements are analyzed.
14
The Libor-OIS spread is used as a proxy for the interbanking money market stress during the crisis. For more
details, see Section 2.
 35

long term refinancing operations (LTROs) in excess of the benchmark allotments are used.15
Secondly, the impact of both the introduction of the Term Auction Facility and the effect of
cuts in the Federal funds and the discount rates by the Fed are quantified. In our study, we
only examine the first phase of the subprime crisis from the summer of 2007 to April 2008
and do not include the Lehman Brothers collapse as well as the various new interventions
tools by the Fed and the ECB.

What are our results? We find that for the early phases of the subprime crisis which began in
July 2007, the rise in the Libor-OIS spread is initially attributed to funding illiquidity,
whereas the credit risk component becomes increasingly important up until the rescue of
Bear Stearns in March 2008. Furthermore, by employing Markov switching models it is
graphically shown that relaxation of money market stress can be related to selected central
bank interventions, whereby the announcement of the LTROs and the TAF seem most
effective in reducing the overall Libor-OIS spreads. Finally, both the bivariate VAR and
GARCH models confirm that the announcements of the TAF and LTROs have a statistically
significant impact on both the level and volatility changes of the Libor spreads. But the
economic magnitudes are not very large, which is supported by the fact that central banks’
actions during the first phase of the subprime crisis, while successful in helping to bring
down Libor spreads, have not led to an end of the liquidity crisis and a containment of the
solvency concerns looming at the time.

This paper is organized as follows. In Section B, major developments in unsecured term

funding markets during the crisis and subsequent central bank interventions to ease
liquidity shortfalls are reviewed. In Section C, estimation results are presented for the
Markov-Switching and VAR models, while Section D discusses the GARCH framework.
Finally, the conclusion and policy implications follow in Section E.

B. Review of Developments and Policy Interventions

Background on term funding markets and Libor fixing

In our empirical analysis below, we focus on the impact of central bank intervention on the
spread between the 3-month London Interbank Offer Rate, or Libor, and the OIS rate. In
essence, the Libor-OIS spread is a measure of the premium that banks pay when borrowing
funds for a pre-determined period relative to the expected interest cost from repeatedly
rolling over funding in the overnight market. In times of sufficient liquidity and in the
absence of market dislocations, these two measures should be close substitutes, as implied by
the expectations hypothesis, such that the interest rate paid on term bank deposits ought to
bear a close relationship with the expectation of the compounded overnight rates over the

15
The benchmark allotment is the ECB's projection of the liquidity provision needed to fulfill its reserve
requirements.
 36

same horizon. During periods of crisis, the widening Libor-OIS spread provides an
appropriate proxy for interbank money market stress in the form of liquidity and credit
premia, quantifying the unwillingness of banks to extend unsecured loans. This is because
the OIS is tied to the Federal funds rate and exhibits only limited credit risk as, like in the
case of most interest rate swaps, no principal is exchanged.

Further motivation for analyzing the effect of central bank intervention on the Libor-OIS
spread in this paper draws on the importance of the Libor instrument, both in terms of the
functioning of securities markets and concerning its macrofinancial implications. With regard
to the former, Libor is a measure of the cost at which banks may borrow unsecured funds,
which over the past decade has become of increasing significance in the light of greater
reliance on the wholesale interbank money markets, as discussed in IMF (2008).
Furthermore, it is used as a risk free rate in discounting and thus pricing derivative contracts
such as forward rate agreements, interest rate swaps and swaptions. Hull (2005) points out
that Libor is the preferred reference rate rather than the yield implied by government
securities as for tax and institutional reasons demand for Treasuries is increased, in turn
implying an interest rate which is too low. As a result, Libor is important in ensuring market
efficiency through accurate pricing of assets and in determining the funding costs of major
financial institutions. In addition, there are also macroeconomic considerations, as the
interbank money markets constitute an important channel for monetary policy transmission.

The Libor rate, the cost of unsecured lending between banks, is set on a daily basis and is
published at 11 a.m. by the British Bankers Association. It is constructed using a survey from
banks that comprise the Libor panel, composition of which is subject to change over time and
which may include foreign banks operating in London. The calculation of the reference rate
is based on the average of quotes rather than that of actual trades, whereby the upper and
lower quartiles are omitted in order to avoid manipulation. As part of this process Libor is set
for 15 maturities ranging from overnight to 12 months in ten different currencies.
It should be noted though that the Libor fixing mechanism exhibits limitations and that the
reference rate is merely a proxy for the interest charged by banks amongst each other.
Quantification of the actual interest rate is difficult as such trades are bilateral and are not
centrally recorded, thus providing no readily available data. Importantly, any quotes are
non-binding and may differ from the subsequent interest rates which are agreed upon.
Furthermore, concerns have been voiced as to whether the banks in the Libor panel have
incentives to make quotes in an accurate fashion, especially during times of financial stress.
Downward biases may be present due to signaling effects, whereby a high offer incurs
reputational damage as it indicates the need to attract significant interest payments. Finally,
there are incentives for banks to influence the Libor fixing process in order to affect the
pricing of securities and thus their respective book values. In response to these issues, the
tails of the quote distribution are discarded, implying that manipulation of the Libor rate
would only be possible through widespread collusion amongst reporting banks. In addition,
transparency has been improved as part of new guidelines set by the British Bankers
 37

Association (see BBA (2008) for further details) by requiring individual financial institutions
to publish their offers.

Review of developments since July 2007

This subsection briefly reviews some of the major developments of the crisis since July 2007,
which are of importance with regard to the interbank money markets. As outlined in more
detail by Frank and others (2008), this period of financial turbulence was triggered by a
credit event, namely the bursting of the subprime mortgage bubble due to falling house prices
and the reversal of interest rates which had been previously at historical lows. Many financial
institutions exhibited exposures to mortgage-backed securities, often in the form of off-
balance sheet entities such as structured investment vehicles (SIVs). These SIVs or conduits
were funded through the issuance of short term asset-backed commercial paper (ABCP) in
order to take advantage of higher yields resulting from longer term investments, and thus
exhibited an inherent maturity mismatch. Due to the increasing uncertainty with regard to
their exposure to and the value of the underlying mortgage-backed assets, investors became
unwilling to roll over the corresponding ABCP. As the problems with the SIVs and conduit
facilities deepened, banks came under increasing pressure to rescue those that they had
sponsored by providing liquidity and by taking their respective assets onto their own balance
sheets.

Following this reabsorbtion of the SIVs, lending within the interbank money markets was
curtailed for reasons of liquidity and credit risk. With regard to the former, increases in the
Libor-OIS spreads, especially at longer dated maturities, are explained by the hoarding of
funding in order to cover further contingent liabilities following asset price declines,
subsequent marking-to-market of securities and forced liquidations. Concerning the latter,
rising credit concerns were priced into the Libor spread as interbank lending is unsecured and
whereby this counterparty risk arises due to the uncertainty of the banks' exposure to troubled
assets. An alternative explanation for the widening of the Libor spreads has been proposed by
Giavazzi (2008) who has put forward the notion of predatory banks. In a model of strategic
behavior amongst financial institutions there are two reasons why excess cash is not lent.
Firstly, if another bank was to fail, its assets could be bought at a depressed price following it
being placed into administration. Secondly, the probability of such an event occurring is
endogenously determined by the amount of liquidity available in the interbank money
markets, such that the optimal strategy may be to hoard any funds.

During the crisis the Libor fixing process itself was also affected. At times of the most
serious market dislocations and heightened risk aversion, term funding at longer dated
maturities was entirely unavailable. In addition to an increase in the level of Libor quotes, the
variance of the individual fixings made by banks also rose significantly. In this context it has
been argued that this was in part due to heterogeneity with regard to credit risk. Financial
institutions exhibited differing exposures to asset-backed securities and to contingent credit
 38

lines in the form of implicit guarantees for SIVs, whereby high liabilities towards these
entities induced increased upward pressure on the respective quotes.

Finally, due to the shortage of U.S. dollar liquidity following the reabsorbtion of conduits
and SIVs, European banks became increasingly engaged in FX swaps, whereby especially
Euro and Sterling were used as the funding currencies in such deals, as discussed by Baba
and others (2008). The spillovers from the interbank to the FX swap market led to a situation
whereby FX swap prices temporarily deviated from their covered interest parity condition,
which further highlighted the international interconnectedness of funding requirements by
banks. These linkages are of importance in the context of our paper as central banks
responded on December 12, 2007 by extending the Term Auction Facility by the Fed through
swap arrangements with the ECB and the Swiss National Bank in order to address these
foreign currency shortages.

Monetary policy implications

Whilst motivating this paper, it was previously argued that the dislocations in the interbank
money markets also have macroeconomic effects such as impairing the transmission
mechanism of monetary policy.

During tranquil times this process operates in that policy rates affect money markets, which
in turn determine the cost of lending to households and companies, and thus the level of
economic activity and price stability. The banking sector is of importance in the transmission
mechanism as it transforms the maturity of loans. Over the past two decades the financial
industry has undergone structural changes, whereby banks have become increasingly reliant
on wholesale funding as compared to retail deposits, in addition to the more recent
emergence of a shadow banking sector. This is comprised of the aforementioned specialized
investment vehicles (SIVs) and other off-balance sheet entities which were devised in order
to circumvent the Basel II capital requirements for risky assets. The funding of long term
investments held by banks and their respective SIVs has occurred increasingly through
issuance of short-term commercial paper and overnight repo agreements, such that a yield
differential is exploited, whilst also creating a maturity mismatch.

During the most recent period of financial stress, the effectiveness of the monetary policy
transmission mechanism has been directly diminished. This is because changes in the policy
rates have only had limited impact on the interbank money markets, such that in effect the
central banks have lost control over the short end of the yield curve. More specifically,
financial institutions have not passed on the cuts in rates to lower the cost of unsecured
borrowing, but rather Libor fixings have remained elevated due to the increase in the
previously discussed liquidity and credit related premia over and above the risk free rate.
Furthermore, these conditions of market dislocations have been amplified due to the recent
structural changes in financial markets, in turn increasing systemic risks. Banks have
 39

exhibited ever greater reliance on wholesale money markets with the respective funding
share of deposits declining from over 50 percent in 1980 to under 20 percent in 2008
(IMF (2008)). At the same time the conditions in the interbank money markets have been
significantly more volatile during the crisis period as compared to the interest rate payments
made to retail customers. As pointed out in Figure 3, widening of Libor spreads and thus
increased funding costs were not geographically confined such that diversification of
liquidity access was curtailed. Combined with the increased mismatch between funding and
investment maturities, a system reliant on market confidence as in the classic bank-run
literature following Diamond and Dybvig (1983) emerged, potentially leading to
self-fulfilling equilibria of investment withdrawals.

Figure 3. U.S., U.K., and Euro area Libor-OIS Spreads

Policy tools

In the empirical analysis in the following section, much focus is placed on the effectiveness
of two main policy tools, namely the supplementary 90-day long-term refinancing operations
(LTROs) by the ECB and on the Term Auction Facility (TAF) by the Fed. These intervention
instruments were devised in order to provide additional liquidity following the impairment of
the interbank money markets, the increased demand for central bank liquidity and to address
the widening spread between secured and unsecured lending.

During periods of tranquility the ECB provides liquidity to market participants through short
term operations (MROs) over 4–5 week reserve maintenance periods (RMPs). In this context,
the projected benchmark allotment is defined as the amount of funding required to allow all
counterparties to fulfill their respective commitments. Following the onset of the subprime
 40

crisis the ECB began injecting additional liquidity into interbank money markets, whereby a
series of fine-tuning operations totaling more than $200 billion were carried out between
August 9 and August 14, 2007 following the emergence of liquidity shortages. Furthermore,
a supplementary LTRO was announced on August 22. Within these interventions additional
liquidity was provided at the beginning of the RMP over and above that required by the
benchmark allotment in order to account for increased uncertainty with regard to liquidity
demands. Emphasis was especially placed on liquidity shortages at longer dated maturities as
this financing was harder to obtain for banks during the crisis due to higher associated risk
premia. By simultaneously withdrawing short term funding the maturity composition rather
than the aggregate amount of reserves was changed, such that monetary policy and interest
rate targets could be achieved throughout.

With end-of-year funding pressures increasing, the Fed on December 12, 2007 announced
a temporary Term Auction Facility such that banks could borrow loans for up to 28 days
maturity, secured against permissible collateral. There were two main reasons for its
introduction. Firstly, liquidity in U.S. markets is normally provided by the Fed through a few
select brokers which limits the number of eligible counterparties for receipt of central bank
funding. During times of severe market stress and funding illiquidity, these intermediaries
hoarded funds, thus impairing their distribution. Secondly, the introduction of the TAF was
intended to overcome the stigma attached to accessing the discount window due to negative
signaling effects in the presence of asymmetric information and limited confidence by market
participants with regard to the health of financial institutions.

The TAF was also linked through a foreign currency swap operation with the ECB and
the Swiss National Bank, allowing them to provide U.S. dollars to their much wider set
of recipient institutions. Other policy tools included the New Primary Dealer Credit
Facility which was introduced by the Fed in March 2008, whilst the Bank of England on
April 21, 2008 launched a special liquidity scheme whereby banks were able to exchange
mortgage-backed securities against UK Gilts for a period of up to three years.

C. Empirical Analysis

Decomposition of the Libor-OIS spread into liquidity and credit components

In this first part of the empirical section the Libor-OIS spread is decomposed into two
components which are associated with liquidity and credit risk premia using the methodology
proposed by the Bank of England (2007). This is related to less formal work by Michaud and
Upper (2008). As previously argued, a no arbitrage condition dictates theoretical equality
between Libor and the correspondingly dated overnight index swap rate. The subsequent
spread between these two measures is mostly explained by liquidity and credit premia,
whereby the latter arises due to the fact that Libor is an interest rate associated with
unsecured lending. The main reason for this decomposition is that it allows us to quantify
changes in the make-up of the Libor-OIS spreads as crisis events unfolded. Furthermore,
 41

it is of interest with regard to motivating future research as it would be possible to distinguish

between the impact of central bank intervention on these two differing constituents, thus
extending the analysis by Taylor and Williams (2008) which does not disentangle these
effects. In this context it is to be expected that the liquidity providing operations such as the
LTROs and the TAF would affect the observable total spread level whilst being unable to
reduce interbank money market stress emanating from counterparty risk.

We make the simplifying assumption that the Libor spread is fully explained by a liquidity
and by a credit component, such that volatility effects of the future expected overnight rates
and market liquidity in general are ignored. The credit premium is derived by employing
credit default swap (CDS) spreads for banks in the Libor panel, whereby it is furthermore
assumed that liquidity and solvency risks are independent and that CDS spreads provide a
fair probability of default.16 In the first stage of the decomposition a no arbitrage argument
under risk neutrality is employed to infer the implied probability of default of Libor panel
banks, by combining the observed market prices of their CDS contracts with a recovery rate
of 40 percent.17 Next, this probability is used to derive the premium above the risk free OIS
rate which is required for investors to be indifferent to accepting the credit risk within the
interbank money market. Finally, these spreads are averaged across the banks in the Libor
panel in order to quantify the credit component of the Libor-OIS spread, whereby the
residual is attributed to a liquidity premium.

In our analysis we extend the sample used by the Bank of England by six months until
April 2008 so that the rescue of Bear Stearns is included. As it can be seen from Figure 4,
during the onset of the subprime crisis the increase in the Libor-OIS spread was mainly
driven by liquidity risks. As discussed in the previous section, funds were extended to off-
balance sheet investment vehicles and liquidity was hoarded by banks, thus reducing the
interbank lending in both the U.S. and Euro area money markets. Liquidity pressures
subsequently declined during the autumn of 2007 but end-of-year effects, driven by window
dressing of balance sheets by financial institutions, raised demand for interbank money
market funds. Interestingly, credit concerns rose continuously until the rescue of Bear Stearns
in mid-March 2008 due to increasing write downs of structured securities, uncertainty with
regard to the value of their underlying assets and heightened risk aversion in general.
Subsequently this trend is reversed, whereby capital markets seemingly re-priced the

16
Clearly, this is not always the case, especially during times of financial turbulence, such as in Iceland where
in 2008 hedge funds speculated on sovereign and corporate default. Furthermore, the effects of government
bailouts and nationalization of financial institutions are ignored. This is of importance as such events may affect
CDS spreads by causing convergence between those of banks and those corresponding to government debt.
17
Clearly, this is a further simplifying assumption as the recovery rate following default will vary across
individual banks depending on their respective financial health, and may also decline as the extent of market
dislocations deepens over the sample period. Also, it is not clear in how far government guarantees affect the
recovery rate as compared to the probability of default itself.
 42

probability of the survival of financial institutions conditional on implicit guarantees

provided by the U.S. government, causing the corresponding CDS spreads to decline
dramatically.

Figure 4. Decomposition of U.S. and Euro area Libor-OIS Spreads.

In addition to analyzing the changing composition of the Libor-OIS spread in levels during
the early phases of the subprime crisis, in Figure 5 the liquidity and credit risk premia for the
Euro area and U.S. Libor spreads are presented in first differences around the time of the
introduction of the first TAF.18 Interestingly, as can be seen from the top panel, before the
implementation of the auction on December 17, 2007 changes in the Euro area spread
associated with liquidity risks were positive on 19 out of 20 trading days, indicating that
money market stress emanating from this risk premium increased systematically before this
event. Subsequently, on 21 out of 24 days following this intervention the changes in the
liquidity component were negative, whereas no such systematic impact is observed for the
corresponding credit component. With regard to the first differences of the decomposed
Libor-OIS spread for the U.S., similar results are found, but where the sign change occurs
following the announcement of the TAF on December 12, rather than on its implementation
date. As previously mentioned, we believe that these results motivate further research in
order to explicitly quantify the effectiveness of central bank intervention on both
components. Especially as the percentage of the Libor-OIS spread which is attributed to
credit risk is increasing throughout the crisis, findings in Figure 3 would suggest that the
impact of liquidity providing instruments has declined.

18
This analysis is related to similar findings in Michaud and Upper (2008) which were written simultaneously
to the publication of IMF (2008) and during the conceptualization of this paper.
 43

Figure 5. Decomposition of Libor-OIS Spreads

 44

Graphical analysis using a Markov-Switching Approach

To assess the effectiveness of central bank intervention in reducing the stress in interbank
money markets, we first employ a graphical analysis that compares the timing of distinct
policy events with changes in the levels and the volatility of Euro area and U.S. Libor
spreads. To this end it is hypothesized that the respective data generating processes are
subject to regime changes. Markov switching models in the first two moments of the Libor
spreads are estimated, such that probabilities of being in certain states of the world can be
derived. These findings build upon work published in IMF (2008). In the Markov Switching
analysis, we focus on the behavior of the overall Libor-OIS spreads rather than the liquidity
and credit component.

A Markov-switching model differs from a standard econometric regression in that the

coefficients and the variance terms may be dependent on an unobserved state variable St,
which is assumed to follow a Markov chain. These models allow for detection of
unobserved structural breaks in the data and subsequent transitions between differing states
of the world. As part of the estimation, inference with regard to smoothed probabilities
is made, which denote the probability of a respective regime
occurring, conditioned ex post on the entire past realizations of rt.

In this paper, the Markov switching framework is used in order to analyze the effect of
central bank intervention on the regime transitions of the first two moments of the Libor
spreads. In both cases, the existence of three states of the world is assumed.19 The
corresponding model for the first differences of the spread levels is based on the univariate
mean-variance specification by Hamilton (1989)

where are the three states of the world. Here, first differences of the Libor
spreads rt are a function of a state dependent constant , whereas the variance of the
subsequent residuals also exhibits respective regime shifts.

The volatility of the first differences of the spreads is captured using the Markov
switching ARCH (SWARCH, hereafter) model proposed by Hamilton and Susmel
(1994). The mean equation is specified as an AR(1) process ,

19
This selection is motivated by the constancy of the Libor-OIS spread before the onset of the subprime crisis,
which is not well captured by a 2 regime specification which would only model increases and decreases in the
spread.
 45

where the conditional variance is parameterized as

where and . Thus the ARCH(q) process in (2) is state

dependent due to multiplication with a scaling factor St which is normalized to unity for the
low volatility regime.2021

In Table 1 the results for the Markov switching estimation are presented based on a sample
ranging from February 1, 2007 until April 4, 2008.22 The mean-variance model in (1) implies
that for the Euro area the average changes in the level of the Libor spreads in the low,
medium and high intercept states of the world are approximately -1, 0 and 1.6 basis points
per day, respectively. This compares to daily changes for the U.S. of -0.8, 0 and 3.4 basis
points such that the coefficient quantifying the upward pressure in the interbank money
markets is twice as high as compared to that for Europe. Before the onset of the crisis the
model indicates for both markets that the data generating process is best characterized by the
middle regime, which is consistent with the observation that during this period Libor spreads
showed very little change and remained approximately constant.

20
In this analysis an ARCH specification is estimated, as the GARCH(p,q) is not nested within the SWARCH
framework, due to its implicit infinite lag representation.
21
In this paper we acknowledge that the model selection can be further refined. Firstly, the mean-variance
specification in (1) could be augmented using an autoregressive parameterization or through the inclusion of
further exogenous regressors explaining the Libor-OIS spread. Secondly, it would be possible to make the mean
equation for the SWARCH model state dependent. Finally, in future research we aim to make the smoothed
probabilities of being in specific regimes a direct function of central bank interventions. Due to the restrictive
modeling tools within the Markov switching framework, this has so far not been possible.
22
The statistical significance of the scaling parameter ɣ indicates the existence of switches in the data
generating process. Inference with respect to gamma is complicated by the fact that corresponding significance
tests exhibit non-standard distributions. Following argumentation by Hamilton and Susmel (1994) it is
concluded that as the associated test statistics are of such great magnitude, the null hypotheses of no regime
changes are rejected for both volatility models, regardless of whether the skew in the distribution is accounted
for or not.
 46

Table 1. Markow Switching Parameters for Leveles and Volatility Models

In the lower half of the Table 1, the scaling parameter , as described above, is presented
for the Markov switching ARCH models for both Libor spreads. During the financial crisis
and the corresponding occurrence of the medium and the highest volatility regimes, the
conditional variance in the European money markets is 17 and 29 times greater relative to
that in the pre-crisis period which is characterized by the normalization of = 1. In the
U.S., this multiplication factor is even more pronounced, standing at 34 and 249, reflecting
the fact that the conditional variance is driven by the large outliers in the changes of the
Libor spread during the period of market turbulence.

Mean-Variance Model

In Figure 6, a graphical representation of findings for the mean-variance models is reported

for both the European and U.S. Libor-OIS spreads. Here, the blue line represents the
smoothed probability of being in the highest intercept state as measured on the right axes,
whereas the shaded bars denote major central bank interventions. Throughout, focus is placed
on the aforementioned LTROs by the ECB, in addition to cuts in both the Federal funds and
discount rates, and the introduction of the Term Auction Facility by the Fed.

The top panel in this figure indicates that stress in the Euro area money markets began to rise
substantially around August 9, 2007, illustrated by multiple upward movements in the Libor
spread of approximately 8 basis points per day. Apart from a shock on September 27, the
magnitude of these pressures decreased, followed by a reduction in the spread during
October until mid-November 2007. On November 19, end-of-year pressures in the interbank
money markets become apparent, as banks hoarded liquidity in order to support their
respective balance sheets during the financial reporting season, in addition to suffering
further write downs and credit related losses. A sign change in the first difference of the
spread occurs on December 18, 2007, after which the Libor rate decreased systematically at
an average rate of about three basis points per trading day.

Corresponding to these data, the Markov switching model implies two main periods during
which a unit probability of being in the highest intercept state is assigned. The first
corresponds to the beginning of the interbank money market stress during August, and the
second to the-end-of-year pressures in November and December. The brief transition around
September 27, 2007 is attributed solely to the corresponding outlier mentioned above.
 47

Regarding the effectiveness of the central bank interventions, the ECB announced and
carried out its first supplementary LTRO on August 22 and 23. As is evident from the upper
panel in Figure 6, the magnitude of the changes in the Libor rates decreases from
approximately 8 basis points, and apart from two exceptions on September 5 and 27, does not
exceed 2 basis points until mid-November.

Thus, the supply of this additional long-term liquidity coincides with a fall in the rate at
which the Libor spread increased.

Following the end-of-year pressures in the money markets, the ECB conducted a further
supplementary LTRO on November 22, but which largely remained ineffective in
encouraging banks to lend to their respective counterparties. Finally, on December 12, the
Fed announced the introduction of the TAF and the implementation of the first auction on
December 17, which is associated with a clear sign change in the first difference of the Euro
area Libor spread. While it increased on average at approximately two basis points per day
before the TAF implementation, decreases of about three daily basis points are observed
afterwards.

In the lower panel of Figure 6, the results of the mean-variance specification for the U.S. are
presented. In comparison to the European Libor, the changes in the spreads are of greater
magnitude, but at the same time are less persistent.

Again, money market stress starts building significantly on August 9, after which the Libor
spread stabilizes and subsequently declines between October 22nd and November 1st. As in
the case above, end-of-year pressures are also present in the U.S. interbank money market.

These periods of market dislocations are captured by increases in the smoothed probability of
being in the highest intercept state in the Markov switching model. As implied by Table 1,
during late August and early September, the average increase in the U.S. Libor spread is
approximately 3.4 basis points per trading day.

This period is followed by a regime switch into the lowest state of the world, implying that
money market stress is reduced by 0.85 basis points per day until approximately
November 16, when end-of-year pressures cause the model to switch back. A further regime
change to the lowest intercept occurs on December 7, 2007.

Following a cut of 50 basis points in the Federal funds target and the discount rates on
September 18, the U.S. Libor spread exhibited its steepest decline over the entire sample
period by falling 35 basis points, as illustrated in the lower panel of Figure 6. Subsequently,
the interbank money market entered a period of relative calm until mid November. In
response to increasing pressures during the end of the year, both rates were cut again by
 48

25 basis points on December 11, which was followed by the TAF announcement and
implementation. As in the case of the Euro area Libor, these events are associated with a sign
change in the movements of the U.S. spread.

Figure 6. Markov Switching Mean-variance Model for Euro area and U.S. Libor-OIS
Spreads
 49

Conditional Variance Model

The graphical representation of the results from the Markov switching ARCH model is
presented in Figure 7 in which the effects of the same central bank interventions, as discussed
above, on the conditional variance of the spreads are analyzed.

Two distinct periods of highest volatility are observed in the case of the Euro area Libor-OIS
spread. The first is driven primarily by the increases in the spreads at the beginning of the
crisis in August 2007, during which time a switch from the lowest to the highest volatility
regime occurs. The econometric model implies that this elevated level of interbank money
market stress subsequently begins to retreat in the beginning of September. The period of
market calm lasts until November 19, 2007, after which the second period of heightened
volatility is observed. This is caused by the previously mentioned end-of-year effects, which
led the Libor spread to increase on average by 2 basis points per trading day. No further
regime switches are recorded during the sample period. Despite the sign change around the
time of the TAF introduction, the absolute value of the changes of the spread does not
decline.

With regard to the effectiveness of central bank intervention, we find some tentative evidence
that the announcement and implementation of the ECB's LTROs had a volatility reducing
effect. As previously described, following the liquidity injection on August 22/23, the
magnitude of the changes in Euro area Libor declined significantly. Due to the autoregressive
parameterization of the Markov switching ARCH model a lag arises though when
quantifying this effect by assigning the corresponding smoothed probabilities of being in the
respective volatility states.

The second panel of Figure 7 contains the findings for the U.S. Libor spread. During the
beginning of the interbank money market stress the data generating process is in the highest
volatility state with approximately unit probability. A shift into the medium regime occurs on
September 28. Unlike in the case of Euro Libor, this period of market calm does not extend
into October, as large negative changes in the spread induce an increase in the volatility,
which is eventually also driven by end-of-year effects. These findings are also consistent
with central bank intervention having reduced stress in the U.S. interbank money markets
when measured in terms of Libor spread volatility. As previously mentioned, the lowering of
the Federal funds target and discount rates on September 18 led to a decline in the spread of
35 basis points, after which the magnitude of Libor changes falls in absolute value. The
Markov switching ARCH model is influenced by this large negative shock, but picks up the
volatility reducing effects by September 27. As in the case of the Euro area Libor, the effect
of the TAF on volatility is limited as despite the induced sign change, the absolute value of
the spread movements does not decline.
 50

Figure 7. Markov Switching ARCH Model for Euro area and U.S. Libor-OIS Spreads

In summary, Markov switching models for both changes in the levels and the volatilities of
the Euro area and U.S. Libor-OIS spreads are employed in order to assess the effectiveness
of central bank intervention on money market pressures. Graphical evidence is provided that
the LTROs by the ECB, and the interest rate cuts and the introduction of the TAF by the Fed
were able to reduce term funding stress.
 51

Bivariate VAR Model

In addition to the descriptive and graphical analysis of the previous subsection, we specify a
bivariate model in which the behavior of the overall Libor-OIS spreads is explained as a
function of a series of variables capturing extraordinary central bank operations. Whilst
approaches such as the Markov switching framework or the univariate specification by
Taylor and Williams (2008) provide first useful insights towards understanding the effect of
central bank policies, they ignore a series of important aspects of the recent crisis. As
previously argued, Libor fixings in Euros and U.S. dollars are likely to display substantial
interdependencies, as funding conditions are increasingly of a global nature, in particular
during periods of crisis.

Similarly, liquidity provisions by central banks in one currency potentially alter funding
conditions in another, as they are transmitted through changes in the default probability of
counterparties, conditions in foreign exchange markets and by affecting overall risk aversion.
Finally, central bank measures were put in place during the recent turmoil which has
explicitly targeted frictions in global liquidity allocation, most notably the extension of the
TAF through swap arrangements with the ECB and the Swiss National Bank.

In order to account for these interdependencies between funding markets a bivariate Vector
Autoregression (VAR) is estimated which quantifies the impact of emergency response to
liquidity stress by jointly modeling changes in the U.S. and Euro area Libor-OIS spreads.23
As before, the LTROs by the ECB and the TAF by the Fed are used as explanatory variables,
whereby differentiation is also made between the announcement and the actual
implementation dates. In addition to these extraordinary liquidity operations by central
banks, the Fed reduced its policy rates during the financial turmoil, which is also included in
the model in order to gauge any possible impact of these more conventional policy tools.
The VAR specification in (3) is estimated for the crisis period spanning from July 1, 2007
until April 3, 2008, where rt is defined as the first difference of the spread between the
three-month Libor and OIS.24 denotes central bank intervention with
this instrument being in the form of a dummy variable which takes on the value of 1 during
the occurrence of the intervention and 0 otherwise. In this context, all policy events in our
sample are 1 day long, whereby in addition to the announcement and implementation of the
first TAF there are 5 LTROs, 6 cuts in the Federal funds and 8 reductions in the discount
rates.

23
Similarly to the Markov switching approach, we focus on the overall Libor-OIS spreads rather than the
liquidity and credit components.
24
With regard to model selection, standard techniques such as information criteria and residual analysis are
employed. Here it is found that the VAR(1) specification in (3) is sufficient due to limited auto- and cross
correlations in the changes of the spreads.
 52

In Table 2 we present the corresponding results for the effectiveness of differing policy
measures on the Libor-OIS spreads of the 6 individual VAR(1) specifications that each
capture a separate type of intervention. With regard to the announcement of the TAF, it is
shown that a compression of the U.S. spread by approximately 9 and 7 basis points at one
and two lags, respectively, is achieved whereas the effect in the Euro area is negligible.
Interestingly, the subsequent introduction of the TAF itself is ineffective with regard to
reducing interbank money market stress, which indicates efficiency in that this new
information has already been priced in.
Table 2. Bivariate VAR Model
 53

The announcement of the LTROs by the ECB has a statistically significant spread reducing
effect in both markets at one lag, albeit only a small impact of 2 basis points in the Euro area
compared to a 7 point reduction in the U.S.. As in the case of the TAF, the actual
implementation is insignificant, further highlighting possible market efficiencies with regard
to information arrival and processing. Finally, the impact of cuts in both the Federal funds
and the discount rates are quantified. In this case the degree of spread compression in the
U.S. money markets exceeds that in Europe substantially for both policy tools. Overall, while
the announcement of the TAF by the Fed and the LTROs by the ECB have a statistically
significant effects, the economic magnitudes are not very large compared to the sharp
increase of the Libor spreads since the beginning of the subprime crisis. While central bank
interventions have been helpful in bringing down the spreads, they have not been successful
in arresting the still very high levels of Libor spreads, compared to the pre-crisis period, as
well as the continuous hoarding of liquidity by financial institutions due to counterparty
concerns.

Finally, we quantify the cumulative effect of central bank intervention on the interbank
money markets. We use parameter estimates from Table 2 to construct impulse response
functions for the change in the spreads following a shock to the binary intervention variable.
More specifically, the forward iteration of (3) is constructed with and without interventions,
whereby the difference in these measures is presented in Figure 8 for selected policies. The
reported 95 percent confidence intervals are obtained using Monte Carlo simulations based
on the asymptotic joint normality of the parameters of the VAR.
 54

Figure 8. Impulse Response Functions

Following from a reparametrization of the results, provided in Table 2, we find that the
dynamic impact of central bank intervention, as measured by the impulse response
functions, is greater in the U.S. money market as compared to that in Europe. The forward
iteration of (3) indicates that the announcement of the LTROs by the ECB caused a
cumulative effect of a 15 basis point reduction in the former market in addition to a 5 basis
point compression in the latter.

With regard to the announcement of the TAF, these quantities are 35 basis points and 1 basis
point, respectively. The impulse response for the Euro area exhibits a slight hump whereby it
is initially negative before turning positive and subsequently declining again. This shape can
be explained by the sign change of the corresponding parameters in Table 2. Interestingly,
the final decline arises due to the positive cross-correlation in the A matrix in (3) indicating
that interdependency between money markets is important, which is an effect not modeled by
Taylor and Williams (2008). Finally, as expected, cuts in the interest rates by the Fed have a
greater effect in the U.S. than in Europe.

Concerning the interpretation of these results it should be noted though that despite the fact
that in absolute value the reduction in the Euro area Libor-OIS spread is small, the policy
interventions by the ECB can be seen as having been helpful in bringing down the spreads.
During the crisis period used in this analysis, the mean spread level has been approximately
 55

55 basis points such that the announcement of the LTRO facility reduced the money market
stress by approximately 10 percent.

In this section it is shown that liquidity providing interventions by central banks achieved a
compression of the Libor-OIS spreads. Alternatively, it can be argued that the volatility of
such spreads is a further adequate proxy for interbank money markets stress as it captures the
uncertainty with regard to future write downs, credit conditions and economic activity in
general.

D. Bivariate GARCH Framework

To complement the descriptive and VAR analysis of the previous section, we estimate a
series of GARCH models in which we explain the conditional variance of Libor-OIS spreads
as a function of their own past realizations and a series of variables capturing extraordinary
central bank operations.25 To get a sense of the underlying dynamics, we first modeled the
volatility of Euro and U.S. dollar term spreads using a univariate GARCH specifications
focusing on intervention variables that were available to central banks when the crisis started
(IMF, 2008). While these GARCH results for most of the ECB intervention variables were
inconclusive, there appeared to be a statistically robust and significant volatility-reducing
effect in the case of the ECB’s supplementary LTRO. The Fed’s interventions via additional
repurchase agreements, in turn, appeared to have had a significantly negative
contemporaneous effect on U.S. dollar spread levels and volatilities. The former, however, is
largely offset by a rebound on the next day, and both effects are sensitive to the chosen lag
structure.

As previously argued, while a univariate GARCH model provides a first useful step towards
understanding the effect of central bank policies on conditional means and variances, it
ignores the dynamic interaction between Libor fixings in Euros and U.S. dollars. To account
for this, we estimate a bivariate GARCH model which evaluates the impact of the central
banks’ emergency response to liquidity stress by jointly modeling U.S. and Euro Libor
spreads. As before, the ECB’s LTRO and the Fed’s TAF are included as explanatory
variables and we also differentiate between the possible effect of the intervention
announcement and the actual intervention date. The estimation is conducted for a sample
spanning from July 1, 2007 to April 3, 2008.

More specifically, the bivariate BEKK GARCH (1,1) model, developed by Engle and Kroner
(1995), is modified in order to include the intervention variables as exogenous regressors.
This model thus allows us to capture the dynamic interactions between both the U.S. and
Euro Libor spreads explicitly, as well as to quantify the effect of any particular intervention

25
Again, we focus on the overall Libor-OIS spreads rather than its liquidity and credit components.
 56

on the first two moments of the Libor spreads. Furthermore, by construction, the conditional
covariance matrix is always positive definite, which overall makes the BEKK GARCH (1,1)
model very tractable for our purposes. The mean equations are specified as

1  b2 y t 1  c1 I t 1   1,t
y tEU  a1  b1 y tEU US

1  b4 y t 1  c 2 I t 1   2 ,t
y tUS  a 2  b3 ytEU US
(4)

where y t  ( y tEU , y tUS ) captures changes in the US and Euro 3-month Libor spreads, I t 1
denotes the intervention of the ECB or Fed (e.g. TAF or LTRO) and  t  ( 1,t ,  2 ,t ) are the
 
residuals with  t  t 1 ~ N 0, H tt . The conditional covariance matrix H t is given by

H t  M  A t 1 t1 A  BH t 1 B   EI t 1 (5)

where M is a scalar and A, B as well as E are diagonal 2  2 matrices. In addition, the model
is estimated with Bollerslev-Wooldridge robust standard errors.

Table 3 provides the findings for the effectiveness of central bank intervention on the
conditional variance of the 3-months Euro and U.S. Libor-OIS spreads. Both the
announcement and the implementation date of the TAF auction have some volatility reducing
impact on the U.S. Libor at 2 lags. Furthermore, the volatility of the Euro Libor spread
significantly declines following the announcement and implementation of the LTRO. The
date of the TAF auction has also a significant impact on the conditional variance of the Euro
Libor spread but, as expected, magnitudes are lower than in the case of the U.S. Libor spread.
Finally, the mean equations of the BEKK models show the same results as in the VAR
specifications.

Overall, the results of the BEKK model are consistent with the findings from the
Markov-Switching approach and the bivariate VAR but should be seen as rather indicative
given some caveats. Firstly, the BEKK model employed here is not a structural model.
Secondly, it does not account for alternative explanatory variables of the Libor-OIS spreads.
Thirdly, some of the findings are not robust across different lags and the differing effects of
policy announcement and its actual implementation are not as pronounced as in the VAR
analysis.
 57

Table 3. Impact of Central Bank Interventions on LIBOR-OIS

Spreads
Volatility of Volatility of
Euro Libor- OIS US Libor-OIS
Announcement of TAF (-1) 36.405 -18.572
(0.317) (0.551)
Announcement of TAF (-2) 33.889 -70.562
(0.196) (0.000)***
Date of TAF Auction (-1) -5.632 -9.252
(0.000)*** (0.709)
Date of TAF Auction (-2) -1.566 -23.504
(0.012)** (0.000)***
Announcement of LTRO (-1) -3.905 -2.791
(0.000)*** (0.959)
Announcement of LTRO (-2) -6.901 4.360
(0.000)*** (0.417)
Date of LTRO (-1) -1.440 12.107
(0.501) (0.418)
Date of LTRO (-2) -10.800 -4.874
(0.000)*** (0.348)
Source: Own Calculations
Note: The mode is computed using Bollershev- Wooldridge robust
standard errors. *** indicates significance at the 1 percent level and **(*)
indicates significance at the 5 (10) percent level. The sample is from July,
1, 2007 to April 3, 2008. OIS= overnight index swap; TAF= term auction
facility; LTRO= long term refinancing operation.

E. Policy Implications and Conclusions

In summary this paper has provided some evidence that a range of central bank policies had
an impact on the dynamics of stress in unsecured interbank markets during the subprime
crisis, as measured by the spread between Libor and OIS rates. In this context, the
supplementary long term refinancing operations by the ECB as well as the Term Auction
Facility by the Fed have been helpful in compressing Libor spreads. But the economic
magnitudes are not very large, which is supported by the fact that central banks’ actions
during the first phase of the subprime crisis, while successful in helping to bring down Libor
spreads, have not led to an end of the liquidity crisis and a containment of the solvency
concerns looming at the time.

In addition to these extraordinary liquidity operations, policy rates have been reduced by
central banks. Whilst these cuts have mainly been in response to changes in the
macroeconomic outlook, financial stability considerations have also been taken into account.
In the case of the Fed, the empirical results suggest that reductions in both the Federal funds
and the discount rates appear to have had statistically significant effects on alleviating
funding pressures.
 58

Finally, further interventions were made by the Fed, such as the provision of guarantees
during the rescue of Bear Stearns in March 2008 and the provision of discount window
access to investment banks. Anecdotal evidence suggests that these policies directly affected
the assessment by capital markets with regard to the probability of the failure of financial
institutions. In the previous section this was discussed in the context of the decomposition of
the Libor-OIS spreads into liquidity and credit related premia, whereby the latter
substantially decreased following these events.

In quantifying the effectiveness of central bank intervention, we believe that there is scope
for further research. More specifically, selection of the Markov switching models could be
refined by making the transition probabilities a direct function of policy events. Furthermore,
the decomposition would be improved by overcoming the restrictive assumptions used in this
paper. Subsequently, it would also be of interest to conduct the inferential analysis separately
on the liquidity and the credit component of the decomposed Libor-OIS spread. If, as
hypothesized, it were the case that the credit premium is not influenced by these
interventions, this would imply that the liquidity injections in the money markets have
become less effective as the crisis unfolded, as the percentage of the Libor-OIS spread which
is attributed to counterparty risk has increased.

In conclusion we discuss some policy implications, whereby focus is placed on the

characteristics which the operational framework of these liquidity providing policies ought to
exhibit, in addition to their potential limitations. In this context, the effectiveness of the
LTROs and the TAF by the ECB and the Fed, respectively, is ensured by providing central
bank funding against a broad pool of eligible collateral. This includes financial instruments
which have either suffered declines in market prices due to illiquidity during the crisis
period, or those which exhibit outright uncertainty with regard to their fundamental value.
Accepting such collateral is and has always been a crucial feature of crisis management, and
does not represent a fundamental departure from long-standing principles, such as those
formulated by Bagehot (1873).

Furthermore, access to liquidity is provided to a broad range of potential counterparties.

Examples of this are the increased eligibility of central bank facilities such as the refinancing
operations by the ECB which allow access to over 500 financial institutions, as well as the
TAF auction and the extension of the discount window facility in order to overcome the
aforementioned distributional limitations due to the hoarding of funds by select brokers. In
response to increased demand for longer dated liquidity, central bank policies placed
emphasis on providing funds at maturities ranging from one week to six months. Finally,
differences between traditional liquidity provisions and arrangements such as the TAF and
LTROs were made in that emergency funding was no longer provided on a discretional basis
in the form of short term bridge financing, but rather within the scope of standard open
market operations which were rolled over continuously.
 59

It should be noted though that the design and implementation of these liquidity operations
pose serious limitations and introduce the risk of commercial banks engaging in regulatory
arbitrage. In this context the main issue is that of incentive compatibility and subsequent
moral hazard. As noted by Goodhart (2007), financial institutions have taken out a protective
put on central bank liquidity during the crisis period. The associated risks have been
increasingly transferred to the public sector, whilst the upside in the form of the yield
differential due to liquidity premia has been taken advantage of. This issue is of central
importance as, due to externalities resulting from the fact that the banking system is a public
good, ex post it is mostly optimal to rescue systemically important entities. In order to
address these concerns it has been suggested that liquidity ought only be provided at a penal
rate against good collateral. This latter consideration is essential in limiting the credit risk
which is assumed by central banks, whilst also providing incentives for financial institutions
to hold higher quality assets.

Furthermore, the construction of an optimal framework for liquidity management

has to take recent structural changes within the banking sector explicitly into
account. Whereas the Basel accords govern the adequacy of the capital holdings of
financial institutions, no such arrangement exists for liquidity. Subsequently, the
percentage of liquid assets held by British banks on their balance sheets has
continuously declined from approximately 30 percent in 1950 to about 1 percent
presently. This is of significance as the maturity mismatch inherent in the financial
industry has simultaneously increased. Banks and their respective SIVs have
become more reliant on wholesale money and the short term commercial paper
markets in order to fund longer dated investments. This in turn has raised the risk of
systemic funding illiquidity because, as seen in Figure 3, stress in these markets
may become geographically correlated. As a response, policy makers have been
challenged to determine the adequate degree of maturity mismatch and holding of
liquid assets, such that banks have a reserve to meet their short-run funding
commitments, whilst ensuring operational efficiency.

Finally, the optimal incidence of insuring against funding illiquidity between the public and
private sectors is to be determined. Complete private coverage for such infrequent events
may be inefficient as optimal risk sharing in the presence of incomplete markets is
unobtainable, in addition to the costs associated with self insurance by holding large amounts
of liquid assets being high. At the same time, incentives are to be provided such that the
public sector acting as lender of last resort to the financial industry during periods of
illiquidity does not make the occurrence of such events more likely. Exactly how these
measures are to be combined within the operational framework by central banks and in the
regulatory oversight of liquidity management is still a question for further research.
 60

References

Aït-Sahalia, Y., J. Andritzky, A. A. Jobst, S. Nowak, and N. Tamirisa, 2009, “How to Stop a
Herd of Running Bears? Market Response to Policy Initiatives during the Global
Financial Crisis,” forthcoming IMF Working Paper (Washington: International
Monetary Fund).

Baba, N., Packer, F., and Nagano, P., 2008, BIS Quarterly Review.

Bagehot, W., 1873, "Lombard Street, a Description of the Money Market."

Bank of England, 2007, "An indicative decomposition of Libor spreads," Quarterly Bulletin,
pp. 498–99.

BIS Quarterly Review, 2008, Bank of International Settlements British Bankers Association:
"Libor Governance and Scrutiny," Proposals by FX and MM committee.

Diamond, D. W., and Dybvig, P. H., 1983, "Bank runs, deposit insurance, and liquidity",
Journal of Political Economy, Vol. 91, pp. 401–19.

Frank, N., Gonzalez-Hermosillo, B., and Hesse, H., 2008, "Transmission of Liquidity
Shocks: Evidence from the 2007 Subprime Crisis," IMF Working Paper 2008/200
(Washington: International Monetary Fund).

Frank, N., and H. Hesse, 2009, "The Effectiveness of Central Bank Interventions During the
First Phase of the Subprime Crisis," IMF Working Paper 2009/ 206 (Washington:
International Monetary Fund).

Giavazzi, F., 2008, "Why does the spread between Libor and expected future policy rates
persist, and should central banks do something about it?," VOX Research.

Goodhart, C., 2007, "Liquidity Risk Management," FMG Special Papers 162, London School
of Economics.

Hamilton, J. D., 1989, "A New Approach to the Economic Analysis of Nonstationary Time
Series and the Business Cycle," Econometrica, Vol. 57, pp. 357–84.

Hamilton, J. D., and Susmel, R., 1994, "Autoregressive Conditional Heteroskedasticity and
Changes in Regime," Journal of Econometrics, Vol. 64, pp. 307–33.

Hull, J., C., 2005, "Options, Futures and Other Derivatives," 6th Edition, Prentice Hall.

IMF, 2008, Global Financial Stability Report, (Washington: International Monetary Fund).

Michaud, F.-L., and C. Upper, 2008, "What drives interbank rates? Evidence from the Libor
panel," Bank of International Settlements.
 61

Taylor, J. B, and Williams, J. C., 2008, "A Black Swan in the Money Market," Federal
Reserve Bank of San Francisco 163.
 62

III. FINANCIAL SPILLOVERS TO EMERGING MARKETS DURING THE GLOBAL FINANCIAL

CRISIS, WITH NATHANIEL FRANK, 200926

In this paper potential financial linkages between liquidity and bank solvency measures in
advanced economies and emerging market (EM) bond and stock markets are analyzed during
the latest crisis. A multivariate GARCH model is estimated in order to gauge
the extent of co-movements of these financial variables across markets. The findings indicate
that the notion of possible de-coupling (in the financial markets) had been misplaced. While
EM stock markets reached their peak in the last quarter of 2007, interlinkages between
funding stress and equity markets in advanced economies and EM financial indicators were
highly correlated and have seen sharp increases during specific crisis moments.

A. Introduction

The recent crisis began in the United States with the bursting of the sub-prime mortgage
market and the unraveling of the securitization process in the summer of 2007, but it initially
did not fully affect emerging markets (EM). In this context, EM stock markets peaked around
November 2007, at a time when the repercussions of the crisis were already apparent in the
U.S. with central banks injecting liquidity into the interbanking markets and major financial
institutions announcing massive writedowns from structured financial products.

The Lehman collapse on September 15, 2008 is seen as a key event, both in advanced
economies but also EM countries, that unleashed a full-blown systemic crisis with global risk
aversion dramatically increasing, asset markets across countries and regions plunging and the
unwinding of carry trades that saw high-yielding EM currencies sharply depreciate within a
short period of time. Even EM countries with sound macroeconomic and financial pre-
conditions, built-up over the previous years, have been strongly affected by the financial
contagion that in late 2008 spilled over to the real sector with export and GDP growth rates
plunging and trade finance being contracting across the world.

This paper examines the financial interlinkages between advanced and EM countries by
focusing on the co-movements of a pertinent number of key financial variables. Specifically,
proxies for general stress in the interbanking market, market volatility and default risk of
major financial institutions in advanced economies are related to stock market, bond spreads
and CDS indices of some selected EM countries.

Since standard correlations are potentially biased when examining spillovers and the
potential for systemic risks to spread (Forbes and Rigobon, 2002), the Dynamic Conditional

26
This chapter is based on Frank and Hesse (2009).
 63

Correlation (DCC) GARCH model by Engle (2002) is used to avoid many of the pitfalls.
This GARCH framework allows us to analyze the co-movement of markets by inferring the
correlations of the changes in the financial variables examined, which in turn is essential in
understanding whether the recent episode of financial distress has become systemic.

The main findings suggest that implied correlations between the U.S. Libor-OIS spread, a
proxy for funding illiquidity, and EMBI+ sovereign bond spreads of Asia, Europe, and Latin
American countries, sharply increase following the onset of the subprime crisis. In addition,
the Shanghai stock market correction in February 2007 led to a temporary spike of the
correlation measures, whereas the Lehman collapse caused the largest increase of co-
movements among these variables. Similarly, the relationship between the S&P 500 and the
EMBI+ regional bond spreads exhibits a potential break during the Chinese episode, after
which correlations increase from the beginning of the subprime crisis, and reach their peak
after the Lehman failure.

In terms of individual country interlinkages, the U.S. Libor spread is related to sovereign
bond and the sovereign CDS spreads of the EM countries Brazil, Russia, and Turkey, Mexico
and South Africa. As before, the Shanghai stock market correction in February 2007 is
evident, in addition to the beginnings of the subprime and the Lehman collapse. The Bear
Stearns rescue in March
2008 also becomes visible with co-movements sharply reversing their down-ward trend prior
to that.

Overall, the findings from the DCC GARCH models indicate that the notion of possible de-
coupling (in the financial markets) had been misplaced. It is true that EM stock markets
reached their peak around November 2007, but interlinkages between funding stress and
equity markets in advanced economies and EM financial indicators were highly correlated
and have seen sharp increases during specific crisis moments. Given the interconnectedness
of global financial markets, investors' increase in global risk aversion from problems in
advanced economies rapidly spilled over into EM countries, as investors sought to pull out
from the latter countries and only invest into the safest and most liquid assets in their home
markets, such as fixed income securities.

The paper this related to the existing literature as follows. It builds upon Frank, Gonzalez-
Hermosillo and Hesse (2008) that analyze liquidity spillovers across asset markets in the
United States as well as in IMF (2008). This is also related to a very substantial literature on
spillovers and contagion that especially our-ished after the Asian Crisis. The identification of
channels of shock transmission across countries is, for instance, discussed in Dungey, Fry,
Gonzalez-Hermosillo and Martin (2005), Dornbusch, Park and Claessens (2000) and Pericoli
and Sbracia (2003). Beirne, Caporale, Schulze-Ghattas and Spagnolo (2008) examine
volatility spillovers from mature to EM countries and test for their changes during crisis
periods. Similarly, some other studies that jointly investigate spillovers of EM and mature
 64

countries are Dungey, Fry, Gonzalez-Hermosillo and Martin (2006, 2007) and Kaminsky and
Reinhart (2003).

In this context, a large body of literature investigated conditional correlations during crisis
periods in order to examine any possible breaks in the underlying data. Examples besides
Forbes and Rigobon (2002) are King, Sentana, and Wadhwani (1994), King and Wadhwani
(1990) and Caporale, Cippollini, and Spagnolo (2005). Investors' risk appetite can rapidly
change during financial crises when suddenly nonrelated asset markets feel the impact by
seemingly unrelated financial shocks. Gonzalez-Hermosillo (2008) and Coudert and Gex
(2007) are some papers that study its importance during crises periods. Finally, some of the
theoretical foundations of contagion are studied by Kodres and Pritzker (2002).

This paper makes several important contributions to the emerging literature on financial
spillovers during the current global financial crisis. It examines the daily co-movements
between key financial variables in advanced economies such as stress in the interbanking
market, market volatility and solvency concerns of large financial institutions with stock
market, bond spread and CDS measures in EM countries. The DCC framework takes time
varying volatility into account and addresses possible feedback effects since unidirectionality
is not imposed. Furthermore, our findings that end-February 2007 was a temporary period
where early signs of stress began to emerge in global markets prior to the time when the
subprime crisis was revealed in mid-2007 is consistent with The Federal Reserve Bank of
Dallas (2008), Gorton (2008) and Gonzalez-Hermosillo (2008) as well as IMF (2008).

The rest of the paper is organized as follows. Section B reviews some of the possible
transmission mechanism of spillovers to EM countries during the global financial crisis.
Section C details the data selection and Section D discusses the empirical methodology.
Section E examines the main results whilst Section F concludes.

B. Transmission of Spillovers to EM Countries During the Subprime Crisis: A

Qualitative Overview

This section examines the role of global market conditions in the current financial crisis and
argues that the Lehman collapse on September 15, 2008, was a key event that led to rapid
spillovers to emerging market countries. The event sharply increased uncertainty across
markets as well as caused a scramble for U.S. dollars with the break-down of the carry trade
and the need for financial institutions to refinance U.S. dollar positions. First, a brief
overview of the different financial linkages across asset markets in the United States during
the crisis is provided, before discussing some of the financial spillovers to EM countries.
 65

The subprime crisis that began in the summer of 2007, was triggered by deteriorating quality
of U.S. subprime mortgages, a credit, rather than a liquidity event.27 This rapidly propagated
across different asset classes and financial markets. Increased delinquencies on subprime
mortgages, driven by rising interest rates for refinancing and falling house prices, resulted in
uncertainty surrounding the value of a number of structured credit products which had these
assets in their underlying portfolios. As a result, rating agencies downgraded many of the
related securities and announced changes in their methodologies for rating such products.
Meanwhile, structured credit mortgage-backed instruments measured by the ABS indices
(ABX) saw rapid declines, and the liquidity for initially tradable securities in their respective
secondary markets evaporated. The losses, downgrades, and changes in methodologies
shattered investors' confidence in the rating agencies' abilities to evaluate risks of complex
securities, a result of which, investors pulled back from structured products in general.

It soon became apparent that a wide range of different financial institutions had exposures to
many of these mortgage-backed securities, often off-balance sheet entities such as conduits
or structured investment vehicles (SIVs).28 Due to the increasing uncertainty with regard to
their exposure to and the value of the underlying mortgage-backed securities, investors
became unwilling to roll over the corresponding ABCP. As the problems with SIVs and
conduits deepened, banks came under increasing pressure to rescue those that they had
sponsored by providing liquidity or by taking their respective assets onto their own balance
sheets. As a result, the balance sheets of those financial institutions were particularly strained
by this reabsorption, which in addition was amplified by losses due to declining asset values.
Consequently, the level of interbank lending declined both for reasons of liquidity and credit
risk and a run for “liquidity” occurred.29 With the evaporation of liquidity in many asset-
backed mortgage securities, in particular in the United States initially, liquidity spirals
occurred where both market and funding liquidity became significantly impaired and
mutually reinforcing (GFSR, 2008; Brunnermeier, 2007). While the Libor-OIS spread, a
proxy for stress in the interbank money markets, widened during the on- set of the crisis and
under the influence of end-of-year effects in December 2007, the Lehman Brothers collapse
exposed the interbank market to even more counterparty and liquidity risk, leading market
participants to globally withdraw from these market segments. Following this event, the

27
See Kifi and Mills (2008) and Dell'Ariccia, Igan and Laeven (2008) for details on the structure of the U.S.
subprime mortgage market and the deterioration of lending standards.
28
The SIVs or conduits were funded through the issuance of short-term asset-backed commercial paper (ABCP)
in order to take advantage of a yield differential resulting in a maturity mismatch.
29
The former is based on a prudency motive whereby banks hoarded liquid assets in order to insure themselves
against contingent liabilities. In contrast, the latter was due to uncertainty with regard to the mortgage exposure
of counterparties and the inability to value their respective assets.
 66

failure of counterparties to honor the delivery of US Treasuries in repo transactions due to

inability or unwillingness drastically soared showing even more stress in funding markets.30
With interbank markets across various advanced economies becoming dysfunctional in early
August 2007, there was clear evidence of a run for “quality” by investors. For example the
price of gold, which is regarded as a storage of value during times of financial turbulence,
rose from $660 per ounce in August 2007 to $1002 around the Bear Stearns rescue by JP
Morgan and the Fed's announcement of the Primary Credit Dealer Facility on
March 16, 2008, after which the gold spot price dropped 10% in a short time.31 In addition,
there was a strong demand for 10-year US Treasuries as a ‘safe’ haven, and accordingly,
yields almost halved between the beginning of the crisis and the Bear Stearns and Lehman
Brothers episodes. The frequency of deviations from the usual bid/ask pattern of the 10-year
US Treasuries also increased.

As turbulence related to the U.S. subprime mortgages heightened, financial markets more
generally showed signs of stress, as investor preference moved away from complex
structured products in a flight to quality and liquidity, and global investors' risk appetite
sharply decreased due to a widespread re-pricing of risk (see Gonzalez-Hermosillo, 2008).
Volatility in various asset classes was affected, mirroring the widening of the Libor-OIS
spread. For instance, a structural break in the VIX index since the Lehman collapse is
apparent, with other implied volatility equity indices also revealing similar patterns. An
inspection of the at-the-money implied volatility of major financial institutions shows a very
close co-movement with their respective CDS spreads.32

Furthermore, hedge funds that held asset-backed securities and other structured products
were burdened by increased margin requirements, driven in turn by greater market volatility.
As a consequence, they attempted to offload the more liquid parts of their portfolios in order
to meet these margin calls and also respond to redemptions by investors. As argued by
Khadani and Lo (2007), quantitatively driven hedge funds were especially engaged in
liquidation sales across different asset classes, thus leading to a transmission of market stress
in the beginning of the subprime crisis. As a result, trading volumes and numbers of trades in
both the bond and the stock markets in the developed and emerging countries increased
markedly, whilst the liquidity surrounding structured investments evaporated.

30
This indicated that despite the higher supply of US Treasuries, market participants had very high demand for
US Treasury collateral and were very concerned about counterparty risk, even though governments had
implemented a systematic response by re-capitalizing major financial institutions and guaranteeing liabilities of
banks.
31
The bankruptcy of Lehman Brothers saw the gold price soar over 20% within a few weeks, as global risk
appetite dramatically deteriorated and precipitated a run for quality across asset classes and markets.
32
Humps occur at the time of the Bear Stearns rescue by JP Morgan in March 2008, during the Fannie and
Freddy bailout by the U.S. government in mid-July 2008 and around the time of Lehman's bankruptcy.
 67

Volatility also spilled over into the foreign currency markets with the carry trades starting to
rapidly unwind at the end of September 2008, whereby this breakdown was reflected by the
implied volatilities of major EM currencies. High-yielding and previous investment
currencies saw large depreciations against the U.S. dollar, while funding currencies such as
the Japanese yen benefited by a repatriation of funds into Japan. There was a scramble for
U.S. dollars, which was reflected in the higher volatility of the euro-U.S. Dollar swap rates.
Relatedly, during the crisis there has been increasing divergence from the assumption of
covered interest rate parity (CIRP). This relationship postulates that the currency forward
premium equals the interest rate differentials of the home and foreign interest rate, such that a
violation would imply possible arbitrage opportunities. The daily deviations from the CIRP
jumped at the time of the Bear Stearns rescue, and then completely broke down for various
EM currencies after Lehman's bankruptcy.

EM countries were less affected during the initial stages of the subprime crisis than advanced
economies, as for example EM equity markets peaked in November 2007. But the
persistence of the market dislocations, the deterioration of economic fundamentals in
advanced economies and rising global risk aversion significantly affected EM countries by
late 2008. In particular, flows to EM equity and debt mutual funds turned negative. Total
foreign assets held by the former peaked in November 2007, but investments in the
equivalent EM debt mutual funds began to fall rapidly beginning in September 2008,
driven by the sharp fall in global risk appetite after the Lehman collapse and fear that EM
economies would be affected by the looming recession in advanced economies. Equity
markets in EM countries saw their gains from the previous boom years wiped out in a short
period of time. Relatedly, while EM corporate spreads (over treasuries) gradually began to
increase following the onset of the subprime crisis, they escalated sharply across the various
EM regions after the Lehman bankruptcy. Similar behavior can be observed for the cost of
corporate credit, especially for high-yield bonds, in the U.S. and Europe. Sovereign spreads
and the costs of insuring against a sovereign default, CDS, soared across a wide range of EM
countries as portfolio outflows and a flight to quality accelerated.

EM countries with large current account deficits and whose banks prior to the crisis have
been most reliant on foreign wholesale funding have been affected the most by the
ramifications of the financial crisis. For instance, the IMF provided substantial financial
support to the Ukraine and Hungary (October 2008), Pakistan (November 2008) and Latvia
(December 2008). EM countries such as South Korea and Russia which had built up large
foreign reserves prior to the crisis increasingly had to employ these in order to stem the
currency depreciation pressures arising from an unwinding of portfolio positions and capital
flight as well as severe strains in their banking sectors.

Initial financial spillovers to EM countries quickly morphed into real sector problems,
whereby economies reliant on declining demand and available trade finance saw their
 68

domestic industrial production and GDP growth rates plunging. In order to counteract the
looming adverse real sector impacts as well as to provide liquidity and credit support to the
domestic banking systems, large fiscal stimulus plans were implemented, such as in China
for over $500 billion in November 2008.

Interestingly, emerging market equity, fixed income and currency markets already saw a
sharp sell-off in February 2007, a relatively short-lived episode, but it revealed how fast and
broad-based a worldwide reappraisal of risk and flight to quality can occur. Starting in late-
February 2007, there was a significant correction in the Shanghai stock market due to an
unwinding of large long equity positions. This reverberated across emerging and mature
markets. At the same time, the price of the ABX (BBB) index (based on CDS written on
subprime mortgages, investment grade tranche) began to decline whilst the outlook on the
U.S. housing market worsened further (see also GFSR (2007)). In particular, carry trades in
high-yielding currencies such as in Brazil, South Africa and Turkey, were rapidly unwound,
causing them to decline and the yen to appreciate. In addition, implied volatilities across a
range of other asset markets, notably fixed-income and equity, sharply increased and stock
markets in previously booming economies such as China, Malaysia, Philippines or Turkey
observed the largest declines. The fall in global risk appetite was broad-based without much
differentiation across regions. Compared to equity markets, sovereign spreads across EM
countries did move in tandem with the general market direction but were less affected.

In the following section our approach to examine the financial interlinkages between
advanced and EM countries during the global financial crisis is presented. In this context, we
focus on the co-movements of a number of pertinent key financial variables such as equity
market and sovereign spreads in EM economies.

C. Data

As mentioned above, in this paper, we use the daily 3-month US dollar Libor-overnight index
swap (OIS) spread as a measure for bank funding liquidity and general stress in the interbank
money market.33 With the onset of the subprime crisis in the summer of 2007, this market
segment exhibited severe dislocation. In addition, S&P 500 stock market returns are included
in the reduced form model, controlling for common shocks. Moreover, the variance serves as
a proxy for market volatility.

As a measure of the default risk of large complex financial institutions, we use the average
credit default swap spread (CDS) of a number of banks, namely those of Citigroup, Bank of
America, JP Morgan, Wachovia, Merill Lynch, Morgan Stanley, Goldman Sachs, Lehman

33
Funding liquidity refers to the availability of funds such that a solvent agent is able to borrow in the market in
order to service his obligations.
 69

Brothers, HSBC, Royal Bank of Scotland, UBS and Deutsche Bank.34,35 Regarding EM
financial variables, EMBI+ spreads for the regions Latin America (LAC), Europe and Asia
are used as a measure of their respective sovereign risks. In terms of individual countries, we
analyze the potential financial-to-financial spillovers to prominent emerging market countries
with open capital accounts which have seen a significant impact due to the financial crisis.
Amongst these are Brazil, Russia, Turkey as well as Mexico and South Africa. We also relate
the advanced economy indicators such as the Libor spread and stock market returns to the
CDS spreads in Brazil, Russia and Turkey. This allows us to analyze co-movements between
default measures of sovereign risk in EM countries and financial stress in advanced
economies.

The data sample encompasses January 3rd 2003 until December 31st 2008. We conduct unit
root tests for the crisis period and formally identify nonstationarity in the data. Therefore,
first differences of the spreads are taken, such that they can be applied to the estimation
framework set out below.

As previously argued, the three main indicators capturing financial stress in the U.S. and
other advanced economies are provided in Figure 9. Funding liquidity pressures in the
interbank market, as measured by the Libor-OIS spread, were negligible prior to the
subprime crisis, after which this proxy drastically increased in late July 2007. Following
central bank interventions in mid August, the Libor spread subsided somewhat before
widening again sharply, driven in part by end-of-year effects as well as by increased losses
and writedowns of major financial institutions. In the run-up to the Bear Stearns rescue,
heightened funding liquidity pressure again became evident, and finally, the Lehman failure
led to an almost breakdown of the interbank money market with a massive dollar shortage
and with margins and haircuts rising across the board, as well as a sharp increase in
counterparty risk.

The S&P 500 peaked in October 2007 but has seen temporary corrections during the
Shanghai stock market crash in late February 2007 as well as in the beginning of the
subprime market turmoil in July 2007. Sharp falls occurred during January 2008 with
financial institutions announcing new writedowns and losses, before the Bear Stearns rescue
in March and after the Lehman collapse in September 2008. Meanwhile, the CDS measure of
LCFIs is characterized by the two spikes, namely during the Bearn Stearns rescue as well as
the Lehman bankruptcy.

34
After the Lehman Brothers collapse, we use the average CDS values for Goldman Sachs, Merrill Lynch and
Morgan Stanley for the Lehman Brothers time series data.
35
Note also that market-traded prices such as CDS spreads contain a liquidity risk component—the risk that an
investor may or may not be able to trade at a price close to the last traded price.Such risks rise during periods of
stress.
 70

Figure 9. U.S. and EM Financial Variables

Source : Bloomberg L.P.

The financial variables for the emerging markets are briefly discussed below. While the
regional EMBI+ spreads for Asia and LAC have remained elevated relative to those of
Europe, between 2003 and 2006 convergence to historically low levels has been observed.
With the onset of the subprime crisis, some moderate widening occurred. Following the
Lehman bankruptcy, global risk aversion sharply increased across asset classes and the
regional EMBI+ spreads jumped to over 800 basis points in late October 2008. Since then
 71

some tightening has been recorded but spreads still remained at very high levels compared to
the pre-subprime period.

EMBI+ and CDS spreads for individual countries such as Brazil, Russia and Turkey exhibit
similar patterns of being compressed before the subprime crisis and then suffering increasing
widening whilst the crisis period unfolded. Stock markets in these countries continued their
upward trend well into 2007 and 2008 with Turkey peaking in October 2007 and Brazil and
Russia in May 2008, before the contagious reversal thereafter. During the financial crisis,
these markets appear to move increasingly in tandem as events unfold in advanced
economies. While relatively small compared to post-Lehman movements, these equity
markets were also affected by the brief Shanghai stock market correction in February 2007,
which resulted in temporary large drops in these equity indices. Finally, it is shown in
Figure 10 that the bond and CDS spreads, as well as equity markets in Mexico and South
Africa follow similar price dynamics compared to those outlined above.
 72

Figure 10. U.S. and EM Financial Variables

Source : Bloomberg L.P.

D. Methodology

We use a multivariate GARCH framework for the estimation, which allows for
heteroskedasticity of the data and a time-varying correlation in the conditional variance.
Specifically, the Dynamic Conditional Correlation (DCC) specification by Engle (2002) is
adopted, which provides a generalization of the Constant Conditional Correlation (CCC)
model by Bollerslev (1990).36 These econometric techniques allow us to analyze the co-
movement of markets by inferring the correlations of the changes in the spreads discussed
above, which in turn is essential in understanding whether the recent episode of financial
distress has become systemic.

36
Given the high volatility movements during the recent financial crisis, the assumption of constant conditional
correlation among the variables in the CCC model is not very realistic especially in times of stress where
correlations can rapidly change. Therefore, the DCC model is a better choice since correlations are time-
varying.
 73

The DCC model is estimated in a three-stage procedure. Let rt denote an n x 1 vector of asset
returns, exhibiting a mean of zero and the following time-varying covariance:

Here, Rt is made up from the time dependent correlations and Dt is defined as a diagonal
matrix comprised of the standard deviations implied by the estimation of univariate GARCH
models, which are computed separately, whereby the ith element is denoted as hit . In other
words in this first stage of the DCC estimation, we fit univariate GARCH models for each of
the five variables in the specification. In the second stage, the intercept parameters are
obtained from the transformed asset returns and finally in the third stage, the coefficients
governing the dynamics of the conditional correlations are estimated. Overall, the DCC
model is characterized by the following set of equations (see Engle, 2002, for details):

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset
returns rt. As defined above, Rt is the time varying correlation matrix and is a function of Qt,
which is the covariance matrix. In the matrix Qt,ι is a vector of ones, A and B are square,
symmetric and  is the Hadamard product. Finally, λi is a weight parameter with the
contributions of Dt21 declining over time, while κ i is the parameter associated with the
squared lagged asset returns. The estimation framework is the same as in Frank, Gonzalez-
Hermosillo and Hesse (2008).

E. Results

The findings in this paper suggest that the implied correlations between the 3-month US
Libor-OIS spread and EMBI+ bond spreads of Asia, Europe and LAC sharply increase after
the subprime crisis (left column of Figure 11). In addition, the China stock market correction
in late February 2007 led to a temporary spike of the correlation measures from 0.20 to
almost 0.50. The Lehman collapse caused the largest increase of co-movements between
these variables. In terms of regional differences, the Asian EMBI+ spread exhibits the largest
correlation for the pre-subprime crisis with some exceptions, followed by Europe and LAC.
All regional spread's co-movements jump up around the China stock market burst in similar
magnitude and move closely during the subprime period. Interestingly, the correlation for the
LAC EMBI+ spread exhibits the greatest rise immediately following the Lehman failure,
compared to those of the other regions.
 74

These results are mirrored in the other multivariate GARCH specifications for the EMBI+
spreads. The relationship between changes in the S&P 500 and the EMBI regional bond
spreads abruptly changes during the Shanghai stock market correction with correlation
magnitudes moving to almost -0.60. Subsequently, the degree of co-movement remains
elevated following the beginning of the subprime crisis and peaks in September 2008. In
terms of regional differences, it appears that the interlinkages between the S&P 500 and the
EMBI spread for LAC dominate the other regional spreads. In addition, the relationship
between the CDS default risk measure and the regional bond spreads highlights a substantial
and persistent increase in correlations beginning in July 2007 and magnitudes remaining high
throughout the crisis period.

In the right column of Figure 11 we examine possible individual country interlinkages with
the Libor-OIS spread. Changes in this measure are related to sovereign bond and CDS
spreads, as well as with stock markets in Brazil, Russia and Turkey. As before, the China
episode in February 2007 is evident, such as market dislocations during the subprime crisis
and the Lehman collapse. The Bear Stearns rescue in March 2008 also becomes visible with
co-movements sharply reversing their downward trend prior to that. Brazil has the largest
correlation for the bond spread, CDS and stock market volatility measures during the crisis
period. This potentially can be attributed to the fact that Brazil has a very open capital
account and has witnessed a dramatic increase in bond risk premia coupled with large foreign
equity outflows precipitating a plunge in the domestic equity market, despite obtaining an
investment grade rating in 2008. Given the relative liquidity of foreign bond and equity
markets in Brazil, mutual and hedge funds were able to unwind these positions to cover their
domestic losses (or margin calls in some cases).
 75

Figure 11. Implied Correlations between U.S. and EM Financial Variables

Source: Own calculations.

Above, the correlations between the U.S. Libor-OIS spread, a proxy for funding liquidity
stress, and bond risk premia and equity market volatility in Brazil, Russia, and Turkey were
quantified. In what follows, we extend this analysis to include CDS changes for LFCIs
during the recent crisis period. In the left column of Figure 12 spikes in the correlation
patterns are less pronounced than in the case of the Libor-OIS spread. Simultaneously, the
 76

co-movement among the CDS measures is highly persistent suggesting that both liquidity
and solvency aspects were central in explaining financial market spillovers.
Finally, the DCC GARCH analysis is extended to the countries Mexico and South Africa.
Results in the right column of Figure 12 indicate similarities to the findings from Brazil,
Russia and Turkey with co-movements significantly increasing during the subprime crisis
period, while providing some evidence that the China stock market correction also led to
temporally higher correlations. As expected and given the proximity of Mexico to the United
States, co-movements of the Mexican financial variables are more pronounced with the
proxies for stress in the interbanking market, stock market volatility as well as default risk
than of South Africa.

Overall, the findings from the DCC GARCH models indicate that the notion of possible de-
coupling (in the financial markets) had been misplaced. Despite EM stock markets reaching
their peak in November 2007, interlinkages between funding stress and equity markets in
advanced economies and EM financial indicators became highly correlated and have seen
sharp increases during specific crisis moments. Given the interconnectedness of global
financial markets, investors' increase in global risk aversion from problems in advanced
economies rapidly spilled over into EM countries, as funds were pulled out from the latter
and subsequently invested into the safest and most liquid assets such as mature market fixed
income securities.

In addition, co-movements between funding stress and bank default risk (proxied by the CDS
measure) in advanced economies with bond spreads as well as stock market returns in EM
countries have been fairly similar in terms of their magnitudes during the financial crisis. As
a result, we believe that these factors are important when analyzing potential financial market
spillovers.
 77

Figure 12. Implied Correlations between U.S. and EM Financial Variables

Source: Own calculations.

 78

F. Conclusion

In this paper the interaction between liquidity and bank solvency measures with stock, bond
and credit markets in EM economies is analyzed during the Global Financial Crisis. A
multivariate GARCH model is estimated in order to quantify the extent of co-movement of
these financial variables across markets. We find that during the recent period of financial
turbulence both the Libor-OIS spread, a proxy of interbank money market pressure, and
the CDS spread, a measure of bank solvency, became more correlated with EM bond, stock
and credit markets. These relationships become especially apparent during the Shanghai
stock market correction, the beginning of the subprime crisis in summer 2007, the Bear
Stearns rescue and the Lehman Brothers bankruptcy. As a result, we provide evidence that
the notion of possible decoupling of financial markets has been misplaced.

In this paper, we did not analyze the exact causal relationships among the financial variables
in advanced as well EM countries. With daily high-frequency data, there are likely to be
significant feedback loops that can affect the causal relationships. We reserve this topic for
future research. One of the key policy implications of the paper is that spillovers need to be
closely attended to especially in light of the interconnectedness of global financial markets.
 79

References

Beirne, J., G. M. Caporale, M. Schulze-Ghattas, and N. Spagnolo, 2008, “Volatility

Spillovers and Contagion from Mature to Emerging Stock Markets,” IMF Working
Paper 08/286 (Washington: International Monetary Fund).

Bekaert, G, C.R. Harvey, and A. Ng, 2005, “Market Integration and Contagion,” Journal of
Business, Vol. 78, No. 1, pp. 39–9.

Capiello, L., R. F. Engle, and K. Sheppard, 2006, “Asymmetric Dynamics in the Correlations
of Global Equity and Bond Returns,” Journal of Financial Econometrics, Vol. 4,
No. 4, pp. 537–72.

Caporale, G.M., N. Pittis, and N. Spagnolo, 2006, “Volatility Transmission and Financial
Crises,” Journal of Economics and Finance, Vol. 30, No.3, pp.376–90.

Coudert, V., and M. Gex, 2007, “Does Risk Aversion Drive Financial Crises? Testing the
Predictive Power of Empirical Indicators,” Centre d’Etudes Prospectives et
d’Informations Internationales, Working Paper No. 2007–02.

Dell’Ariccia, G., D. Igan, and L. Laeven, 2008, “Credit Booms and Lending Standards:
Evidence from the Subprime Mortgage Market, (unpublished; Washington:
International Monetary Fund)

Dornbusch, R., Y.C. Park, and S. Claessens, 2000, “Contagion: How it Spreads and How it
can be Stopped,” World Bank Research Observer, 15, pp. 177–97.

Dungey, M., R. Fry, B. González-Hermosillo, and V. Martin, 2006, “Contagion in

International Bond Markets during the Russian and LTCM Crises,” Journal of
Financial Stability, Vol. 2, pp. 1–27.

Dungey, M., R. Fry, B. González-Hermosillo, and V. Martin, 2005, “Empirical Modeling of

Contagion: A Review of Methodologies,” Quantitative Finance, Vol. 5, pp. 9–24.

Engle, R. 2002, “Dynamic Conditional Correlation: A Simple Class of Multivariate

Generalized Autoregressive Conditional Heteroskedasticity Models,” Journal of
Business & Economic Statistics, Vol. 20, pp. 339–50.

Forbes, K., and R. Rigobon, 2002, “No Contagion, Only Interdependence: Measuring Stock
Market Co-movements,” Journal of Finance, Vol. 57, No. 5, pp. 2223–61.

Federal Reserve Bank of Dallas, 2008, Economic Letter, “Fed Intervention: Managing
 80

Moral Hazard in Financial Crises,” Vol. 3, No. 10 (October).

Frank, N., B. González-Hermosillo, and H. Hesse, 2008, “Transmission of Liquidity Shocks:

Evidence from the 2007 Subprime Crisis,” IMF Working Paper 2008/200
(Washington: International Monetary Fund).

Frank, N., and H. Hesse, 2009, “Financial Spillovers to Emerging Markets during the Global
Financial Crisis,” IMF Working Paper 2009/104 (Washington: International
Monetary Fund). Also published in the Czech Journal of Economics and Finance,
2009, 59(6), 507-521

González-Hermosillo, B., 2008, “Investor’s Risk Appetite and Global Financial Market
Conditions” (Washington: International Monetary Fund).

Gorton, G., 2008, “The Panic of 2007,” Prepared for the Federal Reserve Bank of Kansas
City, Jackson Hole Conference, August.

International Monetary Fund, 2009, Global Financial Stability Report, “Warnings of

Systemic Risk” (Chapter 3), World Economic and Financial Surveys (Washington,
April).

Kaminsky, G.L., and C.M. Reinhart, 2003, “The Center and the Periphery: The
Globalization of Financial Turmoil,” NBER Working Paper No. 9479 (Cambridge:
National Bureau of Economic Research)

Khandani, A. E., and A. Lo, 2007, “What Happened to the Quants in August 2007?” MIT
Working Paper (Cambridge, Massachusetts: Massachusetts Institute of Technology,

Kiff, J., and P. Mills, 2008, “Money for Nothing and Checks for Free: Recent Developments
in U.S. Subprime Mortgage Markets,” IMF Working Paper 07/188 (Washington:
International Monetary Fund).

King, M., E. Sentana, and S. Wadhwani, 1994, “Volatility and Links between National
Stock Markets,” Econometrica, Vol. 62, pp. 901–33.

King, M., and S. Wadhwani, 1990, “Transmission of Volatility between Stock Markets,”
Review of Financial Studies, Vol.3, No.1, pp. 5–33.

Kodres, L., and M. Pritsker, 2002, “A Rational Expectations Model of Financial Contagion,”
Journal of Finance, Vol. 57, pp. 769–99.

Pericoli, M., and M. Sbracia, 2003, “A Primer on Financial Contagion,” Journal of Economic
 81

Surveys, Vol. 17, pp. 571–608.

IV. GLOBAL MARKET CONDITIONS AND SYSTEMIC RISK, WITH BRENDA GONZÁLEZ -
HERMOSILLO, 200937

This chapter examines several key global market conditions, such as a proxy for market
uncertainty and measures of interbank funding stress, to assess financial volatility and the
likelihood of crisis. Using Markov regime-switching techniques, it shows that the Lehman
Brothers failure was a watershed event in the crisis, although signs of heightened systemic
risk could be detected as early as February 2007. In addition, we analyze the role of global
market conditions to help determine when governments should begin to exit their
extraordinary public support measures.

A. Introduction

The IMF has been recently called upon by the international community to deepen its work on
systemic risk, on identification of systemically important institutions and markets, as well as
on developing early warning signals of distress. An important element of anticipating and
identifying systemic events are the role played by underlying “market conditions” and the
ability for events to subsequently further alter fragile market conditions. At its most basic
level, the value of the assets on the books of financial institutions is highly dependent on the
underlying financial environment. When the level of market uncertainty (measured, for
example, by the implicit volatility of asset prices) is high, then even a temporary shock can
lead to defaults and generate significant negative aftershocks, including liquidity spirals
(Brunnermeier and Pedersen, 2009). Similarly, when investors’ risk appetite is low or global
liquidity is tight, even relatively small shocks can have large effects on global financial
markets.

The observation that general market conditions matter for the existence and propagation of
risks through the financial system can be used to examine periods of high vulnerability to
shocks that may become systemic. While most of the tools used to compute early warning
indicators of crisis rely on an assessment of vulnerabilities (as measured, for example, by
large current account deficits or high debt levels), often those vulnerabilities can persist for a
long time, even years, before a crisis occurs. Abrupt changes in market conditions are often
the “trigger” that sets off a financial crisis. Moreover, some of the measures of financial
fragility that are widely used (including credit default swap spreads, but also some more
sophisticated reduced-form models of financial instability) are typically unable to separate
the idiosyncratic credit risk component associated with the potential default of an individual
institution from the intrinsic premium that is attached to the overall distress in market
conditions. Finally, it is also useful to identify when market conditions signal a regime

37
This chapter is based on González-Hermosillo and Hesse (2009).
 82

change such that tranquil periods turn into medium or high volatility states, or when they
reverse to more tranquil periods and the exit of supporting government interventions can be
considered.

We adopt regime switching models using variables that proxy for several key global market
conditions. In particular, the VIX is used as a proxy for market uncertainty,38 the three-month
TED spread as a measure for stress in the interbank market,39 and the euro-U.S. dollar forex
swap as an indicator of U.S. dollar funding pressures in international financial markets.40

We first look qualitatively at the behavior of some global market variables in advanced and
emerging market economies during the financial crisis before presenting the formal findings
of the regime switching models. The analysis presented is focused on two periods, with the
first one concentrating on the events and market signals that led to the peak of the global
crisis in the aftermath of the Lehman collapse on September 15, 2008. This marked a
watershed event that led to rapid spillovers to emerging market economies, sharply
increasing uncertainty across asset markets, a scramble for U.S. dollars with the breakdown
of the carry trade, and the need for financial institutions to refinance their U.S. dollar
positions. The regime switching models indicate a move towards a high volatility state before
the Lehman episode, which are consistent with elevated systemic risks in the financial
system.

The second period of analysis is that which followed massive government intervention in a
number of countries in support of their financial systems. The examination of the most recent
period is of interest because global market conditions and economic activity appear to have
improved since mid-2009, potentially signaling that exit strategies can begin to be
contemplated. The results suggest that, although market conditions have improved
substantially since the spring 2009, the regime switching models examined still signal
moderate pressures on certain market conditions. In particular, the VIX and the forex swap
market both signal a high probability of a medium-volatility regime. In contrast, the TED
spread strongly signals a high probability of being in a low-volatility state.

The paper is organized as follows: Section B provides on overview of systemic risk, while
Section C discusses the role of global market conditions and systemic risk from a qualitative

38
The VIX, the Chicago Board Options Exchange volatility index, is a measure of the implied volatility of the
S&P 500 index options over the next 30 days and is calculated from a weighted average of option prices.
39
The TED spread is the difference between the three-month LIBOR and the three-month treasury bill rate.
40
The spillovers from the interbank market to the foreign exchange swap market has led to periods whereby
foreign exchange swap prices deviated from that implied by covered interest parity conditions. This highlights
the international interconnectedness of banks’ funding requirements through foreign exchange swap markets
and the potential for banks’ inability to obtain funding liquidity.
 83

perspective. The methodology and results are presented in Section D, and we offer some
concluding remarks in Section E.
 84

B. Overview of Systemic Risk

Before evaluating tools that can be useful to detect and measure “systemic risk,” one needs to
define it. “Systemic risk” is a term that is commonly and broadly used. However, it has so far
resisted formal definition and quantification. Indeed, systemic risk typical reflects a sense of
a “broad-based” breakdown in the financial system which is normally realized (ex-post) by a
large number of failures of financial institutions (usually banks). Similarly, a systemic
episode may be simply seen as a extremely acute financial crisis, even though the degree of
severity of the financial stress has proven difficult, if not impossible, to measure.41 Systemic
risk is also viewed as a phenomenon not only measured by its intensity, but also by the breath
of its reach across markets and countries.

Although systemic events are intrinsically related to an aggregate measure of risk, it is

ultimately built up by its components. In this sense, a natural starting point is to begin by
examining the characteristics of individual financial institutions. Some financial institutions
may be simply too-interconnected or large enough that are systemically important. However,
the sum of the fragilities of individual financial institutions may not necessarily equate to the
fragility of the global financial system. Indeed, a body of the literature has made the case that
systemic events are often characterized by contagion, whereby financial crises in some
markets spillover to others in ways that are not characteristic of non-crisis periods. In
particular, channels over and above the market fundamental mechanisms that link countries
and asset markets during non-crisis periods appear only during contagious episodes.42 The
classic example of contagion in the earlier literature is that of bank panics resulting from
information asymmetries. In particular, depositors withdraw their funds simultaneously from
banks, creating a “run” on bank assets that ultimately leads to multiple bank failures (even
from banks that were considered to be sound prior to the run).43 Since the establishment of
deposit insurance schemes in the United States in 1934 and in other advanced economies
several decades ago, and more recently in practically all developing countries following the
Asian crisis in 1996, deposit “runs” have become less frequent than before. However, the

41
Some recent attempts to measure the degree of severity of financial stress in a given country include Illing
and Liu (2006). However, most empirical analyses of multi-country financial crises rely on a binomial notion
whereby the dependent variable takes the value of one during the known crisis period or zero otherwise (e.g.,
Kaminsky and Reinhart (1999); Hardy and Pazarbasiouglu (1999); and Demirguç-Kunt and Detragiache (1998))
with no information about the actual severity of the crises.
42
See, for example, Masson (1999); Dornbush, Park and Claessens (2000); and Dungey, Fry, González-
Hermosillo, and Martin (2005, 2006, 2007). Dungey, Fry, González-Hermosillo, and Martin (forthcoming)
argue that the LTCM/Russian crisis in 1998 and the subprime crisis that began in mid-2007 have been the most
contagious crises in the past decade, based on a sample of advanced and emerging economies whereby credit
and equity market daily data are modeled jointly across countries.
43
Deposit runs are formally modeled as a result of asymmetries of information by Diamond and Dybvig (1983).
More recently, Laeven and Valencia (2008) suggest blanket deposit guarantees to address information
asymmetries.
 85

recent financial crisis that began in mid-2007 has brought to the forefront other types of bank
runs. These more recent “runs” have encompassed runs on wholesale funding (including the
inability to roll roll over funding for banks’ off-balance sheet special purpose vehicles), on
interbank lending, 44 and on banks’ market equity values.45

At the heart of most of these approaches is the notion that “systemic” events are somehow
self-reinforcing and bring into play additional linkages across countries and markets that only
exist during systemic episodes. For example, a liquidity shock which is temporary by nature,
can metamorphose into a default event which further affects liquidity conditions.46 This view
intrinsically suggests that the degree of vulnerability of individual financial institutions may
be related to the degree of stress in global market conditions. For example, when the level of
market uncertainty (measured by the implicit volatility of assets) is high, then even a
temporary liquidity shock can lead to defaults and have exponential aftershocks. Similarly,
when investors’ risk appetite is low or global liquidity is tight, then even relatively small
shocks can have large effects on global financial markets—and vice-versa. 47

From the risk managers’ perspective, systemic risk is often viewed in the context of hedging
positions, such that a diversified portfolio should be able to neutralize market risks. Systemic
events in this context are often measured in practice by the degree of correlation of the
various markets in a diversified portfolio; the higher the correlation among assets, the less
ability that market participants have to hedge the odds. Market participants usually look at
correlations during “normal” times as a gauge to compare correlations when asset markets
become synchronized during periods of stress. Cluster analysis is a tool often used to rank
events based on their degree of correlation. However, from a statistical perspective, observed
high correlations could be simply a function of the non-constant volatility characteristic of
the shocks.48 Thus, basic measures of correlation need to control for the fact that the volatility
of the shocks is typically autocorrelated (heteroskedasticity) during crises periods.

More generally, systemic financial risk is also connected to the notion that there is a regime
shift such that normal periods, and even “contained” crises, become fundamentally different
at some point. In addition to changes in correlations among assets, it may be that higher
moments (including skewness and kurtosis) in the probability density function of the data
generating process also change during systemic events. Complicating this analysis is the fact
that policy actions (such as providing guarantees, capital injections, and various other crisis

44
Issues associated with banks’ restricted interbank lending are examined in IMF (2008).
45
These more recent types of “runs” are discussed by Gorton (2008).
46
Brunnermeier and Pedersen (2009) model this process for the current financial crisis.
47
Different measures of risk appetite are discussed in González-Hermosillo (2008).
48
As argued by Forbes and Rigobon (2002).
 86

resolution schemes) can alter the outcome of events by reducing systemic risks. Some
government actions (including lack of a comprehensive crisis resolution strategy) can also
actually increase systemic risks.

Finally, systemic risk is sometimes reserved for events that trigger a loss of economic value
or confidence in a substantial portion of the financial system that is serious enough to have
significant potential effects on a country’s real economy.49 Similarly, the real sector effects
can be extended to span not only a certain country or group of countries, but most of the
world through direct financial links (such as financial institutions’ parent-subsidiary
connections, cross-border financial flows, trade finance affecting global trade), or indirect
links (rebalancing of investors’ portfolios, deleveraging, and re-pricing of relative risks as
some debt becomes guaranteed but other is not and therefore becomes less attractive). Even
under confidence and lack of information can work as catalysts to spread systemic risks.

Clearly, it is difficult to ascertain when financial crises become “systemic,” and how to
measure the degree of systemic risk. Given the growing complexity and interconnectivity of
the global financial system, it is a daunting task to expect to arrive at a single measure of
systemic risk. Indeed, systemic financial risk may be best viewed as a collection of
measures.50 However difficult it is to measure financial systemic risks, and all the caveats
associated with any such measures, knowing when the system switches into systemic gear is
of critical importance for policymakers. In particular, it is unfeasible to expect to manage
systemic risk if it cannot be gauged, even if imperfectly.

This paper examines financial “systemic” risk from one particular angle in search for
indicators that would have been useful in anticipating systemic events during the current
crisis. The approach chosen to assess systemic risk is to examine regime changes in global
market conditions. This information can be also invaluable in determining when market
conditions have improved sufficiently for policymakers to begin to “exit” the interventions
that were previously introduced to provide support to the financial system in a systemic
crisis.

C. Global Market Conditions and Systemic Risk: A Qualitative View

With interbank markets across various advanced economies becoming clogged in early
August 2007, there was clear evidence towards a “run for quality” by investors. For
example, the gold spot price, which is often used as a crude measure of storage of value,
started its continuous increase in early August 2007 from $660 per ounce and reached its
peak of $1002 around the Bear Stearns rescue by JP Morgan and the Fed’s announcement of

49
Group of Ten (2001).
50
Lo (2009), for example, considers that “systemic” risk should be measured by leverage, liquidity, correlation,
concentration, sensitivities, and connectedness.
 87

the Primary Credit Dealer Facility on March 16, 2008, after which time the gold spot price
dropped 10 percent in a short time.51 In addition, there was a strong demand for 10-year U.S.
Treasuries as a “safe heaven,” and yields almost halved between the onset of the crisis in
August 2007 and the Bear Stearns and Lehman episodes. The bid-ask spread deviated
frequently from its usual pattern.

The “run for quality” was also accompanied by a “run for liquidity.” With liquidity
evaporating in many asset-backed securities, liquidity spirals occurred with a lack of both
market and funding liquidity interacting, significantly impairing funding and asset-backed
markets (IMF, 2008 and Frank, González- Hermosillo, and Hesse, 2008 ). While the Libor-
OIS spread, as a proxy for funding liquidity and general stress in the interbank markets, has
been subject to various humps such as at the onset of the crisis and with end-year effects in
December 2007, the Lehman collapse exposed the interbank market to heightened
counterparty and liquidity risk concerns, with market participants across the world
withdrawing from these market segments. Many central banks had to inject liquidity and, in
effect, substituted for the interbank market. There was a shortage of high-quality collateral
for posting with the central bank with haircuts increasing across Treasury securities and risky
assets. After the Lehman collapse, the failure of counterparties to deliver U.S. Treasuries to
other parties in repo transactions, due to inability or unwillingness drastically soared.52

Volatility in various asset classes was also affected, mirroring the humps of the Libor-OIS
spreads. For instance, a structural break of the VIX since the Lehman collapse is apparent
with other implied volatility equity indices also revealing similar patterns. Volatility also
spilled over into the foreign currency markets with the carry trade starting to rapidly unwind
at the end of September 2008. The implied volatilities of major emerging market (EM)
currencies (based on option prices) were reflecting this breakdown in the carry trade. High-
yielding and previous investment currencies saw large depreciations against the U.S. dollar,
while funding currencies such as the Japanese yen benefited from a repatriation of funds.
There was a scramble for U.S. dollars, which was reflected in the higher volatility of the
euro-U.S. Dollar swap rates. Relatedly, the assumption of covered interest rate parity (CIRP)
has been also violated during the crisis.53 The daily deviations from the CIRP jumped at the
time of the Bear Stearns rescue, and then completely broke down for various EM currencies
after Lehman’s bankruptcy.

51
The bankruptcy of Lehman Brothers saw the price of gold soar over 20 percent within a few weeks, as global
risk appetite dramatically deteriorated and precipitated a run for quality across asset classes and markets.
52
This indicated that despite the higher supply of U.S. Treasury bonds, market participants had very high
demand for U.S. Treasury collateral and were concerned about counterparty risk, even though governments had
announced plans to re-capitalize major financial institutions and guarantee bank liabilities.
53
The CIRP postulates that the currency forward premium equals the interest rate differentials of the home and
foreign interest rate covering the same time period. A violation would indicate possible arbitrage opportunities.
 88

EM countries were less affected in the initial stages of the subprime crisis than countries in
the epicenter; for example, EM equity markets peaked in November 2007. But the
persistency of the crisis, the deterioration of economic fundamentals in advanced economies
and the rise of global risk aversion, hit EM countries with full force in late 2008 after the
Lehman collapse (see also Frank and Hesse, 2009). In particular, flows to EM equity and
debt mutual funds turned negative. Total foreign assets in EM equity mutual funds peaked in
November 2007, but assets in the equivalent EM debt mutual funds only began to fall rapidly
beginning in September 2008, driven by the sharp fall in global risk appetite after the
Lehman collapse and fear that EM economies would be affected by the looming recession in
advanced economies. While EM corporate spreads (over Treasuries) gradually began to
increase following the onset of the subprime crisis, they escalated sharply across the various
EM regions after the Lehman bankruptcy. Similar behavior can be observed for the cost of
corporate credit, especially high-yielding, in the United States and Europe.

EM countries with large current account deficits and whose banks prior to the crisis have
been most reliant on foreign wholesale funding have been affected the most by the
ramifications of the financial crisis. Initial financial spillovers to EM countries quickly
morphed into real sector problems, whereby economies reliant on declining demand and
available trade finance saw their domestic industrial production and GDP growth rates
plunging. In order to counteract the looming adverse real sector impacts as well as to provide
liquidity and credit support to the domestic banking systems, large fiscal stimulus plans were
implemented beginning in late-2008.

Interestingly, emerging market equity, fixed income and currency markets saw a sharp sell-
off in February 2007, a relatively short-lived episode, but it revealed how fast and broad-
based a worldwide reappraisal of risk and flight to quality can occur. Starting in late-
February 2007, there was a significant correction in the Shanghai stock market due to an
unwinding of large long equity positions. This reverberated across emerging and mature
markets. At the same time, the price of the ABX (BBB) index (based on CDS written on
subprime mortgages, investment grade tranche) began to decline while the outlook on the
U.S. housing market worsened further. In particular, carry trades in high-yielding currencies
such as in Brazil, South Africa, and Turkey, were rapidly unwound, causing them to decline
and the yen to appreciate. In addition, implied volatilities across a range of other asset
markets, notably fixed-income and equity, sharply increased and stock markets in previously
booming economies such as China, Malaysia, Philippines, or Turkey observed the largest
declines. The fall in global risk appetite was broad-based without much differentiation across
regions. Compared to equity markets, sovereign spreads across EM countries did move in
tandem with the general market direction but were less affected.

Since the spring of 2009, bolstered by central bank interventions, fiscal stimulus packages
and a nascent pick-up of economic activity in advanced and developing countries, global
market conditions have improved. For instance, volatility across asset classes has
significantly subsided (albeit still at higher levels than the pre-crisis), and especially
 89

emerging markets have benefited from an improvement in global risk appetite. While a
qualitative analysis is a useful starting point to examine the role of key global market
variables in systemic risks, a more formal systematic analysis follows below.

D. Markov-Regime Switching Analysis

We use Markov-regime switching techniques to examine financial stress in a formal way.

Given the intrinsic volatility of high-frequency financial data, especially during periods of
stress, the ARCH Markov-Switching model (SWARCH) by Hamilton and Susmel (1994) is
chosen here because it can differentiate between different volatility states, for example, low,
medium, and high. In particular, univariate SWARCH models are adopted with variables in
first differences to account for the non-stationarity of the variables.

In general, the parameters of the ARCH process can alter. Equation (1) below describes a
Markov chain with yt being a vector of observed variables and s t denoting a unobserved
random variable with values 1, 2, …, K that as a state variable governs the conditional
distribution of yt .

Prob ( st  j | st 1  i, st 2  k ,..., yt 1 , y t 2 ,...)  Prob ( s t  j | s t 1  i )  p ij (1)

It is possible to combine all the transition probabilities pij in a K  K transition matrix. In

our SWARCH framework, the mean equation is an AR(1) process and the variance is time-
varying with the ARCH parameters being state dependent. Formally, the AR(1) process
follows

y t    yt 1   t (2)

The time varying variance ht2 with the error term  t is parameterized as

 t  g S  ~t
 t

~
 t  ht   t (3)
 2 ~2 ~2 ~2 ~2
ht  a0  a1 t 1  a 2  t  2  ...  a q  t  q    d t 1   t 1 ,

where  t ~ N 0,1 , S t  1,2,3 and d t 1 is a dummy variable in which d t 1  1 if ~t 1  0 and

d  0 if ~  0. Hereby, it is assumed that  follows a mean zero process with unit
t 1 t 1 t

variance that is independently and identically distributed (i.i.d.). The ARCH parameters are
 90

thus state dependent due to multiplication with the scaling factor g S t which is normalized to
unity for the low volatility regime.54

E. Results During the Peak of the Crisis

SWARCH Model based on Euro-Dollar Forex Swap

A regime switching model of the euro-U.S. Dollar Forex swap reveals that this variable
moves from a low to a medium volatility regime in the beginning of August 2007, before
entering the high volatility state right after the Lehman collapse in September 2008,
remaining there until the end of November 2008 (Figure 13). Many non-U.S. banks,
especially European ones, faced a shortage of U.S. dollar funding for their conduits and SIVs
beginning in the summer of 2007. As the interbank market for dollar funding became
squeezed due to counterparty and liquidity risks, these banks increasingly engaged in foreign
currency (FX) swap arrangements as well as in the cross-currency swap market (see Baba et
al, 2008). In particular, both the euro and sterling were used as the funding currencies for the
dollar FX swaps. The spillovers from the interbank market to the FX swap market led to a
situation whereby FX swap prices temporally deviated from their covered interest parity
condition. With the turbulence becoming more persistent, many non-U.S. financial
institutions also increasingly engaged in the longer-term cross-currency swap markets. This
episode especially highlighted the international interconnectedness of banks’ funding
requirements through FX swap markets.

54
In this paper, an ARCH specification is estimated, as the GARCH(p,q) is not nested within the SWARCH
framework, due to its implicit infinite lag representation.
 91

Figure 13. Euro-dollar Forex Swap

1.0 150
Probability of being in low-volatility state (left scale) Lehman's
failure
Probability of being in medium-volatility state (left
scale)
100
0.8 Beginning of subprime crisis
Probability of being in high-volatility state (left scale)

Euro-U.S. dollar forex swap (basis point change, right

scale) 50
0.6

0.4
-50

0.2
-100

0.0 -150
Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08
Sources: Bloomberg L.P.; and IMF staff estimates.

As shown in Figure 13, the move of the forex swap into the high volatility state on
September 15, 2008, coincides with the sharp increase in counterparty risk resulting from
Lehman’s failure and a sizeable dollar shortage that occurred with margins and haircuts
increasing on most dollar-denominated assets.

SWARCH Model based on the VIX

After the Lehman episode, the VIX increased to historical heights, and it is of interest to put
the S&P 500 stock market volatility during the current financial crisis into a historical
perspective. Figure 14 shows a daily SWARCH model for VIX from 1998 to the end of
2008. The model has the highest probability of being in the high volatility state during the
Russian Crisis and LTCM default in 1998, the period surrounding the WorldCom scandal
and Brazil’s election in 2002, as well as the beginning of the subprime crisis in the fall of
2007 and the period following the Lehman collapse. In particular, the model also enters the
high volatility state briefly at the time of the Shanghai stock market crash and the first abrupt
ABX (BBB) price decline of investment grade asset-backed subprime mortgages in late-
February 2007. During the Bear Stearns rescue, the VIX was more likely to be in the high
rather than medium volatility state. The Lehman failure then triggered a very fast movement
of the VIX into the high volatility regime where it remained until the end of the sample
period ending December 31, 2008. After the start of the subprime crisis, the VIX only
oscillated between the medium and high regime, in contrast to the predominantly low
volatility regime during 2003–2007.
 92

Figure 14. Markov-Switching ARCH Model of VIX

Absolute change in VIX (left scale) Probability of being in high-volatility state (right scale)
Lehman
25 1
Beginning of subprime crisis
Russian's default
and LTCM WorldCom and
20 Brazil's election
0.9
Turkey crisis Shanghai
crash 0.8
15

Bear 0.7
10 Stearns

0.6
5
0.5
0
0.4
-5 9-11
Dot-com
bubble burst 0.3

-10
0.2

-15 0.1

-20 0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

Sources: Bloomberg L.P.; and IMF staff estimates.

SWARCH Model based on TED Spreads

A similar SWARCH model is estimated for the three-month TED spread (the difference
between Libor and Treasuries). Figure 15 suggests that this indicator of short-term bank
credit risk moved decidedly into a high volatility regime in the beginning of August 2007 and
remained in it until the Bear Stearns’ rescue. The Lehman collapse again triggered a high
volatility regime. As in the VIX model, the SWARCH framework for the TED spread picks
up the Russia and LTCM default in 1998 as well as the September 11 shock. The findings
also imply a role for the sharp Shanghai stock market correction and the first round of ABX
(BBB) price declines in late February 2007, which could be seen as a potential warning
signal about the impending fragilities in the global financial system.

Overall, while the recent persistence of the high-volatility period for the TED spread is
unprecedented over the past decade, that for the VIX is not, suggesting a greater relative
stress in interbank markets during this crisis episode.
 93

Figure 15. Markov-Switching ARCH Model of TED Spread

Change in TED spread (basis points, left scale) Probability of being in high-volatility state (right scale)
100 1
[2]
[1] Beginning of
80 9-11 subprime 0.9
crisis
60 0.8

40 0.7

20 0.6

0 0.5

-20 0.4

-40 0.3

-60 0.2

-80 0.1

-100 0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008
Sources: Bloomberg L.P.; and IMF staff estimates.
Note: [1] = Russian default and LTCM crisis; [2] = Lehman Brothers failure.

F. Results After Massive Government Programs in 2009 to Address the Global Crisis

Since the peak of the global crisis following the bankruptcy of Lehman Brothers in
September 15, 2008, and particularly in the early part of 2009, a number of countries
introduced measures in support of their financial systems which ranged from implicit and
explicit guarantees to capital injections and outright nationalization of banks. In this regard,
Aït-Sahalia et al (2009) find that, for a number of advanced economies and using an event
study methodology, government interventions had a significant impact on the easing stress in
the interbanking market but that this effect was smaller the more prolonged the crisis
became.55 The examination of this most recent period may help policymakers decide whether
indeed market conditions have sufficiently stabilized so that they can start to exit their
massive support provided to the financial system. Casual observation would suggest that
global market conditions and economic activity have improved significantly in 2009. The
regime switching models examined still provide a mixed picture and signal moderate
pressures on certain market conditions through the summer of 2009.56 For example, while the
TED spread strongly signals a high probability of being in a low-volatility state, the VIX and
the forex swap market both signal a high probability of a medium-volatility regime.

55
Frank and Hesse (2009) also find that central banks’ action led to a reduction in Libor spreads in the first
phase of the global financial crisis, even though economic magnitudes were rather small.
56
The last observation in the estimation of the model is July 23, 2009.
 94

SWARCH Model based on Euro-Dollar Forex Swap

U.S. dollar funding pressures have declined dramatically since end-2008, largely as a result
of a number of facilities and swap arrangements between the Fed and several central banks.
However, while the probability of a low-volatility regime has increased to over 0.4 since
early-July 2009, the probability of a medium-volatility state (though declining) still remains
at around 0.6 (Figure 16).

g p Figure 16. Euro-Dollar Forex Swap

(Probability of being in low-, medium-, and high-volatility state)
Low Medium High
1.0
Beginning of
subprime crisis
Lehman's failure

0.8

0.6

0.4

0.2

0.0
Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09
Sources: Bloomberg L.P.; and IMF staff estimates.

SWARCH Model based on the VIX

As discussed, the most recent peak of a probability of being in a high-volatility state was
reached at end-September 2008 and lasted until mid-March 2009, at close to 1 (Figure 17a).
Thereafter, the probability of being in a high-volatility state declined rapidly to reach less
than 0.05 by end-April 2009. Since then, the probability of being in a high-volatility state has
remained at around the same low level.

Despite the sharp decline in the probability of being in a high-volatility regime, market
conditions (based on the VIX as a proxy for market uncertainty) have not yet returned to a
low-volatility state and instead have remained at a high probability (nearly 1) of a medium-
volatility state since early-May 2009 (Figure 17b). The probability of a low-volatility regime
is nearly zero, indicating that market conditions based on the VIX have not yet returned to a
tranquil period.
 95

Figure 17a. Markov-Switching ARCH Model of VIX

Figure 5a. Markov Switching ARCH Model of VIX
(Probability of being in low-, medium-, and high-volatility state) Beginning of
subprime crisis
Low Medium High
Lehman
1.0

0.9

0.8 Russian's
default WorldCom
0.7 and LTCM and
Brazil's
election
0.6
Bear
Stearns
0.5
9-11
0.4

0.3
Dot-com
0.2 bubble burst

0.1

0.0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Sources: Bloomberg L.P.; and IMF staff estimates.
Note: ARCH = autoregressive conditional heteroskedasticy; LTCM = Long-Term Capital Management; VIX = Chicago Board Options Exchange
volatility index.

Figure 17b. Markov-Switching ARCH Model of VIX

Figure 5b. Markov Switching ARCH Model of VIX
(Probability of being in low-, medium-, and high-volatility state)

Low Medium High

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0
Jan-09 Feb-09 Mar-09 Apr-09 May-09 Jun-09 Jul-09
Sources: Bloomberg L.P.; and IMF staff estimates.
Note: ARCH = autoregressive conditional heteroskedasticy; LTCM = Long-Term Capital Management; VIX = Chicago Board Options Exchange
volatility index.
 96

SWARCH Model based on TED Spreads

Similarly, after reaching a peak late in 2008, the probability of being in a high-volatility state
based on TED spreads has declined dramatically since early-April 2009 where it has
remained since then (Figure 18a).

However, in contrast with the VIX, the probability of being in medium-volatility regime is
quite low (less than 0.1) while the probability of being in a low-volatility state has been
higher than 0.9 since late May 2009, where it has remained since then (Figure 18b).57

Thus, while the VIX still signals moderate uncertainty in financial markets, the TED spread
suggests that pressures in interbank markets have subsided dramatically since the spring of
2009.

g Figure 18a. Markov-Switching ARCH Model of TED Spread

g p
(Probability of being in low-, medium-, and high-volatility state)
Low Medium High
9-11 Lehman
1.0

0.9 Beginning
of
subprime
0.8 crisis

0.7
Russian's
default
0.6 and
LTCM
0.5 crisis

0.4

0.3

0.2

0.1

0.0
1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
Sources: Bloomberg L.P.; and IMF staff estimates.
Note: ARCH = autogressive conditional heteroskedasticity; LTCM = Long-Term Capital Management; TED = the spread between the three-month
LIBOR and treasury bill rates.

57
The temporary spike in the probability of a high-volatility state for the TED spread observed in late-March
2009 was likely related to the announcement that the Federal Reserve would start a program of directly buying
U.S. Treasury bonds in the market. Yields on U.S. Treasury bonds dropped significantly following the
announcement, but began to creep up in the days that followed.
 97

Figure 18b. Markov-Switching ARCH Model of TED Spread

Figure 6b. Markov Switching ARCH Model of TED Spread
(Probability of being in low-, medium-, and high-volatility state)

Low Medium High

1.0

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0.0
Jan-09 Feb-09 Mar-09 Apr-09 May-09 Jun-09 Jul-09
Sources: Bloomberg L.P.; and IMF staff estimates.
Note: ARCH = autogressive conditional heteroskedasticity; LTCM = Long-Term Capital Management; TED = the spread between the three-month
LIBOR and treasury bill rates.

G. Conclusion

To summarize, this paper presents a Markov-regime switching technique to examine when

key global market conditions variables such as the VIX, forex swap or the TED spread
moved into a high volatility regime. The findings support the view that the Lehman failure
was a key watershed event in the crisis, but periods of a high volatility state were also present
before Lehman’s failure. In particular, based on the VIX SWARCH model, these earlier
episodes of distress include the Shanghai stock exchange crash and the ABX (BBB) price
decline in February 2007, the beginning of the subprime crisis in August 2007, and the Bear
Stearns rescue in March 2008. The results suggest that the bankruptcy of Lehman Brothers
aggravated what it appeared to be already a crisis characterized by persistent (albeit at times
noisy) signs of a high volatility state. High volatility states can be viewed as a potential
manifestation of systemic risk.

In the aftermath of the Lehman collapse in the fall of 2008, which corresponded to the peak
of the global crisis, a number of countries introduced massive government interventions in
support of their financial systems. The examination of this most recent period is of interest
because global market conditions and economic activity appear to have improved since mid-
2009, potentially signaling that exit strategies can begin to be implemented. However, the
regime switching models examined still signal moderate pressures on certain market
conditions as of end-July 2009. For example, while the TED spread strongly signals a high
probability of being in a low-volatility state, the VIX and the forex swap market both signal a
high probability of a medium-volatility regime.
 98

Overall, the results show that the global market indicators examined here sometimes do not
stay in the high-volatility state for long, with some exceptions such as the TED spread or the
VIX. This suggests they should be used in combination with other tools to help policymakers
detect systemic crises and when those are receding.

The approach presented in this paper can be a helpful tool for policymakers to evaluate when
market conditions are such that even a relatively small shock can lead to systemic events, or
when financial institutions and markets can become distressed as a result of unstable market
conditions, or when conditions have improved sufficiently to begin to withdraw government
support to financial institutions and markets that had been previously embroiled in a systemic
crisis.

A caveat of the methodology used in this paper is that it has been applied univariately to
global market conditions, and future research should attempt to adopt multivariate SWARCH
models that can combine various factors in a coherent and forward-looking manner. In
addition, the states are determined by the existing dataset so once the high volatility periods
exit the dataset, a high volatility state would signify less actual volatility compared to the
crisis period.
 99

References

Aït-Sahalia, Y., J. Andritzky, A. Jobst, S. Nowak, and N. Tamirisa, 2009, “How to Stop a
Herd of Running Bears? Market Response to Policy Initiatives during the Global
Financial Crisis,” Working Paper 09/2004 (International Monetary Fund,
Washington, D.C.).

Baba, Naohiko, Frank Packer, and Teppei Nagano, 2008, “The spillover of money market
turbulence to FX swap and cross-currency swap markets,” BIS Quarterly Review,
pp.73–86.

Brunnermeier, Markus, and Lasse H. Pedersen, 2009, “Market Liquidity and Funding
Liquidity,” Review of Financial Studies 22(6), pp. 2201–2238.

Demirguç-Kunt, A., and Enrica Detragiache, 1998, “The Determinants of Banking Crises in
Developing and Developed Countries,” IMF Staff Papers 45, No. 1, pp. 81–109.

Diamond, Douglas W., and Philip H. Dybvig, 1983, “Bank Runs, Deposit Insurance, and
Liquidity, “Journal of Political Economy,” 91(3), pp. 401-419.

Dornbush, R., Y.C. Park, and S. Claessens, 2000, “Contagion: Understanding How It
Spreads,” The World Bank Research Observer, 15(2), pp. 177–97.

Dungey, Mardi, Renee Fry, Brenda González-Hermosillo, and Vance Martin, 2005,
“Empirical Modeling of Contagion: A Review of Methodologies,” Quantitative
Finance, 5, pp. 9–24.

———, 2006, “Contagion in International Bond Markets during the Russian and LTCM
Crises,” Journal of Financial Stability, 2, pp. 1–27.

———, 2007, “Unanticipated Shocks and Systemic Influences: The Impact of Contagion in
Global Equity Markets in 1998,” North American Journal of Finance and Economics,
18, pp. 155–174.

———, forthcoming, “Are Financial Crises Alike?—From the 1998 Russian/LTCM Crisis to
the 2007 Subprime Debacle and Liquidity Crisis,” IMF Working Paper (Washington:
International Monetary Fund).

Forbes, K., and Rigobon, R., 2002, “No Contagion, Only Interdependence: Measuring Stock
Market Co-movements,” Journal of Finance, 57(5), pp. 2223–61.

Frank, Nathaniel, Brenda Gonzalez-Hermosillo, and Heiko Hesse, 2008, “Transmission of

Liquidity Shocks: Evidence from the 2007 Subprime Crisis,” IMF Working Paper
08/200 (Washington: International Monetary Fund).
 100

Frank, Nathaniel, and Heiko Hesse, 2009, “Financial Spillovers to Emerging Markets during
the Global Financial Crisis,” IMF Working Paper 09/104 (Washington: International
Monetary Fund).

Frank, Nathaniel, and Heiko Hesse, 2009, “The Effectiveness of Central Bank Interventions
during the First Phase of the Subprime Crisis,” IMF Working Paper 09/206
(Washington: International Monetary Fund).

González-Hermosillo, Brenda, 2008, “The Role of International Investors’ Risk Appetite in

Global Financial Crises: 1998–2007,” IMF Working Paper 08/85 (Washington:
International Monetary Fund).

González-Hermosillo, Brenda, and Heiko Hesse, 2009, “Global Market Conditions and
Systemic Risk,” IMF Working Paper 2009/230 (Washington: International Monetary
Fund). Also published in the Journal of Emerging Market Finance, 2011, 10 (2), 227-
252 and in VOX.

Gorton, Gary, 2008, “The Panic of 2007,” Prepared for the Federal Reserve Bank of Kansas
City, Jackson Hole Conference, August.

Group of Ten, 2001, Consolidation in the Financial Sector (Basel: Bank for International
Settlements).

Hamilton, James D., and Raul Susmel, 1994, “Autoregressive Conditional Heteroskedasticity
and Changes in Regime,” Journal of Econometrics, Vol. 64 (September-October),
pp. 307–33.

Hardy, D. C., and Ceyla Pazarbasiouglu, 1999, “Determinants and Leading Indicators of
Banking Crises: Further Evidence,” IMF Staff Papers 46 (3), pp. 247–258.

Illing, Mark, and Ying Liu, 2006, “Measuring Financial Stress in a Developed Country: An
Application to Canada,” Journal of Financial Stability, pp. 243–65.

International Monetary Fund, 2008, Global Financial Stability Report, “Market and Funding
Illiquidity: When Private Risk Becomes Public,” Chapter 3. World Economic and
Financial Surveys (Washington, April).

International Monetary Fund, 2009, Global Financial Stability Report, “Detecting Systemic
Risk,” Chapter 3. World Economic and Financial Surveys (Washington, April).

Kaminsky, G.L., and Reinhart, C.M., 1999, “The Twin Crises: The Causes of Banking and
Balance of Payments Problems,” American Economic Review 89 (3), pp. 473–500.
 101

Laeven, L. and F. Valenci, 2008, “The Use of Blanket Guarantees in Banking Crises,” IMF
Working Paper 08/250 (Washington: International Monetary Fund).

Lo, Andrew, 2008, “Hedge Funds, Systemic Risk, and the Financial Crisis of 2007–2008,”
Written testimony to the U.S. House of Representatives Committee on Oversight and
Government Reform, November 13, 2008.

Masson, Paul, 1999, “Contagion: Monsoonal Effects, Spillovers, and Jumps Between
Multiple Equilibria” in Agenor, P.R., Miller, M. Vines, D. and Weber, A. (eds.), The
Asian Financial Crisis: Causes, Contagion and Consequences. Cambridge University
Press, Cambridge.
V. WHAT DO SOVEREIGN WEALTH FUNDS IMPLY FOR FINANCIAL STABILITY?, WITH TAO
SUN, 200958

This paper examines financial stability issues that arise from the increased presence of
sovereign wealth funds (SWFs) in global financial markets by assessing whether and how
stock markets react to the announcements of investments and divestments to firms by SWFs
using an event study approach. Based on 166 publicly traceable events collected on
investments and divestments by major SWFs during the period from 1990 to 2009, the paper
evaluates the short-term financial impact of SWFs on selected public equity markets in which
they invest. The impact is analyzed on different sectors (financial and nonfinancial), actions
(buy and sell), market types (developed and emerging markets), and level of corporate
governance (high and low score). Results, based on these 166 events, show that there was no
significant destabilizing effect of SWFs on equity markets, which is consistent with anecdotal
evidence.

A. Introduction

Since the beginning of the financial crisis in the summer of 2007, financial stability has been
at the forefront of policy discussions. At the same time, sovereign wealth funds (SWFs) have
become dominant players, as they have injected significant capital in major financial
institutions. Recently in some countries, SWFs were instructed by their governments to
invest into domestic financial institutions and the overall stock market in order to support
battered stock prices. Research on the financial stability implications of these funds has been
slowly emerging, hampered by lack of data on their asset allocations.

There have been many arguments put forth regarding the potential positive and negative
effects of SWFs on global financial markets. For example, some argue that SWFs can play a
stabilizing role in global financial markets. First, many commentators point out that as long-
term investors with no imminent call on their assets, and with mainly unleveraged positions,
SWFs are able to sit out longer during market downturns or even trade against market trends.
In addition, SWFs in some countries, particularly in the Middle East, have recently supported
domestic equity markets and financial institutions. Second, large SWFs may have an interest
in pursuing portfolio reallocations gradually so as to limit adverse price effects of their
transactions. Third, SWFs could, as long-term investors and by adding diversity to the global
investor base, contribute to greater market efficiency, lower volatility, and increased depth of
markets. Fourth, SWF investments may enhance the depth and breadth of markets.

Although SWFs appear to have been a stabilizing force thus far, given their size, there are
circumstances in which they could cause volatility in markets. Having large and often

58
This chapter is based on Sun and Hesse (2009).
 103

intransparent positions in financial markets, SWFs—like other large institutional investors—

have the potential to cause a market disturbance. For instance, actual or rumored transactions
may affect relative valuations in particular sectors and result in herding behavior, adding to
volatility. Deeper markets, such as currency markets, can also be affected, at least
temporarily, by rumors or announcements about changes in currency allocations by central
banks or SWFs. To the extent that SWFs invest through hedge funds that rely on leverage or
are subject to margin requirements, such investments may inadvertently magnify market
changes. For markets to absorb flows from any major investor class without large price
fluctuations, it helps if they can anticipate the broad allocation and risk-preference trends of
such investor classes. Opacity about such trends can lead to inaccurate pricing and volatility.
As regards these financial stability implications of SWFs, both theoretical and empirical
research has begun to be implemented.

Recent capital injections by SWFs in financial institutions have intensified the debate on the
impact on financial stability. SWFs from East Asia and the Middle East were frequently in
the news, as major mature market financial institutions required additional capital. In total,
SWFs have reportedly contributed more than $50 billion of such capital since
November 2007. The capital injections by SWFs have augmented the recipient financial
institutions’ capital buffers and have been helpful in reducing various firm specific risk
premia, at least in the short term, as the injection curtailed the need to reduce bank assets to
preserve capital. The announcements of capital injections from SWFs have assisted in
stabilizing share prices and the elevated CDS spreads, at least over the short run (Global
Financial Stability Report, April 2008). In most cases, after the announcement of new capital
injections, the initial share price reaction to the SWF investments was positive, with
announcements of asset write-downs offset by hand-in-hand capital injections from investor
groups in which the SWF had a significant role. Although other factors are not taken into
account, this initial evidence supports the view that SWFs could have a volatility-reducing
impact on markets.59

This paper, using an event study approach and based on a hand collected database, endeavors
to deepen the analysis of SWFs’ impact on financial stability by differentiating scenarios,
including investments and divestments in advanced and emerging economies, financial and
nonfinancial sectors, higher and lower level of corporate governance. The overall findings
suggest that there is no significant destabilizing effect of SWFs on equity markets. This
empirical study contributes to the emerging academic literature that seeks to analyze the
behavior of SWFs in financial markets.

59
With the continuing increase in banks’ losses and writedowns during the subprime crisis, the rescue of Bear
Stearns, collapse of Lehman Brothers and U.S. government intervention into major financial institutions, the
longer term share price development of banks that obtained initial capital injections from various SWFs, has
been obviously very negative. But the short-term reaction of SWFs financial support has been perceived as very
supportive by the financial market in most cases.
 104

The paper proceeds as follows: Section B briefly reviews the literature and some conceptual
issues. Section C outlines an event study approach and describes data. Section D presents
empirical results. Section E concludes.
B. Literature Review

SWFs are defined as special-purpose investment funds or arrangements owned by the general
government. They are often established out of balance of payments surpluses, official foreign
currency operations, proceeds of privatizations, fiscal surpluses, or receipts resulting from
commodity exports. Their total size has been estimated at $2 trillion to $3 trillion, but many
of them have probably seen large unrealized losses from the ongoing financial crisis
combined with a sharp reduction in oil prices60. These unrealized losses have been higher for
SWFs that have a higher share of equities in their investment portfolio or large illiquid
positions in private equity or hedge funds. Given that SWFs typically have a fairly long
investment horizon, they are likely to sit out these unrealized losses.

Given the lack of publicly available data on SWF asset allocations, a strand of research has
been on the theory side. Lam and Rossi (forthcoming) develop a theoretical model that aims
to examine the impact of SWFs on global financial stability during periods of stress. Their
findings indicate that SWFs have a risk-sharing role in financial markets. As part of the IMF-
coordinated process of the Santiago Principles that provide generally accepted principles and
practices for SWFs, Hammer, Kunzel, and Petrova (2008) examine the asset allocation and
risk management frameworks of SWFs based on a detailed survey. The results show that
SWFs have specific investment objectives in place, adopt an asset approach (mean-variance
style) in determining their asset allocation strategy, utilize common risk measures (e.g., credit
ratings, value-at-risk models, tracking errors, duration, and currency weights) for their risk
management, and have explicit limits in their investment classes and instruments.

Simulations of SWFs’ asset allocations have been undertaken by Kozack, Laxton, and
Srinivasan (forthcoming). Specifically, they create two stylized diversified portfolios, one
mimicking Norway’s SWF and the other representing some well-established SWFs, and they
conduct a scenario analysis of the impact from a further diversification of sovereign assets.
While the calibrations are highly sensitive to the underlying model assumptions, the findings
indicate that advanced economies will see lower capital inflows, while emerging market
countries will be the primary beneficiaries. Their quantitative results are consistent with the

60
A new report by International Financial Services London has revealed that sovereign wealth funds total assets
increased 18 percent to $3.9 trillion in 2008 from $3.3 trillion in 2007. Total assets are now contracted to reach
$8 trillion by 2015, down from their $10 trillion estimated in 2008.
 105

back-of-the envelope calculations of Beck and Fidora (2008), which imply a net capital
outflow from the United States and the euro area and net inflows to emerging market
countries over the medium-term. In the same vein, Jen and Miles (2007) and Hoguet (2008)
points out that there is scope for the global equity risk premium to fall and for real bond
yields to rise if SWFs allocate their assets to equities. In addition, as SWFs increasingly
diversify into global portfolios, their activities may place some downward pressure on the
dollar as they exit dollar-denominated assets.

There has been some empirical research, using equity market indicators and an event study
approach to examine the role of SWFs as major institutional investors. For instance, in an
event study, Chhaochharia and Laeven (2008) find that the announcement effect of SWF
investments is positive. They report that share prices of firms respond favorably when SWFs
announce investments, in part because these investments happen when their targets are in
financial distress. But the long-run performance of equity investments by SWFs tends to be
poor (see Fotak, Bortolotti, and Megginson, 2008, for similar results). Another event study
analysis by Bortolotti, Fotak, Megginson, and Miracky (2009) based on the Monitor Group
database of SWF transactions also finds a positive short-run announcement effect of SWF
investments and negative long run abnormal returns. Dewenter, Han, and Delesta (2009) and
Knill, Lee and Mauck (2009) obtain similar results. Kotter and Lel (2008) show that the
cumulative abnormal return of SWF investments has an announcement effect similar to that
of investments by hedge funds and institutional investors such as CalPERS on stock returns.
In addition, investments by more transparent SWFs have a larger cumulative abnormal return
by an order of 3.5, suggesting that voluntary SWF disclosure might serve as a signaling
device to investors. In addition, Kotter and Lel (2008) also obtain a significant negative but
small announcement impact from SWFs’ divestures. Beck and Fidora (2008) conduct a
country case study of Norway’s SWF and ask whether its exclusion of companies that violate
the ethical guidelines of the ministry of finance exhibit price pressures on those companies.
Their findings suggest no significant negative abnormal returns following the divesture of
these companies.

To summarize, existing research on SWFs suggests that they can be a stabilizing force in
global financial markets. Event studies do not find a destabilizing impact from SWF
investments and divestments in equity markets, while simulations of SWF asset allocations
only imply a gradual shift with modest economic effects. With SWFs improving their
transparency and disclosure over time, the availability of historical SWF transactions would
provide researchers with the necessary data to further examine their implications for financial
stability.

C. Data and Methodology

This empirical research assesses whether stock markets react to the announcements of
investments and divestments to firms by SWFs using an event study approach. The objective
 106

is to investigate the information content of these announcements. Based on 166 publicly

traceable hand collected events of investments and divestments by major SWFs during
1990-2009, this section evaluates the short-term financial impact of SWFs on selected public
equity markets in which they invest. Moreover, the impact will be further analyzed on
different sectors (financial and nonfinancial), actions (buy and sell), market types (developed
and emerging markets), as well as level of corporate governance (higher and lower level).
The results are expected to give some hints on how stock markets react to the capital
investments and divestments by SWFs, and present some implications on SWFs’ stabilizing
role in global financial market. Investigating divestments is of particular interest since if
stock price reactions are abnormally high (relative to the market) there may be destabilizing
effects to the degree that others “front run,” “herd” or otherwise mimic SWFs’ investment
behavior. This might be particularly problematic if prices slip below pre-defined target levels
of other investors, and thus prompting their forced sales.

D. Data

Several SWFs that have bought or sold shares of firms in the advanced and emerging stock
markets are included in the study. Among them are Abu Dhabi Investment Authority, China
Investment Corporation, Government of Singapore Investment Corporation, Kuwait
Investment Authority, Korea Investment Corporation, Libyan Investment Authority,
Mubadala, Qatar Investment Authority, and Temasek. The source of information on the
events is SWFs’ websites and various financial news and reports such as Factiva. Target firm
actual total returns (and price indices) and country stock market returns (and price indices)
are obtained from Datastream International database.61 This search results in a total of
166 investment/ divestment events in 115 unique firms, with some firms receiving multiple
SWF investments through time between 1990 and 2009. This sample is then combined with
firm-level and country level data collected from Bloomberg, and SWF-specific data from
various sources including Truman (2008)62.

Table 4 describes the number of SWF investments and divestments across the country of
target firms, while Table 5 displays the distribution of the sample by the identity of the
acquiring SWF. Given public availability of individual buy and sell transactions, observation
numbers by the two Singapore SWFs GIC and Temasek are dominating the sample. Figure
19 shows the ratios on SWFs’ investments/ divestments in full sample as well as in sub

61
Datastream is the only data vendor that provides total return stock market indices for all the relevant
countries, correcting index returns for the implications of dividend payments, stock splits, and other such
changes.
62
The score of each SWF is from the “total” score of Truman (2008a; 2008b). We take those higher than 40 as
“high”, while those lower than 40 as “low” in the econometric analysis.
 107

samples—in financial and nonfinancial sectors—in developed and emerging markets and by
SWFs with different levels of corporate governance.

Table 4. Country of Target Firms

Country Number
Australia 6
Austria 1
China 17
Egypt 2
France 8
Germany 7
Iceland 1
India 13
Indonesia 5
Italy 6
Japan 2
Malaysia 7
Pakistan 4
Philippines 1
Portugal 2
Singapore 22
South Korea 3
Spain 3
Sweden 2
Switzerland 2
Taiwan Province of China 1
Thailand 2
United Kingdom 31
United States 17
Vietnam 1
Total 166
 108

Table 5. Acquiring SWFs

Number of
SWF Observations Country
Abu Dhabi Investment Authority (ADIA) 26 United Arab Emirates
China Investment Corporation (CIC) 11 China
Government of Singapore Investment Corporation (GIC) 38 Singapore
Kuwait Investment Authority (KIA) 14 Kuwait
Korea Investment Corporation (KIC) 1 Korea
Libyan Investment Authority (LIA) 2 Libya
Mubadala 2 United Arab Emirates
Qatar Investment Authority (QIA) 23 Qatar
Temasek 49 Singapore
Total 166

Figure 19. Ratios of SWF Investments and Divestments

90

80

70

60

50

40

30

20

10

0
Buy in non-financial

Buy by high level

Buy in emerging

Sell in non-financial

Buy by low level

Sell by high level
Sell in emerging

Buy in financial sector

Buy and Sell in Non-

Sell by low level

Sell in financial sector
Buy in developed
Buy and Sell in

Buy and Sell in

Buy and Sell in

Buy

Sell in developed
Sell

Developed

Emerging

economies
economies
Financial

economies
economies
financial

sector
sector

Source: IMF staff estimates.

Note: 1)The SWFs with high level corporate governance refer to those whose total score is higher than 40,
while low level refer to lower than 40 (Truman, 2008a;2008b); 2) the ratios are calculated separately on the
following six sub-groups: i) buy and sell; ii) buy and sell in financial sector, buy and sell in nonfinancial sector;
iii) buy and sell in developed economies, buy and sell in emerging economies, iv) buy in developed economies,
sell in developed economies, buy in emerging economies, sell in emerging economies; v) buy in financial
sector, sell in financial sector, buy in nonfinancial sector, sell in nonfinancial sector; and vi) buy by high level
governance, sell by high level governance, buy by low level governance, sell by low level governance.
 109

E. Methodology

If markets are rational, the effects of an event should be reflected immediately in stock
returns and prices. Thus, a measure of the event’s impact can be constructed using stock
prices and returns observed over a relatively short time period. To benchmark the returns of
the stock relative to the event, the overall stock market returns, in percentage changes, for the
corresponding country are used.

Specifically, the following steps are taken for implementing the event study:

 Determination of the selection criteria for the inclusion of given SWFs. The
sample contains several SWFs, which have bought or sold stakes in financial firms
and nonfinancial firms.

 Collection of a number of such events and compilation of a list of firms and dates
by searching publicly-available databases to find news announcements on SWFs’
actions.

 Identification of the events of SWFs’ investments/divestments. Since the event

date can be determined with precision, as regards to the short-term analysis, we
employ a five-day (seven-day) event window, comprised of two (three) pre-event
days, the event day, and two (three) post-event days. In this way, rumors that precede
the formal announcement can enter the assessment. And as well, in illiquid markets,
prices may take a couple of days to adjust to new information. As robustness tests, we
vary the event window to four pre-event days, the event day, and four post-event
days.

 Definition of the “estimation window.” Following Peterson’s framework (1989), we

will estimate the market model on the 200 trading days ending 30 days prior to the
announcement of the investments/divestments. Ending the sample prior to the event
assures that the “normal” behavior of returns is not contaminated by the event itself.
For robustness tests, we vary the estimation periods (100 days and 300 days) and
using price indices instead of total returns of each firms and economy.

 Prediction of a “normal” return during the event window in the absence of the
event, using a one factor OLS regression equation:63

rit=αi+βirmt+еit,

63
Since the “market model” is most commonly used to generate expected returns and no better alternative has
yet been found despite the weak relationship between beta and actual returns (Armitage, 1995), we use the
market model to predict “normal” return. To test for robustness, a three-factor model could also be employed.
 110

Where rit is the percentage change of returns of the stock relative to the event, rmt is
the percentage change of overall stock market returns, αi and βi are regression
coefficients, and the еit is an error term.

 Calculation of the abnormal return within the event window. Having calculated
estimates of αi and βi with the data from the estimation period, we calculate the
abnormal returns by differencing the actual and estimated returns,

ARit=Rit - Rit* = Rit - (α*+βi*Rmt), where Rit* is the estimated return.

Specifically, the abnormal return observations must be aggregated in order to draw

overall inferences for the event of interest. The aggregation can be along two
dimensions—through time and across securities.

The individual securities’ abnormal returns, in the case of five days, can be
aggregated for each event day, t = t-2, t-1, t, t+1, t+2 during the event window. Given
N events (a total of 166 in the entire sample), the sample average aggregated
abnormal returns (AAR) for period t is
1 N
AARt=  ARit .
N i 1

The average abnormal returns can then be aggregated over the event window to
calculate the cumulative average abnormal return (CAAR) for each firm i.
2
CAARt =  AARt
t  2

 Testing whether the abnormal return is statistically different from zero. Since
the numbers of observation in the event window are limited (five or seven days), we
use t-tests rather than the Z score, the latter usually requiring at least 50 observations
to get a statistically robust results.64

F. Empirical Results

Table 6 presents the AAR and CAAR for the (-2, +2), and (-3, +3) windows. In general, the
AAR is positively associated with SWFs’ buy actions and not significantly negatively with
SWFs’ sell actions in the full sample. Moreover, overall, the results suggest that the share
price’s combined responses to SWFs’ investments and divestments in developed economies

64
The t test is of interest because it can accommodate the differences of the abnormal returns over time and
especially across types of markets. The event study approach shows the explicit impact of SWF actions, since
the methodology is based on individual purchases and sales of publicly available equities.
 111

are significant (Panel C), while those in emerging economies are not (Panel D). In addition,
SWF investments in the financial sector have a larger impact on share prices than in the
nonfinancial sector. These differences in responses may be due to the relatively more
transparent equity markets in developed economies as well as in the financial sector with
potentially higher signaling and information flow.
 112

Table 6. Stock Market Reactions to Announcements of SWF Investments and

Divestments (Total Returns)
Test Statistic of
Event Window Test Statistic of AAR Mean of AAR CAAR Mean of CAAR

Panel A: Buy only, 134 events from 101 firms

(-3,+3) 3.75** 0.22 4.45*** 0.96
(-2,+2) 4.31** 0.27 3.33** 0.77

Panel B: Sell only, 32 events from 23 firms

(-3,+3) -0.08 -0.02 -1.21 -0.19
(-2,+2) 0.00 0.00 -0.31 -0.07

Panel C Buy and Sell in developed economies only, 87 events from 55 firms
(-3,+3) 2.88** 0.18 6.17*** 0.94
(-2,+2) 4.29** 0.21 4.95** 0.72

Panel D: Buy and Sell in emerging economies only, 79 events from 60 firms
(-3,+3) 0.98 0.11 1.14 0.20
(-2,+2) 1.20 0.17 1.67 0.34

Panel E: Buy in developed economies only, 72 events from 51 firms

(-3,+3) 3.13** 0.24 5.82*** 1.21
(-2,+2) 5.47** 0.30 4.44** 0.91

Panel F: Sell in developed economies only, 15 events from 9 firms

(-3,+3) -0.56 -0.16 -2.99** -0.77
(-2,+2) -0.49 -0.19 -1.24 -0.44

Panel G: Buy in emerging economies only, 62 events from 50 firms

(-3,+3) 2.37* 0.20 2.94** 0.69
(-2,+2) 2.07 0.24 2.31* 0.60

Panel H: Sell in emerging economies only, 17 events from 14 firms

(-3,+3) 0.37 0.09 1.62 0.29
(-2,+2) 0.44 0.16 0.99 0.23

Panel I: Buy in financial sector only, 41 events from 24 firms

(-3,+3) 3.09** 0.66 6.38*** 3.40
(-2,+2) 2.72* 0.70 5.13** 2.53
Panel J: Sell in financial sector only, 5 events from 3 firms
(-3,+3) - - - -
(-2,+2) - - - -

Panel K: Buy in non-financial sector only, 93 events from 77 firms

(-3,+3) -0.15 -0.01 -4.06** -0.35
(-2,+2) 0.31 0.04 -1.78 -0.18
Panel L: Sell in non-financial sector only, 27 events from 20 firms
(-3,+3) -0.08 -0.02 -1.21 -0.19
(-2,+2) 0.00 0.00 -0.31 -0.07
Panel M: Buy by high level in governance only, 76 events from 59 firms
(-3,+3) 0.18 0.02 0.84 0.07
(-2,+2) -0.11 -0.02 -0.20 -0.02

Panel N: Sell by high level in governance only, 26 events from 19 firms

(-3,+3) 0.27 0.06 1.04 0.15
(-2,+2) 0.36 0.12 0.85 0.17
Panel O: Buy by low level in governance only, 58 events from 45 firms
(-3,+3) 2.21* 0.50 4.05** 2.22
(-2,+2) 2.68* 0.68 3.03** 1.89
Panel P: Sell by low level in governance only, 6 events from 4 firms
(-3,+3) -1.15 -0.53 -3.67** -2.39
(-2,+2) -1.23 -0.73 -1.96 -1.61

Source: IMF staff estimates.

Note: Since there are no qualified observations before/after the corresponding event dates,
there are no results for the group of "sell in financial sector only (Panel J)".
 113

Different scenarios are tested using these events. Specifically, Panel A of Table 6 reports the
AAR and CAAR around the announcements of SWF investments for the entire sample of
134 observations during the period between 1990 and 2009. The AAR is 0.27 percent and
0.22 percent for windows of (-2, +2), and (-3,+3) around the announcement date, and the
CAAR is 0.77 percent and 0.96 percent, respectively. The sign test statistics for the AAR are
also highly significant for the two windows. Panel B reports the AAR and CAAR around the
announcements of SWF divestments for the entire sample of 32 observations during the
period between 1990 and 2009. The AAR is 0 percent and -0.02 percent for the windows
(-2,+2), and (-3, +3) around the announcement date, and the CAAR is -0.07 percent and
-0.19 percent, respectively. The sign test statistics for the AAR and the CAAR are
insignificant for the two windows.

Panel C reports the AAR and CAAR around the announcements of SWF investments and
divestments for the developed economy sample of 87 observations during the period between
1990 and 2009. The AAR is 0.21 percent and 0.18 percent for the windows (-2, +2), and
(-3,+3) around the announcement date, and the CAAR is 0.72 percent and 0.94 percent,
respectively. The sign test statistics for the AAR and the CAAR are highly significant for the
two windows. Panel D of Table 6 reports the AAR and CAAR around the announcements of
SWF investments and divestments for the emerging economy sample of 79 observations
during the period between 1990 and 2009. The AAR is 0.16 percent and 0.11 percent for the
windows (-2,+2), and (-3, +3) around the announcement date, and the CAAR is 0.34 percent
and 0.20 percent, respectively. The sign test statistics for the AAR and CAAR are
insignificant for the two windows.

The impact is further analyzed on the investments/divestments separately in different market

types (developed and emerging markets), different sectors (financial and nonfinancial), and
level of corporate governance (high and low). In general, according to the AAR, investments
in developed economies (Panel E) and emerging economies (Panel G) are statistically
significant, while divestments in developed economies (Panel F) and emerging economies
(Panel H) are generally statistically insignificant. These demonstrate that SWF investments
produce positive impact in both developed and emerging economies while their divestments
led to little negative impact.65In addition, the positive impact of ARR and CAAR for the
investments by low level governance SWFs are significantly larger than those by high level
governance SWFs because the investment/divestment behaviors of low level governance
SWFs may be more speculative and unexpected, thus triggering larger market impact upon
the announcement of their actions. This is in line with the idea that transparency matters.

65
While the combined impact of investments and divestments in emerging economies (Panel D) is insignificant,
the impact of investment in emerging economies is significant (Panel G). The reason could be the individual
impact of investments was offset by the divestments when both actions are jointly tested.
 114

This could also indicate that the improvement of corporate governance in SWFs would be
helpful in reducing the impact on market volatility66.

As a robustness check, we use the event window of (-4,+4) to test the impact of SWFs’
actions. In addition, we vary the estimation periods (100 and 300 days). Finally, we use price
indices of each firms and economy instead of total return. The results are robust to different
event windows and the estimation periods, and the use of price indices (Table 7).

G. Conclusion

This paper assesses whether and how stock markets react to the announcements of
investments and divestments to firms by SWFs using an event study approach. Based on
166 publicly traceable events collected on investments and divestments by major SWFs
during the period of 1990–2009, we evaluate the short-term financial impact of SWFs on
selected public equity markets in which they invest. The impact is further analyzed on
different sectors (financial and nonfinancial), actions (buy and sell), market types (developed
and emerging markets), countries, and level of corporate governance (high and low). Overall,
this event study does not find any significant destabilizing effect of SWFs on equity markets
as measured by abnormal return behavior, which is consistent with the emerging academic
literature that uses the event study methodology. This study contributes to the slowly
emerging field of empirical studies of SWF behavior in financial markets.

However, it should be noted that the longer-term impact and the potentially stabilizing role of
SWFs as major institutional investors will require a broader set of data and a more rigorous
empirical assessment. The long-run impact of SWF investments could be subject to the
macroeconomic and financial conditions. In the case of recent investments in some U.S. and
European financial institutions under conditions of distress, SWFs’ action could not buffer
those institutions from further large losses. Therefore, it will be hard to draw conclusions for
overall global and regional financial stability only from these 166 events. Other methods to
examine the empirical impact of SWFs would require more detailed knowledge of SWF
investments and their timing and amount—data that are presently not available. Some
progress may be possible with hypothetical scenarios, but hypothetical market responses to
SWF investments require a thorough understanding of how asset allocations are constructed
and the size, depth, and breadth of the corresponding markets.

66
This is in line with the positive market responses to the investments in the entire sample. The reason is that
SWFs with low level of corporate governance accounts for the majority sample of SWF investments.
 115

Table 7. Stock Market Reactions to Announcements of SWF Investments and

Divestments (Price Indices)

Event Window Test Statistic of AAR Mean of AAR Test Statistic of CAAR Mean of CAAR

Panel A: Buy only, 134 events from 101 firms

(-3,+3) 3.84** 0.22 4.71*** 0.98
(-2,+2) 4.09** 0.26 3.46** 0.75

Panel B: Sell only, 32 events from 23 firms

(-3,+3) 0.16 0.03 1.67 0.24
(-2,+2) 0.24 0.07 1.32 0.26

Panel C Buy and Sell in developed economies only, 87 events from 55 firms
(-3,+3) 2.83** 0.20 6.37*** 1.10
(-2,+2) 5.05** 0.25 5.45** 0.87

Panel D: Buy and Sell in emerging economies only, 79 events from 60 firms
(-3,+3) 0.94 0.10 1.33 0.20
(-2,+2) 0.98 0.14 1.58 0.28

Panel E: Buy in developed economies only, 72 events from 51 firms

(-3,+3) 3.22** 0.25 5.66*** 1.23
(-2,+2) 5.5** 0.31 4.44** 0.95

Panel F: Sell in developed economies only, 15 events from 9 firms

(-3,+3) -0.20 -0.05 0.67 0.14
(-2,+2) -0.12 -0.04 1.02 0.28

Panel G: Buy in emerging economies only, 62 events from 50 firms

(-3,+3) 2.58** 0.19 3.41** 0.69
(-2,+2) 1.94 0.21 2.37* 0.53

Panel H: Sell in emerging economies only, 17 events from 14 firms

(-3,+3) 0.40 0.10 1.80 0.32
(-2,+2) 0.47 0.17 1.05 0.24

Panel I: Buy in financial sector only, 41 events from 24 firms

(-3,+3) 2.91** 0.65 6.82*** 3.42
(-2,+2) 2.46* 0.66 5.45** 2.45

Panel J: Sell in financial sector only, 5 events from 3 firms

(-3,+3) - - - -
(-2,+2) - - - -

Panel K: Buy in non-financial sector only, 93 events from 77 firms

(-3,+3) -0.10 -0.01 -3.83** -0.35
(-2,+2) 0.35 0.04 -1.56 -0.17

Panel L: Sell in non-financial sector only, 27 events from 20 firms

(-3,+3) 0.16 0.03 1.67 0.24
(-2,+2) 0.24 0.07 1.32 0.26

Panel M: Buy by high level in governance only, 76 events from 59 firms

(-3,+3) 0.11 0.02 0.68 0.06
(-2,+2) -0.22 -0.04 -0.50 -0.07
Panel N: Sell by high level in governance only, 26 events from 19 firms
(-3,+3) 0.27 0.06 0.91 0.14
(-2,+2) 0.38 0.12 0.92 0.18

Panel O: Buy by low level in governance only, 58 events from 45 firms

(-3,+3) 2.26* 0.51 4.07** 2.26
(-2,+2) 2.72* 0.69 3.04** 1.91

Panel P: Sell by low level in governance only, 6 events from 4 firms

(-3,+3) -0.32 -0.13 1.64 0.88
(-2,+2) -0.40 -0.23 1.21 0.75

Source: IMF staff estimates.

Note: Since there are no qualified observations before/after the corresponding event dates,
there are no results for the group of "sell in financial sector only (Panel J)".
 116

References

Beck, Roland, and Michael Fidora, 2008, “The Impact of Sovereign Wealth Funds on Global
Financial Markets,” ECB Occasional Paper Series No. 91 (Frankfurt: European
Central Bank).

Bortolotti, Bernardo, Veljko Fotak, William L. Megginson, and William F. Miracky, 2009,
“Sovereign Wealth Fund Investment Patterns and Performance“(unpublished;
University of Oklahoma).

Chhaochharia, Vidhi, and Luc Laeven, 2008, “Sovereign Wealth Funds: Their Investment
Strategies and Performance,” CEPR Discussion Paper No. 6959 (London: Center for
Economic Policy Research).

Deventer, Kathryn L., Xi Han, and Paul H. Malesta, 2009, “Firm Value and Sovereign
Wealth Fund Investments, Working Paper, University of Washington

Fotak, Veljko, Bernardo Bortolotti, and William Megginson, 2008, “The Financial Impact of
Sovereign Wealth Fund Investments in Listed Companies” (unpublished; University
of Oklahoma).

George R. Hoguet, 2008, “The Potential Impact of Sovereign Wealth Funds on Global Asset
Prices,” Vision, Vol. 3, Issue 2, pp. 23–30.

International Monetary Fund, “Do Sovereign Wealth Funds have a Volatility-Absorbing

Market Impact?” Global Financial Stability Report, April 2008. (Washington:
International Monetary Fund).

Hammer, Cornelia, Peter Kunzel, and Iva Petrova, 2008, “Sovereign Wealth Funds: Current
Institutional and Operational Practices,” IMF Working Paper 254.

Jen. Stephen and David K. Miles, 2007, “Sovereign Wealth Funds and Bond and Equity
Prices,” Morgan Stanley Research (31 May, 2007).

Knill, April, Bong-Soo Lee, and Nathan Mauck, 2009, “‘Sleeping with the Enemy’ or An
Ounce of Prevention’: Sovereign Wealth Fund Investments and Market Instability,”
Working Paper, Florida State University

Kotter, Jason, and Ugur Lel, 2008, “Friends or Foes? The Stock Price Impact of Sovereign
Wealth Fund Investments and the Price of Keeping Secrets,” International Finance
Discussion Papers No. 940, Board of Governors of the Federal Reserve System.
 117

Kozack, Julie, Douglas Laxton, and Krishna Srinivasan, forthcoming, “Macroeconomic

Implications of Sovereign Wealth Funds,” IMF Working Paper.

Lam, Raphael W., and Macro Rossi, forthcoming, “Sovereign Wealth Funds—Risk Sharing
and Financial Stress,” IMF Working Paper.

Sun, Tao, and Heiko Hesse, 2009, “Sovereign Wealth Funds and Financial Stability—An
Event Study Analysis,” IMF Working Paper 2009/230 (Washington: International
Monetary Fund). Also published in the International Finance Review, 2011, 12, 245-
262 and VOX.

Truman, Edwin M., 2008a, “A Blueprint for Sovereign Wealth Fund Best Practices.”
Peterson Institute Policy Brief, Number PB08-3, April 2008,
http://www.ciaonet.org/pbei/iie/0001182/index.html.

———, 2008b, “The Management of China's International Reserves and Its Sovereign
Wealth Funds.” Paper prepared for the Chinese Academy of Social Sciences Conference
Marking the 30th Anniversary of the Reform and Opening-up, Beijing, China,
December 16-17, 2008, Peterson Institute for International Economics.
 118

VI. RECENT CREDIT STAGNATION IN THE MENA REGION: WHAT TO EXPECT? WHAT
CAN BE DONE? WITH ADOLFO BARAJAS, RALPH CHAMI AND RAPHAEL ESPINOZA, 201067

We examine the recent credit slowdown in emerging markets from three analytical angles.
First, we find that, similar to past history, a credit boom preceded the current slowdown in
many emerging markets, and argue that, going forward, a protracted period of sluggish
growth is likely. Second, we focus on a relatively understudied region, the Middle-East and
North Africa (MENA), using a more detailed banking data. We uncover a key role played by
bank funding, in particular, deposit growth and external borrowing slowed considerably,
despite expansionary monetary policy. Finally, we show that bank-level fundamentals—
capitalization and loan quality—helped to explain differences in credit growth across banks
and countries.
A. Introduction

In many regions of the world, bank credit experienced a marked turnaround in the recent
crisis. After accelerating to peak real annual growth rates often in excess of 50 percent, and
sometimes near 100 percent (Angola, Iraq, Montenegro, Malawi, Sudan) before the global
crisis, credit decelerated sharply, by an average of nearly 30 percentage points, with several
countries experiencing declines of more than 40 percentage points (Figure 20). Continued
sluggishness of bank credit can have serious consequences for economic activity. To the
extent that credit is constrained on the supply side, sectors, firms, and households that are
particularly dependent on bank financing are either forced to scale back their consumption
and investment plans or resort to alternative sources of funding, thus creating a drag on the
economic recovery. In the longer run, slow credit growth will delay financial deepening, in
turn limiting the growth potential of the economy. Furthermore, for oil exporting countries,
spending cutbacks tend to fall disproportionately on the non-oil private sector, for which
alternative sources of funding are scarce, thereby inhibiting the process of economic
diversification. Therefore, policymakers around the world are justifiably concerned regarding
the causes of the credit slowdown and what actions they might take to spur a recovery in
credit.

67
This chapter is based on Barajas, Chami, Espinoza, and Hesse (2010).
 119

Figure 20.
Recent declines in Real Credit Growth (selected countries, year-on-year)
120
Central and Eastern Middle-East and
100
Europe North Africa Emerging Asia
80 Sub-Saharan Africa
Western Hemisphere
60

40

20

0
Paraguay

China,P.R.:Macao
Belarus

United Arab Emirates

Brazil

Jamaica

Togo

Algeria

Malaysia
Bangladesh
Uruguay

Mexico

Zambia

Iraq
Ukraine

Jordan
Chile

Malawi
Costa Rica

Tanzania

Cameroon

Oman

Morocco

Lao People's Dem.Rep

Thailand
Central African Rep.

Senegal
Croatia

Peru

Pakistan
Mozambique

Mali

Qatar
Romania

Dominican Republic

Angola

Kenya

Georgia

Egypt
Kazakhstan

Tunisia
Syrian Arab Republic

Myanmar
Argentina

Ecuador

Bolivia

Armenia
Ethiopia

South Africa

Azerbaijan, Rep. of

Saudi Arabia
Turkey

Colombia

Uganda

Libya

Cambodia

Mongolia

Sri Lanka
Montenegro, Rep. of

Russian Federation

Gabon

Côte d'Ivoire

Yemen, Republic of

Afghanistan, I.R. of

Philippines
Ghana

India
Serbia, Republic of

Iran, I.R. of
Venezuela, Rep. Bol.

Congo, Republic of

Indonesia
-20

‐40

Note: Data is claims on private sector, deflated by the CPI. High point of the arrow is the largest year-on-year growth rate of
real credit since 2006Q1. Low point is the last available data in IMF’s International Financial Statistics.

In a worldwide study of credit growth, Aisen and Franken (2010) report that 95 percent of
the 80 countries in their sample experienced a contraction in real terms in at least one month
following the Lehman Brothers bankruptcy in September 2008. Their study then uncovers
several determinants of credit growth during this period; in particular, the occurrence of a
prior credit boom is associated with lower credit growth, as is a decline in GDP growth of
trading partners. It also finds that structural conditions, such as the degree of financial depth
and integration are relevant predictors of credit growth, and that conventional countercyclical
monetary policy—reductions in policy rates—have been effective in dampening the credit
decline. Similarly, Cihak and Koeva Brooks (2009) focus on recent sluggish credit growth in
the Euro Area, and find that bank soundness—as measured by banks’ distance to default—is
significantly linked to bank-level supply of credit. Using an Instrumental Variable approach,
their analysis also shows that credit growth in turn has an impact on economic activity.

Other studies have also explored adverse consequences of declining credit growth. Abiad,
Dell’Ariccia, and Li (2010) investigate recoveries from recessions, and find that roughly one-
fifth of them are “creditless”, in the sense that real credit growth is zero or negative in the
first three years of the recovery. Creditless recoveries are more likely to occur following a
credit boom and/or a banking crisis, and compared to recoveries with more “normal” credit
growth, they tend to be substantially weaker. Kannan (2010), on the other hand, concentrates
on the aftermath of financial crises, where impaired credit conditions are also linked to a
weaker economic recovery.

Given these documented links between credit conditions and economic activity, this paper
explores the recent decline in credit growth in emerging and developing countries to better
understand the causes of the decline and to suggest avenues for policy. Furthermore, it pays
particular attention to a relatively understudied region, the MENA. While the Aisen and
Franken (2010) study makes an important contribution in analyzing this phenomenon
 120

worldwide, it includes only six MENA countries in its sample,68 and does not incorporate a
measure of quantitative or unconventional monetary policy, which in some cases may have
had an even greater impact on credit growth than movements in policy interest rates. The
paper looks at recent credit growth from three different analytical angles.

First, it uncovers the frequency of credit booms and busts across different regions. The
methodology used, common in the academic literature,69 identifies credit booms as episodes
during which credit is not only growing at a high rate, but also is surpassing its long-run
trend by a “large enough” amount. Findings show that credit booms occurred across regions
in the years before the crisis, with Central and Eastern Europe (CEE) the most affected
region and Sub-Saharan Africa the least affected. Moreover, the historical pattern of credit
surrounding booms suggests that the subsequent sharp credit slowdown is likely to be
followed by a protracted recovery.

Second, the paper conducts a decomposition of banks’ balance sheets in the MENA in the
pre- and post-crisis periods, in line with the approach followed in the Barajas and Steiner
(2002) study of credit stagnation in Latin America in the late 1990s. The main result of the
balance sheet decomposition for this region during the current credit cycle is the dominant
role played by deposits and capital, a marked slowdown of which severely constrained
banks’ ability to lend. This was true for most countries, and this effect was often exacerbated
by difficulties to obtain external financing. On the other hand, there is also evidence that
fiscal and monetary policy—through quantitative means—served to dampen the slowdown in
many countries.

Third, to complement the macro-level analysis the paper delves deeper into the bank balance
sheets of the financial institutions in the MENA region to uncover several key determinants
of bank lending growth, both on the supply and demand side, offering clues as to what pre-
conditions need to be in place for a revival of credit. Bank level panel data regressions for a
subset of eleven MENA countries confirm some findings from the balance sheet
decomposition. On the supply side, deposit growth is found to be the significant driver,
followed by capitalization. Increasing loan loss provisions—indirectly reflecting worsening
loan quality—can be expected to slow lending growth. Lending growth is also associated
with higher overall costs, in response to which banks maintain higher interest margins.
Similarly, favorable macroeconomic conditions, reflecting both supply and demand factors,
are found to spur bank lending. Real GDP growth, and oil prices—in oil-exporting countries
only, however— are associated with stronger lending activity.

68
Egypt, Jordan, Morocco, Saudi Arabia, Sudan, and Tunisia.
69
Gourinchas, Valdés, and Landerretche (2001); Mendoza and Terrones (2004); and Barajas, Dell’Ariccia, and
Levchenko (2007).
 121

Thus, the causes for the sharp credit slowdown have—by differing degrees—spanned both
demand and supply-side factors. On the supply side, banks were subjected to two types of
shocks: (1) an intense cutback in funding, as domestic deposit growth slowed sharply (in
Qatar, for example, deposits declined in nominal terms from mid-2008 to end-2009) and, in
some cases, external borrowing for banks was curtailed (in particular, in Kuwait from mid-
2008 to end-2009); and (2) increased strains on their balance sheets, as profitability fell and
nonperforming loans rose. On the other hand, the economic downturn depressed credit
demand and raised uncertainty about future investment prospects, thus heightening risk
aversion among both banks and prospective borrowers. Finally, as a result of shocks specific
to the region—the failure of Saudi conglomerates, the Dubai crisis, and the difficulties
surrounding investment companies in Kuwait—the credit culture may be undergoing a shift
away from name lending toward an approach based on accurate disclosure and appropriate
risk management.

What are the policy implications of the study? Reviving credit will necessarily take time.
Even as economic activity recovers—thereby lifting credit demand and reducing the
uncertainty that may be weighing on banks’ willingness to lend—credit recovery may lag, as
past experiences from the analysis on the frequency on credit booms and busts indicate.
Specifically, a protracted stagnation in credit can last up to three years before a recovery is
evident.

However, the analysis also shows that certain quantitative policies—by the central bank and
by the government, to restore part of the lost funds to the banking system—have been
effective during the credit slowdown, helping to dampen what could have been even more
serious contractions in credit growth. To the extent possible, these policies should be
maintained in order to avoid a further retrenchment in lending. The bank-level panel data
study suggests that a revival of credit growth will require two interrelated conditions: bank
balance sheets must improve, and the macroeconomic recovery, which in turn influences
deposit growth, must take hold.

As risk aversion may be affecting supply and/or demand for credit, there is also scope for
policy actions to temper it and thus contribute to a more rapid recovery in credit.
Policymakers can remove some of the regulatory uncertainty, particularly after introducing
extraordinary measures in addition to the injection of funds, such as increases in capital and
provisioning requirements, as well as blanket deposit guarantees. Toward the medium-term,
developing local debt markets will be crucial in order to expand the financing options for the
corporate sector, provide a key benchmark for pricing financial instruments in the economy,
and reduce the current high reliance on the banking system in this region.70 While

70
Basher, Dalla and Hesse (2010) examine the prospects and importance of local currency bond markets in the
GCC.
 122

pronounced bank credit cycles may be difficult to avoid in their entirety, their impact on
economic activity might be lessened with a more diversified financial system.

The paper is organized as follows. Section B analyzes the current boom-bust cycle in
emerging and developing countries. Section C presents the results from a balance sheet
decomposition of the credit slowdown, and Section D examines the drivers of credit growth
at the bank level both MENA banks. Section E concludes.

B. The Recent Credit Cycle in Historical and International Perspective

During the pre-crisis years, emerging and developing countries experienced particularly rapid
credit growth, on average peaking at an annual rate of 50 percent (Figure 21a). In order to
compare this recent experience to historical trends, this paper follows the methodology of
Gourinchas, Valdés, and Landerretche (2001) and Barajas, Dell’Ariccia, and Levchenko
(2007): first, a long run trend in credit growth is established, then periods of particularly high
credit growth, or credit booms, are identified as periods in which credit growth exceeds this
long-run trend by a “large enough” amount.
More specifically, the long-run trend is constructed by a rolling, backward-looking Hodrick-
Prescott filter applied to the annual ratio of credit to GDP in each country. A credit boom is
defined as an episode in which the following two thresholds are surpassed: (i) a relative
threshold by which the credit/GDP ratio exceeds its trend by at least 1.5 times the country-
specific historical standard deviation, and (ii) an absolute one whereby the credit/GDP ratio
increased by over 5 percentage points per year.

Applying this methodology over the past 25 years (1983–2008), it is apparent that the pre-
crisis experience is of high and above-trend credit growth in many emerging and developing
economies (Figure 21b). In Eastern and Central Europe, for instance, three countries were
identified as having credit booms: Estonia, Macedonia, and Croatia. Note that Turkey
satisfied the 1.5 standard deviation threshold, but did not qualify as a credit boom because
the increase in the credit-GDP ratio was below 5 percentage points per year.
 0.5
1.5
2.5
3.5
4.5

0
1
2
3
4
5
Turkey, 2006
Estonia, 2007
Macedonia, FYR, 2008

Europe
Eastern
Croatia, 2006
Central and
St. Lucia, 2007
Barbados, 2005
Costa Rica, 2008
Venezuela, Rep. Bol., 2007
Chile, 2008
Western

Honduras, 2007
Belize, 2008
Hemisphere

Bahamas, 2008
Jamaica, 2008

Zimbabwe, 2006
Nigeria, 2008
Uganda, 2008
Sudan, 2006
Sierra Leone, 2008
Malawi, 2008
Mauritius, 2008
Angola, 2008
Seychelles, 2008
123

Niger, 2008
Togo, 2007
São Tomé & Príncipe, 2005
credit/GDP

Liberia, 2008
Sub-Saharan Africa

Zambia, 2008
Congo, Dem. Rep. of, 2008
Swaziland, 2005
Guinea-Bissau , 2008
Senegal, 2005
Benin, 2005

U.A.E., 2008
Saudi Arabia, 2008
Syria, 2005
Sudan, 2006
Iran, 2007
North Africa

Jordan, 2006
Figure 21a. Credit Boom Events in the Last Expansion

Qatar, 2008
Middle-East and

Morocco, 2008

Vanuatu, 2008
Nepal, 2008
Maldives, 2005
Color code: red bar if the credit/GDP ratio increased by over 5 percentage points per year

Bhutan, 2008
Cambodia, 2007
Difference from each country’s historical trend, measured as multiple of standard deviation of

India, 2006
Solomon Islands, 2007
Emerging Asia

Papua New Guinea, 2008

Fiji, 2006
Pakistan, 2005
 124

Figure 21b. Regional Average Differences from Trend

2
1.8
1.6
1.4
1.2
1
0.8
0.6
0.4
0.2
0

SSA
MENA

WHM

CEE

Developing Asia

Oil Exporters
Note: Regional avearages are unweighted averages.

Regional averages reveal that this behavior was a worldwide phenomenon. Although larger
than the average expansion in Western Hemisphere (WHM) countries, credit expansion in
MENA countries was very similar to that of developing Asia, for example. Finally, the recent
credit expansions have been slightly more pronounced among oil exporters in general—in
fact on average, these countries are experiencing a credit boom—thus suggesting a possible
role of oil prices as a factor amplifying credit cycles.

Looking back at the experience over the past 25 years, it is apparent that credit booms are not
new. Countries around the world experienced these episodes at a frequency of around 4
percent, with Central and Eastern Europe the most affected region (this result may be biased
because the data starts only in the 1990s for this region) and Sub-Saharan Africa the least
affected region. (Figure 22). Nonetheless, the pre-crisis years stand out. From 2006–08,
credit booms emerged to an unprecedented degree, making this 3-year period the highest
concentration of such episodes over the past 25 years; in 2008 in particular, about 15 percent
of all countries were experiencing booms, with Developing Asia and the CEE having the
highest occurrence (Figure 23).

Past booms have most often been followed by a pronounced slowdown and prolonged
sluggishness. The retrospective analysis of the 1983–2008 period uncovers this common
pattern of behavior. From a median real growth rate of more than 20 percent, credit slows to
close to zero growth within two years, followed by only 5 percent for at least three years
(Figure 24). In Developing Asia and Sub-Saharan Africa, the median country’s real credit
even growth remained negative 3 years after the peak of the boom. In MENA, the current
slowdown will potentially contribute to a shift in the credit culture in the region, as banks de-
 125

emphasize name lending in favor of a more conventional and arms-length approach to

conducting business.71

Figure 22.
Frequency of Credit Booms throughout the World, 1983–2009
(percentage of country-years experiencing a boom each year)
9%
8%
7%
6%
5%
4%
3%
2%
1%
0%
MENA WHM CEE SSA Developing Oil
Asia Exporters

Figure 23. Boom Frequency

q y over Time
(Percentage of countries experiencing a credit boom each year)
0.3

MENA
0.25
WHM
0.2 CEE
Developing Asia
0.15
Oil exporters

0.1

0.05

0
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008

71
A series of interviews with leading financial sector analysts in the MENA region, conducted during February-
March 2010, revealed divergent views on the future of name lending in the region. While some believed that it
is destined to disappear going forward, most saw its slowdown as the more likely scenario, and that banking
practices in the region were likely to undergo a structural change in the coming years. See IMF (2010) for a
summary of the interviews.
 126

Figure 24. MENA:

Credit Behavior Surrounding Booms
(Median Country by Region)
0.4
Western Hemisphere
0.35
Central and Eastern Europe
0.3

Real creditgrowth (year-on-year)

Sub-Saharan Africa
0.25
Developing Asia
0.2
Middle-East and North Africa
0.15

0.1

0.05

0
-4 -2 0 2 4 6 8 10 12 14 16 18 20 22 24
-0.05
Quarters
-0.1

Note: Unweighted average of real credit growth around boom years identified between
1983 and 2008. Credit was deflated by the CPI index. Boom year in shaded area.

C. Anatomy of the MENA Credit Slowdown

We now focus on the MENA, a region that, according to the previous analysis of credit
booms and busts, appeared to be fairly representative of emerging and developing countries.
This narrower scope allows us to dig more deeply into the drivers of private sector credit.
When comparing the latest post-crisis slowdown period (mid-2008 to the most recent)72 with
a previous expansion period (end-2004 to mid-2008), average real credit growth in MENA
declined by about 17 percentage points (see Table 8), appreciably larger in the oil exporters
(20 percentage points) and in the GCC in particular (22 percentage points). However, as
Table 8 shows, several exceptions arise, in particular, four countries in which credit actually
accelerated significantly: Iraq, Libya, Djibouti, and Lebanon.
To gain further insight into possible drivers of the turnaround in credit, this section analyzes
the major changes occurring on banks’ balance sheets between the two periods. The logic of
the exercise is the following: in order to satisfy accounting identities, credit to the private
sector (CPS) necessarily moves along with movements in other accounts on the balance
sheet, as summarized in the following equation:

CPSt  Dt  K t  NFPSt  CBt  RWt (1)

72
The latest data obtained were for May 2009 (Sudan), June 2009 (Iran), November 2009 (Djibouti), December
2009 (Iraq and Yemen), January 2010 (Kuwait and Tunisia), February 2010 (Oman, Algeria, and Lebanon),
March 2010 (Bahrain, the UAE, Libya, Egypt, Morocco, Pakistan, and Syria), and April 2010 (Qatar, Saudi
Arabia, and Jordan).
 127

where D denotes deposits, K denotes capital and others, and the remaining three terms denote
banks’ net claims on the nonfinancial public sector (NFPS), the central bank (CB), and the
rest of the world (RW), respectively.73 Thus, private sector credit growth (as well as its
change) can be decomposed into changes in these other balance sheet items, which either
contribute to the decline, or offset it:
CPSt Dt Kt NFPSt CBt RWt
    
CPSt 1 CPSt 1 CPSt 1 CPSt 1 CPSt 1 CPSt 1 (2)
Average credit growth over each (expansion and slowdown) period was decomposed in this
manner, and its change from one period to the next was then attributed to the different
balance sheet items in each country. For example, in some cases credit growth declined
together with a slowdown in deposit growth, exacerbated by an increase in net claims on the
central bank (accumulation of bank reserves), and/or an increase in net claims on the rest of
the world (a slowdown in foreign borrowing). Furthermore, net claims on the nonfinancial
public sector declined in some cases (as government deposits increased, for instance),
providing an offset to the decline in credit growth.
The primary shock affecting most countries between the two periods was a marked
slowdown of funding sources, particularly deposits, which severely constrained banks’
ability to lend. With the exception of Jordan, all fourteen countries experiencing credit
slowdowns also saw deposit growth decline noticeably, by 12 percentage points on average.
Thus, the contribution of deposits to the credit slowdown was substantial, ranging from over
4 percentage points in Kuwait to almost 80 percentage points in Algeria. Capital and others
also slowed in most countries, contributing to a credit deceleration of 10 and 24 percentage
points in Egypt and Bahrain, respectively, although in some countries, such as Algeria and
Oman, an acceleration in this category served to dampen the credit slowdown by almost 2
percentage points.74

Banks’ position with central bank in most cases (11 out of 14) dampened the credit
slowdown, either as a result of expanding credit to the banking system or a reduction of bank
reserves. In Algeria, a drawdown in reserves served to offset perfectly the 80 percentage
point effect of the deposit slowdown; in 5 out of 6 of the GCC countries this position
dampened the slowdown by between 3 and 12 percentage points; and in Egypt the effect was
equal to over 20 percentage points. Although the data cannot distinguish between voluntary
and purely policy-induced changes, this analysis suggests that quantitative easing was often
used and with substantial effects on private sector credit growth.

73
Note that each of the net claims terms can be either positive or negative. Based on IFS categories, net claims
on the NFPS were defined as claims on the NFPS minus government deposits; those on the central bank were
defined as reserves and other claims on the central bank minus credit from the central bank; and those on the
rest of the world were defined as foreign assets minus foreign liabilities. Finally, all other items not included in
net claims or in deposits were grouped in a residual category, “capital and others”.
74
The interpretation of capital and others is not as straightforward as that of other categories, as it includes
residual items.
 128

The effect of banks’ positions with the government was more mixed. In nine cases, net
claims on the NFPS served to intensify the credit slowdown, by as much as 18 percentage
points in Egypt and Qatar. As in the case with the central bank, it is not possible to
distinguish voluntary lending to the government as a result of poor prospects in the private
sector from a crowding out effect of increased reliance on the banking sector to fund fiscal
deficits. On the other hand, in a few countries there is evidence that direct funding by the
government provided relief to banks (Saudi Arabia, Iran, and the United Arab Emirates).
Furthermore, in Syria, where credit growth remained constant at just under 20 percent per
year, banks’ net positions with the NFPS more than offset (13 percentage points) the effect of
a decline in deposit growth (8 percentage points).

Finally, the effects of banks’ positions with the rest of the world also differed substantially
across countries. Some countries, such as the UAE, experienced a marked decline in foreign
borrowing (contributing 16 percentage points to the credit slowdown), although generally of
a much smaller magnitude than that in deposits. In other countries, such as Bahrain and
Jordan, a drawdown of banks’ foreign assets served to compensate for lost funding, thus
dampening the credit slowdown, whereas in Saudi Arabia banks built up their foreign assets.
Finally, banks in Qatar were able to
dampen the slowdown both by drawing Figure 25. Decomposition of the Credit
down foreign assets and by borrowing Slowdown in Selected MENA Countries
abroad, with an overall effect of 22 15
(changes in percentage points)

percentage points. Position with nonfinancial public sector

Position with Central bank

10

Two contrasting country examples Position with rest of the world

Deposits
highlight the differences in how the Capital and others

credit slowdowns were reflected in the

5

aggregate balance sheet of the banking

sector, as shown in Figure 25. First, in 0
UAE JOR

the United Arab Emirates, funding

declined sharply. The slowdown in -5

deposits and capital alone would have

led real credit growth to decline by -10

more than 16 percentage points.

Reinforcing this was a decline in -15

external borrowing, accounting for an

additional 16 percentage points.
However, a combination of a fall in
-20

bank reserves (12 percentage points), Note: This graph decomposes declines in private sector credit growth from expansion to slowdown
periods into movements in other balance sheet accounts, classified as the banking system's position with

and an increase in government deposits respect to (i) the central bank, (ii) the nonfinancial public sector, (iii) the external sector, and (iv) deposits
and other liabilities with the domestic private sector. A negative value indicates a contribution to the
slowdown, whereas a positive denotes an offsetting effect, dampening the slowdown. The sum of the
(0.5 percentage points), dampened the four components is equal to the change in the credit growth rate.

credit slowdown, which ultimately amounted to 19 percentage points. In Jordan, on the other
hand, deposit growth actually accelerated between pre-crisis and post-crisis periods, which
would have raised real credit growth by about one percentage point. In addition, some banks
 129

were able to transfer funds from abroad (5 percentage points). Therefore, the decline in real
credit growth between periods was primarily associated with a sizable increase in reserves
with the central bank (slowing real credit by 13 percentage points) and a slowdown in capital
(9 percentage points).

For many MENA countries, the loan-deposit ratio also fell from mid-2008 to early 2010,
with declines ranging from 2 to 13 percentage points (Figure 26). This could reflect (1)
additional funding difficulties, in particular in external borrowing; (2) lack of willingness to
lend on the part of banking systems, because of increased macroeconomic or regulatory
uncertainty post-crisis; or (3) sluggishness in demand for credit, also due to the weak
macroeconomic environment.

Figure 26. Loan-Deposit Ratios in Selected MENAP Countries

(percent )

120
June 2008
Latest date
100
Change

80

60

40

20

0
BHR KWT OMN QAT SAU UAE DZA IRN SDN YMN EGY JOR MAR PAK

-20

Source: International Financial Statistics and authors' calculations.

D. Econometric Analysis of Bank-Level Credit Growth

An analysis of individual bank behavior across a subset of MENA countries in the pre-crisis
period uncovered several key determinants of bank lending, both on the supply and demand
side, offering clues as to what pre-conditions need to be in place for a revival of credit. Panel
data regressions were run for bank-specific credit growth using annual balance sheet and
income statement data obtained from the BankScope database. We mainly used fixed effects
estimations to account for any time-invariant unobserved characteristics.75 To examine
whether country-specific factors have a significant impact on lending growth, country

75
The Hausman specification test indicates that the fixed effects estimator tends to be preferred over the random
effects models. We did not estimate dynamic panel models using GMM for this study given the large
heterogeneity of the banks in the sample.
 130

dummy variables were included into robustness random effects regressions. The sample
consisted of annual data covering 1997–2008 for large commercial, investment, and Islamic
banks in eleven countries (Bahrain, Egypt, Jordan, Kuwait, Lebanon, Morocco, Oman, Qatar,
Saudi Arabia, Tunisia and UAE). 76 As expected, the bank data were very heterogeneous,
comprising consolidated and unconsolidated data, and different years of coverage.77

The basic regression equation was the following:

Cit   i  FUNDit  MACRO jt  INF jt  DumBanki  DumCountry j (3)
where subscripts i, j, and t denote bank, country, and year, respectively. The dependent
variable C is the annual nominal growth rate of credit scaled by lagged total assets. By
controlling for inflation (INF), the remaining right-hand side variables essentially explained
credit growth in real terms. FUND refers to the set of individual bank fundamentals:

Deposit growth is defined as the year change of total deposits scaled by total assets, and a
positive relationship is expected with loan growth. Banks with more funding availability will
be able to better perform their financial intermediation function and should have stronger
lending growth. Higher net margin growth, defined as the change in the net interest rate
margin, scaled by assets, leads to higher bank profits providing banks with higher retained
earnings and the capacity to marginally also increase their lending activities. Bank liquidity
and capitalization are proxied by liquid assets over deposits and equity scaled by assets,
respectively. While higher liquidity buffers tend to signal greater bank soundness, its impact
on lending growth can be two ways. On the one hand, banks’ investments into more liquid
assets could imply foregone illiquid lending elsewhere while on the other hand, the bank
soundness argument could dominate, and stronger banks engage in more marginal lending.
Similarly, capitalization is an important driver of bank lending, as strong capitalized banks
have a higher capacity to extend lending than weakly capitalized banks.

To proxy for worsening loan quality, we use loan loss provisions (scaled by assets), and it is anticipated
that loan quality is negatively related to lending growth.78 We measure banks’ cost effectiveness by the
cost-income ratio. Banks that have higher costs relative to income, possible due to higher wages, a
larger branch network or more loan officers, might have higher marginal lending. We also proxy for the
size of the bank by total assets. Larger banks with a higher branch network or typically larger project
financing might be inclined towards higher lending growth.

76
At the time of the study, balance sheet and income statement information was not yet available for the
majority of the banks for 2009.
77
Consolidated balance sheet and income statement data was used when available, but when consolidated data
was not available for a bank, unconsolidated data was used instead. In addition, some obvious outliers were
eliminated.
78
A more direct measure of loan quality is the nonperforming loan ratio. However, this variable was not
available for a number of banks, and therefore it was not used.
 131

A set of macroeconomic controls was also included to account for supply and demand-side
factors simultaneously affecting all banks in the same country: real GDP growth, aggregate
financial deepening (as measured by the change in the credit-GDP ratio), and the price of oil,
the latter of which was also interacted with a dummy variable indicating whether the given
country is an oil exporter. The lending of banks in oil exporting countries should be more
sensitive to swings in the oil price. Overall, favorable macroeconomic conditions should be
conducive to higher lending growth. In addition, we also differentiate between commercial,
investment and Islamic banks since their willingness and capacity to extend lend might
differ. Investment banks in particular have a different business model with their greater
reliance on wholesale funding and usually lack of deposit funding. Thus, we include dummy
variables for investment and Islamic banks.

The econometric results indicate that, on the supply side, bank-specific fundamentals play an
important role in determining bank-level credit growth. As in the balance sheet
decomposition shown in the previous section, funding was also crucial. Banks with higher
deposit growth tended to expand credit more rapidly. As in the Peek and Rosengren (1995)
and Barajas et al. (2010) studies of credit decline in the U.S., bank capital was shown to be
significantly linked to credit growth; MENA banks with higher prior levels of capital were
able to expand credit more rapidly. Higher loan loss provisions—indirectly reflecting
worsening loan quality79—would tend to slow lending growth. In addition, lending growth
was associated with higher overall costs, in response to which banks maintained higher
interest margins. The findings for liquid assets were ambiguous across the different model
specifications. Finally, there was evidence that Islamic banks tended to increase credit more
rapidly, followed by commercial banks, then investment banks. This suggests that Islamic
banks could have been subject to some catch-up in recent years, and may also have been
influenced by a business model more geared towards investments and lending in high growth
areas such as real estate.80

Similarly, favorable macroeconomic conditions, reflecting both supply and demand factors,
were found to spur bank lending. Real GDP growth, and oil prices—in oil exporting
countries only, however—were associated with stronger lending activity. In addition, banks
were subject to common country-specific trends in credit growth, thus reflecting other
changes in macroeconomic conditions or in attitudes toward risk.

79
Ideally, a more direct measure of loan quality, such as the ratio of nonperforming loans, would have been
used. However, it was not available for a large number of bank-years within this sample, and would have
greatly reduced the number of usable observations.
80
As robustness tests (not shown in Table 2), we also estimated quantile regressions at the 25th and 75th
percentile of credit growth. As expected, for banks with higher lending growth at the 75th percentile, bank
fundamentals such as deposit and net margin growth as well as capitalization were the main drivers. Overall, the
results in the quantile regressions were broadly consistent with the baseline fixed effects regressions.
 132

Finally, other country-specific factors influenced credit growth. All else equal, banks in the E
tended to expand credit more rapidly than in other countries, while there is some evidence
that those in Egypt and Lebanon expanded credit less rapidly than the rest. The latter is
consistent with the finding that these countries were also the only two where pre-crisis
private sector credit had been below trend.

Figure 27 Drivers of Lending Growth in MENA Banks

5
Drivers of Lending Growth in MENA Banks
(percent)
4

Bank-specific drivers Macroeconomic

3 drivers

-1

Note: this figure shows the estimated impact of moving from low to a high value in each of the
explanatory variables driving credit growth. It is calculated as the change in the respective
variable (from the 25th to the 75th percentile) times the estimated coefficient, and is based on
panel regressions on a sample of banks across ten MENA countries during the 1997-2008 period.
Source BankScope and IMF Staff estimates.

The estimated quantitative effects of the explanatory variables were also found to be
substantial. By comparing the predicted difference in credit growth between a relatively low
value (25th percentile) for a given variable with that of a relatively high value (75th
percentile), one gets a measure of the magnitude of each variable’s impact, as shown in
Figure 27.81 The largest quantitative impact is derived from deposit growth; almost 5
percentage points separated annual credit growth of a high deposit growth bank with one in
which deposits were growing relatively slowly. Relative to banks at the lower end of the
distribution, banks with high levels of capitalization and costs tended to display credit growth
of about 2 percentage points higher; those with higher growth in interest margins did so by
about 1½ percentage points; and those with high loan loss reserves increased credit by ½
percentage point less. At the macro level, the difference in trends in countrywide lending—as
proxied by the increase in the credit-GDP ratio—accounted for just over 3 percentage points
in bank-level credit growth, while country-years in which economic activity was particularly
high led to ½ percentage point increase in credit growth. Finally, high oil prices tended to be

81
For this exercise, specification (4) from the regression results reported in Table 2 is used.
 133

associated with roughly one percentage point increase in credit growth, relative to periods of
low oil prices.

The results therefore suggest that a revival of credit growth in the MENA will require two
interrelated conditions: bank balance sheets must improve, and the macroeconomic recovery,
which in turn influences deposit growth, must take hold. The former implies that
capitalization levels increase, asset quality recover, and profit margins be restored so that
banks can embark on (relatively costly) lending activities. In oil-exporting countries, both
supply and demand should respond favorably to the recovery in oil prices.

E. Conclusion

We examine from several angles the post-crisis credit slowdown experienced by many
emerging and developing countries in the wake of the global crisis. We first show that the
slowdown was preceded in many cases by a pre-crisis credit boom. Second, when taking a
longer retrospective view, over the past 25 years, we find that the frequency of booms had
increased in the years 2006 -2008 to unprecedented levels. Third, a prototypical pattern of
post-boom credit growth emerged, suggesting that credit may not recover fully for three
years or more in many of these countries. Fourth, focusing on the MENA region and turning
to possible causes or factors during the slowdown, the balance sheet decomposition and the
econometric analysis pointed to tightness in funding sources both domestic and external,
which limited banks’ ability to lend. However, there was also evidence that monetary policy,
by restoring at least part of the lost funding, served to dampen the slowdown in banks’
lending capacity. Fifth, deteriorating macroeconomic conditions—declining domestic
economic activity and falling oil prices—also played a role in reducing demand for credit and
as well as banks’ willingness to lend. Finally, weakening bank balance sheets, as reflected in
lower loan quality and diminished bank capital, also worked to lower supply of credit.

Thus, reviving credit in emerging and developing economies remains a challenge. For those
countries encountering their slowdown in the aftermath of a credit boom, history provides a
sobering outlook for the next few years: a quick and robust rebound simply should not be
expected. On the other hand, just as previous research showed that conventional monetary
policy had achieved some effectiveness in dampening the slowdown worldwide, this paper’s
analysis showed that unconventional policy—in MENA in particular— had also been
effective. What remains most difficult is restoring banks’ willingness to lend. Of course, as
the economic recovery proceeds, the medium term outlook for bank lending should improve,
and the supply of credit—along with its demand—should begin to recover. But lingering risk
aversion, partly a product of the lack of transparency on the direction of financial regulation
in the wake of the global financial crisis, will undoubtedly prove more difficult to overcome.
Finally, some measure of credit slowdown may in fact be desirable for some time, as banks
shake off the excesses of the past and possibly adapt their practices to a newer approach in
which name lending is phased out in favor of modern, arms-length relationships.
 Table 8. Balance Sheet Decomposition of Changes in Credit Growth in the MENA Region
Average real annual credit growth Decomposition of the change in credit growth Average rate of growth of deposits
Expansion Slowdown Change Contribution (percentage points) Expansion Slowdown Difference
Funding Banks' positions with:
Capital and Nonfinancial Rest of the
Deposits Central bank
others public sector world

Countries with credit slowdowns

Oil exporters
Gulf Cooperation Council (GCC)
Bahrain 26.4 4.1 -22.3 -26.9 -23.9 8.9 -5.1 24.8 24.8 2.4 -22.3
Kuwait 18.9 1.8 -17.1 -4.4 -3.2 -0.7 0.1 -9.2 13.2 10.2 -3.0
Oman 17.8 9.5 -8.3 -12.9 1.8 6.2 -2.3 -0.8 21.7 6.0 -15.7
1
Qatar 54.6 10.0 -44.7 -50.3 -8.0 9.1 -18.1 22.6 44.8 10.5 -34.2
Saudi Arabia 20.0 -0.6 -20.6 -13.7 -4.8 2.9 3.2 -7.6 14.2 3.7 -10.5
U.A.E. 27.0 7.7 -19.3 -17.3 1.1 12.1 0.5 -15.7 21.0 5.1 -15.8
Algeria 17.0 8.4 -8.6 -79.8 1.7 79.0 -12.2 2.7 16.6 -3.5 -20.1
Iran 12.3 -7.4 -19.7 -8.0 -7.5 -2.0 1.0 -3.2 6.6 -0.8 -7.5
Sudan 22.2 8.6 -13.6 -6.4 0.8 0.1 -12.5 4.3 15.6 12.3 -3.2
Yemen 13.2 -9.8 -23.0 -12.7 -0.3 25.7 -19.3 -16.4 8.8 4.7 -4.1
Unweighted average oil exporters 22.9 3.2 -19.7 18.7 5.1 -13.7

Oil Importers
Egypt -0.7 -6.1 -5.4 -9.3 -10.4 20.7 -17.7 11.3 2.8 -2.3 -5.1
Jordan 14.2 -1.7 -15.9 0.8 -9.0 -12.5 0.1 4.8 6.9 8.2 1.4

134
134
Morocco 16.7 11.3 -5.5 -10.6 -1.7 5.8 -0.7 1.7 12.0 4.2 -7.8
Pakistan 5.3 -8.6 -14.0 -12.5 -1.4 7.4 -8.3 0.8 4.6 -5.3 -9.9
Unweighted average oil importers 8.9 -1.3 -10.2 6.6 1.2 -5.4

Unweighted average slowdowns 19.2 2.0 -17.2 15.5 4.0 -11.4

Countries without credit credit slowdowns

Iraq 23.0 32.8 9.7 258.9 -68.3 -50.6 -216.8 86.5 -5.1 43.7 48.9
Libya 1.9 16.7 14.8 -20.7 6.1 40.6 -18.9 7.7 30.1 14.2 -15.9
Djibouti 4.9 25.5 20.6 46.7 0.0 -4.0 -0.4 -24.5 4.8 20.5 15.7
Lebanon 1.1 12.1 10.9 40.2 11.9 -40.4 -12.4 11.6 3.2 16.0 12.8
Syria 19.1 19.6 0.6 -7.6 8.0 11.4 13.1 -24.5 5.8 5.5 -0.4
Tunisia 4.9 7.7 2.8 -0.9 0.9 1.7 -1.0 2.1 8.6 7.0 -1.6

Unweighted average 9.1 19.1 9.9 7.9 17.8 9.9

Source: International Financial Statistics and authors' calculations.

Note: this table decomposes the change in the average annual real growth rate in credit to the private sector from the expansion (2004:12 - 2008:6) to the slowdown (2008:6 - most recent) periods
into five balance sheet categories. The contribution of each category is scaled so that the sum is equal to the change in growth rate, thus a positive (negative) contribution means that it contributes to
an acceleration (deceleration) in credit growth. In addition, the last three columns show the own growth rate of deposits in the two periods, as well as its change.
1 Nominal credit growth rate shown.
 135

Table 9. MENA Countries-Regressions for Bank-Level Loan Growth

Annual Data: 1997-2008
(1) (2) (3) (4) (5) (6)

Bank fundamentals
Deposit Growth 0.307 0.329 0.29 0.286 0.304 0.311
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
Net Margin Growth 2.344 2.603 2.116 2.054 2.159 2.318
(0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)*** (0.000)***
Liquid Assets/ Deposits (-1) -0.003 -0.019 0 0.004 -0.006 -0.013
(0.779) (0.042)** (0.964) (0.678) (0.480) (0.177)
Equity/ Assets (-1) 0.268 0.315 0.224 0.219 0.156 0.168
(0.000)*** (0.000)*** (0.001)*** (0.002)*** (0.005)*** (0.003)***
Loan Loss Reserves/ Assets (-1) -0.351 -0.298 -0.193 -0.139 -0.125 -0.234
(0.000)*** (0.000)*** (0.044)** (0.147) (0.135) (0.006)***
Cost-Income Ratio (-1) 1.854 0.758 2.179 2.25 1.115 0.922
(0.000)*** (0.018)** (0.000)*** (0.000)*** (0.000)*** (0.003)***
Asset Size (-1) 0.004 0.002 0.002 0.001 0 0.001
(0.000)*** (0.000)*** (0.008)*** (0.080)* (0.623) (0.316)
Other bank characteristics
Islamic Dummy 0.042 0.028 0.031
(0.005)*** (0.040)** (0.030)**
Investment Dummy -0.057 -0.053 (0.051)
(0.004)*** (0.003)*** (0.006)***
Macro variables
Growth of Private Credit/ GDP 0.164 0.242 0.244
(0.000)*** (0.000)*** (0.000)***
Growth of Real GDP 0.144 0.111 0.095 0.084
(0.017)** (0.066)* (0.118) (0.174)
Inflation 0.427 0.42 0.494 0.554
(0.000)*** (0.000)*** (0.000)*** (0.000)***
Oil Price Change 0.009 0.008 -0.03
(0.614) (0.671) (0.062)*
Oil Price Change * Oil Exporter 0.074 0.077 0.030
(0.003)*** (0.002)*** (0.165)
Constant -0.033 -0.005 -0.058 -0.066 -0.028 -0.002
(0.013)** (0.665) (0.000)*** (0.000)*** (0.156) (0.921)

Fixed or Random Effects FE RE FE FE RE RE

Country dummies included? No No No No Yes Yes
Observations 1,270 1,270 1,227 1,227 1,227 1,270
Number of banks 154 154 154 154 154 154
R-squared 0.435 0.479 0.491

This table reports the results of fixed effects and random effects regressions for annual growth of bank loans in eleven MENA
countries over the 1997-2008 period. Regressions (5) and (6) include ten country dummies, the coefficients of which are not
reported here. Note that in regression (5) the only country dummy that is significant is that for United Arab Emirates
(positive), while in regression (6), those for Egypt and Lebanon are significant as well (both negative). P-values are shown in
parentheses, * significant at 10%; ** significant at 5%; *** significant at 1%.
 136

REFERENCES
Abiad, Abdul, Giovanni Dell’Ariccia, and Bin Li, 2010, “Creditless Recoveries,”
(unpublished: International Monetary Fund).

Aisen, Ari and Michael Franken, 2010, “Bank Credit During the 2008 Financial Crisis: A
Cross-Country Comparison,” IMF Working Paper 10/47.

Barajas, Adolfo, Ralph Chami, Thomas Cosimano, and Dalia Hakura, 2010, “U.S. Bank
Behavior in the Wake of the 2007-09 Financial Crisis,” IMF Working Paper 10/131
(Washington: International Monetary Fund).

Barajas, Adolfo, Ralph Chami, Raphael Espinoza, and Heiko Hesse, 2010, “Recent Credit
Stagnation in the Mena Region; What to Expect? What Can Be Done?,” IMF
Working Paper 2010/219 (Washington: International Monetary Fund). A version was
also published in World Economics, 2011, 12(2), 153-176.

Barajas, Adolfo, Giovanni Dell’Ariccia, and Andrei Levchenko, 2007, “Credit Booms: The
Good, the Bad, and the Ugly,” (unpublished: International Monetary Fund).

Barajas, Adolfo and Roberto Steiner, 2002, “Why Don’t They Lend? Credit Stagnation in
Latin America,” Staff Papers, International Monetary Fund, Vol. 49, pp. 156–184.

Basher, Syed, Ismail Dalla, and Heiko Hesse, 2010, “Gulf Cooperation Council local
currency bond markets and lessons from East Asia,” VOX Column (May 22, 2010).
Available via the Internet: http://www.voxeu.org/index.php?q=node/5082.

Cihak, Martin and Petya Koeva Brooks, 2009, “From Subprime Loans to Subprime Growth?
Evidence for the Euro Area,” IMF Working Paper 09/69 (Washington: International
Monetary Fund).

Gourinchas, Pierre-Olivier, Rodrigo Valdés, and Oscar Landerretche, 2001, “Lending

Booms: Latin America and the World,” Economía, Vol.1, No. 2, pp. 47–99.

International Monetary Fund, 2010, Regional Economic Outlook: Middle East and Central
Asia, May 2010 (Washington).

Kannan, Prakash, 2010, “Credit Conditions and Recoveries from Recessions Associated with
Financial Crises,” IMF Working Paper 10/83 (Washington: International Monetary
Fund).
 137

Mendoza, Enrique and Marco Terrones, 2008, “An Anatomy of Credit Booms: Evidence
from Macro Aggregates and Micro Data,” IMF Working Paper 08/226 (Washington:
International Monetary Fund).

Peek, J. and Eric Rosengren, 1995, “The Capital Crunch: Neither a Borrower nor a Lender
Be,” Journal of Money, Credit and Banking, Vol. 27, No. 3, pp. 625–638.
 138

VII. FINANCIAL SPILLOVERS AND DELEVERAGING: THE CASE OF ROMANIA, 20121

A. Introduction

Beyond financial spillovers, Romania’s growth trajectory and domestic credit

performance is strongly influenced by developments in Western Europe. According to
the IMF 2012 Spillover Report, a one percent growth shock in Western Europe gives rise to a
shock of about equal size in CESEE. Banking linkages are an important separate conduit for
spillovers. The cross-border banking model used in the Spillover Report finds that a 1 USD
change in cross-border exposure of western banks vis-à-vis CESEE banks translates over
time into a 0.8 USD change in domestic credit. And each extra percentage drop in real credit
growth leads to about 0.3 percentage point reduction to real GDP growth. So any
intensification of the Euro area crisis that would cause disorderly deleveraging of parent
banks could significantly impact private sector credit growth in Romania.2

The risk of disruptive parent funding withdrawals by European banks from CESEE
has been a longstanding concern. Some orderly deleveraging is unavoidable given past
excessive FX driven credit booms and European banks’ desire to shrink non-core assets over
time. Disorderly foreign bank deleveraging can risk a credit crunch, balance of payment
stress and loss of reserves, a sharp depreciation, increases in risk premia as well as spillovers
to the real economy. Excessive deleveraging in CESEE countries has been prevented thus
far, partly thanks to the European Bank Coordination Initiative (EBCI) which encouraged
parent banks to maintain exposure to their subsidiaries.3 The ECB’s LTROs have also
provided some funding relief to parent banks but the LTRO effect is diminishing. Compared
to other emerging market regions, the CESEE has seen larger foreign bank deleveraging
since the Lehman Brothers collapse in September 2008, with the exposure to Asia & Pacific
and Latin America & Caribbean by far exceeding the level in September 2008.

1
This chapter is based on Hesse (2012).
2
The usual caveats of directly translating the average cross-country effect (to the CESEE) onto Romania should
be considered in the above estimates given some country-specific heterogeneities.
3
EBCI (2012) provides an analysis of deleveraging in the CESEE.
 139

Figure 28. Banks’ External Positions

External Positions of BIS-reporting Banks vis-à-vis Emerging Market Regions

(Sept. 2008 = 100, exchange-rate adjusted)

Africa & Middle East Asia & Pacific

CESEE Latin America/Caribbean
160

140

120

100

80

60

40

20

0
Mar-03

Mar-04

Mar-05

Mar-06

Mar-07

Mar-08

Mar-09

Mar-10

Mar-11

Mar-12
Sep-03

Sep-04

Sep-05

Sep-06

Sep-07

Sep-08

Sep-09

Sep-10

Sep-11
Jun-03

Jun-04

Jun-05

Jun-06

Jun-07

Jun-08

Jun-09

Jun-10

Jun-11
Dec-03

Dec-04

Dec-05

Dec-06

Dec-07

Dec-08

Dec-09

Dec-10

Dec-11
Sources: BIS; and IMF staff calculations.

Romania has been strongly impacted by the financial crisis in 2008/09 but also recently
from the intensification of the euro area crisis. Both CDS and Emerging Markets Bond
Index Global (EMBIG) spreads have been steadily increasing again to levels that remain
lower than Hungary but higher than Bulgaria or Poland (Figure 29). Domestic political
tensions in Romania have also contributed to the weak performance of Romanian asset prices
as well as the depreciation of the exchange rate.

Figure 29. CDS and EMBIG Developments

CDS and EMBIG Developments

900 1000
800 900
800
700
700
600 600
500 500
400 400
300
300
200
200 100
100 0
0 1/1/2007 1/1/2008 1/1/2009 1/1/2010 1/1/2011 1/1/2012
1/1/2007 1/1/2008 1/1/2009 1/1/2010 1/1/2011 1/1/2012 EMEuropeEMBIG HungaryEMBIG
BulgariaCDS HungaryCDS PolandCDS RomaniaCDS PolandEMBIG RomaniaEMBIG

Source: Bloomberg. Source: Bloomberg.

This note looks at foreign bank deleveraging and examines how Romania’s asset prices
have been impacted from European crisis spillovers and compare those to peer group
 140

countries. Foreign bank deleveraging has been orderly and moderate so far in Romania, also
partly thanks to the EBCI. Findings from the spillover analysis suggests that Romania’s asset
markets tend to co-move more closely with its regional peers but have been strongly
impacted by the financial crisis in 2008/09 and also recently from the intensification of the
euro area crisis. A GARCH analysis shows that implied co-movements of Romanian asset
prices are higher with peer group countries than with the euro area periphery or euro area
asset prices (e.g. Euro Stoxx). But results also indicate that Romania’s asset prices in some
episodes significantly co-move with GIIPS countries and European risk premia with related
correlation jumps up to 0.5-0.6. Furthermore, an ARCH Markov-Switching model analysis
indicates that Romania’s EMBIG spread recently moved back to a high-volatility state which
could have been also driven by domestic political tensions. Equity market volatility has also
soared again recently.

High estimated correlations of Romania’s asset prices and spreads mean that Romania
is vulnerable from an intensification of the euro area crisis. Continuing domestic political
tensions would bring in an idiosyncratic and adverse component into Romania’s asset prices,
a risk on top of the European common factor. Vulnerabilities especially to financial
spillovers from Europe call for safeguarding sufficient public and financial sector buffers and
implementation of prudent contingency planning, given the negative effect sharp increases in
Romania’s CDS and EMBIG spreads or declines in equity prices would have on Romania’s
financing costs and capital inflows, exchange rate, market sentiment as well as credit and
liquidity risk of the banking sector.

This note is organized as follows: Section B discusses recent trends and causes of foreign
bank deleveraging in Romania, while section C covers the methodology and data of the
GARCH and ARCH Markov-Switching analysis as well as the financial spillover results.
Section D concludes.

B. Foreign Bank Deleveraging

The Romanian banking sector remains vulnerable to spillovers from the euro area and
domestic developments, and deleveraging remains a risk. The banking system is
80 percent foreign owned with Austrian banks dominating the market with 38 percent of
system assets. Subsidiaries of Greek banks hold about 14 percent of system assets and
12 percent of deposits. In particular, Greek banks have orderly deleveraged to cope with a
more limited funding availability. While overall bank capitalization remains strong with
14.7 percent, the liquidity situation has become more heterogeneous among banks, and
funding costs (such as in deposits or the interbank segment) are increasing. Credit growth has
significantly slowed and nonperforming loans continued to rise to 16.8 percent in June,
mainly due to the weak economic activity and the vulnerability of the large legacy of foreign-
currency loans. Prudential provisions almost fully cover nonperforming loans but
profitability is poor, mainly because of the persistent need for higher provisioning, lower
interest rate margins and high overhead costs.
 141

Foreign bank deleveraging has been orderly and moderate so far, also partly thanks to
the EBCI initiative. The total exposure to Romania of the nine largest foreign banks that
participated in the EBCI stood at 94 percent (against March 2009 exposure) but still
compares to 101.3 percent at end-2011. While the EBCI exposure to own subsidiaries has
remained at a similar level between March 2009 and June 2012, the banks’ exposure to non-
financial institutions has been steadily declining, overall by 16 percent in the observation
period. Some banks have reduced their overall exposure to below 80 percent. Overall bank
system parent funding has orderly and moderately declined since end-2011, and at end-July
stood at 89.2 percent of the end-2011 level, a decline from €20.3bn to €18.1bn with some
July acceleration (Figure 30). The system loan-to-deposit (LTD) ratio has remained stable
around 120 percent in recent years while due to the funding currency mismatch, the LTV
ratio in foreign currency has stayed beyond 200 percent.

Figure 30. Romanian Banks’ Parent Funding

EBCI Parent Funding Exposure to Romania Bank System Parent Funding (Bn €)
(Bn €) 20.5
33.5
20.0
33
32.5 19.5

32 19.0

31.5 18.5
31 18.0
30.5
17.5
30
17.0

Source: EBCI. Source: NBR.

A large amount of parent funding has a longer-term maturity structure. The fact that the
majority of banks’ parent funding (close to 70 percent) exhibits a maturity of more than one
year prevents an abrupt withdrawal (Figure 31). Around 12 percent has a maturity of up to
one month and 21.1 percent below six months. For the overall banking system, parent
funding constitutes around 20 percent of total assets.

Figure 31. Parent Funding by Maturity

Parent Funding by Maturity

70.0%
60.0%
50.0%
40.0%
30.0%
20.0%
10.0%
0.0%
up to 1 M 1-3 M 3-6 M 6-12 M 1Y - 2Y >2Y

Source: NBR.
 142

Romania’s overall amount of foreign bank funding is above average for the CESEE.
According to recent BIS data for 2012:Q1, BIS-reporting banks exhibited an overall
exposure to Romanian banks and nonbanks of around 28 percent of GDP compared to the
18 percent average for the CESEE (Figure 32). It ranks lowers than peer countries such as
Hungary or Bulgaria but higher than Poland, Serbia or the Czech Republic. A decomposition
of BIS that Romanian banks receive over 60 percent of the foreign bank exposure while for
the CESEE as a whole this share is around 53 percent with 47 percent going to nonbanks.

Figure 32. CESEE Foreign Bank Funding

CESEE: Funding from BIS-Reporting Banks, 2012:Q1
(BIS-reporting banks' exposure relative to GDP, percent)

70
60
To nonbanks To banks
50
40
30
20
10
0 Turkey
Serbia

Ukraine
Russia
Hungary

Montenegro

Slovak Republic
Bulgaria

Romania

CESEE
Moldova
Croatia

Latvia

Albania
Estonia

Poland

Czech Republic

Belarus
Slovenia

Macedonia, FYR
Lithuania

Bosnia and Herzegovina

Sources: BIS; IMF, WEO; and IMF staff calculations.

The decline in overall exposure of BIS reporting banks to Romania has been moderate
compared to some other CESEE countries. According to Table 10, while Romania’s
decline has been in the average for the CESEE (excl. Russia and Turkey) with a 20 percent
deleveraging ($13.4bn) between 2008:Q3 and 2012:Q1, it is much lower than for instance,
seen in Ukraine (52.8 percent), Latvia (38.3 percent) or Hungary (38 percent). Relative to
GDP, the 7.2 percent decline also compares favorable against many CESEE countries. Part of
the exposure reduction can be explained by the re-absorption of loans by subsidiaries that in
the credit boom period had been outsourced to SPVs and parent- related affiliates abroad.

Deleveraging has been driven by different factors. Some causes for the orderly foreign
bank deleveraging in Romania were weak parent banks (especially Greece), changes in
parent funding strategy (e.g. French banks) or some loss in domestic funding (e.g. Greek
subsidiaries). Further deterioration in the financial sector environment, including soaring
NPLs and continued poor profits, could lead some parents to scale back their long-term
support for the subsidiaries, thus making them more reliant on domestic funding.
 143

C. Financial Spillover Analysis

Methodology and Data

This section analytically examines how Romania’s asset prices have been impacted
from European crisis spillovers and compare those to peer group countries. The adopted
modeling framework takes into account the market and idiosyncratic volatility inherent in
asset prices especially at high-frequency data. First, the Dynamic Conditional Correlation
(DCC) GARCH specification by Engle (2002) is adopted, a multivariate GARCH framework
which allows for heteroskedasticity of the data and a time-varying correlation in the
conditional variance (please see annex 2 for details). Secondly, the ARCH Markov-Switching
model (SWARCH) by Hamilton and Susmel (1994) is utilized here because it can
differentiate between different volatility states of Romania’s asset prices and spreads, that is,
low, medium, and high (please see annex 3 for details). Both models are estimated in first
differences to account for the nonstationarity of the variables in the crisis period.

We choose as the sample period daily data from 2007 to (July 13, 2012). Asset prices and
spreads include Romania’s equity market index, interbank, EMBIG and CDS spreads,
together with asset prices in the peer countries Bulgaria, Hungary and Poland as well as
GIIPS and European risk measures.
 Table 10. External Positions of BIS-reporting Banks vis-à-vis CESEE

External Positions of BIS-reporting Banks vis-à-vis CESEE

Stock (US$ billions, exchange-rate Stock
Change (US$ billions, exchange-rate adjusted) Change (Percent, exchange-rate adjusted) Change (Percent of GDP, exchange-rate adjusted)
adjusted) (Percent of
2008:Q3- 2008:Q3- 2008:Q3-
2008:Q3 2011:Q4 2012:Q1 2012:Q1 2011:Q3 2011:Q4 2012:Q1 2011:Q3 2011:Q4 2012:Q1 2011:Q3 2011:Q4 2012:Q1
2012:Q1 2012:Q1 2012:Q1

CESEE 958.8 789.4 781.5 17.7 -33.7 -18.7 -7.9 -177.3 -4.0 -2.3 -1.0 -18.5 -0.8 -0.4 -0.2 -4.0
CESEE excl. Russia & Turkey 584.5 469.3 463.2 27.5 -24.4 -26.0 -6.1 -121.3 -4.7 -5.3 -1.3 -20.8 -1.4 -1.5 -0.4 -7.2
Emerging Europe 829.1 685.1 675.8 16.7 -35.3 -12.0 -9.3 -153.3 -4.8 -1.7 -1.4 -18.5 -0.9 -0.3 -0.2 -3.8
Albania 0.6 1.4 1.5 11.6 0.0 0.1 0.1 0.9 -3.2 6.5 3.8 142.7 -0.3 0.7 0.4 6.8
Belarus 3.0 3.0 2.8 4.8 -0.3 -0.2 -0.2 -0.2 -8.2 -7.6 -6.2 -5.6 -0.5 -0.4 -0.3 -0.3
Bosnia and Herzegovina 4.5 3.9 3.6 20.9 0.0 0.1 -0.3 -0.9 -1.0 1.7 -7.0 -19.0 -0.2 0.4 -1.6 -4.9
Bulgaria 23.1 18.4 17.5 34.4 -1.2 -0.9 -0.9 -5.6 -5.8 -4.5 -5.0 -24.2 -2.2 -1.6 -1.8 -11.0
Croatia 41.8 38.9 38.4 62.2 -2.3 -0.7 -0.5 -3.4 -5.6 -1.8 -1.2 -8.1 -3.7 -1.1 -0.8 -5.5
Hungary 93.6 61.7 58.7 45.1 -5.5 -6.9 -3.0 -34.9 -7.4 -10.1 -4.9 -37.3 -3.9 -4.9 -2.3 -26.9
Latvia 22.0 13.7 13.6 49.5 0.5 -1.2 -0.1 -8.3 3.2 -8.3 -0.9 -38.0 1.7 -4.4 -0.5 -30.4
Lithuania 21.0 14.1 13.4 31.5 0.6 -1.4 -0.7 -7.6 4.1 -9.0 -5.3 -36.3 1.4 -3.3 -1.7 -18.0
Macedonia, FYR 0.8 1.5 1.6 15.9 -0.1 0.3 0.1 0.8 -4.4 29.0 4.5 102.8 -0.5 3.3 0.7 8.0
Moldova 0.6 0.4 0.4 5.1 0.0 0.0 0.0 -0.2 -0.6 14.4 5.7 -36.1 0.0 0.7 0.3 -2.9
Montenegro 1.4 1.4 1.3 31.4 0.0 -0.1 0.0 0.0 -2.7 -8.7 -3.4 -2.9 -0.9 -2.9 -1.1 -0.9
Poland 125.2 120.8 122.5 25.3 -12.1 -4.8 1.7 -2.7 -8.8 -3.9 1.4 -2.2 -2.3 -0.9 0.3 -0.6
Romania 66.9 54.2 53.5 28.7 -2.1 -2.0 -0.7 -13.4 -3.7 -3.5 -1.2 -20.0 -1.1 -1.1 -0.4 -7.2
Russia 216.3 156.6 153.5 8.0 -0.4 8.4 -3.1 -62.8 -0.2 5.6 -2.0 -29.0 0.0 0.5 -0.2 -3.3

144
Serbia 11.0 11.6 10.1 23.4 -0.3 -0.3 -1.5 -0.9 -2.8 -2.9 -12.7 -8.3 -0.8 -0.8 -3.4 -2.1
Turkey 158.1 163.6 164.9 20.2 -8.9 -1.0 1.3 6.8 -5.1 -0.6 0.8 4.3 -1.1 -0.1 0.2 0.8
Ukraine 39.3 19.9 18.6 10.1 -3.1 -1.3 -1.4 -20.8 -12.7 -5.9 -6.9 -52.8 -1.9 -0.8 -0.7 -11.3
Other CESEE economies 129.8 104.4 105.7 28.9 1.6 -6.7 1.4 -24.0 1.5 -6.0 1.3 -18.5 0.4 -1.7 0.4 -6.6
Czech Republic 51.2 46.1 46.3 22.5 1.2 -1.8 0.2 -4.8 2.7 -3.7 0.4 -9.4 0.6 -0.8 0.1 -2.3
Estonia 19.0 11.2 11.2 51.7 -1.8 -0.7 0.0 -7.7 -13.1 -5.7 0.4 -40.8 -8.1 -3.0 0.2 -35.6
Slovak Republic 25.5 20.0 22.6 24.4 3.1 -2.8 2.5 -3.0 15.8 -12.4 12.6 -11.6 3.2 -2.9 2.7 -3.2
Slovenia 34.1 27.0 25.6 56.8 -1.0 -1.4 -1.4 -8.5 -3.3 -5.0 -5.1 -24.9 -1.9 -2.9 -3.1 -18.9

Sources: BIS; WEO; and IMF staff calculations.

 145

Results

DCC GARCH Model:

Findings from the DCC GARCH equity market model suggest that Romania’s implied
equity market co-movement with a GIIPS equity market average and the Euro Stoxx
appears lower than of Poland but higher than Bulgaria. Romania hovers around 0.4–0.5
in terms of the implied correlation with an occasional correlation jump, corresponding to
volatile episodes (Figure 33). A possible caveat is that any low liquidity in e.g. Romania’s
equity market would possibly distort and amplify the results somewhat.

Figure 33. DCC GARCH Equity Market Model

DCC GARCH Equity Market Model

0.9 1
0.8 0.9
0.8
0.7
0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3
0.3
0.2
0.2 0.1
0.1 0
0 1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012 GIIPS-Euro Stoxx Bulgaria-Euro Stoxx
GIIPS-Bulgaria GIIPS-Poland GIIPS-Romania Poland-Euro Stoxx Romania-Euro Stoxx

0.7

0.6

0.5

0.4

0.3

0.2

0.1

0
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012

GIIPS-Romania Bulgaria-Romania
Poland-Romania Romania-Euro Stoxx

Sources: Bloomberg; and IMF staff calculations.

In terms of CDS spread co-movements, Romania shows the highest implied correlation
with Bulgaria followed by Hungary/ Poland and then an average of the GIIPS CDS
spreads. The average implied correlation between Romania and the GIIPS CDS average
stood at around 0.2–0.3 and sporadic volatility jumps up to 0.4 compared to co-movements
of Romania with Bulgaria, Hungary and Poland of around 0.5–0.8 (Figure 34). The CDS
model with Italy confirms the results.
 146

Figure 34. DCC GARCH CDS Model

DCC GARCH CDS Model

1 0.6
0.9
0.8 0.5
0.7 0.4
0.6
0.5 0.3
0.4
0.2
0.3
0.2 0.1
0.1
0 0
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012 1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012

Bulgaria-Romania Hungary-Romania Bulgaria-GIIPS Hungary-GIIPS

Poland-Romania Romania-GIIPS Poland-GIIPS Romania-GIIPS

1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012

Bulgaria-Romania Italy-Romania
Hungary-Romania Poland-Romania

Sources: Bloomberg; and IMF staff calculations.

The EMBIG spread model finds that Romania’s spread moves closer to Hungary’s and
Poland’s EMBIG spreads than the GIIPS 10-year bond yields over Germany’s 10-year
(GIIPS10y) as well as the Emerging Market Europe EMBIG spread. Comparing
Romania, Hungary and Poland against GIIPS10y indicates that Romania’s EMBIG spread
tends to exhibit a lower DCC GARCH implied correlation to the GIIPS10y for the most part
of the sample period (Figure 35). Results do suggest that Romania as Hungary and Poland
have not been immune to volatility in the GIIPS bond spread over Germany with correlation
jumps up to 0.5-0.6. Overall, an intensification of the Euro zone crisis would likely lead to
heightened financial spillovers to Romania with an increase in risk premia (as measured by
CDS and EMBIG spreads) as well as adverse developments on the domestic equity market.
 147

Figure 35. DCC GARCH EMBIG Model

DCC GARCH EEMBIG Model

0.9 0.7
0.8
0.6
0.7
0.6 0.5
0.5 0.4
0.4
0.3 0.3

0.2 0.2
0.1
0.1
0
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012 0
1/2/2007 1/2/2008 1/2/2009 1/2/2010 1/2/2011 1/2/2012
Romania-Hungary Romania-Poland
Romania-GIIPS10y Romania-EME Romania-GIIPS10y Hungary-GIIPS10y Poland-GIIPS10y

Sources: Bloomberg; and IMF staff calculations.

ARCH Markov Switching Model:

Romania’s EMBIG spread has seen the largest shock post-Lehman but there have been
other episodes of sharp increases. The ARCH Markov Switching model mirrors that. In
particular, the EMBIG stood in the high volatility regime post-Lehman, twice in 2010/ 2011,
and moved decisively back to the high state just recently (Figure 36). Domestic political
tensions could have likely contributed to recent volatility.

The sharp decline of Romania’s equity market since 2007 has been only partially
recovered. As expected, volatility in the equity market has been relatively high, and the
Markov Switching model indicates a recent move back towards the medium volatility
regime. Liquidity conditions in the equity market would have influenced the results.

Romania’s 3m interbank rate has successively declined from 16 percent to below 6

percent between 2009 and the summer of 2012. The Markov Switching model shows that
the decline has been fairly volatile with the model oscillating between the high and medium
volatility state (Figure 36). The fragmentation in the interbank markets could potentially
distort the results.

Overall, examining the different volatility states in the Markov Switching model
framework confirm the findings from the DCC GARCH framework, that is, higher
volatility states in the EMBIG spread and equity market would correspond to higher implied
co-movement in the DCC GARCH models.
 148

Figure 36. ARCH Markov Switching Models

ARCH Markov Switching Models

Romania EMBIG Romania Equity Market

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
1/8/2007 1/8/2008 1/8/2009 1/8/2010 1/8/2011 1/8/2012 1/8/2007 1/8/2008 1/8/2009 1/8/2010 1/8/2011 1/8/2012

Low Medium High Low Medium High

Romania 3m Interbank
1

0.8

0.6

0.4

0.2

0
1/10/2007 1/10/2008 1/10/2009 1/10/2010 1/10/2011 1/10/2012

Low Medium High

Sources: Bloomberg; and IMF staff calculations.

D. Conclusion

Vulnerabilities to financial spillovers from Europe to Romania call for safeguarding

sufficient public and financial sector buffers and implementation of prudent
contingency planning given the negative effect that sharp increases in Romania’s CDS and
EMBIG spreads or declines in equity prices will have on Romania’s financing costs and
capital inflows, exchange rate, market sentiment as well as credit and liquidity risk of the
banking sector. According to the DCC GARCH analysis, Romania’s asset markets and
spreads tend to co-move more closely with its regional peers but have been strongly impacted
by the financial crisis in 2008/09 and also recently from the intensification of the euro area
crisis. Results indicate that Romania’s asset prices significantly co-move with the euro area
periphery and European risk premia with related correlation jumps up to 0.5–0.6.
Furthermore, a Markov-Switching model analysis indicates that Romania’s EMBIG spread
recently moved back to a high-volatility state which could have been also driven by domestic
political tensions. Equity market volatility has also soared again recently.

In light of the uncertain environment and spillover risks from the euro area such as an
acceleration of foreign bank deleveraging, it is important that the NBR continues its
intensive bank supervision and further elaborates its crisis preparedness. Any necessary
measures should be taken to ensure that banks have sufficient capital and liquidity especially
from shareholders. With system deposits limited to fully replace any parent bank
 149

deleveraging, the continuing support of parents will be crucial given, in particular, the large
currency mismatch in the banking system. It is equally important that the NBR, in
coordination with other relevant authorities, stands ready to implement its crisis management
framework and updates detailed contingency plans on an ongoing basis.
 150

References

Bollershev, Tim, 1990, “Modelling the Coherence in Short-run Nominal Exchange Rates: a
Multivariate Generalized ARCH Approach,” Review of Economics and Statistics,
Vol. 72, pp.498–505.

EBCI (2012) “CESEE Deleveraging Monitor,” Report prepared for the Steering Committee
Meeting on July 18, 2012, in Warsaw, Poland.

Engle, R. 2002, “Dynamic Conditional Correlation: A Simple Class of Multivariate

Generalized Autoregressive Conditional Heteroskedasticity Models,” Journal of
Business & Economic Statistics, Vol. 20, pp. 339–50.

Frank, N., B. González-Hermosillo, and H. Hesse, 2008, “Transmission of Liquidity Shocks:

Evidence from the 2007 Subprime Crisis,” IMF Working Paper 08/200 (Washington:
International Monetary Fund). See also blog piece on VOX.

Frank, N., and Heiko Hesse, 2008, “Financial Spillovers to Emerging Markets during the
Global Financial Crisis,” IMF Working Paper 09/104

Hesse, Heiko, 2012, “Financial Sector Linkages in Romania,” Romania Article IV Selective
Issues Paper (SIP). (Washington: International Monetary Fund). A shorter version
was also published in EconoMonitor.
 151

Annex 2. DCC GARCH Methodology

We use a multivariate GARCH framework for the estimation, which allows for
heteroskedasticity of the data and a time-varying correlation in the conditional variance.
Specifically, the Dynamic Conditional Correlation (DCC) specification by Engle (2002) is
adopted, which provides a generalization of the Constant Conditional Correlation (CCC)
model by Bollerslev (1990).85 These econometric techniques allow us to analyze the co-
movement of markets by inferring the correlations of the changes in the spreads discussed
above, which in turn is essential in understanding how the financial crisis has impacted
Romania.

The DCC model is estimated in a three-stage procedure. Let rt denote an n x 1 vector of asset
returns, exhibiting a mean of zero and the following time-varying covariance:

(1)

Here, Rt is made up from the time dependent correlations and Dt is defined as a diagonal
matrix comprised of the standard deviations implied by the estimation of univariate GARCH
models, which are computed separately, whereby the ith element is denoted as hit . In other
words in this first stage of the DCC estimation, we fit univariate GARCH models for each of
the five variables in the specification. In the second stage, the intercept parameters are
obtained from the transformed asset returns and finally in the third stage, the coefficients
governing the dynamics of the conditional correlations are estimated. Overall, the DCC
model is characterized by the following set of equations (see Engle, 2002, for details):

(2)

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset
returns rt. As defined above, Rt is the time varying correlation matrix and is a function of Qt,

85
Given the high volatility movements during the recent financial crisis, the assumption of constant conditional
correlation among the variables in the CCC model is not very realistic especially in times of stress where
correlations can rapidly change. Therefore, the DCC model is a better choice since correlations are time-
varying.
 152

which is the covariance matrix. In the matrix Qt,ι is a vector of ones, A and B are square,
symmetric and  is the Hadamard product. Finally, λi is a weight parameter with the
contributions of Dt21 declining over time, while κ i is the parameter associated with the
squared lagged asset returns. The estimation framework is the same as in Frank, Gonzalez-
Hermosillo and Hesse (2008) or Frank and Hesse (2009).
 153

Annex 3. Markov-Regime Switching Analysis

We use Markov-regime switching techniques to examine financial stress in Romania. Given

the intrinsic volatility of high-frequency financial data, especially during periods of stress,
the ARCH Markov-Switching model (SWARCH) by Hamilton and Susmel (1994) is chosen
here because it can differentiate between different volatility states, for example, low,
medium, and high. In particular, univariate SWARCH models are adopted with variables in
first differences to account for the non-stationarity of the variables.

In general, the parameters of the ARCH process can alter. Equation (3) below describes a
Markov chain with yt being a vector of observed variables and s t denoting a unobserved
random variable with values 1, 2, …, K that as a state variable governs the conditional
distribution of yt .

Prob (st  j | st 1  i, st 2  k ,..., yt 1 , y t 2 ,...)  Prob ( s t  j | s t 1  i )  p ij (3)

It is possible to combine all the transition probabilities pij in a K  K transition matrix. In

our SWARCH framework, the mean equation is an AR(1) process and the variance is time-
varying with the ARCH parameters being state dependent. Formally, the AR(1) process
follows

y t    yt 1   t (4)

The time varying variance ht2 with the error term  t is parameterized as

 t  g S  ~t
 t

~
 t  ht   t (5)
 2 ~2 ~2 ~2 ~2
ht  a 0  a1 t 1  a 2  t  2  ...  a q  t  q    d t 1   t 1 ,
where  t ~ N 0,1 , S t  1,2,3 and d t 1 is a dummy variable in which d t 1  1 if ~t 1  0 and
d t 1  0 if ~t 1  0. Hereby, it is assumed that t follows a mean zero process with unit
variance that is independently and identically distributed (i.i.d.). The ARCH parameters are
thus state dependent due to multiplication with the scaling factor g S t which is normalized to
unity for the low volatility regime.86

86
In this paper, an ARCH specification is estimated, as the GARCH(p,q) is not nested within the SWARCH
framework, due to its implicit infinite lag representation.
 154

VIII. PROGRESS WITH BANK RESTRUCTURING AND RESOLUTION IN EUROPE, WITH

NADEGE JASSAUD, 201387

A. Executive Summary

The European Union (EU) banking system restructuring is under way, but is far from
complete. Some bank restructuring has started, and the level Tier 1 capital ratios of EU
banks have been substantially increased88 (thanks to government back-stops and
capitalization exercises run by the European Banking Authority).89 But system-wide, capital
ratios have been met partly by deleveraging or recalibrations of the risk weights on activities.
Consolidation in the banking sector has been slow, with banks rarely closed.90
Nonperforming loans are building up in banks’ balance sheets, and addiction to central bank
liquidity remains high especially for banks in peripheral countries. Despite the EBA
recapitalization exercise having led to €200 billion of new capital or reduction of capital
needs by European banks, fresh capital is difficult to attract in an environment where
prospects for profitability are uncertain.

Several hurdles impair restructuring and resolution in Europe, and urgent progress
needs to be made:

 First, EU bank resolution tools need to be strengthened, aligning them with the
Financial Stability Board Key Attributes for Effective Resolution. Fast adoption of the
EU resolution directive is welcome, but enhancements are warranted. Swift
transposition should follow.

 Second, restructuring of nonperforming loans (NPLs) should be facilitated. The legal

framework should not slow down restructuring and maximize asset recovery. In
several EU countries, such as Italy, Greece and in Eastern Europe, bankruptcy
reforms lag behind in that, for instance, current practice does not allow the seizure of
collateral in a reasonable timeframe. Banks should also manage more actively their
NPLs, possibly allowing a market for distress assets to emerge in Europe.

 Third, further evolution of the General Directorate for Competition’s (DG COMP)
practices will be needed in systemic cases to ensure consistency with a country’s

87
This chapter is based on Jassaud and Hesse (2013).
88
10 percent in June 2012 against 7 percent in December 2008; 57 EU banks (EBA).
89
Measures related to RWAs were only allowed in limited cases, i.e., where new model rollout had been started
before the start of the exercise.
90
While banks were rarely closed, some have downsized by closing branches, selling or closing business lines
and significantly reducing their staff levels in some cases.
 155

macro-financial framework and support viability of weak banks, recovery of market

access, and credit provision. Increased transparency would give added credibility and
accountability.

 Fourth, disclosure should be significantly enhanced and harmonized by the EBA, to

restore market confidence. In particular, interpretable metrics regarding the quality of
banks’ assets, in terms of NPLs, collateral, probability of defaults (PD) and loan
recovery rates (LGD) are key for assessing the strength of banks and restoring
confidence in the banking system.

B. Introduction91

The global and European financial crises revealed long-standing structural weaknesses
that have yet to be fully addressed in individual banks and in banking systems. In large
part, they reflected weaknesses in the public, household, and corporate sectors, but the banks
themselves contributed to the problems, and the financial sector constituted a feedback
channel that reinforced negative tendencies elsewhere.

In this context, the note looks at experience with bank restructuring in Europe in recent
years, what pressures remain to restructure, the impediments that slow the process, and
what policy actions could be helpful. Thus, the discussion includes, but also goes beyond, a
review of government-led resolution of problem banks.

C. Recent Developments

Before the onset of the crisis, relatively favorable conditions—and, in some economies,
asset price and credit bubbles—masked underlying vulnerabilities. Many financial
systems in Europe were bank dominated, complex and very large in proportion to domestic
GDP. Global assets of the five largest banks were typically more than 300 percent of their
home country’s GDP (Figure 37).92 Credit and asset price bubbles (Reinhart and
Rogoff, 2009; Laeven and Valencia, 2008) built up in several jurisdictions, with sharp
increases in leverage for households, also reflected in many countries in a substantial
increase in house prices. While risks were building up, the overall resilience of banks
improved little. From 2000 to 2007, solvency ratios increased by only 0.2 percent.93 Return

91
Prepared by Nadege Jassaud and Heiko Hesse (both MCM). Marc Dobler, Charles Enoch, Daniel Hardy,
Barend Jansen, Marina Moretti, and Constant Verkoren provided helpful comments and guidance, and Ivan
Guerra and Sarah Kwoh provided research support.
92
Total bank assets account for 283 percent of GDP in the EU, compared to about 65 percent of GDP in the
U.S.
93
From 10.7 percent to 10.9 percent (sample of the largest 90 EU banks included in the 2011 EBA stress test),
Bloomberg.
 156

on equity (ROE) was high, about 17 percent in 2007 for European banks. Leverage of many
large financial institutions also increased, reflecting a reliance on short-term wholesale
funding that was not generally considered a concern.

Figure 37. Assets of EU and U.S. Banking Groups

(2011, in percent of GDP)

250 250

200 % of Domestic GDP 200

% of EU GDP
150 150
% of US GDP
100 100

50 50

0 0
HSBC

Citigroup
Bank of America
ING

BBT & Corp.

Goldman Sachs

Bank of NY Mellon

Capital One
RBS
Barclays

Morgan Stanley
JP Morgan Chase
Societe Generale

Wells Fargo
Lloyds Banking Group

Groupe BCPE
BNP Paribas

Deutsche Bank
Credit Agricole Group
Santander

Sources: Total assets’ data from SNL Financial, GDP data from Eurostat, EU Commission.

The crisis trigger was the U.S. mortgage market—to which some European banks were
heavily exposed—but the developments displayed a number of adverse feedback loops,
such that the crisis deepened and spread. As a result, negative spirals between sovereigns,
banks, and the real economy remain strong. Sovereigns, in turn, are in some cases struggling
when they have to backstop weak banks on their own. Absent collective mechanisms to
break these adverse feedback loops, the crisis has spilled across euro area countries.

D. Crisis Response

One element of the response was a massive extension of government aid to banks in the
form mainly of recapitalization, funding guarantees, regulatory forbearance, and easier
monetary conditions. The amounts involved are very large:

 During recent years, EU governments have committed unprecedented support for

backstopping the financial sector with tax payer money. Over the September 2008–
December 2011 period, member states committed a total of nearly €4.5 trillion,
i.e., 37 percent of the EU GDP.94 The amount of tax payer money effectively used

94
Estimated at €4.9 trillion or 39 percent of EU GDP in October 2012.
 157

(mainly via capital injections, State guarantees issued on bank liabilities, etc)
amounted to €1.7 trillion, or 13 percent of EU GDP (Table 11). Out of the 76 top EU
banking groups, 19 currently have a major or even a 100 percent government stake.

Table 11. EU: Public Interventions in the EU Banking Sector: 2008–1195

(In billions of Euros, unless indicated otherwise)

Used Amounts Approved Amounts

% of GDP % of GDP
Capital injections 288 2.4 598 4.9
Guarantees on bank liabilities 1,112 9.1 3,290 26.8
Relief of impaired assets 121 1.0 421 3.4
Liquidity and bank funding support 87 0.7 198 1.6
Total 1,608 13.1 4,506 36.7
Source: EU Commission (2011c), *: only till Dec 2010.

 Liquidity support has been especially large in the euro area and the
United Kingdom. In the euro area, the European Central Bank (ECB) provided
enhanced support by (i) broadening the scope of eligible assets for central bank
funding and setting up full allotment liquidity facilities for banks; (ii) undertaking
refinancing operations at a fixed and historically low rates; and (iii) extending the
maturity of central bank funding to a historical high via the Long-Term Refinancing
Operations (LTROs); and (iv) actively purchasing assets (Figure 38). A program of
Outright Monetary Transactions (OMT) was announced in September 2012 by the
ECB. National central banks have also granted Emergency Liquidity Assistances
(ELA) in crisis situations. In the United Kingdom, the Bank of England set up an
Asset Purchase Facility (APF), for example, to buy high-quality assets, with
cumulative assets purchased net of sales and redemptions totaling £360 billion (as of
September 2012).

95
Those figures do not include the LTRO amounts––including LTRO, the amount of money committed to
banks stand at 23 percent of EU GDP.
 158

Figure 38. EU: ECB Monetary Financing Operations vis à vis Euro Area Banks
(In billions of Euros)

1400
1200 Main Ref inancing Operations (MRO)
1000 Longer Term Ref inancing (LTRO)
800
600
400
200
0

Jul-11
Jan-08

Jan-09

Jan-10

Jan-12
Jul-08

Jul-09

Jul-10

Jul-12
Jan-11
Apr-08

Apr-09

Apr-10

Apr-12
Apr-11

Oct-11
Oct-08

Oct-09

Oct-10

Oct-12
Source: Bloomberg.

Direct government support measures were normally complemented by action to

restructure the affected banks, in part thanks to EU rules on State aid. According to
DG-COMP, 10-15 percent of the EU banking system is now under the State Aid framework
and undergoing some forced restructuring. Based on a restricted sample of 30 EU large
institutions, banks under EU State Aid rules have been (in the process of) deleveraging, with
up to 19 percent of their total assets, according to Morgan Stanley research, while other
banks that did not fall under DG COMP State rules deleveraged much less (Figure 39).

Figure 39. EU: Deleveraging/Restructuring Plans 1/

(In percent of total assets)

40
State Aid Restructuring
35 Non-restructured
30
25
20 State Aid Restructuring Banks' Average: 19

15
10
Non-restructured Banks' Average: 5
5 %
0
BKT
UBI
Natixis
SG

CASA

BBVA
BOI

Pop

UCG
CS

SAB

DB
Barc
Caixa
HSBC
Lloy
CBK

BNP

RBS

UBS

Erste

PMI
BTO
KBC

SAN
BMPS

BCIV
AIB

ISP

Source: Morgan Stanley.

1/ Banks under formal EU State Aid program as of September 2012.
 159

These direct policy actions went in parallel with supervisory actions on banks to
recapitalize (Figure 40).

Figure 40. EU: Tier 1 Ratio of EU Banks 2008–12 1/

12

10

0
2008-12 2009-12 2010-12 2011-12 2012-06
Source: EBA, sample consists of 57 banks and excludes hybrid instruments.
1/ Tier 1 ratio, excluding hybrid instruments, is used as a proxy for Core Tier 1 ratio.

 Led by the EBA, stress testing and recapitalization exercises resulted in banks
increasing the quantity and quality of their capital. After the 2010 Committee of
European Banking Supervisors CEBS and 2011 EBA EU-wide stress tests,96 the EBA
conducted a recapitalization exercise.97 Capital plans submitted by banks have led to
€200 billion of new capital or reduction of capital needs, for an aggregate capital
shortfall of €115 billion, at end–June 2012. Tier 1 ratios98 are now exceeding
10 percent, against 7 percent in December 2008 (Figure 40),

 In some of the countries subject to most stress, the authorities have embraced
independent third party diagnostics (Annex 4), supplementing the EBA-led stress
testing and recapitalization exercises, to regain market confidence in the system.

96
The second EBA stress test (2011) that included 90 banks examined the resilience of the European banks
against a single adverse macroeconomic scenario, using a core Tier 1 (CT1) capital threshold of 5 percent.
97
The EBA recapitalization exercise recommended a higher core Tier 1 capital (CTI) target of 9 percent by end-
June 2012 after establishing a sovereign buffer against banks’ holdings of government securities based on a
market-implied valuation of those holdings as of September 2011.
98
The Tier 1 excluding hybrid instruments so that it gives a proxy of the core Tier 1 ratio in EBA definition.
 160

E. On-Going Challenges

An environment of very low interest rates, quantitative monetary injections, tolerated

forbearance, and government backstops has helped avoid very abrupt restructuring
and an intense credit crunch, but the underlying pressures remain. Accommodative
monetary policies, for example, aim at dealing with acute liquidity stress and giving some
breathing space. But they are not by themselves a solution, and must be combined with
strong macro policies and comprehensive restructuring strategies (including asset diagnosis,
recapitalization and resolution).

While improving, the economic environment in much of the EU remains weak. The
recession in most of the periphery has been spilling into other EU economies (see IMF WEO,
October 2012). Activity in the euro area is expected to contract by 0.2 percent in 2013 (IMF
WEO Update, January 2013). This reflects delays in the transmission of lower sovereign
spreads and improved bank liquidity to private sector borrowing conditions, and still high
uncertainty about the ultimate resolution of the crisis despite recent progress. Credit
conditions are still tight in some EU countries especially those in the periphery and the
Emerging Economies in the EU (EEE), which threatens the economic recovery.

Recent developments in financial markets have been favorable, although the perceived
risks to financial stability remain elevated. Significant new efforts by European
policymakers—in particular the launching of the OMT program by the ECB in August 2012–
–have somewhat allayed investors’ fears. Tail-risk perceptions have fueled a retrenchment of
private financial exposures to the euro area periphery (see IMF GFSR, October 2012).

NPLs in EU banks continue to rise, outpacing loan growth (Figure 41). Since 2007, loans
to the economy have decreased by 3 percent while NPLs99 increased by almost 150 percent,
i.e., €308 billion in absolute terms. And, this trend shows no sign of reversal, reflecting the
continued macro deterioration in parts of the EU and the absence of restructuring. When
NPLs remain on balance sheets, they absorb management capacity, and continued losses can
weaken banks’ profitability. They can also foster forbearance, thereby deterring new
investors by impairing transparency. In several countries, independent asset quality reviews
and stress tests have facilitated a diagnosis of the quality of banks’ assets, supporting
prospects for private recapitalization.100

99
NPLs have jumped from 2.6 percent in December 2007 to 8.4 percent of total loans in June 2012.
100
Countries under/ near financial assistance (Cyprus, Greece, Ireland, Portugal, and Spain) have all carried out
independent asset quality reviews to regain market confidence in the system. Similarly, Slovenia has carried out
an independent assessment for the three largest banks.
 161

Figure 41. EU Banks NPLs to Total Loans

(EBA 90 SIFI Bank Sample)

% of RWA 3

0
Sep-07
Dec-07

Sep-08
Dec-08

Sep-09
Dec-09

Sep-10
Dec-10

Sep-11
Dec-11

Sep-12
Mar-08

Mar-09

Mar-10

Mar-11

Mar-12
Jun-08

Jun-09

Jun-10

Jun-11

Jun-12
Source: Bloomberg.

NPLs across EU banks differ largely, with those in the “peripheral” countries (Greece,
Ireland, Italy, Portugal and Spain) witnessing the largest increases. For instance, from
December 2007 to June 2012, the NPL ratio for Italy increased by 2.5 times, while in Spain,
the increase was seven times (Figure 42). Ireland stands out with average NPLs of around
30 percent, followed by Hungary and Greece. However, definitions in this area remain non-
harmonized and impair comparability across the EU.101

Figure 42. EU: NPLs to Total Loans

(June 2012 vs. December 2007)

35
30
25
20
15
10
5
0

Average of Jun-12 Average of Dec-07

Sources: Bloomberg, EBA 90 SIFI Banks, September 2012.

101
Across European countries, there can be large differences in NPL definitions, making asset quality
assessment across countries and banks difficult.
 162

Capital ratios have increased, but concerns have been expressed about the consistency
of the Basel risk weights across firms. During the last EBA recapitalization exercise,
30 percent of the shortfall that banks were required to make up was met through reduction in
RWAs, of which €10 billion came through RWA “recalibrations” (validation, roll out or
changes to parameters of internal models). Such recalibrations of RWAs are expected to
continue, contributing to opacity in bank capital computations. The recent Bank of England
Financial Stability Report (November 2012) showed that banks’ RWAs calculations for the
same hypothetical portfolio can be vastly different, with the most prudent banks calculating
over twice the needed capital as the most aggressive banks.

Funding remains a large challenge, especially for banks in the peripheral countries.
Many such banks are heavily reliant on ECB funding with challenges on asset encumbrance
and collateral eligibility due to, for instance, rating downgrades, valuation effects on their
collateral and overall loss of market confidence. Banks in Greece and Ireland have also
substantially used ELA. Following the announcement of the OMT program by the ECB,
funding conditions have somewhat eased for peripheral banks, and some have been able to
issue debt in primary markets; peripheral bank CDS spreads have been easing. But wholesale
funding remains prohibitively expensive for the euro area periphery banks to sustainably
support lending in the current environment.

F. Resolution and Restructuring Framework

To deal with these challenges, the EU needs to enhance the framework for bank
resolution and restructuring. Issues arise relating to bank resolution—on a “gone” or on a
“going” concern basis; rules on State aid; measures to facilitate private sector, market-based
adjustment; and other areas.

The Single Supervisory Mechanism (SSM) will only be one step towards an effective
banking union (BU) as resolution, a deposit guarantee scheme (DGS), and a single
rulebook are essential counterparts. Resolution and a DGS will need to be centralized,
with a common backstop. Meanwhile, as a key element in addressing the crisis, the European
Stability Mechanism (ESM) is being prepared to directly recapitalize banks as well as
providing fiscal support. The European Council decision of June 2012 provided the ESM the
possibility of direct bank recapitalization when an effective SSM is in place (see IMF EU
FSAP FSSA and Goyal et al, 2013).
 163

G. Resolution Framework for Problem Banks

Across the national FSAPs, countries lacked domestic resolution tools.102 In reaction to
the crisis, the United Kingdom created a special resolution regime (SRR) and Germany
adopted a restructuring law, both of which granted the authorities the power to utilize various
resolution tools. Now both countries can sell failing businesses, i.e., to transfer all or part of
the business to a private sector purchaser, and or to create a bridge bank. The German Bank
Reorganization Act (January 2011) also provides for an asset separation tool (the power to
transfer all or part of a business to an entity, even if not a bank, in which the restructuring
fund owns shares) and the possibility to bail-in senior unsecured creditors through a court-led
proceeding on the initiative of the bank.

A critical new EU resolution directive is in preparation. As a national approach to

resolution may well not be appropriate in the EU given the importance of cross-border
banking, and the failure of existing cross-country coordination mechanisms, the European
Commission (EC) has taken steps to harmonize and strengthen domestic resolution regimes.
This should help avoid regulatory arbitrage and make orderly resolution effective and
efficient for cross-border banks. In June 2012, the Commission issued a draft directive for
harmonized crisis management and resolution framework in all EU countries. The Irish
Presidency will make the adoption of the resolution framework a top priority and plans to
adopt it during the first part of 2013. The new national resolution regimes endow EU
countries with strong early intervention powers and resolution tools. The transposition of the
directive into national laws should be accelerated relative to the current deadlines (01/2015,
and 01/2018 for bail-ins).

While the proposed directive will mark a big step forward, further enhancements are
needed (box 1). EU countries need to be endowed with strong early intervention powers. The
FSB has developed new international standards for resolution (Key Attributes) that were
endorsed by the G-20 leaders in 2011. They specify essential features that should be part of
the resolution framework at both the national and international levels for Global Systemic
financial institutions (G-SIFIS). The key objective is to make resolution feasible without
severe systemic disruption and without exposing taxpayers to loss.103

102
FSAP safety nets on Netherlands, Germany, U.K., Spain, and Luxembourg.
103
In recognition of the impending legislative proposals the EBA has been active in developing methods for the
recovery and resolution of failing banks, such as in its efforts for recovery plans, such as developing templates.
 164

Box 1. Proposed Resolution Directive––Risks and Areas for Enhancements1

 Resolution of banks is undermined by the absence of a more effective EU-wide framework

to fund resolution. Binding mediation powers for the EBA and mutual borrowing
arrangements between national funds face inherent constraints (in particular, the EBA
cannot impinge on the fiscal responsibilities of EU member states).

 Passage of the directive will substantially enhance the range of tools available to
resolution agencies in the EU. But the scope of the directive should be widened to include
systemic insurance companies and financial market infrastructures. The European
Commission launched a consultation at the end of 2012 on this issue. All banks should be
subject to the regime, without the possibility of ordinary corporate insolvency
proceedings.

 The breadth and timing of the triggers for resolution should be enhanced by providing the
authority with sufficient flexibility to determine the non-viability of the financial
institution (including breaches of liquidity requirements and other serious regulatory
failings, not just capital/asset shortfalls). There should be provision for mandatory
intervention in the event a specified solvency trigger is crossed.

 The directive affords less flexibility for using certain resolution powers than the key
attributes. For instance, it does not permit exercising the mandatory recapitalization power
and the asset separation tool on a standalone basis. Also, bail-in safeguards should not
prevent departure from pari passu treatment where necessary on grounds of financial
stability or to maximize value for creditors as a whole.

 Depositor preference should be established for insured depositors,2 with the right of
subrogation for the DGS.
______________________
1
Box prepared by Marc Dobler.
2
Staff recommendation related to depositor preference is not drawn from the Key Attributes best practices.

It is desirable to move quickly beyond the harmonized national regimes, and set up a
single resolution mechanism (SRM), ideally with common backstops and safety nets, at
least for the countries participating in the SSM. Just as banks are nowadays too
interconnected to be effectively supervised at a national level, so national resolution regimes
would have difficulty, even under harmonized arrangements, in handling the bigger banks of
the EU. Moreover, there would be limited incentives among national resolution authorities
for least-cost and rapid action to address problems; also, coordination difficulties, especially
for large cross-border banks, in the absence of common backstops, may undermine
 165

effectiveness. To be fully aligned with best practices,104 the resolution authority should seek
to achieve least cost resolution of financial institutions without disrupting financial stability.
It should protect insured depositors, and ensure that shareholders and unsecured, uninsured
creditors absorb losses. The SRM will need a mandate, alongside the SSM, to develop
resolution and recovery plans and intervene before insolvency using well-defined
quantitative and qualitative triggers. It will need strong powers and a range of tools to take
early intervention measures and restructure banks’ assets and liabilities (for example, bail-in
subordinated and senior unsecured creditors, transfer assets and liabilities with “purchase and
assumption,” and separate bad assets by setting up asset management vehicles), override
shareholder rights, establish bridge banks to maintain essential financial services, and close
insolvent banks.

The SRM will need to coordinate closely with the SSM. For instance, there could be
regular formal meetings with the Chair of the supervisory Board of the ECB. Alternatively,
the ECB Chair of the supervisory Board could serve on the board of the SRM, together with
national representatives and representatives of other EU bodies.

Resolution will likely be subject to state aid rules, so the SRM will have to coordinate
closely with DG COMP. Any of the existing agencies would likely have to undergo
operational and possibly legal changes in order to carry out a resolution role; for the time
being it may be worthwhile to use the ESM as the resolution mechanism, but in the medium
term it may be best that the single resolution agency be created from new. This agency can
begin operating once agreement on common resolution funding and backstops are in place.

Borrower restructuring needs to be facilitated, with legal hurdles lifted. The legal
framework should facilitate the restructuring of NPLs and maximize asset recovery. In
several EU countries, including Italy, Greece and Portugal, the IMF is involved in
bankruptcy/insolvency law reform, including by introducing fast track restructuring tools and
out-of-court restructuring process. For instance, repossession of the collateral backing a retail
mortgage may take several years in Italy versus few months in Scandinavia and United
Kingdom. The asset recovery process is also very prolonged in many EEE countries.105
Sometimes in those jurisdictions, the issue is implementation, with banks being unable to
enforce collateral. This can weigh heavily on the value of the bank, making its collateral
worth less and leaving NPLs on their balance sheets. An efficient framework for handling
NPLs is key to rehabilitate viable borrowers and provide the exit of non-viable borrowers.

104
The FSB Key Attributes of Effective Resolution Regimes for Financial Institutions.
105
The European Banking Coordination “Vienna” Initiative (2012) in a working group focused on NPL issues
in Central, Eastern and Southeastern Europe. Recommendations, among others, focused on establishing a
conducive legal framework for NPL resolution, removing tax impediments and regulatory obstacles, as well as
enabling out-of-court settlements.
 166

Active management of NPLs is needed. In principle, NPLs can either be: (i) retained and
managed by banks themselves at appropriately written-down values, while the banks receive
financial assistance from the government for recapitalization; or (ii) relocated or sold to one
or more decentralized “bad banks,” loan recovery companies, or Asset Management
Companies (AMCs) that specialize in the management of impaired assets; (iii) sold to a
centralized AMF set up for public policy purposes (possibly when the size of NPLs reaches
systemic proportions, see Annex 5).

The EU experience with AMCs is at an early stage, although they have been used widely
in many other parts of the world. A number of AMCs were established, including in
Belgium, Denmark, Ireland, Spain, Switzerland, and the United KingdomDiscussions on
possible AMCs are underway in Cyprus and Slovenia; and AMCs were considered but ruled
out in Iceland. While it is early to fully assess the recent experience, it is useful to compare
and contrast features and approaches with AMCs in other countries both past and present;
discuss the rationale behind any deviation from established practice; and draw where possible
some preliminary conclusions.

Government support and State Aid rules for financial sector action

Competition and State Aid policy has served de facto as the main coordinating
mechanism in bank restructuring during the crisis, as the only binding EU framework
available for this purpose.106 DG COMP has the exclusive mandate and power to ensure
that State aid is compatible with the treaty, and that State aid provision is accepted in
exchange for strict conditionality. Member states have provided aid through capital
injections, guarantees and asset purchases. Compensatory measures required by DG COMP
have included divestments, penalty interest rates, management removals, dividend
suspensions and burden sharing (shareholder dilutions, and bail in of subordinated debt).
According to DG Comp, 60 EU banks—accounting for 10–15 percent of the EU banking
assets—underwent a deep restructuring. Under the State aid regime, 20 banks were resolved.

Interventions by DG COMP have been instrumental in imposing restructuring on

banks but have on occasion heightened macro-financial concerns. In particular, there
have been concerns about the speed of decision making and insufficient transparency, and the
impact of compensatory measures on financial stability and economic growth. State aid
decisions have involved relatively long timeframes, and rules not well understood by markets
have at times exacerbated uncertainties. Since DG COMP could only act in response to
national State aid proposals, decisions were taken case-by-case on an individual basis even in

106
The TFEU contains strict limitations on State aid to avoid distorting competition and the internal market.
According to the Article 107 of the treaty, no State aid should be granted in any form which distorts or threatens
competition. However, State aid can be exceptionally allowed under paragraph 3 of Article 107 in cases of
serious disturbances to the economy.
 167

the presence of system-wide problems. The case-by-case approach has led on occasion to
concerns about excessive private sector deleveraging and undesirable macro-financial
outcomes.

State aid management is evolving to respond more flexibly to the crisis, but faces
fundamental challenges. DG COMP is assigned a difficult task in mitigating competitive
distortions, yet preserving financial stability, and limiting the costs to the taxpayers while
ensuring the long term viability of the institutions that receive State aid. The design of
intervention strategies, therefore, sometimes involves significant trade-offs. Procedures have
been accelerated, and sector-wide implications have been taken into account. The ongoing
Spanish arrangement, for example, takes a broader approach. The Commission’s powers
regarding the resolution of banks have been strengthened further, since ESM support to bank
recapitalization is now conditional upon the Commission's approval of those banks'
restructuring plans. The new mechanism has given DG COMP greater influence in the
restructuring and resolution of banks receiving State aid, and led to a significant acceleration
in the approval process. For instance, it took less than six months to approve the restructuring
plans of eight Spanish banks, consistent with the timelines of the European program of
assistance to Spain. Stronger coordination with other institutions is desirable with a view to
achieving the Commission’s objective of “restoring financial stability, ensuring lending to
the real economy, and dealing with systemic risk of possible insolvency.”

DG COMP’s practices in systemic cases can be further enhanced to ensure consistency

with a country’s macro-financial framework and transparency should be enhanced.
Phasing and composition of bank restructuring is critical to mitigate adverse macroeconomic
effects. DG COMP seeks to set the right incentives to make the best use of State aid and
withdraw from state protection as soon as possible. A pricing policy has been established
based on recommendations of the ECB that seeks to limit moral hazard by ensuring a
sufficient degree of burden sharing, although at a level which is still below the remuneration
that would, in the absence of State aid, be requested by the market. However, increased
transparency in pricing and proposed deleveraging would give added credibility to DG
COMP’s efforts, which sometimes appear to be ad hoc. An examination, for instance with
the IMF and ECB, of its policy for determining the remuneration of instruments used for
capital support would be appropriate, to ensure on the one hand that it is not double-hitting a
fragile institution and on the other not simply delaying the institution’s demise, and thereby
undermining financial stability going forward. Similarly, it would be helpful to look again at
the methodology for determining the required degree of bank deleveraging.

DG COMP’s role will change as a dedicated resolution framework for the BU is

developed. The challenge will be to find a balance to foster a more integrated approach
between the Commission as the guardian of competition and institutions that, concomitant
with the BU, will be charged with overseeing bank resolution and safeguarding financial
stability at the EU level. One option would be to foster a permanent coordination mechanism
between DG COMP and financial stability authorities to deal efficiently with the competition
 168

and State aid aspects of future resolution cases. Moreover, as most large EA banks have
presence outside the likely BU perimeter, there is likely to be an important role in
coordinating between the BU resolution authority and those in the remaining EU member
states using the framework of the prospective resolution directive.

H. Disclosure

Publication of EBA stress test results allowed for enhanced transparency, but
remaining data gaps impede market discipline. EBA stress tests allowed for enhanced
transparency with over 3,000 data points disclosed by EU banks. However, consistent public
data across banks are missing on many fronts, including the funding side (collateral
encumbrance, ECB funding, LCR ratios), derivatives portfolio and other off-balance sheet
activities, RWAs, PDs.

NPL definitions are not harmonized across Europe. There can be large differences in NPL
definitions, making asset quality assessment across countries and banks difficult. In
December 2012, the ESMA stressed the need for transparency and the importance of
appropriate and consistent application of impairments (Treatment of Forbearance practices in
IFRS Financial Statements of Financial institutions). While it recognized a certain degree of
judgment in the classification, it suggested some examples of trigger events. Similarly,
practices in terms of write-offs under IFRS are relatively flexible, making comparison across
banks very difficult.

Disclosure of collateral is not mandatory. IFRS does not require disclosing the amount of
collateral, and therefore, when banks disclose a value, there is no consistency. The practices
differ in using Fair value, nominal value, nominal realizable value (capped to the 'gross'
value of the loan) or stressed value. The periodicity (how often data is revalued) and what is
the governance of that process also varies across banks.

The EBA must continue to promote better dissemination of supervisory micro-data

across the EU and to enhance transparency in the disclosure of banks’ risk–related
data. The 2011 stress test exercised showed the value brought by disclosure of detailed
information. As quality assurance is key, the EBA should strive to (i) enhance the quality
assurance process; (ii) promote the disclosure of granular asset quality information; and
(iii) expand depth, and coverage of audits. In addition, the EBA should raise the awareness of
supervisors on asset quality issues, in particular by issuing guidelines for supervisors on best
practices for the conduction of asset quality reviews, addressing some specific sectors, and
urgently pushing for enhancing comparability and completeness of Pillar 3 reports. The EBA
should work with national authorities and coordinate the provision of technical expertise
where needed (cf. TN on EBA).

The EBA should also enhance its work on supervisory convergence. Current work on the
consistency of RWAs should be a priority. Initial work in this area identified divergences in
the application of Internal Ratings Based (IRB) models, differences of
 169

interpretation/implementation of the regulatory framework, and dispersion across banks in

the gap between expected losses on defaulted and non-defaulted assets. This work is of great
relevance for supervisory convergence and the level playing field in the single market. It
should be kept in harmony with Basel Committee on Banking Supervision (BCBS) Level 3
exercises, and followed up with the issuance of guidelines (and perhaps Regulatory
Technical Standards) to ensure consistency.107

107
See also the technical note on the EBA.
 170

Annex 4. Experience with Asset Quality Reviews

Independent Asset Quality Reviews have been conducted in most of the distressed EU
countries. Countries under/near financial assistance (Cyprus, Greece, Ireland, Portugal, and
Spain) have carried out independent Asset Quality Reviews to regain market confidence.108
Self assessments are usually difficult in a crisis environment because supervisors may be
under political pressures to hide losses (Table A4.1).

108
Slovenia has almost conducted an independent assessment.

Ireland
Jan–Mar 2011
In December 2010, as part of
the EU/IMF program, BlackRock
Solutions was engaged to
perform a loan diagnosis of over
€275 billion across the five
largest Irish banks.
The diagnosis had five building
blocks:
 an asset quality review to
assess the quality of
aggregate and individual loan
portfolios and the monitoring
processes employed;
 a distressed credit
operations review to assess
the operational capability and
effectiveness of distressed
Table A4.1. EU: Asset Quality Reviews Conducted in EU Countries: 2008–12
Greece
Aug–Dec 2011
nd
As part of the 2 Memorandum of
Economic and Financial Policies,
BlackRock was engaged to
perform a loan diagnosis over all
Greek banks.
Individual results were
communicated to banks but no
disclosure has been made to the
public.
Portugal
Jul–Nov 2011
Under the EU/IMF program, the
supervisor led detailed asset
quality reviews of the eight
largest national banking groups’
loan portfolios and regulatory
capital (RWA) calculations.
Those eight largest banking
groups account for more than
80 percent of the banking
system’s total assets.
This “Special Inspection
Program” (SIP) was carried out
with support from external
parties, Ernst & Young, PWC and
Oliver Wyman.
Cyprus
Sept–Dec 2012
An asset quality review of the
Cypriot banks will be conducted,
including a stress test exercise.
The Central Bank of Cyprus
appointed the investment
companies Pimco and Deloitte to
conduct the asset quality review
of on 22 institutions, which is a
mix of EU subsidiaries, co-
operative credit institutions, and
domestic banks.
The participating banks account
for 73 percent of the Cyprus
banking system.
The stress test will have a three-
Spain
May–Jun 2012
Olivier and Wyman and Roland
Berger were assigned to assess
the resilience of the main
Spanish banking groups
(14 which hold 88 percent of the
market asset share).
Cumulative credit losses for the
top-down stress test with a three-
year horizon are €250-270 billion
in the adverse scenario and
€170-190 billion in the base
scenario.
The estimated capital needs
range from €51-62 billion and
€16-25 billion in the adverse and

loan portfolio management in
the banks including arrears
management and workout
practices in curing NPLs and
reducing loan losses;
Ireland
Jan–Mar 2011
 a data integrity validation
exercise to assess the
reliability of banks' data;
 a loan loss forecast (LLF)
under base and stress
scenarios; and
 a public communication.
Under the Loan Loss Forecast,
Blackrock estimated future
losses with forecasted financial
statements through end-2013
(three- year horizon) as well as
baseline losses.
The SIP had three different work
streams (WS):
 the valuation of the credit
portfolio,
Greece Portugal
Aug–Dec 2011 Jul–Nov 2011
 a credit risk capital
requirements calculation, and
 a stress test conducted (by
Olivier and Wyman).
The results of the W1 and W2
were made public in
December 2011. The results of
the W3 were not disclosed.
171
year horizon from mid-2012 to base scenario, respectively, and
mid-2015. the capital buffer requirement of
€37 billion for a core Tier 1
threshold of 7 percent.
Cyprus Spain
Sept–Dec 2012 May–Jun 2012
The second part of the
assessment with four domestic
auditors was completed at the
end of September.
 172

Annex 5. Experience with Asset Management Companies in Crisis Countries

In past crisis, AMCs have been extensively used as a way of facilitating bank
restructuring (Sweden, Indonesia, Malaysia, Korea, and Thailand). While there is no
single optimal solution, operational independence, appropriately structured incentives and
commercial orientation are key design features.

In the current EU crisis, a number of AMCs were established, including in Ireland,

Spain, Belgium, Denmark, and the U.K; discussions on possible AMCs are underway in
Cyprus and Slovenia.

Table A5.1. EU: AMCs––Challenges and Key Design Features

Costs and Benefits
AMC allow consolidation of scarce
workout skills and resources in one
agency, and the application of uniform
workout procedures:
 help securitization because of the
larger pool of assets;
 provide greater leverage over debtors
(especially if AMCs are granted
special powers of loan recovery);
 prevent fire sales or destabilizing
spillover effects, as banks deleverage;
and
 allow the good banks to focus on their
core business.
However, asset purchases by an AMC do
not raise banks’ net worth unless the
operation is done at above-market prices,
which should be avoided. Asset
purchases, thus, do not solve a problem
of lack of capital in the banking sector.
The overall cost may be higher than
expected, depending on the legal and
operational environment for loan recovery
and the likelihood of being subject to
political pressure.
Key Design Features
 Governance: operational
independence is necessary to assure
the effective operation of an AMC.
 Structured incentives: the AMC
should not become a “warehouse” of
NPLs and have incentives to ensure
effective and efficient asset
management and asset disposals.
 Commercial orientation: assets
should be purchased at a price as
close to a fair market value as
possible to minimize losses (possibly
considering some form of profit-
109
sharing arrangement).
Funding shall be adequate. the AMC
must have sufficient funds to perform its
intended functions, with the operating
budget separate from funding for asset
takeover. In past crises, funding came
from either the proceeds of government
bond issues or the AMC‘s own bond
issuance backed by the government.
A key advantage of using a company
without a banking license (an AMC)
instead of a―bad bank is that AMCs do
not need to meet regulatory capital and
liquidity requirements, thereby reducing
their overall costs.
EU Crisis Countries
Ireland: the National Asset Management
Agency (NAMA) was set up in
December 2009, to help Irish banks
divest of bad loans (Irish commercial
property) and in turn receive government-
backed securities as collateral against
ECB funding. NAMA aimed to achieve
this task by:
 Acquiring bad loans from the five
participating banks,
 Working pro-actively on a business
plan for acquiring and disposing bad
loans, and
 Protecting and enhancing to the
maximum possible level, value of
these assets.
Spain: the legislation enacted in
August 2012 established the Asset
Management Company for assets arising
from bank restructuring (Sareb ) and
empowers the Fund for the Orderly
Restructuring of the Banking Sector
(FROB) to instruct distressed banks to
transfer problematic assets to it.
Mid–December 2012, Sareb increased its
capital to allow its main private
participants (banks) to become
shareholders.

Sources: Ingves, Stefan and David S. Hoelscher (2005),”The Resolution of Systemic Banking System Crises,” Enoch, Charles, Gillian
Garcia and V. Sundarajan, (2001) “Recapitalizing Banks with Public Funds,” IMF Staff Paper Vol. 48, No.1, Bank of Ireland, FROB
Websites.

109
The Malaysian Danaharta, for example, purchased impaired loans at an average discount of 55 percent, while
banks that sold assets retained the right to receive 80 percent of any recoveries in excess of acquisition costs
that the AMC was able to realize.
 173

References

Bank for International Settlement (2013) “Summary Description of the LCR,” Basel
Committee on Banking Supervision, January 6, 2013, in Basel, Switzerland.

Bank of England (2012), Financial Stability Report, (November 2012).

CEPS (2010), Bank State Aid in the financial crisis, fragmentation or level playing field,
October 2010

Claessens, Stijn, Ceyla Pazarbasioglu, Luc Laeven, Marc Dobler, Fabian Valencia, Oana
Nedelescu, and Katharine Seal, (2011) “Crisis Management and Resolution: Early
Lessons from the Financial Crisis,” IMF Staff Discussion Note No.5.

Cuñat and Garicano (2009), “Did Good Cajas Extend Bad Loans? The Role of Governance
and Human Capital in Cajas’ Portfolio Decisions,” FEDEA monograph.

DG Competition (2011), “The effects of temporary State aid rules adopted in the context of
the financial and economic crisis,” Autumn 2011.

Enoch, Charles, Gillian Garcia and V. Sundarajan (2001) “Recapitalizing Banks with Public
Funds,” IMF Staff Paper Vol. 48, No.1.

European Banking Authority, 2012, Results of the Basel III monitoring exercise based on
data as of 31 December 2011, September 2012.

European Banking Coordination “Vienna” Initiative (2012) “Working Group on NPLs in

Central, Eastern and Southeastern Europe,” March 2012.

EU Parliament report (2011) State aid, crisis rules for the financial sector and the real
economy.

Goyal, Rishi, Petya Koeva Brooks, Mahmood Pradhan, Thierry Tressel, Giovanni
Dell'Ariccia, Ross Leckow, Ceyla Pazarbasioglu, and an IMF Staff Team (2013) “A
Banking Union for the Euro Area,” IMF Staff Discussion Note 13/01.

Haldane (2011) Capital Discipline, January 2011.

Hoelscher, David S. and Marc Quintyn (2003) “Managing systemic crisis,” IMF Occasional
paper No. 224.

Ingves, Stefan and David S. Hoelscher (2005), ”The Resolution of Systemic Banking System
Crises,” in Systemic Financial Crises: Resolving Large Bank Insolvencies, edited by
Douglas Darrell Evanoff, George G. Kaufman, World Scientific Publishing.
 174

International Monetary Fund (2010) Crisis Management and Resolution for a European
Banking System.

———(2011) European Financial Stability Framework Exercise (EFFE).

———(2012), Global Financial Stability Report, April 2012.

———(2012), Global Financial Stability Report, October 2012.

———(2012), World Economic Outlook, October 2012.

———(2013), World Economic Outlook Update, January 2013.

———(2013), “European FSAP: Technical Note on Stress Testing of Banks,” IMF,

(Washington, February).

———(2013), “European FSAP: Technical Note on Financial integration and de-integration

in the EU,” IMF, (Washington, February).

———(2013), “European FSAP: Technical Note on European Banking Authority,” IMF,

(Washington, February).

———(2013), “European FSAP: Financial System Stability Assessment (FSSA),” IMF,

(Washington, February).

Jassaud, Nadege, and Heiko Hesse, 2013, Progress with Bank Restructuring and Resolution
in Europe, IMF European FSAP, Technical Note (Washington: International
Monetary Fund). A shorter version was published in VOX.

Financial Stablity Board (2011), Key Attributes of Effective Resolution Regimes for
Financial Institutions.

JP Morgan (2012), “Deleveraging Versus Growth Financials Sector Outlook 2013,”

November 2012.

Laeven, Luc and Fabián Valencia (2012), “Systemic Banking Crises Database: An Update,”
IMF Working Paper No. 163.
 175

IX. NEXT GENERATION SYSTEM-WIDE LIQUIDITY STRESS TESTING, WITH CHRISTIAN

SCHMIEDER, BENJAMIN NEUENDORFER, CLAUS PUHR AND STEFAN SCHMITZ, 2012110

This paper presents an Excel based framework to run system-wide, balance sheet
data based liquidity stress tests and is part of a new generation stress testing
framework initiated by Schmieder, Puhr, and Hasan (2011). The liquidity framework
includes three elements: (a) a module to simulate the impact of bank-run type
scenarios; (b) a module to assess risks arising from maturity transformation and
rollover risks, implemented either in a simplified manner or as a fully-fledged cash
flow based approach; and (c) a framework to link liquidity and solvency risks; The
framework also allows simulating how banks cope with upcoming regulatory changes
(Basel III) and accounts for the fact that data availability differs widely. A case study
shows the impact of a “Lehman” type event for stylized banks.

A. Introduction

Traditionally speaking, liquidity risk has played an important role, one that was at least as
important as solvency. Only with the introduction of Basel I in 1988, when liquidity risk did
not make it into the framework, solvency gained more importance and the erosion of
capitalization came to a halt (Goodhart, 2008). However, the global financial crisis has
clearly shown that neglecting liquidity risk comes at a substantial price. There is widespread
consensus that banks’ pre-crisis extensive reliance on deep and broad unsecured money
markets is to be avoided in the future (and in current market conditions there is no appetite
for that anyway). Creating substantial liquidity buffers across the board is the explicit aim of
a number of regulatory responses to the crisis, such as the CEBS Guidelines on liquidity
buffers (CEBS 2009b) as well as the forthcoming Basel III liquidity standards, the Liquidity
Coverage Ratio (LCR) and the Net Stable Funding Ratio (NSFR).

Liquidity risks are being triggered by maturity transformation, i.e., usual long-term maturity
profiles of banks on the assets side and short-term maturities on the liabilities side.111 Banks
have commonly relied on retail deposits, and, to some degree, long-term wholesale funding
as supposedly stable sources of funding. Yet, over the last decade large banks have become
increasingly reliant on short-term wholesale funding (especially in interbanking markets) to
finance their rapid asset growth. While at the same time, supply of funding from non-deposit
sources (such as commercial paper placed with money market mutual funds) also soared.
With the unfolding of the global financial crisis, when uncertainties about the solvency of

110
This chapter is based on Schmieder, Hesse, Neuendorfer, Puhr, and Schmitz (2012).
111
Annex 7 provides an overview of the typical distribution of banks’ assets and liabilities.
 176

certain banks emerged, various types of wholesale funding market segments froze, resulting
in funding liquidity challenges for many banks.112
The liquidity stress testing framework presented herein was developed in the context of
recent Financial Sector Assessment Programs (FSAPs)113 and IMF technical assistance
especially in Eastern Europe, extending the seminal work of Čihák (2007), and drawing upon
work at the Austrian National Bank (OeNB). While developing the framework, five key facts
were accounted for: (i) the availability of data varies widely; (ii) liquidity risk has several
dimensions and assessing banks’ resilience vis-à-vis funding risks requires multi-dimensional
analysis; (iii) designing and calibrating scenarios is more challenging than for solvency risks,
mainly as liquidity crises are relatively rare and originate from different sources; (iv) there is
a close link between solvency and liquidity risks; and (v) while the paper and tool present
some economic benchmark scenarios, but these scenarios and economic and behavioral
assumptions used for the tests should depend on bank- and country-specific circumstances,
and current circumstances (i.e., the level of stress), among others. More generally speaking,
the presented liquidity stress testing framework herein does not substitute for sound
economics in designing the tests.

The answer to these multiple dimensions is a framework that is an Excel based, easy-to-use
balance sheet type liquidity stress testing tool that allows running bottom-up tests for
hundreds of banks: First, the tool can be used to run some basic tests in circumstances where
data is very limited to broad asset and liability items. Likewise, a cash flow based module
allows running detailed liquidity analysis like those carried out by banks for the internal
purposes but again can be adapted to a more limited data environment. Second, the
framework includes three broad dimensions (based on four modules) that allow for
complementary views on liquidity risks, including the link to solvency risks. Third, the paper
provides benchmark scenarios based on historical evidence on the one hand and common
scenarios used by FSAP missions on the other. Fourth, the framework allows assessing the
link between liquidity and solvency, albeit additional effort is needed in this context,
including work that captures dynamic aspects of this relationship and spillover effects such
as dynamically examining the link from liquidity to solvency concerns.114 As such, the
framework is meant to provide users with the possibility to run a meaningful system-wide
liquidity stress test within a relatively short period of time, but can also be used for
monitoring purposes.

It is vital to bear in mind that the key benefit of system-wide stress tests is to benchmark
banks against one another, i.e. to run peer comparisons and thereby assess their relative

112
See Annex 9.1 for the evolution of liquidity evaporation during the crisis.
113
Examples include Chile, Germany, India, Turkey and the UK.
114
See IMF (2011) and Barnhill and Schumacher (forthcoming) in that context.
 177

vulnerability to different shocks. Whether and how a shock materializes depends on the
various factors, with behavioral aspects increasingly playing an essential role.115 Hence, it is
also acknowledged that regular liquidity stress testing is not a panacea for a qualitative
judgment by policy-makers in order to complement findings even from well-designed
liquidity stress tests.

While cash flow data reporting, for instance, will become mandatory in the European Capital
Requirements Directive (CRD) IV regulation, it is (for now) still rarely available at
regulatory/ supervisory institutions so we follow a two-pronged approach, distinguishing
between implied cash flow tests and a “real” cash flow approach116, thereby seeking to lift
liquidity tests to a next generation level.117

The framework consists of three elements:

(i) Stress testing funding liquidity based on an implied cash flow approach, with two
different components: (a) a tool to simulate bank-run type scenarios while accounting
for fire sales of liquid assets and/ or central bank liquidity provision subject to eligible
collateral and haircuts;118 and (b) a liquidity gap analysis module that matches assets
and liabilities for different maturity buckets under different stress assumptions,
including rollover risk; the tool also allows for calculating (simplified)119 Basel III
liquidity ratios.

(ii) Cash flow-based liquidity tests—Running this module ideally requires detailed data
on contractual cash flows for different maturity buckets and behavioral data based on
banks’ financial/funding plans. If the latter are not available, the tool can be run on
contractual cash-flows only and behavioral flows can be modeled based on the stress
test assumptions. The calibrated scenarios then denote roll-over assumptions for
contractual cash-outflows and cash-inflows. The former focus on funding risk and the
latter take into account the banks’ objective to maintain its franchise value even under

115
In an environment of unstable short-term funding, the reaction of counterparties to anything from an actual
liquidity squeeze to unjustified rumors can have a highly devastating impact.
116
The idea is that supervisors and regulators can move towards cash flow approaches once data becomes
available. Moreover, the input template could be used as a benchmark for the data collection exercise.
117
While accommodating for such a flexible design it is up to the stress tester to understand the limitations of
sacrificing granularity of data input and the impact on the quality of the results.
118
Market liquidity is thereby captured through haircuts.
119
It is taken into account that full granularity needed to calculate the Basel III liquidity ratios is often not
available.
 178

stress.120 In addition, market funding risk can be captured through haircuts.

Accordingly, the module allows for an intuitive view of each banks’ liquidity risk
bearing capacity in the form of the cumulated counterbalancing capacity at the end of
each maturity bucket. In addition to stress testing, the module is also meant to be used
for liquidity monitoring purposes, for which behavioral cash-flows are particularly
informative.

(iii) Tests linking solvency and liquidity risk—the tool allows linking liquidity and
solvency risk from three complementary perspectives. The assumptions are crucial
for these tests and require sound judgment by the stress tester. First, the module
allows simulating the increase in funding costs from a change in solvency, indicated
by a change in a bank’s (implied) rating.121 Second, the tool enables simulating the
partial or full closure of funding markets (both long and short-term) depending on the
level of capitalization with or without considering solvency stress. Third, it allows
examining the potential impact of concentration in funding and a name crisis (e.g.,
from parent banks) on banks’ liquidity positions.

The output of the tests provides failure and pass rates (in terms of the number of banks and
total assets, respectively), and the estimated funding shortfalls for each bank as well as at the
system level (or group of banks tested). For instance for the fully-fledged cash flow test,
(cumulative) funding gaps and the corresponding (cumulative) counterbalancing capacity for
each maturity bucket are provided after haircuts and roll over rates for each bank and the
aggregate banking system. For the LCR and the NSFR the tests show which banks are likely
to be below the regulatory threshold.

The paper is organized as follows: Section B first provides some generic considerations on
concepts and methods to assess liquidity risks. Section C presents the newly developed
methodological framework. Section D is devoted to designing “extreme yet plausible”
scenarios by focusing on run-off assumptions for different funding sources, issues pertaining
to asset fire sales, collateral and haircuts, as well as on illustrating some benchmark
scenarios. Section E presents an illustrative case study and section F concludes.

120
If behavioral cash-flows are available, the stress test assumptions can be applied to these. While behavioral
cash-flows are more challenging to collect, they allow the stress tester to take into account individual bank
strategies explicitly (e.g. regarding its future funding mix).
121
If available, market implied ratings and liquidity measures (e.g. bid-ask spreads, trading volume in cash and
repo markets) should be used. Alternatively, letter ratings can be used. Calibrating adequate models is a pre-
condition to run such tests.
 179

B. Review of General Concepts to Assess Liquidity Risks

General Considerations and Motivation

Compared to solvency stress tests, particularly market risk, liquidity stress tests are less
developed, for several reasons: (a) liquidity risk management appeared to be “less of an
issue” until the current crisis, hence stress tests have a shorter development history making
use of IT systems (which greatly facilitate this purpose); (b) liquidity crises are very low
frequency-high impact events, which greatly reduces historical cases to calibrate models122;
and (c) all liquidity crises are somehow different, at least if one seeks to analyze how triggers
become manifest in a shortfall of liquidity, reducing the meaningfulness of “standard” stress
assumptions.

What makes liquidity crisis highly challenging is that they usually occur very suddenly,
spread by a mix of facts and rumors, giving banks very little time to react123. This warrants
that liquidity buffers are based on highly conservative principles—an important consideration
for the design of scenarios (section D). A key principle by the U.S. authorities during the
height of the financial crisis late 2008 was to ensure that ailing (investment) banks make it
through a business week, in order to find a viable solution during the weekend. With the
regulatory framework to be established by Basel III, the aim is to assure banks’ resilience
against a “significant stress scenario lasting 30 days” (BCBS 2010b), i.e., to survive a month
of (medium to severe) stress.

Idiosyncratic liquidity crises can be triggered by various events, most notably solvency
problems, but also political instability and fraud for example. Contagion can escalate
idiosyncratic shocks into market-wide shocks, as seen during the recent crisis period.

Annex 6 provides an overview of the evolution of liquidity conditions during the financial
crisis, together with more information on liquidity risks and regulatory action during since
the onset of the great recession.

122
In some cases, central bank support via liquidity provisions has masked the extent of an explicit liquidity
squeeze, with many banks hoarding liquidity and banks faced with funding liquidity challenges merely
substituting their loss of market wholesale and/or retail funding with central bank funding.

123
During a recent market turmoil, fuelled by rumors on potential risks, French banks lost about $60bn of
funding in U.S. short-term markets, especially from U.S. money market mutual funds, within a few days—
which is equal to one third of their U.S. Dollar liabilities and 6 percent of their foreign deposit liabilities, but
“merely” 1.3 percent of their total deposit holdings (J.P. Morgan, Global Asset Allocation Report from 12
August 2011: “Flows & Liquidity: Fears about French banks overdone”). Overall, due to their strong liquidity
position for now (i.e., solid liquidity buffers) banks managed to digest the withdrawal, but the example clearly
shows how sensitive short-term wholesale funding markets can be.
 180

In conceptual terms, the framework seeks to reflect the following, stylized nature of liquidity
crisis, distinguishing between three stages:124
(i) Stage 1—Liquidity crises originate from a sudden dry-out of funding sources.
Initially, the dry-out could be associated with higher costs (and thus lower income125).
Unsecured wholesale funding is the most “vulnerable” (i.e., sensitive) to changes in
the business climate and/or a name crisis, notably due to the fact that usually
considerable funds are at stake, tenors are often short, and counterparties tend to be
more sensitive to bank reputation and market rumors. In the first step, funding costs
will rise (visible both in spreads and, for secured funding, collateral requirements).

(ii) Stage 2—As the situation worsens further, some wholesale funding markets start
closing for single names and/or the whole system, first the long-term markets and
then short-term ones. Unsecured funding is the first source to drain, while secured
funding might remain available, but at higher costs (i.e., higher prices and/or
collateral). At the same time banks start hoarding liquid assets which they can use as
collateral, such as government debt and central bank reserves.

(iii) Stage 3—As a crisis unfolds, bank runs start, which are often subject to contagion and
thus develop into a banking crisis unless this is prevented by policy intervention.
Silent bank runs (e.g. Greece, Ireland in 2010) are by far more common than the text-
book bank run during which retail customer start queuing outside the banks’
branches. In a silent bank run, large corporate and retail depositors start withdrawing
their deposits and move them either to competitors within the banking system or
abroad if the whole banking sector suffers from a systemic crisis. However, despite
the wide-spread availability of deposit insurance systems, “even nowadays” retail
funding considered stable can be subject to a run, as Northern Rock has vividly
demonstrated.126’127 More informed retail depositors (often with higher amounts at
stake) are likely to be the first ones to react to potential crisis indicators of certain
institutions or the system in general.128 Likewise, with competition for retail deposits

124
This stylized process corresponds to empirical evidence, see De Haan and Van den End (2011).
125
It is assumed that only part of the increase in costs can be passed on to customers.
126
The UK deposit insurance scheme entailed a co-insurance component which implied a substantial reward for
depositors who withdrew early, thus thwarting the very rationale for deposit insurance.
127
In fact, the deterioration of fiscal conditions of many sovereigns can undermine deposit insurance systems,
particularly if they are meant to provide unlimited guarantees. An unlimited deposit guarantee—yet informal—
was given by Angela Merkel for retail deposits in Germany at the beginning of the financial crisis, which
appears to have calmed down the general public.
128
Moreover, the recovery of deposits is, at best, subject to administrative burden and usually takes some weeks
despite the fact that the pay-out schedule has been shortened in many countries recently, or could still face a
loss should the deposit insurance scheme not be sufficient to cover all losses
 181

having increased significantly as has the transparency of deposit rates, deposits have
become more volatile in recent years, particularly for the ones driven by yield, which
is further spurred by regulatory evolutions (Basel III, see Annex 6). 129 Another
important reason for the withdrawal of deposits is that typically the amount covered
by deposit insurance is limited, so holdings above a certain limit are subject to
potential losses for the depositors.

The obvious solution to counterbalance bank-run type outflows is liquidating assets through
fire sales. The dilemma for banks is the cost of holding high quality liquid assets, particularly
cash and “prime” government bonds. More illiquid securities are less costly (i.e., qualify as
some substitute for traditional bank business) but subject to higher haircuts, at best, or cannot
be sold at all (i.e., become illiquid) and/or do not qualify as eligible collateral any more.

In case of longer lasting liquidity disturbances, the maturity profile of assets and liabilities
plays an important role, as inflowing assets can then be used to deleverage, provided that (at
least partly) maturing (longer-term) debt can be rolled over. In fact, the analysis of rollover
risks has become an important aspect of liquidity risk analysis as many large banks are facing
a “wall of funding” over the coming years.130 For instance, with cheap funding in advanced
economies due to the low interest rate policy of central banks, capital flows have swelled into
emerging market countries (EC) countries with their domestic banks increasing their reliance
on cheap foreign and short-term funding. It is especially this type funding that can dry up in
(external) shock scenarios, and banks suddenly face rollover problems.

A natural counter-balancing role is played by central bank funding. In case of a severe crisis,
central banks can act as a lender of last resort. For instance, a number of central banks
entered swap agreements with the Fed during the financial crisis so they could supply their
domestic banks with much needed dollar funding. In fact, the Fed became the world’s USD
lender of last resort during the crisis providing liquidity to also large international banks such
as Barclays and UBS besides domestic U.S. financial institutions. 131 Parent banks can also
step in to increase or maintain credit lines to subsidiaries if a subsidiary or branch looses
access to funding sources while the parent retains access to funding (ideally in the required
currency). Yet, the crisis gave rise to episodes of ring-fencing which restricted the
transferability of capital and liquidity during stress times (see Cerutti et al., 2010, for

129
Recently referred to as a “deposit war”.
130
Banks’ debt maturity profiles are monitored in the GFSR, for example.
131
Moreover, in May 2010, the Bank of Canada, the Bank of England, the European Central Bank, the Federal
Reserve, and the Swiss National Bank announced the re-establishment of temporary U.S. dollar liquidity swap
facilities in response to the re-emergence of strains in U.S. dollar short-term funding markets in Europe. Since
then, these were extended twice in terms of time and as recently as mid September 2011 in terms of scope with
especially the ECB providing unlimited 3-month U.S. dollar funding after the re-intensification of funding
strains in Europe.
 182

example). Nevertheless, parent funding (predominantly in Euro for, both, Euro area and non-
Euro area subsidiaries) turned out to be more stable than alternative funding sources (i.e.
non-Euro area subsidiaries’ access to Euro wholesale markets)132, in Central and Eastern
Europe supported by the Vienna initiative.133 CEBS (2009a) suggests that the majority of
instances in which parent institutions did not provide additional liquidity for subsidiaries
were due to idiosyncratic liquidity shocks hitting the parent as a consequence of severe
(perceived) solvency problems of the banking group, i.e. in circumstances where they could
not provide support.

The “typical” balance sheet structure of banks based in OECD countries, ECs and low
income countries (LICs) is displayed in Annex 7. It is shown that the key difference on the
asset side is that banks in OECD countries exhibit a lower level of cash and government
securities in favor for a higher portion of other securities than in ECs and LICs. The total
portion of securities that can be used for fire sales is approximately the same. On the liability
side, banks’ portion of wholesale funding is substantially higher in OECD countries (both
short-term and long-term) than in ECs and LICs (where banks predominantly use customer134
deposits to fund their business), and has grown rapidly during the buildup of the financial
crisis. The data do, however, not confirm the believe that the portion of short-term wholesale
funding is positively correlated to size. Rather, the largest banks (with assets more than $1
trillion) exhibit a slightly lower portion of short-term wholesale funding (14 percent vs. 17
percent on average). Reducing their dependency on (unsecured) short-term wholesale
funding will take time and is costly for the banks that have sizeable portions. The stylized
balance sheets of “average” banks will be used to illustrate the framework in section E and
some additional liquidity patterns.

C. Methodological Aspects

Overview of Recent Methods to Assess Liquidity Risks

A natural starting point to assess liquidity risks is through financial soundness indicators
(FSIs), which provide relevant information on the liquidity position of banks, both vis-à-vis
peers (banks and/or countries) and over time.

132
Parent bank funding is important in two cases: first, the subsidiary is the same currency area, but liquidity
management and (parts of) funding are centralized; second, the subsidiary is in another currency area, but
features a multi-currency balance-sheet (e.g. the subsidiary provides FX-loans). In the latter case non-Euro area
subsidiaries hardly have direct access to long-term stable Euro funding.
133
Within the European Union, transfers of capital and liquidity can, in principle, not be restricted (European
passport). The Vienna initiative sought to prevent the withdrawal of Euro funding from Western European
parent banks in Central and Eastern Europe—to safeguard financial stability, which proved quite successful.
134
i.e., retail and non-bank SME/corporate deposits.
 183

One of the early adopters of cash flow based liquidity stress testing (both top-down and
bottom-up) in recent years has been the Austrian National Bank (for instance OeNB, 2008135),
or more recently Schmitz (2009 and 2010), whose work has heavily influenced the European
approach as well (see, e.g., ECB, 2008).

Van den End (2008) at the Dutch Central Bank developed a stress testing model that tries to
endogenize market and funding liquidity risk by including feedback effects that capture both
behavioral and reputational effects. Using Monte Carlo approach he applied the framework
to the Dutch banks and showed that second round effects have a more substantial impact than
first round effects (i.e., that liquidity risks are highly non-linear), resulting from collective
behavior and suggesting that banks hold substantial liquidity buffers. Wong and Hui (2009)
from the Hong Kong Monetary Authority sought to explicitly capture (i.e., endogenize) the
link between default risk and deposit outflows. As such, their framework allows simulating
the impact of mark-to-market losses on banks’ solvency position leading to deposit outflows;
asset fire sales by banks is evaporating and contingent liquidity risk sharply increases.

An attempt to (fully) integrate (funding) liquidity risks and solvency risk is the Risk
Assessment Model for Systemic Institutions (RAMSI), developed by the Bank of England
(Aikman et al., 2009). The framework simulates banks’ liquidity positions conditional on
their capitalization under stress, and other relevant dimensions, such as a decrease in
confidence among market participants under stress. By now, the framework can be regarded
as the most comprehensive approach to endogenize liquidity risk stress tests in a modeling
framework.

At the IMF, in the context of FSAP stress tests, liquidity tests were originally centered on
Čihák (2007) using bank balance sheet data to perform bank-run type stress tests on a bank-
by-bank level. Besides, a recent chapter of the GFSR (April 2011) focused on systemic
liquidity, based on a Merton-type approach using market data and balance sheet information
to estimate banks’ individual liquidity risk and to calculate the joint probability of all
institutions experiencing a systemic liquidity event accordingly, which can be captured by
means of a systemic liquidity risk index, for example.136 Barnhill and Schumacher
(forthcoming) develop an empirical model that seeks to link solvency risk and liquidity risks,
similar to Van den End (2008) and Wong and Hui (2009). The framework attempts to be
more comprehensive in terms of the source of the solvency shocks, and tries to compute the
(longer term) impact of funding shocks, i.e., deleveraging beyond fire sales.

135
In the liquidity stress tests conducted for the IMF FSAP in 2007, highly adverse scenarios were adopted to
test the resilience of Austrian banks. See section IV.E for further information.
136
The framework can also be used to compute each institution’s contribution to systemic risk and systemic risk
shortfalls, respectively, which could trigger an insurance premium.
 184

On the regulatory side, substantial micro-prudential efforts were undertaken to contain

liquidity risks on a bank-by-bank level: In 2008, the Basel Committee of Banking
Supervision (BCBS) published guiding principles for sound liquidity risk management
(BCBS 2008) and an overhaul of the regulatory framework followed in December 2010
(BCBS 2010b), when the Committee introduced two measures to contain short-term
vulnerabilities on the one hand and excessive maturity mismatch on the other. To this end,
the minimum liquidity standard will incorporate a 30-day Liquidity Coverage Ratio (LCR),
essentially a pre-specified substantial bank-run type stress test lasting a month that banks
have to pass in order to be considered rather safe in the short-term, and a longer-term
structural liquidity ratio, the so-called net stable funding ratio (NSFR) that aims at limiting
maturity mismatch, with a focus on the next 12 months. Both ratios are subject to a transition
phase, during which the ratios will ultimately be calibrated and are scheduled to be fully
implemented by 2015 (LCR) and 2018 (NSFR), respectively.

In addition, several macro-prudential approaches to manage systemic liquidity risk have been
brought forward during the last two years (which have, at least partially, been used in
emerging markets for many years). All approaches aim at introducing incentives to limit
systemic liquidity risks, including through levies, capital charges and introducing minimum
liquidity ratios and haircuts, but implementation seems unlikely at this stage, mainly due to
the complexity of measuring systemic risk. See IMF (2011) for further information.

Finally, in the industry bank level tests are centered on maturity mismatch approaches,
sometimes complemented by stochastic Value-at-Risk components for those funding sources
for which sufficient histories of high frequency data is available. The ultimate goal of
liquidity tests is to determine a banks’ risk tolerance for liquidity risk, i.e. the maximum level
of risk that the bank is willing to accept under stress conditions. Most large European banks
compute their maximum risk tolerance (ECB 2008), for example. The stochastic approach
aims at determining Liquidity at Risk137 (maximum liquidity gap within a certain time
horizon and for a given confidence level) or Liquidity Value at Risk (maximum cost of
liquidity under certain assumptions). While instructive under business as usual and mild
stress scenarios, these models face limitations in stress testing under more severe liquidity
shocks. Given that liquidity risk is a low frequency, high impact risk, historic volatilities and
correlations tend to underestimate funding risk under severe stress, which is highly non-
linear (see ECB 2008, for example).

137
Liquidity at Risk denotes computing a 99.9 percent event based on the cumulative probability distribution (as
is done for market risk and credit risk).
 185

Top-down (TD) or Bottom-Up (BU)?

The most intuitive way to stress test liquidity risks is to use cash flow level data, usually
available only within banks.138 Provided that the cash flow structure and maturity of all cash
flows is monitored through IT systems, the challenge is how to deal with:

(i) The volatility, i.e., cash flows with non-predefined cash flow structures, such as
contingent liabilities (e.g., credit lines) on the asset side and demand deposits or
short-term interbank market access on the liability side as well as
(ii) The strategy of managing maturity mismatch.

For system-wide liquidity stress tests, the subject matter of this framework, there are two
ways to stress test liquidity risk:

(i) Defining common scenarios that are run by banks themselves, so-called bottom-up
(BU) tests, making use of granular data, or

(ii) Collecting data by broader liability and asset types, currency and maturity and
applying scenarios accordingly in a top-down (TD) fashion.

The framework at hand mainly caters to the purpose of running TD stress tests. As such, the
main advantage is to be able to run a set of consistent tests for all banks in the system (and
relevant banks and non-banks outside of it). In principle, the tool could also be used to gather
BU results and run additional sensitivity tests accordingly as outlined below. Table 12
summarizes the relative strengths and weaknesses of the two approaches for liquidity risk,
omitting the hybrid case (TD, run by banks).

An interesting combination of BU and TD approaches are concerted rounds of common

liquidity stress tests (e.g. ECB 2008) which also collect data on banks’ measures taken in the
face of the common scenarios and incorporate second-round effects in an additional TD
round based on the results of the BU exercise. For example, if the majority of banks report
asset sales of particular asset classes in their counterbalancing capacity, the TD analysis
would increase haircuts on those assets; if banks report that they would discontinue reverse
repos, the TD analysis would incorporate a (further) reduction in repo roll-overs.

138
Deutsche Bank, for example, has information on the expected daily cash flows for the next 18 months; both
on-balance and off-balance, by currency, product and organizational division (see Deutsche Bank, 2009, p.95).
 186

Table 12: Comparison of Pros and Cons of Balance Sheet Type TD and BU Liquidity
Type of Test
BU test (run by banks)
TD tests (run by authorities)
Stress Tests
Pros Cons
Cash flow level data, use of Less consistent than TD
models developed by banks,
P&L effects of liquidity
shocks and cost of funding
shocks can be incorporated
more easily.
Consistent approach, Less detailed data, bank-
authority is flexible to run specific situation less
various scenarios, recognized; data are
transparency of situation to outdated rapidly, which can
authority be prevented by a high, but
burdensome frequency of
reporting
Source: Authors

Outcome of Liquidity Stress Tests

The outcome of TD liquidity tests is three-fold:

(i) They show the counter-balancing ability of banks on the one hand (and their specific
limit in case of reverse stress tests)139 to remain liquid,
(ii) They reveal a peer comparison, i.e., the relative performance of banks under liquidity
stress on the other hand, and
(iii) They can provide a link between the joint resistance to liquidity and solvency risks if
the feedback between solvency and liquidity risks is modeled in the TD stress testing
framework.

D. Framework of Next Generation Liquidity Stress Tests

The framework originates from the balance sheet based liquidity stress tests based on Čihák
(2007), and seeks to account for (a) lessons learnt from the crisis on the one hand; and (b) the
evolution of conceptual and regulatory initiatives on the other, i.e., taking into account recent
progress in terms of evidence and conceptual progress as discussed in section B. The

139
Reverse stress test seek to identify maximum stress resistance of banks / the banking system by increasing
the risk factors (e.g., haircuts, run-off rates, etc.) until a predefined threshold (e.g., positive counter-balancing
capacity) is reached..
 187

framework is part of a larger project on next generation balance sheet stress testing at the
IMF, a framework initiated by Schmieder, Puhr, and Hasan (2011).

As such, the tool provides extensions in five dimensions:

(i) A more granular balance sheet structure can be exploited.

(ii) Maturity mismatch is explicitly taken into account through separate tests.

(iii) The framework allows computing (simplified) Basel III liquidity ratios, both the LCR
and NSFR (see also Box A6.1).

(iv) A fully-fledged cash flow test can reveal detailed information on banks’
vulnerabilities provided that granular information is available.

(v) A framework to link liquidity risks and solvency risks addresses liquidity from
complementary angles and allows examining the impact of changes of funding costs
and a (partial) closure of funding sources on solvency and liquidity as well as funding
concentration risks.

More specifically, the innovations of the tool can be summarized as follows: First, the tool
allows for more flexibility and adds additional elements (such as the portion of encumbered
assets or examining banks’ overall interbank exposures) to the implied cash flow tests
established by Čihák (2007). Second, maturity mismatch analyses are extended, with a fully
fledged cash flow test allowing for tests that are similar to the ones run by banks themselves
based on granular contractual and behavioral cash flow data. Third, concentration risk
analysis and the new Basel III ratios are added, which were not available in previous tools.
Fourth and last, the link to solvency seeks to account for lessons learned in the recent past,
and brings in a dynamic element. In the latter dimension, the tool seeks to bring forward
straightforward ways to deal with the issues, while other frameworks (e.g., RAMSI) are more
of a black box nature and need considerable technical effort to be set up. Lessons learned
from the financial crisis were also taken into account for the above improvements.

The key elements of the framework on liquidity risk are displayed in Figure 43. Due to the
lack of empirical cases as argued previously, the calculation of satellite models (i.e.,
econometric models) that link the outflow of deposits to macroeconomic conditions is not
(yet) feasible. However, such models can be used to determine the haircuts for assets under
stress (i.e., market liquidity risk). In addition, satellite models can be used to link banks’
solvency under stress (e.g., capital ratios, or default probabilities) to funding costs.
Accordingly, a multi-period solvency test can link the deterioration of liquidity conditions to
the evolution of bank solvency and vice versa.
 188

Figure 43. Overview on Liquidity Risk Framework

Solvency
(Stress Test) Satellite Expert
models Judgment

Parameter
Funding
Variable
Costs
Setup Assumptions

Core functionality

Calculation Calculation Calculation

(bank runs) (gap analysis) (cash-flows)
Results
(incl. cash-flow template)

Input data

Result
summary

Source: Authors

Table 13 displays the main features of the three modules that constitute the framework,
namely (a) bank run type implied cash flow analysis (ICFA); (b) maturity gap/rollover tests
based on ICFA and a fully-fledged cash flow approach; and (c) integrated solvency/liquidity
tests.

The tests are meant to assess complementary dimensions relevant for liquidity risks, namely
(a) the capacity to withstand a bank-run (short-term counter-balancing capacity); (b) the
extent and capacity to deal with maturity mismatch; and (c) potential threats to liquidity
arising from solvency risks.

The functioning of bank run test is illustrated by means of a case study in section E.
Moreover, Annex 8 provides more detailed information of each individual liquidity stress test
and additional information is given in the tool itself.
 189
Table 13. Overview on the Main Elements of Three Liquidity Tests
Type of Test Description
Assesses banks’ counter-
balancing capacity in case of bank
run type scenario, simulating a
gradual outflow of funding for a
time frame of (a) 5 periods (days,
weeks, months); and (b) fixed
Implied Cash Flow period (30 days/3 months).
Analysis (ICFA) Scenarios account for market
liquidity of assets (in case of fire
sales). The gradual test is usually
run as a reverse test.
(Proxy for) LCR, which assesses
the counterbalancing capacity of
banks for the next 30 days; The
regulatory weights can be
changed to assess sensitivities
The liquidity gap simulation
matches liability and asset
maturities and identifies liquidity
gaps for each maturity bucket and
Maturity under different scenarios. The test
Mismatch/Rollover is available (i) as a simplified
Stress Test version with limited data
requirements and (ii) as a fully-
fledged cash flow test.
(Proxy for) NSFR assesses the
stability of banks’ funding sources
in more structural terms.
Simulates the impact of changes
of solvency and concentration risk
Integrated Liquidity and on liquidity conditions and vice
Solvency Tests versa (the first two modules
require input from a solvency test
and funding cost model,
respectively).
Source: Authors
Most of the tests are deterministic, but the framework can be extended to become more
dynamic. In fact, the framework’s link between funding costs and capitalization has been
Outcome
Which banks “fail” the test?
(test enables peer
comparison); Which portion
of banks remains liquid
under a specific scenario?
How much liquidity shortfall
occurs at the bank and
system level, if applicable?
Which portion of banks
meets regulatory
requirements? How much
liquidity shortfall occurs, if
applicable?
Which portion of banks
remains liquid up to a
specific maturity bucket?
How much is liquidity
shortfall, if applicable?
Which portion of banks
meets regulatory
requirements? How much
liquidity shortfall occurs, if
applicable?
Which portion of banks
remains liquid/solvent under
the specific assumptions?
How much liquidity/capital
shortfall, if applicable?
 190

used in multi-period solvency stress tests based on Schmieder, Puhr and Hasan (2011), for
example in the case of the Germany FSAP. Likewise, the five period implied cash flow test
could be made dynamic, for example by gradually adding additional elements that hit banks
that are performing badly under stress. The closing of funding markets conditional on bank
solvency is another area that would invite such a dynamic analysis, shedding light on
potential short- and medium-term deleveraging effects resulting accordingly.140 As dynamic
designs are highly challenging, they have not been implemented as part of the standard
version of the tool, but future releases might see some of the elements being added.

E. Design of Stress Scenarios

General considerations

In line with the overarching principle for sound stress testing, scenarios should be “extreme
yet plausible”, which is even more important for liquidity risks than it is for solvency risk as
only solid liquidity buffers can ultimately safeguard banks, unless there is a major systemic
event when even those no longer suffice to mitigate a liquidity squeeze. Given that liquidity
crises are infrequent (more so than solvency crises), evidence is scarce and stress levels vary
widely. However, conditions tend to be very unfavorable once there is stress, i.e. stress is
highly non-linear. As a consequence, the tool allows for a range of scenarios with varying
degrees of severity to be run at low cost which we strongly recommend. The output across
the scenarios then provides a clear view of the relative liquidity risk exposures and liquidity
risk bearing capacities of the banks in the system. This allows supervisors to interpret the
results on the basis of their own liquidity risk tolerance for the individual banks in the system
and the aggregate system. Finally, scenarios can be interpreted as tools to condense a wealth
of bank data and assumptions concerning the environment in which banks operate in a way
that is consistent and intuitive. Based thereon supervisors can then scrutinize the funding
structure of those banks that are flagged by the stress test to derive individual policy
conclusions.

The classic alternative to point estimate based scenarios is to “stress it until it breaks” (Ong
and Čihák, 2010) also referred to as reverse stress tests, where tests are used to determine a
set of scenarios that would cause an increasing part of the system (or specific banks) to run
short of liquidity. Reverse stress tests and tests simulating “extreme yet plausible” scenarios
complement each other, and thus there is a good reason to run both, especially for liquidity
risk.141

140
This dimension is highly relevant under the current circumstances in Europe, for example.
141
A challenge is how to deal with the outcome of reverse stress tests in the context of authorities’ stress tests.
Given the sensitivity of liquidity risk an appropriate way to disseminate the results has to be found.
 191

In general, three basic types of inputs can be useful in designing extreme but plausible
scenarios: (a) past experience; (b) expert judgment; or (c) an individual, reverse test type
assessment of the limit for each bank. Scenarios should take into account both market-wide
shocks (a worsening of market conditions and investor confidence) that affect all banks in the
system as well as an idiosyncratic shocks, e.g., due to deterioration of the solvency of single
banks. Given that market confidence in individual banks is more fragile under market-wide
stress, a combined scenario should be taken into account as well (e.g., the LCR is modeled
around such a combined scenario).

If possible, scenarios should be accompanied by a consistent “story line” that underpin the
assumptions on all relevant elements, namely (i) run-off rates for funding; (ii) haircuts for
assets sold at fire sales prices and drawings of contingent liabilities; (iii) the impact of banks’
rating downgrades, i.e. a deterioration of bank solvency. For the analysis of maturity
mismatch, additional parameters (e.g., roll-over rates) need to be modeled in a consistent
manner.

In the case of retail deposits, for example, guiding questions for the design of scenarios and
the development of story lines could be: Which retail deposits are the most vulnerable (e.g.,
foreign currency denominated deposits, deposits held by foreigners abroad, demand deposits
in case of an increase of policy rates from very low levels) and would go first? Would
depositors hoard cash or shift deposits outside the national banking system in the event of a
crisis? Under what conditions would a flight to quality initiate deposit inflows at a subset of
banks in the system and an outflow at others?

F. Run-off Rates for Different Funding Sources

Table 14 provides an overview of the magnitude of a loss of funding based on empirical

evidence as well as parameters used for stress testing in a broader context.

The financial crisis provided ample evidence for solvency and liquidity crises of banks. For
liquidity, probably the most prominent victims were the U.S. investment banks, which
suffered from interbank markets drying up in combination with solvency concerns given their
continuing efforts to raise needed capital (e.g. from Sovereign Wealth Funds). Other banks
became victims of their rapid and aggressive growth strategy and heavy reliance on
wholesale funding, which applies to U.K.’s Northern Rock among others. The former even
experienced a text-book retail bank run, with people queuing in front of the bank’s branches
to withdraw their money, after a silent wholesale run. A third group of banks were those
domiciled in countries with a major recession and/or banking crisis, such as in the Baltics and
in Kazakhstan. Lately banks based in peripheral Europe have become highly dependent on
funding by the ECB, testimony that they are shut out of the interbank market, debt capital
markets, and a protracted outflow of funding, including retail deposits, in some cases.
Selected examples of recent bank runs were summarized in Table 14 and further information
is provided in Annex 9.
 192

Table 14. Magnitude of Runs on Funding—Empirical evidence and Stress Test

Assumptions
Loss of Customer Loss of Wholesale Funding
Deposits
Empirical evidence142
Banking System in Saudi 11 Percent (1 week)
Arabia (August 1990143)
Banesto (ES, 1994) 8 percent (1 week)
Banking System in Argentina Deposits in domestic
(2001) currency: 30 percent
(9 months)

Northern Rock (UK, 2007) 57 percent (12 months) 57 percent (6 months)

Parex Bank (LV, 2008) 25 percent (3 months)
IndyMac (US, June 2008) 7.5 percent (1 week)
Washington Mutual (US, 8.5 percent (10 days)
September 2008)
30 percent (12 days)
DSB Bank (NL, 2009)
Regulatory Parameters Stable: min. 5 (unsecured categories)
LCR (30 days) Less stable: min. 10 Stable SME: min. 5
Less stable SME: min. 10
Non-financial corporate, public sector: 75
All other deposits: 100
(secured)
Repos: 0-25 (quality collateral), 100
otherwise
Regulatory Parameters Stable: 10 (unsecured categories)
NSFR Less stable: 20 Short-term corporate & public sector (< 1
year): 50
Rest: 100
144
Recent FSAPs 10-50 percent (up to 10 to 50 percent for non-bank deposits
80 percent for non-
100 percent for bank funding
resident deposits)
50 to 100 for parent funding
Source: Authors based on publicly available data

142
Other bank runs include MBf Finance Berhad (Malaysia, 1999), Bear Stearns (US, 2008) and Landesbanki
(IS, 2008), for example.
143
Period after the invasion of Kuwait by Iraq.
144
In many cases, the shocks were sensitivity analyses rather than scenario analysis, so the parameters are
higher.
 193

In addition, the two prudential Basel III ratios provide benchmark parameters for run-off
rates of funding sources, the LCR for a period of one month and the NSFR for 12 months
(BCBS 2010b). For retail deposits, the LCR foresees minimum outflow ratios (run-off rates)
are 5 percent for stable retail credit and funding provided by small- and medium-sized
enterprises (SMEs), respectively, and 10 percent for less stable funding. For the NSFR, the
level is twice as high (10 and 20 percent, respectively). Secured wholesale funding is subject
to withdrawal between 0 and 25 percent, provided that it is secured with higher quality
collateral, while unsecured wholesale funding is associated with run-off rates of at least 50
percent (for no-financial corporates), most of it 100 percent (especially for financial
institutions).

European Banks145 use similar parameters for their internal stress tests, with retail deposit
run-off rates mostly at 10 percent (up to 30 percent), and wholesale run-off rates ranging
from 0 to 100 percent (100 percent is assumed by one fifth of the banks in the survey) (ECB
2008).

Table 14 also includes assumptions used by recent Financial Sector Assessment Programs
(FSAPs) in different countries. However, these parameters were, in most parts, to be
understood as input for sensitivity analysis, which is why the severity is higher. Further
information on stress test parameters used in FSAP stress tests is provided in Annex 9.

G. Asset side: Fire Sales & Rollover

The counterbalancing ability of banks depends on their ability to generate cash-inflows from
liquid assets. This includes three elements (a) defining which asset types remain liquid (see
Annex 9 for a distinction of assets according to their liquidity profile adopted by the ECB);
(b) defining market liquidity, i.e., the loss in value (haircut) banks have to accept to sell the
asset; (c) defining the portion of liquid assets that remain unencumbered. In the latter context,
given recent events and the increased importance of secured funding (for example, repos and
covered bonds), it becomes crucial to collect data on the level of unencumbered liquid assets
on the one hand, and making assumptions about their availability under stress on the other,
accounting for potential margin calls.146

The haircuts should differentiate between asset categories, accounting for the level of stress
simulated on the one hand,147 the “quality” of assets (e.g., in case of debt securities the type

145
The survey is based on responses by 30 European banks in 2008.
146
The Basel III definition appears a meaningful benchmark (BCBS 2010b, para. 27).
147
Market prices can be assumed to be substantially lower in case of severe shocks. Driven by the fact that
multiple market participants will try to sell large amounts of the same assets at the same time in response to a
market-wide shock.
 194

and rating of the counterparty) on the other.148 In principle, one could also use haircuts for the
liabilities, to simulate a decrease in the availability of funding due to an increase in collateral
requirements (due to margin calls). In any case, one has to avoid double-counting—
(unencumbered) liquid assets can only be used to either generate cash or maintain the level of
funding (as a substitute of encumbered assets used as collateral that have lost in value).

Potential haircuts to be modeled comprise:

(i) Haircuts for (unencumbered) liquid assets (Table 15).

(ii) Haircuts for encumbered liquid assets (i.e., collateral/margin calls, see Table 15)
(iii) Add-ons (positive haircut) for contingent liabilities (see Table 15)

Deriving model-based haircuts requires a substantial commitment of time and resources, but
comes with the advantage of developing expert knowledge on the value of assets under
stress.149 Alternatively, stress testers can use supervisory haircuts foreseen to be used under
the (comprehensive) Standardized Approach for solvency purposes (BCBS 2006, para.
147f.), for example. These haircuts constitute a proxy for the 99th confidence interval for
different holding periods. Basel III distinguishes between two levels of high quality liquid
assets (so-called “flight to quality” assets) and refers to factors that can be used to define
whether funding remains liquid (BCBS 2010b, para. 22f.). A more granular classification of
marketable assets is the one by the ECB (Annex 9, Table A9.1), which distinguishes between
five categories, where category 1 and partially category 2 would correspond to the Basel III
level 1 assets and category 2 and 3 to Basel III level 2. Category 4 and 5 are assets not
considered as high-quality liquid assets under Basel III, but it is worth to run less severe
scenarios simulating that they are liquid subject to a considerable haircut, for example.150 An
overview of valuation haircuts that are applied to eligible marketable assets by the ECB is
displayed in Table A9.2 in Annex 9 (ECB, 2011).

Table 15 provides an overview of supervisory haircuts as part of the Basel II solvency

framework and the haircuts to be used for the Basel III liquidity tests. It is important to
recognize that the purpose of the parameters is different—for solvency purposes the maturity
is linked to the assets, while the maturity for liquidity purposes depends on the ratio referred

148
Maturities of the assets (and accounting for the holding period, i.e. the timing when the assets are likely to be
fire sold) and currency mismatch could also play a role.
149
Calculating the volatility of market prices of assets allows assigning probabilities for the occurrence of
scenarios. A useful guideline how to do so is provided in BCBS (2006, para. 156ff.).
150
Basel III outlines that the high-quality assets are likely to be comparable to the assets eligible for central
bank funding, but also that “central bank eligibility does not by itself constitute the basis for the categorization
of an asset as a ‘high-quality liquid asset’” (BCBS 2010b, para. 25).
 195

to (and the corresponding time frame). One can see, for example, that the LCR assumes
substantial stress, with equities becoming illiquid, for example.

Table 15. Supervisory Haircuts based on Solvency Regime (BCBS 2006, 2010) and
Basel III Liquidity Regime (BCBS 2010b)
Haircut/Weight for… Basel II (2006, 2010) Basel III (LCR) Basel III (NSFR)
0, includes also short-
Cash 0 0 term securities (less
than 1 year)
Issue Rating for Residual Other Secu-
Sovereign
Debt Security Maturity issuers ritization
< 1 year 0.5 1 2 Sovereign-type: 5;
Sovereign (RW 0%): 0;
AAA to AA-/A-1 1 to 5 years 2 4 8 Corporate (< 1 year):
Corporate: 15
> 5 years 4 8 16 20
(1) Unencumbered < 1 year 1 2 4
A+ to BBB- and
liquid Assets 1 to 5 years 3 6 12 Sovereign (RW 20%,
unrated
> 5 years 6 12 24 Rating A+ to A-): 15
BB+ to BB- All 15 Not eligible
Equity (main
15
index) and Gold
Other equity 25
100 50
Mutual funds (max
Up to 25 (highest haircut applicable to any
of allowed asset
security in the fund)
mix)
(2) Encumbered Haircut on collateral for
liquid Assets n.a. n.a. potential margin calls (3 n.a.
(collateral) notch downgrade)
Lines - Retail: 5;
(3) Add-ons corporate, credit lines: undrawn credit and
n.a. n.a.
(contigent 10; corporate, liquidity liquidity lines: 5
liabilities) lines: 100; other: 100

At this point, reputational considerations—which featured prominently in the ongoing

crisis—need to be built into the scenario assumptions. In particular, stress testers need to take
into account contingent liabilities such as committed credit/liquidity lines to customers and
sponsorships of Special Purpose Vehicles (SPVs) and outflows related to derivatives (i.e.
margin calls). This is a risk that is particularly high under market-wide funding market
dislocations. The LCR provides a valuable benchmark for contingent liabilities, including for
derivatives (in terms of the number of rating downgrades to be simulated).

As a general rule, stress tests should focus on the ability of banks to weather severe but
plausible liquidity shocks as going concern. That implies that the bank is able to maintain its
franchise value. To do so, it needs to keep generating new business (i.e. roll-over maturing
assets) and honor its commitments, which is the underlying assumption under Basel III, but
also by banks (see Deutsche Bank 2010, for example).

The tool also explicitly allows simulating how liquidity support provided by parent banks
and central banks would alter the outcome of the tests. In the former, any estimated liquidity
shortfall of a subsidiary could indicate the possible needed amount of additional parent
funding support. In the latter case, the central bank could assess whether its regular (e.g.
reduction of required reserves requirements or repos) and emergency liquidity support is
 196

sufficient and to determine how much additional liquidity might be need to be earmarked for
worst case situations, e.g. to close funding shortages in specific maturity buckets.

H. Link Between Liquidity and Solvency

There have been some recent attempts to link solvency and liquidity risk, which particularly
applies to (a) funding costs; and (b) the closure of funding markets once solvency conditions
deteriorate further.

The link between solvency and funding costs comprises two dimensions: (i) an increase of
the price to be paid for funding as such–wholesale funding is particularly sensitive to changes
in solvency, but recent competition for retail deposits is an indication that retail deposits also
becomes more price-sensitive going forward; (ii) an increase of collateral needs for secured
funding sources (margin calls).

The former dimension can be derived based on empirical evidence. One way is to use
econometric models to determine the increase in funding costs (i.e., interest expenses) on the
liability side, while also accounting for the effect on earnings on the asset side (interest
income).151 As an illustrative example152, a non-linear relationship between solvency
(measured as implied IRB capitalization) and funding costs has been established for
Germany. The procedure is illustrated in Annex 10 (Figure A10.1) based on an example and
explained below. The funding costs encompass a proxy for an average German bank (the
funding costs were weighted by the portion of each funding source and the pertinent costs,
using market data provided by the OECD and the ECB), based on the sample of all German
banks available in Moody’s KMV. For the illustrative example, the funding costs were
compared against the (one-year) EDF of the German banks for 12 quarters from 2007-2009,
i.e., a period of stress in funding markets. In the next steps, the EDF has been translated into
a capitalization ratio using the IRB formula153, inferring the minimum capital ratio based on
the confidence interval corresponding to the EDF, and adding an additional capital cushion of
2.5 percentage points (in line with the observation that banks hold more capital than the
minimum). It should be noted that the resulting implied capitalization is purely based on

151
The impact of changes in funding costs in net interest income ultimately depends on banks’ ability to pass
through their costs, but short-term developments also depend on the portion of assets and liabilities that can/will
reprice. In the example used for this study, it has been assumed that banks cannot pass on any increase of
funding costs to customers, which is very conservative.
152
Stress testers need to recalibrate it for the situation at hand, which differs widely across countries and banks,
depending on, for example, the country as such, the situation in the financial markets, the fiscal position of a
country, the regulatory environment, etc.
153
This was done by using through-the-cycle credit risk parameters—a probability of default (PD) of 1 percent
and a loss given default (LGD) of 45 percent. The capital requirements for market risk and operational risk have
been assumed to amount to 20 percent of the ones for credit risk for a confidence interval of 99.9 percent.
 197

quantitative elements and subject to the limitations of the Basel IRB model. The implied
rating, both in terms of the letter rating (which uses Moody’s (2010) to create the link
between the EDFs and rating letters) and capitalization should therefore not be confused with
external ratings granted by rating agencies, which are subject to comprehensive analysis
based on both qualitative and quantitative criteria, taking into account implicit government
guarantees, for example, which alters the situation. The ratios could serve as some
conservative, quantitative benchmark which capitalization levels would be needed to reach
certain ratings on a standalone basis. For the computation of the funding costs, it has been
assumed that banks cannot pass on any increase to customers, which is conservative and
could be relaxed (e.g., by using a pass-on rate of 50 percent).

For secured funding, counterparty credit risk plays an important role, as collateral
requirements depend on the rating of a counterpart. Hence, a deterioration of solvency (i.e., a
rating downgrade) leads to an increase of collateral requirements and thus a reduction of
funding. While the impact is highly bank-specific, it is non-linear, at least once banks drop
below the investment grade level. Deutsche Bank, for example, reports that a drop of its
rating by 1 notch results in a loss of funding of about 2 percent, and that the drop is about 6
times for a rating deterioration by 6 notches (Deutsche Bank, 2010).

Once market conditions deteriorate further, driven by general market conditions and/or
idiosyncratic strains at single banks, funding markets will close. An attempt to model to
capture the deterioration of various factors and link it to the closure of funding markets is
part of the RAMSI model, with a calibration for the United Kingdom (see Aikman et al.,
2009).154

The tool provides a template to simulate different scenarios with respect to funding costs on
the one hand and the (partial) closure of funding sources on the other. A key focus of the
tests is peer comparison. It is essential to ensure that the calibration is adequate for the banks
and/or system at hand, which remains at the discretion of the stress tester.

I. Liquidity Stress Tests in Recent FSAPs and Benchmark Scenarios

Table A9.3 in Annex 9 provides an overview of scenarios used for the liquidity stress tests in
selective countries during FSAPs. It should be noted, though, that some tests were meant as
sensitivity tests, whereby they are more conservative as assumptions used for scenario
analysis.

154
Empirical relationships between default probabilities, funding withdrawal, and (in the case of the GFSR)
deleveraging rates have been documented by Van den End (2008) and IMF (2011), but the ultimate impact will
remain very specific to the circumstances at hand.
 198

The scenarios covered a broad scope of potential events, catering for the needs in specific
circumstances, including limited access to parent credit lines (where applicable), separate
run-off rates for foreign deposits, but also the impact of rating downgrade.155

For instance in the case of Austria’s FSAP concluded in 2008, TD stress tests included a
market-wide scenario and six combined scenarios (the market-wide shock plus an
idiosyncratic shock for each of the six participating banks individually). The exercise was
embedded in the solvency stress test which focused on a macro-economic debt crisis. It
employed an implied cash-flow approach based on reported stocks of short-term assets plus
liquid asset and short-term liabilities. The implied cash-flows from short-term assets and the
stocks of liquid assets received different liquidity factors across a range of sensitivity
analysis, to test for variations in the roll-over rates of short-term loans to non-banks (100 and
50 per cent, respectively) and to banks (100 and 0 per cent dependent on the residual
maturity between O/N and 3 months). Four sensitivity analyses were conducted: (a) liquid
bonds minus 25 percent, (b) equity portfolio minus 35 percent, (c) withdrawal of 40 percent
of all interbank short-term funding, and (d) withdrawal of 50 percent of nonbank deposits. In
addition, a scenario analysis that combined a severe disruption of the money and credit
markets (a market shock) with an idiosyncratic shock (a name crisis) for each bank was
performed. The market shock included a decrease in bond and equity market prices of 20 and
30 per cent, respectively. Inflows from interbank loans received a haircut of 5 per cent to
account for potential liquidity problems at counterparties. The idiosyncratic shock assumed a
substantial shortening non-bank deposit outflows (sight deposits -10 per cent, term deposits
with a residual maturity of up to one and three months -20 and -30 per cent, respectively).

In the Austria Article IV consultation the Austrian Central Bank presented the results of a
concerted round of common liquidity stress tests (combining a BU approach based on the
scenarios and a TD approach concerning the second round effects) based on the weekly
standard Austrian maturity mismatch reporting template plus a separate template for
measures taken by banks in the face of the assumed shocks. The market scenario focused on
an assumed return of the Eurosystem to pre-crisis liquidity policy. The volumes of secured
and unsecured interbank deposits were capped at the low averages of the first half of 2008.
Furthermore, the exercise included BU estimates of the P&L effects of the scenarios plus a
substantial widening of the Euribor-OIS spreads (up to 3 M+50 basis points, up to 12 M+75
basis points). The idiosyncratic shock consisted of a significant run-off of retail and non-bank
corporate deposits (-5 percent and -10 percent, respectively, spread out over the first month).
On the wholesale side, each banks faced a 100 percent run-off rate of unsecured wholesale
funding from banks and financial institutions; DCM closed for the bank; 80 percent of repos
are rolled over; committed interbank lines are not available for the bank.

155
The withdrawal of parent support is particularly relevant for systems with foreign-owned banks, for example.
 199

Based on the previous considerations, we define severe benchmark scenarios stress testers
could refer to in order to simulate moderate, medium, severe and very severe stress (Table
16). The level of severity of stress is oriented on the past crisis, relative to levels of stress
observed at the times of the Lehman collapse.156 While the Lehman calibration is not to be
understood as being scientific, it is meant to represent the situation of banks hit hard during
the first month after the Lehman collapse, and, very importantly, it is intuitive. Accordingly,
the moderate scenario is one quarter of Lehman crisis conditions, while medium, severe and
very severe are 0.515, 1 and 22 times Lehman. We have labeled the case study in section E
will assess banks against stress conditions equal to moderate, medium, severe and very
severe stress.

156
Please note that this benchmark scenario remains hypothetical, and is geared towards large banks in OECD
countries. For smaller banks, the benchmark could be different, with even higher run-off rates for customer
deposits in case of a name crisis, for example. Expert judgment is needed to design the most plausible scenario
for the situation at hand.
 200

Table 16. Benchmark Scenarios

Moderate Medium Stress Severe Stress Very Severe

Scenario
Stress Scenario Scenario Scenario Stress Scenario
Severity (x times Lehman/1) 0.25 0.5 1 2
Liquidity Outflows
Customer Deposits
Customer deposits (Term) 2.5 percent 5 percent 10 percent 20 percent
Customer deposits (Demand) 5 percent 10 percent 20 percent 40 percent
Wholesale Funding
Short-term (secured) 5 percent 10 percent 20 percent 40 percent
Short-term (unsecured) 25 Percent 50 Percent 100 Percent 100 Percent
0 Percent need 5 Percent need 10 Percent need 20 Percent need
Contingent liabilities
funding funding funding funding
Liquidity Inflows
Haircut for Cash 0 Percent 0 Percent 0 Percent 0 Percent
Haircut for Government
1 Percent 2 Percent 5 Percent 10 Percent
Securities/2
Haircut for Trading Assets/3 3 Percent 6 Percent 30 Percent 100 Percent
Equity: 10-15;
Equities: 3; Equities: 4-6; Bonds (only LCR
Proxies, specific assets Not liquid
Bonds: 3 Bonds: 3-8 eligible ones): 5-
10
Haircut for other securities 10 Percent 30 Percent 75 Percent 100 Percent
Equities: 25; Equity: 30; Bonds
Equities: 10;
Proxies, specific assets Bonds: 20 (some (only LCR eligible Not liquid
Bonds: 10
not liquid) ones): 20-30
20 Percent (or 30 Percent (or 40 Percent (or
Percent of liquid assets 10 Percent (or
actual figure plus actual figures actual figures
encumbered/4 actual figure)
10 ppt) plus 20 ppt) plus 30 ppt)
1/ The Lehman type scenario would correspond to a scenario encountered by banks that were hit
severely during the 30 day period after the Lehman collapse, i.e. a stress situation within a stress period
rather than an average; The scenario has been put together based on expert judgment, using evidence
as available.
2/ The haircut highly depends on the specific features of the government debt held (rating, maturity,
market depth) and can be higher or lower. The figures displayed herein are meant for high quality
investment grade bonds, taking into account recent market conditions. The same applies for the
remainder of the liquid assets. For the securities in the trading book, it is assumed that they are
liquidated earlier, resulting to lower haircuts.
3/ A haircut of 100 Percent means that the asset is illiquid, i.e., the market has closed.
4/ The figures account for a downgrade of the bank, which triggers margin calls, and higher collateral
requirements for generally. Please note that the unencumbered portion applies to a gradually narrower
definition of liquid assets.
Source: Authors.
 201

As a caveat, it should be highlighted that the level of stress to be considered medium, for
example, depends on the specific circumstances and all liquidity stress tests must be adapted
for the specific country, economic situation, and potential vulnerabilities.157 It is strongly
recommended to run a series of stress to assess the sensitivity of system vis-à-vis stress,
including reverse stress to test the threshold that systems can withstand.

J. Case Study

Case Study Implied Cash Flow Analysis

In order to illustrate the mechanics of the tool, we constructed three stylized banks (Table
17), namely “average” banks (i.e., with median asset and liability structures) based in the
OECD, the ECs and LICs as defined in Annex 7.158

Table 17. Implied Cash Flow Case Study—Sample Banks

Bank OECD EC LIC

Assets 100 100 100
Cash 4.2 11.2 13.5
Government Securities 4.1 7.8 8.3
Other Securities 21.4 8.6 6.5
Customer Loans 52.7 56.2 47.1
Loans to Banks 12.4 12.7 18.5
Other Assets 5.4 3.6 6.1
Liabilities & Equity 100 100 100
Demand Customer Deposits 19.8 23.3 39.6
Term Customer Deposits 27.9 41.8 33.2
Short-term wholesale funding 17 11.2 6.6
Long-term Funding 16.7 7.4 2.9
Other Liabilities 12.3 5.1 6.2
Equity 6.3 11.2 11.6
Contigent Liabilities 21.9 17.6 13

Source: Authors

157
“Moderate” is already a substantial stress event in terms of overall stress, but in some countries this could be
less severe as in others, depending on the quality of the safety nets that are in place (particularly the deposit
insurance system).
158
Please note that the sample of banks (the universe of banks in Bankscope) is biased towards banks in
advanced countries, where a broad coverage is achieved. For emerging markets and even more importantly for
low income countries, the sample is biased towards larger institutions and thus not fully representative.
 202

In the first step, we run ICFA based on the reverse stress test type setting (see Table 13 and
Annex 8). We simulate the impact of moderate, medium, severe and very severe stress
conditions as displayed in Table 16, whereby cumulative impact during 5 weeks, made up by
the gradual impact during each of the 5 weeks, equals the figures as displayed in Table 16.159

Figure 44 (left hand panel) the cumulative outcome of the gradual ICFA test after 5 weeks in
terms of net cash inflows as a percentage of total assets. The two panels on the right hand
side provide more information on the drivers and gradual development for the severe test,
i.e., show the evolution of stress during week one to five rather than the cumulative effect.
The test reveals that all three banks survive the moderate and medium shock for the entire
test horizon (i.e., five weeks). However, the severe shock would be too harsh for both the
stylized OECD and EC bank, while the LIC just passes the test. Under very severe
conditions, all banks fail and the liquidity shortfall is substantial. The very severe scenario
simulates an outflow of 30 (LIC) to 35 (OECD) percent of funding, while only cash and
government securities remain to counterbalance, i.e. 20 percent for the LIC banks and less
than 10 percent of assets for the OECD bank.

The gradual effect for the severe scenario (the two panels on the right hand side1) reveals
that the OECD bank runs short of funding during the third week, while the EC banks would
only fail in the last period. The key reason for the weaker performance of the OECD bank is
that it suffers a higher outflow of wholesale funding (12.8 vs. 8.4 and 5; OECD:EC:LIC) and
thereby also in total (21.8 vs. 19.5 (EC) and 16.7 (LIC)), while the inflow of cash through
fire sales is lower due to its comparably lower buffer with respect to high quality liquid
assets.160

159
Hence, it is assumed that 1/5 of the total funding is withdrawn in the first week, another 1/5 in the second,
and so on. We assume that 30 percent of the other securities are in the trading book and that all short-term
wholesale funding is unsecured.
160
As a caveat, it should be noted that it is assumed that government bonds held by banks remain liquid as such,
which could be too benign especially in ECs and LICs., but also that the haircuts are same, which is unlikely to
happen in reality but simplifies the test. Likewise, the run-off are likely to be higher during “typical” bank runs
in ECs and LICs, but
 203

Figure 44. Outcome of Implied Cash Flow Stress Tests for Stylized Banks
25% Severe Scenario: Analysis of Drivers
Bank OECD EC LIC
20% Total Outflow of Funding 21.8 19.5 16.7
Net Cash Inflow (Percent of Assets)

15% Outflow of Deposits 6.8 9.3 10.4

Outflow of Wholesale Funding 12.8 8.4 5.0
10% Change of other funding 2.2 1.8 1.3
5% Total Cash Inflow 11.4 15.3 16.7
Net Cash Inflow -10.3 -4.1 0.1
0%
Moderate Medium Severe Very severe Survival -
-5% Minimum number of Number of Banks
Percent of
periods of survival illiquid
-10% Banks

-15% 0 0 100.0%
1 0 100.0%
-20% 2 0 100.0%
-25% 3 1 66.7%
4 1 66.7%
OECD EC LIC
5 2 33.3%

Source: Authors

K. Case Study Fully Fledged Cash Flow Analysis

To illustrate the mechanics of the fully fledged cash flow analysis we construct a stylized
cash flow template of a random universal bank. In contrast to the previous section, this
example is purely fictional and emphasises the key prerequesite for the usage: access to
granular, bank specific contractual cash-flow data and plausible assumptions for the
behavioural (planned) future cash flows161.

The table below shows an example of the liquidity parameters of Bank A under the baseline
and under a simple stress scenario. Instead of providing the full range of contractual /
behavioral cash-flows, the example is restricted to the aggregate positions in eight maturity
buckets:

 the sum of cash otuflows in each maturity bucket(1)

 the sum of cash inflows in each maturity bucket (2)
 the net funding gap in each maturity bucket (3), equal to: (2)–(3)
 the counterbalancing capacity cumulated across maturity buckets

Under a given scenario (i.e. set of assumptions) Bank A remains liquid without further
(management) action as long as the latter remains positiv, which is the case in all maturity
buckets in the baseline scenario.

The stress scenario takes into account assumptions about a sudden drop in market confidence
combined with an idiosynchratic shock for Bank A162:

161
The assumptions for the behavioral (planned) future cash flows can be replaced by bank data (i.e. funding
plans) for the year ahead.
162
The assumptions were applied symmetrically for in- and outflows.
 204

 Steady decline in market prices for a bank’s assets available for counterbalancing
liquidity gaps from 10 to 30 percent (depending on asset quality),
 Dry up of funding markets, preventing issuance of new bonds,
 Strong decline in secured and unsecured wholesale funding (ranging from 10 to 60
percent over time),
 Increase of NPLs reflected by a decrease in cash-inflows from customer loans (5 to
10 percent over time) and
 About 80 percent of credit lines granted by Bank A are drawn within one month.

Table 18. Outcome of Fully Fledged Cash Flow Stress Tests for Stylized Banks
Bank A baseline 1 Day 7 Days 1 Month 3 Months 6 Months 6-12 Months 12-24 Months >24 Months
Sum of cash outflows 17,800 6,500 5,850 6,850 7,400 9,750 3,750 7,950
Sum of cash inflows 1,875 4,275 8,925 7,200 6,375 5,100 13,000 23,800
Net funding gap -15,925 -2,225 3,075 350 -1,025 -4,650 9,250 15,850
Cumulative counterbalancing
capacity (after HC and net funding 22,925 20,700 19,725 18,875 15,900 8,400 11,350 7,400
gap)
assumptions 1 Day 7 Days 1 Month 3 Months 6 Months 6-12 Months 12-24 Months >24 Months
Sum of cash outflows 20,340 14,860 5,070 5,090 4,500 6,150 3,750 7,950
Sum of cash inflows 1,545 3,525 7,665 5,670 5,055 4,140 11,835 21,585
Net funding gap -18,795 -11,335 2,595 580 555 -2,010 8,085 13,635
Cumulative counterbalancing
capacity (after HC and net funding 12,900 1,393 170 -15 -833 -5,333 -2,445 -3,990
gap)
Source: Authors

As shown in the table the survival period of the bank under the applied assumptions is
reduced to 1 to 3 months, when the increasing funding gap can’t be covered any more by the
bank’s unencumbered reserves.

L. Conclusion

In this paper we have argued that liquidity risk has–unjustifiably–flown under the regulatory
radar with the advent of the Basel I framework and its focus on banks’ capitalization.
However, the fact that liquidity risk turned out to be one of the key threats to financials
stability throughout the recent financial crisis, lead to reconsideration, with a reemerging
focus on liquidity in industry as well as regulatory circles.

The purpose of this paper and the tool presented therein reflects this development and aims at
providing stress testers with a flexible and easy-to-use platform to assess the liquidity
situation of banks top-down from different angles. The pre-defined tests can easily be
adapted to bank-specific situations and/or specificities of banking systems to be assessed. A
key objective was striking a balance in terms of data requirements and stress test
sophistication, allowing for tests with parsimonious data on the one hand and more complex /
demanding tests on the other.

While the obvious way to stress test liquidity is the use of cash flow data, it is often not
available (yet) at regulatory/supervisory institutions. One of the main contributions of this
 205

paper consists in providing input templates for cash flow based tests that could also serve
regulators/supervisors as a first step towards fully-fledged cash flow analysis based on a
regular data collection from banks. Once available, the cash flow module allows simulating
detailed funding structures of single banks, which enables to draw some broader conclusions
for the system wide situation of banks and potential contagion effects, respectively.
Moreover, the presented tool allows for easy peer comparisons that should always play an
important role for liquidity stress tests and can readily reveal vulnerabilities. Finally, the
paper contributes to existing work on liquidity by modeling the link to solvency stress (tests)
explicitly. Although this should not be misinterpreted as the final solution to this highly
complex problem, the inclusion of a module in the tool to account for this link in an easy to
use fashion should facilitate practitioner’s work significantly.

Future research will focus on better understanding the link between banks’ solvency and
liquidity strains. Both are inherently interrelated, and stand-alone stress tests that only
examine either solvency or liquidity stress testing, potentially risk producing downward
biased results. For example, a bank’s severe funding strain could swiftly mutate into
solvency concerns with the market putting pressure on the bank to increase its capital. The
focus in this paper has been predominantly to analyze the link from solvency to stress testing
but the feedback loop can also originate with liquidity.
 206

Annex 6. Reviewing Liquidity Issues during the Financial Crisis

Investment banks were among the first ones to experience liquidity shortages, due to their
funding mix that relied heavily on the wholesale market as well as interconnectedness in the
financial system that led many banks to start hoarding liquidity during the systemic crisis
episodes because of counterparty risks.163 For the same reason164, namely uncertainty on the
solvency conditions, funding also drained for banks that had aggressively collected deposits
before the crisis, such as Icelandic banks (see Ong and Čihák, 2010) and the systemically
important financial institutions (SIFIs) that were heavily exposed to securitized products and
asset-backed commercial paper (ABCP) funding. While the latter banks were subject to
wholesale bank runs, the most prominent recent “victim” of a (pure) liquidity squeeze was
Northern Rock, which was subject to a classical bank run with customers queuing to
withdraw their money after a wholesale funding run and emergency liquidity assistance from
the Bank of England (see Section D and Annex 9 for an outline of major recent liquidity
shocks).

The liquidity squeeze particularly affected the most sensitive liquidity channels, namely
unsecured cross-border funding as well as foreign currency swaps in countries as diverse as
Australia, Korea, and Kazakhstan. Foreign currency lending (and thereby funding) played an
important role in Central and Eastern Europe, for instance in Hungary and Poland, where
banks increasingly used foreign exchange swaps to fund their domestic lending activities.
With the unfolding of market turbulences in international money markets after the Lehman
Brothers demise U.S. dollar funding dried up. The situation for Euro funding was less
precarious as the foreign-owned subsidiaries and branches refinanced their Euro exposure
largely via their parent banks.

Below, we will review in more details the events and channels of contagion. The financial
crisis in general and liquidity problems more specifically began with the deteriorating quality
of U.S. subprime mortgages, a credit, rather than a liquidity event. A wide range of different
financial institutions had exposures to many of the related mortgage-backed securities, often
off-balance sheet entities such as conduits or structured investment vehicles (SIVs). The
SIVs or conduits were funded through the issuance of short-term ABCP in order to take
advantage of a yield differential but resulting in a maturity mismatch. Due to the increasing
uncertainty with regard to their exposure to and the value of the underlying mortgage-backed
securities, investors became unwilling to roll over the corresponding ABCPs (IMF, 2008,
2010; Frank et al, 2008).

163
Demirguc-Kunt and Huizinga (2009), for example, found that business models that rely heavily on non-
deposit short-term funding and non-interest income appear to be riskier in terms of liquidity.
164
A CEBS report on lessons learnt from the crisis (Committee of European Banking Supervisors 2009a) found
that the majority of the EU cross-border banking groups that faced severe idiosyncratic liquidity problems were
also subject to substantial doubts concerning their solvency or even insolvent, e.g. the Icelandic banks.
 207

As the problems with SIVs and conduits deepened, banks came under increasing pressure to
rescue those that they had sponsored by providing liquidity or by taking their respective
assets onto their own balance sheets. As a result, the balance sheets of those financial
institutions were particularly strained by this reabsorption, which in addition was amplified
due to declining asset values and the evaporation of market liquidity in structured products.
A further strain on banks’ balance sheets came from warehousing a higher than expected
amount of mortgages and leveraged loans, the latter usually passed on to investors in order to
fund the highly leveraged debt deals of private equity firms. Both the market for mortgages
and leveraged loans dried up from the collapse of transactions in the mortgage-related
securitization market and collateralized loan obligations (CLOs). Banks also felt obliged to
honor liquidity commitments to alternative market participants, such as hedge funds and
other financial institutions that also suffered from the drain of liquidity. With regard to
alternative channels of liquidity provision, stress in the FX swap markets and the negative
reputational signal resulting from using the Fed discount window limited options further.

Consequently, the level of interbank lending declined both for reasons of liquidity and credit
risk. In addition, the money market disruptions at the beginning of the crisis (August 2007)
led to a general shift to a more conservative risk tolerance. Before the onset of the crisis,
banks had relied on a (perceived) “insurance function” of unsecured money markets against
negative liquidity shocks. As this perception evaporated as quickly as market depth and
breadth of the unsecured money markets, banks shifted to self-insurance (liquidity hoarding).
Subsequently, money markets were severely affected especially in advanced countries in the
form of lower supply of liquidity, a shortening of tenors, and a widening of the Libor–
overnight index swap (OIS) spreads, which in turn led to increased funding costs. Some
banks were shut-out of the market completely. The funding situation of many banks was
exacerbated by a deliberate shortening of funding maturities of many banks in the first phase
of the crisis. Spreads over mid-swap increased for banks. Hoping that the situation would
improve, many banks postponed issuance.

When debt capital markets closed for banks in late 2008, many banks had accumulated a
substantial issuance back-log. Finally, the evident deterioration of market and funding
liquidity conditions had implications with regard to the solvency position of banks for several
reasons. First, financial institutions saw a decline in the values of the securitized mortgages
and structured securities on their balance sheets, which in turn resulted in extensive write-
downs. The drying up of many of these markets and the built-in leverage of many of the
products, increased the uncertainty with respect to their valuation and consequently with
respect to the solvency situation of many banks. Second, funding liquidity pressures forced
rapid deleveraging during this period, further depressing asset prices. Third, funding costs
increased due to rising money market and debt capital market spreads, which was amplified
by the fact that many financial institutions had become increasingly reliant on funding from
wholesale money markets. Jointly, these pressures with the key role that solvency concerns
 208

played resulted in a decline in the capital ratios throughout the banking sector, and as a result
of which credit default swap (CDS) spreads increased significantly across the industry during
the crisis. At the same time, increased uncertainty with respect to asset quality and valuation
led investors to raise the bar for banks; before the onset of the crisis a core-tier 1 ratio of 6
percent was generally considered sufficient. During the crisis that requirement increased to
10 percent. (“10 is the new 6”).

Re-enforcing liquidity spirals and a re-pricing of risk occurred when, on the one hand,
market illiquidity turned into funding illiquidity, such as when the French bank BNP Paribas
announced in August 2007 it would refuse to accept withdrawals from three of its investment
funds. Funding illiquidity also led to market illiquidity, when for instance, European banks in
late 2007 required dollar funding through foreign exchange swaps, but due concerns over
counterparty credit risk, liquidity, typically obtained in the underlying swap market dried up.
The collapse of Lehman Brothers in September 2008 was then the watershed event that
caused a near break-down of secured and unsecured interbank and debt capital markets with
sharply increasing counterparty risks and banks hoarding liquidity (in reaction to increased
funding liquidity risk and tightening risk tolerance), haircuts and dollar shortages as well as
led to rapid spillovers to emerging market countries, and soaring uncertainty across asset
markets.

These liquidity spillovers have been facilitated by recent structural changes in the financial
markets and by financial innovation during the last decade. In this context, banks have
become increasingly reliant on wholesale funding and short term liquidity lines. Also,
increased complexity of securities has led to great information asymmetries among market
participants. Favorable macroeconomic conditions, especially low interest rates in recent
years, have increased investors’ risk appetite and the demand for high yield products in order
to satisfy profit margins. Finally, increased correlations between returns of differing asset
classes due to algorithmic trading, such as by quantitative hedge funds, has heightened the
vulnerability with regard to the transmission of illiquidity.

In any case, the vast availability of underpriced liquidity in the pre-crisis period and the
eventual evaporation of funding liquidity with the onset of the subprime crisis in the summer
of 2007 proved challenging for many financial institutions. Solvency risks165 quickly
morphed into liquidity risks and vice versa in some cases, even though many of the rescued
banks surpassed minimum regulatory capital requirements, as funding did not only become
more expensive, but key funding markets closed entirely.166
165
The simultaneous drying up of market liquidity in some asset markets (i.e. ABS) lead to increased
uncertainty with respect to the solvency of institutions with high exposures in these asset classes.
166
In fact, the link between solvency and liquidity is highly complex, with effects working both ways. Banks
with solvency problems are natural candidates to run into liquidity traps (once markets dry up), but solvent
banks can, under certain circumstances, also be hit by strains on the liquidity side.
 209

Additional bank runs were prevented with the help of policy measures such as an increase of
the level of deposit insurance, providing banks with access to central bank funding (i.e.,
through a lender of last resort) and by guaranteeing interbank market exposure.167
Nevertheless, two years after the beginning of the financial crisis bank funding remains
problematic, subject to elevated costs and/or dependent on public support measures, with
confidence not yet re-established (see GFSR, April 2011). Recently, banks based in
peripheral Europe have become highly dependent on funding by the European Central Bank
(ECB)—one of the signs that interbank funding markets have not yet fully recovered since
the onset of the crisis.168

With the underpricing of liquidity risk prior to the crisis, a return to the same pre-crisis
liquidity pattern is not expected. Furthermore, there is widespread consensus that banks’ pre-
crisis extensive reliance on deep and broad unsecured money markets is to be avoided in the
future (and in current market conditions there is no appetite for that anyway). Creating
substantial liquidity buffers across the board (i.e., to be followed by all banks) is the explicit
aim of a number of regulatory responses to the crisis, such as the CEBS Guidelines on
liquidity buffers (CEBS 2009b) and the Liquidity Coverage Ratio (LCR), one of the two new
Basel III liquidity standards. As time goes on, liquidity management needs to be prepared for
the materialization of tail risks, which is, the simultaneous closure of various funding and
assets markets and the tapping of off-balance sheet positions—highly positive correlations as
in the case of solvency risks.169

167
An overview of market interventions during the financial crisis and their effectiveness can be found in IMF
(2009).
168
Some Irish banks, for example, suffered a silent deposit run over a period of a few months when large
corporate clients withdrew deposits.
169
See ECB (2008b, pp. 35) for empirical evidence on the short-comings of banks’ liquidity stress tests exposed
by the crisis.
 210

Box A6.1 Regulatory Initiatives to Stress Test Liquidity Risk

A. Basel III
 Basel III takes into account lessons learnt during the crisis, namely that liquidity risks can trigger
solvency problems in banks and vice versa, as illustratively shown at the example of U.S.
investment banks. Basel III (BCBS 2010b) is based on two minimum standard ratios for funding
liquidity, namely the LCR and the NSFR. The former is meant to put banks into a position to
withstand sudden funding stress for one month, while the second ratio attempts to limit the
maturity mismatch conditional on banks’ asset composition, and over-reliance on short-term
wholesale funding in particular. In addition to that, the BCBS has published basic principles of
liquidity management, essentially guidance on the risk management and supervision of liquidity
risks for banks and supervisors (BCBS 2008). Additional analysis on liquidity risk in broader
terms has also been undertaken by the Committee of the Global Financial System (CGFS).

 Both ratios are introduced with a transitional observation period, in order to provide banks with
time to adjust, and subject to additional revisions. For the LCR, the observation period beings in
2011 and the ratio will be introduced in 2015 (BCBS 2010a). The NSFR is scheduled to be
introduced in 2018, with a monitoring period starting in 2011. A recent Quantitative Impact Study
(QIS 6) revealed average LCRs of 83 percent for Group 1 banks and 98 percent for Group 2
banks, based on data from 23 countries, with 46 percent of the banks meeting the standard (BCBS
2010c). For the NSFR, 43 percent of the banks met the standard and the averages of Group 1 and
2 banks were at 93 percent and 103 percent, respectively. On average, European banks
underperformed the other banks (CEBS 2010). The main conclusion from the QIS is that the
liquidity risk exposure and the liquidity risk bearing capacity of banks differed widely across the
international sample, depending on their maturity profile (for the NSFR), the portion of stable
funding (customer deposits) and liquid assets, respectively.

 Basel III also enforces monitoring activities for liquidity risk, with a focus on contractual maturity
mismatch, concentration of funding, the availability of unencumbered liquid assets, currency-
specific liquidity assessment (through LCR) and market-related monitoring tools (2010b).

B. U.K. Liquidity Regulation Adopted in October 2009

In 2009, the then U.K. Financial Services Authority (FSA) introduced revised liquidity standards
(Policy Statement 09/16). In November 2010, however, the FSA announced that it will reconsider the
calibration of the standards with a view not phase in its own liquidity rules unilaterally, but the
implementation of the qualitative elements is under way. The new standard includes the following
elements and could be treated as general guidelines for further evolutions in terms of liquidity
management:

 Improved control and system requirements for sound liquidity risk management.
 Adequate liquidity and self-sufficiency.
 Stricter stress testing scenarios including short- and long-term stress scenarios.
 Individual liquidity guidance to each firm.
 Comprehensive and detailed examination of contracts (e.g., maturity buckets, asset type, or
currency).
 New definitions of liquid assets and risk-based buffers.
 Granular and frequent reporting requirements (daily, weekly, monthly, quarterly).
 211

Annex 7. Cross-Country Funding Pattern

Figure A7.1 shows the average170 asset (left hand side) and liability (right hand side)
composition of banks in OECD countries171, emerging market countries (ECs) and low
income countries (LICs) based on the end 2010 situation172 for the universe of banks
available in Bankscope.173 The asset side includes the off-balance sheet items (mostly
guarantees, committed credit lines and other contingent liabilities), whereby the total is above
100 percent.174

The graphs show that the portion of long-term customer loans on the banks’ total business is
about 50 percent, slightly higher for ECs (56 percent) and OECD banks (53 percent) and
somewhat lower for banks in LICs (47 percent) and that total loans (customer loans and loans
to banks) account for about two thirds of the balance sheet in all three regions. The main
differences show up in terms of the composition of cash and securities held by banks: in
terms of the most liquid assets (cash and government securities), LICs are best off (24
percent), followed by the ECs (19 percent), while OECD banks exhibit only 8 percent on
average. However, banks in OECD countries hold more than 21 percent other securities,
which can be used for fire sales provided that they are liquid. In sum, banks hold the same
portion of securities which can, in principle be sold or pledged to generate liquidity (30
percent). In terms of off-balance sheet items, the OECD countries lead (22 percent) by a low
margin, driven by a few countries with very substantial off-balance sheet items.

On the liability side, the differences are more pronounced, as the reliance on deposits is far
higher in the ECs (65 percent) and LICs (73 percent) than in OECD countries (48 percent),
while the opposite is true for wholesale funding, both short- (OECD: 17 percent: EC: 11
percent; LIC: 7 percent) and long-term (OECD: 17 percent: EC: 7 percent; LIC: 3 percent).
Through the long-term funding, OECD banks make up a noteworthy portion of the gap in
terms of deposit funding compared to ECs and LICs. It is also shown that banks in ECs and
LICs are substantially less leveraged than banks in OECD countries, which increases the gap

170
In the first step, the unweighted average was computed for each country and then the unweighted average for
the three country types.
171
The figures for the OECD country banks are closely in line with a study by the BCBS/FSB (“An assessment
of the long-term economic impact of stronger capital and liquidity requirements”, August 2010) for a sample of
6,600 banks in 13 OECD countries for the period from 1993 to 2007.
172
For most banks, the data were from end 2010, otherwise mostly from 2009.
173
It should be noted that the data contains a higher portion of banks in OECD countries, reflected by the fact
that the median OECD bank is smaller than the median EC bank (but larger than the median LIC bank).
174
The total percentage of all assets is 100 percent for the on-balance sheet items plus the off-balance sheet
positions as a percentage of total assets.
 212

once more. The ultimate question becomes how the composition of funding sources related to
the maturity profile, which is where maturity gap analysis, the second test category, comes
into play.

Figure A.7.1. Composition of Assets (left) and Liabilities (right) for banks in OECD
countries, ECs and LICs
OECD Banks (n = 12,400) OECD Banks (n = 12,400)
4.2%
4.1% 6.3% Demand deposits

Cash and Cash-like Term Deposits

21.9% 19.8%
Government Securities 12.3%
5.4% 21.4% Short-term wholesale
Other securities
funding
Customer Loans
12.4% 16.7% Long-term funding
Loans to Banks
27.9%
Other assets Other Liabilities
52.7%
Offbalance 17.0%
Equity

Emerging Market Banks (n = 3,000) Emerging Market Banks (n = 3,000)

Demand deposits

11.2% 7.8% Cash and Cash-like 5.1% 11.2%

17.6% Term Deposits
3.6%
Government Securities 23.3%

8.6% Other securities Short-term wholesale

7.4%
12.7%
funding
Customer Loans
11.2% Long-term funding
Loans to Banks
Other assets Other Liabilities
56.2% 41.8%
Offbalance
Equity

Low Income Country Banks (n = 500) Low Income Country Banks (n = 500)
Demand deposits

6.1% 13.0% 13.5% Cash and Cash-like 11.6% Term Deposits

Government Securities 2.9% 6.2%
8.3%
Other securities 39.6% Short-term wholesale
6.5% 6.6%
funding
18.5% Customer Loans
Long-term funding
Loans to Banks
Other assets Other Liabilities
47.1% 33.2%
Offbalance
Equity

Source: Authors based on universe of Bankscope data for last available date (mostly end
2010). Please note that the sample of banks is biased towards advanced countries, and for
emerging markets and low income countries tends to capture the larger banks only.
Note: All figures are relative to total assets; the asset side figures include off-balance sheet
items, whereby the total is above 100 percent
 213

Annex 8. Details on all Modules of the Stress Testing Framework

Implied Cash Flow Analysis (ICFA)

Overview

The global financial crisis has shown that a deposit run on a seemingly non-systemic
financial institution such as Northern Rock can have serious implications not only
domestically but also in cross-border terms. The ICFA module allows simulating a run on
bank funding sources, both on wholesale deposits (in case of a name crisis and/or a general
confidence crisis) as well as on retail and wholesale deposits (in case of a severe name crisis
or a systemic banking crisis). This stress testing component extends Čihák (2007) by
allowing for greater granularity on the asset and liability side of banks’ balance sheets. The
test pays particular attention to defining which assets remains liquid, and defining the level of
market liquidity under stress, i.e., setting haircuts. By allowing for a gradual withdrawal of
funds during 5 periods (e.g., days, weeks or months), the module allows assessing the point
where specific banks and the system more generally become illiquid in a reverse test type
manner; The module also contains a similar, cumulative liquidity test for 30 days/3 months,
and keeps in line with Basel III by allowing running a (simplified) Basel III LCR test.175

Assumptions

In the assumptions’ worksheet stress testers have to specify which asset categories remain
liquid, the level of haircuts, if applicable, as well as the portion of assets that are
unencumbered under stress. The liabilities are divided into demand and time deposits (both
further broken down into retail/wholesale and domestic/foreign currency with the option to
break down even further) as well as wholesale short-term and long-term funding on the
interbank market. The user has to input assumptions for the percentage of deposit
withdrawals and other funding sources per period (e.g., a day, week, month) in the 5
day/other period (30) day test. There is flexibility to decompose the deposit categories
according to specific data availability.

The user can decide whether the assumptions apply to all banks uniformly. In the other case,
there is a switch button to “bank-specific” from “market-wide/ uniform” in the results’
worksheet that allows the user to manually input the assumptions for the asset and liability
side of each bank. Please also note that funding withdrawal does not assume an explicit
policy reaction but they can be implicitly incorporated by, for instance, modifying the level
of haircuts and eligibility of unencumbered securities. Finally, any deposit withdrawal would
also lead to a release of the according required reserves. It is advisable to include the required

175
It should be noted that running the LCR test requires granular data and/or expert judgment.
 214

reserves separate from cash and balances with the central bank on the asset side and then
proportionally to the assumed deposit run-off rates include a release of required reserves with
a haircut.176

For the LCR test, the user can make assumptions for the withdrawals of (less) stable deposits,
which are defined as minimum ratios. While the other ratios are pre-defined, the tool allows
simulating different parameter levels to test the sensitivity of the system.

Results

The results’ worksheet for the 5 period test provides the liquidity situation of each bank
before the stress test. It then lists for each of the 5 periods the outflow of funding (by broad
categories), the cumulative impact, as well as the inflow from fire sales of liquid assets for
the first period. For each day, the change in net cash flows since the beginning of the test is
computed, and most importantly, whether any bank becomes illiquid. The ICFA module
therefore allows for an explicit examination whether and how long any bank can withstand a
shock. Similarly, in the 30 day/3 months deposit run, the ex-ante liquidity position of each
bank is listed, and then the outflow of funding and inflow of assets, the change in net cash
flows and whether banks become illiquid. For all tests, a potential shortfall of funding is
computed.

The designed LCR test already shows which banks are likely to be below the required ratios.
A fully-fledged test requires comprehensive data, though. The outcome of the LCR test
reveals the stock of high quality liquid assets, as well as potential cash outflow and cash
inflows, and thereby the ratio of liquid assets (i.e. cash inflows to cash outflows). If the ratio
is above 1, a bank is considered “safe”, whereas the opposite is the case for ratios below 1.
Again, a potential liquidity shortfall is calculated for each bank and the system as a whole,
together with the LCR ratio(s).

Deposit runs are usually rare but when they happen they can cause large damages to the
affected banks, their depositors, the financial system and its reputation. In the case of
Northern Rock, the deposit run even exacerbated an already fragile financial system. The
ICFA test is flexible enough to allow for different types of deposit runs (retail versus
wholesale) and for a run on foreign deposits if currency mismatches play an important role in
the banking system. The level of detail on both the asset and liability side makes the ICFA
test more realistic especially since there are often differences among banks, and asset
categories exhibit different liquidity levels and haircuts.

176
For example, if the average level of required reserves is 10 percent and the average deposit run-off rate is 20
percent, then this would amount to a freeing of required reserves of ca. 20 percent so a haircut of 80 percent
could be used.
 215

Maturity mismatch and liquidity rollover stress testing

Overview

The global financial crisis has shown that many of the failed financial institutions suffered
from a liquidity maturity mismatch caused by very short-term (wholesale) funding (and
thereby a sizeable duration gap), making them vulnerable to a loss of confidence and
counterparty credibility and an eventual liquidity squeeze. The liquidity gap analysis module
matches liability and asset maturities and identifies liquidity gaps at each maturity bucket and
under different scenarios. There are three different types of tests: (a) a static maturity gap
analysis of each bank and the overall banking system; (b) a static maturity gap analysis
taking into account rollover risks; and (c) a dynamic maturity gap analysis taking into
account rollover risks. In the latter case potential liquidity gaps can be closed by free liquid
assets at lower maturities (if available). This liquidity stress testing module allows for a clear
examination of the funding structure by maturity buckets of each individual bank and the
overall system. The module also implements the Basel III NSFR test.

Assumptions

The assumptions’ worksheet differentiates between the liabilities (overall liabilities and more
granular buckets, such as interbank and long-term debt, if available) by maturity that cannot
be rolled over (on a continuous basis) and the various asset categories. For the assets, the user
has to input the level of illiquidity of each asset category and the asset-specific haircut for a
fire sale to cover potential liquidity gaps at specific maturity buckets. There is also an
assumption on the portion of loans that are reinvested when maturing.

The user can decide whether the assumptions apply to all banks uniformly. In the other case,
there is a switch button to “bank-specific/manual” from “market-wide/ uniform” in the
results’ worksheet that allows the user to manually input the assumptions for the asset and
liability side of each bank.

As all tests the liquidity rollover tool is designed for a non-intervention of central bank
liquidity. But the assumptions’ worksheet enables the user to specify the additional available
central bank and intra-group funding (as percentage of bank liabilities) that could be used ex-
post the stress test to cover part or the entire liquidity shortfall.

Finally, no explicit assumptions on the parameter have to be made for the NSFR, but weights
different from the pre-defined ones could be simulated to assess sensitivity.
 216

Results

The results’ worksheet first provides a descriptive (static) maturity mismatch analysis
showing the total system and bank liquidity shortfall by maturity in amounts and in the
number of banks that experience a possible shortfall in each maturity bucket. Second, the
liquidity tool becomes dynamic and allows for free and liquid assets to close liquidity gaps in
other maturity buckets. Suppose that bank X has a liquidity shortfall in maturity bucket “Due
within 1 to 3 months,” then excess liquidity in the maturity bucket “Demand” is
automatically allocated to the shortfall maturity bucket. If not sufficient, then excessive
liquidity from the next available maturity will be used until all liquidity gaps are closed or in
the worst case, a shortfall position for a bank can be discerned. Ultimately, the test assesses
up to what horizon banks remain liquid, and different thresholds to pass the tests can be set
by the stress tester.

With the assumption on additional (collaterized) central bank funding (by percentage of
liabilities), the user can specific how much central bank liquidity would be needed to close
the liquidity gap by bank and system. Alternatively, the user can use common benchmarks
from past liquidity crises domestically (or cross-country) or central bank regulations to
discern the impact of central bank funding on alleviating the liquidity shortfalls.

The NSFR test to examine a structural maturity mismatch first calculates the available stable
funding and then the required stable funding based on the different categories from the Basel
III proposal followed by the ratio for each bank and the system and the according shortfall/
surplus.

Cash Flow Data based Stress Testing

Overview

A key prerequisite to carry out cash flow based liquidity tests is access to a wide range of
data. Even though Basel III requires a maturity mismatch approach to liquidity monitoring in
the future, only few jurisdictions already have such a monitoring tool in place.177 The
difference between cash flow tests run by banks and those run by authorities for monitoring
purposes is that the latter requires standardized templates, which then allows simulating the
impact of common shocks based on a uniform method.

The input data consist of contractual gross cash-flows in various buckets of residual
contractual maturities (e.g. 1 day, 1 week, 1 month, 3 months, 6 months, 12 months, 2 years

177
Given the implementation of Basel III via CRD IV framework in the European Union, uniform cash-flow
templates for liquidity reporting / stress testing are likely to become a standard in other jurisdictions as well.
 217

and more than two years). In addition, stock data is required for many items. The definition
of items is usually tailor-made for liquidity stress tests and does not necessarily mimic
traditional accounting / supervisory information.

The cash flow template in this paper / tool is structured in three broad categories: cash-
inflows, cash-outflows, and the counterbalancing capacity (including stocks of liquid assets
and haircuts). The template distinguishes between contractual in- and outflows178 which are
already fixed and behavioral cash flows which cover the expected cash flows banks use for
their liquidity planning. Figures should be provided at the consolidated level (or sub-
consolidated level for local subsidiaries of foreign banks). In order to enhance usability and
for usage of the cash flow template as a monitoring tool, the user can always check the CF
monitoring tabs that provide a bank specific overview on structure and the funding situation
itself. If foreign currencies play an important role for a banking system, the template can be
duplicated and submitted for all other significant foreign currencies.179

As is the case for all risk analysis, plausibility and robustness checks are required. In this
context, the collection of comprehensive data provides a check on the quality of liquidity risk
management at the banks in the jurisdiction and their compliance with the BCBS Principles
of sound liquidity risk management and supervision.

The following table provides a detailed description of the default items within all three broad
categories of the cash flow module:

178
The template does not include the non-financial business related cash-flows, for example, from wages,
facility management (office rents) and similar items.
179
Given the variation in business models and activities across banks, a standardized template implies that
banks only have to provide data on items and currencies in which they are exposed to material liquidity risk.

Table A8.1. Cash Flow Items
Cash-flows
Contractual Outflows
Own issuances due
Unsecured wholesale funding due from
non-financial corporates
Unsecured wholesale funding due from
financial corporates
Unsecured wholesale funding due from
financial institutions
Unsecured wholesale funding due from
government/ public entities
Unsecured wholesale funding due from
institutional networks
Secured wholesale funding due, secured
by sovereign debt 0 percent r/w
Secured wholesale funding due, secured
by sovereign debt 20 percent r/w,
covered bonds up to AA-, non-financial
corporates
Secured wholesale funding due, secured
by equity
Secured wholesale funding due, secured
by other instruments
218
The following positions contain all contractual (already
fixed) and behavioural (expected) Cash Outflows. If a
position that has a material liquidity risk is not covered by
the predefined outflow items, they have to be aggregated
within position for other cash flows.
All positions have to be split according to their
contractual maturity into the corresponding buckets.
Within the contractual Cash Flows no rollover of existing
liabilities is assumed to take place. Behavioural
(expected) cash flows reflect the banks funding plans for
the following 12 month period.
Symmetry should always be obeyed and there should be
no double counting of cash flows in the template at all.
This section contains all contractual (already fixed)
outflows split into 8 maturity buckets. For contractual
cash flows the stock value should always equal the sum
of all maturity buckets (since they cover an infinite time
horizon)
This position refers to the outflow of maturing
commercial papers (bonds, private placements, CDs,
FRN, etc.) issued by the bank itself. The outflows also
contain the principal the bank has to pay periodically.
They have to be split according to their contractual
maturity into the corresponding buckets.
These wholesale positions refer to outflows resulting
from maturing liabilities from various entities that are not
secured by repos or similar. They usually differ from
retail deposits in deposit size, higher concentration and
higher professionalism of the counterparties.
Financial institutions are Monetary Financial institutions
(MFIs) excluding central banks (predominantly banks
and credit institutions) while financial corporates refer to
other undertakings that provide financial services (e.g.
insurance).
The line for Institutional networks is only relevant for
credit institutions that are part of a tiered sector structure
(e.g. cooperative banks with an apex institution) and
should be used to reflect deposits placed by other
members of that network.
In contrast to the items above this section refers to
outflows from maturing liabilities that are secured (e.g.
repos, etc.). The items are split according to the quality
of the security instruments used.
A repo will create a cash outflow at its maturity date;
correspondingly the security which has been repoed out
will enter as a positive value in the maturity bucket in
which the repo transaction matures in the
Counterbalancing Capacity (security inflow).
Risk weights for asset classes follow the Basel II
standardized approach.

Table A8.1. Cash Flow Items (continue)
Repos due with central banks
Retail (incl. SME) funding due, term
deposits
Retail (incl. SME) funding due, demand
deposits
Outflows from derivatives (other than FX-
Swaps)
Outflows from maturing FX-Swaps
Other contractual outflows
Behavioral Outflows
Expected new loans
Expected new financial investments
219
Outflows resulting from maturing open market operations
with Central Banks are included in this section. The
security which is repoed out will enter the corresponding
unencumbered CB eligible collateral position in the
Counterbalancing Capacity in the same maturity bucket
(security inflow). Rollover is not assumed to take place
here.
Contractual retail and SME deposit maturities should be
reported in this section.
Term deposits are commonly reported according to their
maturity buckets, demand deposits should be reported in
the "1 Day" maturity bucket.
Outflows from derivatives refer to contractual flows only.
The Outflows from maturing FX Swaps (including also
cross currency swaps) result from the rescindment of the
FX Swap at the end of its maturity. The Cash Flows have
to be split according to the currencies with material
liquidity risk. (e.g. If USD outflows are swapped for EUR
inflows create an EUR outflow and a USD inflow at
maturity)
This position is intended to include all other contractual
outflows that do not match the description for the
predefined items above, but have an impact on liquidity
risk. Typically this would be significant outflows like
dividends or tax payments (no operating expenses)
Due to the fact that a serious estimation of expected in-/
outflows can't be quantified for an indefinite time horizon
the behavioural section covers only the maturity buckets
up to the next 12 months.
Assumptions should always be conservative and reflect
the current macroeconomic conditions as well as the
experience of the bank in former time periods. Ideally
they are based on solid statistical evidence. The data
would usually stem from the banks’ business plans for
the next year.
This position should reflect a modelled / planned
estimation of outflows resulting from all new loans the
bank is going to grant within the next 12 months
(Wholesale and Retail/SME). This estimation should be
conservative and should be in line with former periods.
If a bank plans to invest in financial assets (e.g. bonds).
If these will be a part of the counterbalancing capacity,
they have to be reflected also in the corresponding line
of that section, so that the final position of the cumulated
counterbalancing capacity remains unchanged (security
inflow).

Table A8.1. Cash Flow Items (continue
Expected outflow and stock volume of
undrawn committed credit/liquidity lines of
financial institutes (incl. SPVs)
Expected outflow and stock volume of
undrawn committed credit/liquidity lines of
others (corp., gov, etc.)
Expected outflows from new FX-Swaps
Expected other behavioral outflows
Contractual Inflows
Maturing loans to financial institutions
Other maturing loans (including
installments)
paper in own portfolio maturing
Reverse repos, secured by sovereign
debt 0 percent r/w
Reverse repos, secured by sovereign
debt 20 percent r/w, covered bonds up to
AA-, non-financial corporates
Reverse repos, secured by equity
Reverse repos, secured by other
instruments
Inflows due to maturing FX-Swaps
Inflows from derivatives (other than FX-
Swaps)
Other contractual inflows
220
The stock value should reflect the sum of all committed
lines the bank has granted. Banks should report the
maximum exposure to an (unconsolidated) SPV based
on the SPV's current debt maturities. The outflows
should reflect conservative expectations of the lines that
are going to be drawn over the next 12 months and
should take into account macroeconomic conditions,
past experience and statistical assumptions.
This covers the outflows due to new FX Swaps a bank is
expecting within the next 12 months. The Cash Flows
have to be split according to the currencies with material
liquidity risk. (e.g. If USD outflows are swapped for EUR
inflows create an EUR outflow and a USD inflow at
maturity)
All other outflows that the bank expects to happen in the
next 12 months that do have a material impact on the
liquidity situation and that do not fit into a category above
should be included here.
This section contains all contractual (already fixed)
inflows split into 8 maturity buckets. For contractual cash
flows the stock value should always equal the sum of all
maturity buckets (since they cover an infinite time
horizon)
This position covers all inflows that result from
contractual (already fixed) credit claims split by financial
institutions (interbank deposits) and all other entities.
This line covers inflows that result from maturing papers
and should also include principal from marketable
securities held by the reporting institution. There should
be no double-counting of inflows reported on other lines.
If the maturing paper influences the counterbalancing
capacity it has to be also subtracted in the corresponding
CBC maturity bucket (security outflow).
A reverse repo will create in inflow at maturity date.
Reverse repos should be booked with the cash inflow at
maturity in these lines corresponding to the risk weights.
Risk weights for asset classes follow the Basel II
standardized approach. The counterbalancing capacity
will be reduced by the corresponding amount (security
outflow).
The Inflows from maturing FX Swaps (including also
cross currency swaps) result from the rescindment of the
FX Swap at the end of its maturity. The Cash Flows have
to be split according to the currencies with material
liquidity risk.
The inflow of derivatives refer to contractual flows only
This position is intended to include all other contractual
inflows that do not match the description for the
predefined items above. (e.g. sale of a business unit)

Table A8.1. Cash Flow Items (continue)
Behavioral Inflows
Expected new debt issuances
Expected new retail deposits
Expected new secured wholesale funding
Expected new unsecured wholesale
funding
Expected inflows due to new FX Swaps
Expected other behavioral inflows
Counterbalancing Capacity -
contractual and behavioural Security
Flows
221
Due to the fact that a serious estimation of expected in-/
outflows can't be quantified for an indefinite time horizon
the behavioural section covers only the maturity buckets
up to the next 12 months.
Assumptions should always be conservative and reflect
the current macroeconomic conditions as well as the
experience of the bank in former time periods. Ideally
they are based on solid statistical evidence. The data
would usually stem from banks’ funding plans for the
next year.
This position refers to the expected inflow created by
new placements of debt / own issuances a bank is
planning within the next 12 months.
The expected funding by retail and wholesale deposits
within the next 12 months has to be conservative and
should be based upon the expected macroeconomic
conditions and should reflect experience from former
periods.
Inflows from expected new repo transactions refer to
secured wholesale funding. If a repo causes also an
outflow in commercial papers the corresponding
positions within the expected outflow section and
position within the counterbalancing capacity have to be
adjusted appropriately.
This covers the inflows due to new FX Swaps a bank is
expecting within the next 12 months. The Cash Flows
have to be split according to the currencies with material
liquidity risk. (e.g. If USD outflows are swapped for EUR
inflows create an EUR outflow and a USD inflow at
maturity)
All other inflows that the bank expects to happen in the
next 12 months that do have a material impact on the
liquidity situation and that do not fit into a category above
should be included here.
The ‘Counterbalancing Capacity’ contains information on
the institutions’ holdings of liquid assets. Assets are
divided into relevant subgroups based on asset
characteristics, such as central bank eligibility. Like in the
Cash Flow section, the CBC is also differs between
contractual and behavioural security flows. (if for
example repos from behavioural in-/outflows trigger a
change in the liquid asset composition used for
counterbalancing, these effects should be reported in the
behavioural security flow section of the CBC)
All the assets reported in the counterbalancing capacity
must be unencumbered.
 222

Table A8.1. Cash Flow Items (concluded)

The column ‘stock’ contains the current unencumbered
stock of assets available to the institution. The Maturity
buckets contain contractual and expected flows of
securities in the counterbalancing capacity.
Institutions should apply haircuts (orange box) reflecting
conservative assessments about the marketability of the
assets in each class and the possibility to be used in
repo transactions.

Cash-inflows are not to be accounted for in the ‘Cash

and Central Bank reserves’ position in the
‘Counterbalancing Capacity’ to avoid double counting.
Negative flows should be reported for securities at their
maturity date or at the maturity of reverse repos. Positive
flows should be reported at the maturity of repo
transactions, or at the settlement date or any purchases.
Corresponding cash flows should be reported in the
inflows or outflows section of the template.
Source: Authors

Assumptions

In the Assumption’s worksheet the stress tester specifies the ratio of the various contractual
and behavioral cash flows that will be rolled over on a continuous basis for each time bucket.
Concerning the security flows to counterbalance the net outflow of funding the user has to
apply haircuts on the contractual and behavioral security flows with regard to the stress
scenario. For potential counterbalancing assets the user can also set haircuts for asset prices.

The stress tester can either choose to use general assumptions that apply to all banks
uniformly or to set bank specific haircuts and roll over rates in order to differentiate between
single banks due to an idiosyncratic stress scenario.

Results

The result sheet contains the (cumulative) funding gaps and the corresponding (cumulative)
counterbalancing capacity for each maturity bucket after haircuts and roll over rates. These
can be compared across banks and with the aggregated banking system. A positive
counterbalancing capacity over the observation period is the main indicator for the funding
situation of the bank and serves as pass criterion for the fully-fledged cash flow analysis.
However, the user can adjust the pass criterion for each maturity bucket in which banks have
to remain liquid.
 223

Liquidity and solvency

Overview

There is a natural link between solvency and liquidity, and as the recent global financial
crisis has shown, they can reinforce each other. For example, during the European crisis in
the spring of 2010, market concerns about solvency of some sovereigns as well as their
domestic banks led to a liquidity crisis with banks finding it harder to access wholesale and
interbank funding and haircuts increasing for collaterals at the ECB. Downgrades by rating
agencies exacerbated this situation.

This liquidity stress testing tool allows simulating the link between liquidity and solvency
from three different complementary perspectives. First, it simulates the increase in funding
costs from a change in solvency (based on a simplified macro-financial credit risk model,
which should be re-calibrated to country-specific circumstances) or a rating downgrade.
Second, the tool allows simulating the (partial) closure of funding markets (both long and
short-term) depending on the level of capitalization. Central bank and intra-group funding as
well as the sale of liquid assets (subject to a haircut) can partly compensate for the liquidity
drain. Third, it examines the impact of concentration risk on funding, both in terms of name
concentration (through wholesale funding) and concentration in specific currencies, again
with the possibility of additional central bank funding and sales of liquid assets. To cater to
the needs of the first two tests, the liquidity framework hosts a simplified solvency template.
For most sophisticated tests, a fully-fledged solvency tool (such as Schmieder, Puhr and
Hasan 2011) should be used.

Assumptions

For the first stress test on the increase in funding costs arising from a change in solvency and
a rating downgrade, the user can either refer to the bank-specific one-year PD or the
equivalent external rating. The mapping between the ratings and the PD is based on empirical
evidence observed by Moody’s during the last three decades (from 1983 to 2009). The link
between funding costs and a bank’s default probability has to be determined based on a
regression model (there is a pre-defined model, but country-specific circumstances vary
widely). Depending on the number of notches a bank is simulated to be downgraded the
funding costs will increase accordingly, which feeds back to a solvency ratio under stress.
The test reveals the additional loss of solvency (capitalization) under stress, and is meant to
be linked to a solvency test. It is crucial to decide on the portion of the funding costs to be
passed on to customers, which alters the situation of banks.

For the second stress test, the user needs to specify whether either Tier 1 capital or the total
capital ratio is used for the liquidity/ solvency test. Then assumptions have to be made for the
threshold for the closure of funding markets (long and short-term) conditional on the capital
 224

ratio. It is important that the capital threshold is based on historical evidence and/or expert
judgment. Finally, the user decides whether there’s additional inflow of cash through central
bank funding and/or intragroup support (intragroup support could also be negative) as well as
the inflow of assets (e.g. through fire sales and subject to an appropriate asset-specific
haircut).

For the third stress test on the simulation of the impact on funding concentration, there are
two different tests (a) in the first test, the user selects the number of the largest (non-intra-
group) liquidity providers assumed to default (0-5) as well as the underlying recovery rate;
and (b) the second test simulates the liquidity position of a bank if funding for different
currencies closes. Again, assumptions have to be made with respect to intragroup funding
and central bank funding.

The user can decide whether some of the assumptions apply to all banks uniformly. The
manual data entry worksheet allows for bank-specific assumptions on the inflow of cash
from fire asset sales. For the concentration risk solvency test both the central bank funding
and change in intra-group funding can vary by bank.

Results

The results’ worksheet provides an overview for the three liquidity and solvency stress
testing modules. For the first stress test on the link between ratings/ PD and funding costs,
the key result is the ex-post total capital (tier 1) ratio without/ with the increase in funding
costs (which reveals the additional number of failures due to higher funding costs). The
liquidity stress test without the funding cost component is just the simplified macro-financial
credit risk stress test. Adding the funding costs explicitly links solvency to liquidity. This
provides an easy examination whether credit or liquidity risks (in terms of their contribution
to the ex-post capital ratio) are more important for individual banks and the system. The user
can specify a desired hurdle ratio for the chosen capital ratio (tier 1 or total) which then gives
the number of banks that fail due to the funding shock alone.

For the second stress test on the closure of funding costs depending on the capital level, the
results show the loss of funding (short and long term), the available assets to compensate
from additional central bank and intra-group funding and sale of liquid assets and finally,
whether any bank or the system has become illiquid from the liquidity shock.

In similar vein, for the third stress test on the impact of concentration risks on bank funding,
the results break down the overall loss of funding due to concentration risk (from the default
of a number of liquidity providers or due to a closure of funding for a specific currency). It
then lists available assets (both central bank funding, intra-group funding and liquid assets) to
compensate for the loss of funding as well as whether any bank/ the system have become
illiquid.
 225

Annex 9. Additional Information on Scenario Specification

Historical Scenarios

Deposit run on Northern Rock in 2007

Northern Rock was a former building society that, having demutualized in 1997, began to
expand rapidly. By the end of 2006, its assets had grown six-fold to £101 billion. This
expansion was aided by the increasing reliance on wholesale funding, comprising about
three-quarters of total funding, with over 40 percent alone in residential mortgage-backed
securities (as shown in Annex 7, the OECD average is 28 percent). Only a quarter of total
funding (OECD average: 44 percent) came from deposits, down from nearly two-thirds in
1997.

Beginning in August 2007, concerns about exposure to U.S. subprime mortgage assets led
wholesale markets to seize up. Northern Rock came under pressure, as it was able to find
only limited liquidity in wholesale markets. A retail run, the first significant bank run in the
U.K. since 1878, began on September 14. Northern Rock faced heavy withdrawals, and its
share price halved. Although most withdrawals were made through the internet, phone, or
mail, lines forming outside some branches were the most visible sign of the run. The run was
stopped only when the Chancellor, on September 17, announced arrangements to guarantee
all existing deposits in Northern Rock, “during the current instability in the financial
markets.”

In terms of withdrawal magnitudes, according to Shin (2009), between December 2006 and
December 2007, overall retail funding fell from 24.4 to 10.5 billion pounds. While typical
branch based customer deposits only fell from 5.6 to 3 billion pounds, withdrawals on phone,
internet and offshore deposits were relatively larger. In addition, the wholesale funding
squeeze was substantial with overall wholesale funding falling from 26.7 billion pounds in
June 2007 to 11.5 billion pounds in December 2007.

Deposit run on Latvian Parex Bank in 2008

Latvia’s second-largest bank (the largest domestic bank), Parex, was nationalized to maintain
its solvency in 2008 by the national government after a run on deposits took it to the brink of
bankruptcy. Parex, with 3.1bn lats ($5.6bn) in assets, lost 25 percent of its deposits from end-
August to end-November 2008 (i.e., within 3 months).

Wholesale funding squeeze for Kazak Banks

Over 2002–07, banks were able to sustain rapid expansion of their balance sheets through
high levels of foreign borrowing. Banking sector external debt, facilitated by high economic
growth and a burgeoning oil sector, grew to about 44 percent of GDP by 2007 (nearly
 226

$45 billion) from around 6 percent in 2002. During this period, the loan to deposit ratio
nearly doubled, peaking above 200 percent in 2007, which was among the highest relative to
comparable countries. Limited (tenge) deposits—as well as a lack of tenge term liquidity—
encouraged banks to take advantage of cheap foreign capital, which was largely channeled to
risky sectors, and to borrowers without foreign currency income streams.

Half the funding for the banking sector in Kazakhstan came from the wholesale market. A
combination of structural weaknesses and external factors left the Kazakhstani banking
system highly vulnerable to the sudden stop in capital flows. This contributed to the failure of
two large banks and two smaller institutions. The authorities were forced to intervene in two
top banks, take stabilizing equity stakes in two other leading banks, and provide widespread
liquidity support, including the targeted placement of deposits of state owned enterprises
throughout the system.

Deposit runs on Icelandic Banks

The size of Iceland’s banking sector was about nine times the country’s gross domestic
product (GDP) at the end of 2007, funded largely by external debt. The banking system was
dominated by three large commercial banks, Kaupthing Bank hf. (“Kaupthing”), Landsbanki
Íslands hf. (“Landsbanki”) and Glitnir banki hf. (“Glitnir”). The banks had relied heavily on
market funding for their operations, and had previously been criticized for a lack of
diversification in their funding profile, in particular, for the low proportion of deposits in
their funding. As a result, these banks intensified their focus on gathering deposits, and
successfully so. At the end of 2007, some 40 percent of their funding was in the form of
deposits, up from 28 percent in 2006, with more than two-thirds sourced from non-residents.

Iceland’s banking sector collapsed in early-October 2008, following severe liquidity and
solvency problems at the banks and collapse of the exchange rate. On September 29, 2008,
the Prime Minister announced that an agreement had been reached between the Government
and the largest owners of Glitnir, the country’s third largest bank, whereby the government
would contribute new share capital and take up a 75 percent stake in the bank. A week later,
on October 6, Iceland's parliament, the Althing, passed emergency legislation enabling the
government to intervene extensively in Iceland's financial system. On October 7, the FME
put Landsbanki into receivership; Glitnir and Kaupthing followed on October 8 and 9,
respectively. By that stage, the three banks combined had amassed debt of an estimated $61
billion—about 12 times the size of Iceland’s economy—and were unable to secure short-term
funding to continue servicing their obligations. A number of private interbank credit facilities
to Icelandic banks were shut down, and banks were unable finance their debts through short-
term borrowing. In an attempt to alleviate depositor concerns, the government offered an
unlimited guarantee to all depositors in banks and branches in Iceland. By that stage,
however, deposit runs on the overseas branches of Icelandic banks had already started (Ong
and Čihák, 2010).
 227

Haircuts for Liquid Assets

Treatment of Marketable Securities by the ECB (2011)

Table A9.1. Liquidity Categories for Marketable Assets Used by the European Central
Bank (ECB 2011)
Category I Category II Category III Category IV Category V
Central Local and Traditional Credit Asset back
Government regional covered institution securities
Debt government bonds debt
Instruments debt instruments
instruments (unsecured)
Debt Jumbo Debt Debt
instruments covered instruments instruments
issued by bonds issued by issued by
Central Bank corporate financial
and other corporations
issuers other than
credit
institutions
(unsecured)
Agency debt Other
instruments covered
bank bonds
Supranational
debt
instruments

Table A9.2. Haircuts Applied to Eligible Market Securities (ECB 2011)

Credit Residual Category I Category II Category III Category IV C
quality Maturity V
(years) Fixed Zero Fixed Zero Fixed Zero Fixed Zero
Coupon Coupon Coupon Coupon Coupon Coupon Coupon Coupon
0-1 0.5 0.5 1 1 1.5 1.5 6.5 6.5
1-3 1.5 1.5 2.5 2.5 3 3 8.5 9.0
AAA 3-5 2.5 3 3.5 4 5.0 5.5 11.0 11.5 16
to A- 5-7 3 3.5 4.5 5 6.5 7.5 12.5 13.5 /1
7-10 4 4.5 5.5 6.5 8.5 9.5 14.0 15.5
>10 5.5 8.5 7.5 12 11 16.5 17.0 22.5
0-1 5.5 5.5 6.0 6.0 8.0 8.0 15.0 15.0
1-3 6.5 6.5 10.5 11.5 18.0 19.5 27.5 29.5
BBB+ 3-5 7.5 8.0 15.5 17.0 25.5 28.0 36.5 39.5 NA
to 5-7 8 8.5 18.0 20.5 28.0 31.5 38.5 43.0
BBB- 7-10 9 9.5 19.5 22.5 29.0 33.5 39.0 44.5
>10 10.5 13.5 20.0 29.0 29.5 38.0 39.5 46.0
 228

Scenarios used in FSAPs

Table A9.3. Liquidity Risks Stress Tests as Part of the Recent FSAPs

Liquidity Outflows Example 1 Example 2 Example 3

Deposits
Austria (2007): 50 percent Ireland (2006): 10,30 and
Retail Russia (2008): 30 percent
of ST nonbank customers 50 percent
Spain (2005): 10-20
Russia (2008): 30 percent
Serbia (2009): 2 percent percent of demand
Corporate of current and 5 percent of
daily for 5 days deposits (retail and
time deposits
corporate)
Government UAE (2007): 35 percent
Belarus (2008): 25,50 and
75 percent (both
Non-resident UAE (2007): 30 percent
households and
corporates)
Lithuania (2007): 80
Foreign bank
percent
Other funding
Lithuania (2007): 100
UAE: 30 percent of
Interbank Russia (2008): No access percent of domestic
foreign interbank funding
interbank deposits
Romania (2009): Limited
Credit Lines granted by banks access to committed and
uncommitted lines
Lithuania (2007): 50 Serbia (2009): Full
Romania (2009): Limited
Parent Funding percent of liabilities to withdrawal with maturity
access
parent of less than 1 year
South Africa (2008):
Contingent liabilities Tapping 50 percent of
committed credit lines
Liquidity Inflows Example 1 Example 2 Example 3
Varies widely, most Isle of Man (2008):
conservative: only cash Interbank funding and
Definition of Liquid Assets
and government bonds intra-group funding not
remains liquid available
Austria (2007): Decrease
Ireland (2006): 10, 20 Russia (2008) and South
in value of liquid bonds by
Haircuts on Liquid Assets percent for debt securities Africa (2008): 20 percent
25 percent and equity by
and government bonds. on liquid assets
35 percent
Austria (2007): Eligible
colllateral in secured mkt
shrinks by 30 percent and
Shrinkage of Eligible Collateral below
30 percent of the eligible
Threshold
assets become ineligible
(i.e., their rating falls below
single A).
Note: ST = Short-term

Source: Authors
 229

Annex 10. Link Between Solvency and Liquidity

Figure A10.1. Schematic Overview for the Calibration of Funding Costs

Legend
Implied Capitalization Yellow: Input
(here: from BII IRB model) Green: Output

Time series of PDs Time series of Implied Funding Costs

(here: Moody’s KMV) (here: OECD, Market Data)

Historical Link Portion passed

(linear/non- on to
linear) customers?

Implied Rating Increase of Funding

(here: evidence from costs (benchmark,
Moody’s 2010) used for stress test)

Illustrative example for a sample of large German banks

Economic Change of
EDF or PD Funding costs
capital ratio Funding
Rating (One-year, (spread above T-
(Basel II spread (CAR
Percent) bills, bps)
(quasi-IRB) Elasticitity)
AAA 0.00004 8.7 28.1%
AA+ 0.00006 8.7 27.3% 0.00
AA 0.0001 8.7 26.2% 0.00
AA- 0.001 8.9 21.2% 0.00
A+ 0.002 9.0 19.7% 0.00
A 0.026 11.9 14.3% -0.01
A- 0.032 12.7 13.9% -0.02
BBB+ 0.1 21.0 11.7% -0.04
BBB 0.139 25.9 11.1% -0.08
BBB- 0.291 44.6 9.9% -0.15
BB+ 0.682 92.7 8.5% -0.35
BB 0.728 98.4 8.4% -0.57
BB- 1.791 229.4 7.1% -1.03
B+ 2.45 310.5 6.7% -2.01
B 3.827 480.2 6.2% -3.16
Note: capital ratio includes 2.5 ppt voluntary buffer above regulatory
minimum; funding costs without equity costs

Source: Authors based on Moody’s KMV and OECD data

 230

REFERENCES

Acharya, Viral V., Thomas F. Cooley, Matthew P. Richardson, and Ingo Walter, 2010,
“Regulating Wall Street: The Dodd-Frank Act and the New Architecture of Global
Finance” (New York: Wiley and Sons).

Aikman, David, Piergiorgio Alessandri, Bruno Eklund, Prasanna Gai, Sujit Kapadia,
Elizabeth Martin, Nada Mora, Gabriel Sterne and Matthew Willison, 2009, Funding
liquidity risk in a quantitative model of systemic stability, Bank of England Working
Paper no. 372.

Boss, M., , G. Fenz, G. Krenn, J. Pann, C. Puhr, T. Scheiber, S. W. Schmitz, M. Schneider,

E. Ubl), 2008, “Stress Tests for the Austrian FSAP Update 2007: Methodology,
Scenarios and Results,” Financial Stability Report 15, Austrian National Bank: Vienna,
June 2008, 68-92

Barnhill, Theodore Jr. and Liliana Schumacher, forthcoming, “Modeling Correlated Systemic
Liquidity and Solvency Risks in a Financial Environment with Incomplete
Information,” Forthcoming IMF Working Paper.

Basel Committee on Banking Supervision (BCBS), 2006, ‘International Convergence of

Capital Measurement and Capital Standard,’ Bank for International Settlements

———, 2008, Principles for Sound Liquidity Risk Management and Supervision (“Sound
Principles”. www.bis.org/publ/bcbs144.htm.

———, 2010a, Basel III: A global regulatory framework for more resilient banks and
banking systems, December. http://www.bis.org/publ/bcbs189.htm.

———, 2010b, Basel III: International framework for liquidity risk measurement, standards
and monitoring, December. http://www.bis.org/publ/bcbs188.htm

———, 2010c, “Results of the comprehensive quantitative impact study”, December.

http://www.bis.org/publ/bcbs186.htm

Basel Committee on Banking Supervision (BCBS) and Financial Stability Board (FSB),
2010, “An assessment of the long-term economic impact of stronger capital and
liquidity requirements”, August 2010, p.48.

Brunnermeier, Markus, Andrew Crocket, Charles Goodhart, Martin Hellwig, Avi Persaud,
and Hyun Song Shin, 2009, “The Fundamental Principles of Financial
Regulation,”Geneva Reports on the World Economy 11.
 231

Cao, Jin, and Gerhard Illing, 2009, “Endogenous Systemic Liquidity Risk,” CESIFO
Working Paper No. 2627 (April).

Cerutti, Eugenio, Anna Ilyina, Yulia Makarova and Christian Schmieder, 2010, Bankers
Without Borders? Implications of Ring-Fencing for European Cross-Border Banks,
IMF WP no. 10/247, http://www.imf.org/external/pubs/cat/longres.cfm?sk=24335.0.

Čihák, Martin, 2007, “Introduction to Applied Stress Testing”, IMF WP no. 59.

Committee of European Banking Supervisors, 2009a, Report on Supervisors’ Lessons Learnt

- A Liquidity Risk Perspective, CEBS 2009 107, London.

———, 2009b, Guidelines on liquidity buffers and survival periods, London.

http://www.eba.europa.eu/documents/Publications/Standards---
Guidelines/2009/Liquidity-Buffers/Guidelines-on-Liquidity-Buffers.aspx

———, 2010, Results of the comprehensive quantitative impact study. http://www.c-

ebs.org/Publications/Quantitative-Impact-Study.aspx

Demirguc-Kunt, Asli and Harry Huizinga, 2009, Bank Activity and Funding Strategies: The
Impact on Risk and Returns, World Bank Policy Research Working Paper No. 4837

Deutsche Bank, 2009, “Basel II Pillar 3 Report”.

www.db.com/ir/en/download/Basel_II_Pillar_3_Report_2009_FINAL.pdf

———, 2010, “Basel II Pillar 3 Report 2010,”

http://www.db.com/ir/en/download/Basel_II_Pillar_3_Report_2010.pdf.

European Central Bank, 2008, “EU banks liquidity stress tests and contingency funding
plans”, http://www.ecb.int/pub/pdf/other/eubanksliquiditystresstesting200811en.pdf

———, 2011, “The Implementation of Monetary Policy in the Euro Area,” General
Documentation on Eurosystem Monetary Policy Instruments and Procedures, February
2011, http://www.ecb.int/udl.html?doc_id=gendoc_en.

Farhi, Emmanuel, Mikhail Golosov, and Aleh Tsyvinski, 2009, “A Theory of Liquidity and
Regulation of Financial Intermediation,” Review of Economic Studies, Vol. 76, No. 3,
pp. 973–92.
 232

Frank, N., B. González-Hermosillo, and H. Hesse, 2008, “Transmission of Liquidity Shocks:

Evidence from the 2007 Subprime Crisis,” IMF Working Paper 08/200 (Washington:
International Monetary Fund).

Goodhart, Charles, 2008, “Liquidity risk management”, Special issue on liquidity No. 11,
Banque de France, Financial Stability Review, February 2008.

Goodhart, Charles, 2009, “Liquidity Management,” Jackson Hole Financial Stability and
Macroeconomic Policy Symposium, Federal Reserve Bank of Kansas City, August.

Jobst, Andreas A., forthcoming, “Measuring Systemic Risk- Adjusted Liquidity,” IMF Working
Paper (Washington: International Monetary Fund).

De Haan, Leo, and Jan Willem van den End, 2011, “Banks’ responses to funding liquidity
shocks: lending adjustment, liquidity hoarding and fire sales,”, Dutch Central Bank
Working Paper no. 293, Available via the Internet:
www.dnb.nl/en/binaries/working%20paper%20293_tcm47-253006.pdf.

International Monetary Fund, 2008, Global Financial Stability Report, “Market and Funding
Illiquidity: When Private Risk Becomes Public” (Chapter 3), World Economic and
Financial Surveys (Washington, April).

———, 2009, Global Financial Stability Report, ‘Market Interventions during the Financial
Crisis: How Effective and How to Disengage?’ (Chapter 3), October 2009.

———, 2011, Global Financial Stability Report, ‘How to address the systemic part of
liquidity risk?’ (Chapter 2), April 2011.

Moody’s, 2010, “Corporate Default and Recovery Rates”, 1920-2009, Special Comment,
February.

Oesterreichische Nationalbank (OeNB, Austrian Nationalbank), 2008, Financial Stability

Report 16. December.

Ong, Li Lian, and Martin Čihák, 2010, ‘Of Runes and Sagas: Perspectives on Liquidity
Stress Testing Using an Iceland Example,’ IMF Working Paper Series 10/ 156.

Perotti, Enrico, and Javier Suarez, 2009, “Liquidity Risk Charges as a Macroprudential
Tool,” CEPR Policy Insight No. 40 (London: Centre for Economic Policy Research).
Available via the Internet: www.cepr.org/pubs/PolicyInsights/
CEPR_Policy_Insight_040.asp.

______, forthcoming, “Regulation of Liquidity Risk,” CEPR Policy Insight (London).

 233

Shin, Hyun Song, 2009, ‘Reflections on Northern Rock: The Bank Run that Heralded the
Global Financial Crisis,” Journal of Economic Perspectives, Vol. 23(1), pp. 101-119.

Schmieder, Christian, Puhr, Claus and Maher Hasan, 2011, “Next Generation Balance Sheet
Stress Testing”, IMF Working Paper 2011/83 (Washington: International Monetary
Fund).

Schmieder, Christian, Heiko Hesse, Benjamin Neuendorfer, Claus Puhr and Stefan W.
Schmitz, 2012, “Next Generation System-Wide Liquidity Stress Testing”, IMF
Working Paper 2012/3 (Washington: International Monetary Fund). A shorter version
was also published in VOX.

Schmitz, Stefan W., 2009, “Central banks and systemic liquidity risk: the role of stress
testing”, presented at the 6th Annual RiskCapital Conference, Brussels, June 2009

———, 2010, “Liquidity Regulation and Stress Tests”, presented at RiskMinds, Geneva,
December 2010

———, 2011, “The Impact of the Basel III Liquidity Standards on the Implementation of
Monetary Policy,” Available at SSRN: http://ssrn.com/abstract=1869810.

Schmitz, Stefan W., Ittner, A., 2007, “Why central banks should look at liquidity risk”,
Central Banking, Vol. 17(4), 32-40

Valderrama, Laura, 2010, “Macroprudential Regulation under Repo Funding,” IMF Working
Paper No. 10/220 (Washington: International Monetary Fund).

Van den End, Jan Willem, 2008, “Liquidity Stress Tester: A macro model for stress-testing
banks’ liquidity risk, “Dutch National Bank Working Paper No. 175, May 2008.

Wong, Eric and Cho-Hoi Hui, 2009, “A liquidity risk stress- testing framework with
interaction between market and credit risks,” Hong Kong Monetary Authority Working
Paper 06/ 2009.
 234

X. EUROPEAN FSAP: TECHNICAL NOTE ON STRESS TESTING OF BANKS, WITH DANIEL

HARDY, 2013180

A. Executive Summary

The European authorities are strengthening bank stress testing procedures and their
application. Following the poor reception of the 2010 exercise, the 2011 solvency stress
testing and recapitalization exercises were marked by extensive consistency checks, more
transparency about methodology and data, for example, regarding sovereign exposures, and
higher hurdle rates. The exercises succeeded in prompting banks to increase the quantity and
quality of their capitalization, and contributed to a reduction in uncertainty and an increase in
the credibility of the process. However, despite banks raising more than €200 billion as a
result of the recapitalization exercise, confidence in European banks is not fully restored, in
part because the market suspects some banks of having been insufficiently transparent—
including as part of the stress testing exercises—about their losses and exposures to problem
sectors.

Lessons from the past stress tests are being used to strengthen and streamline
procedures for the planned 2013 exercise. That exercise is likely to involve three-year
projections under a baseline and a stressed scenario. Much effort has been, and will be put
into ensuring that methodologies are consistently applied while reducing, as far as possible,
costs to participating banks. A major objective it to generate detailed analysis relevant to the
assessment of banks’ capital plans during the gradual transition to Capital Requirement
Directive (CRD) IV requirements, rather than pass/fail results based on a single metric.

There remain a number of controversial issues, but experience suggests that the
benefits of a bold approach outweigh the risks. A high degree of transparency, including
on reference date data and on sensitivity to differences in definitions of input data,
strengthens rather than weakens confidence and market functioning. If the 2013 exercise is to
focus on supervisory issues such as an assessment of banks’ plans to implement the solvency
elements of CRD IV, then it should be consistently designed for that purpose, and also
presented as such; otherwise, markets are likely to follow past form and be fixated on capital
shortfalls and relative weaknesses.

As the acuteness of the crisis diminishes, the identification of other vulnerabilities and
issues, such as funding risks and structural weaknesses, will gain in relative importance.
Most major banks now seem comparatively well capitalized, but funding remains
problematic (for example, because of reliance on official funding and asset encumbrance in

180
This chapter is based on Hardy and Hesse (2013).
 235

some banks), while the sector faces deep structural challenges relating to low profitability
and growth and the longer-term impact of regulatory changes.

In light of these considerations, the note elaborates on certain main recommendations,

some of which are primarily the responsibility of NSAs. In these areas, work is already
on-going or can start soon, but often some time will be needed to complete the task:

 Continue development of the efficiency and effectiveness of consistency checks, for

example, by facilitating timely communication and agreement on templates.

 Continue to publish a wide range of detailed information on participating banks, and

especially on their reference period condition and (sovereign) exposures.

 The practice adopted in the 2011 EBA recapitalization exercise of recognizing the full
risks attached to all banks’ sovereign securities’ exposure, and at a minimum relevant
data should be published.

 Move to standardize definitions of NPLs, loan classifications, provisioning, etc.,

while initiating reviews of input asset quality data.181 Some issues here will take time
to be resolved, so the sooner the process starts the better. NSAs are responsible for
the provision of consistent and interpretable banking sector data, but the EBA has an
important role in coordinating and driving forward activities.

 Implement checks on the sensitivity of the results to model assumptions and bank
circumstances, including sensitivity to differences in asset quality and definitions.

 Further refine benchmarks and satellite models, especially regarding pre-impairment

income, risk-weighted assets (RWA), and funding costs, in order to ensure more
comparability and consistency across banks’ bottom-up results.

 Ensure that the 2013 stress testing exercise generates operational recommendations
and supporting indicators for supervisors, rather than being reduced to a perception of
a pass/fail metric.

 Incorporate as far as possible banks’ funding and capitalization plans in the stress test
projections, including the effects of the phase out of the Long-Term Refinancing
Operations (LTRO) provided by the European Central Bank (ECB); further efforts
could be made to assess the sensitivity of results to likely changes in balance sheet
composition, rather than assuming that it stays static.

181
The definitions should be as consistent as possible while recognizing real differences, for example, in loss
given default rates across countries and across time.
 236

 Ensure full coordination between future EBA and ECB and Europan Systemic Risk
Board (ESRB) stress testing exercises. For the 2013 exercise, the ECB could play a
very active role, not only in making macro projections and top-down stress testing,
but also, for example, in the review of input asset quality data.

 Continue EBA coordination of the EU-wide stress tests under the new Single
Supervisory Mechanism (SSM), working with ECB and NSAs, and ensuring of
quality control. The ECB should run supervisory stress tests for the banks in the SSM,
while ESRB should focus its contributions on macro-prudential issues, such as the
identification and calibration of systemic risk factors and the use of stress test results
in formulating policy advise. Over the medium term, shift some of EBA’s efforts to
running tests on hitherto relatively neglected topics such as structural issues and
funding vulnerabilities. In this context, stress tests could be designed to incorporate
more longer-term and cross-sector factors (for example, using contingent claims
analysis and incorporating nonbank financial institutions) that relate to structural and
macro-prudential issues, and to calibrate prudential requirements. However,
competing EBA, ECB/SSM and ESRB stress tests are to be avoided.

 Following the 2011 internal EBA liquidity risk assessment, develop further liquidity
stress testing, including through cash flow-based approaches and making use of the
relevant reporting templates being developed in the context of CRD IV. Ensure
disclosure and transparency of the reporting templates and overall liquidity stress
testing approach, while safeguarding sensitive results.

B. Introduction182

Stress testing has become an essential and very prominent tool in the analysis of
financial sector stability and development of financial sector policy. Starting with
the 2010 test led by the Committee of European Banking Supervisors (CEBS), and reinforced
by the 2011 test and the bank recapitalization exercise led by the EBA, the output of EU-
wide stress tests has been viewed as essential information on the health of the system.
Moreover, the reliability of the results and the efficiency with which they were generated
(especially the recapitalization exercise) have greatly influenced the credibility of the
European and national authorities involved. This prominence demands that future stress
testing exercises be very carefully designed and executed.

Stress testing in itself can have only a limited impact unless it is tied to action. In current
circumstances in much of Europe, stress testing alone would mainly reconfirm the over-
leverage of the public, household or corporate sectors, and the structural impediments to
faster overall growth, in addition to unfavorable conjunctural conditions. The publication of

182
Prepared by Daniel Hardy and Heiko Hesse.
 237

stress test results with enough supporting material (including on the initial condition of
banks) can indeed be helpful in reducing uncertainty; even banks that are revealed to be
relatively weak may benefit if the market paralysis engendered by great uncertainty is
relieved. But stress tests are of value mainly when they are followed up by concrete and swift
actions by the authorities (supervisory and others) and by bank managers that improve the
condition of banks and of banks’ clients. Therefore, the informational role of stress testing
and its link to policy actions are the underlying themes of this note.

This note focuses on bank stress testing led by the EBA, and in particular the
forthcoming 2013 exercise and the associated data quality issues. The NSAs, ECB and
ESRB conduct their own tests for various purposes; there will be discussion of those that
have more or less Europe-wide relevance. Moreover, consideration will be given to both
solvency and liquidity aspects of stress testing, and how priorities are likely to evolve in the
post-crisis environment, especially with the introduction of the SSM.

C. Background

The 2010 CEBS-led stress testing exercise, which can be viewed as the start of EU-wide
stress testing and which was initiated near the start of the financial crisis, was relatively
poorly received. The stress scenario was regarded as too mild in the circumstances, and
there was little assurance that banks had not been able to incorporate an optimistic bias into
the results. Limited information disclosure did little to relieve the intense uncertainty
prevalent at that time. The sample of banks included some that quickly proved to pose
systemic risks in certain countries.

The design of the 2011 EBA-led exercise partly reflected the lessons learnt, notably on
the need for quality and consistency controls and on transparency, and was better
received. Even though the final estimated capital shortfall was modest, that result was
largely the product of many banks—especially those with relatively weak capital buffers—
preemptively increasing their capitalization and what with hindsight appears to be unduly
optimistic baseline and stress scenarios, including with regard to the treatment of sovereign
risk. Three main quality control mechanisms were: the banks’ own controls; those by NSAs
(e.g., supervisory judgment); and the quality assurance process led by EBA. For the latter,
EBA formed a Quality Assurance Task Force (QATF) with secondees from NSAs, the ECB
and the ESRB, who challenged their peers in other NSAs on the consistency of the banks’
bottom-up assumptions, methodologies and results. Compared to the 2010 stress test, EBA
also improved its off-site review by checking bank input data for errors, ensuring the correct
adoption and application of the stress testing methodologies, and using statistical benchmarks
(mainly cross-sectional) for probabilities of default (PDs), loss given default (LGDs), and
default rates by counterparties, country and sector. The top-down stress test performed by the
ECB and ESRB played an essential part in order to benchmark the bottom-up results of the
banks.
 238

For the 2011 stress test, EBA’s board of supervisors decided not to include market risk
haircuts to the banks’ sovereign exposures in the banking book, but did publish
relevant data. Only the banks’ sovereign holdings in the trading book would be subject to
mark-to-market (MTM). Given the intensification of the euro area sovereign debt crisis, this
assumption was debatable and criticized (including by the IMF), but the enhanced disclosure
and transparency of the banks’ sovereign exposures allowed market analysts to calculate their
own sovereign haircuts and eventually the capital shortfall of banks in the sample.

The subsequent recapitalization exercise contained some elements common to stress

testing, and further enhanced the credibility of the institutions involved (Box 2).
Importantly, all sovereign securities’ holdings were subject to MTM. The recapitalization
exercise was not a full stress test since it did not include a macro scenario or capture banks’
ongoing funding strains. Most banks have met the 9 percent core Tier 1 (CT1) capital
requirement; the exceptions are banks in unusual circumstances where action is being taken
especially where government governments apply (Box 2). One important implication of this
achievement is that banks already more or less have the capital necessary to meet
requirements under Basel III or the EU’s Capital Adequacy Directive (CRD) IV, even were
the requirements to be applied in full or imposed through market discipline.

Box 2. Capital Outcome of the 2011 Stress Test and Recapitalization Exercises

The EBA stress and recapitalization exercises have helped identify weak banks and increase capital
buffers. The second EU-wide stress test (July 2011) identified €25bn of capital shortfall using a single adverse
macro scenario. Even though banks raised €50bn fresh capital in the first four months of 2011, confidence
remained tenuous as the sovereign debt crisis intensified in 2011.

In December 2011 the EBA recapitalization exercise recommended the achievement of 9 percent CT1 by
end-June 2012, after establishing a sovereign buffer against banks’ holdings of government securities
based on a market-implied valuation of those holdings. The aggregate capital shortfall after including the
sovereign capital buffer amounted to €115 billion for 37 banks including those under restructuring (out of 71
banks in the sample), with the largest shortfall on Greek banks (€30 billion) followed by Spain (€26 billion),
Italy (€15 billion) and Germany (€13 billion). Taking out banks under deep restructuring (Dexia, Volksbank
and WestLB), the Greek banks as well as Bankia left 27 banks with a capital shortfall totaling €76 billion.

The banks’ capitalization plans, in aggregate, more than covered the capital shortfall identified by EBA.
Direct capital measures accounted for the majority of the plans, with the remainder comprising changes to bank
risk weight models, asset disposals, and reductions in lending, which mostly comprised actions taken under EU
State Aid rules. EBA explicitly discouraged banks from shedding assets in order to meet the 9 percent capital
target, by requiring that banks cover the shortfall mainly through capital measures of the highest quality. EBA
subsequently published an overview of the capital plans that banks submitted to regulators and then to EBA at
end- January 2011.183

The EBA recapitalization exercise lead to an additional €200 billion in capital generation or release by

183
The banks’ capital plans were submitted by national supervisors to EBA at end January 2011.
 239

June 2012, while government backstops were provided to the weakest banks. A few banks under
restructuring and recapitalization programs did not achieve the target on time.
D. The 2013 Bank Solvency Stress Testing Exercise

The authorities have decided to conduct another coordinated bank solvency stress
testing exercise, but with more emphasis on supervisory issues and less on a pass/fail
metric. The initial plan envisages that the exercise would be conducted mainly in the second
half of 2013, with preparatory work beginning as soon as possible. The main supervisory
issue is the assessment of the realism, consistency and robustness of banks’ capital plans to
meet the phased-in capital requirements under CRD IV, which will affect minimum
capitalization levels, the definitions of various sorts of capital, and the definition of risk-
weighted assets (RWA).

The lessons from past stress testing exercises will have to be incorporated into the
design and execution of the forthcoming exercise, but modified as needed in light of
current conditions and the exercise’s objectives. The improvements in efficiency and
effectiveness seen since the 2010 CEBS exercise should be extended, and in particular,
the 2011 stress testing and recapitalization exercises offer additional lessons. Nonetheless,
adjustments need to be made to allow for the fact that the situation of European banks is
more diverse (also due to ongoing fragmentation and asset quality pressures) and also less
uncertain than in 2010 or 2011—some operate in program countries and others operate in a
comparatively benign macroeconomic environment, some are heavily dependent on central
bank refinancing and others have ample and excess liquidity (often deposited back at the
ECB). Moreover, the tests need to be geared towards generating output and recommendations
that are relevant for supervisory purposes, rather than those that are needed in an acute crisis
situation. This section concentrates on identifying ways to reconcile these features in various
aspects of the design of the exercise.

E. Publication and Transparency

The publication of detailed data on major European banks in the context of the 2011
stress testing and recapitalization exercises contributed to reducing uncertainty
markedly and to the credibility of those exercises. The authorities were praised for
providing enough information (over 3,000 series, notably on sovereign exposures) that
market analysts could check and run their own projections based on alternative scenarios and
assumptions on banks’ treatment of their sovereign exposures. It is inevitable that analysts
will want to assess the situation of banks assuming the immediate full implementation of
CRD IV (partly already happening), and banks may be basing their own planning on this
assumption. Were relevant data not provided, the market would look on the exercise with
increased skepticism.

The authorities will have to publish data from the forthcoming exercise, in at least as
much bank-by-bank detail and also covering the initial situation of individual banks at
the reference date, if not necessarily all projections. To do otherwise would at best miss
 240

an opportunity to reduce uncertainty, which has contributed to the fragmentation of funding

markets, and could lead to suspicions that the authorities have bad news to hide. There
should be a presumption that also test results would be published in detail. Even if the
authorities do not highlight certain series such as the projected evolution of banks’
profitability, the information is valuable in the context of structural pressures on the sector.
For instance, the 2011 stress test did not publish as comprehensive bank-specific data on the
banks’ starting positions as projections from the adverse scenario. However, there may be
scope to keep confidential some details. For example, publication of results from sensitivity
analysis may be more confusing than reassuring, in part because the market could take those
as benchmark results and penalize banks that do poorly in the sensitivity stress tests. The
experience with sensitivity tests performed in the 2011 exercise could be useful here.

Confidentiality will have to be maintained over certain aspects of the supervisory

recommendations. Some recommendations may relate to a bank’s confidential business
planning, its detailed funding plans, or supervisors’ own policies. But a possible negative
market impact should not in itself be grounds for non-publication, since such market
discipline is desirable. Consistent treatment across banks would be essential, not least to
maintain a level playing field for competition.

In this connection, it is worth stressing that full disclosure of banks’ sovereign

exposures (including that in the banking book) will be essential, and that the tests will
need to recognize fully the attendant risks. Given that the 2011 EBA recapitalization
exercise involved marking to market (MTM) banks’ sovereign securities’ exposure in the
banking book (available for sale, AFS, and held to maturity, HTM), the market could be
critical of a reversal. Admittedly, the Basel III rules envisage that just the trading and AFS
books be marked to market with a gradual phase in period, while the HTM book would not
be subject to MTM. While authorities will have to trade off disclosure with the taken MTM
approach, at a minimum relevant data should be published. Nonetheless, if current market
conditions persist, for most banks sovereign exposures are likely to be a smaller source of
losses (or could contribute positively) in the baseline of the 2013 exercise. A relapse at least
to recent peak sovereign spreads would be seen as constituting a plausible but not very
extreme scenario. Hence a conservative approach would probably not be disruptive. In any
case, analysts would be able, based on the disclosed detailed banking data, to calculate each
bank’s estimated haircut on its total sovereign securities portfolio.

F. Consistency and Quality Control Mechanisms

The authorities are making strong and commendable efforts to improve on the quality
control mechanisms that were successfully deployed in the 2011 exercise. For example,
reporting templates will incorporate various checks, and there is expected to be early contact
between the authorities and banks to ensure that the methodology, benchmarks, and reporting
forms are well understood. The 2011 experience suggests that the ECB top-down macro
stress test need to be prepared and used as a cross-check at an earlier stage, but to this end
 241

“clean up” data needs to be provided to the EBA earlier on.184 In particular, NSAs need to
make a quick but thorough check of data as soon as it is received.

The role of the ESRB/ ECB top-down stress test could be further strengthened. The
ECB has indeed indicated an interest in being more closely involved in various aspects of the
exercise, including the input data review (see below). Besides the above mentioned
importance of the top-down stress test used at an earlier stage in the quality assurance
process, such tests could also be used to challenge the banks’ bottom-up results by
introducing different modeling approaches or including effects that cannot be captured in the
bottom-up exercise, such as systemic and feedback effects. Such an expanded role could be
especially useful for sensitivity analysis around the adverse scenario.

The efficiency of the quality control process would be enhanced by allowing EBA staff
to be in direct contact with banks, rather than channeling communication through
NSAs. Staff from the NSAs should be present in conference calls or physical meetings with
EBA staff and the involved banks so the NSAs remain well-informed. This would help to
avoid unnecessary delays encountered in the 2011 stress test in communication among banks,
NSAs and the EBA.

As part of this process, the authorities need to continue to build up time series of
statistical benchmarks for PDs, LGDs, and default rates by granular counterparties,
countries and sectors, as well as ensure consistent application by banks of point-in-time
(PIT) estimates of PDs and LGDs. Those benchmarks should be cross-checked with the
estimated PDs and LGDs of the ECB that are also being used to challenge banks and are
adopted by banks under the standardized approach. Banks should use their PIT PD and LGD
parameters for the bottom-up stress test and not through-the-cycle (TTC) equivalents. The
use of PIT parameters is important because results need to be sensitive to the scenarios, and
the PIT PDs are relatively forward-looking. Stress tests are meant to say something about the
ability of banks to survive bad points in time, so TTC parameters are not fully relevant to an
assessment of resilience to conjunctural shocks 185

G. Input Data Review

The banking systems of the program countries have been subject to detailed asset
quality reviews (AQR), and the question therefore arises of how to ensure consistency of
input data. Elaborate and expensive “deep dive” AQRs have been carried out in individual

184
In the 2011 EBA stress test, the ECB contributed the adverse scenario and top-down stress test, besides
participating in the stress testing and quality assurance task forces.
185
It is possible that, if PDs and LGDs are not sufficiently sensitive to the scenario, and RWAs decline over the
scenario due to losses, the positive impact on capital (from lower the RWA—the denominator) may offset the
limited impact of losses on capital (the numerator).
 242

banking systems in Europe, and have formed a solid base upon which to conduct crisis stress
tests. The IMF-Commission-ECB “Troika” very much supported these efforts. Yet, data from
some of the non-program countries conceivably may contain importantly flaws, and merely a
lack of consistency will make results difficult to interpret and could interfere with the internal
market.

Inconsistencies may arise despite the (almost) universal application of International

Financial Accounting Standards (IFRS) and a system of internal and external audits
with supervisory oversight. First, IFRS allows some room for local differences in
definition, for example, of a nonperforming loan (NPL) or renegotiated/ restructured loan.
Second, interpretation of common definitions may differ across countries or banks. Third,
interpretations may differ over time. In current circumstances, a bank may more readily
choose to roll over a problem loan and make modest provisions, partly to help its borrower
and partly to make its own results look better. It should be noted here, however, that
consistency does not imply uniformity, as accounting differences may reflect underlying
differences; PDs and LGD rates may genuinely differ, for example, because of large
differences in bankruptcy laws and loan work-out arrangements.

Yet, undertaking a full-blown AQR across the EU would be very expensive, time
consuming, and possibly counter-productive. Besides the practical difficulties and
expense, announcing a comprehensive AQR would cast doubt on the integrity of past stress
testing exercises, national authorities, bank management, and the accounting and audit
professions. Furthermore, consideration would have to be also given to undertaking an AQR
for the assets of European banks outside the EU, which enterprise would add greatly to the
complexity and cost.

It is recommended that the authorities, coordinated by the EBA, make rapid progress
in unifying definitions of NPLs and provisioning criteria. Efforts in this direction have
been under way for some time, but now there should be momentum behind the project. Full
implementation of all aspects might take place after the 2013 stress testing exercise, but that
would not be a great drawback. There would need to be guidance offered to national
authorities and the accounting and audit professions. Overall, the provision of consistent and
comparable banking sector data lies in the (national) supervisors’ accountability.

On balance, it would be worthwhile to conduct a limited review of input data, especially

on asset quality, but with a focus on problem sectors and without greatly impeding the
stress testing exercise. It may be possible identify some sectors—based on expert judgment,
statistical analysis, or experience with the program country AQRs—which are worth
investigating more closely. EBA is considering choosing one specific asset portfolio (which
can be different by countries) for the asset classification review as input for the 2013 stress
test, an approach which seems sensible. EBA with the NSAs should conduct a detailed
analysis which of the four chosen portfolios commercial real estate, SMEs, forborne
residential mortgages or level 3 trading book assets should apply to each country in the EBA
 243

sample. The review of input data would complement the enhanced system of consistency
checks built into the stress testing procedures.

Issues of special concern for such a review of input data are likely to include: lender
forbearance, impairment deficiencies, risk weighting, and RWA calculations by banks;186
collateral valuations and credit risk mitigation techniques; and treatment of restructured
loans. The approach would also need to enhance confidence in the reliability of the internal
credit rating systems operated by banks.187 The main concern should be to make the reference
period data as reliable as possible, as judged by the situation at that time: the baseline
projection is meant to capture the evolution of impairments going forward. The initial focus
should be on banks included in the stress test, but in due course the unified definitions should
be used across all banks. Countries that recently underwent third party diagnostics of their
banking systems would be exempted from part of the exercise to avoid a mere repetition of
effort.

Delaying the stress testing for the sake of undertaking an exhaustive input data review
would bring some benefit, but also important costs if the postponement is major or of
uncertain length: the review may end up being very long drawn out, and meanwhile the
regulatory framework is changing, so the stress testing hurdle rates and other conditions
would alter.188 In addition, investors could come to expect the revelation of additional losses
in banks’ portfolios, so a delay could potentially lead to destabilizing market uncertainty.189
One approach would be to define certain limited but valuable objectives (e.g., review of a
significant sample of exposures to one problem sector in each country) that could realistically
be achieved within six to nine months, and launch the stress testing exercise when these
objectives are met. Methodological preparations for the exercise could proceed in parallel.
Further refinements to data consistency would be left to later. Costs and benefits of a delay

186
For instance, the recent Bank of England Financial Stability Report (November 2012) shows that banks’
RWA calculations for the same hypothetical portfolio can differ vastly, with the most prudent banks calculating
over twice the needed capital than the most aggressive banks.
187
Depending on the scope and time available, some additional elements are possible for the review of the
chosen loan portfolio, such as a review of loan alteration practices, determining the reasonability of the
methodology used to estimate TTC risk drivers and cycle-smoothing techniques, or providing an opinion on the
overall reliability and integrity of the TTC risk weights and PIT loss estimates. The comprehensive AQRs
conducted in Euro area program countries could provide further elements that could be relevant for the review
of the input data.
188
For example, the liquidity coverage ratio (LCR) would come into force.
189
It is possible that the exercise will reveal a sizable “hole” in the capitalization of some banks, even before
any projections are made. The authorities will need to think in advance of how to handle such a situation, for
example, through immediate remedial supervisory action and exclusion of the affected banks from the regular
stress test.
 244

need to take into account also the interest of the ECB in participating in the 2013 stress test in
connection with the forthcoming Single Supervisory Mechanism.

Public explanation of the effort will need to be handled with care, and measures taken
to ensure that the exercise is recognized to be rigorous but limited.190 In the publication
stage, EBA could promote the disclosure of granular asset quality information to enhance
transparency and reduce market uncertainty about banks’ asset quality. However, some data
may be highly market sensitive; rules of engagement in such case should be worked out in
advance, especially if the ECB is involved in the context of the SSM. Also, evidence of
under-provisioning that is unquestionably consistent with IFRS might prompt tougher
guidance in the stress testing methodology on the future evolution of losses, rather than being
reflected in published stock data at the reference date. The experiences of data disclosure
from the external AQRs already conducted could be useful here. The credibility of the
exercise would be increased by the involvement of outside evaluators or at least peer
reviewers.

H. Refinement of Satellite Models

The authorities have collected data and undertaken analysis which allows the plausible
projection of many variables of interest in a baseline and an adverse scenario, but
certain important series have proven to be especially difficult to model and deserve
more research attention. In the 2011 stress testing exercise, the projections of some series
differed greatly across banks for reasons that were at best unintuitive, and these peculiarities
may have weakened confidence in the overall results. The following series are important but
have proven difficult to forecast:

 Non-interest, non-trading income, which is of increasing importance to many banks

and which may be disproportionately sensitive to a severe downturn. Banks’
projections of their fee income could be subject to some guidance by EBA, especially
for the adverse scenario, to avoid banks using fee projections to compensate for loan
impairments;

 Trading income, which depends on banks’ own-account trading activity of both on-
and off-balance sheet items. Financial instruments in bank portfolios could be re-
valued at the prices prevailing in the scenario, rather than through the use of satellite
models, which might not adequately capture banks’ trading income;

190
Using a term other than “Asset Quality Review” might be one element of the communication strategy that
distinguishes this effort from the more comprehensive AQRs undertaken in program countries. “Asset quality
data exercise” or “input data review” might be suitable titles.
 245

 Funding costs, which depend both on exogenous or macroeconomic conditions such

as the sovereign’s credit rating and on the bank’s own situation (Box 3); and

 Risk weighted assets, which may be affected by shifts in risk weights and write-offs
even if the overall balance sheet is static.

Box 3. EBA Stress Tests and Bank Funding Costs

EBA incorporated a cost of funding shock in their 2011 solvency stress test, which was
linked to sovereign stress. Specifically, EBA assumes that the banks were subject to static
balance sheets and faced stable wholesale and retail funding needs. Higher cost of funding in
the baseline and adverse scenarios could arise due to: higher short- and long-term interest
rates, increased banks’ spreads (which depended just on a bank’s respective sovereign’s
spreads), collateral value declines, and more expensive deposits. Funding costs did not
depend on the change in a bank’s projected situation in the course of the scenario.

An important funding and capital link was not considered in the 2011 exercise: As seen
during the financial crisis, the banks’ cost of funding is linked to the banks’ capital buffers.
Banks with lower solvency levels have either seen their funding costs sharply increase, or
market funding channels closed entirely. Some recent FSAPs have incorporated this feedback
mechanism, albeit somewhat crudely. In this context, a bank benefits from a reversal of this
negative feedback when it is recapitalized. Establishing a link between funding costs and
capitalization would benefit from panel data (and not only based on a cross section) given the
link is possibly dynamic and non-linear. For sensitivity analysis, higher hurdle rates that are
compatible with sustainable funding costs could be used. The 2011 exercise benefited from
the EBA liquidity risk assessment in 2011 in terms of cost of deposits for different sources
and maturities. In light of the ongoing financial fragmentation, the 2013 exercise would
benefit from further refining the cost of deposit channel. Ideally, the cost of funding
methodology used should take into account banks’ applications of their internal pricing
mechanisms, which often include hedging of funding cost changes.

Furthermore, the 2011 stress test considered only sovereign assets as collateral for
central bank and wholesale funding, and not other securities such as corporate or bank
bonds. Hence, the assumption of a drop in the sovereign asset value emanating from the
increased sovereign spreads did not cover a wide range of collateral assets. The sovereign
asset collateral value also did not include ECB haircuts but only the estimated bond price
haircut.
 246

I. Achieving Supervisory Orientation

The forthcoming stress testing exercise is meant to be mainly for supervisory purposes,
as opposed to the past emphasis on crisis management and the assessment of bank
capitalization. Translating this intention into the design and practice of the stress test, and
transcending the market perception of past tests, will require careful preparation. Market
participants and analysts are likely to compare results across banks and try to quantify
capitalization needs. If the capital needs are not “enough,” analysts may question the rigor of
the tests.

To achieve the supervisory purposes, the tests should yield recommendations for
supervisors as bank managers, and generate relevant indicators. Some of these
recommendations, which may have to remain confidential, might include indications of the
areas on which supervisors should focus their attention during the coming period
(e.g., lending practices in especially vulnerable sectors or sustainability of funding). To this
end, it may be useful to generate relatively detailed projections, for example, for loan quality
by sector and by country, capitalization by country, and profitability measures. Supervisory
colleges could then discuss the implementation of these recommendations.

The authorities intend in particular to use the exercise to evaluate banks’ plans to
comply with the evolving capitalization requirements under CRD IV. The approach is as
follows: each bank will provide its dynamic capital plan, which includes also related planned
adjustments among its assets and liabilities (e.g., a shift out of assets with high capital
weights or out of short-term market funding). Each will also provide data on its balance sheet
positions at the reference period, which will probably be end-2012, and projections of its
profit and loss under the given the provided baseline and stress scenarios and satellite model
guidance, but with a static balance sheet.191 A three-year projection period is envisaged. The
authorities, after checking plausibility and consistency, will substitute the projected losses,
etc. from the two scenarios (baseline and adverse) into the bank’s capital plan. The
authorities would then assess whether the bank’s capitalization level falls below, or close to,
the CRD IV minimum requirements, such as on the definition of capital and RWA, which are
progressively tighter over the projection period due to the gradual phase-in period of CRD
IV. If a bank’s plan looks precarious or based on implausible assumptions, the relevant
supervisor would demand a revision.

The use of a static balance sheet over three years may be justified mainly on the
grounds of tractability and the desire to facilitate comparability. The stress testing
exercises are already highly complex, and allowing balance sheets to change would greatly

191
The balance sheet is “static” in that it is not managed by the bank, but it will change as loan quality varies
and capital is accumulated or depleted. Furthermore, the authorities envisage incorporation of the ending of the
ECB’s LTRO and the replacement of this financing with more expensive market financing.
 247

add to the complexity: the methodology and consistency checks would need to be more
complex, because they would need to cover the capital plans. Also, peer reviews would be
less informative.

Yet, the static balance sheet approach has distinct risks and other drawbacks if the aim
is primarily supervisory. First, precisely because static balance sheets facilitate
comparisons (more so than with dynamic balance sheets), market analysts will be better able
to interpret the results as a “beauty contest” among banks, as they seek out those that look
comparatively or absolutely under-capitalized. Second, there is an inconsistency between the
treatment of the considerable number of banks under restructuring plans agreed in the context
of EU state aid rules, which will be assumed (as in the 2011 exercise) to implement those
plans, and the others. Third, there is an inconsistency between the macro projections, where
monetary and financial aggregates change over the envisaged three year stress testing
horizon, and the assumption of static balance sheets of monetary financial institutions.
Fourth, there may be instances where a static balance sheet is inconsistent with other
regulatory changes, such as those prompted by ESRB recommendations on foreign currency
lending. Finally and perhaps most importantly, projections using a static balance sheet may
not be very relevant for evaluating banks’ plans that involve significant balance sheet
adjustment. It is possible that a bank has a robust, plausible capital plan, that is consistent
with the plans of others and the macroeconomic forecast, but the plan looks inadequate when
projections from a static balance sheet are inserted into it. Were the supervisor to raise
objections and require action, the bank could argue that its original plan includes all the
additional action needed. Relevant in this connection is the fact that many banks are currently
being prompted by upcoming regulatory changes (CRD IV, CRR) and the LTRO phase-out
to aggressively adjust their balance sheets by de-risking activities and by decisively changing
their funding profile. Market analysts also facilitate such a balance sheet adjustment by
focusing not on the gradual CRD IV regulatory thresholds but on the fully phased in ones.

Furthermore, banks’ capital plans are likely to also include funding elements aimed at
dealing with LTRO exit and achieving compliance with the prospective Basel III LCR,
and any supervisory assessment should take account of these elements. Banks with an
estimated LCR shortfall have a number of ways in their funding plans to become
compliant.192 There is likely to also be some heterogeneity across European banks depending,
for instance, on their funding profile, risk appetite and geographical activities. The need to
achieve LCR compliance shortly after the projection period could imply that some European
banks will be further forced to cut back on short-term wholesale funding and increase
funding maturity, with consequences for their asset side (e.g., deleveraging and sale of non-
core business).

192
For instance, they can lengthen the maturity of their (unsecured) wholesale funding beyond 30 days, promote
deposits, reduce costly uncommitted credit lines or increase their proportion of liquid assets in their balance
sheets and deleverage on activities that are low interest yielding but are funded by short-term liquidity.
 248

The possibility of incorporating more “dynamic” elements into the balance sheet
projections should therefore be reviewed, and is encouraged, while comparability of the
CRD IV definition should be ensured.193 It may be reasonably easy to provide banks with
guidance on the evolution of major balance sheet components that are consistent with the
macroeconomic scenario (for example, aggregate growth in deposits and credit by country or
use of loan-to-deposit limits) and indicate that they should avoid strategies that rely on “deus
ex machina” (such as the sale of an unprofitable business at a handsome price). Such a
differentiation would also allow for the ongoing process of financial fragmentation and de-
integration. As in the case of the 2011 recapitalization exercise, banks should not be allowed
to optimize their RWAs. Adherence to such guidance would reduce the need to subject plans
to preliminary evaluation before the stress test is performed.194 To ensure comparability
across banks and jurisdictions, the stress testing CRD IV definitions and hurdle rates during
the phase-in period should allow for no national discretion.

More effort in assessing the robustness of results, including to the assumption of static
balance sheets, would contribute to their usefulness to supervisors and to their overall
credibility. Sensitivity tests might involve, for example, re-running top-down tests with
slight variations in the macro scenario or the satellite models, to see the extent to which
results vary (possibly in a non-linear manner). Top-down analysis could be used to quantify
the effects of changing balance sheet size and competition to reflect projected aggregate
changes (e.g., money supply and credit stock evolution), banks’ own plans, and the
consistency of these elements. It may also be worthwhile to run tests for sensitivity to
variations in input data.

Sensitivity to macroeconomic assumptions and projections needs to be assessed.

Macroeconomic assumptions in the baseline and adverse scenarios play a crucial role in
solvency stress tests, and can be key drivers for banks’ loan losses.195 It might be useful to use
the ECB/ ESRB top-down stress test for a country-specific sensitivity analysis of the adverse
macroeconomic scenario. This could provide a sensitivity of banks’ resilience to the severity,
or lack, of the adverse macroeconomic scenario, especially since a common scenario might
affect banks in specific jurisdictions in very diverse ways.

193
Note that the external banking stress test conducted for Spain was based on a dynamic balance sheet
assumption and banks’ capital plans.
194
In any case, assumptions need to be made consistent. For example, if the balance sheet is static, a well-
capitalized bank cannot be expected to retain any dividends.
195
For the 2011 EBA stress test, the EC provided the baseline scenario while the adverse scenario was given by
the ECB/ ESRB. EBA identified the microprudential risk factors and the ESRB and the ECB mapped them into
the macroeconomic scenarios.
 249

J. Future Priorities

Solvency and Structural Issues

As the situation of the banking sector changes and supervisory institutions evolve, it is
worth considering where best to allocate limited stress testing resources. There is already
a great deal of stress testing and simulations done not only by the EBA but also by the NSAs,
for their own stability analysis and supervisory purposes; and by financial institutions
themselves for risk management generally, internal capital adequacy assessment process
(ICAAP), and recovery planning/living will purposes. Some streamlining would be welcome.
EBA could also have some enhanced role on giving guidance to banks on their recovery
plan/living will stress testing.

Over the medium term, EBA could shift efforts to running tests on hitherto relatively
neglected topics such as structural issues and funding vulnerabilities. In this connection,
competing EBA, ECB/SSM and ESRB stress tests are to be avoided. Under the new SSM
architecture, EBA should continue to closely coordinate the EU-wide stress tests with ECB
and NSAs, and ensure quality control; the ECB should run supervisory stress tests for the
banks in the SSM; while ESRB should focus its contributions on macro-prudential issues,
such as the identification and calibration of systemic risk factors and the use of stress test
results in formulating policy advise. The EBA regulation gives it a mandate to oversee stress
testing; its comparative advantages lie in such areas as (i) providing benchmarks and satellite
models, especially for host country operations;196 (ii) ensuring that NSAs benefit from the
latest techniques and apply them with full rigor;197 and (iii) exercising its mandate to ensure
that best use is made of stress testing by NSAs, e.g., in setting supervisory priorities and in
evaluating banks’ recovery plans. Consistency in scenario building may sometimes be
desirable, but may be of lesser importance for supervisory purposes for many banks. In this
light, the EBA may wish to focus in 2014–15 on improving liquidity stress testing and its
integration with solvency tests, which is a relatively new area, rather than devoting so much
of its limited resources to another comprehensive solvency test during this period (see
below). Also, the EBA may have occasion to assess the use of stress testing by NSAs as part
of its peer review process.

It would be valuable to run stress tests and related simulations designed to incorporate
more long-term factors and generate lessons that relate more to structural issues. As
emphasized above, the European banking system faces a prolonged period of low interest
rates, possibly low growth, increased regulatory burden such as Basel III and CRD IV, and

196
The EBA is well placed to provide common benchmarks for the hosted operations of banks that come from
several home countries.
197
The EBA is already active in this area, as evidence by its guidance on ICAAP evaluations and review of
practices.
 250

demographic change, developments which will put pressure on profitability, the supply of
savings, competition, etc. Hence, the stress tests scenarios need to encompass a longer time
horizon; incorporate structural shifts (e.g., ongoing deleveraging and changes of bank
funding profiles) affecting the balance sheet and income; and emphasize more other metrics,
such as profitability, and changes in RWA. It should be noted that stress tests and simulations
are only one instrument in the toolkit to examine the structural challenges faced by banks,
and complement other quantitative and qualitative approaches. The ESRB would be the
leader for efforts in these areas—in addition to its analysis of more conjunctural issues and
nonbank sectors—which would be guided by the emerging consensus on best practice in
macro-financial stress testing (Box 4).

Box 4. Principles for Macro-financial Stress Testing

A recent Fund document has brought together principals that summarize good
practices and strategies for macro-financial stress testing (IMF 2012c):

 Define appropriately the institutional perimeter for the tests.

 Identify all relevant channels of risk propagation.
 Include all material risks and buffers.
 Make use of the investors’ viewpoint in the design of stress tests.
 Focus on tail risks.
 When communicating stress test results, speak smarter, not just louder.
 Beware of the “black swan.”

Stress tests that make full use of market data (such as those based on contingent claims
analysis, CCA) need to be developed further and used to complement balance sheet-
based tests, at least where such data are available.198 These methods, which are already
deployed and being further developed by the ECB and some national central banks, are
especially suited to capturing cross-sectoral and funding issues, for example, by treating
banks, nonbank financial institutions, and nonfinancial corporations on a consistent and
integrated basis, and by linking sovereign and bank balance sheets. Furthermore, these
models are intrinsically non-linear, and thus differentiate between behavior and pricing in
“normal” times and in under stress conditions.

Another area for attention over the medium term is the calibration of prudential
requirements. Many prudential requirements (such as the 8 percent capital adequacy
requirement under Basel II) seem to be largely the produce of historical accident rather than a
deliberate evaluation of costs versus (stability) benefits. While work in this area is
challenging, it would be comforting to know, say, the change in the proportion of banks that

198
See Gray and Jobst (2011) and Gray et al (2007) for details on the CCA approach.
 251

failure under a given shock as a capital requirement is varied. A thorough analysis that took
into effect structural implications might be suitable for a joint effort by the ESRB and the
EBA.

In this connection, stress tests might be used to investigate the stability effects of the
growth in “shadow banks.” The shadow banking sector is diverse, and some parts might be
of much greater systemic importance (for example, due to linkages to banks, or affects on
aggregate credit supply) than others. Even simple stress tests might shed light on how
important it might be to tighten the regulatory and supervisory framework for this sector.

K. Liquidity Stress Testing

The financial crisis has highlighted the need to better integrate solvency and liquidity
stress testing. A sharp rise in their euro and US dollar funding costs, or quantitative
rationing, was often the trigger for the failure of banks during the crisis, and for the
difficulties that many European banks continue to face. As mentioned above, EBA in
their 2011 stress test introduced a cost of funding shock, which, among others, was linked to
the sovereign debt spread. The EBA in 2011 conducted a less formal liquidity risk
assessment, which indirectly captured fire sales through collateral haircuts.199 Elsewhere, the
ECB in their recent Financial Stability Report has incorporated an explicit funding volume
shock and deleveraging path into the ECB macro stress testing framework (see annex 11 for
details). The IMF (2012a and d) has been also incorporating a dynamic deleveraging path in
their analysis.

In the medium term, EBA could intensify its work on liquidity stress testing, especially
in the context of the phasing in of detailed common reporting templates on maturity
mismatches, cost of funding, and asset encumbrance, as part of CRD IV. EBA already
has experience with liquidity stress testing especially from their 2011 cash flow based
assessment of European banks. Such a liquidity risk assessment would test the resilience of
European banks to various funding shocks (deposits, wholesale and off-balance sheet). It
would also consider the banks’ behavior to more limited liquidity support such as, for
instance, the tightening of central banks’ collateral requirements, and include risks from asset
encumbrance (box 5). The output could also feed into ESRB’s work stream on systemic risk
assessments. The starting point for EBA could be lessons learnt from the 2011 internal EBA
cash flow based liquidity risk assessment. EBA could provide guidance on liquidity stress
testing issues, and ensure some consistency of approaches by NSAs. EBA would likely need
to boost its staff resources as well as adjust its medium term work plan to incorporate such

199
The exercise was generally well-designed, and some of its features will be a useful in preparation for the
introduction of the LCR. Granular cash flow data including by currencies and maturity buckets, for broad
sample of European banks (54), was compiled and checked. Multiple scenarios capturing the banks’ main
liquidity risks and counterbalancing capacities were analyzed. On this basis, recommendations were conveyed
to banks through the NSAs.
 252

additional work on liquidity stress testing. EBA should also ensure the disclosure and
transparency of the reporting templates and the overall liquidity stress testing approach,
while safeguarding sensitive results.

Box 5. Asset Encumbrance and Liquidity Risk Assessments

Excessive asset encumbrance levels lower the resilience of a bank to further funding
shocks by constraining its access to funding backed by suitable collateral, and may
undermine investor confidence. It also subordinates other unsecured creditors such as
depositors. In such circumstances, a tightening of central bank collateral requirements can
reduce a bank’s unencumbered eligible collateral to dangerously low levels.

Based on a survey of 53 European banks, the ESRB finds a large dispersion among
banks, and significant constraints faced by a subset of banks. Increased ECB liquidity
provision has contributed to very high asset encumbrance levels among some banks,
especially in the periphery. To deal with excessive asset encumbrance levels, the ESRB
proposed action to improve related risk management in banks, enhance supervisory
monitoring on asset encumbrance, and market transparency, where the ESRB recommends
EBA to develop guidelines. For the longer run, the ESRB will consider whether to have a
formal encumbrance framework that may alleviate the pro-cyclicality of excessive asset
encumbrance.

A cash flow-based liquidity stress test, such as used by EBA in 2011, offers certain
advantages (see Schmieder et al, 2012, and the Annex 11). A cash flow-based module
along the lines of the 2011 internal EBA liquidity risk assessment or the forthcoming EBA
cash flow based maturity mismatch template allows running detailed liquidity analysis, and
hence it is well suited to capture a bank’s funding resilience and its liquidity risk bearing
capacity. Cash flow-based liquidity stress testing allows for detailed maturity buckets and
can be also adapted to different currencies. Liquidity risk exposure (net funding gap,
cumulated net funding gap) and liquidity risk bearing capacity are clearly separated in the
cash flow template. The template incorporates securities flows and ensures consistency
between cash-flows and securities flows. This is especially important given the role
unsecured and secured wholesale funding play for many large banks. Off-balance sheet
activities such as FX swaps or credit and liquidity lines can be easily incorporated as well.

Weaknesses of the cash flow approach include the high data intensity as well as initial
set-up costs. While banks typically use a cash flow-based approach for internal liquidity
monitoring and liquidity stress testing, regulatory liquidity ratios are often based on stock
accounting data with often less data granularity than the cash flow based templates. The
phase-in of EBA cash flow based maturity mismatch templates will provide regulators and
banks with standardized templates that would need to be regularly filled out and reported. As
with the EBA solvency stress tests, it is suggested that EBA staff have access with NSA
 253

colleagues to banks for a consulting/feedback process, and direct interaction with banks’
liquidity risk managers, so as to facilitate the roll-out of such cash flow templates.
 254

ANNEX 11. APPROACHES TO LIQUIDITY STRESS TESTING

A. Literature Review

There have been a number of liquidity stress testing approaches in the literature with a
few studies attempting to link solvency and funding risk.

 Schmieder et al. (2012) provide an Excel based framework that allows running liquidity
tests informed by banks’ solvency conditions, and to simulate the increase in funding
costs resulting from a change in solvency. Drawing on this framework, Hesse, Salman
and Schmieder (2013, forthcoming) focus on scenario design and building integrated
macro-financial scenarios that take into account various dimensions of potential shocks
at the same time - solvency and liquidity risks in particular.

 The IMF GFSR (2012a, 2012b) conducts a dynamic deleveraging analysis which
includes an assumed funding shock on deposits and wholesale funding for the European
banks in the sample.

 The Systemic Risk-adjusted Liquidity (SRL) model of Jobst (2012) combines option
pricing with market information and balance sheet data to generate a probabilistic
measure of the frequency and severity of multiple entities experiencing a joint liquidity
event. It links a firm’s maturity mismatch between assets and liabilities impacting the
stability of its funding with those characteristics of other firms, subject to individual
changes in risk profiles and common changes in market conditions.

 Van den End (2008) at the Dutch Central Bank developed a stress testing model that
tries to endogenize market and funding liquidity risk by including feedback effects that
capture both behavioral and reputational effects. A number of central banks and bank
supervisors have been successfully using the Monte Carlo framework of Van den End
(2008).

 Wong & Hui (2009) from the Hong Kong Monetary Authority sought to explicitly
capture the link between default risk and deposit outflows. Their framework allows
simulating the impact of mark-to-market losses on banks’ solvency position leading to
deposit outflows; asset fire sales by banks is evaporating and contingent liquidity risk
sharply increases.

 Barnhill & Schumacher (2011) developed a more general empirical model,

incorporating the previous two approaches that attempts to be more comprehensive in
terms of the source of the solvency shocks and compute the longer term impact of
funding shocks.
 255

 Another attempt to integrate funding liquidity risks and solvency risk is the Risk
Assessment Model for Systemic Institutions (RAMSI) developed by the Bank of
England (Aikman et al., 2009). The framework simulates banks’ liquidity positions
conditional on their capitalization under stress, and other relevant dimensions, such as a
decrease in confidence among market participants under stress.

 The current Basel Research Task Force on liquidity stress testing is also looking at the
solvency and liquidity link.

B. Integrating Liquidity and Solvency Risks and Bank Reactions in Stress Tests200

Banks have numerous ways to react to credit and funding shocks. High-quality capital
and profits are usually the first line of defense, and retained earnings can help buffer
banks’ capital levels. Banks have an inherent capacity to generate liquid assets by using
high-quality eligible securities as collateral for market or central bank funding if interbank
markets freeze. As seen post-Lehman, fire sales of securities are also an option, but at a
considerable cost in an environment of sharply declining asset prices. Deleveraging,
especially targeted at assets with higher risk weights, is also a way to raise capital adequacy
ratios by reducing RWAs. In practice, banks have been using a combination of these, as well
as other hybrid measures, ranging from debt-to-equity conversions to issuance of convertible
bonds to optimizing risk-weighted assets, to react to shocks.

Incorporating banks’ reactions to shocks is a critical input into the design of

informative stress tests, especially over longer time horizons. This, however, requires
modeling solvency and liquidity shocks in a coherent manner because first, when banks react
to financial stress, the source of the shock (solvency or liquidity) is not always clear; and
second, the measures banks take in reaction to these shocks have both capital and liquidity
aspects that are not easy to disentangle.

A relatively simple (but somewhat ad hoc) way to integrate solvency and liquidity
shocks is to conduct two-round stress tests, with a bottom-up (BU) first round and a
top-down (TD) second round. If, for example, the majority of banks report in the BU first-
round test asset sales of particular asset classes in response to the shock, the TD second-
round test could impose haircuts on those assets; if banks report that they would discontinue
reverse repos, the analysis would incorporate a reduction in repo roll-overs. The
quantification of these haircuts or roll-over rates could be based on historical information,
cross-country experience, or expert judgment.

200
This section draws on IMF, 2012c.
 256

C. Liquidity Risks Analysis by Authorities

The ECB does not conduct stand-alone liquidity stress tests of European banks. It
indirectly does incorporate funding and liquidity stress via its contribution to the EBA
solvency stress tests where a funding cost shock is assumed (see below). Furthermore, in the
June 2012 ECB Financial Stability report, an explicit funding volume shock is incorporated
into the ECB macro stress testing framework. Specifically, in a contagion and deleveraging
scenario originating from non EU/IMF program countries such as Belgium, France, Italy and
Spain, it is assumed that European banks can only refinance 50 percent of their wholesale
funding which matures in 2012-2013. Deposit withdrawal rates range between 0 percent for
countries with an AAA rating to 20 percent for countries below investment grade. The
scenario embeds a deleveraging path whereby banks fire sell more liquid assets to cover the
wholesale funding gap, and around one third of this gap is covered by loan reductions.
Similarly, losses in deposits lead to loan deleveraging. The scenario also includes differential
increases in interest rates as well as stock market declines. Banks’ solvency positions are
calculated by changes in their profits, credit risk parameters and RWAs. The scenario leads
to a 0.2 euro area GDP decline in 2012 and of 0.7 in 2013, both relative to baseline of -0.3
and 1 percent, respectively.

Findings for the contagion and deleveraging scenario suggest a sizable drop of banks’
capital. In aggregate, from a baseline average core Tier 1 ratio of 9.3 percent at end 2013
(end–2011 is 8.7 percent) the scenario causes a capital drop to 7 percent. If one adds an
additional domestic demand shock, core Tier 1 capital further drops to 6.4 percent. If one
adds the EBA mandated temporary sovereign buffer from the recapitalization exercise to the
contagion and deleveraging scenario, the core tier capital increases from 7 to 7.4 percent. The
macro stress test does not consider the LTRO effect on reducing banks’ funding costs which
the ECB estimates at between 0.2-0.6 percent of core Tier 1 capital. The ECB mentions that
the analysis does not incorporate any second round effects on the banks from the limited
bank funding availability.

The ECB approach does not take into consideration its collateral and haircut policies or
banks’ heterogeneous asset encumbrance levels. As mentioned above, asset encumbrance
levels for peripheral banks have significantly increased during the ongoing financial crisis
with many banks also resorting to the ECB LTROs and benefiting from collateral loosening
by increasing pledging credit claim type collateral subject to high and conservative ECB
haircuts. Banks that already suffer from high asset encumbrance levels have diminished
counterbalancing capacity to withstand severe liquidity stress so the design of liquidity stress
tests should ideally also include information about banks’ asset encumbrance or liquidity
from eligible collateral ex post haircuts.
 257

IMF liquidity stress tests for FSAPs have often been based on a stock and not on a cash
flow approach.201 FSAP stress testers would then use stock balance sheet information
(sometimes not decomposed by maturity buckets) and conduct a bank run type analysis on
deposits and wholesale funding whereby banks use liquid assets as their counterbalancing
capacity. In some FSAP cases, the Basel III liquidity ratios LCR (which embeds a quasi 30-
day cash flow funding run) and the NSFR were simulated. For instance, in the Spain and UK
FSAPs, implied cash flow tests (sensitivity of banks to outflow of funding over 5 and 30
days) were conducted (IMF, 2011a and 2012b). Both FSAPs also included a simulation of
banks’ Basel III liquidity measures. The Germany FSAP also included an implied cash flow
analysis of funding shocks to consecutive periods (IMF, 2011b), while the France FSAP
included a cash flow-based liquidity stress test using maturity buckets (IMF, 2012e). In the
U.S. FSAP, the liquidity stress test was based on a basic analysis of the maturity mismatches
of major banks (same coverage as the Pillar 1 of solvency analysis) and assumes that banks
are unable to refinance maturing loans (IMF, 2010). The forthcoming paper by Hesse, Jobst,
Ong and Schmieder (2013) provides an overview of IMF FSAP liquidity stress testing
exercises.

The cash flow based liquidity stress tests of the Oesterreichische Nationalbank (OeNB)
are relatively advanced. During the financial crisis in 2008, the Austrian Market Authority
(FMA) and OeNB required banks to submit a weekly cash flow based report based on a new
liquidity reporting template (see OeNB, 2009, as well as Schmitz and Ittner, 2008, for more
details). The cash flow approach is forward-looking by including banks’ contractual cash
out- and inflows as well as banks’ expected counterbalancing capacity. The template also
distinguishes between different currencies. The difference between cash flow tests run by
banks and those run by the OeNB for monitoring purposes is that the latter requires
standardized templates, which then allows simulating the impact of common shocks based on
a uniform method. A key prerequisite to carry out cash flow based liquidity tests is access to
a wide range of data on contractual cash flows for different maturity buckets and possibly
behavioral data based on banks’ financial/funding plans.

D. Basel III and Liquidity Stress Testing

The Basel III liquidity coverage ratio (LCR) amount to a quasi-liquidity stress test, and
its phase in period from 2015 could compel many European banks to close their current
funding gaps. The BCBS has published basic principles of liquidity management, essentially
guidance on the risk management and supervision of liquidity risks for banks and supervisors
(BCBS 2008). Overall, the LCR can be viewed as a 30-day medium shock stress test. With
the publication of the final Basel III LCR rules (see BIS, 2013), banks with an estimated

201
Jobst, Ong and Schmieder (forthcoming) provide an overview of FSAP solvency stress testing, and
Schmieder, Hasan and Puhr (2011) offer a flexible stress testing framework.
 258

LCR shortfall have a number of ways in their funding plans to become compliant.202 The Net
Stable Funding Ratio (NSFR) is currently being reviewed by the Basel Committee.

202
For instance, they can lengthen the maturity of their (unsecured) wholesale funding beyond 30 days, promote
deposits, reduce costly uncommitted credit lines or increase their proportion of liquid assets in their balance
sheets and deleverage on activities that are low interest yielding but are funded by short-term liquidity. There is
likely to also be some heterogeneity across European banks depending, for instance, on their funding profile,
risk appetite and geographical activities.
 259

References

Aikman, David, Piergiorgio Alessandri, Bruno Eklund, Prasanna Gai, Sujit Kapadia,
Elizabeth Martin, Nada Mora, Gabriel Sterne and Matthew Willison, 2009,
―Funding Liquidity Risk in a Quantitative Model of Systemic Liquidity, Bank of
England Working Paper, June, No. 372.

Barnhill, Theodore and Liliana Schumacher, 2011, “Modeling Correlated Systemic Liquidity
and Solvency Risks in a Financial Environment with Incomplete Information,” IMF
Working Paper No. 11/263.

Basel Committee on Banking Supervision (BCBS), 2008, Principles for Sound Liquidity
Risk Management and Supervision (“Sound Principles.”
www.bis.org/publ/bcbs144.htm.

———, 2012, Results of the Basel III monitoring exercise as of December 31, 2011,
September 2012.

———, (2013) “Summary Description of the LCR,” Basel Committee on Banking

Supervision, January 6, 2013, in Basel, Switzerland.

European Banking Authority, 2012, Results of the Basel III monitoring exercise based on
data as of December 31, 2011, September 2012.

Gray, Dale F, and Andreas Jobst, (2011), “Modeling Systemic Financial Sector and
Sovereign Risk,” Sveriges Riksbank Economic Review, September.

Gray, Dale F., Robert C. Merton and Zvi Bodie, (2007), “Contingent Claims Approach to
Measuring and Managing Sovereign Credit Risk, Journal of Investment Management,
Vol. 5, No. 4, pp. 5-28.

Hardy, Daniel, and Heiko Hesse, 2013, “Stress Testing of Banks,” European FSAP,
Technical Note. (Washington: International Monetary Fund). A shorter version was
also published in VOX.

Hesse, Heiko, Andreas A. Jobst, Li Lian Ong, and Christian Schmieder, (2013), “An IMF
Framework for Macroprudential Liquidity Stress Testing: Application to S-25 and
Other G-20 Country FSAPs,” IMF Working Paper, forthcoming.

Hesse, Heiko, Ferhan Salman, and Christian Schmieder, (2013), “Running Integrated Risk
Scenarios to Identify Financial Vulnerabilities,” forthcoming.
 260

International Monetary Fund, 2010, “United States: Publication of Financial Sector

Assessment Program Documentation—Technical Note on Stress Testing,” IMF
Country Report No. 10/244 (Washington, July), available at
http://www.imf.org/external/pubs/ft/scr/2010/cr10244.pdf.

———,2011a, “United Kingdom FSAP Update: Stress Testing the Banking Sector Technical
Note,” IMF Country Report No. 11/227 (Washington, July), available at
http://www.imf.org/external/pubs/ft/scr/2011/cr11227.pdf.

———, 2011b, “Germany: Technical Note on Stress Testing,” IMF Country Report No.
11/371, (Washington, December), available at
http://www.imf.org/external/pubs/ft/scr/2011/cr11371.pdf.

———, 2012a, Global Financial Stability Report, April 2012.

———,2012b, “Spain: Financial System Stability Assessment,” IMF Country

Report No. 12/137 (Washington, July), available at
www.imf.org/external/pubs/ft/scr/2012/cr12137.pdf.

———, 2012c, “Macro-financial Stress Testing—Principles and Practices,” Policy Paper,

August 2012.

———, 2012d, Global Financial Stability Report, October 2012.

———, 2012d, “France: Financial System Stability Assessment,” IMF Country

Report No. 12/341 (Washington, July), available at
http://www.imf.org/external/pubs/ft/scr/2012/cr12341.pdf.

Jobst, Andreas A., 2012, “Measuring Systemic Risk-Adjusted Liquidity (SRL) — A Model
Approach,” IMF Working Paper No. 12/209.

Jobst, Andreas A., Li Lian Ong and Christian Schmieder, forthcoming, “An IMF Framework
for Macroprudential Bank Solvency Stress Testing: Application to S-25 and Other G-
20 Country FSAPs,” IMF Working Paper.

Oesterreichische Nationalbank (OeNB, Austrian Nationalbank), 2009, Financial Stability

Report 18. December.

Schmieder, Christian, Puhr, Claus and Maher Hasan, 2011, “Second Generation Applied
Stress Testing—Solvency Module,” IMF Working Paper No. 11/83 (Washington:
International Monetary Fund).
 261

Schmieder, Christian, Hesse, Heiko, Neudorfer, Benjamin, Puhr, Claus, and Stefan W.
Schmitz, 2012, “Next Generation System-Wide Liquidity Stress Testing,” IMF
Working Paper No. 12/03.

Schmitz, Stefan, and Andreas Ittner. 2007. “Why central banks should look at liquidity risk.”
In: Central Banking XVII (4). May. 32–40.

Van den End, Jan Willem, 2008, “Liquidity Stress Tester: A macro model for stress-testing
banks’ liquidity risk, “Dutch National Bank Working Paper No. 175, May 2008.

Wong, Eric and Cho-Hoi Hui, 2009, “A liquidity risk stress- testing framework with
interaction between market and credit risks,” Hong Kong Monetary Authority
Working Paper 06/ 2009.
 262

XI. HOW TO CAPTURE MACRO-FINANCIAL SPILLOVER EFFECTS IN STRESS TESTS?, WITH

FERHAN SALMAN AND CHRISTIAN SCHMIEDER, 2014203

One of the challenges of financial stability analysis and bank stress testing is how to
establish scenarios with meaningful macro-financial linkages, i.e., taking into account
spillover effects and other forms of contagion. We come up with an approach to simulate the
potential impact of spillover effects based on the “traditional” design of macro-economic
stress tests. Specifically, we examine spillover effects observed during the financial crisis and
simulate their impact on banks’ liquidity and capital positions. The outcome suggests that
spillover effects have a highly non-linear impact on bank soundness, both in terms of
liquidity and solvency.

A. Introduction

Stress testing has garnered broad attention during recent years, which has spurred numerous
conceptual developments.204 Yet, overarching approaches to establish macro-financial
linkages, and explicitly capture the non-linearity of shocks (originating from spillover effects
and other types of contagion) are still evolving. Such linkages have seen a particularly
significant growth during the last decade (e.g. Frank et al, 2008) and are therefore an
important dimension to be captured by meaningful empirical analysis. This paper focuses on
the design of stress tests to capture spillover effects and demonstrates the potential impact
based on a case study.

The first part of the paper deals with the establishment of macro-financial scenarios which
are explicitly informed by spillover effects. Scenario design for macroeconomic stress tests is
typically based on an “indirect approach” (Jobst and others, 2013; see Figure 11.1): (i) first,
economic and financial variables are estimated conditional on a macroeconomic scenario; (ii)
in the second step, the trajectories of the economic and financial variables are translated into
bank solvency and liquidity205 measures based on so-called “satellite” or “auxiliary” models.
Three approaches have commonly been used to predict economic and financial variables
under stress (see Foglia 2009): (i) a structural econometric model; (ii) vector autoregressive
methods; and (iii) pure statistical approaches. The satellite models commonly take the form
of (panel) regression models. The “direct approach” is based on projections of the actual
solvency and liquidity parameters without an explicit link to the state of economic and
financial variables. While this approach could be equally meaningful in terms of the outcome
of stress tests, it does not allow for a detailed story-telling and can underestimate the
importance of non-linear macro-financial factors for bank-specific stress tests.
203
This chapter is based on Hesse, Salman and Schmieder (2014).
204
For work on stress testing at the IMF, for example, see Jobst and others (2013).
205
For liquidity stress tests, most tests have typically relied on the “direct” approach.
 263

Modeling contagion effects and their impact typically constitute a challenge (see Jobst and
others 2013, for example). By definition, spillover effects and other dynamic contagion
effects are implicitly captured in past data, but not necessarily if one uses structural
econometric models - usually perceived as being “best practice.” Even if potential spillover
events are captured in past data, this data might not be representative for a future scenario if
e.g. linkages between economies and banks have become gradually more intense over time.
In this study, we focus on spillover effects originating from the recent sovereign debt crisis.
Other spillover catalysts could be, for instance, a macroeconomic downturn in a major world
economy as well as the failure of a large financial institution such as in the case of Lehman
Brothers.

We aim to come up with a stress testing approach that captures spillover effects in detail. Our
solution is an amended version of the indirect approach: the starting point is to establish a
macroeconomic scenario, typically not informed by potential spillover effects, at least not
explicitly. In the second step, the potential marginal increase of stress due to spillover effects
is estimated by translating the spillover effects into reduced output paths, i.e., an adverse
macroeconomic scenario.

In terms of the stylized design of macro-economic stress tests (Figure 45), we thereby
implicitly incorporate a quasi-feedback loop into the linear design of traditional stress tests –
through a sensitivity type approach.206 The approach could also include a test for interbank
bank contagion, as shown in Figure 45. We build on previous IMF work to establish an
explicitly iterative process, i.e., establish a scenario informed by initial spillover effects based
on a structural econometric approach, compute the impact on banks’ solvency parameters, re-
compute the resulting spillover effects and feed them back to the structural model etc. until
an equilibrium is reached.207 The approach presented herein uses proxies for the “ultimate”
impact of spillovers for different advanced and emerging economies conditional on the
evolution of sovereign spreads in the Euro area periphery (that serves as the stress catalyst).
Dynamic effects can also be captured via “direct” approaches, as done by Jobst and Gray
(2013), for example, but renders the outcome a reduced-form type.208

Specifically, we infer from market data the magnitude of sovereign spread spillovers effects
resulting from an increase in peripheral EU sovereign debt spreads, while controlling for

206
Further information on macroeconomic scenarios used for FSAPs can be found in Jobst and others (2013).
207
At the IMF, such analyses were carried out by combining the work of Schmieder, Puhr and Hasan (2011)
and Vitek and Bayoumi (2011) as part of early warning analysis and vulnerability exercises. It should be noted
that running such an approach requires close cooperation between staff running macroeconomic forecasts and
staff simulating the impact of stress at the bank level (typically done by financial stability departments).
208
See also IMF (2011, 2012) for further information on related work.
 264

changes in the market sentiment (i.e., risk aversion) and macroeconomic factors. Using
market data, we seek to capture point-in-time and dynamic time series’ effects, while
recognizing the limitations of using market data, i.e., that they might not necessarily “only”
reflect underlying vulnerabilities and risks. The translation of sovereign spread spillovers into
a loss of output is based on recent work at the IMF (Vitek and Bayoumi, 2011).

Two approaches are used to capture the spillover effects in sovereign debt markets: panel
regressions and a GARCH model. The panel regressions, which are used to establish an
“average” impact of spillover effects during periods of stress on AM and EM countries,
respectively, suggest that increasing sovereign risk in the Euro periphery was a major driving
force behind spillover effects. As expected, risk aversion, measured through changes in the
VIX and high yield spreads, is found to increase during periods of financial stress, exhibiting
a non-linear pattern. Country-specific macroeconomic factors also matter, but to a lesser
degree, and their impact does not appear to change significantly under periods of stress.

Figure 45. Stylized Design of Stress Tests

Source: Authors

GARCH models were run to obtain more granular spillover effects, such as the country-
specific co-movements between peripheral European GIIPS209 sovereign debt spreads and the
corresponding spreads in the banks’ home countries (i.e., the 25 most systemically important
209
GIIPS refers to Greece, Ireland, Italy, Portugal and Spain.
 265

financial systems, the “S-25” sample) for specific points in time. The study reveals
significant differences in terms of the spillovers across countries, with a higher impact
observed for most core Euro area countries (in particular during peak periods of the crisis)
than for Scandinavian countries, Switzerland, the UK and most non-European countries. The
findings also show a flight-to-quality element, i.e., a negative co-movement of GIIPS spreads
with German Bunds and U.S. Treasuries.

In the second part of the paper, we illustrate how the established spillover effects would feed
through to banks based on a case study for 154 large international banks from the “S-25”
country sample. The impact of different degrees of spillover on banks’ solvency and liquidity
positions is compared with baseline type conditions (which corresponds to realized stress
scenarios in recent years unlike in “normal” times). Stress at the bank level is simulated
based on a recently developed IMF stress testing framework for liquidity (Schmieder, Hesse,
Neudorfer, Puhr and Schmitz, 2012) and benefits from work on solvency (Schmieder, Puhr
and Hasan, 2011; Hardy and Schmieder, 2013), which together allow running integrated
solvency and liquidity scenarios.210

The outcome suggests that spillover effects have a highly non-linear impact on bank
soundness, both in terms of liquidity and solvency. It is thereby shown (once more) that the
design of stress scenarios is a highly crucial element of stress testing, and is sensitive with
respect to the outcome of stress tests.211 The magnitude of the impact on bank solvency and
liquidity could serve as a benchmark for other studies, while recognizing that future spillover
channels could be highly different, both in terms of direction and magnitude. In this sense,
our study could help to identify potential systemic vulnerabilities ex ante, a role that stress
tests have not necessarily played in the past for a number of reasons (see Borio, Drehmann
and Tsatsaronis 2012, for example).

The paper is organized as follows. Section B investigates financial spillovers at the sovereign
and bank level, based on panel regressions and a GARCH model framework.
Section C provides a brief overview of the stress testing framework used to simulate the
impact of spillover effects on bank liquidity and solvency. Section D shows the impact of
different degrees of spillover based on a case study. Finally, section E concludes and offers
some avenues for future research. The Annex also shows an illustrative country example.

210
The frameworks were developed in the context of recent FSAPs and IMF technical assistance, extending the
seminal work of Čihák (2007), and drawing upon work at the Austrian National Bank (OeNB).
211
See also Taleb and others (2012) how to test the sensitivity (i.e., non-linearity) of the outcome of stress tests.
 266

B. Financial Spillovers from the Euro periphery to the Rest of the World

Panel Approach

Financial market linkages across economies have grown significantly in recent decades,
which was felt strongly when the financial crisis started in 2008 with the failure of Lehman
Brothers, and later continued to become a sovereign debt crisis especially in the European
periphery. AM financial spillovers have been a dominant determinant of AM and EM
financial soundness during the previous years.

Recent studies identified three important factors for spillover effects (e.g., Caceres and
Unsal, 2011): (i) a stress spillover catalyst – in this study AM sovereign debt yields; (ii) risk
aversion in global markets; and (iii) country-specific risk factors.

Herein, we sought to establish benchmark parameters to simulate spillover effects at the bank
level. Initially, we construct a risk premium variable for our sample of 35 countries.212 The
risk premium is the spread between 10 year domestic treasuries to U.S. Treasuries for non-
European AM countries, to German Bunds for AM countries in Europe, and to the JP
Morgan Emerging Markets Bond Index (EMBI) for the EM countries.213

Based on random effects’ panel regressions the sovereign spreads are regressed on three sets
of peripheral spreads: average spreads for (i) the European peripheral countries (GIIPS); (ii)
for the GIP (Greece, Ireland, Portugal); and (iii) for IT-ES (Italy and Spain);
Risk aversion) is identified by two variables, high yield spreads and the VIX. The former is
the difference between yields to maturity of Moody’s Aaa rated and Baa1 rated U.S.
corporate bonds. The latter is the implied volatility for S&P 500 index options. Trade
openness, liquidity (proxied by M2 to GDP and the level of reserves to GDP), inflation rates,
GDP growth, the current account, the level of public debt and deficits to GDP ratios are used
as macroeconomic control variables to capture country-specific cyclical effects.

The regressions are estimated for two time periods based on quarterly data: (i) 2006–2012
and (ii) 2008–2012. The choice of the two sample periods is meant to capture the impact of
the systemic stress.

The results (displayed in Tables A12.1-2 in Annex 12) present various model specifications
considered useful to identify drivers of spillover stress and their actual impact, respectively.

212
The sample of countries includes Australia, Austria, Belgium, Brazil, Canada, China, Cyprus, Denmark,
Finland, France, Germany, Greece, Hong Kong SAR, Hungary, India, Ireland, Italy, Japan, Korea,
Luxembourg, Malta, Mexico, Netherlands, Norway, Poland, Portugal, Russia, Singapore, Slovenia, Spain,
Sweden, Switzerland, Turkey, United Kingdom, and the United States.
213
The panel regressions adjust for exchange rate changes.
 267

Using the sovereign debt spreads of the 35 sample countries as the dependent variable, Table
A12.1 shows the outcome for 2006–2012 and Table A12.2 for 2008–2012:

The results confirm previous studies in that all three factors – i.e., a catalyst, risk aversion
and country-specific factors are actually important to explain financial stress (measured in
terms of sovereign spreads), at least for the current financial crisis:

 Increasing sovereign risk in the Euro periphery was found to be a catalyst for
spillover effects,
 The global perception of risk magnifies stress conditions as do expected future
interest rates;
 Country-specific macroeconomic factors also matter, but to a lesser degree.
 While the impact214 of country-specific factors does not appear to change
significantly under stress, the impact of the former two factors is higher during 2008-
2012, i.e., in the period covering only the crises years (compared to the full sample
period).

For the longer sample period (i.e., 2006–2012) a one percentage point change in Euro
periphery sovereign spreads (i.e., GIIPS and GIP) translates into a 0.2–0.3 percentage point
change of sovereign debt spreads in the 35 sample countries (Table A12.1). Global risk
aversion (measured by changes in high yield spreads) has an even higher impact - a one
percentage point change in high-yield spreads translates into around 0.6 percentage point
change in sovereign spreads. As global risk aversion and high-yield spreads are highly
correlated during episodes of stress, the joint impact on the peripheral spreads is exacerbated
– which is illustratively in a comparison of the coefficients in Tables A12.1 and A12.2. The
transmission of risk premium shocks from Italy and Spain to the countries in the sample is
more pronounced than for the GIPs. Depending on the model specification the availability of
domestic liquidity and trade openness also contribute to some degree to spillovers.215

The outcome for the crisis period only (covering the years from 2008–2012, Table A12.2)
indicates that the coefficients for all three major drivers, i.e., European periphery shocks,
global risk aversion, as well as the slope of the US yield curve are higher than for the period
including pre-crisis years (Table A12.1). A one percent shock to Euro periphery spreads
translates into a 0.5 percentage point increase in the risk premium of the 35 sample countries
if the shock originates in the GIPs and a one percentage point increase in spreads if it
originates in Italy and Spain. Hence, it seems that the size of the peripheral European country

214
Measured in terms of the R-squared and the actual coefficients.
215
For robustness check, a separate set of regressions were run to estimate the impact of expectations of higher
interest rates, represented by the slope of the US Treasury yield curve on the global risk premium. Results
indicate that a steepening of the curve implies higher costs of borrowing for the periphery countries.
 268

determines the size of spillovers, as expected. Moreover, global risk aversion shocks also
translate almost one-to-one into spreads.

C. DCC GARCH Approach

The panel regression approach provided the average spillover effect on countries’ sovereign
spreads. Below, we complement the above by estimating country-specific daily co-
movements, in order to differentiate more between countries, and to come up with the range
of the potential spillover impact observed over time. We use a multivariate GARCH
framework for the estimation, which allows for heteroskedasticity of the data and a time-
varying correlation in the conditional variance. Specifically, the Dynamic Conditional
Correlation (DCC) specification by Engle (2002) is adopted, which provides a generalization
of the Constant Conditional Correlation (CCC) model by Bollerslev (1990).216 The DCC
GARCH models are estimated in first differences to account for the non-stationarity of the
variables in the crisis period.

These econometric techniques allow us to analyze the daily co-movement of the GIIPS
spreads and the sovereign bond spreads of our sample of AMs and EMs. The GIIPS spreads
are included into the model as a conditioning variable, as is the VIX. The methodology is
therefore closely aligned to the one of the panel regression and further explained in Annex
13.

We choose as the sample period daily data from 2007 to end August 2012, with a view to
cover the full crisis period. As before, for the European AMs we measure the risk premium
of 10-year instruments as the difference between the average GIIPS spread as well as those
of the domestic treasuries to German Bunds. For the non-European countries, the spread to
the 10-year U.S Treasury bonds is calculated and for EM countries we use the EMBI Global
spread and the HSBC Asian U.S. Dollar spread for Asian countries.

As expected, our findings suggest that the spread between GIIPS to German Bunds exhibits a
higher degree of co-movement with the risk premia for European countries than non-
European countries (Figures 46–49). In particular, implied DCC GARCH correlations with
the GIIPS spread were as high as 0.7–0.8 for Austria, Belgium, France, and the Netherlands
during episodes of systemic stress (Figure 46, upper panels). In contrast, the GIIPS co-
movement with the UK spread to German Bunds is relatively low and oscillates between 0
and 0.2, while the model implied correlation with the Swiss spreads reaches a maximum of
0.4 (Figure 46, left hand panel at bottom). The results also show that the spreads of the

216
Given the high volatility movements during the recent financial crisis, the assumption of constant conditional
correlation among the variables in the CCC model is not very realistic especially in times of stress where
correlations can rapidly change. Therefore, the DCC model is a better choice since correlations are time-
varying.
 269

Scandinavian countries, namely Denmark, Norway, Sweden, and Finland (with higher
average levels though),217 on average exhibit a lower co-movement with the GIIPS spread
than their continental European peers (Figure 46, right hand panel at bottom). The outcome
does also suggest a constant level of stress, with some easing towards the end of the
observation period, a finding which also applies to the non-European sovereigns.

Co-movements of the GIIPS spread with Australian and Canadian spreads (relative to U.S
Treasury bonds) are rather low with implied correlations up to 0.2 (Figure 47). Looking at
the Asian countries Hong Kong, Japan, and Singapore shows a somewhat higher correlation
with the GIIPS spread of up to 0.3 and with one jump to 0.4. In terms of EM countries,
results suggest that China’s co-movement with the GIIPS spread is rather subdued compared
to the other EMs Brazil, Mexico, Russia, and Turkey (Figure 48). Out of this EM sample,
Turkey has the highest implied correlation with the GIIPS during episodes of system stress at
up to 0.6.

Since the onset of sovereign debt crisis by 2009, the average GIIPS interest rates exhibit a
negative correlation with both the German Bund and U.S. Treasury interest rates (Figure 49).
Since 2009, the implied correlation has turned negative for both countries, with lows at -0.4
(US) and -0.6 (Germany), indicating a sudden flight to safety, in line with other recent
studies (IMF 2011, for example).

217
Finland is the only Euro area country within the sample, which seems to explain the higher level of
correlations.
 270

Figure 46. Estimated GARCH Correlations GIIPS with European

Countries
0.9 0.8
0.8 0.7
0.7 0.6
0.6
0.5
0.5
0.4
0.4 GIIPS-Belgium GIIPS-Austria
0.3
0.3 GIIPS-France GIIPS-Netherlands
0.2 0.2

0.1 0.1

0 0
-0.1 -0.1
-0.2

0.5 0.6
0.45
0.5
0.4
0.35
0.4
0.3
GIIPS-Denmark
0.25 0.3
GIIPS-Switzerland GIIPS-Finland
0.2
GIIPS-UK 0.2 GIIPS-Norway
0.15
0.1 GIIPS-Sweden
0.1
0.05
0 0
-0.05

Source: Bloomberg and Authors’ Calculations

Figure 47. Estimated GARCH Correlations GIIPS with Non-European Countries

0.5 0.25

0.4 0.2

0.3
0.15

0.2 GIIPS-HK
GIIPS-Australia
GIIPS-Japan 0.1
GIIPS-Canada
0.1 GIIPS-Singapore
0.05
0
0
-0.1

-0.2 -0.05

Source: Bloomberg and Authors’ Calculations

 271

Figure 48. Estimated GARCH Correlations GIIPS with EM Countries and Korea

0.6 0.7

0.5 0.6

0.5
0.4
0.4
0.3 GIIPS-Brazil GIIPS-Poland
0.3
GIIPS-China GIIPS-Russia
0.2
GIIPS-Mexico 0.2
GIIPS-Turkey
0.1 0.1

0 0

-0.1
-0.1
-0.2

0.3

0.25

0.2

0.15

0.1 GIIPS-India

0.05 GIIPS-Korea

-0.05

-0.1

-0.15

Source: Bloomberg and Authors’ Calculations

Figure 49. Estimated GARCH Correlations GIIPS with Germany and the U.S.
1

0.8

0.6

0.4

0.2 GIIPS-Germany

0 GIIPS-U.S.

-0.2

-0.4

-0.6

-0.8

Source: Bloomberg and Authors’ Calculations

Note: Unlike the other GARCH models, the average GIIPS interest rates are taken and not the GIIPS
spread to German Bunds
 272

D. Liquidity and Solvency Stress Testing

The area of stress testing has seen a number of advances during recent years. Our study uses
a recently developed IMF liquidity stress testing framework to run integrated solvency and
liquidity stress tests. The liquidity stress testing framework presented in Schmieder et al
(2012) was developed in the context of recent FSAPs (Financial Sector Assessment
Program)218 and IMF technical assistance, extending the seminal work of Čihák (2007), and
drawing upon work at the Austrian National Bank (OeNB).219 An overview of recent
academic and policy research on integrating liquidity and solvency stress testing is given in
box 6.

In this study, the focus is on scenario design, namely building integrated scenarios for
solvency and liquidity risks that take into account spillover effects and feedback loops.220 The
central question becomes how the findings established in section B can be used to inform
bank-level stress tests.

Nevertheless, while we attempt to condense a wealth of information and assumptions to

establish integrated scenarios this should not, in any sense, give a false sense of precision.
Instead, we recommend running a whole range of scenarios which can build upon the ones
established in the study, with varying degrees of severity. Reverse stress tests can be also
included.221 This is an important way forward to obtain a better understanding of key
solvency and liquidity risks faced by banks, and to gain a more comprehensive view on their
respective risk tolerance.

218
Examples include Chile, Germany, India, Spain, Turkey, and the UK.
219
It is complemented by a previously developed solvency stress testing tool by Schmieder, Puhr and Hasan
(2011). While developing the solvency and liquidity stress testing frameworks, four key facts were accounted
for, which constitute key challenges of contemporaneous financial stability analysis: (i) the availability of data
varies widely, and lack of data is common; (ii) both solvency and liquidity risk have various dimensions, which
requires multi-dimensional analysis, thereby integrating risks; (iii) designing and calibrating scenarios is
challenging, even more so for liquidity risk than for solvency risk (mainly as liquidity crises are relatively rare
and originate from different sources); and (iv) communication of stress test results is a key integral part of the
exercise. The answer to these multiple dimensions are Excel based balance sheet type frameworks.
220
The exercise thereby reflects key principles for liquidity stress testing put forward by the Basel Committee in
the aftermath of the first wave of shocks following the default of Lehman Brothers (BCBS 2008).
221
The work by Taleb and others (2012) and Schmieder and Hardy (2013), for example, could be useful to
consider in this context.
 273

Box 6. Integrating Liquidity, Solvency Risks and Bank Reactions in Stress Tests
Banks have numerous and overlapping ways to react to credit and funding shocks. High-quality
capital and profits are usually the first line of defense, and retained earnings can help buffer banks’ capital
levels. In terms of liquidity, banks have an inherent counterbalancing capacity to generate liquid assets by
using high-quality eligible securities as collateral to generate market funding or, if interbank markets freeze
entirely, central bank funding. As seen post- Lehman, fire sales of securities can also be an option to
generate liquidity, but at a considerable cost in an environment of sharply declining asset prices.
Deleveraging, especially targeted at assets with higher risk weights, is also a way to raise capital adequacy
ratios by reducing risk-weighted assets (RWAs). In practice, banks have been using a combination of these,
as well as other hybrid measures, ranging from debt-to-equity conversions to issuance of convertible bonds
to optimizing risk-weighted assets, to react to shocks.

Incorporating banks’ reactions to shocks is a critical component for the design of informative stress
tests, especially over longer time horizons. This, however, requires modeling solvency and liquidity
shocks in a coherent manner because first, when banks react to financial stress, the source of the shock
(solvency or liquidity) is not always clear; and second, the measures banks take in reaction to these shocks
have both capital and liquidity aspects that are not easy to disentangle.

Recently, a number of analytical approaches have attempted to integrate solvency and liquidity more
systematically.
222
 Empirical work includes Van den End (2008) at the Dutch Central Bank and Wong & Hui (2009)
223
from the Hong Kong Monetary Authority , for example. Barnhill & Schumacher (2011) developed
a more general empirical model, incorporating the previous two approaches that attempt to be more
comprehensive in terms of the source of the solvency shocks and compute the longer term impact of
funding shocks.
 Schmieder and others (2012) provide an Excel based framework that allows running liquidity tests
informed by banks’ solvency conditions, and to simulate the increase in funding costs resulting from
a change in solvency.
 An integrated approach to model funding liquidity risks and solvency risk is the Risk Assessment
Model for Systemic Institutions (RAMSI) developed by the Bank of England (Aikman et al., 2009).
The framework simulates banks’ liquidity positions conditional on their capitalization under stress,
and other relevant dimensions, such as a decrease in confidence among market participants under
stress. A recent attempt by the Austrian National Bank to come up with an integrated framework and
to overcome operational challenges identified with previous work on integrated models, the Applied
Risk, Network and Impact assessment Engine (ARNIE), should also be mentioned (OeNB, 2013).

For an overview of liquidity stress tests, including the link to solvency, see also BCBS (2013). Hardy and
Hesse (2013) examine the EBA stress tests.
Source: Based on Oura and Schumacher (2012)

222
Van den End (2008) developed a stress testing model that tries to endogenize market and funding liquidity
risk by including feedback effects that capture both behavioral and reputational effects. A number of central
banks and bank supervisors have been successfully using the Monte Carlo framework of Van den End (2008).
223
The authors sought to explicitly capture the link between default risk and deposit outflows. Their framework
allows simulating the impact of mark-to-market losses on banks’ solvency position leading to deposit outflows;
asset fire sales by banks is evaporating and contingent liquidity risk sharply increases.
 274

Liquidity Stress Testing Approach

We apply an implied cash-flow approach to simulate the impact of bank-run type stress
scenario. The banks’ liabilities are broken down into demand and term deposits, short-term
wholesale funding (including bank and secured funding), derivatives’ funding as well as
long-term funding such as senior debt or subordinated debt. On the asset side, we include a
range of potentially liquid asset positions such as cash, government, trading and investment
(both available-for-sale and held-to-maturity) securities, loans and advances to banks and
reverse repos and cash collateral. Given European periphery banks’ increasing collateral use
of pools of loans (such as covered bonds) for liquidity, we also include a crude definition of
banks’ loan level as a portion of their total assets.

Solvency Stress Testing Approach

We use rules of thumb for solvency stress testing as proposed by Hardy and Schmieder
(2013) and thereby a simplified solvency test.224 Credit losses, banks’ pre-impairment income
and the trajectories of Risk-Weighted Assets (RWAs) for a 2-year horizon were simulated
based on the GDP trajectories, with and without spillover effects. The capital shortfall was
measured against a tier 1 capital ratio (Tier 1 capital/Risk-weighted Assets) of 6 percent,
below which a bank is considered undercapitalized.225

E. Integration of the Financial Spillover Analysis with the Stress Testing Approach

Our integrated approach to simulate stress at the bank level is illustratively shown in Figure
50:

1. Scenario design: We use the GDP trajectories of a specific macroeconomic scenario,

the WEO baseline scenario for 2013–14 as of April 2012, and add the spillover stress
component.
2. Spillover analysis: The outcome of the spillover analysis (see above), measured
through a widening of sovereign spreads, worsens the macroeconomic scenario, and
is used as a sensitivity analysis. The translation of the spillover effects into the
revised macroeconomic trajectories is based on recent IMF work.
3. Soundness of banks: The scenario is translated into bank level stress parameters to
simulate both the banks’ solvency and liquidity positions, drawing on work by Hardy
and Schmieder (2013) and Schmieder, Hesse, and others (2012), respectively.

224
However, it should be noted that the evidence is based on a comprehensive set of data from 16,000 banks
during the last 15 years (as available).
225
Please note that this specific choice is meant for illustration only—through a similar level as used for the
European stress tests conducted in 2010 and 2011, for example.
 275

We use bank-level data from Bankscope (from end-June 2012) for large Systematically
Important Banks (SIBs). In total, our sample includes 154 large banks from the following 26
countries:

Austria, Australia, Belgium, Brazil, Canada, Switzerland, China, Germany, Denmark,

Finland, France, UK, HK, India, Japan, Korea, Luxembourg, Mexico, Netherlands, Norway,
Poland, Russia, Sweden, Singapore, Turkey, and the USA.

Our sample comprises almost the full EBA sample for the European banks (except for the
banks in the GIIPS countries) and includes the largest banks in the non-European countries.
In total, it captures $84 trillion of bank assets (i.e., about 50 percent of the assets held by
banks worldwide), $39 trillion non-bank deposits and around $7 trillion of government
securities held by banks.

Figure 50. Overview of the Concept to Simulate Stress at the Bank Level

Bank solvency parameters:

GDP  Credit Losses
trajectory,  Security P/L impact
Scenario
adjusted for  Pre-impairment income
(eg WEO)
spillover
effects
Bank liquidity parameters
 Haircuts (Market
Liquidity)
IMF Spillover analysis  Outflow of funding
Panel/GARCH

Translation into bank-level stress

scenario: Overall soundness of bank
Solvency: Hardy/Schmieder
Liquidity: Schmieder/Hesse/at al
Source: Authors

Scenarios

We refer to four different scenarios: The April 2012 WEO baseline scenario for 2013–14
(Scenario 1); and three spillover scenarios (referred to as scenarios 2.x) conditional on
 276

scenario 1 – scenarios that banks could potentially face in case increasing degrees of
spillovers affect the general growth trend.

Specifically, scenario 1 is adjusted for an increase of GIIPS spreads by 100 (scenario 2a),
200 (2b) and 300 (2c) basis points, respectively. We further distinguish between the spillover
impact observed during periods of substantial financial stress (using the panel regression for
2008–12 and the GARCH model for 2010–12) and during periods of less significant stress
(using the panel regression for 2006–12 and the GARCH model for 2008–12), i.e. refer to a
total of six spillover scenarios (2a/1, 2a/2, 2b/1, 2b/2, 2c/1, 2c/2).

For the banks’ solvency, we simulate their Tier 1 capital ratios by end-2014, based on the
evolution of the main solvency dimensions (banks’ income and losses). For liquidity, we
determine the impact of a worst-case idiosyncratic shock to the bank’s liquidity profile on top
of the impact on liquidity resulting from the macroeconomic/spillover scenarios. Illustrative
examples are provided in Annex 15 (solvency) and 16 (liquidity).

Impact on bank solvency

As outlined above, we use the outcome of the 2012 IMF Spillover Report, which simulates
the impact of a 300bp increase in peripheral countries’ spreads (including a lower yield
increase for core countries) on European countries’ GDP paths based on the IMF G-35 model
(drawing upon Vitek and Bayoumi, 2011).

Annex 15 provides an illustrative example for a stylized Austrian bank. In the first step, the
increase of Austrian sovereign debt spreads is simulated, using the evidence established in
section B. A 100 basis point shock of GIIPS spreads (scenario 2a) would thereby result in an
increase of Austrian spreads by 24 basis points for less significant spillover stress (scenario
2a/1) and 50 basis points (2a/2) for more substantial spillover stress. Measured relative to the
April 2012 WEO baseline scenario for Austria, suggesting real GDP growth rates of 1.8
percent (2013) and 2.2 percent (2014), spillover analysis carried out at the IMF (2012) would
predict a drop of real GDP growth by about 0.45 percentage points for scenario 2a/1 (less
significant spillover stress), whereby the GDP trajectory becomes 1.4 percent (2013) and 1.8
percent (2014). For a period with more significant spillover (scenario 2a/2), the impact is
about twice (0.9 percentage points), whereby the GDP trajectory is 0.9 percent (2013) and
1.3 percent (2014). For a 200 basis point shock (scenario 2b), growth drops by 1.7 percentage
points and for 300 basis points (scenario 2c) by 2.6 percentage points (per year) under
substantial spillover conditions (Annex 15).
 277

We then use the satellite models by Hardy and Schmieder (2013) to determine banks’ loan
impairment levels and pre-impairment income for 2013 and 2014.226 For a stylized bank with
loss impairment rates of 0.5 percent and a pre-impairment return on capital of 10 percent in
2012227, loan impairment rates are simulated to decrease slightly under the baseline scenario
and mild spillover conditions, while they would increase (non-linearly) under increasing
levels of spillover stress. The same pattern holds for pre-impairment income. This input is
used to simulate the bank’s capital, Risk-weighted assets (RWAs)228, and capital ratio. Again,
the same pattern holds, with a decrease of the stylized banks’ capital ratio to 7.5 percent
under the most severe scenario, which is above the hurdle rate in terms of Tier 1 capital to
pass the stress test (6 percent).

The outcome of this solvency stress test applied to the 154 banks presented in Figure 51
shows that the large international banks would be in a position to digest the baseline scenario
plus some level of spillover stress, while additional stress in the Euro area periphery results
would have a highly non-linear impact on potential capital needs. The non-linearity results
from two factors: (i) the non-linearity in the satellite models for loan impairment rates and
pre-impairment income; and (ii) the effect of the kick-in of capital needs for banks that fall
below the hurdle rate.

Figure 51. Outcome of Solvency Stress Tests

25
Capital needs (Tier 1, bn USD) for different scenarios

20

15

10

0
Baseline (Sc 1) Sc 2a (Shock 100bps) Sc 2b (Shock 200bps) Sc 2c (Shock 300bps)
Less substantial spillover stress More substantial spillover stress

Source: Authors

226
For simplification, we assume that banks’ are affected according to their domestic scenarios, i.e., that their
business is pre-dominantly based in their home country.
227
In a few cases, the latest available figures were from 2011.
228
The RWAs are simulated based on work by Schmieder and others (2011), assuming point-in-time credit risk
parameters.
 278

Impact on bank liquidity

For the liquidity stress test, we simulate the impact of stress on both banks’ market liquidity
(i.e., their ability to fire sale assets) and funding liquidity (i.e., the potential outflow of
funding).229 Again, we assume that the bank is affected by the shock in its home country.230

The link between the level of stress and bank liquidity is established based on empirical work
of Schmieder, Hesse and others (2012). We link the GDP trajectories implied by the changes
of sovereign spreads to funding shocks experienced by the most affected banks during the
Lehman crisis. In other words, we simulate highly adverse idiosyncratic liquidity shocks
conditional upon macroeconomic conditions.

In line with (very limited) empirical evidence, we expect the relationship between the shock
and the potential adverse impact on the bank level to be highly non-linear (as implied by the
scenarios in Annex 14, and in addition to the non-linearity for the banks hitting the hurdle
rate, as for capital). Under a worst case scenario, banks would experience a shock equal to a
“Lehman Brothers type” scenario, the “severe stress scenario” in Annex 11.3 (this shock
level represents how the stress at the time of the Lehman Brothers event affected the banks
that were most severely hit, i.e. overlays a market shock with an idiosyncratic liquidity
shock). The stress level relative to the one experienced by banks at the time of the Lehman
Brothers crisis is established via the cumulative GDP trajectory under stress compared to the
long term average. For the stylized example presented in Annex 16, the stress level is at 0.65,
i.e., the benchmark funding stress parameters (for the “severe stress scenario”) in Annex 14
have to be multiplied by 0.65. The funding available for the specific banks under the ECB’s
Long Term Refinancing Operations (LTROs) is inferred from country-level data and used as
a cushion for the relevant European banks.

Figure 52 shows the outcome of this liquidity stress test. Under the baseline scenario all
banks have sufficient liquidity, as expected. Adding spillover stress triggers a non-linear
increase of liquidity needs (which occur in case the liquidity needs exceed the available
liquidity generated via fire sales), and more substantial spillover stress makes the stress
highly non-linear. Measured against Tier 1 capital rather than total assets, the substantial
spillover stress leads to a maximum liquidity shortfall of 20 percent for the entire bank
sample for scenario 2c/2 (300bp spread shock, significant spillover stress) and close to 6

229
Unlike for the solvency scenario, we do not simulate stress for a specific point in time; rather, the simulated
stress conditions reflect a worst-case situation resulting from the general macroeconomic conditions as well as
an idiosyncratic shock to the bank conditional.
230
In other words, it is assumed that all of its assets are based in the home country, which is a crude
simplification.
 279

percent for scenario 2b/2 (200bp spread shock), compared to 0.3 percent and 1 percent if
measured against total assets.231

Figure 52. Outcome of Liquidity Tests in Terms of Assets

1.2%
Liquidity Needs as a Percentage of Total Assets
1.0%

0.8%

0.6%

0.4%

0.2%

0.0%
Baseline (Sc 1) Sc 2a (Shock Sc 2b (Shock Sc 2c (Shock
100bps) 200bps) 300bps)

Less Substantial Spillover Stress Substatial Spillover Stress

Source: Authors

F. Conclusion

This study attempted to contribute to an important challenge faced by current financial

stability analysis, namely to capture spillover effects and other types of contagion that
ultimately determine macro-financial stress at the bank level.

By integrating recent IMF work on financial spillover analysis and stress testing, we use a
novel framework that allows shedding some light on the potential impact of spillover effects
on bank-level solvency and liquidity. Nevertheless, we recognize that significant additional
effort and evidence is needed to make the modeling of dynamic macro-financial linkages
more robust, not least due the many potential channels of spillover and contagion, the fact
that the use of crude data available for stress tests is subject to uncertainty, and other factors
that contribute to uncertainty (such as mixed evidence for the use of market data).

231
We did not explicitly model a central bank response as the Lender of Last Resort (LOLR) to mitigate the
estimated liquidity shortfall. In reality and as seen during the crisis period, central banks would provide large
liquidity support to solvent banks subject to an appropriate haircut.
 280

The outcome of the stress tests suggests that spillover effects observed for the sovereign debt
markets in recent years have a highly non-linear impact on bank soundness, both in terms of
liquidity and solvency. This implies (once more) that the design of stress scenarios is a
crucial element of stress testing, and is very sensitive with respect to the outcome of stress
tests. The approach used in this paper is meant to be menu for future analyses of the impact
of potential spillovers. Sensitivity analysis and reverse stress tests appear to be an important
complement in this context.
 ANNEX 12. OUTCOME OF PANEL REGRESSIONS ASSESSING SPILLOVER RISKS

Table A12.1. Panel Regressions, 2006Q1–2012Q2 (Dependent variable: Sovereign Spreads of 35 sample countries)
(Quarterly data)
Explanatory (1) (2) (3) (4) (5) (6)
VARIABLES1

GIIPS spread 0.237*** 0.244***

(0.045) (0.047)
GIP spread 0.288*** 0.289***
(0.046) (0.047)
Italy/Spain spread 0.611*** 0.653***
(0.09) (0.094)
High-yield spread 0.666*** 0.621*** 0.357
(0.242) (0.229) (0.30)
VIX 0.348 0.342 -0.070

281
(0.238) (0.229) (0.291)
Openness 0.015 0.015 0.031* 0.030* 0.025 0.025
(0.017) (0.017) (0.016) (0.016) (0.020) (0.021)
M2/GDP 0.080*** 0.078*** 0.061*** 0.060*** 0.053*** 0.051**
(0.017) (0.017) (0.016) (0.016) (0.020) (0.020)
Constant 0.297** -0.632 0.256* -0.660 0.700*** 0.997
(0.131) (0.744) (0.136) (0.718) (0.166) (0.912)

R-squared (within) 0.77 0.70 0.79 0.73 0.79 0.78

Observations 415 415 435 435 454 454
T 25 25 23 23 26 26
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
1
Right Hand side variables are in logs.
 Table A12.2. Panel Regressions, 2008Q1–2012Q2 (Dependent variable: Sovereign Spreads of 35 sample countries)
(Quarterly data)
Explanatory (1) (2) (3) (4) (5) (6)
VARIABLES1

GIIPS spread 0.492*** 0.463***

(0.105) (0.106)
GIP spread 0.511*** 0.479***
(0.090) (0.090)
Italy/Spain spread 1.002*** 0.998***
(0.173) (0.175)
High-yield spread 1.042*** 1.033*** 0.735**
(0.299) (0.279) (0.366)
VIX 0.823** 0.813*** 0.517
(0.322) (0.301) (0.397)
Openness 0.018 0.017 0.034* 0.033* 0.033 0.032

282
(0.021) (0.021) (0.019) (0.019) (0.027) (0.027)
M2/GDP 0.078*** 0.075*** 0.057*** 0.056*** 0.045* 0.043*
(0.020) (0.020) (0.018) (0.018) (0.025) (0.025)
Constant -0.133 -2.418** -0.216 -2.459** 0.308 -1.117
(0.222) (1.084) (0.222) (1.022) (0.246) (1.307)

R-squared (within) 0.93 0.78 0.91 0.78 0.91 0.85

Observations 321 321 357 357 341 341
T 18 18 18 18 18 18
Standard errors in parentheses, *** p<0.01, ** p<0.05, * p<0.1
1
Right Hand side variables are in logs.
 283

Table A12.3. Main Explanatory Variables

Factor Variable Description
 GIIPS spread  Average of Euro
 periphery sovereign
 spreads to German
 Bunds
 Sovereign Risk
GIP spread Average of Greece, Ireland
and Portugal sovereign
spreads to German Bunds
Italy/Spain spread (IS spread)  Average of Italy
and Spain
sovereign spreads
to German Bunds.

 High-yield spread Difference between yields

 Risk aversion to maturity of AAA rated
and BAA rated corporate
US bond
VIX Implied volatility of S&P
500 index options.
 Openness Sum of imports and
 Macroeconomic exports to GDP ratio
environment M2/GDP Broad money to GDP ratio

Source: Authors
 284

ANNEX 13. OUTLINE OF THE DCC GARCH METHOD

The DCC model is estimated in a three-stage procedure. Let rt denote an n x 1 vector of asset
returns, exhibiting a mean of zero and the following time-varying covariance:

(1)

Here, Rt is made up from the time dependent correlations and Dt is defined as a diagonal
matrix comprised of the standard deviations implied by the estimation of univariate GARCH
models, which are computed separately, whereby the ith element is denoted as hit . In other
words in this first stage of the DCC estimation, we fit univariate GARCH models for each of
the five variables in the specification. In the second stage, the intercept parameters are
obtained from the transformed asset returns and finally in the third stage, the coefficients
governing the dynamics of the conditional correlations are estimated. Overall, the DCC
model is characterized by the following set of equations (see Engle, 2002, for details):

(2)

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset
returns rt. As defined above, Rt is the time varying correlation matrix and is a function of Qt,
which is the covariance matrix. In the matrix Qt,ι is a vector of ones, A and B are square,
symmetric and  is the Hadamard product. Finally, λi is a weight parameter with the
contributions of Dt21 declining over time, while κ i is the parameter associated with the
squared lagged asset returns. The estimation framework is the same as in Frank, Gonzalez-
Hermosillo and Hesse (2008) or Frank and Hesse (2009).
 285

ANNEX 14. BENCHMARK STRESS SCENARIOS

Moderate Medium Stress Severe Stress Very Severe

Scenario
Stress Scenario Scenario Scenario Stress Scenario
Severity (x times Lehman/1) 0.25 0.5 1 2
Liquidity Outflows
Customer Deposits
Customer deposits (Term) 2.5 percent 5 percent 10 percent 20 percent
Customer deposits (Demand) 5 percent 10 percent 20 percent 40 percent
Wholesale Funding
Short-term (secured) 5 percent 10 percent 20 percent 40 percent
Short-term (unsecured) 25 Percent 50 Percent 100 Percent 100 Percent
0 Percent need 5 Percent need 10 Percent need 20 Percent need
Contingent liabilities
funding funding funding funding
Liquidity Inflows
Haircut for Cash 0 Percent 0 Percent 0 Percent 0 Percent
Haircut for Government
1 Percent 2 Percent 5 Percent 10 Percent
Securities/2
Haircut for Trading Assets/3 3 Percent 6 Percent 30 Percent 100 Percent
Equity: 10-15;
Equities: 3; Equities: 4-6; Bonds (only LCR
Proxies, specific assets Not liquid
Bonds: 3 Bonds: 3-8 eligible ones): 5-
10
Haircut for other securities 10 Percent 30 Percent 75 Percent 100 Percent
Equities: 25; Equity: 30; Bonds
Equities: 10;
Proxies, specific assets Bonds: 20 (some (only LCR eligible Not liquid
Bonds: 10
not liquid) ones): 20-30
20 Percent (or 30 Percent (or 40 Percent (or
Percent of liquid assets 10 Percent (or
actual figure plus actual figures actual figures
encumbered/4 actual figure)
10 ppt) plus 20 ppt) plus 30 ppt)
1/ The Lehman type scenario would correspond to a scenario encountered by banks that were hit
severely during the 30 day period after the Lehman collapse, i.e. a stress situation within a stress period
rather than an average; The scenario has been put together based on expert judgment, using evidence
as available.
2/ The haircut highly depends on the specific features of the government debt held (rating, maturity,
market depth) and can be higher or lower. The figures displayed herein are meant for high quality
investment grade bonds, taking into account recent market conditions. The same applies for the
remainder of the liquid assets. For the securities in the trading book, it is assumed that they are
liquidated earlier, resulting to lower haircuts.
3/ A haircut of 100 Percent means that the asset is illiquid, i.e., the market has closed.
4/ The figures account for a downgrade of the bank, which triggers margin calls, and higher collateral
requirements for generally. Please note that the unencumbered portion applies to a gradually narrower
definition of liquid assets.
Source: Schmieder, Hesse and others (2012)
 286

ANNEX 15. ILLUSTRATIVE EXAMPLE FOR THE SOLVENCY TEST

The table provides an illustrative example for a hypothetical bank in Austria.

Step 1.1: Spillover impact in sovereign debt markets observed for Austria

Scenario Impact on Austria, average for 2006- Impact on Austria during peak spillover
(increase of 2012 stress (2008-2012)
GIIPS sovereign
Increase of Increase of
debt spreads Source Source
spreads (bps) spreads (bps)
by…)
Table A12.1, spec Table A12.2, spec (1),
24.4232 49.8234
100 bps (2a) (2233), Figure 1, top Figure 1, top right
(=24*1.017) (=49*1.017)
right hand panel hand panel
Same, linear Same, linear increase
200 bps (2b) 48.8 99.6
increase assumed assumed
Same, linear Same, linear increase
300 bps (2c) 73.2 149.4
increase assumed assumed
Source: Authors

232
The average impact of stress (in terms of GIIPS spreads) on Euro area countries is 24 basis points (based on
the panel analysis, see Table I.1) and for Austria the relative severity of this impact approximately matches the
impact observed for the EU, i.e. it is 1.0 times of this level (GARCH analysis, average impact from 2008–12
based Figure 1, top right hand panel, relative to average of the average impact for other EU countries).
233
We used the higher impact on the GIIPS spreads from Table I.1 and Table I.2, i.e. specification 2 (Table I.1)
and 1 (Table I.2), respectively, i.e. 24 bps and 49 bps, respectively.
234
The average impact of stress on Euro area countries is 49 basis points (based on the panel analysis, Table I.2)
and for Austria the impact is again estimated to be at a similar level (GARCH analysis, average impact from
2008–12, Figure 1, top right hand panel, relative to average of the average impact for other EU countries).
 287

Step 1.2: GDP trajectory for Austria, adjusted for the impact of spillovers

 GDP Elasticity of widening of spreads for Austria estimated for two year period from
2013-2014: 3.5 (based on IMF , 2012)235

Trajectory based on evidence for 2006-2012 (i.e., less significant spillovers)

Scenario 2012 2013 2014 Cumulative

deviation of output
from 2012 real
GDP growth level
(2013-14), ppts
Baseline (1) 0.9 1.8 2.2 2.2
0.9 1.4 1.8 1.3
(2a/1)
(=1.8-0.5*3.5*0.244) (=2.2-0.5*3.5*0.244)
0.9 0.9 1.3 0.4
(2b/1)
(=1.8-0.5*3.5*0.488) (=2.2-0.5*3.5*0.488)
0.9 0.5 0.9 -0.4
(2c/1)
(=1.8-0.5*3.5*0.732) (=2.2-0.5*3.5*0.732)

Trajectory based on evidence for 2008-2012 (i.e., more significant spillovers)

Scenario 2012 2013 2014 Cumulative

deviation of output
from 2012 real
GDP growth level
(2013-14), ppts
Baseline (1) 0.9 1.8 2.2 2.2
0.9 0.9 1.3 0.4
(2a/2)
(=1.8-0.5*3.5*0.498) (=2.2-0.5*3.5*0.498)
0.9 0.1 0.5 -1.2
(2b/2)
(=1.8-0.5*3.5*0.996) (=2.2-0.5*3.5*0.996)
0.9 -0.8 -0.4 -3
(2c/2)
(=1.8-0.5*3.5*1.494) (=2.2-0.5*3.5*1.494)

235
The GDP elasticities of sovereign debt spreads vary between 0.5 (e.g. Brazil) and 3.5.
 288

Step 2: Simulation of the impact at the bank level (Example for stylized bank)236

Change of key solvency parameters

Loan impairment rates (Percent of Pre-impairment income (Percent of total

credit exposure) capital)
Scenario 2012 2013 2014 2012 2013 2014
Baseline 0.5 0.4 0.4 10 10.3 10.5
2a/1 0.5 0.45 0.4 10 10.15 10.3
2b/1 0.5 0.5 0.45 10 10 10.1
2c/1 0.5 0.55 0.5 10 9.8 10
2a/2 0.5 0.5 0.45 10 10 10.1
2b/2 0.5 0.7 0.6 10 9.7 9.8
2c/2 0.5 0.9 0.8 10 9.2 9.5
Note: credit growth is assumed to be constant (for simplification)

Evolution of Risk-weighted Assets (RWAs) and Capital

RWAs (Indexed) Capital

Scenario 2012 2013 2014 2012 2013 2014
Baseline 100 90 90 10 10.58 11.21
2a/1 100 95 90 10 10.57 11.18
2b/1 100 100 95 10 10.56 11.15
2c/1 100 105 100 10 10.54 11.12
2a/2 100 100 95 10 10.56 11.15
2b/2 100 120 111 10 10.52 11.08
2c/2 100 140 132 10 10.47 10.99
Note: For simplification, RWA elasticity to credit losses assumed to be 0.5, i.e., for a 1 percentage
point change of credit loss rates RWAs will change by 0.5 percentage points.

Evolution of the Bank’s Capital Ratio

Capital Ratio (= Capital/RWA, Percent)

Scenario 2012 2013 2014
Baseline 10.0 11.8 12.5
2a/1 10.0 11.1 12.5
2b/1 10.0 10.6 11.7
2c/1 10.0 10.0 11.1
2a/2 10.0 10.6 11.7
2b/2 10.0 8.8 9.9
2c/2 10.0 7.5 8.3

236
See Hardy and Schmieder (2013) for further information.
 289

ANNEX 16. ILLUSTRATIVE EXAMPLE FOR LIQUIDITY

This annex provides an illustrative example for a hypothetical bank in Austria.

Step 1: GDP trajectory for Austria, adjusted for the impact of spillovers

The first steps uses the same GDP trajectories as for solvency (see Annex 15). Accordingly,
the severity of the liquidity shock is simulated relative to the Lehman Brothers benchmark
scenario in Annex 14. Specifically, based on the observation that the cumulative U.S. real
GDP growth deviated by about 8 percentage points from the long-term average, the
corresponding figures are computed for each of the scenarios. For Austria (and for the other
European countries), the baseline growth rates for 2013-2104 (i.e., 2 percent) are (for
simplicity) used as a proxy for the long-term trend. For scenario 2c/2, the cumulative
deviation from the baseline is 5.2 percentage points. For the severity of the liquidity test, we
therefore use the stress parameters for the severe scenario in Annex 14 multiplied by a factor
of 0.65 (=5.2/8).

Step 2: Simulation of the impact at the bank level (Example for stylized bank)237

Relevant asset and liability balance sheet items are shocked based on the severity of each
scenario, i.e., the stress factor (such as 0.65) multiplied by the respective stress parameters.
The balance sheet items are taken from Bankscope. For the LTRO, the available total funding
was assigned to the single banks based on their size, using the available evidence for the total
at the country level.

In the table below, scenario 2c/2 is simulated for a stylized bank based on Austria. The
composition of the banks’ asset and liabilities resemble those of an average OECD bank.238
The stress factor reduces the haircuts and outflows of the benchmark scenario. In the
example, the bank is able to generate an inflow of 21.5 units of assets, compared to a
required level of 13.7 units, whereby the bank remains liquid.

237
See Schmieder and others (2012) for further information.
238
See Schmieder and others (2012), p. 38, for more information.
 290

Assets (of stylized bank)

Portion of Haircut, Percent Haircut Available
total239 (Annex 14) Scenario 2c/2 assets (fire
sales)
Cash and cash-like 4 0 0 4.0
Government securities 6 5 3 5.8
Trading securities 5 30 20 4.0
Other securities 15 75 49 7.7
Loans 60 NA NA
Other 10 NA NA
Liabilities (of stylized bank)
Portion of Outflow, Percent Outflow Required
total240 (Annex 14) Scenario 2c/2 funding
Customer Term deposits 30 10 6.5 2
Customer Demand 2.6
deposits 20 20 13
Secured short-term 1.3
wholesale funding 10 20 13
Unsecured short-term 6.5
wholesale funding 10 100 65
Long-term funding 20 0 0 0
Equity based funding 10 0 0 0
Contingent liabilities 20 10 6.5 1.3

239
Aligned to the average composition of OECD banks‘ balance sheets. See Schmieder and others (2012), p. 38.
240
Aligned to the average composition of OECD banks‘ balance sheets. See Schmieder and others (2012), p. 38.
 291

References

Aikman, David, Piergiorgio Alessandri, Bruno Eklund, Prasanna Gai, Sujit Kapadia,
Elizabeth Martin, Nada Mora, Gabriel Sterne and Matthew Willison, 2009, Funding
liquidity risk in a quantitative model of systemic stability, Bank of England Working
Paper no. 372.

Austrian National Bank (OeNB), 2013, “ARNIE in Action: The 2013 FSAP Stress Tests for
the Austrian Banking System.”

Barnhill, Theodore Jr and Liliana Schumacher, 2011, “Modeling Correlated Systemic

Liquidity and Solvency Risks in a Financial Environment with Incomplete
Information,” IMF Working Paper 11/263.

Basel Committee on Banking Supervision (BCBS), 2008, “Principles for Sound Liquidity
Risk Management and Supervision - final document”, September.

Basel Committee on Banking Supervision (BCBS), 2013, “Liquidity stress testing: a survey
of theory, empirics and current industry and supervisory practices”, BCBS Working
Papers No 24, October.

Bollershev, Tim, 1990, “Modelling the Coherence in Short-run Nominal Exchange Rates: a
Multivariate Generalized ARCH Approach,” Review of Economics and Statistics,
Vol. 72, pp.498–505.

Borio, Claudio, Drehmann, Matthias and Kostas Tsatsaronis, 2012, “Stress-testing macro
stress testing: does it live up to expectations?,” BIS working paper no. 369, available
at www.bis.org/publ/work369.pdf.

Caceres, Carlos, and D. Filiz Unsal, 2011, “Sovereign Spreads and Contagion Risks in Asia,”
IMF Working Paper no.11/134

Čihák, Martin, 2007, “Introduction to Applied Stress Testing,” IMF Working Paper No.
07/59 (Washington, D.C.: International Monetary Fund), available at
http://www.imf.org/external/pubs/ft/wp/2007/wp0759.pdf.

Engle, R., 2002, “Dynamic Conditional Correlation: A Simple Class of Multivariate

Generalized Autoregressive Conditional Heteroskedasticity Models,” Journal of
Business & Economic Statistics, Vol. 20, pp. 339–50.

Foglia, Antonella, 2009, “Stress Testing Credit Risk: A Survey of Authorities' Aproaches”,
International Journal of Central Banking, September, available at
http://www.ijcb.org/journal/ijcb09q3a1.htm.
 292

Frank, N., B. González-Hermosillo, and H. Hesse, 2008, “Transmission of Liquidity Shocks:

Evidence from the 2007 Subprime Crisis,” IMF Working Paper 08/200. See also blog
piece on VOX.

Frank, N., and H. Hesse, 2009, “Financial Spillovers to Emerging Markets during the Global
Financial Crisis,” IMF Working Paper 09/104.

Hardy, Daniel C., and Christian Schmieder, 2013, “Rules of Thumb for Bank Solvency
Stress Testing,” IMF Working Paper no. 13/232.

Hesse, Heiko, Ferhan Salman, and Christian Schmieder, 2014, “How to Capture Macro-
Financial Spillover Effects in Stress Tests,” IMF Working Paper 2014/103.
Forthcoming in the Journal of Financial Perspectives.

International Monetary Fund, 2011a, “Safe Assets: Financial System Cornerstone?,”

available at www.imf.org/external/pubs/ft/gfsr/2012/01/pdf/c3.pdf

-----------, 2011b, “Euro Area Policies: Spillover Report for the 2011 Article IV Consultation
and Selected Issues”, July, available at
http://www.imf.org/external/pubs/cat/longres.aspx?sk=25056.0

-----------, 2012, 2012 Spillover Report, July 2012, available at

www.imf.org/external/np/pp/eng/2012/070912.pdf

Jobst, Andreas A., and Dale F. Gray, 2013, “Systemic Contingent Claims Analysis—
Estimating Market-Implied Systemic Solvency Risk,” IMF Working Paper No. 13/54
(Washington: International Monetary Fund), available at
http://www.imf.org/external/pubs/ft/wp/2013/wp1354.pdf.

Jobst, Andreas, Ong, Li Lian, and Christian Schmieder, 2013, “A Framework for
Macroprudential Bank Solvency Stress Testing: Application to S-25 and Other G-20
Country FSAPs,” IMF Working Paper No. 13/68.

Oura, Hiroko and Liliana Schumacher, 2012, “Macrofinancial Stress Testing—Principles and
Practices,” IMF Policy Paper, August 2012.

Schmieder, Christian, Puhr, Claus and Maher Hasan, 2011, “Next Generation Balance Sheet
Stress Testing,” IMF Working Paper No. 11/83.

Schmieder, Christian, Hesse, Heiko, Neudorfer, Benjamin, Puhr, Claus and Stefan W.
Schmitz, 2012, “Next Generation System-Wide Liquidity Stress Testing,” IMF
Working Paper no. 03/12.
 293

Taleb, Nassim N., Canetti, Elie, Kinda, Tidiane, Loukoianova, Elena, and Christian
Schmieder, 2012, “A New Heuristic Measure of Fragility and Tail Risks: Application
to Stress Testing,” IMF Working Paper no. 216/2012.

Van den End, Jan Willem, 2008) “Liquidity Stress Tester: A macro model for stress-testing
banks’ liquidity risk”, Dutch National Bank Working Paper No. 175, May.

Vitek, Francis, and Tamim Bayoumi, 2011, “Spillovers from the Euro Area sovereign debt
crisis: A macroeconometric model based analysis,” CEPR Discussion Paper, 8497.

Wong, Eric and Cho-Hoi Hui, 2009, “A liquidity risk stress- testing framework with
interaction between market and credit risks,” Hong Kong Monetary Authority
Working Paper 06/2009.
 294

XII. REPROFILING AND DOMESTIC FINANCIAL STABILITY: RECENT EXPERIENCES, 2014241

This paper analyses the experience with past cases of reprofiling to assess whether they had
destabilizing effects on the domestic banking system. It examines several past maturity
extensions (Cyprus, Jamaica, Pakistan, and Uruguay) and finds that destabilizing effects did
not materialize. Several factors contributed to the generally successful outcomes under
maturity extensions: financial stability concerns were taken into account in the design of the
restructuring and program strategy; banks mainly held their sovereign assets as held-to-
maturity (HTM); a reprofiling was not assessed to be an impairment event requiring a write
down of these assets (e.g., Cyprus, Jamaica); regulatory incentives for banks were provided
(e.g., Jamaica or Uruguay); capital and liquidity support mechanisms were established (e.g.,
Jamaica) or were present (Cyprus); the amount of bank holdings of sovereign bonds in most
cases was not very large; and some forbearance was used. The Jamaican case illustrates
how a restructuring was designed to be light in order to ensure a limited impact on the
financial system. The paper also proposes possible measures that could help protect the
banking system during a reprofiling and encourage participation by domestic banks in the
exchange. Finally, the paper examines financial stability implications of a creditor bailout.
Although a reprofiling may have some disruptive effects, a bailout does not necessarily
insulate the domestic financial system, as the Greek experience demonstrates.242

A debt restructuring or reprofiling can potentially have several disruptive effects on

domestic banks and other financial institutions holding government debt. In general,
domestic banks can be affected by sovereign stress because of the role of the sovereign as
backstop to the financial system and through direct exposure to the sovereign. Some of the
possible effects include the following:

 Banks can suffer mark-to-market losses on their holdings of government bonds which
could lead them to being undercapitalized.

 A concern about the health of domestic banks could lead to deposit runs in the
domestic financial system which could affect otherwise healthy banks.

 Banks’ short-term external wholesale funding could become more expensive or even
evaporate.

 Banks could face larger haircuts on their sovereign exposure for collateral operations
(e.g., interbank or central bank repo operations) from the increased sovereign yields
and asset valuation changes.

241
This chapter is based on Hesse (2014).
242
“Reprofilings” are proxied by cases involving limited (face-value preserving) maturity extensions that lead to
moderate NPV reductions.
 295

 Exchange rate depreciation associated with worsening market sentiment could feed
into higher bank FX funding costs and could expose unhedged FX (corporate)
borrowers.

A reprofiling can potentially also have beneficial effects on the banking system. If a
reprofiling improves the prospects for debt sustainability and is seen to help avoid a
potentially deeper debt restructuring subsequently, it can speed up the recovery thereby
reducing debt overhang and benefiting the financial sector. Compared to an alternative were
markets may have priced in a deep debt restructuring, a reprofiling would remove the
uncertainty and could lead to a relief rally and market-to-market gains on bank balance
sheets. In general, the extent of the gains or losses on banks’ balance sheets would depend on
the portfolio structure of bond holdings.

The Fund’s experience indicates that past debt reprofiling cases have not had
destabilizing effects on the domestic financial system. Staff reviewed the experience in
four Fund-supported programs that entailed a reprofiling (Cyprus, Jamaica, Pakistan, and
Uruguay). The detailed analysis finds that in none of these cases did the reprofiling have a
disruptive impact on the domestic banking system. Several factors helped to contain the
effects on the banking system: financial stability considerations were taken into account in
the design of reprofiling; banks mainly held their sovereign assets as HTM and the
reprofiling was not assessed to be an impairment event requiring a write-down of these
holdings; regulatory incentives for banks to participate in the bond exchange were provided
(see Boxes 7 and 8); bank holdings of sovereign debt were not large in some cases;
domestically held debt was excluded from reprofilings in some cases; and capital and
liquidity support mechanisms were established or were present and some forbearance was
used. Box 7 provides details on each country case.

 For example, in Jamaica, the restructuring was intentionally designed to be light so as

to minimize the impact on the financial system.

 In Cyprus, banks did not need to write down their holdings of domestic sovereign
debt as the reprofiling was not assessed to be an impairment event, given that bonds
were close to maturity and principal and coupon were being preserved.243

 Where domestic financial stability was a concern in past reprofiling cases, program
design explicitly provided for bank recapitalization and/or liquidity support to
successfully limit the impact on the financial sector. In any event, such funds often
did not need to be fully utilized or were not used at all.

243
Nevertheless, while these measures in the Cyprus and Jamaica cases were important, the debt reprofilings
could have had a market impact on their domestic banking system had there not been a credible design of safety
measures to limit any potential fallout. In addition, the approach in Cyprus was part of a broader program
strategy which included a restructuring of Cypriot banks with the bail-in of bank creditors.
 296

 Finally, Cyprus and Uruguay are examples where a reprofiling worked smoothly and
in parallel with an ongoing program to address vulnerabilities in the banking system
and address problem banks. For example, a bail-in of bank creditors closed the
imminent capital hole in some of the Cypriot banks.

Some caveats need to be borne in mind. While the review of experience shows that
reprofilings have worked relatively smoothly in the past, this evidence should not be taken as
conclusive. There are always risks in the design of a restructuring that need to be adequately
addressed through the design of the restructuring, and more generally, the design of the
Fund-supported program. Past reprofilings were also not done under the regime being
considered. While temporary forbearance seems to have worked in the past in containing the
impact on the banking system, it can adversely affect credit growth. It is also possible that the
benign outcomes in past reprofilings could have been the result of better initial
macroeconomic conditions compared to those countries that undertook a deep debt
restructuring. However, this does not seem to be the case: table 19 shows that initial
macroeconomic conditions were very similar in past cases of reprofilings and deeper debt
restructurings.

Table 19. Indicators of Fundamentals and Policy Track Record

Indicators of Fundamentals and Policy Track Record

Reprofiling Debt reduction

Mean Median Mean Median

Public Debt (percent of GDP, year before

96.8 93.0 100.2 101.6
restructuring)

Current Account Balance (percent of GDP, average

-6.4 -7.7 -5.7 -5.5
over three years before restructuring)

Overall Fiscal Balance (percent of GDP, average over

-5.0 -4.6 -3.4 -2.9
three years before restructuring)

Primary Balance (percent of GDP, average over three

0.5 0.4 0.7 0.9
years before restructuring)

Source: Staff reports, WEO

Reprofilings: Belize (2007), Cyprus, Dominican Republic, Grenada, Jamaica (2010), Jamaica (2013),
Pakistan, Paraguay, Uruguay
Debt reductions: Argentina, Belize (2013), Ecuador (2000), Ecuador (2009), Greece, Russia, Seychelles, St.
Kitts & Nevis, Ukraine

These findings notwithstanding, program design would in any event seek to address
such concerns, as typically common in Fund-supported programs. As in past
restructurings, program design would be expected to build in measures to safeguard financial
 297

stability during and after the reprofiling.244 In the specific circumstances of a currency union
with highly interconnected financial markets, limiting the impact on the domestic banking
system as well as overall contagion to other members would also need to rely on credible
system-wide backstops, with prompt action taken to strengthen such backstops if necessary.
While the exact set of measures to be used would depend on country specific considerations,
some of the measures used in the past have included the following:

 Creation of backstops, in the event that stress tests suggest possible financial
stability concerns. Examples include Jamaica’s use of a Financial Sector Stability
Fund (FSSF) in 2010 and 2013, and the Fund for Fortifying the System of Banks in
Uruguay. Fund program resources could potentially contribute to the backstop.

 Increased central bank liquidity provision prior to and during debt reprofiling
episodes. Central banks typically play a fundamental role during debt reprofiling
episodes as the lender-of-last-resort. Central bank liquidity provisions would be
especially important if domestic banks lose market funding on account of an SD
rating, and could include a temporary expansion of eligible collateral if banks face
large haircuts on collateral of government bonds for repo purposes. Where relevant
and needed, FX swap lines with other central banks could provide some temporary
FX funding buffers to the domestic banking sector, and indirectly to corporate, should
there be a squeeze on FX funding. The availability of ample FX funding would also
enhance the confidence in the financial sector, including amongst potentially-fickle
depositors. Box 8 provides an overview of how central banks have provided banking
sector liquidity during the past reprofiling cases of Cyprus, Jamaica, Pakistan and
Uruguay.

 Regulatory measures. Various measures could be taken to help banks and nonbanks
participate in the bond exchange and limit some potential direct as well as indirect
impact. These include applying different risk weights on old versus new bonds,
providing liquidity support subject to eligible collateral for viable financial
institutions, and granting some temporary forbearance (e.g., on provisioning
requirements).245 Box 9 provides an overview of the accounting treatment of banks’
holdings of domestic sovereign debt.

244
Under the proposal, specific financial sector measures would differ on a case-by-case basis. Past
recommendations, as described above, cannot necessarily be a guide to future advice in different cases.
245
Regulators would need to strike the right balance in the use of forbearance. In this context, forbearance would
typically involve allowing affected financial institutions more grace time to book any bond losses from the bond
exchange than is stipulated under IFRS accounting standards. Forbearance is in general not recommended, since
it could undermine market confidence in the domestic banks if provisioning and capital holes are suspected as a
result of adverse bond yield developments around and after the time of the bond exchange. If financial stability
concerns are a serious issue following the debt reprofiling, however, temporary relaxation of some regulations
could be justified as a last resort, if considered necessary to avoid “cliff effects” on affected banks.
 298

 Capital and deposit controls. In the extreme case where nonresident capital outflows
(such as bank deposits) could destabilize the financial system over the short-run, prior
to or following a debt reprofiling, temporary capital controls—as were deployed in
Cyprus and Iceland—might be also part of the crisis management toolkit. Capital
controls would need to be implemented carefully, with clear communication and a
credible exit strategy.

 Bank resolution tools, such as bail-in. If banks were found to be undercapitalized

prior to or as a consequence of a debt reprofiling, it would be appropriate from a
financial stability perspective—and to safeguard limited public resources—to require
bank shareholders and possibly other bank creditors to contribute to a resolution of
the problem. For instance, in the case of Cyprus, the bail-in of bank creditors closed
the imminent capital hole of the main Cypriot banks.

The alternative to reprofiling—bailing out holders of sovereign debt—would not

necessarily eliminate risks of financial instability or the need to build in appropriate
program safeguards.246 In past programs, countries which benefited from access to Fund
resources did not always avoid banking crises, particularly in cases where a restructuring was
eventually needed. A case in point is Greece, where the banking sector was perceived as
relatively sound in 2010 when the first Fund-supported program started. With the severity of
the Greek recession, ongoing sovereign debt problems and fears about a Greece euro exit, the
domestic banking system increasingly faced existential liquidity, asset quality and capital
problems even before a restructuring was envisaged. Strong program measures were taken in
response, including on bank recapitalization and restructuring as well as by the ECB on
substantial liquidity support. Box 10 provides an overview of relevant banking sector
developments in Greece during the crisis.

246
For the purposes of this annex, a bailout is defined as cases where creditors are paid out before an eventual
restructuring.
 299

Box 7. Impact of Sovereign Debt Maturity Extensions on Domestic Bank’s Balance Sheets
This box examines several past maturity extensions (Cyprus, Jamaica, Pakistan, and Uruguay)
and finds that they did not have destabilizing effects on the banking system. Staff reports and
other sources were examined for each case to obtain information on the banking system impact. The
criterion used to assess whether the bond exchange had a material impact on the banking sector was if,
as a direct result of the bond exchange, any bank in the country needed either additional provisioning
or recapitalization. The cases highlight a number of factors that contributed to the preservation of
financial stability in the wake of a maturity extension, and may also help explain the short duration of
the SD downgrade in most cases (less than two months except in Pakistan where it lasted 11 months).

Cyprus (2013): A few weeks ahead of when its (primarily domestically-held) bonds were originally
due, Cyprus exchanged them with new bonds with the same coupon, and extended maturities through
2019-2023 (IMF, 2013a). The main reason Cypriot banks did not have to book losses during the bond
exchange was that banks held the affected sovereign bonds as Hold-to-Maturity (HTM) and they
assessed, with the tacit consent of the regulator, that there was no impairment event. This allowed
Cypriot banks to maintain the newly exchanged bonds as HTM and not move them to the Available for
Sale (AFS) portfolio, where fair value measurement would have been required (as market prices were
well below par). The temporary SD assessment of the rating agencies did not bind Cypriot banks to
book losses with the affected sovereign bond holdings as the sovereign did not default on its payments
and issued new bonds with the same face value and other terms. Finally, the prior bail-in of bank
creditors closed the imminent capital hole of the main Cypriot banks.

Jamaica (2010, 2013): Overall, the restructuring was designed to be light, in order to avoid
destabilizing the financial system. In both bond exchanges, most of the affected domestic debt was
primarily held by banks as HTM, and as there was no impairment event banks did not take any
immediate additional provisioning or capital hit from the debt reprofiling exercise. Rating agencies
ruled Jamaica’s domestic government debt as SD but then upgraded the sovereign following the
successful completion of the bond exchange. A Financial Sector Stability Fund (FSSF)—set up to help
with any capital and liquidity support for financial institutions— was not tapped (see Grigorian and
others, 2012).

Pakistan (1999): Pakistan’s restructuring of external sovereign debt in November 1999 had a limited
impact on the domestic banking sector. The restructuring involved a slight nominal increase in
principal outstanding for two of the three Eurobonds, to roll in unpaid interest, and the offered terms
were relatively attractive to the creditors. About 30 percent of restructured bonds were held by
domestic investors. One participating domestic bank received a capital injection 6 months after the
exchange following a bank audit, though the undercapitalization does not seem to be related to the
earlier bond exchange (IMF, 2001).

Uruguay (2003): The overall immediate direct impact on the domestic banking system from the
domestic and external bond exchange in 2003 was small. More than 50 percent of the bonds were held
by domestic creditors, mostly retail investors, while domestic bank holdings of government bonds
were relatively low, at less than 5 percent of total bank assets, and mostly held as HTM. In this case,
the fact that domestic banks were mostly holding the sovereign debt as HTM did not matter since the
bank supervisor provided strong regulatory incentives for banks to participate in the exchange (see
Box 8).
 300

Box 8. Central Bank Liquidity Provision in Past Reprofiling Cases

In general, central banks provide liquidity to banks subject to eligible collateral and appropriate
haircuts. Banks need to be solvent to have access to central bank liquidity (varying maturities and
frameworks). If banks lose eligible collateral, they can often gain access to Emergency Liquidity
Assistance (ELA) with a wider pool of collateral accepted. But this is typically more expensive and
with more conditions attached, given the higher level of credit risk for the central bank.
A debt reprofiling would immediately affect the regular central bank liquidity provision. A debt
reprofiling typically leads to a Selected Default (SD) rating of the affected sovereign debt securities,
which cease to be eligible for regular central bank liquidity. Banks will then have to access ELA until
the bond exchange is successfully completed and the SD rating is lifted. Past experience from some
successful debt reprofiling episodes indicates that the duration of the SD tends to be rather short: 0.2
months in Cyprus (July 2013), 0.5 months in Uruguay (May 2003), and 0.7–1.3 months in Jamaica
(2010, 2013), with Pakistan (July 1999) an outlier at 11 months.1 Even in the Greek debt restructuring
case in March 2012 the SD rating only lasted 2¼ months. So the length of ELA provision during a debt
reprofiling SD can be typically minimized unless other factors are at play (e.g., bank collateral scarcity
or bank solvency concerns).
The central bank liquidity provision framework in past debt reprofiling cases has been
heterogeneous:
Cyprus (2013): At the time of the debt reprofiling in June 2013, the affected Cypriot banks already
had only liquidity access through ELA from the Central Bank of Cyprus and not the ECB and
Eurosystem. Thus, no ECB and Eurosystem liquidity provision was involved. Even if prior to the bond
exchange, Cypriot bank were drawing regular ECB liquidity, ELA during the SD duration (0.2
months) would have been very short.
Jamaica (2010, 2013): In the case of these bond exchanges, the authorities set up a Financial Stability
Support Fund (FSSF) to help with any liquidity (with new bonds as collateral) and solvency support
for financial institutions that exchanged the majority of their sovereign holdings. For instance in the
2010 exchange, the FSSF was established to provide liquidity in case of external funding or deposits
withdrawals, or if assets under management were affected by the debt reprofiling (Grigorian and
others, 2012). Financial institutions qualified to access the FSSF if they at least exchanged 90 percent
of their old bonds. Banks also had access to the Bank of Jamaica temporary discount window subject
to liquid collateral. Affected banks were subject to a maximum liquidity maturity (6 months) or
increased regulatory intervention. In the end, no Jamaican bank has drawn on the FSSF liquidity
support (IMF, 2013c).
Pakistan (1999): Following the bond exchange in December 1999, the State Bank of Pakistan
provided liquidity through its lender-of-last resort (LOLR) discount window and also offered liquidity
through its regular open market operations (though there was no take up in the immediate months after
the exchange). There was no destabilizing impact on the domestic banking sector after the bond
exchange.
Uruguay (2003): In the May 2003 bond exchange, the central bank only allowed new bonds as
eligible collateral, thus encouraging domestic banks to participate in the debt reprofiling (IMF, 2003).
Prior to the exchange, the central bank announced that old bonds would no longer be eligible for
liquidity assistance. The old bonds were effectively deemed not-marketable, which would have
penalized the banks’ provisioning and capital adequacy ratios, had they failed to participate in the
exchange.
__________________________________
1
Pakistan’s sovereign debt was downgraded to selective default in January 1999 because of external arrears accumulated in
late 1998. A rescheduling agreement with the Paris Club—reached in January 1999—compelled the Pakistan government to
seek comparable treatment with its private external creditors. The Eurobond exchange was completed in December 1999,
with an upgrade in sovereign rating.
 301

Box 9. Accounting Treatment of Bank Holdings of Government Bonds

The impact of changes in bond valuations on a bank’s balance sheet depends on the account in
which the securities are held. In general, banks’ balance sheets are directly impacted from valuation
changes in their trading sovereign portfolios, while a prudential filter on Available-for-Sale (AFS)
securities excludes a regulatory capital impact (in Basel II), and HTM securities are not marked to
market.
Trading Portfolio
Banks have to mark-to-market (MTM) their trading book, typically on a daily basis, affecting
their regulatory capital calculation, via the profit & loss (P&L) account. Thus, if a bank has not
already divested its trading holdings of sovereign bonds by the time of a reprofiling, it would stand to
make MTM gains following the exchange, as yields have usually fallen after past cases of maturity
extensions (partially compensating for any MTM losses prior to the reprofiling).
Available for Sale (AFS) Portfolio
For AFS sovereign securities, unrealized gains and losses are taken to OCI (Other
Comprehensive Income), which is excluded from the capital calculation via a prudential filter in
Basel II. This avoids swings in capital from, for example, large changes in yields. If, as part of an
impairment test, a bank assesses that there is objective evidence that an AFS sovereign asset has
become impaired, impairment losses are recorded in P&L and the cumulative losses that have been
recognized in OCI are recycled to P&L (subject to certain criteria) and would therefore impact capital.
The impairment loss can be reversed if, after recognizing the loss, the fair value of the AFS sovereign
asset increases. As a general matter, it is not clear if a reprofiling would be treated as an impairment
event, although the cases reviewed in this paper show that in practice it has not been treated as such
(Cyprus, Jamaica).1
While the prudential filter for AFS is to be phased out under Basel III, the European CRDIV/
CRR allows an exemption. Basel III eliminates the prudential filter with a 20 percent phase-in
period a year. Thus, after the phase-in period, banks’ regulatory capital could be subject to swings.
However, the European CRDIV/CRR legislation introduced an exemption that allows national
discretion to maintain the prudential filter so that banks’ common tier 1 capital from the exposure to
central government securities will not be subject to any unrealized gains and losses.2 Some EU
countries such as Ireland or Italy already allowed their banks to keep the AFS prudential filter under
Basel III.
Held to Maturity (HTM) Portfolio
HTM sovereign securities are held as longer-term investments, and thus are not marked-to-
market. Similar to AFS, HTM securities are also subject to an accounting impairment test. Any
impairment loss recognition and measurement would happen for fair presentation of financial
statements at reporting date (e.g., quarterly or yearly). For both AFS and HTM, incurred impairment
losses are taken to the P&L and directly affect capital. As mentioned above, only market fluctuations
for AFS are taken to OCI and recycled (due to the prudential filter) when there are impairment losses.
In certain conditions, reclassifications from AFS into HTM are allowed. For instance, during the
global financial crisis in October 2008, international accounting standards (IAS 39/ IFRS 7) were
amended. As a consequence, many bank supervisors allowed their banks to reclassify their sovereign
assets. For example, the Turkish bank regulator in October 2008 allowed domestic banks to reclassify
on a one-time basis their AFS securities to HTM, thereby avoiding the need for mark-to-market pricing
and recording capital write-downs when securities prices fall. Philippines allowed a similar one-time
reclassification during the 2008 crisis. In contrast, there are conditions by which a HTM security
would have to be reclassified as AFS and re-measured at fair value.3
 302

Box 9. Accounting Treatment of Bank Holdings of Government Bonds (concluded)

If domestic banks hold sovereign bonds as HTM and those are then reprofiled before maturity,
there has been flexibility in practice as to whether banks were able to keep them as HTM or had
to classify them as AFS and re-measure them at fair value. According to paragraph 9 in IAS39, a
bank that has sold or re-classified a sizable amount of HTM before maturity during the current
financial year or the previous two financial years is not allowed to classify any financial assets as
HTM. Importantly, there are exceptions if, for instance, the sale or re-classification of the HTM
security occurs close to the maturity date (e.g., three months), the bank has already collected the
significant amounts of the original principal through schedule payments or prepayments, or they are
related to an “isolated event that is beyond the entity’s control, is nonrecurring and could not have
been reasonably anticipated by the entity.”
Overall, there appears to be some accounting scope during a debt reprofiling episode (including
potential forbearance) for a bank not to move its HTM securities to AFS and re-measure at fair
value. This can occur if the debt reprofiling happens just before the maturity date (as e.g., in the case
of Cyprus in 2013), or if the bank deems the reprofiling as an isolated and unforeseen event.
Forbearance could occur as a result of (i) an objective assessment at the level of the bank (as part of an
impairment test and following evidence); (ii) a captive auditor or (iii) regulatory rules that might
facilitate it. For instance, the 100 percent risk-weight on FX bonds during the 2010 Jamaica bond
exchange was phased-in during a 2-year period. Even if banks keep the sovereign debt as HTM
following the bond exchange, an impairment test based on objective evidence would eventually
indicate that the bank would need to provision if bond prices are well below the par value. Overall, it is
fair to say that a reprofiling will not per se require a reclassification.
___________________________
1
The relevant accounting references are made in IAS 39: “The cumulative loss that had been recognized in OCI
shall be reclassified from equity to P&L…,” when there is objective evidence that the AFS asset is impaired.
Furthermore, “if, in a subsequent period, the FV of a debt instrument classified as AFS increases, and the
increase can be objectively related to an event occurring after the impairment loss was recognized in P&L, the
impairment loss shall be reversed, with the amount of the reversal recognized in P&L.”
2
The CRR article 467 allows the ASF prudential filter. The forthcoming new accounting rules IFRS 9 are still
unclear whether the AFS category will be eliminated or kept in some form.
3
According to IAS39, Article 51, “if, as a result of a change in intention or ability, it is no longer appropriate to
classify an investment as held to maturity, it shall be reclassified as available for sale and re-measured at fair
value, and the difference between its carrying amount and fair value shall be accounted for.”
 303

Box 10. Banking Sector Developments in Greece during the Crisis1

The Greece experience showed that a bailout does not necessarily insulate the domestic financial
system. The banking system was perceived to be relatively sound when the Stand-by Arrangement
program began in May 2010. The bank capital ratio was 11.7 percent, aided by a recapitalization in
2009, but balance sheets came under pressure from higher nonperforming loans (NPLs) once the
economy weakened. Moreover, liquidity conditions tightened in 2009 due to banks losing wholesale
market access and some deposit outflows.
As the recession intensified and liquidity tightened, the Greek financial sector became
increasingly vulnerable. Financial sector distress was a result of the protracted recession and
sovereign debt problems. By 2011, deleveraging in the financial sector and restructuring of state-
owned banks was perceived as necessary. ATE, the largest state-owned bank and the only Greek bank
to fail the Europe-wide stress tests in mid-2010, had to be recapitalized. Sizable deposit outflows
began in the second half of 2011, fanned by fears of a Greek euro exit.
The ECB provided substantial and extraordinary liquidity support. From May 2010, the ECB
suspended the link between sovereign credit ratings and eligibility of collateral for refinancing
operations and intervened directly in the government bond market under the Securities’ Market
Program (SMP). The ECB also began to accept uncovered bank bonds guaranteed by the government
as collateral eligible for refinancing operations.
As expected, the debt restructuring eliminated the banks’ capital but appropriate remedial
measures were implemented under the program. Greek banks were heavily exposed to the
sovereign, holding government bonds with a book value of about €40 billion (after some initial June
2011 impairments). By contrast, core capital was €22 billion, or about the same magnitude as the
capital needs arising from the Private Sector Involvement (PSI). Only €1.5 billion was drawn from the
Hellenic Financial Stabilization Fund (HFSF) during the SBA-supported program, but the banks’
capital needs subsequently dwarfed the HFSF provision. As of the fourth review, the purpose of the
HFSF changed from a means of topping up capital for banks that had tried and failed to raise private
capital to providing a substantial injection of public funds for banks that had been severely affected by
the PSI and the deep recession: the amount needed for the HFSF in the context of the EFF was
estimated at €50 billion.
__________________________________
1
This discussion is based on IMF (2013b).
 304

References

Hesse, Heiko, 2014, “Reprofiling and Domestic Financial Stability: Recent Experiences,”
Annex to the IMF board paper “The Fund’s Lending Framework and Sovereign
Debt,“ 2014

Grigorian, D. A, T. Alleyne, and A. Guerson, 2012, “Jamaica Debt Exchange,” IMF Working
Paper 12/244 (Washington: International Monetary Fund).

International Monetary Fund, 2013a, “Cyprus: First Review under the Extended
Arrangement under the Extended Fund Facility,” IMF Country Report 13/293
(Washington).

———, 2013b, “Greece: Ex Post Evaluation of Exceptional Access under the 2010 Stand-By
Arrangement,” IMF Country Report 13/156 (Washington).

———, 2013c, “Jamaica: Request for an Extended Arrangement under the Extended Fund
Facility,” IMF Country Report 13/126 (Washington).

———, 2003, “Uruguay: Article IV Consultation and Third Review Under the Stand-By
Arrangement—Staff Report,” IMF Country Report 03/247 (Washington).

———, 2001, “Pakistan: 2000 Article IV Consultation and Request for a Stand-By
Arrangement—Staff Report,” IMF Country Report 01/24 (Washington).
 305

XIII. IS BANKS’ HOME BIAS GOOD OR BAD FOR DEBT SUSTAINABILITY? WITH TAMON
ASONUMA AND SAID BAKHACHE, 2015247

Motivated by the recent increase in domestic banks’ holdings of domestic sovereign debt
(i.e., home bias) in the European periphery, this paper analyzes implications of banks’ home
bias for the sovereign’s debt sustainability. The main findings, based on a sample of
advanced (AM) and emerging market (EM) economies, suggest that home bias generally
reduces the cost of borrowing for AMs and EMs when debt levels are moderate to high. A
worsening of market sentiments appears to dimish the favorable impact of home bias on cost
of borrowing particularly for EMs. In addition, for AMs and EMs, higher home bias is
associated with higher debt levels, and less responsive fiscal policy. The findings suggest that
home bias indeed matters for debt sustainability: Home bias may provide fiscal breathing
space, but delays in fiscal consolidation may actually delay problems until debt reaches
dangerously high levels.

A. Introduction

Issues related to the entrenched sovereign-bank nexus particularly home bias—banks’

holdings of sovereign domestic debt—have gained prominence during the financial
crisis in recent years as public debt was rising especially in the European periphery.248
Prompted by foreign investors’ flight as well as cheap long-term refinancing operation
(LTRO) funding from the European Central Bank (ECB), many peripheral banks absorbed
sizable domestic sovereign debt both from the secondary and primary markets. The
entrenched sovereign-bank nexus has raised concerns regarding the health of the banking
sector as well as its potential impact on debt sustainability of the sovereign.249 It is worth
noting that the recent increase in home bias is not unique to the European periphery, but has
generally been observed across many advanced economies that have seen a rise in public
debt (Figure 53).

247
This chapter is based on Asonuma, Bakhache and Hesse (2015).
2
Banks’ home bias typically denotes the preference of domestic banks for holding domestic sovereign debt
instruments compared to other sovereign debt instruments. However, the two most commonly used measures of
home bias in the literature (holding of domestic sovereign debt in percent of total assets, and in percent of total
debt) tend to also capture other factors such as investor base diversification.
249
This was one of the reasons that the 2011 EBA stress test findings were not perceived as credible so the
recapitalization exercise forced European banks to mark-to-market all of their securities holdings (e.g., IMF
2011 and 2012).
 306

Figure 53. Banks’ Holding of Domestic Sovereign Claims/Total Bank Assets

and Public Debt

Sources: Arslanalp and Tsuda (2012); IMF IFS; IMF WEO.

This paper examines the impact of banks’ home bias—banks’ holding of domestic
sovereign debt in total assets—on advanced markets (AMs) and emerging markets
(EMs) economies by addressing four questions that have implications for debt
sustainability:

(1) Is the cost of borrowing lower for sovereigns with higher home bias? Yes, for both
AMs and EMs using bond spreads and domestic bond yields respectively. The
negative relationship between home bias and the domestic cost of borrowing is milder
for EMs. Worsening of market sentiments tend to temper the effect of home bias on
borrowing cost particularly for EMs.

(2) Is the level of public debt higher in countries with greater home bias? Our panel
regressions show that this is the case for both AMs and EMs.

(3) Is a primary balance adjustment slower in sovereigns with higher home bias? Our
empirical results show that indeed sovereigns with a higher home bias are less willing
to conduct fiscal consolidation. Even if foreign investors have reduced their exposure
to domestic sovereign debt markets, the presence of domestic banks ablility to absorb
the domestic debt issuances can provide a significant breathing space to struggling
sovereigns but this deepens the negative sovereign-bank feedback loop and could
potentially delay needed fiscal adjustment.250

250
In the European periphery, external market pressures and soaring sovereign yields have forced peripheral
countries to start implementing some of the overdue reforms. Decisive monetary policy by the ECB has
certainly helped to provide a backstop to the peripheral domestic banks, which was the predominant factor for
compressing peripheral sovereign spreads.
 307

(4) Do sovereigns with higher home bias enter into debt distress at a higher level of
public debt? Our findings suggest a positive relationship between home bias and the
level of debt at which countries are assessed to have experienced debt difficulties.

The empirical results strongly suggest that home bias matters for debt sustainability.
High home bias which in some cases is tantamount to having a captive investor base may
provide fiscal breathing space, but delays in fiscal consolidation may actually postpone
problems until debt reaches dangerously high levels. The breathing space is largely the result
of the favorable impact of high home bias on rollover risk which is particularly evident
during crisis periods, but is not likely to yield better fiscal outcomes. The empirical analysis
in this paper which examines the multi-faceted impact of home bias provides analytical
support for anecdotal evidence in this regard. For example, during the recent crisis period
countries with a captive domestic investor base faced less market pressure on rollover needs,
and therefore enjoyed more breathing space while attempting difficult fiscal consolidations
(or during periods of lax fiscal discipline).

Our main findings broadly hold in robustness checks. For instance, dropping outliers such
as Greece and Japan from the country sample does not change the empirical results. The
findings are also generally robust to alternative regression methodologies as well as different
home bias measures. The only exception is that while we find a negative relationship
between our preferred home bias measure (holdings of domestic sovereign debt relative to
total assets) and AM borrowing costs, the relationship is positive for the home bias measure
that has total public debt as the denominator. This finding is not be surprising because while
our preferred home bias measure mainly captures the banks’ preference for sovereign debt,
the other measure mainly reflects the diversification angle. A more diversified investor base
(i.e., lower home bias) is associated with lower spreads. These findings are consistent with
the literature. For example, Arslanalp and Poghosyan (2014) and Andritzky (2012) find that
the diversification home bias measure and borrowing costs are positively related. However,
Acharya and Steffen (2013), using the bank preference home bias measures find a negative
relationship in line with our results.

A number of factors could explain banks’ home bias. Domestic sovereign debt tends to
enjoy a preferential regulatory treatment with a zero risk-weighting. In this context however,
risk weights on other assets, including foreign sovereign debt, might differ significantly
between countries which potentially could contribute to cross-country variations in home
bias. The increase in home bias during and after the recent crisis period across many
countries benefited from the higher importance of domestic sovereign debt for central bank
collateral (as well as market funding). There could be structural factors (such as market
infrastructure or lack investment opportunities) or business cycle considerations. The supply
of public debt has also increased especially in many advanced economies after the crisis.
While this paper does not formally address the potential determinants of home bias, it
provides some conceptual discussion to contextualize recent developments.
 308

Given the focus of this paper on the impact of banks’ home bias on debt sustainability,
it does not address a number of key issues related to home bias. First the concept of home
bias generally goes beyond the banking sector and includes all domestic investors including
non-bank institutions such as pension or insurance companies. While these entities can be
sizable in some countries, the paucity of cross country data limits the ability to include them
in the analysis. Second, while the empirical panel regression methodology attempts to
account for endogeneity issues, e.g., between home bias and public debt, by using
instruments, there might still be the possibility of reverse causality in some circumstances.
For instance, a government faced with increased rollover risks could use moral suasion with
domestic banks (e.g., primary dealers) to increase their holdings of sovereign debt.

The paper is organized as follows: Section B briefly discusses the existing literature on
home bias. Section C presents the empirical analysis on the relationship between home bias
and borrowing costs, level of public debt, primary balance adjustment, and the level of debt
at which countries enter distress. A discussion of robustness checks is also included. Section
D discusses a number of additional issues related to home bias, particularly its determinants.
Section E concludes.

B. Literature Review

The growing literature on sovereign debt home bias and its implications for debt
sustainability is relatively heterogeneous:

 One strand of literature examines potential causes for home bias in bonds. For
instance, Fidola and others (2006) find that exchange rate volatility induces a stronger
home bias in bonds.251, 252 Portes and others (2001) show that government bonds
respond less to information frictions than corporate bonds or equity.

 Existing research points to different effects of home bias on sovereign bond

yields. For example, Arslanalp and Poghosyan (2014) and Andritzky (2012), who
examine the diversification angle of sovereign claims, find that an increase in the
share of government debt held by domestic investors leads to an increase in sovereign
bond yields in AMs. Similarly, Ebeke and Lu (2014) show that an increase in share of
government bonds held by domestic residents has reduced bond yields in EMs.
Acharya and Steffen (2013), on the other hand, find that home bias (measured by

251
The measure of home bias here is based on the share of domestic banks’ holdings of domestic sovereign
claims in total holdings of sovereign claims. It reflects the degree to which banks are overweight in domestic
sovereign claims and underweight in foreign sovereign claims, as compared to the benchmark portfolio on
sovereign claims.
252
Applying a similar approach, Chan, and others (2005) show that transaction costs are less important than
informational asymmetries in explaining equity home bias.
 309

banks’ holding of domestic sovereign debt relative to total assets) actually helped to
lower spreads in the European periphery after the systemic crisis.253

 Beyond the impact on bond yields, the consequences of home bias are found to
be multi-faceted. For instance, countries with high home bias tend to have high
public debt (BIS, 2011) and face higher spillover risks of sovereign stress to the
banks (Merler and Pisani-Ferry, 2012). Related, a high concentration of domestic
sovereign claims in banks’ balance sheets results in an increase of spillover risks of
sovereign stress to banks (Acharya and others 2012). Domestic banks tend to also
increase exposure to sovereign claims when they are hit by country-specific and
common shocks on yields (Battistini and others, 2013). Moreover, domestic banks,
especially large banks increase their exposure to sovereign claims during sovereign
defaults (Gennaioli and others, 2014a). From a theoretical perspective, Gennaioli and
others (2014b) show that home bias reduces the probability of default on public debt
due to high cost of default on the domestic economy.254, 255

This paper contributes to the existing literature by taking a broad perspective on the
implications of home bias for debt sustainability. On the impact of home bias on
borrowing costs, we also explicitly account for the role of public debt levels as well as
market sentiment when examining the relationship between home bias and borrowing cost
and account for potential differences between AMs and EMs. The paper also sheds light on
the role of banks’ home bias in explaining cross country differences in public debt levels,
primary balance adjustments, and the level of debt at the time of fiscal distress.

C. Empirical Analysis on Home Bias

We analyze the role of home bias in debt sustainability by examining the relationship
between home bias and (A) borrowing costs of sovereigns, (B) the level of public debt, (C)
the fiscal primary balance of sovereigns, and (D) the level of debt at which sovereigns enter
debt distress.

Throughout the paper, we focus on the home bias indicator that reflects the banks’
holding of domestic sovereign claims in total assets as in Acharya and Steffen (2013).
This allows for the widest country coverage in the sample (22 AMs and 29 EMs) and longer
coverage of time horizon (1999–2012) in annual frequency, which reflects banks’ preference

253
Jaramillo and Zhang (2013) find that an increase in bond yields due to high debt to GDP ratio is partly offset
if this debt is in the hands of real money investors—domestic nonbanks and national and foreign central banks.
254
The above studies measure home bias as the share of domestic banks’ sovereign holdings in total bank assets.
255
In a theoretical model of defaultable sovereign debt and banks, Boz and others (2014) find that sovereign
default amplifies the business cycle as banks’ losses due to a default hampers lending to firms.
 310

on domestic sovereign claims over other assets. Two other measures of home bias are used
for robustness checks in Section H.256

Home Bias = Banks’ holding of domestic sovereign claims / Banks’ total assets257

The choice of macro variables in the regression analysis follows the academic literature
on borrowing costs, debt accumulation, and fiscal reaction function. In line with Ghosh
and others (2011), control variables for the fiscal reaction function include the level of lagged
public debt, output gap, fiscal expenditure gap, and trade openness. Following Ardagna and
others (2007), conventional determinants of bond yields / spreads include the level of lagged
debt, GDP growth rate and inflation rate. The VIX is used as a proxy for global risk aversion
of investors. Details on macro variables are provided in Annex 18.

AMs and EMs with high home bias tend to have high public debt in both the pre- and
post-global financial crisis periods (Table 20 and Figure 54). The average public debt
level of AMs and EMs whose home bias is in the top (bottom) quartile of the distribution is
higher (lower) than the the mean of the corresponding sample. This evidence is found to be
robust with any sample period during 1999–2012. Home bias in EMs on average tends to be
higher than that in AMs because of potentially narrower portfolio allocation options available
for banks in EMs and relatively limited access to international capital markets.

Table 20. Summary of Home Bias Indicators (average, 2005–07 and 2009–11)
Advanced markets Emerging markets
Home Bias Public debt / GDP Home Bias Public debt / GDP
2005-7 Mean (whole sample) 5.5 56.4 12.8 51.1
(debt - 2007) Mean of observations
- Above 75th percentile 11.3 95.1 28.6 75.9
- Below 25th percentile 1.8 37.8 1.7 20.3
2009-11 Mean (whole sample) 6.5 79.9 13.5 54.0
(debt - 2011) Mean of observations
- Above 75th percentile 14.8 120.2 26.7 78.8
- Below 25th percentile 1.3 61.1 3.5 24.1
Sources: Arslanalp and Tsuda (2012); IMF IFS; IMF WEO.

256
Details of the country sample, calculations of home bias measures, and sources of data are reported in Annex
13.1.
257
The home bias measure uses banks’ holding of claims to general government given data constraints in some
countries that make it more difficult to consistently examine the home bias measure for only the central
government. Similarly, debt-to-GDP and the primary balance-to-GDP refer to general government. In contrast,
both the sovereign borrowing costs and sovereign default events relate to the central government.
 311

Figure 54. Average Public Debt (2007) and Home Bias (average, 2005–07)
in AM and EM1

Sources: Arslanalp and Tsuda (2012), IMF IFS, IMF WEO.

1
Excluding Japan, average public debt for AMs is 76 percent of GDP in countries whose HB falls in the top quartile of
the distribution.

D. Borrowing Costs of Sovereigns

Sovereign’s Borrowing Costs—Panel Analysis

We assess the impact of home bias on sovereigns’ borrowing costs in AMs and EMs
based on a panel regression approach. Our method is a two-step generalized method of
moment (GMM) estimation with housing price and lagged credit-to-GDP ratio as instruments
to control for home bias in order to deal with potential endogeneity issues. The housing price
variable is considered to be appropriate as one of the instruments because it is correlated with
banks’ holdings of domestic sovereign claims but not with bond yields or spreads. In
addition, the lagged credit-to-GDP variable is predetermined, and thus it is correlated with
neither bond yields nor spreads but affects the portfolio allocation of banks. These two
instruments are significant and have significant explanatory power supported by an Adj-R2 of
0.54 (AM), 0.66 (EM) and 0.59 (AM & EM).

Model specifications closely follow Ardagna and others (2007) in terms of determinants
of spreads and shown as:

ri ,t   1bi ,t 1   0 ( hbi ,t  hb t )   1 (( hbi ,t  hb t ) * bi ,t 1 )   2 (( hbi ,t  hb t ) * VIX t )  x i ,t   y t    it

--- (1)
 312

where ri ,t captures long-term bond yields or spreads of country i at t ,258,259 bi ,t is the level of
public debt of country i at t , respectively. The variable hbi ,t is the home bias for country at t
, hb t is the sample average of home bias at time t , VIX t is a proxy for global risk aversion
at time t , and xi ,t is a vector of macroeconomic variables. The first lagged debt term reflects
the effect of debt on borrowing costs pointed out by Ardagna and others (2007).260 While the
second term reflects the direct influence of home bias, the third and fourth terms capture how
home bias interacts with debt and global risk aversion, respectively. To account for the multi-
dimensional impact of home bias on borrowing costs, we introduce a non-linear function of
home bias. Though a majority of home bias observations fall around the sample median, a
sizable fraction of observations is close to zero. An introduction of an interactive term of
home bias and debt (VIX)—home bias multiplied by these variables—provides asymmetric
and biased effects in the range of home bias. To correct asymmetry and bias effects and
capture symmetrically and precisely its interaction with the level of debt and market
sentiment (VIX), the home bias variable is entered as a deviation from its sample median.

Findings suggest that for AMs with moderate to high debt levels, borrowing costs
(measured by the spreads over German/US bonds) generally decline as home bias
increases especially in normal times (Figure 54, panel A). Lower spreads with high home
bias reflect reduced expectation of default whenever domestic banks own a sizable portion of
domestic sovereign claims because of the anticipated high cost of default. As shown in the
panel regression results (3rd column of Table A20.1), bond spreads are negatively influenced
by interactive terms of home bias and debt whereas positively affected by interactive terms of
home bias and VIX (a proxy for investor risk aversion). The former effect clearly dominates
the latter leading to a decrease in spreads due to higher home bias when the level of debt is
above 50 percent of GDP. This is also highlighted in a downward sloping curve of bond
spreads. When debt is below 50 percent of GDP, the interaction between home bias and debt
level is low, and therefore the interaction between home bias and VIX dominates. Thus the
overall impact of home bias on bond spreads is smaller compared to the case of high debt. In
cases of high debt levels and large spreads, sovereigns enjoy larger reductions in spreads due

258
Throughout our empirical analysis, we focus on yields or spreads of long-term bonds, i.e. benchmark 10-year
bond yields or equivalent bond yields in the local market. There could be some possibility that the impact of
home bias on bond yields/spreads varies across different maturity of bonds, but data limitations prevent a closer
look at this issue.
259
For AMs, we use bond spreads against the German bonds for European countries and the US bonds for non-
European countries since we are interested in how borrowing costs for sovereigns deviate from those of “risk-
free” bonds, i.e. the German or US bonds. For EMs, we use sovereign bond yields.
260
Using a panel of 16 OECD countries over several decades, Ardagna and others (2007) use both linear and
nonlinear (including quadratic term) model specifications to have an increasing function of spreads and find that
the effect of the debt level on interest rates is nonlinear only for countries with above-average levels of debt.
 313

to a “real money investors”—domestic nonbanks and national and foreign central banks point
increase in home bias (relative to its median). This is highlighted by a steeper slope of the
line associated with debt at 100 percent of GDP.261 These results mentioned above remain
robust if we omit Greece, Japan and Portugal from the sample (4th column of Table A20.1).

However, during crisis periods when risk aversion rises, the negative impact of home
bias on spreads diminishes and may turn positive (Figure 55 panel B).262 For example
when VIX is 80 points above the sample mean and debt is relatively low (60 percent of
GDP), the interaction of home bias and VIX could surpass the interaction of home bias and
debt. The intuition is that banks demand higher risk premia during periods of increased risk
aversion, while continuing to hold more domestic sovereign claims.263 Coefficient on control
variables have the expected signs and significance: higher growth rate and higher
institutional quality significantly reduce bond spreads, whereas an increase in credit increases
bond spreads (3rd column of Table A20.1).

261
In addition, the steepness of the slope is explained by the fact that a small change in the ratio of the home bias
indicator is generated by a large increase in the numerator. There also can be some difference between AMs and
EMs since banks’ total assets are significantly large in AMs (440 percent of public debt), compared to those in
EMs (220 percent).
262
Throughout our analysis, we consider a “crisis period” as a period when investor risk aversion, proxied by the
VIX, deviates substantially from its sample mean (22.7 point).
263
In a similar vein, Broner and others (2014) theoretically find that discrimination between domestic and
foreign creditors in turbulent times provide incentives for domestic purchases of debt. Creditor discrimination
could be due to sovereigns’ incentive to avoid defaults on domestic debt or ad hoc domestic regulations
imposed during these turbulent periods.
 314

Figure 55. Bond Spreads and Home Bias in AMs

(A) “Normal Times”1 (B) “Increased risk aversion” Period2

Sources: IMF WEO; and Fund staff estimates.

1
VIX is fixed at sample mean (22.7 point).
2
Debt is fixed at 60 percent of GDP.

Similar to the results for AMs, home bias is also in general negatively associated with
bond yields in EMs. This also reflects the lower expectation of sovereign default when
domestic banks hold sovereign debt. However, home bias seems to have a milder impact on
cost of borrowing as seen in the flatter lines in Figure 56 panel A and 3rd column of Table
A20.2. On the contrary, market sentiments seem to play a bigger role in EMs than AMs. In
particular, for debt level of 60 percent of GDP the impact of home bias on cost of borrowing
becomes positive when VIX is slightly above its mean as seen Figure 56 panel B. Control
variables enter with expected signs and significance: higher growth rate and lower inflation
rate remarkably reduce bond spreads, whereas an increase in investor’ risk aversion (VIX)
and US long-term interest rates increases bond yields in EMs (3rd column of Table A20.2).
 315

Figure 56. EM Sovereigns Borrowing Costs in the Domestic Market

(A) “Normal Times”1 (B) “Increased risk aversion” Period2

Sources: IMF WEO; and Fund staff estimates.

1
VIX is fixed at sample mean (18.5 point).
2
Debt is fixed at 60 percent of GDP.

The relatively large importance of market sentiments in EMs is supported by the strong
correlation between the VIX and EM bond spreads. We apply a multivariate generalized
autoregressive conditional heteroskedasticity (GARCH) framework on a sample of monthly
data from 1999 to end August 2013. The GARCH allows for heteroskedasticity of the data
and a time-varying correlation in the conditional variance. The methodology is explained in
Annex 19. For AMs, we calculate the spread to German Bunds (based on 10-year
instruments) for European countries, while for the non-European countries; the spread to the
10-year U.S Treasury bonds is calculated. For EMs we use the EMBI Global spread because
of the lack of high frequency domestic yield data. The dynamic conditional correlation
(DCC) GARCH models are estimated in first differences to account for the nonstationarity of
the variables in crisis periods. Figure 57 illustrates the GARCH findings for a selective
sample of EM and AM countries, respectively. The estimated correlation coefficients
between spreads and the VIX increase for both EMs and AMs during periods of financial
stress, e.g., 2008 and 2011, but overall co-movements are higher for EM. The average
estimated co-movement between EM spreads and the VIX is 43 percent for the full sample
period (1999–2013) compared with 14 percent for AM countries. The large difference holds
for different sample periods (pre-Lehman, during and post global financial crisis) as well
 316

(Table 21).264 The relatively low co-movement in AMs is consistent with the safe asset role
AM sovereign bonds play particularly during crisis periods.

Figure 57. Estimated GARCH Correlations with VIX

0.9 0.6

0.8 0.5
0.7
0.4
0.6
0.3 VIX-Ireland
VIX-Mexico
0.5 VIX-Italy
VIX-Russia 0.2
0.4 VIX-Portugal
VIX-Turkey
0.1 VIX-Spain
0.3

0.2 0

2/1/1999
2/1/2000
2/1/2001
2/1/2002
2/1/2003
2/1/2004
2/1/2005
2/1/2006
2/1/2007
2/1/2008
2/1/2009
2/1/2010
2/1/2011
2/1/2012
2/1/2013
12/1/2000
11/1/2001
10/1/2002

12/1/2011
11/1/2012
2/1/1999
1/1/2000

9/1/2003
8/1/2004
7/1/2005
6/1/2006
5/1/2007
4/1/2008
3/1/2009
2/1/2010
1/1/2011

-0.1

-0.2

Sources: Bloomberg and Fund staff estimates.

Table 21. Average EM and AM Estimated GARCH Correlations with VIX

Correlation with VIX Mexico Poland Russia Turkey China Malaysia PhilippineBrazil Peru Venezuel Bulgaria Hungary EM Av
Average 58% 41% 54% 46% 25% 35% 46% 57% 43% 46% 46% 14% 43%
Pre-Lehman 58% 40% 54% 46% 24% 34% 47% 55% 39% 44% 45% 13% 42%
During GFC 60% 40% 57% 49% 25% 35% 45% 66% 57% 53% 47% 14% 46%
Post-GFC 58% 42% 54% 46% 26% 36% 46% 59% 47% 51% 46% 16% 44%

Ireland Italy Portugal Spain Austria Finland France NetherlandAustralia Canada Japan UK AM Av
Average -1% 28% 9% 26% 29% 14% 11% 6% -6% 23% 17% 7% 14%
Pre-Lehman -2% 27% 9% 25% 29% 13% 11% 5% -5% 23% 17% 6% 13%
During GFC 2% 32% 13% 27% 32% 18% 16% 9% -11% 15% 10% 9% 14%
Post-GFC 1% 29% 9% 26% 30% 14% 11% 6% -7% 25% 19% 9% 14%

Note: The pre-Lehman period is before September 2008, the Global Financial Crisis (GFC) between September
2008 and June 2009, while the post- GFC period is July 2009–August 2013 (end of sample period). Correlation
between domestic yields and EMBI spreads in annual frequency is 0.38 over the period 1999–2012.
Sources: Bloomberg and Fund staff estimates.

E. Public Debt

We analyze whether home bias contributes to a high public debt level in AM and EM
possibly through reduced borrowing costs. To avoid the endogeneity problem (as in the
borrowing cost regressions), we apply a two-step generalized method of moment (GMM)
estimation using housing price and lagged credit-to-GDP ratio as instruments for home
bias.265 Our model specifications are as follows:

264
The focus here is on the overall correlation magnitudes between AM and EM. Low correlation magnitudes
are unlikely to be statistically significant.
265
Similar to the borrowing costs regressions, we use the housing price and lagged credit-to-GDP variables as
instruments. The housing price variable is correlated with banks’ holdings of domestic sovereign claims but not
with public debt. Moreover, the lagged credit-to-GDP ratio is predetermined and thus not correlated with public
debt. These two instruments have enough explanatory power supported by a high Adj-R2.
 317

bi ,t   1 hbi ,t  x i ,t    it --- (2)

bi ,t   2 hbi ,t 1  x i ,t    it --- (3)

where bi ,t is the level of public debt at t , hbi ,t and hbi ,t 1 are home bias at t and t 1 ,
respectively, and xi ,t is a vector of macroeconomic variables.

An increase in home bias in AMs and EMs is associated with a high public debt level
(Figure 58 and Table A20.3).266 Given the potentially high demand of banks for domestic
sovereign claims, sovereigns, ceteris paribus, tend to issue relatively larger amounts of debt
(Figure 58). Panel regression results on AM and EM samples confirm that home bias has
concurrent effects on the level of debt (2nd and 4th columns of Table A20.3). Moreover, the
subsequent debt level is significantly influenced by home bias due to its persistent feature (3rd
and 5th columns of Table A20.3). Other macro variables result in expected signs and
significance: an increase in credit leads to increase in public debt whereas sovereigns with
high inflation rates, low institutional quality and high degree of capial openness tend to
accumulate high debt (7th column of Table A20.3). As mentioned in the introduction, we do
recognize circumstances of reverse causality. While analytically we accounted for these
endogeneity issues, there are still cases where a rise in public debt might lead sovereigns to
use moral suasion to ensure that new debt can be safely placed with domestic banks, which
increases the home bias indicator.267

266
For the regressions on public debt on a combined sample of AMs and EMs, we introduce a dummy variable
for AMs to account for heterogeneity between the AM and EM samples.
267
Disentangling potential feedback mechanisms between home bias and public debt can be difficult despite our
efforts to control for such endogeneity by using strong instruments, i.e., house prices and the lagged credit-to-
GDP ratio.
 318

Figure 58. Public Debt-to-GDP and Home Bias (Average, 2005–07)

Public debt/GDP Public Debt/GDP in 2007 and Home Bias ave. 2005-7
(2007)
200
AM+EM Debt = 55.3 + 2.1 * Home bias
Japan
(***, 9.63) (***, 0.61)
Linear regression Lebanon
160

St. Kitts and Nevis

120

80
Brazil Argentina

40 Mexico Turkey

0
0 5 10 15 20 25 30 35 40 45
Home Bias - Banks' Holding of Domestic Sovereign Claims/Banks'
Total Assets, Average 2005-7

Source: Fund staff estimates.

F. Primary Balance Adjustments

This section considers whether the primary fiscal balance tends to adjust less to lagged
debt level in countries with high home bias. Our model specifications closely follow
Ghosh and others (2011) to include both square and cubic terms of lagged debt to capture
two inflexion points in the fiscal reaction function. Specifically, Ghosh and others (2011)
explain the appropriateness of the nonlinear fiscal reaction function as follows: At a very low
level of debt, there is little (or even a slightly negative) relationship between lagged debt and
the primary balance. As debt increases, the primary balance rises but the responsiveness
eventually begins to weaken, and then actually decreases at high levels of debt.

pbi ,t   1bi ,t 1   2 bi2,t 1   3 bi3,t 1   0 ( hbi ,t  hb t )   1 (( hbi ,t  hb t ) * bi ,t 1  x i ,t    it -- (4)

where pbi ,t is the primary balance at time t , bi ,t 1 , bi2,t 1 , bi3,t 1 are linear, quadratic and cubic
terms of lagged public debt at t , hbi ,t is the country-specific home bias and hb t is the sample
median of home bias at t , and xi ,t is a vector of macroeconomic variables. The fourth and
fifth terms on the right hand side of equation capture effects of home bias per se and the
interaction between home bias and lagged debt on current primary balance, respectively. As
before, we apply a two-step generalized method of moment (GMM) fixed effects estimation
 319

using housing price and lagged credit-to-GDP ratio as instruments for home bias to deal with
potential endogeneity issues.268 The fixed effects estimation accounts for significant
variations in primary balance adjustments across countries.269

We find that fiscal policy is less responsive to lagged public debt in AMs and EMs with
a higher home bias (Figure 59).270 The average primary balance of countries whose home
bias falls in the top quartile of distribution is substantially lower than that of the whole
sample for a given level of lagged debt. If a country’s home bias deviates from the median by
5 percent, its primary balance is lower by 0.6 percent of GDP compared to the sample mean
(Table 22). Given domestic banks’ interest in domestic sovereign claims, sovereigns might
be less willing to commit to fiscal consolidation, ceteris paribus, despite a high level of
debt.271 As shown in the panel regression results (2nd column of Table A20.4), the primary
balance is negatively affected by home bias (entered as a deviation from its median). Overall,
an elevated home bias might provide the sovereign with more fiscal breathing space since it
potentially reduces the cost of borrowing, even when the debt level is high. As expected,
signs and significance of other macroeconomic variables are the same with Ghosh and others
(2011): primary balance responds positively to the output gap and trade openness, but
negatively to temporary increases in government outlays (captured by government
expenditure gap). In other words, having a captive domestic investor base can reduce rollover
risks for the sovereign and that, ceteris paribus, might translate into lower fiscal
consolidation. While we attempt to address endogeneity issues in the panel regression
framework, we recognize that reverse causality could be an issue in some circumstances
(similar to the home bias-public debt relationship).

268
These two instruments have enough explanatory power supported by a high Adj-R2.
269
Mauro and others (2013) find that there are significant variations in primary balance adjustments across
countries and time periods.
270
Due to data limitations (particularly for primary balance-to-GDP ratio and home bias in EMs), the regression
is based on the combined sample of AMs and EMs.
271
A close relationship between sovereigns and domestic banks creates amplification effects, which could result
in an insufficient degree of fiscal consolidation: With banks willing to absorb domestic sovereign claims, they
also anticipate a low risk of sovereign default given the sovereign is unlikely to impose high default costs on the
banks. In turn and with a captive domestic investor base, sovereigns are less willing to commit to fiscal
consolidation despite a high debt level.
 320

Figure 59. Fiscal Policy Reactions and Home Bias

Sources: IMF WEO; and Fund staff estimates.

Table 22. Estimated Fiscal Policy Reactions

Primary balance-to-GDP (%)
Public debt-to-GDP,
lagged (%) HB 5% below Median of whole HB 5% above
Median observation Median

80 -0.4 -0.9 -1.5

100 0.7 0.1 -0.5
120 1.8 1.2 0.6
160 3.1 2.5 1.9
Sources: IMF WEO; and Fund staff estimates.
 321

G. Debt under Distress

AMs and EMs with high home bias tend to experience debt difficulties at a higher level
of public debt (Figure 60). This finding is based on an event study analysis using a sample
of 17 episodes of debt difficulties based on the methodology of Baldacci and others (2011)
and Cruces and Trebesch (2013). In particular, Baldacci and others (2011) define debt
distress events for AMs as (1) Default: a sovereign not current on its debt obligations
(Standard and Poor definition); (2) Restructuring and rescheduling: any operation which
alters the original terms of the debtor-creditor contract; (3) IMF financing: in excess of 100
percent of quota; or (4) Inflation: greater than 35 percent per annum. For EMs, debt distress
events are defined as: (1) Default: arrears on principal or interest payments to commercial or
official creditors; (2) Restructuring and rescheduling: any operation which alters the original
terms of the debtor-creditor contract; or (3) IMF financing: addressing liquidity issues
associated with sovereign debt distress.

There is a positive relationship between the level of debt at which countries are assessed
to have experienced debt difficulties and home bias (3 year average). The intuition is that
whenever a large share of domestic sovereign claims is held by domestic banks, sovereigns
are less likely to default because defaults would have severe adverse effects on banks
(Gennaioli and others, 2014b). Moreover, due to relatively lower borrowing costs, sovereigns
can tolerate higher levels of debt. Therefore, debt accumulates further before debt
sustainability concerns arise.

Figure 60. Debt Distress and Home Bias

Debt Level in Debt = 75.05 + 2.63 * Home bias

Time of Distress
(In percent of GDP) (***, 19.25) (*, 1.37)
KNA 2011 JAM 2012
160
GRC 2010 JAM 2010

120 PRT 2011

IRL 2010 BLZ 2006
BLZ 2012
ATG 2010HUN 2008
80
BGR 2002 ARG 2001
VEN 2005
40
ECU 2008 ROU 2009
UKR 2008
AM & EM under debt distress
LVA 2008
Linear regression
0
0 5 10 15 20 25
Home Bias - Banks' Holding of Domestic Sovereign Claims /
Banks' Total Assets (3 year average)

Sources: Baldacci and others (2011); and Cruces and Trebesch (2013).
 322

H. Robustness Tests

To support our main empirical findings, we use two other measures of home bias which
reflect banks’ sovereign portfolio allocation (A) and diversification of sovereign claims
by residency of investors (B). Following the existing academic literature, these measures are
defined below (see also Annex 17).

Home Bias (A) = Banks’ holding of domestic sovereign claims / Banks’ holding of
sovereign claims

Home Bias (B) = Banks’ holding of domestic sovereign claims / Public debt

The main empirical results hold when we use the home bias measure (A) which is only
available for AMs. AMs’ borrowing costs are substantially lower for countries with high
home bias (2nd column of Table A20.5). Furthermore, higher home bias in AMs contributes
to higher public debt concurrently and subsequently (2nd and 3rd columns of Table A20.6).
Fiscal policy is less responsive to lagged public debt in AM countries with higher home bias
(2nd column of Table A20.7).

Using the second measure (B) gives different results for AMs, namely that an increase
in banks’ holding of sovereign claims as a percent of total debt leads to an increase in
borrowing costs in AMs. This is not surprising given that this measure in fact captures
investor diversification rather than banks’ assets allocation preference. When foreign
investors’ purchases of domestic sovereign claims increase (indicating a low home bias),
borrowing costs in AM are reduced, partly due to higher competition (3rd column of Table
A20.5). This is consistent with findings in the academic literature (e.g., Arslanalp and
Poghosyan, 2014, Andritzky, 2012, and Warnock and Warnock, 2009). For EMs in contrast,
an increase in this measure reduces domestic borrowing costs similar to our results using our
preferred home bias measure (4th columns of Table A20.5). It appears that for EMs, the
influence of banks’ portfolio allocation dominates investor diversification effect.

Quantile regression techniques using our preferred measure of home bias also support
the previous findings and partly strengthen them. We use the median least squares (MLS)
estimator, which minimizes the median square of residuals rather than the average and thus
reduces the effect of outliers. For instance, we focus on the higher percentile of countries
with home bias and find that the borrowing costs of AM countries are more sensitive to the
home bias variable than in the panel regression framework (5th column in Table A20.1). This
is not surprising since one would expect a higher sensitivity of the home bias and sovereign
borrowing link for countries in say the 75th or 90th percentile distribution.
 323

I. Other Home Bias Issues

While the paper does not analyze the determinants of home bias, it is worth noting that
a multitude of factors could potentially explain differences in home bias across
countries.

 The preferential treatment of domestic sovereign debt in the context of banking

sector regulatory frameworks is universal (see also IMF, 2014). The zero risk-
weighting for capital requirements has substantially contributed to the elevated home
bias. In this context however, risk weights on other assets, including foreign
sovereign debt, might differ significantly between countries which potentially could
contribute to cross-country variation in home bias.

 In the context of financial sector vulnerabilities, domestic sovereign debt has

become increasingly important as central bank collateral as well as for secured
wholesale funding combined with more demanding liquidity requirements (e.g., the
Liquidity Coverage Ratio).

 The supply of public debt has substantially increased in many advanced

countries during the crisis, and led to domestic banks absorbing much of new
sovereign debt issuances especially when there was a foreign investor retrenchment.272
In particular, this occurred in an environment of increased global risk aversion and is
typically accompanied by other home bias factors (as discussed here).

 Structural factors, for example, the availability of other investment

opportunities relative to the size of the banking sector could affect domestic
banks’ holding of domestic sovereign debt. Market infrastructure such as liquidity
or the presence of capital controls (especially in EMs) could also lead to differences
in home bias across countries. In some countries there may be a seemingly inherent
preference of local investors (including banks) to invest in domestic sovereign debt
due to, for instance, asymmetric information.

 Business cycle considerations could also play a role, as banks might move towards
more domestic sovereign debt due to decline in investment opportunities and for
flight-to-safety reasons during economic downturns.

The increased home bias for European peripheral banks during the crisis reflects a
number of factors. For instance, with foreign investors’ flight from the peripheral sovereign
debt market, peripheral banks absorbed much of the sovereign debt sales of nonresidents as

272
If public debt is increasing but the share of foreign investors or domestic banks’ preferences remain the same,
home bias would not increase per se.
 324

well as new debt issuances.273 The ECB LTROs also provided peripheral banks with cheap
and abundant liquidity to increase their sovereign exposure, which sizably contributed to
banks’ pre-provisioning profits (especially after the Outright Monetary Transactions (OMT)
caused declining sovereign yields) at a time of deteriorating asset quality and supervisory
requirement for continuing NPL provisioning. With credit growth in peripheral countries
significantly declining during the crisis, banks tended to invest in sovereign debt during times
of crisis (flight-to-safety). A higher level of sovereign debt also augmented banks’ eligible
amount of collateral for both ECB and other market funding.

Increasing home bias has raised some concerns regarding its impact on the health of the
banking sector in the European peripheral countries. According to European Banking
Authority (EBA) data as of June 2013 for a cluster of banks, Italian banks’ investments in
domestic sovereign bonds represent 204 percent of their core equity, which is among the
highest for European banks, and similar for Spanish banks with 156 percent. Interestingly,
German banks also have a sizable sovereign exposure to their sovereign relative to their
capitalization (214 percent) but with obviously less sovereign risk given the higher perceived
safety of the German sovereign. The concerns about European banks’ sovereign exposure led
to the mark-to-market (MTM) of all sovereign securities in the EBA recapitalization exercise
in late 2011.

The sovereign exposure of peripheral banks could potentially decline given the ECB LTRO
expiry in 2015. To counteract potential market concerns from a deepened sovereign-bank
nexus and to signal financial strength to rating agencies in the run up to the European Asset
Quality Review (AQR) and stress test, some peripheral banks have started to reduce their
sovereign exposure (e.g., via LTRO repayments). Peripheral sovereign debt markets have
seen a resurgence of interest by foreign investors in the recent period including successful
debt issuances by countries such as Ireland and Portugal. Solvent credit demand could
increase with the economic recovery potentially also helping to reduce peripheral banks’
sovereign debt holdings. Nonetheless, the continued need to finance large deficits implies
that the overall share of peripheral banks’ debt holdings is likely to remain high unless
nonresident investors and other peripheral non-banking institutions substantially increase
their shares.

273
For instance, Acharya and Steffen (2013) find that increased home bias contributed to compressing peripheral
bond spreads after the European crisis. Results by Arslanalp and Poghosyan (2014) also show that foreign
investor outflows have significantly raised bond yields in Italy and Spain.
 325

Figure 61. European Banks’ Domestic Holdings of Sovereign Debt

Domestic holdings of sovereign debt, as

percent of Core Tier 1
250
214 210
204
200 180 178
Source: EBA.
156
Data as of
150
1H2013
115 108
106
97 94
100 88 86
62 57
53 52
45
50 29 23 22

0
DE MT IT BE SI ES LU PT PL CY IE GR HU NL AT NO FR GB SE DK FI

J. Conclusion

The paper empirically examined home bias and the sovereign-bank nexus from a broad
perspective. It attempted to empirically address such pertinent issues such as the link
between home bias and borrowing costs, public debt, fiscal policy reactions as well as debt
distress. To account for possible cross-country and time series heterogeneity, a large sample
of AM and EM countries was chosen for a long sample period between 1999–2012.

The main empirical results indicate that AM and EM countries seem to benefit from a
higher home bias in terms of lower borrowing costs. Furthermore, market sentiments play
a more important role for EMs and the negative relationship between home bias and EM
borrowing costs could become positive during crisis periods. As experienced by previous
sovereign debt crises, EMs tend to be highly sensitive to sudden changes in risk sentiments
which explain the sharp increase in EM borrowing costs in spite of large home bias.

We also find that countries with higher home bias tend to have larger debt levels and to
undertake less fiscal adjustments. In other words, sovereigns with a highly captive
domestic banking system are more likely to exhibit an impaired and delayed fiscal reaction
function and might overly rely on the ability of domestic banks to fund the sovereigns. While
this allows the sovereign to carry higher debt burdens, it also leads to stress at times when
debt reaches dangerously high levels and when the sovereign-bank nexus has become more
entrenched.

The empirical findings are well reflected in the European debt crisis experience. The
experience has shown that especially peripheral countries with a very captive domestic
banking sector have been more reluctant to undertake necessary fiscal reforms. External
market pressures and soaring sovereign yields have forced peripheral countries to start
implementing some of the overdue reforms. Decisive monetary policy by the ECB has
certainly helped to provide a backstop to the peripheral domestic banks.
 326

The empirical analysis could be extended in a number of ways. It is possible to integrate

non-bank holdings of domestic sovereign debt into the home bias coverage. Furthermore, an
extension could aim to better deal with potential concerns on reverse causality, e.g., between
home bias and public debt, by using some case studies. While the regression framework
attempts to control for the standard macro-financial and institutional factors, the set of
explanatory variables could be expanded.
 327

ANNEX 17. Computations of Home Bias Indicators

(1) Banks’ holding of domestic sovereign claims / Banks’ total assets

AMs: Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark,
Finland, France, Germany, Greece, Ireland, Italy, Japan, Korea, Luxembourg, Netherlands,
Portugal, Slovenia, Spain, Sweden, United States.

EMs: Antigua and Barbuda, Argentina, Belize, Brazil, Bulgaria, Chile, Ecuador,
Estonia, Egypt, Hungary, Indonesia, Jamaica, Jordan, Latvia, Lebanon, Lithuania, Malaysia,
Mexico, Pakistan, Peru, Philippines, Poland, Romania, Russia, Saint Kitts and Nevis,
Thailand, Turkey, Ukraine, Venezuela.

Nominator
- Banks’ holding of domestic sovereign claims (A)
(1) IFS “CLAIMS ON GENL GOVT IN CTY”
complemented by
(2) Arslanalp and Tsuda (2012, IMF WP)
“Domestic banks’ holding of domestic sovereign claims”
Six countries: Australia, Canada, New Zealand, Norway, Switzerland,
U.K.

Denominator
- Banks’ total assets = (A) + Banks’ holding of claims on nonresidents (B)
+ Banks’ holding of claims on central bank (C)
+ Banks’ holding of claims on other sectors (D)

- Banks’ holding of claims on nonresidents (B)

IFS “CLAIMS ON NONRESIDENTS”
- Banks’ holding of claims on central bank (C)
IFS “CLAIMS ON CENTRAL BANK”
- Banks’ holding of claims on other sectors (D)
IFS “CLAIMS ON OTHER SECTORS”

(2) Banks’ holding of domestic sovereign claims / Banks’ holding of sovereign claims
AMs: Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark,
Finland, France, Germany, Greece, Hong Kong SAR, Ireland, Israel, Italy, Japan, Korea,
Luxembourg, New Zealand, Netherlands, Norway, Portugal, Singapore, Slovenia, Spain,
Sweden, Switzerland, United Kingdom, United States.

Nominator-(defined above)
Denominator
- Banks’ holding of sovereign claims = (A) + Banks’ holding of other sovereign
claims (E)

- Banks’ holding of other sovereign claims (E) = Foreign assets (F)

* Proxy (ratio of banks’ exposure to foreign public claims) (G)
 328

- Foreign assets
IFS “FOREIGN ASSETS”
- Proxy (Ratio of banks’ exposure to foreign public claims) (G) = Banks’ exposure
to foreign public sector / banks’ exposure to total foreign claims
- Banks’ exposure to foreign public sector
BIS banking sector statistics (Table 9E) “G: Public sector claims”
- Banks’ exposure to total foreign claims
BIS banking sector statistics (Table 9E) “S: Foreign claims”
* available for Belgium, France, Germany, Italy, Japan, Spain, Switzerland,
U.K., U.S. For others, European average or non-European average is used.
** Data is available for the period over 2010–12, 2010 data is used for
period prior 2009.

(3) Banks’ holding of domestic sovereign claims / Public debt of sovereign

(sample available AM & EM)
Nominator - (defined above)
Denominator
- Public debt IMF WEO
 329

ANNEX 18. Details and Sources of Macroeconomic Variables

Variable Description Frequency Source

Dependent variables
Long-term bond yields In percent Annual / IMF’s World Economic
Monthly Outlook (WEO) / IFS
EMBI stripped spreads In percent Annual / Bloomberg
Monthly
Debt-to-GDP In percent Annual WEO database
Primary balance to GDP In percent Annual WEO database
ratio
Explanatory variables
Lagged debt to GDP ratio In percent Annual WEO database
GDP growth rate In percent Annual WEO database
Inflation rates Three year lagged moving average Annual Staff calculations based on
CPI inflation WEO database
Exchange rate depreciation In percent Annual Staff calculations based on
WEO database
Institutional quality index Smaller (larger) values indicating Annual International Country Risk
higher (lower) political risk. Guide (ICRG) dataset.
Credit to GDP ratio In percent Annual IFS database
Capital account openness Higher indices indicating a high Annual Chin and Ito (2006)
degree of capital account openness
VIX Chicago Board Options Exchange Annual / Bloomberg
Market Volatility Index Monthly
U.S. long-term bond yields In percent Annual WEO database
Output gap Difference between actual and Annual Staff calculations based on
potential (calculated using the WEO database
Hodrick-Prescott filter real GDP)
Government expenditure Difference between actual and Annual Staff calculations based on
Gap potential (calculated using the WEO database
Hodrick-Prescott filter real GDP)
Trade openness Sum of exports and imports to Annual Staff calculations based on
GDP (in percent) WEO database
Oil price Log of (trend) oil price applied to Annual Staff calculations based on
oil exporters only. WEO database
 330

ANNEX 19. Outline of the DCC GARCH Method

The Dynamic Conditional Correlation (DCC) specification by Engle (2002) is adopted,

which provides a generalization of the Constant Conditional Correlation (CCC) model by
Bollerslev (1990).274 The DCC framework allows us to analyze the monthly co-movement of
both AM and EM spreads against the VIX, which proxies for global risk sentiment.
Specifically, each of the DCC GARCH models includes the VIX as well as four EM or AM
spreads.

The DCC model is estimated in a three-stage procedure. In general, let rt denote an n x 1

vector of asset returns, exhibiting a mean of zero and the following time-varying covariance:
(A1)

Here, Rt is made up from the time dependent correlations and Dt is defined as a diagonal
matrix comprised of the standard deviations implied by the estimation of univariate GARCH
models, which are computed separately, whereby the ith element is denoted as hit . In other
words in this first stage of the DCC estimation, we fit univariate GARCH models for each of
the five variables in the specification. In the second stage, the intercept parameters are
obtained from the transformed asset returns and finally in the third stage, the coefficients
governing the dynamics of the conditional correlations are estimated. Overall, the DCC
model is characterized by the following set of equations (see Engle, 2002, for details):

(A2)

Here, S is defined as the unconditional correlation matrix of the residuals εt of the asset
returns rt. As defined above, Rt is the time varying correlation matrix and is a function of Qt,
which is the covariance matrix. In the matrix Qt,ι is a vector of ones, A and B are square,
symmetric and  is the Hadamard product. Finally, λi is a weight parameter with the
contributions of Dt21 declining over time, while κ i is the parameter associated with the
squared lagged asset returns. The estimation framework is the same as in Frank, Gonzalez-
Hermosillo and Hesse (2008) or Frank and Hesse (2009).

274
Given the high volatility movements during the recent financial crisis, the assumption of constant conditional
correlation among the variables in the CCC model is not very realistic especially in times of stress where
correlations can rapidly change. Therefore, the DCC model is a better choice since correlations are time-
varying.
 ANNEX 20. Home Bias Regression Tables
Table A20.1. Regression of Bond Spreads—(1) AM
(1) HB HB/Debt (2)Baseline - HB. (3) Omitting Greece, (4) Quartile regression
Dependent variable: Bond Spreads 1/ HB/Debt, HB/VIX Japan, Portugal
IV pooled estimation IV pooled estimation IV pooled estimation Quartile regression

1 (Public debt/GDP, lagged)

0.055***
(0.013)
0.054***
(0.013)
0.038**
(0.010)
-0.0068
(0.016)

1 (Public debt/GDP, square, lagged)

- - - 0.000014
(0.00006)
1.361*** 1.255** 1.226*

0 (Deviation of home bias from median) (0.493) (0.515) (0.620)

-0.027*** -0.027*** -0.026*** -0.013**

1 (Deviation of home bias from median * Public debt/GDP, (0.0065)
(0.009) (0.012) (0.012)
level, lagged)

2 (Deviation of home bias from median * VIX)

- 0.004
(0.012)
0.007
(0.013)
0.0084
(0.021)

1 (GDP growth rate)

-0.045***
(0.012)
-0.045***
(0.016)
-0.024*
(0.014)
-0.139***
(0.052)

2 -0.089 -0.087 0.026 0.178

331
(Inflation rate, 3-year MA) (0.101) (0.101) (0.109) (0.124)

3 (Institutional quality)
-0.046***
(0.012)
-0.045***
(0.012)
-0.028**
(0.011)
-0.118***
(0.022)

4 (Exchange rate depreciation)

-0.009*
(0.005)
-0.009*
(0.005)
-0.008*
(0.006)
-0.019
(0.019)

5 (Credit-to-GDP ratio)
0.009**
(0.004)
0.009**
(0.004)
0.005**
(0.002)
-0.007**
(0.003)

1 (VIX)
0.011*
(0.006)
0.007
(0.016)
0.004
(0.017)
-0.017
(0.031)
Adj. R-squared 0.252 0.253 0.202 -

Sample of years 1999–2012 1999–2012 1999–2012 1999–2012

Sample of observations 313 313 276 313

Root MSE 0.681 0.682 0.636 -

Note: ***, **, * show significance at 1%, 5%, and 10%. Error term assumed to follow an AR(1) process.
1/ Bond spreads is defined as difference between yields of countries’ long-term bonds and those of the U.S. bonds (non-European countries) or those of
German bonds (European countries).
 332

Table A20.2. Regression of Bond Yields—(2) EM

Dependent variable: Local Currency Bond Yields (long-term)
(1) HB, HB/VIX (2) Baseline—HB, HB/debt,
HB/VIX
IV pooled IV pooled
0.046*** 0.050***
1 (Public debt/GDP, lagged) (0.010) (0.015)
-0.223** -0.202**
0 (Deviation of Home bias from median) (0.088) (0.088)
1 (Deviation of Home bias from median *
- -0.0007
(0.002)
Public debt/GDP, level, lagged)
2 (Deviation of Home bias from median *
0.012***
(0.0035)
0.012***
(0.0036)
VIX)
-0.258*** -0.264***
1 (GDP growth rate) (0.063) (0.066)
0.201* 0.202*
2 (Inflation rate, 3-year MA) (0.102) (0.105)
-0.0012 -0.002
3 (Exchange rate depreciation) (0.021) (0.021)
0.054 0.068
4 (Capital account openness) (0.214) (0.217)
0.0028 0.0028
5 (Credit-to-GDP ratio) (0.015) (0.015)
0.057** 0.055**
1 (VIX) (0.026) (0.026)
0.977*** 0.960***
2 (U.S. long-term bonds) (0.313) (0.313)
Adj. R-squared 0.747 0.742
Sample periods 1999–2012 1999–2012
Sample of observations 113 113
Root MSE 1.574 1.578

Note: ***, **, * show significance at 1%, 5%, and 10%. Error term assumed to follow an AR(1) process.
 Table A20.3. Regression of the Public Debt/GDP—AM, EM, AM&EM
(A) AM (B) EM (C) AM & EM
(1) IV pooled (2) Least (1) IV (2) Least (1) IV pooled
regression Square—pooled pooled Square—pooled regression (2) Least
regression estimation regression Square—pooled
regression
Dependent variable: Public debt to
GDP ratio
2.307** - 1.104*** - 0.420 -
1 (Home bias) (1.127) (0.392) (0.518)
3.753*** 1.919***
(0.520) (0.157)
2.529***

1 (Home bias, lagged) (0.242)

-1.742*** -2.342*** -0.913*** -0.863*** -1.356*** -0.907***
1 (Output gap) (0.669) (0.666) (0.320) (0.318) (0.458) (0.307)
-1.095 -0.414 -0.132 0.043 -0.427 0.188
2 (Government expenditure gap) (1.146) (1.166) (0.165) (0.120) (0.264) (0.160)
-0.074* -0.137*** -0.082 0.287*** -0.131*** -0.052***
3 (Trade openness) (0.045) (0.047) (0.070) (0.064) (0.041) (0.032)
-4.318** -10.604*** 0.093*** 0.032*** 0.172*** 0.087***
4 (Inflation rate, 3-year MA)

333
(1.901) (1.870) (0.028) (0.023) (0.041) (0.032)
- - -5.186*** -1.655** -4.502*** -1.018
5 (Oil price, lagged) (1.145) (0.774) (1.414) (0.782)
18.223*** 13.179*** -4.301*** 4.430*** -4.502*** 7.821***
6 (Capital account restriction) (3.036) (2.618) (1.073) (0.977) (1.414) (1.251)
0.087 0.197*** -0.111 -0.207** 4.712** 0.181***
7 (Credit-to-GDP ratio) (0.068) (0.042) (0.126) (0.092) (2.051) (0.049)
-1.243** -0.857** -0.225 -0.567 -1.094*** -0.865***
8 (Institutional quality) (0.484) (0.406) (0.350) (0.774) (0.393) (0.279)
118.25** 96.480*** 56.22 44.079*** 121.442*** 63.275***
Constant (46.961) (34.614) (36.016) (27.290) (34.916) (20.937)
- - - - 37.076*** 23.936***
Dummy variable for AM (7.891) (6.174)
Sample period 1999–2012 1999–2012 1999–2012 1999–2012 1999–2012 1999–2012
Adj. R-squared 0.420 0.635 0.879 0.588 0.387 0.493
Sample of observations 202 207 70 184 272 391
Sample of countries 19 20 9 18 28 38
Root MSE 21.654 24.392 11.084 21.892 23.68 27.388
Wald chi-squared 210.73 - 2227.36 - 352.74 -
Note: ***, **, * show significance at 1%, 5%, and 10%. Estimation including capital adequacy ratio is also examined, but due to availability of data on capital adequacy ratio, sample
period and observation are limited to 2008–10 and 75. Thus, we do not report the results in the table.
 Table A20.4. Regression of the Fiscal Reaction Function—AM&EM
Dependent variable: Primary balance to GDP (1) HB indicator— (2) HB indicator— (3) HB indicator— (4) HB indicator— (5) HB indicator—
ratio constant constant/linear linear interactive quadratic interactive cubic interactive

1 (Public debt/GDP, lagged)

-0.143*
(0.084)
-0.130
(0.083)
-0.150*
(0.085)
-0.151*
(0.086)
-0.148*
(0.085)

2 (Public debt/GDP, square, lagged)

0.00166**
(0.0007)
0.0014*
(0.00075)
0.00179**
(0.00074)
0.00181**
(0.00074)
0.0018**
(0.00072)

3 (Public debt/GDP, cubic, lagged)

-0.0000048**
(0.0000020)
-0.0000044**
(0.0000020)
-0.0000052***
(0.0000020)
-0.0000051***
(0.0000020)
-0.0000047**
(0.0000019)

0 (Deviation of HB from median)

-0.096**
(0.050)
-0.169*
(0.092)
- - -

1 (Deviation of HB from median * Public

- 0.00096
(0.00099)
-0.00024
(0.00046)
- -

debt/GDP, lagged)

2 (Deviation of HB from median * Public

- - - -0.0000023
(0.0000031)
-

debt/GDP, square, lagged)

3 (Deviation of HB from median * Public
- - - - -0.000000022
(0.000000019)
debt/GDP, cubic, lagged)

1 (Output gap)
0.151***
(0.050)
0.140***
(0.050)
0.155***
(0.052)
0.157***
(0.052)
0.158***
(0.052)

334
2 (Government expenditure gap)
-0.065**
(0.028)
-0.060**
(0.027)
-0.062**
(0.029)
-0.062**
(0.028)
-0.062***
(0.028)

3 (Trade openness)
0.070***
(0.020)
0.072***
(0.020)
0.071***
(0.020)
0.070***
(0.020)
0.070***
(0.020)

4 (Inflation rate, lagged)

-5.869
(7.790)
-7.372
(7.793)
-7.124
(7.881)
-7.013
(7.800)
-6.879
(7.730)

5 (Oil price, lagged)

-12.734***
(3.887)
-13.118***
(3.834)
-12.491***
(4.002)
-12.404***
(4.020)
-12.305***
(4.018)

6 (Capital account openness)

0.060
(0.346)
-0.009
(0.348)
0.105
(0.344)
0.124
(0.347)
0.148
(0.348)

7 (Credit-to-GDP ratio)
-0.067***
(0.013)
-0.067***
(0.013)
-0.068***
(0.013)
-0.068***
(0.013)
-0.068***
(0.013)
Adj. R-squared 0.424 0.434 0.418 0.417 0.417

Sample of years 1999–2012 1999–2012 1999–2012 1999–2012 1999–2012

Sample of observations 453 453 453 453 453

Sample of countries 45 45 45 45 45

Root MSE 2.423 2.417 2.433 2.432 2.423

Transformed DW 1.670 1.664 1.680 1.682 1.685

Note: ***, **, * show significance at 1%, 5%, and 10%. Country-specific fixed effect included, and error term assumed to follow an AR(1) process. All
specifications are regressed by two-step GMM fixed effects estimation
 Table A20.5. Robustness Check for Bond Spreads Regression—AM, EM
(A) Banks’ holding of domestic (B) Banks’ holding of domestic sovereign claims /
sovereign claims / Total sovereign Public Debt–AM & EM
Dependent variable: claims—AM
Bond spreads-AM Bond spreads-AM Bond yields-EM
0.032*** 0.052*** 0.049***
1 (Public debt/GDP, lagged) (0.012) (0.012) (0.014)
- 0.657** -0.201**
0 (Deviation of Home bias from median) (0.274) (0.087)
1 (Deviation of Home bias from median *
-0.0012**
(0.00047)
-0.014***
(0.005)
-0.001
(0.016)
Public debt/GDP, level, lagged)
2 (Deviation of Home bias from median *
0.0018**
(0.00086)
0.0019
(0.006)
0.012***
(0.0035)
VIX)
-0.049*** -0.045*** -0.264***
1 (GDP growth rate) (0.016) (0.013) (0.066)
-0.059 -0.086 0.201*
2 (Inflation rate, 3-year MA) (0.095) (0.101) (0.105)
-0.0085 -0.009* -0.002
3

335
(Exchange rate depreciation) (0.0053) (0.005) (0.021)
- - 0.068
4 (Capital account openness) (0.217)
0.009** 0.009** 0.0026
5 (Credit-to-GDP ratio) (0.004) (0.004) (0.015)
0.012* 0.007 0.059**
1 (VIX) (0.007) (0.015) (0.026)
- - 0.952***
2 (U.S. long-term bonds) (0.312)
Adj. R-squared 0.228 0.252 0.744
Sample periods 1999–2012 1999–2012 1999–2012
Sample of observations 313 313 113
Sample of countries 25 25 16
Root MSE 0.689 0.682 1.576
Note: ***, **, * show significance at 1%, 5%, and 10%. Error term assumed to follow an AR(1) process.
 Table A20.6. Robustness Check for Public Debt Regression
(A) Banks’ holding of domestic sovereign claims / (B) Banks’ holding of domestic sovereign
Total sovereign claims—AM claims / Public Debt—AM & EM
Dependent variable: Public debt to GDP (1) IV pooled regression (2) Least Square— (1) IV pooled regression
ratio pooled regression

1 (Home bias)
1.573*
(0.902)
2.075*
(1.224)

1 (Home bias, lagged)

- 0.148**
(0.074)
-

1 (Output gap)
-0.846
(1.482)
-1.569***
(0.185)
-0.926
(1.191)

2 (Government expenditure gap)

-0.383
(01.443)
-0.182
(0.216)
0.182
(0.419)

3 (Trade openness)
0.096*
(0.056)
0.045
(0.033)
-0.184**
(0.085)

4 (Inflation rate, 3-year MA)

-10.951
(10.192)
-2.860**
(1.278)
0.163***
(3.884)

5 (Oil price, lagged)

- - 0.270
(3.884)

6 (Capital account openness)

27.191***
(4.355)
-6.780
(3.982)
5.562**
(2.460)

336
7 (Credit-to-GDP ratio)
0.333
(0.249)
-0.019
(0.049)
0.00015
(0.092)

8 (Institutional capacity/political risk)

-1.441*
(0.786)
-0.347*
(0.198)
0.231
(0.343)
- 52.8** -
Constant (19.682)
Dummy variable for AM - - -

Sample period 2005–12 2005-2012 2005–12

Adj. R-squared - 0.490 -

Sample of observations 172 127 334

Sample of countries 26 26 39
Root MSE 33.885 4.682 41.454

Note: ***, **, * show sgnificance at 1%, 5%, and 10%. Estimation including capital adequacy ratio is also examined, but due to availability
of data on capital adequacy ratio, sample period and observation are limited to 2008–10 and 75. Thus, we do not report the results in the
table.
 337

Table A20.7. Robustness Check for Fiscal Reaction Function—AM

(A) Banks’ holding of domestic
Dependent variable: Primary balance to GDP sovereign claims / Total sovereign
ratio claims - AM

1 (Public debt/GDP, lagged)

-0.270*
(0.108)

2 (Public debt/GDP, square, lagged)

0.0033***
(0.00070)

3 (Public debt/GDP, cubic, lagged)

-0.0000094**
(0.0000031)

0 (Deviation of HB from median)

-0.117*
(0.067)

1 (Deviation of HB from median * Public

0.0019**
(0.00094)
debt/GDP, lagged)

1 (Output gap)
0.811***
(0.102)

2 (Government expenditure gap)

-0.450***
(0.097)

3 (Trade openness)
0.048**
(0.020)

4 (Inflation rate, lagged)

0.978
(0.350)

5 (Oil price, lagged)

19.756***
(19.756)

6 (Capital account openness)

-1.141
(0.1.422)

7 (Credit-to-GDP ratio)
-0.041
(0.015)
Adj. R-squared 0.851

Sample of years 2005–12

Sample of observations 181

Sample of countries 28

Root MSE 2.060

Transformed DW 1.691

Note: ***, **, * show significance at 1%, 5%, and 10%. Country-specific fixed effect
included, and error term assumed to follow an AR(1) process.
 References

Acharya, V. V., Drechsler, I., and P. Schnabl, 2012, “A Tale of Two Overhangs: The Nexus
of Financial Sector and Sovereign Credit Risks,” Banque de France Financial
Stability Review, No.16, April 2012.

Acharya, V. V., and S., Steffen, 2013, “The Greatest Carry Trade Ever? Understanding
Eurozone Bank Risks,” NBER Working Paper No. 19039, May 2013.

Andritzky, J. R., 2012, “Government Bonds and Their Investors: What Are the Facts and Do
They Matter?” IMF Working Paper 12/158, (Washington).

Ardagna, S., Caselli, F., and T. Lane, 2007, “Fiscal Discipline and the Cost of Public Debt
Service: Some Estimates for OECD Countries,” The B.E. Journal of
Macroeconomics, Vol. 7(1); pp. 1–35.

Arslanalp, S., and T., Poghosyan, 2014, “Foreign Investor Flows and Sovereign Bond Yields
in Advanced Economies,” IMF Working Paper 14/27, (Washington).

Arslanalps, S., and T. Tsuda, 2012, “Tracking Global Demand for Advanced Economy
Sovereign Debt,” IMF Working Paper 12/284, (Washington).

Asonuma, Tamon, Said Bakhache, and Heiko Hesse, 2015, “Is Banks’ Home Bias Good or
Bad for Debt Sustainability?“ IMF Working Paper 2015/44. A shorter version was
published in VOX.

Baldacci, E., Petrova, I., Belhocine, N., Dobrescu G., and S., Mazraani, 2011, “Assessing
Fiscal Stress,” IMF Working Paper 11/100 (Washington).

Bank of International Settlement, 2011, “The Impact of Sovereign Credit Risk on Bank
Funding Condition,” CGFS Paper, No. 43.

Battistini, N., Pagano, M., and S. Simonelli, 2013, “Systemic Risk and Home Bias in the
Euro Area,” European Commission Economic Papers, No. 494.

Bollerslev, T., 1990, “Modelling the Coherence in Short-run Nominal Exchange Rates: A
Multivariate Generalized ARCH Approach,” Review of Economics and Statistics,
Vol. 72; pp. 498–505.

Boz, E., D’Erasmo, P., and B. Durdu, 2014, “Sovereign Risk and Bank Balance Sheets: The
Role of Macroprudential Policies,” manuscript, (Washington: International Monetary
Fund).
 339

Broner, F., Erce, A., Martin, A., and J. Ventura, 2014, “Sovereign Debt Markets in Turbulent
Times: Creditor Discrimination and Crowding-out Effects,” Journal of Monetary
Economics, Vol. 61; pp. 114–42.

Chan, K., Covrig, V., and L. Ng, 2005, “What Determines the Domestic Bias and the Foreign
Bias?” Evidence from Mutual Fund Equity Allocations Worldwide,” Journal of
Finance, Vol. 60(3); pp.1495–534.

Cruces, J., and C. Trebesch, 2013, “Sovereign Defaults: The Price of Haircuts,” American
Economic Journal: Macroeconomics, Vol. 5(3), pp. 85–117.

Ebeke, C., and Y. Lu, 2014, “Emerging Market Local Currency Bond Yields and Foreign
Holdings in the Post-Lehman Period—A Fortune or Misfortune?” IMF Working
Paper 14/29 (Washington).

Engle, R., 2002, “Dynamic Conditional Correlation: A Simple Class of Multivariate

Generalized Autoregressive Conditional Heteroskedasticity Models,” Journal of
Business & Economic Statistics, Vol. 20; pp. 339–50.

Fidola, M., Fratzcher, M., and C. Thimann, 2006, “Home Bias in Global Bond and Equity
Markets,” ECB Working Paper No. 685.

Frank, N., B. González-Hermosillo, and H. Hesse, 2008, “Transmission of Liquidity Shocks:

Evidence from the 2007 Subprime Crisis,” IMF Working Paper 08/200, see also blog
piece on VOX (Washington: International Monetary Fund).

Frank, N., and H. Hesse, 2009, “Financial Spillovers to Emerging Markets during the Global
Financial Crisis,” IMF Working Paper 09/104 (Washington).

Gennaioli, N., Martin, A., and S. Rossi, 2014a, “Banks, Government Bonds, and Default:
What do the Data Say?” IMF Working Paper, 14/120 (Washington).

Gennaioli, N., Martin, A., and S. Rossi, 2014b, “Sovereign Default, Domestic Banks, and
Financial Institutions,” Journal of Finance, Vol. 69 (2); pp.819–866.

Ghosh, A. R., Kim, J. I., Mendoza, E. G., Ostry, J. D., and M. Qureshi, 2011, “Fiscal Fatigue,
Fiscal Space and Debt Sustainability in Advanced Economies,” NBER Working
Paper No. 16782.

International Monetary Fund, 2014 “Spain: Financial Sector Reform—Final Progress

Report,” February (Washington).

———, 2012, “Global Financial Stability Report, April (Washington).

———, 2011, “Global Financial Stability Report, September (Washington).

 340

Jaramillo, L., and S. Zhang, 2013, “Real Money Investors and Sovereign Bond Yields,” IMF
Working Paper 13/254 (Washington).

Mauro, P., Romeu, R., Binder, A., and A. Zaman, 2013 “A Modern History of Fiscal
Prudence and Profligacy,” IMF Working Paper 13/5 (Washington).

Merler, S., and J., Pisani-Ferry, 2012, “Hazardous Tango: Sovereign-Bank Interdependence
and Financial Stability in the Euro area,” Banque de France Financial Stability
Review, No. 16, April 2012.

Portes, R., Rey, H., and Y. Oh, 2001, “Information and Capital Flows: The Determinants of
Transactions in Financial Assets,” European Economic Review, Vol. 45; pp. 783–96.

Warnock, F. E., and V. C. Warnock, 2009, “International Capital Flows and U.S. Interest
Rates,” Journal of International Money and Finance, Vol. 28; pp. 903–19.