ESSAYS ON THE GLOBAL FINANCIAL CRISIS

Edited by Heiko Hesse

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

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 ith . 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 2

1tD 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

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.

Correlation (ABCP, Libor)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (ABCP, CDS)

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Libor, CDS)

-0.15

-0.05

0.05

0.15

0.25

0.35

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Returns, CDS)

-0.35

-0.25

-0.15

-0.05

0.05

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

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)

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (ABCP, Returns)

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (ABCP, CDS)

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (ABCP, Two)

-0.15

-0.05

0.05

0.15

0.25

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)

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

Correlation (Libor, Returns)

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Libor, CDS)

-0.2

-0.1

0

0.1

0.2

0.3

0.4

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Libor, Two)

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Returns, CDS)

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (Returns, Two)

-0.4

-0.3

-0.2

-0.1

0

0.1

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

Correlation (CDS, Two)

-0.15

-0.05

0.05

0.15

0.25

03/01/2003 03/01/2004 03/01/2005 03/01/2006 03/01/2007 03/01/2008

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

Asset-backed commercial

paper spread1

LIBOR spread3

Credit default swap2

0

50

100

150

200

250

300

350

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

0

50

100

150

200

250

300

350

Aggregate Bank Credit Default Swap Rate and Selected Spreads(In basis points)

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

1Spread between yields on 90-day U.S. asset-backed commercial paper and on three-month U.S. Treasury bill.

2The 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.

3Spread between yields on three-month U.S. LIBOR and on three-month U.S. overnight index swap.

-0.04

-0.02

0.00

0.02

0.04

0.06

0.08

0.10

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

-0.04

-0.02

0.00

0.02

0.04

0.06

0.08

0.10

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

Source: Bloomberg L.P.1Spread between yields on five-year off-the-run bond and on five-year on-the-run bond.

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

Figure A1.2. On-the-Run/Off-the-Run Five-Year U.S. Treasury Bond Spread1

32

-60

-40

-20

0

20

40

60

80

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

-60

-40

-20

0

20

40

60

80

Asset-backed commercial paper spread 1/

LIBOR spread 2/

United States: Selected Spreads(First difference; in basis points)

Sources: Bloomberg L.P.; and IMF staff estimates.1Spread between yields on 90-day U.S. asset-backed commercial paper and on three-month U.S. treasury bill.2Spread between yields on three-month U.S. LIBOR and on three-month U.S. overnight index swap.

-15

-10

-5

0

5

10

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

-15

-10

-5

0

5

10

15

Stock market returns

CDS

United States: S&P 500 Stock Market Returns and Credit Default Swap (CDS)1

(First difference; in basis points)

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

1CDS 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.

Figure A1.3. United States: Selected Spreads (First difference; in basis points)

Figure A1.4. United States: S&P 500 Stock Market Returns and Credit Default Swap (CDS)1

(First difference; in basis points)

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

ttUSt

EUt

EUt Icybybay ,11112111

ttUSt

EUt

USt Icybybay ,21214132 (4)

where ),( US

tEUtt yyy captures changes in the US and Euro 3-month Libor spreads, 1tI

denotes the intervention of the ECB or Fed (e.g. TAF or LTRO) and ),( ,2,1 ttt are the

residuals with tttt HN ,0~1 . The conditional covariance matrix tH is given by

1111 ttttt EIBBHAAMH (5)

where M is a scalar and A, B as well as E are diagonal 22 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 ith . 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 2

1tD 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 ty being a vector of observed variables and ts denoting a unobserved

random variable with values 1, 2, …, K that as a state variable governs the conditional distribution of ty .

Prob ,...),,...,,|( 2121 ttttt yyksisjs Prob ijtt pisjs )|( 1 (1)

It is possible to combine all the transition probabilities ijp in a KK 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

ttt yy 1 (2)

The time varying variance 2

th with the error term t is parameterized as

,~~...~~

~

~

211

2222

2110

2ttqtqttt

ttt

tSt

daaaah

h

gt

(3)

where 1,0~ Nt , 3,2,1tS and 1td is a dummy variable in which 11 td if 0~

1 t and

01 td if .0~1 t 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

90

thus state dependent due to multiplication with the scaling factor tSg 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

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.

0.0

0.2

0.4

0.6

0.8

1.0

Jan-06 May-06 Sep-06 Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08

-150

-100

-50

0

50

100

150

Probability of being in low-volatility state (left scale)

Probability of being in medium-volatility state (leftscale)Probability of being in high-volatility state (left scale)

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

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

Beginning of subprime crisis

Lehman'sfailure

92

Figure 14. Markov-Switching ARCH Model of VIX

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.

-20

-15

-10

-5

0

5

10

15

20

25

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Absolute change in VIX (left scale) Probability of being in high-volatility state (right scale)

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

Russian's default and LTCM

9-11

WorldCom and Brazil's election

Dot-com bubble burst

Turkey crisis Shanghai crash

Bear Stearns

Beginning of subprime crisis

Lehman

93

Figure 15. Markov-Switching ARCH Model of TED Spread

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.

-100

-80

-60

-40

-20

0

20

40

60

80

100

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Change in TED spread (basis points, left scale) Probability of being in high-volatility state (right scale)

Sources: Bloomberg L.P.; and IMF staff estimates.Note: [1] = Russian default and LTCM crisis; [2] = Lehman Brothers failure.

[1]

9-11

Beginning of subprime

crisis

[2]

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).

Figure 16. Euro-Dollar Forex Swap

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.

g p(Probability of being in low-, medium-, and high-volatility state)

0.0

0.2

0.4

0.6

0.8

1.0

Jan-07 May-07 Sep-07 Jan-08 May-08 Sep-08 Jan-09 May-09

Low Medium High

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

Beginning of subprime crisis

Lehman's failure

95

Figure 17a. Markov-Switching ARCH Model of VIX

Figure 17b. Markov-Switching ARCH Model of VIX

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

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Low Medium High

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.

Russian's default and LTCM

9-11

WorldCom and Brazil's election

Dot-com bubble burst

Bear Stearns

Beginning of subprime crisis

Lehman

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

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Jan-09 Feb-09 Mar-09 Apr-09 May-09 Jun-09 Jul-09

Low Medium High

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.

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

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.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Low Medium High

g g p(Probability of being in low-, medium-, and high-volatility state)

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.

Russian's default and LTCM crisis

9-11

Beginning of

subprime crisis

Lehman

97

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

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.

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Jan-09 Feb-09 Mar-09 Apr-09 May-09 Jun-09 Jul-09

Low Medium High

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

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.

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

108

Table 5. Acquiring SWFs

Figure 19. Ratios of SWF Investments and Divestments

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.

SWFNumber of

Observations Country

Abu Dhabi Investment Authority (ADIA) 26 United Arab EmiratesChina Investment Corporation (CIC) 11 ChinaGovernment of Singapore Investment Corporation (GIC) 38 SingaporeKuwait Investment Authority (KIA) 14 KuwaitKorea Investment Corporation (KIC) 1 KoreaLibyan Investment Authority (LIA) 2 LibyaMubadala 2 United Arab EmiratesQatar Investment Authority (QIA) 23 QatarTemasek 49 SingaporeTotal 166

0

10

20

30

40

50

60

70

80

90

Buy

Sel

l

Buy

and

Sel

l in

Fina

ncia

l

Buy

and

Sel

l in

Non

-fi

nanc

ial

Buy

and

Sel

l in

Dev

elop

ed

Buy

and

Sel

l in

Em

ergi

ng

Buy

in d

evel

oped

econ

omie

s

Sell

in d

evel

oped

econ

omie

s

Buy

in e

mer

ging

econ

omie

s

Sel

l in

emer

ging

econ

omie

s

Buy

in f

inan

cial

sec

tor

Sell

in f

inan

cial

sec

tor

Buy

in n

on-f

inan

cial

sect

or

Sell

in n

on-f

inan

cial

sect

or

Buy

by

high

leve

l

Sel

l by

high

leve

l

Buy

by

low

leve

l

Sell

by

low

leve

l

Source: IMF staff estimates.

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

AARt=1

N

1

N

iti

AR .

The average abnormal returns can then be aggregated over the event window to calculate the cumulative average abnormal return (CAAR) for each firm i.

CAARt =

2

2ttAAR

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)

Event Window Test Statistic of AAR Mean of AARTest Statistic of

CAAR Mean of CAAR

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

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

(-3,+3) 2.88** 0.18 6.17*** 0.94(-2,+2) 4.29** 0.21 4.95** 0.72

(-3,+3) 0.98 0.11 1.14 0.20(-2,+2) 1.20 0.17 1.67 0.34

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

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

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

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

(-3,+3) 3.09** 0.66 6.38*** 3.40(-2,+2) 2.72* 0.70 5.13** 2.53

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

(-3,+3) -0.15 -0.01 -4.06** -0.35(-2,+2) 0.31 0.04 -1.78 -0.18

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

(-3,+3) 0.18 0.02 0.84 0.07(-2,+2) -0.11 -0.02 -0.20 -0.02

(-3,+3) 0.27 0.06 1.04 0.15(-2,+2) 0.36 0.12 0.85 0.17

(-3,+3) 2.21* 0.50 4.05** 2.22(-2,+2) 2.68* 0.68 3.03** 1.89

(-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)".

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

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

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

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

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

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

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

Panel A: Buy only, 134 events from 101 firms

Panel B: Sell only, 32 events from 23 firms

Panel C Buy and Sell in developed economies only, 87 events from 55 firms

Panel D: Buy and Sell in emerging economies only, 79 events from 60 firms

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

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

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

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

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

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

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

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

(-3,+3) 2.83** 0.20 6.37*** 1.10(-2,+2) 5.05** 0.25 5.45** 0.87

(-3,+3) 0.94 0.10 1.33 0.20(-2,+2) 0.98 0.14 1.58 0.28

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

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

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

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

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

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

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

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

(-3,+3) 0.11 0.02 0.68 0.06(-2,+2) -0.22 -0.04 -0.50 -0.07

(-3,+3) 0.27 0.06 0.91 0.14(-2,+2) 0.38 0.12 0.92 0.18

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

(-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)".

Panel A: Buy only, 134 events from 101 firms

Panel B: Sell only, 32 events from 23 firms

Panel C Buy and Sell in developed economies only, 87 events from 55 firms

Panel D: Buy and Sell in emerging economies only, 79 events from 60 firms

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

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

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

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

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

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

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

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

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

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

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

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

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)

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

‐40

-20

0

20

40

60

80

100

120M

onte

negr

o, R

ep. o

f

Ukr

aine

Rom

ani

a

Bel

aru

s

Rus

sian

Fed

era

tion

Turk

ey

Serb

ia, R

epub

lic o

f

Cro

atia

Vene

zuel

a, R

ep.

Bo

l.

Par

agua

yB

razi

l

Col

ombi

a

Cos

ta R

ica

Uru

gua

y

Arg

enti

na

Jam

aica

Peru

Me

xico

Dom

inic

an R

epu

blic

Ecu

ador

Chi

leB

oliv

ia

Mal

awi

Ang

ola

Con

go,

Re

publ

ic o

f

Gha

naZa

mbi

a

Moz

am

biq

ue

Uga

nda

Gab

on

Tanz

ania

Ethi

opia

Ken

ya

Ce

ntra

l Afr

ican

Rep

.

Togo

Cô

te d

'Ivoi

reSo

uth

Afr

ica

Sene

gal

Mal

i

Cam

ero

on

Kaza

khst

anIr

aq

Aze

rbai

jan,

Rep

. of

Geo

rgia

Arm

enia

Uni

ted

Ara

b Em

irat

esQ

ata

r

Om

an

Liby

a

Iran

, I.R

. of

Syri

an A

rab

Rep

ubl

icJo

rda

n

Yem

en,

Re

publ

ic o

f

Sau

di A

rabi

a

Mor

occo

Alg

eri

a

Tun

isia

Egyp

t

Ca

mbo

dia

Afg

hani

stan

, I.R

. of

Lao

Pe

ople

's D

em

.Rep

Mon

golia

Chi

na,P

.R.:M

acao

Mya

nmar

Indo

nesi

a

Indi

a

Sri L

anka

Ba

ngla

desh

Paki

stan

Mal

aysi

a

Tha

iland

Phi

lipp

ines

Central and Eastern Europe

Western HemisphereSub-Saharan Africa

Middle-East and North Africa Emerging Asia

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.

123

Figure 21a. Credit Boom Events in the Last Expansion Difference from each country’s historical trend, measured as multiple of standard deviation of

credit/GDP Color code: red bar if the credit/GDP ratio increased by over 5 percentage points per year

0

0.5

1

1.5

2

2.5

3

3.5

4

4.5

5

Turk

ey, 2

00

6Es

ton

ia, 2

00

7M

aced

on

ia,

FYR

, 2

00

8C

roat

ia, 2

00

6

St. L

uci

a, 2

00

7B

arb

ado

s, 2

00

5C

ost

a R

ica,

20

08

Ven

ezu

ela,

Rep

. B

ol.,

20

07

Ch

ile, 2

00

8H

on

du

ras,

20

07

Bel

ize,

20

08

Bah

amas

, 20

08

Jam

aica

, 2

00

8

Zim

bab

we,

20

06

Nig

eria

, 20

08

Uga

nd

a, 2

00

8Su

dan

, 20

06

Sier

ra L

eon

e, 2

00

8M

alaw

i, 2

00

8M

auri

tiu

s, 2

00

8A

ngo

la, 2

00

8Se

ych

elle

s, 2

00

8N

iger

, 20

08

Togo

, 20

07

São

To

mé

& P

rín

cip

e, 2

00

5Li

ber

ia,

20

08

Zam

bia

, 20

08

Co

ngo

, Dem

. R

ep.

of,

20

08

Swaz

ilan

d, 2

00

5G

uin

ea-B

issa

u ,

20

08

Sen

egal

, 2

00

5B

enin

, 20

05

U.A

.E.,

20

08

Sau

di A

rab

ia,

20

08

Syri

a, 2

00

5Su

dan

, 20

06

Iran

, 20

07

Jord

an, 2

00

6Q

atar

, 20

08

Mo

rocc

o, 2

00

8

Van

uat

u, 2

00

8N

epal

, 20

08

Mal

div

es, 2

00

5B

hu

tan

, 20

08

Cam

bo

dia

, 20

07

Ind

ia, 2

00

6So

lom

on

Isla

nd

s, 2

00

7P

apu

a N

ew G

uin

ea,

20

08

Fiji,

20

06

Pak

ista

n, 2

00

5

Central and Eastern Europe

Western Hemisphere

Sub-Saharan AfricaMiddle-East and North Africa

Emerging Asia

124

Figure 21b. Regional Average Differences from Trend

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-

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

MEN

A

WH

M

CEE

SSA

Dev

elo

pin

g A

sia

Oil

Exp

ort

ers

Note: Regional avearages are unweighted averages.

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)

Figure 23. Boom Frequency over Time

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.

0%

1%

2%

3%

4%

5%

6%

7%

8%

9%

MENA WHM CEE SSA Developing Asia

Oil Exporters

0

0.05

0.1

0.15

0.2

0.25

0.3

1989

1990

1991

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

2006

2007

2008

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

MENA

WHM

CEE

Developing Asia

Oil exporters

126

Figure 24. MENA:

Credit Behavior Surrounding Booms (Median Country by Region)

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:

t t t t t tCPS D K NFPS CB RW ( 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).

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

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

Re

al c

red

itgr

ow

th (

year

-on

-ye

ar)

Western Hemisphere

Central and Eastern Europe

Sub-Saharan Africa

Developing Asia

Middle-East and North Africa

Quarters

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:

1 1 1 1 1 1

t t t t t t

t t t t t t

CPS D K NFPS CB RW

CPS CPS CPS CPS CPS CPS

( 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

-20

-15

-10

-5

0

5

10

15

UAE JOR

Position with nonfinancial public sector

Position with Central bank

Position with rest of the world

Deposits

Capital and others

(changes in 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 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 four components is equal to the change in the credit growth rate.

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 down foreign assets and by borrowing abroad, with an overall effect of 22 percentage points.

Two contrasting country examples highlight the differences in how the credit slowdowns were reflected in the aggregate balance sheet of the banking sector, as shown in Figure 25. First, in the United Arab Emirates, funding declined sharply. The slowdown in deposits and capital alone would have led real credit growth to decline by more than 16 percentage points. Reinforcing this was a decline in external borrowing, accounting for an additional 16 percentage points. However, a combination of a fall in bank reserves (12 percentage points), and an increase in government deposits (0.5 percentage points), dampened the 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

Figure 25. Decomposition of the Credit Slowdown in Selected MENA Countries

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

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.

-20

0

20

40

60

80

100

120

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

June 2008

Latest date

Change

(percent)

Source: International Financial Statistics and authors' calculations.

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:

it i it jt jt i jC FUND MACRO INF DumBank DumCountry ( 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

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.

-1

0

1

2

3

4

5

Drivers of Lending Growth in MENA Banks(percent)

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.

Bank-specific drivers Macroeconomicdrivers

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

Expansion Slowdown Change Expansion Slowdown Difference

DepositsCapital and

othersCentral bank

Nonfinancial

public sector

Rest of the

world

Countries with credit slowdowns

Oil exportersGulf 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

Qatar1

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 ImportersEgypt -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

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

Average rate of growth of depositsAverage real annual credit growth

Funding Banks' positions with:

Contribution (percentage points)

Decomposition of the change in credit growth

Source: InternationalFinancial 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.

134134

135

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

(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

Annual Data: 1997-2008

This table reports the results of fixed effects and random effects regressions for annual growth of bank loans in eleven MENAcountries 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

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

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

0

20

40

60

80

100

120

140

160

Mar

-03

Jun-

03Se

p-03

Dec

-03

Mar

-04

Jun-

04Se

p-04

Dec

-04

Mar

-05

Jun-

05Se

p-05

Dec

-05

Mar

-06

Jun-

06Se

p-06

Dec

-06

Mar

-07

Jun-

07Se

p-07

Dec

-07

Mar

-08

Jun-

08Se

p-08

Dec

-08

Mar

-09

Jun-

09Se

p-09

Dec

-09

Mar

-10

Jun-

10Se

p-10

Dec

-10

Mar

-11

Jun-

11Se

p-11

Dec

-11

Mar

-12

Africa & Middle East Asia & PacificCESEE Latin America/Caribbean

External Positions of BIS-reporting Banks vis-à-vis Emerging Market Regions(Sept. 2008 = 100, exchange-rate adjusted)

Sources: BIS; and IMF staff calculations.

0

100

200

300

400

500

600

700

800

900

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

BulgariaCDS HungaryCDS PolandCDS RomaniaCDS

0

100

200

300

400

500

600

700

800

900

1000

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

EMEuropeEMBIG HungaryEMBIG

PolandEMBIG RomaniaEMBIG

CDS and EMBIG Developments

Source: Bloomberg. Source: Bloomberg.

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

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

30

30.5

31

31.5

32

32.5

33

33.5

EBCI Parent Funding Exposure to Romania (Bn €)

Source: EBCI.

17.0

17.5

18.0

18.5

19.0

19.5

20.0

20.5

Bank System Parent Funding (Bn €)

Source: NBR.

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

up to 1 M 1-3 M 3-6 M 6-12 M 1Y - 2Y >2Y

Parent Funding by Maturity

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

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.

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

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

0

10

20

30

40

50

60

70

Croa

tiaSl

oven

iaEs

toni

aLa

tvia

Hun

gary

Bulg

aria

Lith

uani

aM

onte

negr

oRo

man

iaPo

land

Slov

ak R

epub

licSe

rbia

Czec

h Re

publ

icTu

rkey

Bosn

ia a

nd H

erze

govi

naM

aced

onia

, FYR

Alba

nia

Ukr

aine

Russ

iaM

oldo

vaBe

laru

s

CESE

E

To nonbanks To banks

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.

144

Table 10. External Positions of BIS-reporting Banks vis-à-vis CESEE

2008:Q3 2011:Q4 2012:Q1 2012:Q1 2011:Q3 2011:Q4 2012:Q12008:Q3-2012:Q1

2011:Q3 2011:Q4 2012:Q12008:Q3-2012:Q1

2011:Q3 2011:Q4 2012:Q12008:Q3-2012: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 -4.0CESEE 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 -7.2Emerging 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 6.8Belarus 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 -0.3Bosnia 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 -4.9Bulgaria 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 -11.0Croatia 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 -5.5Hungary 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 -26.9Latvia 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 -30.4Lithuania 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 -18.0Macedonia, 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 8.0Moldova 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 -2.9Montenegro 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 -0.9Poland 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 -0.6Romania 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 -7.2Russia 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 -3.3Serbia 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 -2.1Turkey 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 0.8Ukraine 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 -6.6Czech 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 -2.3Estonia 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 -35.6Slovak 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 -3.2Slovenia 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.

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

Change (Percent of GDP, exchange-rate adjusted)Stock

(Percent of Stock (US$ billions, exchange-rate

adjusted)Change (Percent, exchange-rate adjusted)Change (US$ billions, exchange-rate adjusted)

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

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.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

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

GIIPS-Bulgaria GIIPS-Poland GIIPS-Romania

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

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

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

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

GIIPS-Euro Stoxx Bulgaria-Euro Stoxx

Poland-Euro Stoxx Romania-Euro Stoxx

DCC GARCH Equity Market Model

Sources: Bloomberg; and IMF staff calculations.

146

Figure 34. DCC GARCH CDS Model

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.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

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

Bulgaria-Romania Hungary-Romania

Poland-Romania Romania-GIIPS

0

0.1

0.2

0.3

0.4

0.5

0.6

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

Bulgaria-GIIPS Hungary-GIIPS

Poland-GIIPS Romania-GIIPS

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

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

DCC GARCH CDS Model

Sources: Bloomberg; and IMF staff calculations.

147

Figure 35. DCC GARCH EMBIG Model

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.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

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

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

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

Romania-GIIPS10y Hungary-GIIPS10y Poland-GIIPS10y

Sources: Bloomberg; and IMF staff calculations.

DCC GARCH EEMBIG Model

148

Figure 36. ARCH Markov Switching Models

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

0

0.2

0.4

0.6

0.8

1

1/8/2007 1/8/2008 1/8/2009 1/8/2010 1/8/2011 1/8/2012

Romania EMBIG

Low Medium High

0

0.2

0.4

0.6

0.8

1

1/8/2007 1/8/2008 1/8/2009 1/8/2010 1/8/2011 1/8/2012

Romania Equity Market

Low Medium High

0

0.2

0.4

0.6

0.8

1

1/10/2007 1/10/2008 1/10/2009 1/10/2010 1/10/2011 1/10/2012

Romania 3m Interbank

Low Medium High

ARCH Markov Switching Models

Sources: Bloomberg; and IMF staff calculations.

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 ith . 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 2

1tD 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 ty being a vector of observed variables and ts denoting a unobserved

random variable with values 1, 2, …, K that as a state variable governs the conditional distribution of ty .

Prob ,...),,...,,|( 2121 ttttt yyksisjs Prob ijtt pisjs )|( 1 (3)

It is possible to combine all the transition probabilities ijp in a KK 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

ttt yy 1 (4)

The time varying variance 2

th with the error term t is parameterized as

,~~...~~

~

~

211

2222

2110

2ttqtqttt

ttt

tSt

daaaah

h

gt

(5)

where 1,0~ Nt , 3,2,1tS and 1td is a dummy variable in which 11 td if 0~1 t and

01 td if .0~1 t 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

tSg 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)

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.

0

50

100

150

200

250

0

50

100

150

200

250

ING

San

tand

er

HS

BC

Bar

clay

s

RB

S

BN

P P

arib

as

Cre

dit A

gric

ole

Gro

up

Deu

tsch

e B

ank

Lloy

ds B

anki

ng G

roup

Soc

iete

Gen

eral

e

Gro

upe

BC

PE

JP M

orga

n C

hase

Ban

k of

Am

eric

a

Citi

grou

p

Wel

ls F

argo

Gol

dman

Sac

hs

Mor

gan

Sta

nley

Ban

k of

NY

Mel

lon

Cap

ital O

ne

BB

T &

Cor

p.

% of Domestic GDP

% of EU GDP

% of US GDP

.

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)

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)

Source: Morgan Stanley. 1/ Banks under formal EU State Aid program as of September 2012.

0200400600800

100012001400

Jan

-08

Apr

-08

Jul-0

8

Oct

-08

Jan

-09

Apr

-09

Jul-0

9

Oct

-09

Jan

-10

Apr

-10

Jul-1

0

Oct

-10

Jan

-11

Apr

-11

Jul-1

1

Oct

-11

Jan

-12

Apr

-12

Jul-1

2

Oct

-12

Main Ref inancing Operations (MRO)

Longer Term Ref inancing (LTRO)

0

5

10

15

20

25

30

35

40

CB

KB

OI

KB

CLl

oy AIB SG

BK

TU

BI

Nat

ixis

BM

PS

BN

PB

TOR

BS

CS

BC

IVP

opU

BS

UC

GE

rste

HS

BC

CA

SA

SA

BP

MI

BB

VA ISP

Cai

xaS

AN

Bar

cD

B

State Aid Restructuring Banks' Average: 19

Non-restructured Banks' Average: 5 %

Non-restructuredState Aid Restructuring

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/

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.

0

2

4

6

8

10

12

2008-12 2009-12 2010-12 2011-12 2012-06

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)

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)

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.

0

3

6

9

Sep

-07

Dec

-07

Mar

-08

Jun-

08

Se p

-08

Dec

-08

Mar

-09

Jun-

09

Se p

-09

Dec

-09

Mar

-10

Jun-

10

Se p

-10

Dec

-10

Mar

-11

Jun-

11

Se p

-11

Dec

-11

Mar

-12

Jun-

12

Se p

-12

% o

f R

WA

0

5

10

15

20

25

30

35

Average of Jun-12 Average of Dec-07

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.

______________________ 1Box prepared by Marc Dobler. 2Staff 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.

Table A4.1. EU: Asset Quality Reviews Conducted in EU Countries: 2008–12 Ireland Greece Portugal Cyprus Spain

Jan–Mar 2011 Aug–Dec 2011 Jul–Nov 2011 Sept–Dec 2012 May–Jun 2012

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 loan portfolio management in the banks including arrears management and workout practices in curing NPLs and reducing loan losses;

As part of the 2nd 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.

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. The SIP had three different work streams (WS): the valuation of the credit

portfolio,

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-year horizon from mid-2012 to mid-2015.

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 base scenario, respectively, and the capital buffer requirement of €37 billion for a core Tier 1 threshold of 7 percent.

Ireland Greece Portugal Cyprus Spain

Jan–Mar 2011 Aug–Dec 2011 Jul–Nov 2011 Sept–Dec 2012 May–Jun 2012

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.

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.

The second part of the assessment with four domestic auditors was completed at the end of September.

171

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 Key Design Features EU Crisis Countries

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.

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-sharing arrangement).109

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.

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 Stress Tests

Type of Test Pros Cons

BU test (run by banks) Cash flow level data, use of models developed by banks, P&L effects of liquidity shocks and cost of funding shocks can be incorporated more easily.

Less consistent than TD

TD tests (run by authorities) Consistent approach, authority is flexible to run various scenarios, transparency of situation to authority

Less detailed data, bank-specific situation less recognized; data are outdated rapidly, which can 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

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.

ParameterVariable

Setup

Solvency(Stress Test) Expert

Judgment

Input data

Satellite models

Assumptions

Resultsummary

Results(incl. cash-flow template)

FundingCosts

Calculation(gap analysis)

Calculation(bank runs)

Calculation(cash-flows)

Core functionality

189

Table 13. Overview on the Main Elements of Three Liquidity Tests

Type of Test Description Outcome

Implied Cash Flow Analysis (ICFA)

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 period (30 days/3 months). Scenarios account for market liquidity of assets (in case of fire sales). The gradual test is usually run as a reverse test.

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?

(Proxy for) LCR, which assesses the counterbalancing capacity of banks for the next 30 days; The regulatory weights can be changed to assess sensitivities

Which portion of banks meets regulatory requirements? How much liquidity shortfall occurs, if applicable?

Maturity Mismatch/Rollover Stress Test

The liquidity gap simulation matches liability and asset maturities and identifies liquidity gaps for each maturity bucket and under different scenarios. The test is available (i) as a simplified version with limited data requirements and (ii) as a fully-fledged cash flow test.

Which portion of banks remains liquid up to a specific maturity bucket? How much is liquidity shortfall, if applicable?

(Proxy for) NSFR assesses the stability of banks’ funding sources in more structural terms.

Which portion of banks meets regulatory requirements? How much liquidity shortfall occurs, if applicable?

Integrated Liquidity and Solvency Tests

Simulates the impact of changes of solvency and concentration risk on liquidity conditions and vice versa (the first two modules require input from a solvency test and funding cost model, respectively).

Which portion of banks remains liquid/solvent under the specific assumptions? How much liquidity/capital shortfall, if applicable?

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

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 Deposits

Loss of Wholesale Funding

Empirical evidence142

Banking System in Saudi Arabia (August 1990143)

Banesto (ES, 1994)

Banking System in Argentina (2001)

Northern Rock (UK, 2007)

Parex Bank (LV, 2008)

IndyMac (US, June 2008)

Washington Mutual (US, September 2008)

DSB Bank (NL, 2009)

11 Percent (1 week)

8 percent (1 week)

Deposits in domestic currency: 30 percent (9 months)

57 percent (12 months)

25 percent (3 months)

7.5 percent (1 week)

8.5 percent (10 days)

30 percent (12 days)

57 percent (6 months)

Regulatory Parameters

LCR (30 days)

Stable: min. 5

Less stable: min. 10

(unsecured categories)

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

NSFR

Stable: 10

Less stable: 20

(unsecured categories)

Short-term corporate & public sector (< 1 year): 50

Rest: 100

Recent FSAPs144

10-50 percent (up to 80 percent for non-resident deposits)

10 to 50 percent for non-bank deposits

100 percent for bank funding

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)

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

Haircut/Weight for… Basel III (LCR) Basel III (NSFR)

Cash 00, includes also short-term securities (less

than 1 year)Issue Rating for Debt Security

Residual Maturity

SovereignOther

issuersSecu-

ritization< 1 year 0.5 1 2

1 to 5 years 2 4 8> 5 years 4 8 16< 1 year 1 2 4

1 to 5 years 3 6 12> 5 years 6 12 24

BB+ to BB- All 15Equity (main index) and GoldOther equityMutual funds (max of allowed asset mix)

(2) Encumbered liquid Assets (collateral)

n.a.Haircut on collateral for potential margin calls (3

notch downgrade)n.a.

(3) Add-ons (contigent liabilities)

n.a.

Lines - Retail: 5; corporate, credit lines: 10; corporate, liquidity lines: 100; other: 100

undrawn credit and liquidity lines: 5

Sovereign-type: 5; Corporate (< 1 year):

20

50Up to 25 (highest haircut applicable to any

security in the fund)

Sovereign (RW 0%): 0; Corporate: 15

Sovereign (RW 20%, Rating A+ to A-): 15

n.a.

n.a.

100

AAA to AA-/A-1

(1) Unencumbered liquid Assets

25

Basel II (2006, 2010)

0

A+ to BBB- and unrated

15

Not eligible

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

Source: Authors.

ScenarioModerate

Stress ScenarioMedium Stress

ScenarioSevere Stress

ScenarioVery Severe

Stress Scenario

Severity (x times Lehman/1) 0.25 0.5 1 2

Customer deposits (Term) 2.5 percent 5 percent 10 percent 20 percentCustomer deposits (Demand) 5 percent 10 percent 20 percent 40 percent

Short-term (secured) 5 percent 10 percent 20 percent 40 percentShort-term (unsecured) 25 Percent 50 Percent 100 Percent 100 Percent

Contingent liabilities0 Percent need

funding5 Percent need

funding10 Percent need

funding20 Percent need

funding

Haircut for Cash 0 Percent 0 Percent 0 Percent 0 PercentHaircut for Government Securities/2

1 Percent 2 Percent 5 Percent 10 Percent

Haircut for Trading Assets/3 3 Percent 6 Percent 30 Percent 100 Percent

Proxies, specific assetsEquities: 3; Bonds: 3

Equities: 4-6; Bonds: 3-8

Equity: 10-15; Bonds (only LCR eligible ones): 5-

10

Not liquid

Haircut for other securities 10 Percent 30 Percent 75 Percent 100 Percent

Proxies, specific assetsEquities: 10; Bonds: 10

Equities: 25; Bonds: 20 (some

not liquid)

Equity: 30; Bonds (only LCR eligible

ones): 20-30Not liquid

Percent of liquid assets encumbered/4

10 Percent (or actual figure)

20 Percent (or actual figure plus

10 ppt)

30 Percent (or actual figures plus 20 ppt)

40 Percent (or actual figures plus 30 ppt)

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.

Liquidity OutflowsCustomer Deposits

Wholesale Funding

Liquidity Inflows

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.

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

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.

Bank OECD EC LICAssets 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 6.1Liabilities & 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 11.6Contigent Liabilities 21.9 17.6 13

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

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.

-25%

-20%

-15%

-10%

-5%

0%

5%

10%

15%

20%

25%

Moderate Medium Severe Very severe

Net

Cas

h In

flow

(P

erce

nt o

f A

sset

s)

OECD EC LIC

Bank OECD EC LICTotal Outflow of Funding 21.8 19.5 16.7 Outflow of Deposits 6.8 9.3 10.4 Outflow of Wholesale Funding 12.8 8.4 5.0 Change of other funding 2.2 1.8 1.3Total Cash Inflow 11.4 15.3 16.7Net Cash Inflow -10.3 -4.1 0.1

Severe Scenario: Analysis of Drivers

Minimum number of periods of survival

Number of Banks illiquid

Survival -Percent of

Banks

0 0 100.0%1 0 100.0%2 0 100.0%3 1 66.7%4 1 66.7%5 2 33.3%

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

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

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 7,950Sum of cash inflows 1,875 4,275 8,925 7,200 6,375 5,100 13,000 23,800Net 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 gap)

22,925 20,700 19,725 18,875 15,900 8,400 11,350 7,400

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 7,950Sum of cash inflows 1,545 3,525 7,665 5,670 5,055 4,140 11,835 21,585Net funding gap -18,795 -11,335 2,595 580 555 -2,010 8,085 13,635Cumulative counterbalancing capacity (after HC and net funding gap)

12,900 1,393 170 -15 -833 -5,333 -2,445 -3,990

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

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

4.2%4.1%

21.4%

52.7%

12.4%

5.4%

21.9%

OECD Banks (n = 12,400)

Cash and Cash-like

Government Securities

Other securities

Customer Loans

Loans to Banks

Other assets

Offbalance

11.2% 7.8%

8.6%

56.2%

12.7%

3.6% 17.6%

Emerging Market Banks (n = 3,000)

Cash and Cash-like

Government Securities

Other securities

Customer Loans

Loans to Banks

Other assets

Offbalance

13.5%

8.3%

6.5%

47.1%

18.5%

6.1% 13.0%

Low Income Country Banks (n = 500)

Cash and Cash-like

Government Securities

Other securities

Customer Loans

Loans to Banks

Other assets

Offbalance

19.8%

27.9%

17.0%

16.7%

12.3%

6.3%

OECD Banks (n = 12,400)

Demand deposits

Term Deposits

Short-term wholesale funding

Long-term funding

Other Liabilities

Equity

23.3%

41.8%

11.2%

7.4%

5.1% 11.2%

Emerging Market Banks (n = 3,000)

Demand deposits

Term Deposits

Short-term wholesale funding

Long-term funding

Other Liabilities

Equity

39.6%

33.2%

6.6%

2.9% 6.2%

11.6%

Low Income Country Banks (n = 500)

Demand deposits

Term Deposits

Short-term wholesale funding

Long-term funding

Other Liabilities

Equity

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.

218

Table A8.1. Cash Flow Items

Cash-flows

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.

Contractual Outflows

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)

Own issuances due

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.

Unsecured wholesale funding due from non-financial corporates

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.

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

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.

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

219

Table A8.1. Cash Flow Items (continue)

Repos due with central banks

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.

Retail (incl. SME) funding due, term deposits

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.

Retail (incl. SME) funding due, demand deposits

Outflows from derivatives (other than FX-Swaps) Outflows from derivatives refer to contractual flows only.

Outflows from maturing FX-Swaps

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)

Other contractual outflows

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)

Behavioral Outflows

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.

Expected new loans

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.

Expected new financial investments

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).

220

Table A8.1. Cash Flow Items (continue Expected outflow and stock volume of undrawn committed credit/liquidity lines of financial institutes (incl. SPVs)

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.

Expected outflow and stock volume of undrawn committed credit/liquidity lines of others (corp., gov, etc.)

Expected outflows from new FX-Swaps

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)

Expected other behavioral outflows

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.

Contractual Inflows

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)

Maturing loans to financial institutions This position covers all inflows that result from contractual (already fixed) credit claims split by financial institutions (interbank deposits) and all other entities.

Other maturing loans (including installments)

paper in own portfolio maturing

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).

Reverse repos, secured by sovereign debt 0 percent r/w 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).

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

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.

Inflows from derivatives (other than FX-Swaps) The inflow of derivatives refer to contractual flows only

Other contractual inflows 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)

221

Table A8.1. Cash Flow Items (continue)

Behavioral Inflows

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.

Expected new debt issuances This position refers to the expected inflow created by new placements of debt / own issuances a bank is planning within the next 12 months.

Expected new retail deposits 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.

Expected new secured wholesale funding

Expected new unsecured wholesale funding

Expected inflows due to new FX Swaps

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)

Expected other behavioral inflows

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.

Counterbalancing Capacity - contractual and behavioural Security Flows

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 Government Debt Instruments

Local and regional government debt instruments

Traditional covered bonds

Credit institution debt instruments (unsecured)

Asset back securities

Debt instruments issued by Central Bank

Jumbo covered bonds

Debt instruments issued by corporate and other issuers

Debt instruments issued by financial corporations other than credit institutions (unsecured)

Agency debt instruments

Other covered bank bonds

Supranational debt instruments

Table A9.2. Haircuts Applied to Eligible Market Securities (ECB 2011)

Credit quality

Residual Maturity (years)

Category I Category II Category III Category IV C V

Fixed Coupon

Zero Coupon

Fixed Coupon

Zero Coupon

Fixed Coupon

Zero Coupon

Fixed Coupon

Zero Coupon

AAA to A-

0-1 0.5 0.5 1 1 1.5 1.5 6.5 6.5

16 /1

1-3 1.5 1.5 2.5 2.5 3 3 8.5 9.0 3-5 2.5 3 3.5 4 5.0 5.5 11.0 11.5 5-7 3 3.5 4.5 5 6.5 7.5 12.5 13.5

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

BBB+ to

BBB-

0-1 5.5 5.5 6.0 6.0 8.0 8.0 15.0 15.0

NA1-3 6.5 6.5 10.5 11.5 18.0 19.5 27.5 29.5 3-5 7.5 8.0 15.5 17.0 25.5 28.0 36.5 39.5 5-7 8 8.5 18.0 20.5 28.0 31.5 38.5 43.0

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

Source: Authors

Liquidity Outflows Example 1 Example 2 Example 3Deposits

Retail Russia (2008): 30 percentAustria (2007): 50 percent of ST nonbank customers

Ireland (2006): 10,30 and 50 percent

CorporateRussia (2008): 30 percent of current and 5 percent of

time deposits

Serbia (2009): 2 percent daily for 5 days

Spain (2005): 10-20 percent of demand deposits (retail and

corporate)Government UAE (2007): 35 percent

Non-resident UAE (2007): 30 percent

Belarus (2008): 25,50 and 75 percent (both households and

corporates)

Foreign bankLithuania (2007): 80

percentOther funding

Interbank Russia (2008): No accessUAE: 30 percent of

foreign interbank funding

Lithuania (2007): 100 percent of domestic interbank deposits

Credit Lines granted by banksRomania (2009): Limited access to committed and

uncommitted lines

Parent FundingLithuania (2007): 50

percent of liabilities to parent

Romania (2009): Limited access

Serbia (2009): Full withdrawal with maturity

of less than 1 year

Contingent liabilitiesSouth Africa (2008):

Tapping 50 percent of committed credit lines

Liquidity Inflows Example 1 Example 2 Example 3

Definition of Liquid Assets

Varies widely, most conservative: only cash and government bonds

remains liquid

Isle of Man (2008): Interbank funding and intra-group funding not

available

Haircuts on Liquid Assets

Austria (2007): Decrease in value of liquid bonds by 25 percent and equity by

35 percent

Ireland (2006): 10, 20 percent for debt securities and government bonds.

Russia (2008) and South Africa (2008): 20 percent

on liquid assets

Shrinkage of Eligible Collateral below Threshold

Austria (2007): Eligible colllateral in secured mkt shrinks by 30 percent and 30 percent of the eligible assets become ineligible

(i.e., their rating falls below single A).

Note: ST = Short-term

229

Annex 10. Link Between Solvency and Liquidity

Figure A10.1. Schematic Overview for the Calibration of Funding Costs

Illustrative example for a sample of large German banks

Source: Authors based on Moody’s KMV and OECD data

Time series of PDs(here: Moody’s KMV)

Time series of Implied Funding Costs(here: OECD, Market Data)

Historical Link(linear/non-

linear)

Implied Capitalization(here: from BII IRB model)

Implied Rating(here: evidence from

Moody’s 2010)

Increase of Funding costs (benchmark,

used for stress test)

Portion passedon to

customers?

LegendYellow: Input

Green: Output

RatingEDF or PD (One-year, Percent)

Funding costs (spread above T-

bills, bps)

Economic capital ratio

(Basel II (quasi-IRB)

Change of Funding

spread (CAR 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

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.

183 The banks’ capital plans were submitted by national supervisors to EBA at end January 2011.

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

239

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

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.

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

Source: Bloomberg and Authors’ Calculations

Figure 47. Estimated GARCH Correlations GIIPS with Non-European Countries

Source: Bloomberg and Authors’ Calculations

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

GIIPS-HK

GIIPS-Japan

GIIPS-Singapore

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

GIIPS-Belgium

GIIPS-France

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

GIIPS-Switzerland

GIIPS-UK

0

0.1

0.2

0.3

0.4

0.5

0.6

GIIPS-Denmark

GIIPS-Finland

GIIPS-Norway

GIIPS-Sweden

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

GIIPS-Austria

GIIPS-Netherlands

-0.05

0

0.05

0.1

0.15

0.2

0.25

GIIPS-Australia

GIIPS-Canada

271

Figure 48. Estimated GARCH Correlations GIIPS with EM Countries and Korea

Source: Bloomberg and Authors’ Calculations

Figure 49. Estimated GARCH Correlations GIIPS with Germany and the U.S.

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

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

GIIPS-Brazil

GIIPS-China

GIIPS-Mexico

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

GIIPS-India

GIIPS-Korea

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1

GIIPS-Germany

GIIPS-U.S.

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

GIIPS-Poland

GIIPS-Russia

GIIPS-Turkey

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. Empirical work includes Van den End (2008)222 at the Dutch Central Bank and Wong & Hui (2009)

from the Hong Kong Monetary Authority223, 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

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

Scenario (eg WEO)

GDP trajectory,

adjusted for spillover effects

Bank solvency parameters: Credit Losses

Security P/L impact Pre-impairment income

Bank liquidity parameters Haircuts (Market

Liquidity) Outflow of funding IMF Spillover analysis

Panel/GARCH

Overall soundness of bank Translation into bank-level stress

scenario: Solvency: Hardy/Schmieder

Liquidity: Schmieder/Hesse/at al

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

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.

0

5

10

15

20

25

Baseline (Sc 1) Sc 2a (Shock 100bps) Sc 2b (Shock 200bps) Sc 2c (Shock 300bps)

Capital needs (Tier 1, bn USD) for different scenarios

Less substantial spillover stress More substantial spillover stress

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

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.

0.0%

0.2%

0.4%

0.6%

0.8%

1.0%

1.2%

Baseline (Sc 1) Sc 2a (Shock100bps)

Sc 2b (Shock200bps)

Sc 2c (Shock300bps)

Liquidity Needs as a Percentage of Total Assets

Less Substantial Spillover Stress Substatial Spillover Stress

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.

281

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 VARIABLES1

(1) (2) (3) (4) (5) (6)

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 (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.

282

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 (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 Sovereign Risk

GIIPS spread Average of Euro periphery sovereign spreads to German Bunds

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.

Risk aversion

High-yield spread Difference between yields to maturity of AAA rated and BAA rated corporate US bond

VIX Implied volatility of S&P 500 index options.

Macroeconomic

environment

Openness Sum of imports and exports to GDP ratio

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 ith . 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 2

1tD 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

Source: Schmieder, Hesse and others (2012)

ScenarioModerate

Stress ScenarioMedium Stress

ScenarioSevere Stress

ScenarioVery Severe

Stress Scenario

Severity (x times Lehman/1) 0.25 0.5 1 2

Customer deposits (Term) 2.5 percent 5 percent 10 percent 20 percentCustomer deposits (Demand) 5 percent 10 percent 20 percent 40 percent

Short-term (secured) 5 percent 10 percent 20 percent 40 percentShort-term (unsecured) 25 Percent 50 Percent 100 Percent 100 Percent

Contingent liabilities0 Percent need

funding5 Percent need

funding10 Percent need

funding20 Percent need

funding

Haircut for Cash 0 Percent 0 Percent 0 Percent 0 PercentHaircut for Government Securities/2

1 Percent 2 Percent 5 Percent 10 Percent

Haircut for Trading Assets/3 3 Percent 6 Percent 30 Percent 100 Percent

Proxies, specific assetsEquities: 3; Bonds: 3

Equities: 4-6; Bonds: 3-8

Equity: 10-15; Bonds (only LCR eligible ones): 5-

10

Not liquid

Haircut for other securities 10 Percent 30 Percent 75 Percent 100 Percent

Proxies, specific assetsEquities: 10; Bonds: 10

Equities: 25; Bonds: 20 (some

not liquid)

Equity: 30; Bonds (only LCR eligible

ones): 20-30Not liquid

Percent of liquid assets encumbered/4

10 Percent (or actual figure)

20 Percent (or actual figure plus

10 ppt)

30 Percent (or actual figures plus 20 ppt)

40 Percent (or actual figures plus 30 ppt)

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.

Liquidity OutflowsCustomer Deposits

Wholesale Funding

Liquidity Inflows

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.

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 (increase of GIIPS sovereign debt spreads by…)

Impact on Austria, average for 2006-2012

Impact on Austria during peak spillover stress (2008-2012)

Increase of spreads (bps)

Source Increase of

spreads (bps) Source

100 bps (2a) 24.4232

(=24*1.017)

Table A12.1, spec (2233), Figure 1, top

right hand panel

49.8234 (=49*1.017)

Table A12.2, spec (1), Figure 1, top right

hand panel

200 bps (2b) 48.8 Same, linear

increase assumed 99.6

Same, linear increase assumed

300 bps (2c) 73.2 Same, linear

increase assumed 149.4

Same, linear increase 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

(2a/1) 0.9 1.4

(=1.8-0.5*3.5*0.244) 1.8

(=2.2-0.5*3.5*0.244) 1.3

(2b/1) 0.9 0.9

(=1.8-0.5*3.5*0.488) 1.3

(=2.2-0.5*3.5*0.488) 0.4

(2c/1) 0.9 0.5

(=1.8-0.5*3.5*0.732) 0.9

(=2.2-0.5*3.5*0.732) -0.4

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

(2a/2) 0.9 0.9

(=1.8-0.5*3.5*0.498) 1.3

(=2.2-0.5*3.5*0.498) 0.4

(2b/2) 0.9 0.1

(=1.8-0.5*3.5*0.996) 0.5

(=2.2-0.5*3.5*0.996) -1.2

(2c/2) 0.9 -0.8

(=1.8-0.5*3.5*1.494) -0.4

(=2.2-0.5*3.5*1.494) -3

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 credit exposure)

Pre-impairment income (Percent of total 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

total239 Haircut, Percent

(Annex 14) Haircut

Scenario 2c/2 Available 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

total240 Outflow, Percent

(Annex 14) Outflow

Scenario 2c/2 Required funding

Customer Term deposits 30 10 6.5 2 Customer Demand deposits 20 20 13

2.6

Secured short-term wholesale funding 10 20 13

1.3

Unsecured short-term wholesale funding 10 100 65

6.5

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.

243Nevertheless, 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

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

Mean Median Mean Median

Public Debt (percent of GDP, year before

restructuring)96.8 93.0 100.2 101.6

Current Account Balance (percent of GDP, average

over three years before restructuring)-6.4 -7.7 -5.7 -5.5

Overall Fiscal Balance (percent of GDP, average over

three years before restructuring)-5.0 -4.6 -3.4 -2.9

Primary Balance (percent of GDP, average over three

years before restructuring)0.5 0.4 0.7 0.9

Source: Staff reports, WEO

Reprofiling Debt reduction

Indicators of Fundamentals and Policy Track Record

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

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.

244Under 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.

245Regulators 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.

246For 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. __________________________________ 1Pakistan’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.

___________________________ 1The 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.” 2The 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.

3According 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.

__________________________________ 1This 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).

2Banks’ 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.

249This 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

250In 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

251The 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.

252Applying 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

253Jaramillo 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.

254The above studies measure home bias as the share of domestic banks’ sovereign holdings in total bank assets.

255In 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)

Sources: Arslanalp and Tsuda (2012); IMF IFS; IMF WEO.

256Details of the country sample, calculations of home bias measures, and sources of data are reported in Annex 13.1.

257The 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.

Home Bias Public debt / GDP Home Bias Public debt / GDPMean (whole sample) 5.5 56.4 12.8 51.1Mean of observations - Above 75th percentile 11.3 95.1 28.6 75.9 - Below 25th percentile 1.8 37.8 1.7 20.3Mean (whole sample) 6.5 79.9 13.5 54.0Mean of observations - Above 75th percentile 14.8 120.2 26.7 78.8 - Below 25th percentile 1.3 61.1 3.5 24.1

Advanced markets Emerging markets

2005-7 (debt - 2007)

2009-11 (debt - 2011)

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. 1Excluding 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:

itttitttitittittititi yxVIXhbhbbhbhbhbhbbr ,,21,,1,01,1, )*)(()*)(()(

--- (1)

312

where tir , captures long-term bond yields or spreads of country i at t ,258,259 tib , is the level of

public debt of country i at t , respectively. The variable tihb , is the home bias for country at t

, thb is the sample average of home bias at time t , tVIX is a proxy for global risk aversion

at time t , and tix , 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

258Throughout 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.

259For 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.

260Using 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).

261In 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).

262Throughout 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).

263In 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.

1VIX is fixed at sample mean (22.7 point). 2Debt 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. 1VIX is fixed at sample mean (18.5 point). 2Debt 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

Sources: Bloomberg and Fund staff estimates.

Table 21. Average EM and AM Estimated GARCH Correlations with VIX

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:

264The focus here is on the overall correlation magnitudes between AM and EM. Low correlation magnitudes are unlikely to be statistically significant.

265Similar 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.

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

2/1

/19

99

1/1

/20

00

12

/1/2

00

0

11

/1/2

00

1

10

/1/2

00

2

9/1

/20

03

8/1

/20

04

7/1

/20

05

6/1

/20

06

5/1

/20

07

4/1

/20

08

3/1

/20

09

2/1

/20

10

1/1

/20

11

12

/1/2

01

1

11

/1/2

01

2

VIX-Mexico

VIX-Russia

VIX-Turkey

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

0.6

2/1

/19

99

2/1

/20

00

2/1

/20

01

2/1

/20

02

2/1

/20

03

2/1

/20

04

2/1

/20

05

2/1

/20

06

2/1

/20

07

2/1

/20

08

2/1

/20

09

2/1

/20

10

2/1

/20

11

2/1

/20

12

2/1

/20

13

VIX-Ireland

VIX-Italy

VIX-Portugal

VIX-Spain

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%

317

ittititi xhbb ,,1, --- (2)

ittititi xhbb ,1,2, --- (3)

where tib , is the level of public debt at t , tihb , and 1, tihb are home bias at t and 1t ,

respectively, and tix , 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

266For 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.

267Disentangling 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)

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.

ittitittittititititi xbhbhbhbhbbbbpb ,1,,1,03

1,32

1,21,1, *)(()( -- (4)

where tipb , is the primary balance at time t , 1, tib , 21, tib , 3

1, tib are linear, quadratic and cubic

terms of lagged public debt at t , tihb , is the country-specific home bias and thb is the sample

median of home bias at t , and tix , 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

ArgentinaBrazil

Turkey

Lebanon

Mexico

Japan

St. Kitts and Nevis

0

40

80

120

160

200

0 5 10 15 20 25 30 35 40 45

Public debt/GDP (2007)

Home Bias - Banks' Holding of Domestic Sovereign Claims/Banks' Total Assets, Average 2005-7

Public Debt/GDP in 2007 and Home Bias ave. 2005-7

AM+EM

Linear regression

Debt = 55.3 + 2.1 * Home bias(***, 9.63) (***, 0.61)

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).

268These two instruments have enough explanatory power supported by a high Adj-R2.

269Mauro and others (2013) find that there are significant variations in primary balance adjustments across countries and time periods.

270Due 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.

271A 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

Sources: IMF WEO; and Fund staff estimates.

Public debt-to-GDP, lagged (%)

Primary balance-to-GDP (%)

HB 5% below Median

Median of whole observation

HB 5% above 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

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

Sources: Baldacci and others (2011); and Cruces and Trebesch (2013).

ARG 2001

JAM 2012JAM 2010

VEN 2005

KNA 2011

HUN 2008

BGR 2002

GRC 2010

ATG 2010

ROU 2009

BLZ 2012

PRT 2011BLZ 2006

LVA 2008

UKR 2008

IRL 2010

ECU 2008

0

40

80

120

160

0 5 10 15 20 25

Debt Level in Time of Distress (In percent of GDP)

Home Bias - Banks' Holding of Domestic Sovereign Claims / Banks' Total Assets (3 year average)

AM & EM under debt distress

Linear regression

Debt = 75.05 + 2.63 * Home bias(***, 19.25) (*, 1.37)

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

272If 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.

273For 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

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.

214 210 204

180 178

156

115 108 10697 94 88 86

62 57 53 52 4529 23 22

0

50

100

150

200

250

DE MT IT BE SI ES LU PT PL CY IE GR HU NL AT NO FR GB SE DK FI

Domestic holdings of sovereign debt, as percent of Core Tier 1

Source: EBA.Data as of 1H2013

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 /

Monthly IMF’s World Economic Outlook (WEO) / IFS

EMBI stripped spreads In percent Annual / Monthly

Bloomberg

Debt-to-GDP In percent Annual WEO database Primary balance to GDP ratio

In percent Annual WEO database

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

CPI inflation Annual Staff calculations based on

WEO database Exchange rate depreciation In percent Annual Staff calculations based on

WEO database Institutional quality index Smaller (larger) values indicating

higher (lower) political risk. Annual International Country Risk

Guide (ICRG) dataset. Credit to GDP ratio In percent Annual IFS database Capital account openness Higher indices indicating a high

degree of capital account openness Annual Chin and Ito (2006)

VIX Chicago Board Options Exchange Market Volatility Index

Annual / Monthly

Bloomberg

U.S. long-term bond yields In percent Annual WEO database Output gap Difference between actual and

potential (calculated using the Hodrick-Prescott filter real GDP)

Annual Staff calculations based on WEO database

Government expenditure Gap

Difference between actual and potential (calculated using the Hodrick-Prescott filter real GDP)

Annual Staff calculations based on WEO database

Trade openness Sum of exports and imports to GDP (in percent)

Annual Staff calculations based on WEO database

Oil price Log of (trend) oil price applied to oil exporters only.

Annual Staff calculations based on 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 ith . 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 2

1tD 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).

274Given 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.

331

ANNEX 20. Home Bias Regression Tables

Table A20.1. Regression of Bond Spreads—(1) AM

Dependent variable: Bond Spreads 1/ (1) HB HB/Debt (2)Baseline - HB.

HB/Debt, HB/VIX (3) Omitting Greece, Japan, Portugal

(4) Quartile regression

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)

0 (Deviation of home bias from median)

1.361***

(0.493)

1.255**

(0.515)

1.226*

(0.620)

1 (Deviation of home bias from median * Public debt/GDP,

level, lagged)

-0.027***

(0.009)

-0.027***

(0.012)

-0.026***

(0.012)

-0.013** (0.0065)

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 (Inflation rate, 3-year MA) -0.089

(0.101) -0.087

(0.101) 0.026

(0.109) 0.178

(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

1 (Public debt/GDP, lagged)

0.046*** (0.010)

0.050***(0.015)

0 (Deviation of Home bias from median) -0.223** (0.088)

-0.202**(0.088)

1 (Deviation of Home bias from median *

Public debt/GDP, level, lagged)

- -0.0007(0.002)

2 (Deviation of Home bias from median *

VIX)

0.012*** (0.0035)

0.012***(0.0036)

1 (GDP growth rate) -0.258***

(0.063) -0.264***

(0.066)

2 (Inflation rate, 3-year MA) 0.201* (0.102)

0.202*(0.105)

3 (Exchange rate depreciation) -0.0012 (0.021)

-0.002(0.021)

4 (Capital account openness) 0.054

(0.214) 0.068

(0.217)

5 (Credit-to-GDP ratio) 0.0028 (0.015)

0.0028(0.015)

1 (VIX) 0.057**(0.026)

0.055**(0.026)

2 (U.S. long-term bonds)

0.977*** (0.313)

0.960***(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.

333

Table A20.3. Regression of the Public Debt/GDP—AM, EM, AM&EM (A) AM (B) EM (C) AM & EM

Dependent variable: Public debt to GDP ratio

(1) IV pooled regression

(2) Least Square—pooled

regression

(1) IV pooled

estimation

(2) Least Square—pooled

regression

(1) IV pooled regression (2) Least

Square—pooled regression

1 (Home bias) 2.307**(1.127)

- 1.104*** (0.392)

- 0.420 (0.518)

-

1 (Home bias, lagged)

3.753*** (0.520)

1.919*** (0.157)

2.529***

(0.242)

1 (Output gap) -1.742***

(0.669) -2.342***

(0.666) -0.913***

(0.320) -0.863***

(0.318) -1.356***

(0.458) -0.907***

(0.307)

2 (Government expenditure gap) -1.095

(1.146) -0.414

(1.166) -0.132

(0.165) 0.043

(0.120) -0.427

(0.264) 0.188

(0.160)

3 (Trade openness) -0.074*(0.045)

-0.137*** (0.047)

-0.082 (0.070)

0.287*** (0.064)

-0.131*** (0.041)

-0.052***(0.032)

4 (Inflation rate, 3-year MA) -4.318**(1.901)

-10.604*** (1.870)

0.093*** (0.028)

0.032*** (0.023)

0.172*** (0.041)

0.087***(0.032)

5 (Oil price, lagged) - - -5.186***

(1.145) -1.655** (0.774)

-4.502*** (1.414)

-1.018(0.782)

6 (Capital account restriction) 18.223***

(3.036) 13.179***

(2.618) -4.301***

(1.073) 4.430*** (0.977)

-4.502*** (1.414)

7.821***(1.251)

7 (Credit-to-GDP ratio) 0.087

(0.068) 0.197*** (0.042)

-0.111 (0.126)

-0.207** (0.092)

4.712** (2.051)

0.181***(0.049)

8 (Institutional quality) -1.243**(0.484)

-0.857** (0.406)

-0.225 (0.350)

-0.567 (0.774)

-1.094*** (0.393)

-0.865***(0.279)

Constant 118.25**(46.961)

96.480*** (34.614)

56.22 (36.016)

44.079*** (27.290)

121.442*** (34.916)

63.275***(20.937)

Dummy variable for AM - - - - 37.076***

(7.891) 23.936***

(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.

334

Table A20.4. Regression of the Fiscal Reaction Function—AM&EM Dependent variable: Primary balance to GDP ratio

(1) HB indicator—constant

(2) HB indicator—constant/linear

(3) HB indicator—linear interactive

(4) HB indicator—quadratic interactive

(5) HB indicator—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

debt/GDP, lagged)

- 0.00096 (0.00099)

-0.00024 (0.00046)

-

-

2 (Deviation of HB from median * Public

debt/GDP, square, lagged)

- - - -0.0000023 (0.0000031)

-

3 (Deviation of HB from median * Public

debt/GDP, cubic, lagged)

- - - - -0.000000022 (0.000000019)

1 (Output gap) 0.151*** (0.050)

0.140*** (0.050)

0.155*** (0.052)

0.157*** (0.052)

0.158*** (0.052)

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

335

Table A20.5. Robustness Check for Bond Spreads Regression—AM, EM

Dependent variable:

(A) Banks’ holding of domestic sovereign claims / Total sovereign

claims—AM

(B) Banks’ holding of domestic sovereign claims /

Public Debt–AM & EM

Bond spreads-AM Bond spreads-AM Bond yields-EM

1 (Public debt/GDP, lagged)

0.032*** (0.012)

0.052*** (0.012)

0.049*** (0.014)

0 (Deviation of Home bias from median) -

0.657** (0.274)

-0.201** (0.087)

1 (Deviation of Home bias from median *

Public debt/GDP, level, lagged)

-0.0012** (0.00047)

-0.014*** (0.005)

-0.001 (0.016)

2 (Deviation of Home bias from median *

VIX)

0.0018** (0.00086)

0.0019 (0.006)

0.012*** (0.0035)

1 (GDP growth rate) -0.049***

(0.016) -0.045***

(0.013) -0.264***

(0.066)

2 (Inflation rate, 3-year MA) -0.059

(0.095) -0.086

(0.101) 0.201* (0.105)

3 (Exchange rate depreciation) -0.0085

(0.0053) -0.009* (0.005)

-0.002 (0.021)

4 (Capital account openness) - - 0.068

(0.217)

5 (Credit-to-GDP ratio) 0.009** (0.004)

0.009** (0.004)

0.0026 (0.015)

1 (VIX) 0.012* (0.007)

0.007 (0.015)

0.059**(0.026)

2 (U.S. long-term bonds)

- - 0.952*** (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.

336

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.

Table A20.6. Robustness Check for Public Debt Regression

(A) Banks’ holding of domestic sovereign claims / Total sovereign claims—AM

(B) Banks’ holding of domestic sovereign claims / Public Debt—AM & EM

Dependent variable: Public debt to GDP ratio

(1) IV pooled regression (2) Least Square—pooled regression

(1) IV 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)

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)

Constant - 52.8**

(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

337

Table A20.7. Robustness Check for Fiscal Reaction Function—AM

Dependent variable: Primary balance to GDP ratio

(A) Banks’ holding of domestic sovereign claims / Total sovereign 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

debt/GDP, lagged)

0.0019** (0.00094)

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.